BNC Handbook : Exploring the British National Corpus With SARA

title :
Edinburgh Textbooks in Empirical Linguistics
author : Aston, Guy.; Burnard, Lou.
publisher : Edinburgh University Press
isbn10 | asin : 0748610545
print isbn13 : 9780748610549
ebook isbn13 : 9780585132310
language : English
Dodd, A.--1952---(Anthony),--SARA (Computer file) , British National
subject Corpus, English language--Data processing.--Great Britain ,
Computational linguistics.
publication date : 1998
lcc : PE1074.5.A88 1998eb
ddc : 410.285
Dodd, A.--1952---(Anthony),--SARA (Computer file) , British National
subject : Corpus, English language--Data processing.--Great Britain ,
Computational linguistics.
cover

Page i

The BNC Handbook

page_i

Page ii

EDINBURGH TEXTBOOKS IN EMPIRICAL LINGUISTICS

CORPUS LINGUISTICS
by Tony McEnery and Andrew Wilson
LANGUAGE AND COMPUTERS
A PRACTICAL INTRODUCTION TO THE COMPUTER ANALYSIS OF LANGUAGE
by Geoff Barnbrook
STATISTICS FOR CORPUS LINGUISTICS
by Michael P. Oakes
COMPUTER CORPUS LEXICOGRAPHY
by Vincent B.Y. Ooi
EDITORIAL ADVISORY BOARD
Ed Finegan
 University of Southern California, USA
Dieter Mindt
Freie Universität Berlin, Germany
Bengt Altenberg
Lund University, Sweden
Knut Hofland
Norwegian Computing Centre for the Humanities, Bergen, Norway
Jan Aarts
Katholieke Universiteit Nijmegen,The Netherlands
Pam Peters
Macquarie University, Australia
If you would like information on forthcoming titles in this series, please contact
Edinburgh University Press, 22 George Square, Edinburgh EH8 9LF

page_ii

Page iii

EDINBURGH TEXTBOOKS IN EMPIRICAL LINGUISTICS

Series Editors: Tony McEnery and Andrew Wilson

The BNC Handbook

Exploring the British National Corpus with SARA

Guy Aston and Lou Burnard

EDINBURGH UNIVERSITYPRESS

page_iii

Page iv

EDINBURGH UNIVERSITY PRESS

© Guy Aston and Lou Burnard, 1998
Edinburgh University Press
22 George Street, Edinburgh EH8 9LF
Printed and bound in Great Britain by the University Press, Cambridge
A CIP record for this book is available from the British Library
ISBN 0 7486 1054 5 (cased)
ISBN 0 7486 1055 3 (paperback)

page_iv

Page v

CONTENTS

I: Corpus Linguistics and the BNC 1

1 4
Corpus Linguistics

1.1 4
What is a Corpus?

1.2 5
What Can You Get Out of a Corpus?

1.3 10
How have Corpora been Used?

1.3.1 10
What Kinds of Corpora Exist?

1.3.2 12
Some Application Areas

1.3.3 13
Collocation

1.3.4 15
Contrastive Studies

1.3.5 18
NLP Applications

1.3.6 19
Language Teaching

1.4 21
How Should a Corpus Be Constructed?

1.4.1 21
Corpus Design

1.4.2 24
Encoding, Annotation, and Transcription

2 28
The British National Corpus

2.1 28
How the BNC was Constructed

2.1.1 28
Corpus Design

2.1.2 33
Encoding, Annotation, and Transcription

2.2 36
Using the BNC: Some Caveats

2.2.1 37
Source Materials

2.2.2 38
Sampling, Encoding, and Tagging Errors

2.2.3 39
What is a BNC Document?

2.2.4 40
Miscellaneous Problems

3 42
Future Corpora
II: Exploring the BNC with SARA 45

1 48
Old Words and New Words

1.1 48
The Problem: Finding Evidence of Language Change

1.1.1 48
Neologism and Disuse

1.1.2 48
Highlighted Features

1.1.3 49
Before You Start

1.2 49
Procedure

1.2.1 49
Starting SARA

1.2.2 49
Logging On

1.2.3 50
Getting Help

1.2.4 50
Quitting SARA

1.2.5 51
Using Phrase Query to Find a Word: 'Cracksmen'

page_v

Page vi

1.2.6 51
Viewing the Solutions: Display Modes

1.2.7 52
Obtaining Contextual Information: the Source and Browse Options

1.2.8 54
Changing the Defaults: the User Preferences Dialogue Box

1.2.9 57
Comparing Queries: 'Cracksmen' and 'Cracksman'

1.2.10 58
Viewing Multiple Solutions: 'Whammy'

1.3 60
Discussion and Suggestions for Further Work

1.3.1 60
Caveats
 1.3.2 61
Some Similar Problems

2 63
What is More Than One Corpus?

2.1 63
The Problem: Relative Frequencies

2.1.1 63
'Corpus' in Dictionaries

2.1.2 63
Highlighted Features

2.1.3 64
Before You Start

2.2 64
Procedure

2.2.1 64
Finding Word Frequencies Using Word Query

2.2.2 65
Looking for More Than One Word Form Using Word Query

2.2.3 66
Defining Download Criteria

2.2.4 68
Thinning Downloaded Solutions

2.2.5 70
Saving and Re-opening Queries

2.3 72
Discussion and Suggestions for Further Work

2.3.1 72
Phrase Query or Word Query?

2.3.2 72
Some Similar Problems

3 74
When is Ajar not a Door?

3.1 74
The Problem: Words and Their Company

3.1.1 74
Collocation

3.1.2 75
Highlighted Features

3.1.3 75
Before You Start

3.2 76
 Procedure

3.2.1 76
Using the Collocation Option

3.2.2 78
Investigating Collocates Using the Sort Option

3.2.3 81
Printing Solutions

3.2.4 82
Investigating Collocations without Downloading: Are Men as Handsome as
Women are Beautiful?

3.3 83
Discussion and Suggestions for Further Work

page_vi

Page vii

3.3.1 83
The Significance of Collocations

3.3.2 85
Some Similar Problems

4 86
A Query too Far

4.1 86
The Problem: Variation in Phrases

4.1.1 86
A First Example: 'The Horse's Mouth'

4.1.2 86
A Second Example: 'A Bridge too Far'

4.1.3 86
Highlighted Features

4.1.4 87
Before You Start

4.2 87
Procedure

4.2.1 87
Looking for Phrases Using Phrase Query

4.2.2 91
Looking for Phrases Using Query Builder

4.3 95
Discussion and Suggestions for Further Work
 4.3.1 95
Looking for Variant Phrases

4.3.2 95
Some Similar Problems

5 98
Do People Ever Say 'You Can Say That Again'?

5.1 98
The Problem: Comparing Types of Texts

5.1.1 98
Spoken and Written Varieties

5.1.2 98
Highlighted Features

5.1.3 99
Before You Start

5.2 99
Procedure

5.2.1 99
Identifying Text Type by Using the Source Option

5.2.2 100
Finding Text-type Frequencies Using SGML Query

5.2.3 103
Displaying SGML Markup in Solutions

5.2.4 105
Searching in Specific Text-types: 'Good Heavens' in Real and Imagined
Speech

5.2.5 107
Comparing Frequencies in Different Text-types

5.3 108
Discussion and Suggestions for Further Work

5.3.1 108
Investigating Other Explanations: Combining Attributes

5.3.2 110
Some Similar Problems

6 112
Do Men Say Mauve?

6.1 112
The Problem: Investigating Sociolinguistic Variables

6.1.1 112
Comparing Categories of Speakers

6.1.2 112
Components of BNC Texts
 6.1.3 114
Highlighted Features

6.1.4 115
Before You Start

6.2 115
Procedure

6.2.1 115
Searching in Spoken Utterances

page_vii

Page viii

6.2.2 116
Using Custom Display Format

6.2.3 117
Comparing Frequencies for Different Types of Speaker: Male and Female
'Lovely'

6.2.4 123
Investigating Other Sociolinguistic Variables: Age and 'Good Heavens'

6.3 125
Discussion and Suggestions for Further Work

6.3.1 125
Sociolinguistic Variables in Spoken and Written Texts

6.3.2 126
Some Similar Problems

7 129
'Madonna Hits Album' Did it Hit Back?

7.1 129
The Problem: Linguistic Ambiguity

7.1.1 129
The English of Headlines

7.1.2 129
Particular Parts of Speech, Particular Portions of Texts

7.1.3 130
Highlighted Features

7.1.4 131
Before You Start

7.2 131
Procedure

7.2.1 131
Searching for Particular Parts-of-Speech with POS Query: 'Hits' as a Verb
 7.2.2 133
Searching within Particular Portions of Texts with Query Builder

7.2.3 135
Searching within Particular Portions of Particular Text-types with CQL Query

7.2.4 137
Displaying and Sorting Part-of-Speech Codes

7.2.5 138
Re-using the Text of One Query in Another: 'Hits' as a Noun

7.2.6 139
Investigating Colligations Using POS Collating

7.3 140
Discussion and Suggestions for Further Work

7.3.1 140
Using Part-of-Speech Codes

7.3.2 141
Some Similar Problems

8 143
Springing Surprises on the Armchair Linguist

8.1 143
The Problem: Intuitions about Grammar

8.1.1 143
Participant Roles and Syntactic Variants

8.1.2 144
Inflections and Derived Forms

8.1.3 145
Variation in Order and Distance

8.1.4 146
Highlighted Features

8.1.5 146
Before You Start

8.2 147
Procedure

page_viii

Page ix

8.2.1 147
Designing Patterns Using Word Query: Forms of 'Spring'

8.2.2 150
Using Pattern Query
 8.2.3 150
Varying Order and Distance between Nodes in Query Builder

8.2.4 153
Checking Precision with the Collocation and Sort Options

8.2.5 154
Saving Solutions with the Listing Option

8.3 157
Discussion and Suggestions for Further Work

8.3.1 157
Using Two-way Links

8.3.2 158
Some Similar Problems

9 161
Returning to More Serious Matters

9.1 161
The Problem: Investigating Positions in Texts

9.1.1 161
Meaning and Position

9.1.2 161
Laughter and Topic Change

9.1.3 162
Highlighted Features

9.1.4 162
Before You Start

9.2 163
Procedure

9.2.1 163
Searching in Sentence-initial Position: 'Anyway' and 'Anyhow'

9.2.2 165
Customizing the Solution Display Format

9.2.3 168
Sorting and Saving Solutions in Custom Format

9.2.4 170
Searching at Utterance Boundaries: Laughs and Laughing Speech

9.3 176
Discussion and Suggestions for Further Work

9.3.1 176
Searching Near Particular Positions

9.3.2 177
Some Similar Problems

10 179
What Does 'SARA' Mean?

10.1 179
The Problem: Studying Pragmatic Features

10.1.1 179
How Are Terms Defined?

10.1.2 179
Highlighted Features

10.1.3 180
Before You Start

10.2 180
Procedure

10.2.1 180
Looking for Acronyms with POS Query

10.2.2 181
Classifying Solutions with Bookmarks

10.2.3 182
Searching in Specified Texts

10.2.4 184
Serendipitous Searching: Varying the Query Type

10.2.5 186
Finding Compound Forms: Combining Phrase and Word Queries

page_ix

Page x

10.2.6 187
Viewing Bookmarked Solutions

10.2.7 189
Including Punctuation in a Query

10.3 191
Discussion and Suggestions for Further Work

10.3.1 191
Using Bookmarks

10.3.2 191
Punctuation in Different Query Types

10.3.3 192
Some Similar Problems

III: Reference Guide 195

1 196
Quick Reference Guide to the SARA Client
1.1 196
Logging on to the SARA System

1.2 197
The Main SARA Window

1.3 199
The File Menu

1.3.1 200
Defining a Query

1.3.2 200
Defining a Word Query

1.3.3 202
Defining a Phrase Query

1.3.4 203
Defining a POS Query

1.3.5 204
Defining a Pattern Query

1.3.6 206
Defining an SGML Query

1.3.7 207
Defining a Query with Query Builder

1.3.8 210
Defining a CQL Query

1.3.9 212
Execution of SARA Queries

1.3.10 213
Printing Solutions to a Query

1.4 213
The Edit Menu

1.5 215
The Browser Menu

1.6 216
The Query Menu

1.6.1 216
Editing a Query

1.6.2 217
Sorting Solutions

1.6.3 218
Thinning Solutions

1.6.4 218
Options for Displaying Solutions
 1.6.5 221
Additional Components of the Query Window

1.6.6 222
Saving Solutions to a File

1.6.7 223
Displaying Bibliographic Information and Browsing

1.6.8 224
The Collocation Command

1.7 225
The View Menu

1.7.1 225
Tool Bar Command

1.7.2 226
Status Bar Command

1.7.3 227
Font

1.7.4 227
Colours

1.7.5 227
User Preferences

page_x

Page xi

1.8 228
The Window Menu

1.9 229
The Help Menu

1.10 229
Installing and Configuring the SARA Client

2 230
Code Tables

2.1 230
POS Codes in the CLAWS5 Tagset

2.2 234
Text Classification Codes

2.3 238
Dialect Codes

2.4 239
Other Codes Used
 3 240
SGML Listing Format

4 242
Bibliography

page_xi

Page xiii

PREFACE
The British National Corpus is a collection of over 4000 samples of modern British English, both spoken and written, stored in electronic form and
selected so as to reflect the widest possible variety of users and uses of the language. Totalling over 100 million words, the corpus is currently being
used by lexicographers to create dictionaries, by computer scientists to make machines 'understand' and produce natural language, by linguists to
describe the English language, and by language teachers and students to teach and learn it to name but a few of its applications.
Institutions all over Europe have purchased the BNC and installed it on computers for use in their research. However it is not necessary to possess a
copy of the corpus in order to make use of it: it can also be consulted via the Internet, using either the World Wide Web or the SARA software system,
which was developed specially for this purpose.
The BNC Handbook provides a comprehensive guide to the SARA software distributed with the corpus and used on the network service. It illustrates
some of the ways in which it is possible to find out about contemporary English usage from the BNC, and aims to encourage use of the corpus by a
wide and varied public. We have tried as far as possible to avoid jargon and unnecessary technicalities; the book assumes nothing more than an interest
in language and linguistic problems on the part of its readers.
The handbook has three major parts. It begins with an introduction to the topic of corpus linguistics, intended to bring the substantial amount of
corpus-based work already done in a variety of research areas to the non-specialist reader's attention. It also provides an outline description of the
BNC itself. The bulk of the book however is concerned with the use of the SARA search program. This part consists of a series of detailed task
descriptions which (it is hoped) will serve to teach the reader how to use SARA effectively, and at the same time stimulate his or her interest in using the
BNC. There are ten tasks, each of which introduces a new group of features of the software and of the corpus, of roughly increasing complexity. At
the end of each task there are suggestions for further related work. The last part of the handbook gives a summary overview of the SARA program's
commands and capabilities, intended for reference purposes, details of the main coding schemes used in the corpus, and a select bibliography.
The BNC was created by a consortium led by Oxford University Press, together with major dictionary publishers Longman and Chambers, and
research centres at the Universities of Lancaster and Oxford, and at the British Library. Its creation was jointly funded by the Department of Trade and
Industry and

page_xiii

Page xiv

the Science and Research Council (now EPSRC), under the joint Framework for Information Technology programme, with substantial investment from
the commercial partners in the consortium. Reference information on the BNC and its creation is available in the BNC Users' Reference Guide
(Burnard 1995), which is distributed with the corpus, and also from the BNC project's web site at http://info.ox.ac.uk/bnc.
This handbook was prepared with the assistance of a major research grant from the British Academy, whose support is gratefully acknowledged.
Thanks are also due to Tony Dodd, author of the SARA search software, to Phil Champ, for assistance with the SARA help file, to Jean-Daniel Fekète
and Sebastian Rahtz for help in formatting, and to our respective host institutions at the Universities of Oxford and Bologna. Our greatest debt however
is to Lilette, for putting up with us during the writing of it.

page_xiv

Page 1

I:
CORPUS LINGUISTICS AND THE BNC

page_1
 Page 2

Introduction
This part of the BNC Handbook attempts to place the British National Corpus (BNC) within the tradition of corpus linguistics by providing a brief
overview of the theoretical bases from which that tradition springs, and of its major achievements. We begin by defining the term corpus pragmatically,
as it is used in linguistics, proceeding to review some uses, both actual and potential, for language corpora in various fields. Our discussion focuses
chiefly on the areas of language description (with particular reference to linguistic context and collocation, and to contrastive and comparative studies),
of Natural Language Processing, and of foreign language teaching and learning.
We then present an overview of some of the main theoretical and methodological issues in the field, in particular those concerned with the creation,
design, encoding, and annotation of large corpora, before assessing the practice of the British National Corpus itself with respect to these issues. We
also describe some particular characteristics of the BNC which may mislead the unwary, and finally suggest some possible future directions for corpora
of this kind.

page_2

Page 3

corpus ... pl. corpora ... 1. The body of a man or animal. (Cf. corpse.) Formerly frequent; now only
humorous or grotesque. 1854 Villikins & his Dinah (in Mus. Bouquet, No. 452), He kissed her cold corpus
a thousand times o'er. 2. Phys. A structure of a special character or function in the animal body, as corpus
callosum, the transverse commissure connecting the cerebral hemispheres; so also corpora quadrigemina,
striata, etc. of the brain, corpus spongiosum and corpora cavernosa of the penis, etc.; corpus luteum L.
luteus, -um yellow (pl. corpora lutea), a yellowish body developed in the ovary from the ruptured Graafian
follicle after discharge of the ovum; it secretes progesterone and other hormones and after a few days
degenerates unless fertilization has occurred, when it remains throughout pregnancy. 1869 Huxley Phys. xi.
298 The floor of the lateral ventricle is formed by a mass of nervous matter, called the corpus striatum. 1959
New Biol. XXX. 79 As in mammals, glandular bodies known as corpora lutea are produced in the ovaries of
viviparous (and also of some oviparous) reptiles, in places from which the eggs have been shed at ovulation.
3. A body or complete collection of writings or the like; the whole body of literature on any subject. 172751
Chambers Cycl. s.v., Corpus is also used in matters of learning, for several works of the same nature,
collected, and bound together.. We have also a corpus of the Greek poets.. The corpus of the civil law is
composed of the digest, code, and institutes. 1865 Mozley Mirac. i. 16 Bound up inseparably with the whole
corpus of Christian tradition. 4. The body of written or spoken material upon which a linguistic analysis is
based. 1956 W. S. Allen in Trans. Philol. Soc. 128 The analysis here presented is based on the speech of a
single informant.. and in particular upon a corpus of material, of which a large proportion was narrative,
derived from approximately 100 hours of listening. 1964 E. Palmer tr. Martinet's Elem. General Linguistics
ii. 40 The theoretical objection one may make against the 'corpus' method is that two investigators operating
on the same language but starting from different 'corpuses', may arrive at different descriptions of the same
language. 1983 G. Leech et al. in Trans. Philol. Soc. 25 We hope that this will be judged.. as an attempt to
explore the possibilities and problems of corpus-based research by reference to first-hand experience, instead
of by a general survey. 5. The body or material substance of anything; principal, as opposed to interest or
income. 1884 Law Rep. 25 Chanc. Div. 711 If these costs were properly incurred they ought to be paid out
of corpus and not out of income. phr. corpus delicti (see quot. 1832); also, in lay use, the concrete evidence
of a crime, esp. the body of a murdered person. corpus juris: a body of law; esp. the body of Roman or civil
law (corpus juris civilis). 1891 Fortn. Rev. Sept. 338 The translation.. of the Corpus Juris into French. 1922
Joyce Ulysses 451 (He extends his portfolio.) We have here damning evidence, the corpus delicti, my lord, a
specimen of my maturer work disfigured by the hallmark of the beast. 1964 Sunday Mail Mag. (Brisbane)
13 Sept. 3/3 An enthusiastic trooper, one of a party investigating river, dam and hollow log in search of the
corpus delicti, found some important evidence in a fallen tree. corpus vile Pl. corpora vilia Orig. in phr. (see
quot. 1822) meaning 'let the experiment be done on a cheap (or worthless) body'. A living or dead body that
is of so little value that it can be used for experiment without regard for the outcome; transf., experimental
material of any kind, or something which has no value except as the object of experimentation. 1822 De
Quincey Confess. App. 189 Fiat experimentum in corpore vili is a just rule where there is any reasonable
presumption of benefit to arise on a large scale.] 1865 C. M. Yonge Clever Woman I. iii. 80 The only
difficulty was to find poor people enough who would submit to serve as the corpus vile for their charitable
treatment. 1953 Essays in Criticism III. i. 4, I am not proposing to include among these initial corpora vilia
passages from either Mr Eliot's criticism or Dr Leavis's.

Definitions of corpus from OED 2

Page 4
1
Corpus Linguistics
1.1
What is a Corpus?
We shall discuss what a corpus is by looking at how the word is used, in particular by linguists. What kind of an object is a corpus, and what is it likely
to be useful for?
We learn the sense of a newly-encountered word in different ways. Young children experimentally combine or mutate words to see which uses meet with
approval; older ones do the same in the process of defining peer groups based on a shared exotic vocabulary. In both cases, meaning is exemplified or
confirmed by repeated, socially sanctioned, usage. One of the objectives of traditional linguistics was to overcome this requirement of exposure to
''language in use'' an impractical option for those wishing to learn a new language in a short time, or to understand a language no longer spoken anywhere
by defining powerful general principles which would enable one to derive the sense of any newly-encountered word simply by applying etymological or
morphological rules. Knowles (1996), arguing that linguistic theory is above all a matter of organizing linguistic knowledge in this way, points for example
to the success with which such models have been used in training generations of schoolchildren to understand Latin or Greek unseens.
While only experience can tell us what a word "is understood to mean", such analytic methods tell us what a word "ought to mean". A modern dictionary
combines the strengths of both methods, by organizing evidence of usage into an analytic framework of senses.
What, then, does the word 'corpus' actually mean? We might do worse than consider the five distinct senses listed in the second edition of the Oxford
English Dictionary as a starting point (see figure on preceding page). Of these, two particularly refer to language. The first is that of "A body or
collection of writings or the like; the whole body of literature on any subject". Thus we may speak of the 'Shakespearean corpus', meaning the entire
collection of texts by Shakespeare. The second is that of "the body of written or spoken material upon which a linguistic analysis is based". This is the
sense of the word from which the phrase 'corpus linguistics' derives, and in which we use it throughout this book. The two senses can, of course,
overlap as when, for example, the entire collection of a particular author's work is subjected to linguistic analysis. But a key distinction remains. In the
words of John Sinclair, the linguist's corpus is "a collection of pieces of language, selected and ordered according to explicit linguistic criteria in order to
be used as a sample of the language" (Sinclair 1996). It is an object designed for the purpose of linguistic analysis, rather than an object defined by
accidents of authorship or history.

page_3

Page 5

As such, corpora can be contrasted with archives or collections whose components are unlikely to have been assembled with such goals in mind (see
further Atkins et al 1992). Given this emphasis on intended function, the composition of a corpus will depend on the scope of the investigation. It may
be chosen to characterize a particular historical state or a particular variety of a particular language, or it may be selected to enable comparison of a
number of historical states, varieties or languages. Varieties may be selected on geographical (for example, British, American, or Indian English),
sociological (for example, by gender, social class, or age group), or generic bases (for example written vs. spoken; legal or medical; technical or
popular; private or public correspondence). Generally the texts to be included in a corpus are defined according to criteria which are external to the
texts themselves, relating to the situation of their production or reception rather than any intrinsic property they may have. Discovery of such intrinsic
properties (if any) may, indeed, be the purpose of the exercise.
Corpora stored and processed by computer, once the exception, are now the norm. It is worth noting however that there is very little in the practice of
corpus linguistics which could not equally well be done in principle by non-automatic means. However, in general, corpora are understood to be
computer-processable corpora.
The British National Corpus (BNC) consists of a sample collection which aims to represent the universe of contemporary British English. Insofar as it
attempts to capture the full range of varieties of language use, it is a balanced corpus rather than a register-specific or dialect-specific one; it is also a
mixed corpus, containing both written texts and spoken ones transcriptions of naturally-occurring speech.
1.2
What Can You Get Out of a Corpus?
A corpus can enable grammarians, lexicographers, and other interested parties to provide better descriptions of a language by embodying a view of it
which is beyond any one individual's experience. The authoritative Comprehensive Grammar of the English Language (Quirk et al 1985) was
derived in part from evidence provided by one of the first modern English corpora, the Survey of English Usage. Svartvik and Quirk (1980: 9) observe
that: "Since native speakers include lawyers, journalists, gynaecologists, school teachers, engineers, and a host of other specialists, it follows: (a) that no
individual can be expected to have an adequate command of the whole 'repertoire': who, for example, could equally well draft a legal statute and
broadcast a commentary on a football game? (b) that no grammarian can describe adequately the grammatical and stylistic

page_5

Page 6
properties of the whole repertoire from his own unsupplemented resources: 'introspection' as the sole guiding star is clearly ruled out."
A corpus which is designed to sample the entire 'repertoire' offers a tool for the description of properties with which even the grammarian may not be
personally familiar. Corpus-based descriptions have produced a few surprises, sometimes contradicting the received wisdom. Sampson (1996)
describes how he became a corpus linguist as a result of his experience with theories of recursive 'central embedding' in sentences such as 'the mouse
the cat the dog chased caught squeaked', where component clauses nest within each other like Russian dolls. Most discussions of this phenomenon had
used linguistic intuition to analyze entirely imaginary sentences, claiming that such constructions were in some sense 'unnatural', though syntactically
feasible. However, when Sampson turned to look at corpus data, he found that such centrally embedded structures were actually far from rare, and used
in ways which appeared entirely 'natural'. While it does not eliminate linguistic intuition in classifying and evaluating instances, the use of corpora can
remove much of the need to invent imaginary data, and can provide relatively objective evidence of frequency.
The utility of a corpus to the lexicographer is even more striking: careful study of a very large quantity and wide range of texts is required to capture and
exemplify anything like all the half-million or more words used in contemporary British English. It is no coincidence that dictionary publishers have played
major roles in setting up the two largest current corpora of British English: the Bank of English (HarperCollins) and the BNC (Oxford University Press,
Longman, Chambers); or that, in the increasingly competitive market for English language learners' dictionaries, four new editions published in 1995 (the
Collins Cobuild Dictionary, the Cambridge Dictionary of International English, the Longman Dictionary of Contemporary English, the Oxford
Advanced Learner's Dictionary) should all have made the fact of their being 'corpus-based' a selling point.
Linguists have always made use of collections of textual data to produce grammars and dictionaries, but these have traditionally been analyzed in a
relatively ad hoc manner, on the basis of individual salience, with a consequent tendency to privilege rare and striking phenomena at the expense of
mundane or very high frequency items. Corpora, in particular computer-processable corpora, have instead allowed linguists to adopt a principle of 'total
accountability', retrieving all the occurrences of a particular word or structure in the corpus for inspection, or (where this would be infeasible) randomly
selected samples. This generally involves the use of specialized software to search for occurrences (or co-occurrences) of specified strings or patterns
within the corpus. Other

page_6

Page 7

software may be used to calculate frequencies, or statistics derived from them, for example to produce word lists ordered by frequency of occurrence
or to identify co-occurrences which are significantly more (or less) frequent than chance.
Concordances are listings of the occurrences of a particular feature or combination of features in a corpus. Each occurrence found, or hit, is displayed
with a certain amount of context the text immediately preceding and following it. The most commonly used concordance type (known as KWIC for
'Key Word In Context') shows one hit per line of the screen or printout, with the principal search feature, or focus, highlighted in the centre.
Concordances also generally give a reference for each hit, showing which source text in the corpus it is taken from and the line or sentence number. It is
then up to the user to inspect and interpret the output. The amount of text visible in a KWIC display is generally enough to make some sense of the hit,
though for some purposes, such as the interpretation of pronominal reference, a larger context may have to be specified. Most concordancing software
allows hits to be formatted, sorted, edited, saved and printed in a variety of manners.
Frequency counts are implicit in concordancing, since finding all the occurrences of a particular feature in the corpus makes counting the hits a trivial
task. Software will generally allow numbers to be calculated without actually displaying the relevant concordance an important feature where thousands
or even millions of occurrences are involved. Frequency counts can be elaborated statistically, in many cases automatically by the concordancing
software, but should be interpreted with care (see further 2.2.4 on page 40).
Concordances and frequency counts can provide a wide variety of linguistic information. We list some of the kinds of questions which may be asked,
relating to lexis, morphosyntax, and semantics or pragmatics.
A corpus can be analyzed to provide the following kinds of lexical information:
How often does a particular word-form, or group of forms (such as the various forms of the verb 'start': 'start', 'starts', 'starting', 'started') appear in the
corpus? Is 'start' more or less common than 'begin'? The relative frequency of any word-form can be expressed as a z-score, that is, as the number of
standard deviations from the mean frequency of word-forms in the corpus. The number of occurrences of a word-form in the entire BNC ranges from
over 6 million for the most frequent word, 'the', to 1 for 'aaarrrrrrrrggggggghhhhhh' or 'about-to-be-murdered'. The mean frequency is approximately
150, but the standard deviation of the mean is very high (over 11,000), indicating that there are very many words with frequencies far removed from the
mean.

page_7

Page 8

With what meanings is a particular word-form, or group of forms, used? Is 'back' more frequently used with reference to a part of the body or a
direction? Do we 'start' and 'begin' the same sorts of things?
How often does a particular word-form, or group of forms, appear near to other particular word forms, which collocate with it within a given distance?
Does 'immemorial' always have 'time' as a collocate? Is it more common for 'prices' to 'rise' or to 'increase'? Do different senses of the same word have
different collocates?
How often does a particular word-form, or group of forms, appear in particular grammatical structures, which colligate with it? Is it more common to
'start to do something', or to 'start doing it'? Do different senses of the same word have different colligates?
How often does a particular word-form, or group of forms, appear in a certain semantic environment, showing a tendency to have positive or negative
connotations? Does the intensifier 'totally' always modify verbs and adjectives with a negative meaning, such as 'fail' and 'ridiculous'?
How often does a particular word-form, or group of forms, appear in a particular type of text, or in a particular type of speaker or author's language?
Is 'little' or 'small' more common in conversation? Do women say 'sort of' more than men? Does the word 'wicked' always have positive connotations for
the young? Is the word 'predecease' found outside legal texts and obituaries? Do lower-class speakers use more (or different) expletives?
Whereabouts in texts does a particular word form, or group of forms, tend to occur? Does its meaning vary according to its position? How often does it
occur within notes or headings, following a pause, near the end of a text, or at the beginning of a sentence, paragraph or utterance? And is it in fact true
that 'and' never begins a sentence?
A corpus can also be analyzed to provide the following kinds of morphosyntactic information:
How frequent is a particular morphological form or grammatical structure? How much more common are clauses with active than with passive main
verbs? What proportion of passive forms have the agent specified in a following 'by' phrase?
With what meanings is a particular structure used? Is there a difference between 'I hope that' and 'I hope to'?
How often does a particular structure occur with particular collocates or colligates? Is 'if I was you' or 'if I were you' more common?

page_8

Page 9

How often does a particular structure appear in a particular type of text, or in a particular type of speaker or author's language? Are passives more
common in scientific texts? Is the subjunctive used less by younger speakers?
Whereabouts in texts does a particular structure tend to occur? Do writers and speakers tend to switch from the past tense to the 'historic present' at
particular points in narratives?
And, finally, a corpus can be analyzed to provide semantic or pragmatic information. Rather than examining the meanings and uses of particular forms,
we can use it to identify the forms associated with particular meanings and uses:
What tools are most frequently referred to in texts talking about gardening?
What fields of metaphor are employed in economic discourse?
Do the upper-middle classes talk differently about universities from the working classes?
How do speakers close conversations, or open lectures? How do chairpersons switch from one point to another in meetings?
Are pauses in conversation more common between utterances than within them?
What happens when conversationalists stop laughing?
Not all of these types of information are equally easy to obtain. In using concordancing software, specific strings of characters have to be searched for.
In order to disambiguate homographs or to identify particular uses of words or structures, it may be necessary to inspect the lines in the output,
classifying them individually. Thus while it is relatively easy to calculate the frequency of a word-form and of its collocates, it may be more difficult to
calculate its frequency of use as a particular part of speech, with a particular sense, or in a particular position or particular kind of text.
To help in such tasks, computer corpora are increasingly marked up with a detailed encoding which encompasses both external characteristics of each
text and its production, and internal characteristics such as its formal structure. Such information will typically include details of what kind of text it is and
where it comes from, details relating to the structure of the text and the status of particular components division into chapters, paragraphs, spoken
utterances, headings, notes, references, editorial comments etc., as well as any linguistic annotation, indicating for instance the part-of-speech value or
the root form of each word. Such encoding permits the user to search for strings or patterns in

page_9

Page 10

particular kinds, parts or positions of texts, or with particular types of linguistic annotation.
It can be equally difficult to find instances of particular syntactic, semantic or pragmatic categories unless these happen to have clear lexical correlates, or
the corpus markup clearly distinguishes them. For instance, the markup of the BNC might be used to find occurrences of highlighting (typically through
italics or underlining in the original), to investigate headings and captions, to generate a list of the publishers responsible for the texts in the corpus, or to
identify those texts published by specific publishers.
While the examples just cited have all concerned analyses within a particular corpus, it is evident that all these areas can also be examined contrastively,
comparing data from corpora of different languages, historical periods, dialects or geographical varieties, modes (spoken or written), or registers. By
comparing one of the standard corpora collected twenty years ago with an analogous corpus of today, it is possible to investigate recent changes in
English. By comparing corpora collected in different parts of the world, it is possible to investigate differences between, for instance, British and
Australian English. By comparing a corpus of translated texts with one of texts originally created in the target language, it is possible to identify linguistic
properties peculiar to translation. By comparing a small homogeneous corpus of some particular kind of material with a large balanced corpus (such as
the BNC), it is possible to identify the distinctive linguistic characteristics of the former.
1.3
How Have Corpora Been Used?
This section describes a few major corpora which have previously been created and discusses some of the work done with them, to illustrate current
concerns in the field.
1.3.1
What Kinds of Corpora Exist?
We begin by listing some of the main corpora developed for English in the past, grouped according to the main areas of language use they sample. For a
fuller annotated list, see Edwards (1993) or Wichmann et al (in press).
geographical varieties
The earliest corpus in electronic form, compiled at Brown University in 1964, contained 1 million words of written American English published in 1961
(Kucera and Francis 1967). The Brown corpus has since been widely imitated, with similarly-designed corpora being compiled for British (the
Lancaster-Oslo-Bergen corpus or LOB: Johansson 1980), Indian (the Kolhapur Corpus of Indian English: Shastri 1988), Australian (the Macquarie
Corpus of Australian English: Collins and Peters 1988) and New Zealand varieties (the Wellington corpus:

page_10

Page 11

Bauer 1993). The International Corpus of English project (ICE) is currently creating a corpus with similarly-designed components representing each of
the major international varieties of contemporary English (Greenbaum 1992).
spoken language corpora
The earliest computer corpora, such as Brown and LOB, were collections of written data. A number of corpora consisting of transcripts of spoken
English have since been developed. These vary enormously both in the types of speech they include and in the form and detail of transcription employed
(see 1.4.2 on page 26). The best-known is probably the London-Lund Corpus, a computerised version of just under half-a-million words of the Survey
of English Usage conversational data (Svartvik 1990; Svartvik and Quirk 1980), which has been widely used in comparisons with the LOB corpus of
written English. The Corpus of Spoken American English under development at Santa Barbara (Chafe et al 1991) is collecting a similar quantity of
American conversational data.
mixed corpora
The major large mixed corpus to precede the BNC was the Birmingham collection of English texts, developed at the University of Birmingham with the
dictionary publishers Collins during the 1980s as a basis for the production of dictionaries and grammars (see e.g. Sinclair 1987). This originally
contained 7.5 million words, growing eventually to nearly 20 million, of which approximately 1.3 million were transcripts of speech. The collection has
continued to grow since, having now been incorporated into the 300 million word Bank of English (see 1.4.1 on page 21).
historical varieties
The most extensive corpus of historical English is probably the Helsinki corpus of English texts: diachronic and dialectal (Kytö 1993). The corpus has
three parts, corresponding with three historical periods (Old, Middle, and Early Modern English); within each period, there are samples of different
dialects, permitting not only diachronic comparisons but also synchronic comparisons of different geographical varieties.
child and learner varieties
A number of corpora have been compiled relating to particular categories of language users, in particular children who are acquiring English as their first
language, and foreign learners of English. They are sometimes termed special corpora (Sinclair 1996), because they document uses of language which
are seen as deviant with respect to a general norm. Instances include the Polytechnic of Wales corpus of child language (O'Donoghue 1991), and the
International Corpus of Learner English (ICLE) being created at Louvain (Granger 1993).

page_11

Page 12

genre- and topic-specific corpora

Other corpora have been designed to include only samples of language of a particular type, for example dealing with a particular topic, or belonging to a
particular genre or register. There are many examples, ranging from psycholinguistically motivated experiments such as the HCRC map task corpus
(Anderson et al 1991), consisting of 128 transcribed performances of map-reading tasks, to corpora created for other purposes, such as the Hong
Kong corpus of computer science texts, designed to support analysis of technical vocabulary (Davison 1992). In the USA, the Linguistic Data
Consortium has produced a large number of corpora of specific genres of speech and writing on CD-ROM, ranging from telephone conversations to
stock-exchange reports.
multilingual corpora
Monolingual corpora of languages other than English are not mentioned here for reasons of space, but a number of multilingual corpora containing texts
in both English and one or more other languages have been developed. Some are fairly heterogeneous collections, while others are carefully constructed
ensembles of texts selected on the basis of similar criteria in each language. In the former category, the European Corpus Initiative (ECI) has produced
a multilingual corpus of over 98 million words, covering most of the major European languages, as well as Turkish, Japanese, Russian, Chinese, Malay
and more (Armstrong-Warwick et al 1994). In the latter category, an EU-funded project called PAROLE is currently building directly comparable
corpora for each major European language.
1.3.2
Some Application Areas
The range of corpus-based descriptive work is well documented by Altenberg's bibliographies of corpus linguistics (Altenberg 1990, 1995), and is also
covered in a number of introductory textbooks on the field. Recent examples include Sinclair (1991), Stubbs (1996) and McEnery and Wilson (1996);
Leech and Fligelstone (1992) and Biber et al (1996) provide accessible short introductions. In this section, we review a handful of studies in order to
illustrate some of the areas in which corpus-based work has been carried out, and to raise some of the key methodological issues. No claim to
completeness of coverage is intended, as the field is both very varied and rapidly expanding. For up-to-date information, and for a wider (more corpus-
like!) perspective, the reader could do a lot worse than to search the World Wide Web for pages on which the phrase 'corpus linguistics' appears.
Corpus-based research naturally grounds its theorizing in empirical observation rather than in appeals to linguistic intuition or expert knowledge. It thus

page_12

Page 13

emphatically rejects one of the major tenets of Chomskian linguistics, namely that the linguist's introspection provides the only appropriate basis for
describing language, insofar as "information about the speaker-hearer's competence ... is neither presented for direct observation nor extractable from
data by inductive procedures of any known sort" (Chomsky 1965: 18). Corpus users have taken varying positions on these issues, ranging from the
'weak' view, that sees corpus data as complementing the 'armchair' linguist's intuitive insights by providing real-life examples and a reliable testbed for
hypotheses (see 8.1 on page 143), to the 'strong' view, according to which corpus data should always override intuition, and discussion should be
confined solely to naturally-occurring examples. In either case, corpus-based work has wider affinities than many other branches of linguistics, since the
study of language-in-use has something to offer historical, political, literary, sociological, or cultural studies, and has profited from the resulting synergy.
Our discussion focuses on four application areas: the emergence of collocation as a key component in linguistic description; the opportunities afforded
by corpus-based methods for contrastive studies of different languages, varieties and registers; the use of corpora in natural language processing
(NLP); and finally, their use in foreign language teaching.
1.3.3
Collocation
One of the forefathers of contemporary corpus linguistics, J.R. Firth, observed that part of the meaning of the word 'ass' consists in its habitual
collocation with an immediately preceding 'you silly' (Firth 1957: 11). (Whether this use is still current some fifty years later is a question the BNC can
answer: there are in fact only 8 occurrences of 'silly ass' in the corpus, none of them preceded by 'you'.) There are a great many cases in English where
the occurrence of one word predicts the occurrence of another, either following or preceding it. Kjellmer (1991) notes such examples as 'billy', which
predicts 'goat' or 'can' following, and 'bail', which predicts 'jump' or 'stand' preceding. Such collocational patterns tend to be highlighted by KWIC
concordances, since these show just the few words which precede and follow the keyword or focus, and can typically be sorted according to these
words. It is also relatively easy to calculate the frequency with which a particular collocate appears within a certain range of the focus its collocation
frequency within a given span and to compare such frequencies to find the most common collocates occurring, say, up to two words before 'ass'.
Jones and Sinclair (1974) claim that the probabilities of lexical items occurring in English are generally affected by collocational norms within a span of
up to four words. Co-occurrence of two or more words within a short

page_13

Page 14

space can be important insofar as that co-occurrence is expected and typical (whether in the language in general, in a particular text-type, or in the style
of a particular speaker or author), or insofar as it is unexpected and atypical. Sinclair (1991) argues that recurrent collocational patterns effectively
distinguish different senses of the same word a 'silly ass', while potentially a quadruped, is statistically a biped and that consequently collocational
frequencies can be used to disambiguate word senses. In this he builds on Firth's view that for the lexicographer, ''each set of grouped collocations may
suggest an arbitrary definition of the word, compound or phrase which is being studied'' (Firth 1957: 196). From a converse perspective, deviation from
a collocational norm 'since breakfasts immemorial', say can be a means of generating particular effects, such as irony (Louw 1993).
The tendency for one word to occur with another has both grammatical and semantic implications. The collocation of a word with a particular
grammatical class of words has been termed colligation. For instance, unlike 'look at', the verb 'regard' appears always to colligate with adverbs of
manner, as in 'She regarded him suspiciously' (Bolinger 1976). From a semantic perspective, the habitual collocations of some words mean that they
they tend to assume the positive or negative connotations of their typical environments a particular semantic prosody. For example, Sinclair (1991)
notes that the verb 'set in' has a negative prosody, because things which typically set in are 'rot', 'decline' etc., making it extremely difficult to use this verb
with positive implications. In the same way, the typical collocations of many apparently neutral terms may reveal deep-seated cultural prejudices: Stubbs
(1996: 186ff.) notes how the high-frequency collocates of terms such as 'Welsh' or 'Irish' tend to reinforce nationalistic stereotypes.
Other than in set phrases, collocations and their frequencies are not generally accessible to intuition. They can however be easily identified and quantified
by computational methods, in corpora which are sufficiently large for the purpose. Work based on the Birmingham collection of English texts, revealing
the extent of collocational patterning in English, has contributed to change current views of psycholinguistic organization, by providing important
evidence that lexical items are to a large extent co-selected rather than combined individually, following what Sinclair terms an idiom principle rather
than an open-choice one. A collection of concordances showing the most frequent collocates of some 10,000 words in the Bank of English has
recently been published on CD-ROM (Cobuild 1995). Much discussion and research has also been dedicated to the development of appropriate
measures of the strength of collocational

page_14

Page 15

links (Dunning 1993; Stubbs 1995), and to the automatic listing of significant collocations.
1.3.4
Contrastive Studies
The construction of the LOB corpus of British English, on closely parallel lines to the Brown corpus of American English, and their subsequent
morphosyntactic annotation (see 1.4.2 on page 24), stimulated a variety of comparative studies, facilitated by the wide distribution of both corpora on a
single CD-ROM by ICAME, a highly influential organization of European corpus linguists based at the University of Bergen in Norway. This section
reviews some examples of contrastive studies, involving both different corpora and different components of a single corpus, with the purpose of
illustrating some of the methodological issues involved.
Comparing Geographical Varieties and Languages
Hofland and Johansson (1982) and Johansson and Hofland (1989) report detailed studies of word frequencies in the Brown and LOB corpora,
showing, for instance, that 49 of the 50 most frequent words in each corpus are the same. Contrasts concern not only such areas as spelling (e.g.
'colour' vs. 'color'), and different choices of synonyms (e.g. 'transport' vs. 'transportation', 'film' vs. 'movie'), but also different subject matter (e.g. 'tea'
vs. 'coffee', 'London' vs. 'Chicago'). Leech and Falton (1992) suggest that some of these differences in frequency may indicate cultural, rather than
simply linguistic differences. Noting, for instance, the considerably more frequent use in Brown of military terms, such as 'armed', 'army', 'enemy',
'forces', 'missile(s)', 'warfare', they suggest that this may reflect a greater concern in the US with military matters (remembering that 1961, the year of the
Brown texts, was also that of the Cuban missile crisis). And faced with the greater frequency in LOB of conditional and concessive conjunctions ('if',
'but', 'although', 'though') and words denoting possibility or uncertainty ('possible', 'perhaps', 'unlikely' etc.), they speculate that this may conform to the
stereotype of the "wishy-washy Briton who lacks firmness and decisiveness" (Leech and Falton 1992: 44).
In reaching these (tentative) conclusions, they note that a relatively small number of words can be analyzed in this way. The LOB and Brown corpora
each contain only 50,000 word types (less than the number of headwords in a single-volume dictionary), and among the less frequent words, relative
frequency or infrequency may be due to sampling bias. Even at higher frequency levels, differences may be the product of a skewed distribution across
texts. While the influence of these sources of error can be reduced by comparing groups of words identified by semantic or other criteria, careful
examination of concordances remains necessary to check how often the word or words in question are in fact

page_15

Page 16

used with a particular sense not all occurrences of the word 'film' refer to cinema, and 'tea' is a meal in some parts of Britain, as well as a drink.
Multilingual, Comparable and Parallel Corpora
There is an increasing tendency to apply corpus techniques to the task of comparing different languages. Where a corpus consists of texts selected using
similar criteria in two or more languages, comparisons can be made at many different levels, ranging from lexicogrammatical preferences to rhetorical
organization. One particularly interesting type of multilingual corpus is the parallel corpus, consisting of texts that are actually translations of each other:
prototypical instances are official documents produced in multilingual environments, such as the UN and EU, or the Canadian Hansard, which is
published in both English and French. Such corpora have clear utility for the study of translation itself, as well as providing a useful focus for contrastive
studies of the differences between particular languages.
To facilitate comparison, the texts in parallel corpora may be aligned, identifying equivalences on a sentence-by-sentence, phrase-by-phrase, or word-
by-word basis, and much effort has gone into the development of software to align parallel texts automatically. The major fields of application have so
far been in developing and testing machine translation packages and producing computerized translation aids (such as bilingual dictionaries and
terminology databanks), but such corpora also have much to teach about the universals and specifics of language, and the process of translation. For
instance, the English-Norwegian parallel corpus project (Johansson and Hofland 1993; Johansson and Ebeling 1996; Aijmer et al 1996) lists among its
fields of investigation not only the similarities and differences in the lexicogrammatical, rhetorical, and information structure of texts in the two languages,
but also such questions as:
To what extent are there parallel differences in text genres across languages?
In what respects do translated texts differ from comparable original texts in the same language?
Are there any features in common among translated texts in different languages (and, if so, what are they)?
Comparing Diachronic Varieties
English-language corpus-building dates back at least thirty years. The continued availability of these pioneering corpora has made possible a range of
contrastive studies investigating changes in the English language over time. There is also a growing interest in the construction of specifically designed
diachronic corpora which sample language production over much longer time periods.

page_16

Page 17

Examples of the first kind include a 1991 version of LOB, using identical sampling criteria as far as possible, recently completed at Freiburg (Mair
1993). As well as facilitating diachronic comparison of particular linguistic features, such corpora may also provide a useful yardstick for comparing
studies based on the larger corpora of recent years with ones based on the smaller corpora which preceded them.
Notable examples of the second kind include the Archer corpus (Biber, Finegan and Atkinson 1994), which contains samples of eleven different
registers from different historical periods, and the Helsinki corpus of English texts: diachronic and dialectal (Kytö 1993: see 1.3.1 on page 10). Kytö et
al (1995) provides a useful checklist of new projects in this expanding field.
Change over time can also be investigated by contrasting the usage of different age groups, as further discussed in the next section.
Comparing Categories of Users
Where corpora provide information as to the social and linguistic provenance of speakers and hearers, or of writers and readers, they can be used to
compare the language of different groups according to such variables as region, age, sex, and social class, provided that a demographically balanced
sample of language users has been taken. For instance, Stenström (1991) used the London-Lund corpus to study the relationship between gender and
the use of expletives relating to religion, sex and the body in speech. She found that while female speakers tended to use such words more often than
male, they tended to choose expletives from a 'heaven' group, while male speakers used ones relating to 'hell' and 'sex'. Investigating the function of
these words in the light of their position in the utterance, she also found that the male speakers used them to emphasize their own contributions, whereas
female speakers used them to give responses and invitations to continue.
Comparing Different Uses of Language
The construction of the London-Lund corpus of spoken English spawned a large number of studies comparing speech with writing, generally using the
LOB corpus as evidence of the latter. These have highlighted differences in the relative frequencies of words and structures in the two modes (in speech
the most common word is 'I', while in writing it is 'the'), as well as facilitating the identification and description of features whose use appears to be
specific to the spoken language, most notably discourse-structuring elements such as 'well', 'I mean', and 'you know'.
Corpora have also been extensively used to investigate the ways in which genres differ linguistically, attempting to characterize genres by the relative
frequency of particular features. Conversely, insofar as texts can be categorized statistically according to linguistic features, what correspondence is there
between those categorizations and the lists of text-types employed, for instance, in

page_17

Page 18

corpus design? Biber (1988) compared the frequencies of a range of linguistic features which can be automatically counted (word length, type/token
ratio, nouns, prepositions, first/second/third person pronouns, tense, voice, aspect, 'wh-' relative clauses, synthetic vs. analytic negation, 'private' vs.
'public' verbs, etc.) in samples from texts of different types in the LOB and London-Lund corpora, using cluster and factor analysis to identify eight basic
classes of text, grouped by similar scores on five dimensions of variation. Biber and Finegan (1994) employ similar methods to investigate variation
within texts, showing that in medical research articles, frequencies of a number of lexicogrammatical phenomena vary according to the section of the
article sampled.
1.3.5
NLP Applications
There has recently been an increased awareness of the potential which corpus methods offer for tackling a number of problems in the field of natural
language processing (NLP), that is, the development of automatic or semi-automatic systems for analyzing, 'understanding', and producing natural
language.
Corpora are increasingly used in the development of NLP tools for applications such as spell-checking and grammar-checking, speech recognition, text-
to-speech and speech-to-text synthesis, automatic abstraction and indexing, information retrieval, and machine translation. A major problem to be faced
by all NLP systems is that of resolving ambiguity, be this selecting which of two or more possible orthographic transcriptions might match a given
acoustic input ('whales', 'Wales', or 'wails'?), or deciding whether an instance of the word 'bank' refers to a financial institution or a landscape feature,
and hence how it should be translated into, say, French, or how the text that contains it should be classified for retrieval purposes.
The limited results achieved in such areas using traditional rule-based models of language have led to an increasing interest in probabilistic models,
where probabilities are calculated on the basis of frequencies in corpus data (Church and Mercer 1993). Traditional spell-checkers, for instance, are
based only on a dictionary of possible orthographic forms in the language, so that they fail to recognize errors which are nonetheless acceptable forms
(such as 'form' for 'from'). Performance in such cases can be improved by considering the probability that the form typed by the user will occur after the
previous word, where this probability has been calculated by analyzing a corpus for the language concerned. (For instance, it is highly unlikely that the
word between 'the' and 'typed' in the previous sentence could be 'from'.)
The analysis of frequencies of particular features in corpora underlies a wide variety of NLP applications based on probabilistic techniques, such as:

page_18

Page 19

categorization of specific texts, for instance by identifying their type, semantic field, and keywords as a basis for automatic indexing and abstracting, or
extracting terminology;
refinement of question-answering and information retrieval systems, enabling them to employ or suggest additional or alternative search terms to
interrogate textual databases, on the basis of collocational regularities in corpus data, and to filter retrieved information by checking its conformity to the
typical collocational patterns of the search terms proposed;
improvement of multilingual retrieval of texts, and identification of terminological equivalents in different languages, on the basis of lexical and
collocational equivalences identified in parallel multilingual corpora (see 1.3.4 on page 15).
A further use of corpora in NLP is as testbeds to evaluate applications, be these theoretically motivated or probabilistic. Probabilistic models of
language can, to a certain extent, be self-organizing, and in this respect corpora can provide training instruments for software which learns probability
through experience, or refines an initial model in a bootstrapping process (Atwell 1996). A system which needs to disambiguate the term 'bank', for
instance, can analyze a corpus to learn that the landscape sense generally collocates with 'river', 'flower', etc., while the financial one collocates with
'merchant' and 'high street'. Such uses typically call for substantial annotation of corpora in order to reduce ambiguity in the training materials, and NLP
applications in many cases overlap with applications designed to annotate corpora in various ways (see 1.4.2 on page 24).
1.3.6
Language Teaching
The growing variety of corpus applications in the field of English language teaching is reviewed by Murison-Bowie (1996). Corpora have already had a
considerable influence in the creation of new dictionaries and grammars for learners, where the use of corpus data has allowed:
more accurate selection of words and senses for inclusion, based on frequency of occurrence;
introduction of information concerning the relative frequency of each word and of the different senses of each, and their use in different genres and
registers;
citation of actual rather than invented examples, selected to illustrate typical uses and collocations.
Sinclair (1987) provides a detailed discussion of these issues in reference to the creation of the Collins Cobuild Dictionary.

page_19

Page 20

Kennedy (1992) reviews the long tradition of pre-electronic corpus work in language teaching. Many of the studies he discusses aimed to identify the
most frequent words and grammatical structures in the language, with a view to optimizing the design of syllabuses and the grading of materials. Such
goals have received new impetus from the availability of electronic corpora. Analysis of the Birmingham collection of English texts underlay the selection
of the 'lexical syllabus' proposed by Willis (1990); Grabowski and Mindt (1995) used the Brown and LOB corpora to create a list of irregular verbs
ordered according to frequency, arguing that by following this order in syllabus design, teaching should achieve maximum yield for the student's effort,
irrespective of when the learning process is broken off.
Corpus data have also provided a means of evaluating conventional syllabuses. Ljung (1991) compares the lexis of textbooks of English as a foreign
language with that of a corpus of non-technical writing, while Mindt (1996) compares the treatment of future time reference in textbooks and learner
reference grammars with corpus data. Such studies use corpora to highlight actual frequency of occurrence, which, while not the only criterion for
deciding syllabus content or the form of materials (Widdowson 1991), can clearly provide teachers and textbook writers with an important tool to
assess the pedagogic suitability and adequacy of particular choices (Biber et al 1994).
There is also a growing interest in providing teachers and learners with direct access to corpora as resources for classroom or individual work.
Fligelstone (1993) suggests that learners can use corpora to find out about the language for themselves and hence to question prescriptive specifications,
for instance by exploring the nature of idioms and collocations, rhetorical questions, the use of sentence-initial 'and', etc. Similarly, Aston (in press)
argues that with appropriate training, advanced learners can use large corpora as reference tools which overcome many of the limitations of existing
dictionaries and grammars by providing a much larger number of more contextualized examples. Corpora may not only be a source of information about
the language in question: Fligelstone notes that they can also provide encyclopedic knowledge, making them a useful tool to gather ideas about a subject
in order to write or talk about it, while Aston (1995) suggests that concordancing software enables learners to browse the corpus texts in a serendipitous
process where they not only analyze language but experience it as communicative use. In such ways the growing availability of corpora offers learners a
new kind of resource which can complement the traditional dependency on teacher, textbook, and reference book.

page_20

Page 21

1.4
How Should a Corpus be Constructed?
We noted above that a corpus is not a random collection of text. In this section, we review some of the major issues relating to corpus construction. We
discuss first some basic design principles, concerning size, sampling practice, and composition, and then consider the various kinds of encoding,
annotation, and transcription policies which may be adopted.
1.4.1
Corpus Design
In designing a corpus to address a particular purpose, two groups of criteria must be considered. On the one hand the size of the corpus and of its
component parts, and on the other the material actually selected for inclusion, may each have crucial effects on its usability.
Corpus Size and Sample Size
The frequency of different word forms in a corpus generally follows a Zipfian distribution (Zipf 1935), whereby the second most frequent word
occurs approximately half as often as the most frequent one, the third most frequent word approximately one third as often, and so on. All but the most
frequent words are extremely rare. Corpora therefore need to be very large and heterogeneous if they are to document as wide as possible a range of
uses of as many linguistic features as possible.
Even where they are relatively frequent, features which are unevenly distributed across different types of text in the corpus may not be adequately
represented. Sinclair (1991: 24) notes of the Brown and LOB corpora that they only provide reliable sources of information concerning relatively
frequent words that occur in a wide range of texts. They are much less reliable for words which occur only in certain text-types, because "the sub-
categories necessary to balance the sample are not in themselves reasonable samples because they are too brief". Increasing corpus size can go some
way to solving these problems by providing larger samples for each sub-category.
While an increase in size provides more data, it also tends to entail less detailed analyses: it is striking how many descriptive studies have analyzed only
small corpora (or small samples of larger ones), often because of the need to inspect and categorize data manually. The analysis of larger corpora is
heavily dependent on the use of automatic or semi-automatic procedures able to identify particular linguistic phenomena (see 1.4.2 on page 24). The
availability of such procedures is still limited in many areas.
Sinclair (1991: 24ff.) has argued that the static sample corpus consisting of a fixed collection of data should ideally give way to the monitor corpus
where information could be gleaned from a continuous stream of new text "as it passes through a set of filters which will be designed to reflect the
concerns of researchers." In the lexicographical field, for instance, procedures might be

page_21

Page 22

designed to capture new word forms or usages, and shifts in frequency of use. Such a corpus would allow the user to detect phenomena which would
be inadequately represented in even a very large sample corpus, and to monitor changes in the language as they took place. The Bank of English project
at the University of Birmingham puts this idea into practice (Sinclair 1992). At the time of writing, this corpus contains over 300 million words, but is
continually expanding and being monitored by a set of software tools which categorize incoming data automatically for particular purposes.
In order to include both a wide range of text-types and a large number of different texts of each type, early corpora included relatively brief extracts
from each text sampled: the one million-word Brown and LOB corpora each consisted of randomly-selected 2000-word samples from 500 texts, of 15
different types. A corpus composed of short samples of equal length is less likely to give skewed results due to the influence of particular source texts,
but is of little use for the study of large textual features such as narrative organization, or of within-text variation (Biber and Finegan 1994). To permit
the study of such phenomena, Sinclair (1996) has argued that large corpora should be composed of whole texts wherever possible.
The continued growth in the size of corpora has generally implied an increase in sample sizes as well as in the number of samples. However, the
inclusion of complete texts may not always be possible, either for copyright reasons, or because the notion of 'completeness' is inappropriate or
problematic (Is a newspaper a complete text? Is each story in a newspaper a complete text?). Complete texts may also vary greatly in size, giving rise to
problems of balance.
Corpus Composition
A corpus which claims to 'characterize the state of a language' must define both the linguistic universe which has been sampled and the sampling
procedures followed. Is it intended to characterize only the speech and writing of competent native-speakers? If so, how are the latter to be defined? Is
it to include as wide a variety of different types of language as possible? Should its composition reflect the relative frequencies with which these different
types occur? And if so, should these frequencies be calculated on the basis of reception (the language people hear and read), production (the language
people speak and write), or both? In the first case priority will be given to those text-types which are most widely and frequently experienced, such as
casual conversation, everyday workplace and service encounters, television, radio and the popular press. In the second case, while much of the
everyday dialogue content may be similar, the rest of the corpus may look very different, since most texts are produced for small audiences.

page_22

Page 23

Many of the criteria for the composition of a corpus are determined by its intended uses. The Survey of English Usage aimed to describe the
grammatical repertoire of adult educated speakers of English with the aid of a corpus which was "reasonably representative of the repertoire of
educated professional men and women in their activities, public and private, at work and at leisure, writing and speaking" (Quirk 1974: 167). Given
these goals, it was designed to sample a wide range of text types, deciding the proportions of each largely on their assumed frequency of production in
the language as a whole. Most large mixed corpora have tended to follow the Survey in aiming to cover users' production repertoires (though not merely
the educated, professional and adult) by drawing up a list of text types to be included, deciding the proportions of the corpus to be constituted by each,
and then selecting texts for each type using a combination of random and controlled sampling techniques.
There are, of course, many different ways of characterizing texts, and hence many text typologies. One of the more fundamental distinctions is between
spoken and written materials. The Survey of English Usage contained equal quantities of each, and subsequent compilers of mixed corpora have
generally agreed that ideally this proportion should be respected though there is no particular reason for thinking that writing and speech are equally
present in either production or reception of the language as a whole. However, since it is for the moment much more expensive to obtain speech data,
which has to be recorded and transcribed, than written texts, many of which are already available in machine-readable form, large mixed corpora
generally contain much smaller proportions of speech.
More complex text typologies have been based on such concepts as the field (that is, the topic and purpose) of the text; on sociolinguistic factors
determining its tenor (for example, the context in which the text is produced or received, the participants and their inter-relationships, etc.); or on its
mode (that is, whether speech is monologue or dialogue, face-to-face or broadcast, prepared or spontaneous; or whether writing is published or
unpublished). Atkins et al (1992) list as many as 29 parameters to be considered in constructing a 'balanced' corpus. Within each category, reception
criteria may be used to complement production ones, for instance by preferring bestsellers to remaindered novelettes when sampling published fiction.
There is ample evidence that word frequencies and other linguistic features vary widely within different text-types, both with respect to each other, and
with respect to the whole of a corpus, whatever typology is employed. For instance, in most forms of speech, 'sure' is more common than 'certain',
while in written social science texts the opposite is the case (Biber and Finegan 1989).

page_23

Page 24

Biber (1993) demonstrates that analyses based on restricted samples cannot be generalized to language as a whole. It has consequently been argued
that a balanced corpus is useful in a specific application only to the extent that it includes an adequate sample of the category in question, which can be
separated out and treated as a corpus in its own right. However, balanced corpora can at least provide a baseline against which variation amongst pre-
defined categories can be measured. Halliday (1992: 69) argues ''if we recognize departure from a norm, then there has to be a norm to depart from. If
we characterize register variation as variation in probabilities, as I think we must, it seems more realistic to measure it against observed global
probabilities than against some arbitrary norm such as the assumption of equiprobability in every case.''
1.4.2
Encoding, Annotation, and Transcription
Simple lexical analysis of a corpus of written texts requires only a computer-processable version of the text. However, for the full range of analytic
possibilities sketched out in section 1.2 on page 5 above, some thought must also be given to the ways in which the text and its context are to be
encoded, that is, the way in which particular features of them can be made explicit and hence processable. This is especially important for corpora
derived from spoken language, where the process of transcription immediately confronts the analyst with many difficult theoretical and methodological
issues.
Encoding and Annotation
A corpus may simply consist of sequences of orthographic words and punctuation, sometime known as plain text. However, texts are not just
sequences of words; they have many other features worthy of attention and analysis. At the very least, we want to distinguish and describe the different
texts in the corpus, as well as their different components. Such text descriptions may include bibliographic details, details of the nature of the sample
(complete or otherwise), or classification in terms of the parameters employed in designing the corpus.
When describing the components of written texts (other than words), it is useful to indicate the boundaries of chapters, sections, paragraphs, sentences,
etc., and the specialized roles of headings, lists, notes, citations, captions, references, etc. For spoken texts, indications of the beginnings and ends of
individual utterances are essential, as is an indication of the speaker of each. It may also be desirable to encode paralinguistic phenomena such as
pausing and overlap, and non-verbal activity such as laughter or applause. For either kind of text, it may be helpful to include editorial information about
the status of the electronic text itself, for example to mark corrections or conjectures by the transcriber or editor.

page_24
 Page 25

A further type of information which may be provided is linguistic annotation of almost any kind, attached to components at any level from the whole
text to individual words or morphemes. At its simplest, such annotation allows the analyst to distinguish between orthographically similar sequences (for
example, whether the word 'Frank' at the beginning of a sentence is a proper name or an adjective), and to group orthographically dissimilar ones (such
as the negatives 'not' and '-n't'). More complex annotation may aim to capture one or more syntactic or morphological analyses, or to represent such
matters as the thematic or discourse structure of a text.
Types of linguistic annotation that have been employed with corpora include the following:
part-of-speech or word-class
Placing a tag alongside each word in the corpus to indicate its word-class can disambiguate different grammatical uses of a word such as 'works', which
may be a plural noun or a singular verb. Such part-of-speech tagging can also help identify grammatical patterns. While many existing corpora have
been manually or automatically tagged in this way, no standard set of part-of-speech tags has yet been defined.
lemmatization and morphological analysis
Lemmatization involves the use of tags to indicate the relationship of each word-form to its root (e.g. that 'took' is the past tense form of 'take'). It
enables derived and inflected forms of a word to be retrieved and counted along with its root. While lemmatization provides a useful way of grouping
data for many descriptive purposes, particularly in languages with many declensions and conjugations, corpus-based research suggests that different
forms of lemmas do not always share the same meaning, but tend to occur in distinctive contexts (Sinclair 1991).
word-sense
Tags can also be used to distinguish different senses of the same word (e.g. 'table' meaning 'piece of furniture' as opposed to 'data in rows and
columns'), on the basis of an existing dictionary or thesaurus. While word-sense annotation is quite extensively used in the fields of machine translation
and information retrieval (Guthrie et al 1994), it clearly prejudges the issue for corpora whose goals include lexicographic description.
syntactic role
The parsing of corpora involves the addition of annotation to indicate the grammatical function of each word, describing the structure of each sentence
as a set of labelled bracketings or tree. A number of small parsed corpora have been derived from pre-existing larger corpora. These include the
Gothenberg corpus(Ellegård 1978) derived from parts of Brown; the LOB corpus treebank (Leech and Garside 1991)

page_25

Page 26

derived from LOB; and the Susanne corpus (Sampson 1994), which combines results from both. While most parsing has been done by hand,
considerable research effort has gone into the development of automatic parsers. Notable examples include the English Constraint Grammar Parser
developed at Helsinki (Karlsson 1994) and the TOSCA system developed at Nijmegen (van Halteren and Oostdijk 1993). As syntactic analysis is often
necessary to decide who did what to whom in English, parsed corpora have an important role in many NLP applications. Parsing schemes are, however,
highly theory-dependent, and there is relatively little consensus in the field.
pragmatic annotation
In a sentence like 'Not there they won't', pronouns, deixis, and ellipsis refer to concepts which are (probably) more fully expressed elsewhere in the text.
Identifying those concepts is often important for natural language understanding systems and for machine translation. Substantial work has been carried
out on procedures to insert pragmatic annotation linking such anaphoric features to their antecedents (Fligelstone 1992; Garside 1993). Annotation
indicating other pragmatic features such as the function of individual speech acts and overall discourse structure (Coulthard 1994) has been so far limited
to very small samples, given the lack of consensus as to relevant units and categories, and of explicit algorithms for their identification.
Much of the concern with annotated corpora derives from the need to provide training materials and testbeds for software which will annotate text
automatically for various NLP applications (see 1.3.5 on page 18). Some descriptive linguists have on the other hand argued that corpora should only be
made available as plain text, given that annotation always involves introducing an interpretation of some kind (Sinclair 1991).
Transcribing Speech
No transcript, however detailed, is able to provide an the information on which hearers draw when understanding speech (Cook 1995). French (1992)
proposes four levels of transcription, ranging from a broad orthographic representation of the words spoken, to a narrow phonetic transcription with
detailed prosodic information. The act of transcribing spoken data is in itself a kind of encoding, making explicit an interpretation of the original sound
wave. Even a broad transcription generally implies, for instance, distinguishing different speakers' utterances, dividing these into sentences or prosodic
units, and dividing the latter into words, with a disambiguation of homophonous elements with distinct orthographic forms. Narrower transcription
involves additional encoding of prosodic features, such as stress and intonation, as

page_26

Page 27

well as of paralinguistic ones such as changes in voice quality, pausing, and non-vocal events (phone rings, applause, shifts in position and eye-contact,
etc.).
For studies of phonology and dialectology, an orthographic transcription is clearly unlikely to be adequate unless supported by more detailed phonemic
or phonetic data. Such coding systems are not discussed here. Their complexity, requiring highly-specialized expertise, has meant they are only currently
available for relatively small corpora of speech. Given its costs, there has generally been a trade-off between size and detail in corpus transcription.
Edwards and Lampert (1993) provide a detailed survey of different transcription systems, both in terms of the features encoded and their representation
(for a summary, see Edwards 1995).

page_27

Page 28

2
The British National Corpus
2.1
How the BNC was Constructed
Here we review the status of the British National Corpus with respect to the corpus construction issues discussed in the previous section. The discussion
follows the same order as that used above.
2.1.1
Corpus Design
Corpus Size and Sample Size
Looking back over the history of computer corporal, we can see that corpus sizes have increased by roughly one order of magnitude per decade. It has
been estimated that the 100 million word BNC would take 4 years to read aloud, at 8 hours a day. The Associated Press newswire, by comparison,
generates some 50 million words per year. The overall size of the BNC corresponds to roughly 10 years of linguistic experience of the average speaker
in terms of quantity though not, of course, in quality, given that it aims to sample the language as a whole, rather than that experienced by any particular
type of speaker.
Most samples in the BNC are of between 40,000 and 50,000 words; published texts are rarely complete. There is, however, considerable variation in
size, caused by the exigencies of sampling and availability. In particular, most spoken demographic texts, which consist of casual conversations, are
rather longer, since they were formed by grouping together all the speech recorded by a single informant. Conversely, several texts containing samples of
written unpublished materials such as school essays or office memoranda are very short.
Corpus Composition
The BNC was designed to characterize the state of contemporary British English in its various social and generic uses. A more detailed discussion of the
design criteria and their implementation is provided in chapters two and three of the BNC Users' Reference Guide (Burnard 1995). In selecting texts
for inclusion in the corpus, account was taken of both production, by sampling a wide variety of distinct types of material, and reception, by selecting
instances of those types which have a wide distribution. Thus, having chosen to sample such things as popular novels, or technical writing, best-seller
lists and library circulation statistics were consulted to select particular examples of them.
Although the BNC distinguishes several different geographical, sociological, and generic varieties, it does not necessarily provide a reliable sample for
any particular set of such criteria. Different considerations applied to the procedures used when choosing material for inclusion in the written and spoken
parts of the corpus, which are therefore discussed separately.

page_28

Page 29

Written Texts
Ninety per cent of the BNC is made up of written texts, chosen according to three selection features: domain (subject field), time (within certain
dates) and medium (book, periodical, unpublished, etc.). In this way, it was hoped to maximize variety in the language styles represented, both so that
the corpus could be regarded as a microcosm of current British English in its entirety, and so that different styles might be compared and contrasted.
Each selection feature was divided into classes and target percentages were set for each class. Thus for the selection feature 'medium', five classes
(books, periodicals, miscellaneous unpublished, miscellaneous unpublished, and written-to-be spoken) were identified. Samples were then selected in
the following proportions: 60 per cent from books, 30 per cent from periodicals, 10 per cent from the remaining three miscellaneous sources. Similarly,
for the selection feature 'domain', 75 per cent of the samples were drawn from texts classed as 'informative', and 25 per cent from texts classed as
'imaginative'. The following list illustrates each selection criterion, and indicates the actual numbers of texts and words in each category words being
counted according to the criteria described in 2.1.2 on page 34.
Domain
The evidence from catalogues of books and periodicals suggests that imaginative texts account for less than 25 per cent of published output.
Correspondence, reference works, unpublished reports, etc. add further to the bulk of informative text which is produced and consumed. Nevertheless,
the overall distribution between informative and imaginative text samples in the BNC was set to reflect the influential cultural role of literature and
creative writing. The target percentages for the eight informative domains were arrived at by consensus within the project, based loosely upon the pattern
of book publishing in the UK during the past 20 years or so.
texts percentage words percentage
Imaginative 625 19.47 19664309 21.91
Arts 259 8.07 7253846 8.08
Belief and thought 146 4.54 3053672 3.40
Commerce and finance 284 8.85 7118321 7.93
Leisure 374 11.65 9990080 11.13
Natural and pure science 144 4.48 3752659 4.18
Applied science 364 11.34 7369290 8.21
Social science 510 15.89 13290441 14.80
World affairs 453 14.11 16507399 18.39
Unclassified 50 1.55 1740527 1.93

page_29

Page 30

Time
Informative texts were selected only from 1975 onwards, imaginative ones from 1960, reflecting their longer 'shelf-life', though most (75 per cent) of the
latter were published no earlier than 1975.
texts percentage words percentage
19601974 53 1.65 2036939 2.26
19751993 2596 80.89 80077473 89.23
Unclassified 560 17.45 7626132 8.49

Medium
This categorization is broad, since a detailed taxonomy or feature classification of text medium would have led to such a proliferation of subcategories as
to make it impossible to represent them all adequately. The labels used were intended to be comprehensive in the sense that any text can be assigned
with reasonable confidence to these macro categories. 'Miscellaneous published' includes brochures, leaflets, manuals, advertisements. 'Miscellaneous
unpublished' includes letters, memos, reports, minutes, essays. 'Written-to-be-spoken' includes scripted television material, play scripts etc.
texts percentage words percentage
Book 1488 46.36 52574506 58.58
Periodical 1167 36.36 27897931 31.08
Misc. published 181 5.64 3936637 4.38
Misc. unpublished 245 7.63 3595620 4.00
To-be-spoken 49 1.52 1370870 1.52
Unclassified 79 2.46 364980 0.40

Written texts are further classified in the corpus according to sets of descriptive features (see 5.2.3 on page 103 for a complete list). This information
was recorded to allow more delicate contrastive analysis of particular sets of texts. These descriptive features were monitored during the course of data
collection, and, in cases where a free choice of texts was available, text selection took account of the relative balance of these features. For example,
although no relative proportions were pre-defined for different target age groups, it was

page_30

Page 31

possible to ensure that the corpus contained texts intended for children as well as texts intended for adults.
Spoken Texts
Ten percent of the BNC is made up of transcribed spoken material, totalling about 10 million words. Roughly equal quantities were collected in each of
two different ways:
a demographic component of informal encounters recorded by a socially-stratified sample of respondents, selected by age group, sex, social class and
geographic region;
a context-governed component of more formal encounters (meetings, debates, lectures, seminars, radio programmes and the like), categorized by
topic and type of interaction.
This dual approach was chosen in the absence of any obvious objective measures that might be used to define the target population or to construct a
sampling frame for spoken language. Demographic sampling techniques alone would have resulted in the omission from the corpus of many types of
spoken text produced only rarely in comparison with the total output of all speech producers. Examples include broadcast interviews, lectures, legal
proceedings, and other texts produced in situations where broadly speaking there are few producers and many receivers. Reliance on a previously
drawn-up list of spoken text types alone would have been very difficult to justify, given the lack of agreement on ways of categorizing speech and the
impossibility of determining the relative proportions of each type..
The following classifications apply to both the demographic and context-governed components:
Region where text captured
texts percentage words percentage
South 296 32.34 4728472 45.61
Midlands 208 22.73 2418278 23.33
North 334 36.50 2636312 25.43
Unclassified 77 8.41 582402 5.61

Interaction type
texts percentage words percentage
Monologue 218 23.82 1932225 18.64
Dialogue 672 73.44 7760753 74.87
Unclassified 25 2.73 672486 6.48

page_31

Page 32

Crowdy (1995) describes in more detail the procedures adopted both for sampling and for transcribing the spoken part of the BNC. For the
demographic component, random location sampling procedures were used to recruit 124 adults (aged over 15) from across the United Kingdom, with
approximately equal numbers of both sexes and from each of five age groups and four social classes. Each recruit used a portable tape recorder to
record their own speech and the speech of people they conversed with over a period of up to a week. Additional recordings were gathered for the
BNC as part of the University of Bergen COLT Teenager Language Project (Stenström and Breivik 1993). This project used the same recording
methods and transcription scheme, but selected only respondents aged 16 or below.
As with any sampling method, some compromise had to be made between what was theoretically desirable and what was feasible within the constraints
of the BNC project. There is no doubt that recruiting 1000 people would have given greater statistical validity but the practical difficulties and cost
implications made this impossible. Nevertheless, the total number of participants in all conversations was well in excess of a thousand, producing a total
of 4.2 million words of unscripted conversational English.
The context-governed component consists of 762 texts (6.1 million words). As in the written component, the range of text types was selected
according to previously-defined criteria, based in the first place on domain:
texts percentage words percentage
Educational and informative 144 18.89 1265318 20.56
Business 136 17.84 1321844 21.47
Institutional 241 31.62 1345694 21.86
Leisure 187 24.54 1459419 23.71
Unclassified 54 7.08 761973 12.38

Each of these categories was divided into the subcategories monologue (40 per cent) and dialogue (60 per cent), and within each category a range of
contexts defined as follows:
educational and informative
Lectures, talks and educational demonstrations; news commentaries; classroom interaction etc.
business
Company and trades union talks or interviews; business meetings; sales demonstrations etc.

page_32

Page 33

institutional
Political speeches; sermons; local and national governmental proceedings etc.
leisure
Sports commentaries; broadcast chat shows and phone-ins; club meeting and speeches etc.
The overall aim was to achieve a balanced selection within each category, taking into account such features as region, level, gender of speakers, and
topic. Since the length of these text types varies considerably news commentaries may be only a few minutes long, while some business meetings and
parliamentary proceedings may last for hours an upper limit of 10,000 words per text was generally imposed.
2.1.2
Encoding, Annotation, and Transcription
The encoding of the BNC is designed to capture an extensive variety of information. It includes the various design features described in the previous
section, bibliographic details, and a great deal of detail about the structure of each text, that is, its division into sections or chapters, paragraphs, verse
lines, headings, etc. for written text, or into speaker turns, conversations, etc. for spoken texts.
In the corpus, such information must be represented in a manner which distinguishes it from the words of the plain text. In the BNC this is achieved
through the use of special purpose markup, following a scheme known as the Corpus Data Interchange Format (CDIF), itself strongly influenced by
the Text Encoding Initiative's Guidelines for the encoding of electronic text (Sperberg-McQueen and Burnard 1994). The purpose of CDIF is to allow
the portability of corpora across different types of hardware and software environments, and the comparability of different corpora, as well as to make
it easy to search for occurrences of particular encoded features, or for linguistic phenomena which occur in particular contexts.
Contextual information common to all texts is described in an initial corpus header. Contextual information specific to a given text is listed in a text
header which precedes each text. Detailed structural and descriptive information is marked at appropriate positions within each text. The BNC Users'
Reference Guide includes a full description of the markup conventions employed, from which the following brief description is taken.
CDIF uses an international standard known as SGML (ISO 8879: Standard Generalized Mark Up Language), now very widely used in the electronic
publishing and information retrieval communities. In SGML, electronic texts are regarded as consisting of named elements, which may bear descriptive
attributes and can be combined according to a simple grammar, known as a document type

page_33

Page 34

definition (Goldfarb 1990). In an SGML document, element occurrences are delimited by the use of tags. There are two forms of tag, a start-tag,
marking the beginning of an element, and an end-tag marking its end. Tags are delimited by the characters < and >, and contain the name of the
element, preceded by a solidus (/) in the case of an end-tag. For example, a heading or title in a written text will be preceded by a tag of the form
<HEAD> and followed by a tag in the form <HEAD>. Everything between these two tags is regarded as the content of an element of type < HEAD>.
Instances of elements may also be assigned particular attributes. If present, these are indicated within the start-tag, and take the form of an attribute
name, an equal sign and the attribute value. For example, the <HEAD> element may take an attribute TYPE which categorizes it in some way. A main
heading will thus appear with a start-tag <HEAD TYPE=MAIN>, and a subheading with a start-tag <HEAD TYPE=SUB>. Attribute values are used for a variety of
purposes in the BNC, notably to represent the part of speech codes allocated to particular words by the CLAWS tagging scheme, described below.
End-tags are omitted for the elements <S>, <W> and <C> (i.e., for sentences, words, and punctuation). For all other non-empty elements, every
occurrence in the corpus has both a start-tag and an end-tag. In addition, attribute names are omitted for the elements <W> and <C> to save space. A
list of the elements employed in the BNC is provided in the BNC Users' Reference Guide.
A restricted range of characters is used in element content: specifically, the upper- and lower-case alphabetics, digits, and a subset of the common
punctuation marks. All other characters are represented by SGML entity references, which take the form of an ampersand followed by a mnemonic for
the character, and terminated by a semicolon where this is necessary to resolve ambiguity. For example, the pound sign is represented by the string
&pound;, the character é by the string &eacute; and so forth. The mnemonics used are taken from standard entity sets, and are also listed in the
BNC Users' Reference Guide.
Since the publication of the BNC, its encoding scheme, or derivations from it, have been widely adopted by other corpus building projects. It also forms
the basis of the 'Corpus Encoding Standard' (CES) recommended by the European Union's Expert Advisory Group on Language Engineering
Standards (EAGLES).
Part-of-Speech Annotation
Leech et al (1994) describe how the 100 million words of the BNC were automatically tagged, using the CLAWS4 system developed at Lancaster
University, originally by Roger Garside, with additional

page_34

Page 35

software developed by Michael Bryant. CLAWS was first used for the tagging of the LOB Corpus (Garside et al 1987) using an annotation scheme
described in Johansson et al (1986). The CLAWS system automatically attaches a tag to each word indicating its grammatical class, or part of speech
(POS), and to each punctuation mark. The aim is to provide distinct codings for all classes of words having distinct grammatical behaviour. A set of 58
POS codes, known as the C5 tagset, is used in the first release of the BNC; these are listed in section 2.1 on page 230.
CLAWS generally assumes that an orthographic word (separated by spaces from the adjacent words) is the appropriate unit for tagging. The following
particular cases should however be noted:
A single orthographic word may contain more than one grammatical, or L-word: thus in the case of contractions, such as 'she's', 'they'll', 'we'd', 'don't',
'won't', 'd'ya', 'gotta', ' 'twas', and of possessives ('John's', 'pupils'' etc.), separate tags are assigned to each grammatical component of the orthographic
word. A list of cases where orthographic words are treated as multiple L-words is given in the BNC Users' Reference Guide.
The opposite circumstance is also quite common, where two or more orthographic words behave as a single grammatical word: for example, compound
prepositions such as 'instead of' and 'up to' are assigned a single preposition tag. Foreign phrases such as 'hoi polloi' or 'viva voce I are also tagged as
single items. Again, a list of such multi-word L-words is given in the BNC Users' Reference Guide.
Otherwise, spaces determine treatment as single or multiple L-words. Note that compounds, where orthographic practice is often uncertain (for example
'fox holes', 'fox-holes', and 'foxholes', all of which appear in the BNC), may occur tagged as two words (if they are separated by spaces), or as one (if
they appear as a single word, or with a hyphen). Truncated words in spoken data are also tagged as separate L-words which are unclassified.
For approximately 4.7% of the L-words in the corpus, CLAWS was unable to decide between two possible POS codes. In such cases, a two-value
POS code known as a portmanteau is applied. For example, the portmanteau code VVD-VVN means that the word may be either a past tense verb
(VVD), or a past participle (VVN).
The automatic tagging system had an overall error-rate of approximately 1.7%, excluding punctuation marks. With such a large corpus, there was no
opportunity to undertake post-editing to correct annotation errors. However, in a successor project, due to complete in 1997, the Lancaster team has
been

page_35

Page 36

refining the tagging of the whole corpus, using as input a new set of data derived from a 2% sample of the BNC which was manually checked and
corrected.
Speech Transcription
The spoken texts of the BNC are transcribed at a level roughly corresponding to French's (1992) level 2 (see 1.4.2 on page 26). It is an orthographic
transcription, rather than a phonological or prosodic one, in which changes of speaker, pausing, overlap, and a variety of non-verbal and non-vocal
events identified by the transcribers are made explicit by means of SGML markup.
The basic unit is the utterance, marked as a <U> element, with an attribute WHO whose value specifies the speaker, where this is known. Overlapping
speech is marked using a system of pointer elements, explained in 9.2.4 on page 172. Pausing is marked using a <PAUSE> element, with an indication
of its length if this seems abnormal. Gaps in the transcription, caused either by inaudibility or the need to anonymize the material, are marked using the
<UNCLEAR> or <GAP> elements as appropriate. Truncated forms of words, caused by interruption or false starts, are also marked, using the
<TRUNC> element.
A semi-rigorous form of normalization is applied to the spelling of non-conventional forms such as 'innit' or 'lorra'; the principle adopted was to spell
such forms in the way that they typically appear in general dictionaries. Similar methods are used to normalize such features of spoken language as filled
pauses, semi-lexicalized items such as 'um', 'err', etc. Some light punctuation was also added, motivated chiefly by the desire to make the transcriptions
comprehensible to a reader, by marking (for example) questions, possessives, and sentence boundaries in the conventional way.
Paralinguistic features affecting particular stretches of speech, such as shouting or laughing, are marked using the <SHIFT> element to delimit changes
in voice quality. Non-verbal sounds such as coughing or yawning, and non-speech events such as traffic noise are also marked, using the <VOCAL>
and <EVENT> elements respectively; in both cases, the values supplied for the DESC attribute specifies the phenomenon concerned. It should,
however, be emphasized that the aim was to transcribe as clearly and economically as possible rather than to represent all the subtleties of the audio
recording.
2.2
Using the BNC: Some Caveats
Although the BNC was designed to represent the full variety of the English language at a particular point in time, users of the corpus should be cautious
in generalizing from the evidence it provides about the frequency of particular linguistic phenomena in the language. The very all-inclusiveness of the
BNC means that it necessarily contains instances of untypical, irregular, accidental, and

page_36

Page 37

possibly erroneous phenomena. Defining 'erroneous' in the context of corpus linguistics is not methodologically neutral, for no corpus is error free, and
indeed to decide finally that some aspects of a corpus are erroneous may be a non-trivial operation.
We list some features of the BNC which may mislead the unwary, relating to
the nature of the materials included in the corpus;
the sampling, encoding and annotation procedures adopted;
the division of the corpus data into separate documents, and the division of those documents into text headers and the texts themselves.
Many of these difficulties are inherent in the materials making up the BNC. Others are attributable to errors or inconsistencies in the process of
sampling, encoding, and annotating the corpus. Corrigible errors of these kinds will be corrected, as resources permit, in later versions of the BNC.
2.2.1
Source Materials
As with any other large corpus, the texts themselves contain a wide variety of potentially deceptive features. These include:
orthographic error and variation
Texts were not systematically proof-read or corrected prior to inclusion in the corpus, with the result that any errors in the original will normally be
reproduced in the BNC. In particular, printing errors in the original may lead to incorrect division of the text into words, as when a dash is printed as a
hyphen, or a full stop without a following space is interpreted as an abbreviation rather than a sentence boundary. Written English also accepts a wide
range of spelling variants, as with '-ise' and '- ize' forms. This is particularly problematic where compounds are concerned, since the number of words
may also vary: for instance, the forms 'busybody', 'busy-body' and 'busy body' are all present in the BNC.
features of spoken performance
Spontaneous speech not only involves a range of conventionally 'non-verbal' elements whose orthographic transcription is relatively unstandardized
('uhuh', 'haha', 'heheh' and the like), but also what are (by written standards) lexical and grammatical errors. False starts, changes of mind etc. may be
cut short or subsequently corrected by their producers, with the effect that the transcript retains both the original and its repetition or reformulation.
variation in transcription practice
Every effort was made to achieve accuracy and consistency in transcriptions, but unquestionably some errors

page_37

Page 38

remain. It should not be assumed, for example, that one transcriber's 'erm' or 'going to' is necessarily phonetically distinct from another's 'um' or 'gonna'.
non-standard usage
Although a check was made that all identifiable authors represented in the corpus were native speakers of British English, no such check was (or could
have been) applied to the authors of unattributed material, or to all speakers in the spoken part of the corpus. In fact there are many non-standard
usages, not all of them produced by non-native speakers or writers. It should also be borne in mind that both speech and writing contain wilfully deviant
forms, as in puns or poetry. The demographically sampled part of the corpus contains a number of jokey discussions of the likely use of the tape
recordings. In the written part of the corpus, extracts from electronic mail discussion lists, and personal letters, contain equally informal and unorthodox
usages.
quotation and allusion
Apparent inconsistencies may also be due to the presence of quotations or allusions, or of explicit references to linguistic features. Quotations in
languages other than English are also occasionally to be found, which may lead to confusion where they include forms which are identical to English
words for example, a fragment in German may contain many occurrences of the word 'die' but have nothing to do with mortality.
homographs
Foreign words are only one category of unexpected homographs in the corpus: others include names, abbreviations, and acronyms, as well as misprints.
The wide range of material included in the BNC means that almost any possible variant is likely to appear. Analyses of word frequency should pay
particular attention to homographic forms: for example, any attempt to repeat Holmes' (1994) study of changes in gender-related usages with the BNC
data, would need to take as much care as she does to separate out occurrences of 'Miss' as a term of address from the verbal and nominal usages, and
to distinguish 'Ms' from the disease 'MS' and the abbreviation 'ms' before attempting to tabulate their relative frequencies.
2.2.2
Sampling, Encoding, and Tagging Errors
The BNC is probably more richly encoded and annotated than any other corpus of comparable size. Despite the best endeavours of all involved, the
need to complete the project on time and within a tight budget meant that errors and inconsistencies remain in the corpus. Some typical errors are listed:
sampling errors
Some texts, or parts of texts, appear in the corpus more than once. This particularly applies to newspaper materials, and is

page_38

Page 39

a consequence of the way such texts were prepared and selected for inclusion in the corpus.
encoding errors
Not all structural features of the texts are consistently encoded. For instance, not all quotations are marked as SGML <Q> elements, many being simply
implied by the appearance of inverted commas; list items are sometimes tagged as <P> (paragraph) elements; not all headings are correctly identified,
and so forth. In general, items which are tagged at all are tagged correctly, but inferences about the frequency distribution of structural elements in the
BNC should be made only with great caution.
tagging errors
There are part-of-speech tagging errors and inconsistencies. Although the part-of-speech tagging has high overall accuracy, errors in part-of-speech
assignment to words which have more than one possible POS value can still be frequent. Furthermore, the parameters of the tagging system were
modified during the tagging process, in the light of experience. This means that in the first release of the corpus some instances of the same sequence of
words have received different taggings. Similar inconsistencies may be observed in the case of phrases which are treated as a single L-word, or of
orthographic words which are treated as two or more L-words: for example, the phrase 'innit' (for 'isn't it') appears in the first release of the corpus both
as a single lexical item, with a single POS code, and (correctly) as a phrase, with three.
It is hoped to correct many of the errors listed in these three categories in later versions of the BNC.
2.2.3
What is a BNC Document?
The overall organization of the BNC entails the following problems:
BNC documents
None of the 4124 BNC documents making up the corpus should properly be regarded as a complete written or spoken text. Either for reasons of
length or, more frequently, for reasons of copyright, only samples of original written texts are included, of size between 40 and 50 thousand words.
Care should therefore be exercised when interpreting such features as co-occurrence and position within a document. The sampling method (beginning,
middle, or end) is included along with other text classification information provided by the <CATREF> element in the header of each text. Some BNC
documents, by contrast, contain more than one text in the everyday sense of the word. Documents taken from newspapers and periodicals, for example,
are likely to be composed of a considerable number of articles. For spoken demographic material,

page_39

Page 40

all the conversations recorded by a particular respondent are combined to form a single document. In these cases co-occurrence within the document
clearly does not imply co-occurrence within the same article or conversation. For spoken texts in general, the notion of completeness is rather hard to
define: it cannot be assumed that an entire speech event has been successfully recorded in every case. Even where an entire event was recorded, it may
have been only partially transcribed, whether for ethical or technical reasons.
Text headers
Each BNC document contains a header as well as the text itself. The header contains information about the text of a largely repetitive nature.
Consequently, searches which include headers in their scope may find a surprisingly high number of occurrences of words such as 'publication' or
'press'. The amount of information provided in headers varies considerably from text to text: for instance only a few texts have lists of keywords
included in the header.
2.2.4
Miscellaneous Problems
Further sources of potential misinterpretation have to do with the composition of the BNC. Some results may be biassed by the fact that the corpus was
collected at a particular time, with the result that certain 'buzzwords' occur more frequently than might otherwise have been the case. Others may be
influenced by atypically frequent recurrences in one or a few particular texts.
Although the BNC contains many different kinds of text, it should not be assumed that it contains every possible kind of material. There are almost as
many ways of classifying texts as there are text-classifiers; if therefore you are seeking texts in a category not specifically identified by the BNC's own
classification scheme, you may find it very hard to identify them, even if they are present in the corpus, or they may be present in only very small
quantities.
In general, it should be borne in mind that numbers of occurrences in a corpus are quite likely to be too small to be interpreted with confidence: half the
word-forms in the corpus occur only once. Nor should it be assumed, where higher frequencies are involved, that these are necessarily adequate for any
statistical procedure. In principle, corpora lend themselves to quantitative analysis: it is relatively easy to draw up frequency lists of words, collocations,
or grammatical patterns, and to compare these in particular categories of use or user. It may also seem obvious to use traditional statistical tests (such as
chi-square or t-test) to assess whether such frequencies, or the differences between them, are significant. However, the use of such tests involves some
questionable assumptions:

page_40

Page 41

The statistics traditionally used in frequency-based analyses assume that observations are independent of each other (i.e. that the probability of seeing a
particular event is constant such as the probability of seeing a head or a tail when tossing a coin). In texts, however, the likelihood of a particular word
occurring at any point is heavily constrained by neighbouring choices and by the nature of the text indeed probabilistic tagging techniques such as those
employed by the CLAWS program (see 2.1.2 on page 34) are based precisely on the failure of this assumption. This calls into question the use of
statistics based on this assumption to interpret language data. It is a separate decision, perhaps carried out after consideration of non-statistical data,
whether or not particular statistical values are 'linguistically interesting' (Stubbs 1995: 33).
Many statistical procedures assume that the events being analyzed are normally distributed, an assumption which can only be made for relatively
common events. The reliability of the assumption of normal distribution can be calculated using the formula np (1-p) >5, where n is the number of
events in the sample, and p the probability of the event in question occurring as the next event. As most words, and even more so phrases and
collocations, are extremely rare, the value of this formula is near enough to np, meaning that a reliable inference can only be made when the expected
frequency of an event is 5 or more. This condition is quite likely not to be satisfied even in the case of very large corpora, particularly where
occurrences in limited sections of the corpus are concerned, or specific collocations required.

page_41

Page 42

3
Future Corpora
It may not be too soon to ask how the widespread use of computer-held corpora has changed and is changing the study of language in its widest sense,
and to hazard some practical consequences. We have already commented on the way in which corpus usage encourages a probabilistic rather than a
rule-based approach to fundamental issues of language description. However the question of how these two approaches can best be integrated into a
consistent theoretical framework, as well as practically combined in order to maximize the advantages of each for NLP work, remains largely open.
Another widely-noted trend is the blurring of traditional linguistic categories as corpus-based analyses reveal complex patternings of language. For
instance, the categorical distinction between lexis and grammar is being increasingly questioned, as the extent of collocational patterning of lexis, and the
idiosyncracy of the grammar of individual words, are revealed (Sinclair 1991; Francis 1993). Understanding of such features will predictably increase
with growth in the number and the size of corpora available. The progressive accumulation of corpus evidence is also providing material that enables
linguists to chart changes in the language as they occur, and to understand more deeply the diachronic processes involved.
Natural language processing applications currently provide much of the momentum for new developments in corpus creation, with a trend towards ever-
larger corpora, towards greater standardization of sampling procedures and of encoding, and towards more sophisticated and reliable automatic
annotation procedures. As the number and the dimensions of available corpora increase, so both the need for more sophisticated automatic analyses,
and the feasibility of creating tools to produce them increase.
Advances in computer storage and processing capabilities suggest that it will soon be commonplace to include digital audio along with the transcriptions
which make up corpora of speech data. This will help overcome some of the limitations of current transcripts, though the process of automatically
aligning the two, or of generating the transcription from the audio, still poses many technical problems. There is an interesting synergy of applications
here with recent developments in the computer processing of ancient texts, where there is a need to align a digital image of a manuscript with a
'diplomatic' transcription of it. In both cases, the transcription is an essential indexing aid to the original.
Efforts towards standardization, going hand in hand with efforts to expand the availability of directly comparable corpora for many different languages,
are of particular importance in a European context, where there are nine official languages as well as a political commitment to support several other so-
called 'minority' languages. Corpus-based techniques are seen by many as central

page_42

Page 43

in the development of the European Language Industry, which is attracting significant levels of investment, both private and governmental.
One function served by large reference corpora such as the BNC may be supplied by other resources in the near future. It is already commonplace to
comment on the immense quantities of raw electronic text now available from the World Wide Web, and to speculate about ways in which it might be
organized as a linguistic corpus resource. It seems probable that as the sophistication of web indexing, search and retrieval systems increases, it will
become easier for individuals to create corpora of their own design, representing particular areas of language use by searching for relevant components
on the web.
Whatever form corpora of the future may take, it can be predicted that the range and quantity of corpus applications will increase in both academic and
everyday fields, with a variety of potential social consequences. Corpus-based analyses may come to influence areas such as public communication,
where speakers and writers may begin to rely on demographically-organized corpus evidence as a means of matching their language to a particular
target group or groups, or the interpretation of language in legal contexts, either as a means of demonstrating what is generally meant by a particular
expression, or of revealing stylistic idiosyncracies in questions of disputed authorship (Coulthard 1993). In language teaching increasing access to
corpora may modify the traditional role of the teacher as an authority about the use of the language to be learned, and reduce the sense of inferiority felt
by many non-native speaker teachers. More generally, there is much to be said about the way in which thinking about language, particularly the English
language, is politicized, and hence about the political implications of changing the basis on which assessments of correctness or appropriateness of
usage are made.

page_43

Page 45

II:
EXPLORING THE BNC WITH SARA

page_45

Page 46

Introduction
This part of the BNC Handbook contains a series of ten tasks, which introduce you to the main features of the SARA software and illustrate how you
can use it to obtain information from the BNC. The tasks investigate several issues which we noted in section 1.2 on page 5 as being important
application areas for corpus-based analysis. The topics covered range from the meanings and contexts of particular words and phrases to differences
between the language of writing and speech, or of speakers of different ages and sexes. Although these issues are largely linguistic, we have tried to
make the discussion as accessible as possible for the non-specialist reader and the tasks do not presuppose any particular competence in linguistics.
Each task has three parts:
an outline of the problem or issue to be analysed, and the relevant software features to be used;
step-by-step instructions for performing the analysis;
a summary of the main points to emerge, together with some suggestions for further practical work of a similar nature.
The tasks are designed as a sequence, progressively introducing different features of the corpus and of the software. You are recommended to work
through them in the order they are presented here. Reference information concerning specific features of the software can most easily be found in the
final part of this handbook, which provides a summary overview of the program. For more specific details concerning the BNC, you should consult the
BNC Users' Reference Guide.
The tasks are intended for teaching purposes only, and in no way claim to provide an exhaustive treatment of the linguistic problems discussed, which
would in all cases require considerably closer and more extensive analysis. Their primary aim is to familiarize you with the corpus and the software,
giving some idea of their potential as a means of investigating such problems. In particular, since SARA can take a long time to provide large sets of
results, we have tended to choose examples which involve rather small numbers. These numbers are generally far too small to permit reliable inferences
about the language as a whole.
We tested all the exercises described here using version 930 of the SARA system running against release 1.1 of the BNC. In some cases, our results
may be different from yours if you run the the same query against a different version of the corpus or of the software. In particular, the frequencies
reported for several words in version 1.0 of the BNC (the first released version) are somewhat

page_46

Page 47

lower than those for version 1.1. New versions of the corpus and software are announced as they become available at http:/ /info.ox.ac.uk/bnc on the
BNC website.
We have assumed that you have some acquaintance with the Microsoft Windows environment, and are familiar with its terminology. If you don't know
what a dialogue box is, or how to double-click your mouse, or how to re-size or iconify a window, you may find it helpful to have a quick look at any
introduction to the Microsoft Windows environment before beginning to work through what follows. The buttons displayed on the screen when using
SARA are reproduced and briefly explained on the inside cover of this handbook.

page_47

Page 48

1
Old Words and New Words
1.1
The Problem: Finding Evidence of Language Change
1.1.1
Neologism and Disuse
One of the most widespread uses of large corpora of contemporary language is to identify changes in vocabulary. Many recently-published dictionaries
of English have used corpora to hunt for neologisms, or for evidence that words or senses have fallen into disuse, in order to decide what words and
senses they should include. This task takes two examples, using SARA to look in the BNC for evidence of the use of one old word which such
dictionaries exclude, and of one new word which they include.
An Old Word: 'Cracksman'
In his turn-of-the-century stories about the gentleman-thief Raffles, E.W. Hornung refers to the latter as a 'cracksman'. Cited in the Shorter Oxford as a
slang term first found in the 19th century, 'cracksman' is absent from recent corpus-based dictionaries such as the Collins COBUILD, the Cambridge
International Dictionary of English, the Oxford Advanced Learner's Dictionary, and the Longman Dictionary of Contemporary English. This
first example examines whether the BNC provides evidence of contemporary use of the term 'cracksman' and its plural, 'cracksmen', with the meaning
of burglar.
A New Word: 'Whammy'
The word 'whammy' was introduced in the 1995 editions of all the dictionaries just cited. All define a 'whammy' as an unpleasant or difficult experience,
and note the phrase 'double whammy'. This second example examines the BNC for evidence of this use, and looks to see whether 'whammy' is also
found in other phrases or with other senses.
1.1.2
Highlighted Features
This task shows you:
how to start and leave SARA, and obtain on-line help;
how to look for a word in the corpus using the PHRASE QUERY option;
how to display solutions either on separate pages or as a one-per-line (KWIC) concordance display, using the CONCORDANCE option;
how to scroll through and select individual solutions;
how to display bibliographic details of the source of a selected solution and browse the text from which it comes, using the SOURCE and BROWSE options;
how to copy a selected solution to the Windows clipboard using the COPY option;
how to change the default display settings using the VIEW PREFERENCES option;

page_48

Page 49

how to switch between windows showing solutions from different queries using the WINDOW menu.
1.1.3
Before You Start
This task assumes that:
you are familiar with the basic conventions of operation under Microsoft Windows using a mouse, such as moving and sizing windows, starting, exiting
and switching between programmes, clipboard operation, and file management;
the SARA software has been correctly installed on your computer;
your computer has a TCP/IP connection to the machine running the server;
you know your SARA username and password, which you should obtain from your local network manager or SARA adviser.
1.2
Procedure
1.2.1
Starting Sara
The BNC icon should be visible on your desktop.

Double-click on the icon and wait for the program to load. The ABOUT SARA box will be displayed,
1 showing the version of the SARA software you are using.

2 Click on OK (or press ENTER). The LOGON dialogue box should be displayed, showing the server version
of the SARA software installed, and prompting you for your username and password. SARA works by
linking your computer to the one where the corpus is stored, which functions as a server. You use
your computer (the client) to formulate queries about the BNC and send them to the server, and to
manipulate the solutions to these queries returned by it. To start the program, a connection
between the client and server must be established. If this cannot be done, a 'Cannot connect to
server' message will be displayed. Clicking on OK will then display the COMMUNICATIONS dialogue
box, which allows you to RETRY the connection, if necessary changing the server specification, or
to CANCEL the display and exit from the program. You should not normally alter the settings in this
dialogue box without consulting your network manager.

1.2.2
Logging On

Type in your username, then press TAB and type in your password. Your password will not be displayed
3 as you type it.

4 Click on OK (or press ENTER). If you entered an incorrect username or password, the L OGON dialogue
box will be re-displayed. Otherwise a message

page_49

Page 50

box will confirm that you are logged on to the SARA server. It may also provide messages from the
network manager about the system you are using. If you make a mistake while entering your
username and password, clicking on the CANCEL button will clear the contents of the dialogue box .
If you repeatedly enter an incorrect username or password, a 'Cannot log on to server' message
will be displayed. Clicking on OK will clear this message and take you to the COMMUNICATIONS
dialogue box (see 1.2.1 on the preceding page).
You can change your password using the VIEW menu PREFERENCES option (see 1.2.8 on page 56) but
only after you have logqed on using your current password.
5 Click on OK (or press ENTER) to remove the message box. The main window will now be revised to
display the title SARA-bnc. A BNC icon will also appear in the bottom left-hand corner: you should resist
the temptation to click on this icon. The purpose of this icon is to show which corpora or subcorpora
are currently available for searching. In the current version of this software, only the full BNC
corpus is available. Clicking on the icon will simply open a view of the first text in the corpus;
closing this will close the corpus and SARA will have nothing to search!
6 If necessary, use the Windows buttons to enlarge the SARA-bnc window to full screen size.
7 Click on the VIEW menu, and check that TOOLBAR and STATUS BAR are both ticked.

You are now ready to formulate queries about the contents of the BNC. When you use SARA, you will usually be asking it to find examples of the
occurrences of particular words, phrases, patterns, etc. within the BNC. We refer to the request you make of the system as a query; the set of
examples or other response which this request produces we refer to as the solutions to the query. The distinction is important, because SARA allows
you to save and manipulate queries and their solutions independently.
1.2.3
Getting Help
At the bottom of the screen, on the status bar, is a reminder that you can get help at any point by pressing Fl. You can also do this from the H ELP menu,
or by clicking on the CONTEXT HELP button situated at the right-hand end of the toolbar. Clicking on this button creates a cursor which you can move to
any button or box and then click to display help concerning that object. The buttons on the toolbar and their names are shown on the inside cover of this
handbook
1.2.4
Quitting Sara
When you want to leave the program, select EXIT under the FILE menu (or press ALT+F4).

page_50

Page 51

1.2.5
Using Phrase Query to Find a Word: 'Cracksmen'
Let us begin by investigating which forms of the word 'cracksman' appear in the BNC. We will start by looking for occurrences of the plural form
'cracksmen'.

Move the mouse to the toolbar, and click on the PHRASE QUERY button. The Phrase Query dialogue box
8 will be displayed. You can also reach this dialogue box by selecting NEW QUERY then PHRASE from the
FILE menu.
9 Type in the string cracksmen, then click on OK (or press ENTER) to send the query to the server. The
options in the Phrase Query dialogue box are described in 1.3.3 on page 202. They allow you to
CANCEL the dialogue box and return to the previous window, losing any input you have entered; to
SEARCH HEADERS, looking for solutions to the query within both the texts and the text headers,
which contain information about the corpus and its component texts (for fuller details, see 5.1.1
on page 98); and to IGNORE CASE (the default setting), searching for solutions regardless of their use
of upper and lower case letters. If the Ignore case box is unchecked, SARA searches only for
solutions which match the case pattern used in the input string a considerably slower process (for
examples see 1.3.2 on page 61 and 10.2.4 on page 184).
10 Wait while the server searches for solutions and downloads them to the client. A new window called
Query 1 will open to display them. You may briefly see a red BUSY dot flashing on the status bar at
the bottom of the window as text begins to appear. The red dot means that your computer is
exchanging data with the server: it will stop flashing as soon as downloading is complete. You
should not attempt to give new commands while the red dot is flashing, but if you want to
abandon a search you can either press ESC before downloading begins, or simply close the query
window after downloading has commenced.

1.2.6
Viewing the Solutions: Display Modes
Enlarging the Context
There is exactly one occurrence of the word 'cracks-men' in the BNC, which you are currently looking at. However the amount of context displayed is
probably too small for you to understand how the word is being used.

Position the mouse on the solution and double-click the right mouse button. The amount of context will
11 be increased, up to the maximum download length (see 1.2.8 on page 55).

Page and Line Display Modes

Depending on the default settings being used to display solutions, you will see either:
a page display, showing this solution with a few lines of context, or

page_51
 Page 52

a line display, showing this solution on a single line, with a dashed surround.
In both display modes, the query focus (the word 'cracksmen') will be shown in a highlight colour: in Line display mode, it will be in the centre of the
line.

Click on the CONCORDANCE button on the toolbar to change the display mode (or select CONCORDANCE
12 under the QUERY menu). The CONCORDANCE button allows you to switch between Page and Line
display modes. Whichever of the two modes you are in, clicking on the button will change to the
other. Clicking on it a second time will return you to your original display mode. You can also
toggle the display mode by checking or unchecking the CONCORDANCE option under the QUERY
menu. For full details, see 1.6.4 on page 218.
To change the default display mode and the highlight colour, use the VIEW menu (see 1.2.8 on page
54).

Character Display
You will see that in Page display, the solution contains the sequence 'wrong &hellip;' at the end of the sentence following that containing 'cracksmen'. All
characters outside the basic alphabet are represented in the BNC by SGML entity references (see 9.2.2 on page 165, 2.4 on page 239), which SARA
converts into appropriate symbols as far as possible when displaying solutions. There is no way of displaying three dots (horizontal ellipsis) as a single
character under Microsoft Windows, and so here SARA displays the unconverted entity reference. You can change the way in which entity
references are displayed by configuring your own CUSTOM display format (see 1.2.8 on page 55, 9.2.2 on page 165).
1.2.7
Obtaining Contextual Information: the Source and Browse Options
On the status bar you will see the code ANL, followed by the number 368. The source text from which the current solution is taken is indicated by the
three-character text identifier code, displayed in the third box from the left on the status bar. In each text in the BNC, sentences are consecutively
numbered, and the number of the sentence from which the current solution comes is shown in the fourth box on the status bar.
The codes used for text identifiers do not provide any information about the nature of the source text. To see the full bibliographic details (author, title,
and publication details for written texts; participant details for spoken texts), you can use the SOURCE option. This also allows you to browse the whole of
the source text. A complete bibliographical listing of the texts in the BNC, in order of their text identifier codes, is provided in the BNC Users'
Reference Guide.

page_52

Page 53

Click on the SOURCE button on the toolbar (or select SOURCE from the QUERY menu). The BIBLIOGRAPHIC
13 DATA box will be displayed, showing that this solution comes from a written text, taken from a murder
story published in 1991. The OK button in the Bibliographic data box returns you to the solutions
display.
14 Click on the BROWSE button in the Bibliographic data box to display the source text. SARA will switch to
browse mode, opening a bnc window to display the source text. This will be represented as a series of
SGML elements between angle brackets. Initially only the element identifying the BNC document or
<BNCDOC> (the highest level of structure) will be shown, followed by a red line and box indicating the
point in the document structure at which the solution occurs.
15 Click in the red box to show the portion of text which contains the solution. The text will start to be
expanded, showing its header and various subdivisions (such as chapters or sections). When the whole
text has been downloaded from the server, the paragraph which contains the current solution will be
displayed in its fully expanded form, with the query focus highlighted. Every word in this paragraph
should have a tag next to it (shown between angle brackets).
16 If you cannot see these tags, select the B ROWSER menu and check TAGS. If you look carefully, you will
see that each sentence is preceded by an <S> tag with its consecutive number; each word is preceded
by a <W> tag with its part-of-speech code (see 7.1.2 on page 129), and each punctuation mark by a
<C> tag.
The text will be easier to read if these tags are removed from the display.

Select the BROWSER menu and uncheck TAGS. Then scroll through the text till you find the highlighted
17 query focus. The <P> element containing the solutions will be expanded as formatted text, with each
sentence beginning on a new line. The <S>, <W> and <C> tags have been removed from the display. You
 can choose to have these tags omitted from browse mode displays the next time you start SARA
by unchecking SHOW TAGS in the BROWSER section of the USER PREFERENCES dialogue box (see 1.2.8 on
the following page).
18 Use the scroll bar to show the text immediately preceding the expanded paragraph. You will see that the
preceding text is represented as a series of <P> elements, each of which represents a paragraph of the
text. These unexpanded elements are preceded by plus signs; whereas the <P> element which has been
expanded to show its internal structure is preceded by a minus sign.

page_53

Page 54

Click on the minus sign to remove the expansion of this paragraph. The minus sign will be replaced by a
19 plus, showing that this element is not expanded, and a red box will be added to the display, to indicate
the position of the solution.
20 Use the scroll bar to scroll back to the beginning of the browser display. You will see that the fifth
<DIV1> element in the text (the one which contains the solution) is expanded to show its internal structure
as a <HEAD> element (i.e. a heading) followed by a series of < P> elements.
21 Click on the plus sign next to the <HEAD> element. You will see that the heading to this section of the text
reads '5' it is in fact the fifth chapter of the book. The BROWSE option is the only way in which you can
examine the whole text from which a particular solution is taken, rather than just the context
immediately surrounding the query focus. You cannot save or print Browse displays.
22 Click in the Query1 window (or select QUERY1 from the WINDOW menu) to return to the solutions display.
You will be returned to the solutions to your query for 'cracksmen'.

Copying Solutions to the Windows Clipboard

When using SARA, you may often want to save a particular solution in order to analyze or quote it elsewhere. Start by saving the current solution.

Click on the COPY button on the toolbar (or select COPY from the EDIT menu). You can now paste this
23 solution into a Windows word processor, such as Word, Notepad or Write. Solutions are copied
together with the source information displayed on the status bar. The amount of context copied
and the format will generally correspond to those visible in Page display mode (see 8.2.5 on page
154). Remember that copyinq to the clipboard deletes the clipboard's previous contents. To save
an entire set of solutions to file, use the LISTING option (see 8.2.5 on page 154); to print a Line
mode display use the PRINT option (see 3.2.3 on page 81).

1.2.8
Changing the Defaults: The User Preferences Dialogue Box
Before formulating any further queries, let us change the default query display settings, so that solutions will from now on automatically be downloaded in
an appropriate format.

Under the VIEW menu, select PREFERENCES. The USER PREFERENCES dialogue box will be displayed.
24

From this box you can change SARA's default settings for a variety of features. We will start by requesting a maximum amount of context (1000
characters) and a maximum number of hits (100).

page_54

Page 55

Download Parameters
These specify the maximum amounts of data that should be sent from the server to your computer in response to a query.

Click on the MAX DOWNLOAD LENGTH box and type in 1000. The greatest maximum download length that
25 can be specified is 2000 characters. This includes SGML markup, and corresponds to around 150
 words of plain text.
26 Click on the MAX DOWNLOADS box and type in 100. There is no intrinsic limit to the maximum number of
solutions that can be downloaded, but the current version of the software cannot display more than
2000 solutions. On most systems, it will take a long time to download more than a few hundred.

Default Query Options: Format

These specify the way in which SGML markup is to be displayed on the screen.

Click on the PLAIN radio button. PLAIN is the fastest display format. It shows words and punctuation only.
27 Entity references (see 1.2.6 on page 52) are converted to their conventional typographical equivalents.
The query focus is shown in the highlight colour selected under the VIEW FONT and COLOURS options (see
1.7.4 on page 227). Format options are fully described in 1.6.4 on page 219. POS displays
different parts of speech in different fonts and colours (see 7.2.4 on page 137), according to the
default FONT and COLOURS selections. The default CUSTOM format behaves differently in Line and
Page display modes: in Line mode, paragraph and utterance boundaries are shown as vertical
lines, whereas in Page mode the appearance of a printed page is reproduced (see 6.2.2 on page
116). You can edit Custom format to display particular features as you wish (see 9.2.2 on page
165). SGML displays the text with full SGML markup, showing all elements and attributes in
angle-bracketed tags, and entity references in their coded form (see 1.2.6 on page 52; for an
example see 5.2.3 on page 103).

Default Query Options: Scope

These specify the amount of context to be displayed for each solution, within the limits set by the Max download length as defined in the Download
parameters (see above).

Click on the PARAGRAPH button. This will display one paragraph of written text, or one utterance of
28 spoken text. It will generally provide enough context to understand the sense in which the query focus is
used, and the nature of the source text. The SCOPE options are fully described in 1.6.4 on page 219.
AUTOMATIC will normally display the sentence (or a similarly-sized unit) in which the query focus
occurs, unless the query has to be satisfied within the span of a given number of words, in which
case as many sentences as are necessary to show that span will be displayed (see 8.2.3 on page
152); SENTENCE displays the sentence in which

page_55

Page 56

the query focus occurs; MAXIMUM will display the Max download length specified in the Download
parameters, trimmed to complete sentences. Since you cannot in any case show more text than
specified as the Max download length, you may not see a complete sentence or paragraph if the
Max download length is small or the sentence or paragraph is particularly long, regardless of the
scope selected.

Default Query Options: View

These options specify the layout of the window in which solutions will be displayed.

Check QUERY and ANNOTATION to show the text of the query and a space for notes above the solutions.
29
30 Check the CONCORDANCE box so as to display solutions in Line mode. The VIEW options are fully
described in 1.6.5 on page 221. QUERY displays the text of the query above the list of solutions.
This provides a useful way of checking complex queries, as well as of learning the syntax of CQL
(the SARA Corpus Query Language: see 7.2.3 on page 135). It also shows the number of
solutions, and details of any thinning which has taken place (see 2.2.3 on page 66). ANNOTATION
opens a space for notes above the list of solutions, which can be saved with the query and re-
displayed when you re-open it (see 4.2.2 on page 94). CONCORDANCE indicates whether solutions are
to be displayed in Line (i.e. concordance) or Page mode (see 1.2.6 on page 51).
Browser Options
The checkbox indicates whether low-level tags (<S>, <W>, <C>) are to be shown in browser displays (see 1.2.7 on page 52). Changing this setting
will only take effect the next time you log on to SARA.
Other Default Settings
To change the default display font and colours, use the FONT and COLOURS options under the VIEW menu (see 1.7.4 on page 227). The other options in
the USER PREFERENCES dialogue box are fully described in 1.7.5 on page 227. COMMS enables you to change the server address, port and timeout
interval, while PASSWORD enables you to change the password you use to log on to SARA. CANCEL returns you to the current solutions window
without changing the current defaults.
Applying Changed Defaults
Any change you make in the USER PREFERENCES dialogue box will apply to all further queries in this session, as well as the next time you log on to SARA.
It will not however affect queries for which solutions have already been downloaded and displayed on the screen, nor will it affect the BROWSER display in
the current session (see above). You can change the settings for individual sets of solutions after they have been downloaded by using the OPTIONS (see
5.2.3 on page 103), QUERY TEXT (see 2.2.3 on page 67), ANNOTATION (see 2.2.3 on page 67) and CONCORDANCE options (see 1.2.6 on page 51) under
the QUERY menu.

page_56

Page 57

Click on OK to return to the solutions window. Since the alterations made to the defaults only affect the
31 display of solutions to subsequent queries, the display in the current solutions window (Query1) will be
unchanged.

1.2.9
Comparing Qeries: 'Cracksmen' and 'Cracksman'
We are now ready to perform another query, this time looking for occurrences of the singular form 'cracksman'.

Click on the PHRASE QUERY button on the toolbar. The Phrase Query dialogue box will be displayed.
32
33 Type in the string cracksman and click on OK (or press ENTER). SARA will open a new solutions
window entitled Query2.

If you have changed the default options according to the instructions in the last section, you should find
that:
the text of the query ('cracksman') is displayed at the top of the window;
this is followed by a space where you can write notes if you click in it;
the four solutions are displayed in Line mode, the first of them being current.
34 Click on the CONCORDANCE button on the toolbar to switch to Page display mode. You should find that a
full paragraph of context (23 lines) is displayed for the first solution.
35 Press PGDN to show the next solution. Repeat to see the following solutions. Of the four solutions, three
are references to the title of Hornung's book.

Apart from the references to the book title, there is thus only one example for each of the forms 'cracksman' and 'cracksmen' in the corpus. Are these
occurrences merely idiosyncratic uses of an archaism by an isolated writer? You can see whether these solutions both come from the same source text
by comparing their text identifier codes, which are displayed in the third box from the left on the status bar.

Read the last 'cracksman' example and note its text identifier.
36
37 Click on the SOURCE button on the toolbar to display the bibliographic data for this text.
38 Click on OK (or press ENTER) to return to the solutions display.
39 Click in the Query1 window, if it is visible, or select Q UERY1 from the WINDOW menu.
 page_57

Page 58

Check the text identifier and source for this solution. You will see that the source texts are two different
40 novels. A difference in text identifier codes is not a completely reliable indicator that solutions
come from different source texts. For written materials, each distinct BNC document is a sample
of up to 45,000 words from a distinct text. However, some very large and miscellaneous texts
such as the Dictionary of National Biography were sampled more than once, as were newspapers
and periodicals, with a view to representing the wide variety of types of language such texts
contain. Consequently there are a few cases where different text identifiers indicate the same
source text. For spoken texts, the material gathered demographically (i.e. the collections of
spontaneous informal conversation) is grouped so that each respondent's conversations make up
a single text. This has not however been done for other kinds of spoken material, with the result
that such things as radio phone-ins or meetings are occasionally split across several texts.

We can perhaps infer that the word 'cracksman' has not totally died out in contemporary British English, at any rate within the genre of thrillers. It would
however seem to be very very rare, which would justify its absence from the one-volume dictionaries cited earlier, many of which are designed primarily
with foreign learners in mind.
1.2.10
Viewing Multiple Solutions: 'Whammy'
'Cracksman' and 'cracksmen' are very rare in the BNC. We now look at the word 'whammy', which we would expect to occur rather more frequently,
since it is included in all the dictionaries cited in 1.1.1 on page 48.

Click on the PHRASE QUERY button on the toolbar. Wait for the dialogue box to be displayed, then type in
41 the stringwhammy and click on OK (or press ENTER) to send the query to the server. If you have
followed all the steps in this task up to now, the results should be displayed in a window called Query3,
using the default display options selected in 1.2.8 on page 54, i.e. in Line mode.
42 Wait until the red BUSY dot on the status bar has stopped flashing, showing that all the solutions have
been downloaded. The solutions are displayed in the order they appear in the corpus, i.e. according to
their text identifier code and sentence number. In the second box from the left on the status bar, you will
see the numbers 1:46 (31). These numbers indicate that a total of 46 solutions have been downloaded,
taken from 31 different texts, and that the current solution is 1, i.e. the first in the display. The difference
between the second and the third numbers provides an indicator of dispersion of solutions across
texts: if the difference is large (relative to the total number of solutions), this implies that the
solutions tend to be concentrated in particular texts, rather than being evenly dispersed.

page_58

Page 59

If necessary, use the Windows buttons to enlarge the solutions window to full screen size.
43

Changing The Current Solution

Many of the solutions contain the expression 'double whammy'. Can a 'whammy' be more than double?

Click on the Windows scroll bar to scroll through the solutions until you reach the final one. You will find
44 a 'triple Conservative tax whammy' in the last solution, and a 'quadruple whammy' twelve lines above it.

45 Look at the first of the numbers in the second box from the left on the status bar. You will see that the
current solution is still number 1. Using the scroll bar (unlike the keyboard cursor keys) to scroll through
a Line mode display does not change the current solution.
46 Click on the last solution in the list to make it the current solution. You will see that the first of the
numbers in the second box on the status bar is now the same as the second number, i.e. the current
solution is now the last solution in the list.
Scrolling Through Solutions Using The Arrow Buttons
You can also move through a list of solutions (in either Line or Page display mode) by using the A RROW buttons on the toolbar. The inner pair of arrow
buttons take you to the previous or next solutions, while the outer pair take you to the first or last solutions. Clicking on any of these buttons also
changes the current solution.

Click repeatedly on the inner backward arrow button to make the current solution number 38. The
47 number of the current solution is displayed on the Status bar, and changes as you click on the arrows. In
Line mode, the current solution is always displayed with a dashed surround.
48 Click on the CONCORDANCE button on the toolbar to switch to Page display mode. Use the inner forward
arrow button to look through this and the following two solutions, which give an account of the origins of
the expression 'double whammy'. When looking up a rare word in the corpus, it is quite common to
find examples which define or explain the meaning of the word.
49 Click on the CONCORDANCE button to return to Line display mode, then click on the outer backward
arrow button to return to the first solution. From the number display on the status bar, you can see that
the first solution in the display is now the current one.
50 Look through the initial solutions again to see if they all use the word 'whammy' in the same sense.
Following a few 'double whammy' examples, you will see a group of solutions which use the word in
such phrases as 'whammy bar' and 'whammy pedal'.

page_59

Page 60

Unless you already know what a 'whammy bar' is, click on the first of these examples to make it the
51 current solution, then click on the CONCORDANCE button to switch to Page display mode.

52 Use the inside forward arrow button (or the PGDN key) to page through the other solutions in this group,
until you have roughly understood this second meaning of 'whammy'. While you can use the arrow
buttons on the toolbar to scroll through solutions in either Line or Page display mode, the
keyboard cursor keys behave differently according to the display mode. To scroll through solutions
one at a time, you must use the up and down arrow keys in Line mode, and the PGUP and PGDN
keys in Page mode. In Line mode, pressing the PGUP and PGDN keys moves you through the
display half a screenful at a time, whereas in Page mode, pressing the up and down arrow keys
moves you within the single solution displayed. Scrolling through solutions using the cursor keys
or arrow buttons also changes the current solution.

Results
It seems clear from the solutions that as well as the 'double whammy' sense of misfortune, 'whammy' quite often has the sense of 'whammy bar' (or
pedal) on an electric guitar. This meaning is not included in any of the dictionaries mentioned perhaps because it is less frequent, or considered a
specialized use.

Close each of the open windows individually by pressing C TRL-F4, or select CLOSE ALL from the WINDOW
53 menu. SARA will switch to browse mode, displaying the beginning of the corpus in a bnc window.

54 Press ALT-F4 to leave SARA, or else iconify the browse window, and proceed to experiment with the
queries described in the next section.

1.3
Discussion and Suggestions for Further Work
1.3.1
Caveats
Before drawing conclusions from frequency data, you should evaluate the precision of your query that is, whether all the solutions being counted
concern the phenomenon you are interested in. You should look carefully at the solutions themselves, which may include unexpected proper names,
foreign words, quotations, deliberate archaisms, misspellings, or misprints. You should also evaluate the query's recall that is, whether it is likely to have
found all the occurrences of the phenomenon you are interested in, or whether the latter may take other forms. Finally, you should consider the texts and
the corpus from which the numbers have been derived. Even relatively high frequencies may have limited dispersion, being due to the influence of just
one or two texts. Conversely, if a word is absent from the corpus, this need not mean it does not occur in the language: at most it may suggest that it
occurs relatively infrequently,
 page_60

Page 61

or in text-types which are under-represented in the corpus in question. Such methodological issues are further illustrated in the following exercises.
1.3.2
Some Similar Problems
'Thine' and 'Wight'
The first example in this task suggested that the word 'cracksman' is still occasionally used in contemporary English. Perform queries to find evidence in
the BNC for contemporary use of two other relatively archaic words, 'thine' (other than in addressing God), and 'wight' (other than in the name 'Isle of
Wight'). To increase precision, type in the search terms in lower case, and uncheck IGNORE CASE in the Phrase Query dialogue box. This will avoid
references to God and to the Isle of Wight. Read through the solutions to check that the word is not merely mentioned or quoted from an older
English text, or an attempt to imitate an archaic style.
The main non-religious use of 'thine' is in a quoted saying: 'Know thine enemy'. The 'wight' query demonstrates how occurrence in the corpus
does not always mean that a word is normally used in contemporary English. Of the 17 solutions, several appear to be misprints of 'might' or
'weight'; some reproduce deliberate mispronunciations in a punning conversation ('Quite wight'); some are mentions of Chaucer's 'gentil
wight'; while others are references to the names of creatures in J.R.R. Tolkien's fairy world based largely on Old English vocabulary.
'Avatars'
Neal Stephenson's novel Snow Crash (1993) tells the story of a struggle for power in two worlds, that of twenty-first century America, and that of the
Metaverse, a virtual reality where the characters interact assuming computer-generated physical forms. Stephenson terms these computerized personae
their 'avatars'. Does the BNC provide any evidence for the use of 'avatar' in this sense? If you look up both singular and plural forms, you will find
a total of 10 solutions. Using Page display mode to see a full paragraph of context, you will see that the main use of 'avatar' is in relation to
gods and demons arguably a similar use to Stephenson's?
'Psykers'
Lookup the word 'psyker' (and its plural 'psykers') in the BNC. Do you feel there is evidence of sufficient use to warrant its inclusion in a dictionary of
contemporary English? Taken together, the two forms occur a total of 21 times. However if you notice the display of the number of texts from
which the solutions are taken on the status bar, you will see that they have zero dispersion. As you can find out by using the SOURCE button,
they all come from a single science-fiction story.
'In-Your-Face'
One of the new words in the 1995 edition of the Collins COBUILD dictionary is the adjective 'in-your-face' (or 'in-yer-face'), which is defined as
follows: ''If you say that someone has an in-your-face attitude, you

page_61

Page 62

mean that they seem determined to behave in a way that is unconventional or slightly shocking, and that they do not care what people think of them.''
Is this word attested in the BNC? Does it always have this meaning? Which of the spellings is more widely used? While the COBUILD definition
takes 'in-your-face' to refer to people, in the corpus examples it is mainly used to describe drama and music. The 'in-yer-face' spelling occurs
only once. However, it is possible that we have not found all the occurrences, since we cannot be sure that 'in your/yer face' is always
hyphenated. You will see how to design queries to include orthographic variations of this kind in 10.2.5 on page 186.

page_62

Page 63

2
What is More than One Corpus?
2.1
The Problem: Relative Frequencies
2.1.1
'Corpus' in Dictionaries
In the last task we used the BNC to find evidence of the occurrence of particular forms and senses of words. In this exercise, we examine the evidence
it provides about the relative frequencies of different forms and senses. There is, of course, a big difference between saying that one form or sense is ten
times as common as another in a particular corpus, and saying the same thing of the language as a whole; this issue of the representativeness of a
corpus is discussed in 2.2 on page 36. Here we are primarily concerned with ways of obtaining numeric information from the BNC rather than with its
interpretation.
This task investigates plural forms of the word 'corpus' in the BNC, and the senses in which they are used. According to the entry for 'corpus' in the
second edition of the Oxford English Dictionary reproduced at the start of this Handbook, the plural form of this noun is 'corpora'. The definitions
given include five main senses and several phrasal uses, all of them rather esoteric. This suggests that 'corpus' is likely to be a fairly rare noun in the BNC
we shall see in a moment just how rare.
Three smaller recent dictionaries, the Collins COBUILD, the Longman Dictionary of Contemporary English, and the Cambridge International
Dictionary of English, give only the 'text collection' meanings (senses 3 and 4 in the OED2), but list both 'corpora' and 'corpuses' as alternative plural
forms. What evidence is there of the regularized plural 'corpuses' in the BNC? And is this regularized form only found for the 'text collection' sense?
2.1.2
Highlighted Features
This task shows you:
how to find the frequency of a word or group of words in the BNC using the WORD QUERY option;
how to look for occurrences of a word or group of words using the WORD QUERY option;
how to download only one solution from each source text using the options in the TOO MANY SOLUTIONS dialogue box;
how to mark and delete particular solutions using the THIN option;
how to save and re-open a query.
It assumes you already know how to:
log on to SARA and display the toolbar and status bar (see 1.2.1 on page 49);

page_63

Page 64

change the default settings using the View Preferences option (see 1.2.8 on page 54);
look up a word using Phrase Query (see 1.2.5 on page 51).
2.1.3
Before You Start
Log on to SARA and wait for the SARA-bnc window to be displayed (see 1.2.2 on page 49). Under the VIEW menu, check that the TOOLBAR and
STATUS BAR options are ticked, and then use the PREFERENCES option (see 1.2.8 on page 54) to set the default settings as follows:

MAX DOWNLOAD LENGTH 1000 characters

MAX DOWNLOADS
100
FORMAT
Plain
SCOPE
Paragraph
VIEW: QUERY and ANNOTATION
checked
CONCORDANCE
checked
BROWSER: SHOW TAGS
unchecked

Click on OK (or press ENTER) to return to the SARA-bnc window.

2.2
Procedure
2.2.1
Finding Word Frequencies Using Word Query
The WORD QUERY option allows you to specify a list of one or more words, and to search for occurrences of all of them. It also enables you to find out
quickly how often any word or group of words occurs in the BNC.
 Click on the WORD QUERY button on the toolbar to display the Word Query dialogue box. You can also
1 reach the Word Query dialogue box by choosing FILE from the menu bar, then NEW QUERY, then
WORD.
2 Type in the string corpus and click on the LOOKUP button. An alphabetical list will be displayed showing
all the word forms in the BNC which begin with the characters 'corpus'. Notice that the list contains
hyphenated forms like 'corpus-based', as well as some foreign phrases like 'corpus delicti' and
'corpus juris'. Eachforin in the list is a separate entry in the BNC word index, or L-word. As well as
conventional orthographic words, L-words include conventional and foreign phrases which
function as single units, such as 'in spite of' or 'corpus juris', as well as clitics, i.e. morphological
components used in contracted and possessive forms, such as ''s', 'ca' and 'n't'. For a list of clitics
and phrasal L-words used in the BNC, see the BNC Users' Reference Guide.

page_64

Page 65

If the index contains more than 200 L-words which begin with the string entered, an error message
will be returned. Either a longer string or a more precise pattern should be specified (see 8.2. 1 on
page 147).
3 Click on the first word in the list, i.e. 'corpus'. Clicking on a word in the list selects it and shows how often
it occurs in the BNC. You will see that 'corpus' occurs 724 times, with a z-score of 0.0518. z-scores
indicate how frequent an L-word (or group of L-words) is compared with the mean frequency of all
the L-words in the corpus, which is 149.3. The z-score is the number of standard deviations from
the mean frequency: a positive score indicates that a word is more frequent than the mean, a
negative one that it is less frequent. Thus 'corps' occurs 0.0518 times the standard deviation more
frequently than the mean.
4 Use the scroll bar to scroll through the list past 'corpuscle', 'corpuscular', etc. until you come to 'corpuses'.
Click on 'corpuses' to select it and display its frequency and z-score. 'Corpuses' occurs 9 times, with a z-
score of -0.0126.

The other form we are interested in, 'corpora', does not appear in the current list of matching words, since it does not begin with the characters 'corpus'.

Click in the input box and delete the final us of 'corpus'. Then click on L OOKUP again. The longer list
5 which now appears contains some very strange L-words, such as 'corpus' and 'corp-apple'. The
word index includes compounds, abbreviations, foreign words, and misprints, clitics and phrasal L-
words, as well as cases where two words had no space between them in the source text. The more
peculiar L-words rarely occur more than a few times in the corpus, so that for most purposes they
can be safely ignored.
6 Scroll through the list of matching words until you find 'corpora', and click on it to select it. You will see
that 'corpora' occurs 111 times in the BNC, over ten times more often than 'corpuses'. The Latin plural
form given by OED2 is thus clearly the more frequent one.

2.2.2
Looking for More than One Word Form Using Word Query
While these figures tell us something of the relative frequency of the two forms, they do not tell us if they are both used over the full range of senses. To
compare their uses we must examine the contexts in which they occur.
We shall do this by downloading occurrences of both forms. W ORD QUERY allows you to select several entries from a list of matching words and
combine them into a single query. After selecting a first item, you simply add other items by holding down the CTRL key when you click on them.

page_65

Page 66

Check that you have selected 'corpora' (i.e. that it is highlighted), then scroll down the list of matching
7 words till you find 'corpuses'. Hold down CTRL and click on 'corpuses' to add it to the selection. The total
 number of occurrences of 'corpora' and 'corpuses' will now be displayed, along with their combined z-
score.
8 Now click on OK to send the query to the server. You can also select a group of consecutive items in
the list by clicking on the first item and then draging the mouse, or by then holding down the SHIFT
key and clicking on the last of the group. The options in the WORD QUERY dialogue box are fully
described in 1.3.2 on page 200. If the PATTERN box is checked, SARA interprets the string entered as
a pattern, which may include alternatives and tvildcards (see 8.2.1 on page 147). This means that
if you look up 'corpus' with PATTERN checked, the list will contain only the L-word 'corpus', instead
of all the L-words beginning with the characters 'corpus' (for an example, see 3.3.1 on page 83).
COPY copies the input string to the Windows clipboard (for examples, see 7.2.1 on page 131).
CANCEL leaves the dialogue box without starting a query, while CLEAR deletes any previous input and
selections from the dialogue box.

2.2.3
Defining Download Criteria
Selecting from the Solutions
If you have set the Maximum downloads number at 100 in the default Download parameters, the T OO MANY SOLUTIONS dialogue box will be displayed.
Where the number of solutions to the query is smaller than the number specified in the default settings, all the solutions are automatically downloaded
and displayed. Where, on the other hand, a greater number of solutions is found, the Too many solutions dialogue box is displayed prior to any
downloading.
The dialogue box tells you:
how many solutions to the query have been found by the server;
in how many different texts these solutions occur;
the Maximum downloads number specified as a default under VIEW PREFERENCES (see 1.2.8 on page 54). The number of solutions found will
normally be the same as the total frequency of the selected items in the Word Query dialogue box. It may however differ where a selected word
also occurs as part of a phrasal L-word. For instance, while the index lists the frequency of 'corpus' as 724, a Word Query for 'corpus' will find
776 solutions, since the latter will also include occurrences of the phrases 'corpus juris', 'habeas corpus', etc.
You can choose how many and which of these solutions to download, namely:
the INITIAL n solutions found in the corpus, where n is any number specified in the DOWNLOAD HITS box (this is the fastest alternative);

page_66

Page 67

a RANDOM set of n solutions;

ALL the solutions found;
only ONE PER TEXT (where the number of texts is greater than the number in the D OWNLOAD HITS box, this alternative can be combined with the INITIAL or
RANDOM or DOWNLOAD ALL options). The ONE PER TEXT option discards all but the first solution in any text.
The number initially displayed in the DOWNLOAD HITS box is always that specified in the Max Downloads box under VIEW PREFERENCES (see 1.2.8 on page
54). You can change this number temporarily by typing a new number in the box, or by clicking on the DOWNLOAD ALL radio button. If the O NE PER TEXT
box is checked, clicking on the Download all button will change the number to the number of texts. If it is not, the number will change to the total
number of hits.
In this example, we shall use One per text as the criterion for downloading. The total number of solutions found (120) is very much larger than the
number of different texts in which they appear (17). This limited dispersion in the corpus suggests that 'corpora' and 'corpuses' are at least in some of
their senses highly specialized: they appear in very few texts, but in those texts they appear on average several times. Given publishers' tendencies to
standardize spellings, it seems unlikely that both forms will occur in the same text.
For the moment, we will assume that relatively little information will be lost by downloading only one solution per text, thereby reducing the total number
of solutions to 17. We will return to check the validity of this assumption in section 2.3.2 on page 72.

Check the ONE PER TEXT box.

9
10 Click on OK (or press ENTER) to start downloading the solutions.

The Query Text and Annotation Displays

Provided you checked the QUERY and ANNOTATION boxes under VIEW PREFERENCES in 2.1.3 on page 64, the text of the query which was sent to the server
win be displayed above the solutions, along with details of any thinning procedure adopted in the TOO MANY SOLUTIONS dialogue box. In addition there will
be a blank pane for notes below the query text. The query is shown in the SARA Corpus Query Language, or CQL for short. While not necessaryfor
simple searches, you may at some stage want to formulate queries directly in CQL (see 7.2.3 on page 135). Displaying the query text as you
use SARA is a convenient way of learning the CQL syntax. "us you will see that the query finding any instance of 'corpora' or 'corpuses' is
expressed as "corpora" ¦ "corpuses". Each word form is enclosed in double quotation marks, while the vertical line indicates a
disjunction The thinning procedure is shown as OPT

page_67

Page 68

(one per text), with the number of downloaded solutions (17) and the total number of hits found
on the server (120) in brackets.
11 Click in the Annotation pane between the query text and the solutions. A cursor will appear at the point
where you can type in notes. For instance, you might like to note the meaning of the query text. Any
annotations can be saved along with the query by using the SAVE option under the FILE menu (or by
clicking on the SAVE button on the toolbar: see 2.2.5 on page 70). You can also copy text to the
annotation pane from the Windows clipboard using SHIFT+INSERT.
12 Click anywhere in the solutions display to leave the Annotation pane. You can switch the QUERY TEXT
and ANNOTATION displays on and off in the current window from the QUERY menu.

Aligning Solutions
If this is your first query since starting SARA, the solutions will be displayed in the Query 1 window. Provided you checked CONCORDANCE under VIEW
PREFERENCES in 2.1.3 on page 64, they win be displayed in Line mode, with the query focus ('corpuses' or 'corpora') highlighted in the centre of each
line.
If you glance through the solutions, you will see that all the occurrences of 'corpuses' and 'corpora' are aligned to begin in the same column. You can
change the alignment to the centre or to the final character of the query focus using the ALIGN button on the toolbar.

Click on the ALIGN button on the toolbar. You will see that the solutions are now aligned around the
13 central point in the focus, and that the pattern on the ALIGN button on the toolbar has changed to a
symbol indicating central alignment.
14 Click again on the ALIGN button to see the third possibility, i.e. right alignment. The pattern on the button
changes, and the solutions are aligned on the final character of the query focus. Where, as in this
example, the query focus varies in length, changing the alignment can be a convenient way of
highlighting the patterns immediately preceding or following the query focus. For instance, in the
right-aligned display, solutions where the word following the queryfocus is 'of' are now more
clearly identifiable.
15 Click again on the ALIGN button to return to left alignment.

2.2.4
Thinning Downloaded Solutions
With such a small number of solutions it is relatively easy to count the instances of each form and sense by inspection. However we shall take the
opportunity to divide them up into groups using the THIN option.
Glancing through the 17 solutions, you will see that two main senses of 'corpuses' or 'corpora' are present:

page_68

Page 69

collections of texts or knowledge;

body parts ('corpora lutea' etc.).
There is however also one instance which is a citation from the Latin (with an English gloss), "Ne polluantur corpora. Lest our bodies be polluted". Let
us start by discarding this solution, as it is irrelevant to a study of the plural forms of 'corpus' in English.
 Click on this solution to select it. The dashed surround shows that it is now the current solution.
16
17 Mark this solution by double-clicking on it (or pressing the space bar). The solution will now be
displayed in reverse video.
18 From the QUERY menu, select THIN then REVERSE SELECTION. The list of solutions will be re-displayed, this
time without the one you marked. You will see that the total number of solutions indicated on the status
bar is now only 16, and that the thinning description in the Query Text pane records this further selection
as SEL 16/17. The T HIN menu is fully described in 1.6.3 on page 218. While REVERSE SELECTION
keeps the unmarked solutions, deleting marked ones, SELECTION does the opposite, keeping the
marked solutions and deleting unmarked ones. RANDOM keeps a specified number of randomly
selected solutions, while ONE PER TEXT keeps only the first solution from each source text: these
options are analogous to those in the TOO MANY SOLUTIONS dialogue box (see 2.2.3 on page 66), with
the difference that using THIN allows you to see all the solutions before reducing their number. The
RANDOM and ONE PER TEXT options do not distinguish between marked and unmarked solutions, and
any previous marking will be lost in the thinned display.
You cannot recover solutions once they have been removed by thinning, unless you have saved the
query in its previous state (see 2.2.5 on the next page).

We now want to divide the remaining solutions according to the sense of the query focus.

Glance through the solutions to decide whether the 'body part' or 'text collection' sense is the less
19 frequent. Overall, there appear to be fewer solutions with the 'body part' sense.

20 Scroll through the solutions using the arrow buttons on the toolbar. Press the space bar (or double-click)
on all those with the 'body part' sense to mark them. If you mark a solution by mistake, pressing the
space bar or double-clicking a second time will unmark it.
21 If you are uncertain of the sense used in a particular solution, click on the C ONCORDANCE button on the
toolbar to display the solution in Page mode and show a larger amount of context. Click on the

page_69

Page 70

CONCORDANCE button a second time to return to Line display mode. You may also be able to expand the
context even further by double-clicking on the right mouse button (see 1.2.6 on page 51).

2.2.5
Saving and Re-opening Queries
You should now have marked all the solutions which have the 'body part' sense. If you were now to use the S ELECTION option under the THIN menu, you
would obtain a display of only these solutions. If instead you used the REVERSE SELECTION option, you would delete them, obtaining a display of only
those solutions with the 'text collection' sense. We shall now create a second copy of the solutions in another window, so that by thinning each window
using a different criterion, we can obtain separate displays of the two sense groups. As we cannot display different sets of solutions for a single query,
we must first make a second copy of the query by saving it with two different names, one for each window.

Click on the SAVE button on the toolbar. The File Save As dialogue box will be displayed, proposing the
22 name query1.sqy for the query.

23 Click on OK to save the query.

24 If you are asked whether a preceding file with the same name should be replaced, click on OK (or press
ENTER). When saving is complete, you will see the solutions window has been renamed QUERY1.SQY.
Mere displayed solutions correspond to a saved query, the window name is that of the query file,
and is displayed in upper case.
SARA can save two kinds of files: Query files, which record queries, and Listing files, which
record solutions (see 8.2.5 on page 154). Unless the defaults are overridden, they are
automatically assigned .SQY and .SGM extensions respectively. To save queries and solutions
for future use, you are advised to use these file type extensions, but to choose filenames which will
act as appropriate mnemonics.
You now need to save a second copy of the same query, this time with a different name.

Select SAVE AS from the FILE menu. The File Save As dialogue box will be displayed once more, again
25 proposing to save the file as query1.sqy.

26 Type in a new name for the query (such as query1a) and click on OK. You will see that the title of
the current solutions window now shows the new name. As well as the text of the query itself, saving
a query records the display options selected at the time (including annotations); any bookmarks
(see 10.2.2 on page 181); any thinning selections, whether made when downloading solutions or
subsequently (see 2.2.3 on page 66, 2.2.4 on page 68). Saved queries which were randomly
thinned are re-opened with the same randomization.

page_70

Page 71

Now re-open your first copy of the query in a new window.

Under the FILE menu, click on QUERY1.SQY to open it. (Alternatively, click on the OPEN button on the
27 toolbar, then select query1.sqy from the File OPEN dialogue box). A new window will be opened with the
title QUERY1.SQY. This new window should contain exactly the same solutions as the other window
(QUERY1A.SQY.). The other options on the FILE menu are described in 1.3 on page 199. Viese
allow you to formulate a NEW QUERY of any type; to CLOSE the current window; to Print the current
solutions display (one-per-line) or to PREVIEW and change the PRINT SETUP (see 3.2.3 on page 81);
and to EXIT from SARA.

We shall now thin the solutions in each window according to their senses.

From the WINDOW menu, select TILE to display both windows. Unfortunately the solutions are now barely
28 visible, the screen being mainly occupied by the Query Text and Annotation panes. Let us therefore
remove these from the display.
29 Under the QUERY menu, click on QUERY TEXT to uncheck it. The Query Text pane will be removed from
the window.
30 Again under the QUERY menu, uncheck ANNOTATION. The Annotation pane will also be removed from the
window, which should now show only the solutions.
31 Click in the other window to make it the active window.
32 Following the same procedure as above, remove the Query Text and Annotation panes. You should now
be able to see most of the solutions, and to scroll through them using the scroll bar.
33 In the QUERY1A.SQY window, the solutions with the 'body part' sense are marked. Mark the same
solutions in the QUERY1.SQY window by double-clicking on them.
34 Under the Query menu, select THIN followed by SELECTION. This window will now include only the marked
solutions those with the 'body part' sense.
35 Click in the other window to make it the active window (or select it from the WINDOW menu).
36 Under the QUERY menu, select THIN followed by REVERSE SELECTION. This window will now include only
the unmarked solutions those with the 'text collection' sense.

You will see that the solutions in this window contain instances of both 'corpuses' and 'corpora', whereas the solutions in the other window only contain
instances of 'corpora'.

page_71

Page 72

Results
The plural 'corpuses' occurs only in the 'text collection' sense in the BNC. While the numbers involved are too small to draw reliable conclusions, it is
nonetheless interesting that the dictionaries cited offer no information on this point.
2.3
Discussion and Suggestions for Further Work
2.3.1
Phrase Query or Word Query?
We have so far seen two different ways of formulating a query. While in many cases either procedure can be used, note that only P HRASE QUERY allows
you to:
make a query case-sensitive (for instance to include or exclude proper names: see 1.3.2 on page 61, 10.2.4 on page 184);
search the text headers as well as the texts (for instance to include occurrences of keywords, or of a particular author or publisher name);
search for multi-word phrases (see 4.2.1 on page 87);
search for orthographic words which are not treated as L-words in the word index (e.g. 'polloi', which only appears in the index under 'hoi polloi'; 'don't'
or 'gonna', which are listed under the clitics 'do', 'n't', 'gon' and 'na'; etc.);
specify punctuation the query (see 10.2.7 on page 189).
Only WORD QUERY, on the other hand, allows you to:
generate a list of words matching a pattern from which to make selections (see 2.2.1 on page 64, 8.2.1 on page 147);
include multiple selections in a query (see 2.2.3 on page 66);
consult the index to see whether a word occurs in the corpus and with what frequency;
consult the index to see whether a phrase or clitic is treated as a single L-word.
These last two options are illustrated in the problems in the next section.
2.3.2
Some Similar Problems
The Dispersion of Different Forms
In the task above, we only examined one solution from each text, assuming that no text would use both of the plurals of 'corpus'. Is this in fact the case?
Are publishers always consistent? Carry out a WORD QUERY for 'corpuses', then thin the solutions to ONE PER TEXT and note their number. Then
carry out a second query for 'corpora'. You can either download one solution per text, or simply read off the number of texts containing
solutions from the Too many solutions dialogue box.

page_72

Page 73

The sum of the two numbers is 18 whereas the number of texts found in the query in 2.2.3 on page 66 was only 17. So there is in fact one text
in the corpus where both plural forms are used. You can find out which one it is by comparing the text identifier codes in the two sets of
solutions. If you look at the bibliographic data for this text using the SOURCE option, you can also discover its publisher.
Strong and Weak Verb Forms
Some English verbs have both strong and weak past tense or past participle forms, which may not be used with the same frequency. In the one million
word LOB corpus of British English from the early 1960s, for instance, the strong form 'dreamt' occurs only once, whereas the weak form 'dreamed'
occurs 14 times (Hofland & Johansson 1982). Does the BNC confirm that 'dreamt' is less common than 'dreamed'? What about 'knelt' and 'kneeled'?
These instances of strong and weak forms are chosen deliberately. In contrast with 'burned/burnt', say, 'dreamed' and 'dreamt' and 'kneeled'
and 'knelt' have clearly distinct pronunciations, so their frequencies should not be influenced by the spelling preferences of the transcribers of
the spoken BNC texts. They are also only used as verbs, unlike 'burned' and 'burnt', which may have different relative frequencies as verbs and
as adjectives.
If you use WORD QUERY to look them up in the word index, you will find substantial differences in the relative frequencies of both 'dreamed' and
'dreamt' (776:284) and 'kneeled' and 'knelt' (24:546). Being derived from a far larger corpus, these figures probably provide a more reliable
estimate of relative frequency than those derived from LOB. The LOB figures are unfortunately too small to warrant hypotheses as to whether
the relative frequencies of 'dreamed' and 'dreamt' have changed since the 1960s.
A Topical Reference: 'Annus Horribilis'
In her 1992 Christmas speech, Queen Elizabeth referred to that year as an 'annus horribilis', punning on the Latin expression 'annus mirabilis'. Look up
both of these expressions in the corpus, and find out which is more frequent, and what the uses of each refer to. If you look up the string annus
using WORD QUERY, you will find that 'annus', 'annus horribilis' and 'annus mirabilis' are all present in the word index, with a combined total of
45 occurrences. As you will discover, there are good reasons to include all three forms in the query, which you can do either by selecting all of
them, or just 'annus'. Download all the solutions, then use the THIN option to remove those which are irrelevant.
'Annus horribilis' (including associated misspelled variants) is the more common form, but with a much narrower range of reference virtually
always alluding to the Queen's speech, and very often in quotation marks. This illustrates how frequency may be affected by the topicality of a
particular event at the moment when the corpus texts were created, and may or may not reflect wider trends in the use of the language.

page_73

Page 74

3
When is Ajar Not a Door?
3.1
The Problem: Words and Their Company
3.1.1
Collocation
The playground riddle ''When is a door not a door? When it's ajar'' plays on the fact that the word 'ajar' is closely associated with the word 'door' in
English. Intuitively, we would therefore expect 'ajar' to co-occur with 'door' in texts, or collocate with it, with greater than chance frequency (see 1.3.3
on page 13). However, not only doors can be ajar. OED2 defines this sense of the word as follows:
Of a door or window: on the turn, slightly opened.
[ ... ]
1708 Swift Abol. Chr. Wks. 1755 II. i. 90 opening
a few wickets, and leaving them at jar.
1786 Beckford Vathek (1868) 92 With a large door
in it standing ajar.
1815 Scott Ld. of Isles v. iii, But the dim lattice
is ajar.

The examples here indicate some other possible referents. They also indicate that 'ajar' may occur either as a complement of copular verbs such as 'be'
and 'stand' ('the door stood ajar'), or of complex transitive verbs such as 'leave' ('he left the door ajar'). They provide no information, on the other
hand, concerning possible adverbial modifiers indicating, for example, degree (does one say fractionally ajar'?). Nor do they possible metaphorical uses.
And there is no information concerning the relative frequency of alternative referents.
OED2 also gives a second sense of 'ajar':
In a jarring state, out of harmony, at odds.
1860 Hawthorne Marble Farm (1879) I. xiii. 129 Any
accident. that puts an individual ajar with
the world.
1877 H. Martineau Autobiog. I. 83 My temper was
so thoroughly ajar.

This sense would seem to be distinguished from the first by having different collocates we would probably not speak of 'putting' a door ajar, or of a
door being 'thoroughly' ajar.
This task looks at 'ajar' in the corpus to see:

page_74

Page 75

in which of these two senses 'ajar' is mainly used;

what things tend to be 'ajar';
in what verb complement structures 'ajar' appears;
what adverbs modify 'ajar'.
3.1.2
Highlighted Features
This task shows you:
how to investigate collocates and their frequencies using the COLLOCATION option;
how to highlight collocates adjacent to the query focus using the SORT option;
how to PRINT solutions in one-per-line (KWIC) format;
how to use the PATTERN option in WORD QUERY to look up a particular word-form in the index.
It assumes you already know how to:
adjust the default display settings using the View Preferences option (see 1.2.8 on page 54);
carry out a Phrase Query (see 1.2.5 on page 51) and a Word Query (see 2.2.2 on page 65);
adjust downloading procedure in the Too many solutions dialogue box (see 2.2.3 on page 66);
align displayed solutions (see 2.2.3 on page 68).
3.1.3
Before You Start
Log on to SARA and wait for the Sara-BNC window to be displayed (see 1.2.2 on page 49). Using the PREFERENCES option under the VIEW
menu, set the default settings as follows:

MAX DOWNLOAD LENGTH 400 characters

MAX DOWNLOADS
200
FORMAT
Plain
SCOPE
Paragraph
VIEW: QUERY and ANNOTATION
checked
CONCORDANCE
checked
BROWSER: SHOW TAGS
unchecked

page_75

Page 76

3.2
Procedure
3.2.1
Using the Collocation Option
The COLLOCATION option allows you to find out exactly how often a particular word occurs in a set of solutions within a given number of words on either
side of the query focus.
First, let us find all the occurrences of 'ajar' in the corpus.

Click on the WORD QUERY button to display the Word Query dialogue box.
1
2 Type in the string ajar and click on LOOKUP.
3 Click on 'ajar' in the matching words display to select it, then click on OK to send the query to the server.
4 Wait for the solutions to be displayed in the Queryl window. There are 133 occurrences of 'ajar', so it
may take some time to download them all.
5 Scroll through the concordance display to look at the solutions. You will see that 'door' occurs in a large
number of them.

Now let us find out just how often 'door' appears as a collocate.

From the QUERY menu, select COLLOCATION. The Collocation dialogue box will be displayed. The box
6 initially shows:

the name of the current query (Query 1);

 the number of hits on which collocation frequencies will be calculated. Where all the solutions to the
query have not been downloaded, the USE DOWNLOADED HITS ONLY option will be available (see 3.2.4 on
page 82);
an empty COLLOCATE window, where you can type or paste in words to find out how often they collocate
with the query focus;
the collocation SPAN, i.e. the maximum number of words from the focus within which an occurrence is to
be counted as a collocation.

Click in the COLLOCATE window and type in door. The collocate must be a single L-word which is
7 included in the Sara word index, a single punctuation symbol, or an SGML element (entered between
angle brackets). Phrases or patterns will return zero scores.
8 Click on the CALCULATE button (or press ENTER). The word 'door' will appear listed beneath the COLLOCATE
window. The numbers in the display indicate:
the collocation frequency, i.e. the number of times the collocate appears within the selected span in the
solutions (25);

page_76

Page 77

the collocation ratio, i.e. the ratio of the collocation frequency to the number of solutions (0.19 = 19%).

Assuming that 'ajar' always follows its referent (we do not say 'an ajar door'), this means that almost one in five 'ajar's is immediately preceded by the
word 'door' (as in 'he left the door ajar'). Particularly frequent collocates (especially function words such as 'the', 'of', 'that' etc.) may occur more
than once within the span in a particular solution. In such cases the collocation ratio may be greater than 1.00. Any such cases of upwards
collocation (Sinclair 1991) will be displayed in the highlight colour.
Changing the Collocation Span
A glance at the solutions shows that 'door' may also appear several words before 'ajar'. To include cases where 'door' is a non-adjacent collocate, you
must increase the span within which collocation is calculated.

Click in the SPAN box and change the value to 3, then click on the CALCULATE button. You will see that the
9 numbers for 'door' have changed to 71 and 0.53. In other words over half the solutions for 'ajar' have
'door' as a collocate within a span of 3 words on each side of the focus. The collocation span may be
varied up to a maximum of 9, i.e. 9 words to each side of the first word of the query focus. Span is
calculated in L-words (see 2.2.1 on page 64); punctuation symbols also count as words. The default
span value is 1, i.e. one word to either side of the first word of the query focus.
While tending to increase collocation frequency, increasing the span decreases the probability that
this frequency indicates a strong association between the collocate and the focus, since it makes it
more likely that these occur in different clauses, sentences or even paragraphs dealing with
different topics. For instance, were you to use a span of 7 or more to calculate occurrences of
'OED2' as a collocate of 'next' in this sentence, you would also include the 'OED2' from the first
sentence of the next section.

Investigating Other Potential Collocates

The OED2 entry (see 3.1.1 on page 74) suggests that nouns which are semantically similar to 'door' may also collocate with 'ajar'. You can investigate
these by calculating the collocation frequencies of words which you have observed in the solutions or expect to be present in them.

Click in the COLLOCATE window and typewindow. Then click in the SPAN box and set the value to 9, so
10 as to maximize collocation frequency. You will see that the collocate list now contains both 'door' and
'window', in alphabetical order. Notice that the collocation frequency and ratio for 'door' have been
recalculated for the new span. 'Window' collocates very much less frequently with 'ajar' than 'door' does:
only 5 times, or in a ratio of

page_77
 Page 78

0.04, as opposed to 108 times (ratio 0.81) for 'door'. You will also notice that the collocate 'door' is
now displayed in a different colour. This is because its collocation ratio is close to being that of an
upward collocate (between 0.80 and 1.0).

Printing Collocation Data

So far we have only considered the singular forms 'door' and 'window'. From the point of view of understanding the semantic categories which can be
'ajar', it may also be relevant to check the collocation frequencies of the plural forms.

Type windows in the collocate window, and reduce the span to 3. Then click on the CALCULATE
11 button. 'Windows' collocates only once with 'ajar' in this span; the collocation frequency of 'window' is
reduced to 4.
12 Now type the form doors and click on CALCULATE. 'Doors' occurs 5 times in this span.

At the end of this task, we shall compare the collocation frequencies of 'door' and 'window' to see whether they differ significantly. To obtain a record
of these numbers, let us print the current contents of the Collocation dialogue box.

Click on the PRINT button. The Windows Print dialogue box will be displayed.
13
14 Check that the printer indicated is available, and click on OK to print. In the current release of the
software, this is the only way of saving the results of a collocation analysis.

3.2.2
Investigating Collocates Using the Sort Option
Other words which come into mind as possible collocates of 'ajar' are degree adverbs such as 'a little', and copular verbs such as 'be', 'stand', etc. As
the COLLOCATION option only allows you to investigate specific words, you will probably need to return to the concordance display in order to identify
further words which may recur as collocates.
One way to highlight such words is to use the SORT option. When solutions to a query are first downloaded, they are displayed in the order in which they
appear in the corpus. Using SORT enables you to re-order solutions, grouping those which have similar patternings.

Click on CLOSE to leave the Collocation dialogue box and return to the concordance display. Closing
15 the Collocation dialogue box does not delete its contents: re-selecting COLLOCATION will return you
to the previous display.
16 Select SORT from the QUERY menu to display the Sort dialogue box. SORT is only available from a
solution display in Line mode.

page_78

Page 79

The SORT option allows you to specify two sets of criteria, or keys, by which downloaded solutions are to be sorted. For each key you can choose:
DIRECTION whether to sort the solutions by the:
LEFT the words which precede the query focus;
CENTRE the query focus itself (this is relevant where the query is satisfied by a range of different: forms, as in the 'corpora/corpuses' example in
the last task: see 2.2.2 on page 65);
RIGHT the words which follow the query focus.
ORDER whether to display the sorted solutions in:
ASCENDING order (AZ, so that 'antelope' precedes 'zebra');
DESCENDING order (ZA, so that 'zebra' precedes 'antelope').
SPAN
the number of words to consider in sorting. The maximum span is 5 words. With this span, when sorting by the LEFT, Sara will discriminate between
solutions which have the same four words preceding the focus, but the fifth preceding word is different.
The default values for both Primary and Secondary keys are for left sorting, in ascending order, with a span of one word meaning that solutions will be
ordered alphabetically by the word immediately to the left of the query focus. The Secondary key is used where the Primary key does not discriminate
between solutions. The COLLATING options in the Sort dialogue box are fully described in 1.6.2 on page 217. They determine the alphabet which
will be used in sorting. ASCII orders words orthographically using the ASCII character sequence, roughly corresponding to A-Z followed by a-
z. IGNORE CASE (the default collating order) instead treats upper and lower case letters as identical. IGNORE ACCENTS uses the ASCII sequence, but
treats accented and unaccented letters as identical. POS orders words according to their part-of-speech codes rather than their orthographic
form; this option is only available when solutions are displayed in POS format (see 7.2.4 on page 137).
In the current query, using the default sort and collating values will mean that 'door ajar' will precede 'window ajar', but that 'a door ajar' and 'the door
ajar' will not be distinguished. Solutions in which particular words recur directly preceding 'ajar' will also be grouped together.

Click on the SORT button and wait for SARA to display the sorted concordance.
17
18 Check that the solutions are aligned by the left of the query focus, clicking on the A LIGN button if
necessary (see 2.2.3 on page 68).

page_79

Page 80

Use the scroll bar to scroll through the solutions. You will see that the solutions where 'door' directly
19 precedes 'ajar' now form a clearly visible group. You should also see some other groups of solutions
where the same word recurs prior to 'ajar' for instance 'slightly ajar' and 'left ajar'.

Collocate Positions
Many collocates have fixed positions, for instance immediately before or after the word they collocate with. While we have seen that the distance
between 'door' and 'ajar' varies, the positions of other collocates of 'ajar' may be fixed. For instance, do 'slightly' and 'left' always directly precede
'ajar'? Used together, the SORT and COLLOCATION options can provide answers to such questions.

Count the number of occurrences of 'slightly ajar' in the sorted solutions display.
20
21 Select COLLOCATION from the QUERY menu to re-display the Collocation dialogue box.
22 Find the collocation frequency of 'slightly' in a span of 1, then of 3, 5, and 7 words. The collocation
frequency is 15 in each case. This is also the same as the number of 'slightly ajar's. 'Slightly' thus appears
to have a fixed position as a collocate of 'ajar', directly preceding it.
23 Click on CLOSE to return to the solutions display.
24 Repeat the previous four steps for 'left', increasing the span progressively from 1 to 7 words. You will
find that it may occur at any point within a span of 5 words from 'ajar'.

Phrasal Collocates
Sorting also provides a useful heuristic to identify phrases, rather than just single words, which collocate with a particular query focus. For instance, in
some of the solutions where 'door' precedes 'ajar', 'door ajar' forms part of a larger recurrent phrase.

Select SORT from the QUERY menu. The SORT dialogue box will be displayed, showing the selections of
25 the last sort performed.

26 Change the Primary key span to 3, then click on SORT. The solutions will be re-sorted, grouping those
where the same three words precede 'ajar'.
27 Scroll through the solutions, and make a list of cases where three or more share the same two or more
words preceding 'ajar'. Your list should include 'leave the door ajar', 'the Gates of Heaven Ajar', 'a little
ajar', 'a door stood ajar', 'it was ajar', etc.
28 Using the COLLOCATION option, vary the span to see if any of the words in these sequences, such as
'gates' and 'little', always occur in the same position. You will find that 'little' only occurs as the first word
to the left of 'ajar'.
 page_80

Page 81

Click on CLOSE to close the Collocation dialogue box. As well as using SORT to highlight collocations
29 where the query focus forms part of a recurrent phrase, you can also highlight cases where the
query focus forms part of a recurrent sequence of word classes, or colligations (Sinclair 1991), by
sorting solutions with POS rather than ASCII or IGNORE CASE collating (see 7.2.4 on page 137).

3.2.3
Printing Solutions
Other Senses of 'Ajar'
Is the second sense of 'ajar' provided by OED2, that of 'out of harmony', present in these solutions?
Rather than scrolling through the sorted solutions, looking for instances which might have this meaning, you may find it easier to work with a printout.

Click on the PRINT button on the toolbar (or select PRINT from the FILE menu). The Print dialogue box will
30 be displayed, showing your currently selected Windows printer.

31 If your printer supports landscape orientation, but this is not currently selected, click on S ETUP to display
the Windows PRINT SETUP dialogue box. You can also access PRINT SETUP at any time from the FILE
menu.
32 Click on the LANDSCAPE orientation radio button, then on OK to return to the Print dialogue box.
Landscape orientation allows a larger context to be printed for each solution.
33 Click on OK to begin printing. The solutions in the window will be printed as a one-per-line (KWIC)
concordance, corresponding to Line display mode in the format currently selected. The corresponding
text identifier and sentence number for each solution will be printed at the beginning of each line. You can
get a rough idea of the appearance of a printout with the current settings by selecting PRINT
PREVIEW from the FILE menu.

Now examine the solutions.

Mark any solutions which look as if they may involve the 'out of harmony' sense of 'ajar'. If you do not
34 have a printout, double-click on them in the Line display. You can safely ignore those where 'ajar' is
preceded by 'door', 'left', 'slightly', 'stood', etc.
35 Check any candidates by switching to Page display mode to see a wider context. You will see there are
no solutions where 'ajar' is used with the 'out of harmony' sense, unless you count it to be implied in the
following: "The court accepted that the minister would not be expected to hear such representations as if
he were a judge. The minister would not be expected to approach the matter with an empty mind, but his
mind should, in the words, of the court, at least be ajar."

page_81

Page 82

3.2.4
Investigating Collocations Without Downloading: Are Men as Handsome as Women are Beautiful?
In looking at collocates of 'ajar', we were able to download all the solutions and study these individually. For queries which have large numbers of
solutions, it may be impractical to download all of them, but it is still possible to use SARA to calculate the frequencies of particular collocates in all the
solutions.
Taking the corpus as a whole, the collocates of the words 'women' and 'men' seem likely to differ. For instance, we might expect the former to be
'beautiful', but the latter to be 'handsome'. In this section we look to see if the BNC confirms this stereotype.

Click on the PHRASE QUERY button to open the Phrase Query dialogue box.
36
37 Type in the string men and click on OK to send the query to the server. The Too many solutions
dialogue box will be displayed, showing that there are 38,892 solutions.
To investigate the frequencies of particular collocates in these solutions, it is only necessary to download one of them.

Change the DOWNLOAD HITS number to 1, then click on OK. Only the first solution in the corpus will be
38 downloaded.

39 Select COLLOCATION from the QUERY menu. The Collocation dialogue box will be displayed, showing the
number of hits for the query as 1.
40 Uncheck the USE DOWNLOADED HITS ONLY box. The number of hits indicated will change to the total number
of solutions that were found to this query, i.e. 38,892.
41 Type the string beautiful in the collocate window and click on the CALCULATE button (or press
ENTER). The collocation frequency and ratio will be shown.
42 Now enter the string handsome in the collocate window and repeat the operation. You will see that
men appear more frequently handsome than beautiful in the corpus.
43 Increase the span to 3, then 5, then 7 words. You will notice that at a span of 5 words, 'beautiful'
becomes a more common collocate than 'handsome'. While 'beautiful' is less closely associated with
'men' than is 'handsome', it is a much more common word overall.
44 Repeat the procedure for 'women'. You will see that 'handsome' is much less common as a collocate than
'beautiful'. Again, notice that increasing the span for which collocations are calculated has a
proportionally greater effect on the more weakly associated collocate: for 'women' the proportion of
'beautiful'

page_82

Page 83

to 'handsome' is, with a span of 1, 70:2; with a span of 5, 105:10; with a span of 9, 134:22.

The numbers of occurrences of 'women' and 'men' in the corpus are very similar, but there are clear differences in their collocates, men being more
rarely beautiful and women more rarely handsome.

3.3
Discussion and Suggestions for Further Work
3.3.1
The Significance of Collocations
By using the COLLOCATION option, we have seen that things ajar are generally doors, but also occasionally windows and gates (and even minds and
lips!), which are, are left, or stand slightly or a little ajar; that men are rarely beautiful and women rarely handsome. To assess the significance of these
results, however, we must consider both the collocation ratios of these words with the query focus, and their relative frequencies in the corpus overall.
Fairly frequent recurrence of a relatively rare word as a collocate is likely to be of greater significance than the very frequent recurrence of a very
common word, such as 'the' or 'of'. Thus, while we have seen there are very many more beautiful women than there are handsome ones in the corpus
(70:2 with a span of 1), we cannot judge whether the former is a stronger association without taking into consideration the fact that 'handsome' is a much
less common word in English.
You can get a rough idea of the relative significance of the collocation ratios of two words in a particular set of solutions by comparing these with their
z-scores (see 2.2.1 on page 64). For instance, the collocation ratios of 'door' and 'window' with 'ajar', for a span of 3 words, were 0.53 and 0.03
respectively (see 3.2.1 on page 77). Now use WORD QUERY to find out their z-scores.

Click on the WORD QUERY button on the toolbar.

45
46 In the Word Query dialogue box, type in the string door and check the PATTERN box before clicking on
LOOKUP. If the P ATTERN box is checked, only the word which precisely matches the string 'door' will be
listed, rather than all the words which begin with those characters.
47 Select 'door' in the matching words list and read off its z-score. 'door' has a z-score of 2.2725 in the
corpus.
48 Do the same for 'window'. 'window' has a z-score of 0.9408.
49 Divide the collocation ratio for each word by its z-score. 'door' has a much stronger collocational link
 with 'ajar' than 'window' does (0.238 vs 0.032).

Corpus linguistics has developed a variety of methods for assessing collocational significance. The best known is probably the mutual information
score (Church & Hanks 1990), which compares the frequency of co-occurrence of

page_83

Page 84

two words in a given span with their predicted frequency of co-occurrence, i.e. that which would be expected were these words each randomly
distributed in the corpus. Mutual information can be calculated using the following formula (Stubbs 1995):
I = log2 ((f(n,c) x N)/(f(n) x f(c))

where f(n,c) is the collocation frequency, f(n) is the frequency of the node word (the query focus), f(c) is the frequency of the collocate, and
N is the number of words in the corpus.
For instance, using WORD QUERY to look up the overall frequencies in the corpus, the mutual information scores for 'men' and 'handsome', and for
'women' and 'beautiful' in a span of 3 can be calculated as follows:
'handsome/men'
f(n,c) = 23 (the collocational frequency of 'handsome' with 'men' in a span of 3: see 3.2.4 on page 82);
f(n) = 1683 (the overall frequency of 'handsome');
f(c) = 38,892 (the overall frequency of 'men');
N = 100,000,000 (the approximate total number of words in the BNC);
I = log2 ((23 x 100,000,000)/(1683 x 36,892)) = log2 (35.1) = 5.13.
'beautiful/women'
f(n,c) = 92 (the collocational frequency of 'beautiful' with 'women' in a span of 3: see 3.2.4 on page 82);
f(n) = 8670 (the overall frequency of 'beautiful');
f(c) = 39,916 (the overall frequency of 'women');
N = 100,000,000 (the approximate total number of words in the BNC);
I = log2 ((92 x 100,000,000)/(8670 x 39,916)) = log2 (26.6) = 4.75.
It thus emerges that 'handsome' and 'men' have a marginally higher mutual information score than 'beautiful' and 'women'. Church and Hanks (1990)
suggest that for a span of 3 words on either side of the focus, a mutual information score greater than 3 may indicate a significant collocational link. The
formula should not however be relied on where the collocates are relatively infrequent as a rule of thumb, where the product of their individual
frequencies is less than the size of the corpus. For this reason, in the case just examined it would have been more prudent to also include collocations of
'beautiful' and 'handsome' with the singular forms 'man' and 'woman' before making the calculation.

page_84

Page 85

3.3.2
Some Similar Problems
Time Immemorial
How often does 'immemorial' have 'time' as a collocate, and is this always in the phrase 'time immemorial'? Use WORD QUERY to download all 81
occurrences of 'immemorial' in the corpus. Calculate the collocation ratio of 'time' with a span of 1, and then increase the span to see if the
collocation ratio also increases. (As collocation span is calculated on either side of the query focus, you should also sort all the solutions by
the first word to the right in order to make sure that there are no cases where 'time' follows rather than precedes 'immemorial'.)
'Time' collocates with 'immemorial' 45 times within a span of 1, so this is quite the most common use of the latter word. Increasing the span to 9 only
increases the collocation frequency by 1 (to 46), suggesting that the position of 'time' as a collocate is fixed.
Wild Things
Which are more commonly wild in the corpus: animals or flowers? As there are over 5000 occurrences of 'wild', you should download the minimum
number of solutions, and uncheck the USE DOWNLOADED HITS ONLY box.
'Flowers' and 'animals' collocate with 'wild' with almost identical frequency in a span of 1, but at wider spans, 'animals' becomes the more
frequent, reflecting its higher z-score. The singular 'animal' is more frequent than the singular flower' at all spans, but both are far less
common than the plural forms.
The Clapham Omnibus
How frequently do 'Clapham' and 'omnibus' occur in collocation? What longer phrasal collocates appear in the solutions? In order to examine phrasal
collocates, you will have to download all the solutions. As Sara can take a long time to download and sort, you should generally aim to keep
numbers as low as possible. If you first use WORD QUERY to look up the frequencies of both 'clapham' and 'omnibus', you can then send the
query for the less frequent form, downloading all the solutions and then calculating the collocation frequency of the other. Sorting the
solutions using the 5 words to the left as Primary key, and the 5 words to the right as Secondary key will group, inter alia, occurrences of the
phrase 'the man on the Clapham omnibus'.
Mad Hatters
What proportion of BNC hatters are mad? Carry out one WORD QUERY for 'hatter' and another for 'hatters', and see if their collocation ratios
with 'mad' are different. You should use a span of several words if you are to include cases of 'mad as (a) hatter(s)' as well as 'mad hatter(s)':
varying the span from 1 to 4 should provide an indication of the relative frequencies of these two forms.
There are only two cases where 'hatters' has 'mad' as a collocate, and both of them appear to be misprints of 'hatter's. The singular 'hatter', on the other
hand, has 'mad' as a collocate from 28 to 44 percent of the time, according to the span selected. There thus appears to be a strong collocational link for
the singular form: for a span of 3 there are 26 occurrences, with a mutual information score of over 13 (see 3.3.1 on page 83). And this is excluding the
two cases misprinted as 'hatters' (not to mention a further 'loopy' one).

page_85

Page 86

4
A Query Too Far
4.1
The Problem: Variation in Phrases
4.1.1
A First Example: 'The Horse's Mouth'
Many idiomatic phrases occur in environments which, while less fixed in form than the phrase itself, vary relatively little. Consider 'a stone's throw'. You
may 'live within a stone's throw', but perhaps not 'go a stone's throw from here'; you may live 'a stone's throw away', but probably not 'a stone's throw
round the corner'. Accounts of idioms in dictionaries rarely describe such limitations, giving at most a few examples of environments they consider
typical. The Cambridge International and Collins COBUILD dictionaries, for instance, illustrate the phrase 'from the horse's mouth' with the examples
''hear something from the horse's mouth'' and "straight from the horse's mouth". These examples suggest that 'hear' and 'straight' are frequent collocates
of 'from the horse's mouth' (the Longman Dictionary of Contemporary English in fact lists the expression as 'straight from the horse's mouth'). This
task examines the BNC to find out what parts of the phrase appear to be fixed, and the range of the environments in which it occurs.
4.1.2
A Second Example: 'A Bridge Too Far'
Even relatively fixed parts of an idiomatic phrase can occasionally be varied, often for ironic or humorous effect. A headline in The Guardian in July
1995 referred to a Bosnian peace conference as 'a fudge too far'. Native speakers of English will recognize this as a punning variant of 'a bridge too far',
playing on the latter idiom's meaning of an overambitious objective. 'A bridge too far' is absent from the dictionaries mentioned, and even were it
present, we would probably not expect to find variants like 'fudge' illustrated. The second part of this task uses the BNC to investigate the range of such
variants.
4.1.3
Highlighted Features
This task shows you:
how to look for occurrences of a phrase using PHRASE QUERY;
how to look for variant forms of a phrase using the ANYWORD wildcard character;
how to look for alternative forms of a phrase using the QUERY BUILDER;
how to edit queries for re-use using the EDIT option on the QUERY menu;
how to sort solutions to highlight variant forms and environments using PRIMARY and SECONDARY keys.
It assumes you already know how to:

page_86
 Page 87

adjust default settings using the View Preferences option (see 1.2.8 on page 54);
find occurrences of a word using Phrase Query and Word Query (see 1.2.5 on page 51, 2.2.1 on page 64);
look up the frequency of a word in the SARA index using the Word Query Pattern option (see 3.3.1 on page 83);
adjust downloading procedure in the Too many solutions dialogue box (see 2.2.3 on page 66);
mark and thin solutions (see 2.2.4 on page 68);
sort solutions using a Primary key (see 3.2.2 on page 78).
4.1.4
Before You Start
Log on to SARA and wait for the SARA-bnc window to be displayed. Using the VIEW menu PREFERENCES option, set the default settings as follows:

400 characters
MAX DOWNLOAD LENGTH
MAX DOWNLOADS 50
FORMAT Plain
SCOPE Automatic
VIEW: QUERY and ANNOTATION checked
CONCORDANCE checked
BROWSER: SHOW TAGS unchecked

4.2
Procedure
4.2.1
Looking for Phrases Using Phrase Query
To investigate variation in the form and environment of a phrase, we must decide which features of the phrase to treat as fixed, and which as potentially
variable. All the dictionaries cited in 4.1.1 on the facing page treat 'from the horse's mouth' as a fixed sequence. Should we therefore design the query to
look for occurrences of 'from the horse's mouth', 'the horse's mouth', 'the horse's', or 'the X's mouth'? The first of these options will reduce the risk of
spurious solutions, but at the cost of excluding all variation within the sequence. The third and fourth options will capture such variation, but risk finding
many spurious solutions which have nothing to do with the idiom, such as 'the horse's bridle' or 'the patient's mouth'. In this example, we shall start from
the most restrictive option (maximizing precision), and then progressively relax it so as to increase recall.

page_87

Page 88

Formulating the Phrase Query

You have already seen how to use Phrase Query to look for a word in the corpus (see 1.2.5 on page 51). As its name suggests, Phrase Query also accepts
a sequence of words as input. Any number of words may be given, but the query cannot exceed 200 characters in total length.

Click on the Phrase Query button to display the Phrase Query dialogue box.
1
2 Type in the string from the horse's mouth
3 Click on OK (or press ENTER) and wait for SARA to download the solutions in the Query1 window.
There are 9 solutions.
4 Look through these solutions to check that they all involve metaphorical uses of the phrase, rather than
literal references to veterinary dentistry. Assuming you have checked QUERY in the VIEW PREFERENCES
option (see 4.1.4 on the preceding page), you will also see the text of the query displayed in CQL
(Corpus Query Language) format. It reads: 'from the horse\'s mouth'. The backslash
symbol indicates that the apostrophe following it has a literal, not a parenthetical function.
If you want to search for occurrences of possessive or other contracted forms (such as 'can't',
'wanna' and 'you'd've'), you must use PHRASE QUERY rather than WORD QUERY, since these forms are
 not listed in the word index, other than under their component clitics (see 2.2.1 on page 64). You
may type such forms into the Phrase Query dialogue box either as one word or else separated by
spaces, e.g. you 'd 've. A list of contracted forms which are treated as multiple L-words in the
BNC is given in the BNC Users' Reference Guide.

Looking for Variants: the Query Edit Option

Now let us look for possible variants of 'from the horse's mouth'. Rather than typing in a new query, you may find it easier to modify the current one.
You can do this using the EDIT option under the QUERY menu.
First let us remove the word 'from'.

Select QUERY then EDIT. You will be returned to the Phrase Query dialogue box, still showing the current
5 query.

6 Click on the beginning of the string and use the D ELETE key to cancel the wordfrom.
7 Click on OK to send the revised query to the server, and wait for the solutions to be downloaded. The
number of solutions displayed will increase to 20. Note that these solutions replace the previous ones
in the Query1 window. If you wish to keep the previous solutions, you should design a new query
rather than editing the current one.

page_88

Page 89

Scroll through the solutions to see what words other than 'from' appear before the query focus, and
8 whether these solutions involve literal or metaphorical uses of 'the horse's mouth'. You should find that as
well as several literal uses ('flies enter the horse's mouth'), there are also some new metaphorical ones with
prepositions other than 'from', such as 'go to the horse's mouth'.

From these 20 solutions, let us now remove the literal horses, and sort the metaphorical ones according to the words which precede them.

Double-click on the literal solutions to mark them, and use THIN then REVERSE SELECTION to remove them
9 from the list.
10 Use the SORT option to sort the remaining solutions by the left with a span of 3.

This will group together several phrasal collocates found to the left of 'the horse's mouth' 'heard it from', 'straight from', and 'go to'. Most of the solutions
appear to follow either one of these patterns or a combination of them (such as 'go straight to the horse's mouth').
Looking for Further Variants
The first word in the original phrase, the preposition 'from', thus appears to be variable. Is there any indication in the corpus that the final word of the
phrase, 'mouth', may also vary?

Select QUERY then EDIT. You will be returned to the Phrase Query dialogue box.
11
12 Click on the beginning of the input string and add the word from at the beginning, then delete the word
mouth from the end. Then click on OK to send the query to the server. 18 solutions win be displayed.

Now see whether any of these solutions involve metaphorical uses.

Sort the solutions by the right with a span of 3, in order to group them according to the words following
13 the query focus. There appear to be no metaphorical uses other than with 'mouth'.

Variants Within the Phrase: The Anyword Wildcard

We have so far considered variation on the boundaries of the original phrase: initial prepositions other than 'from', and final parts of the body other than
'mouth'. The results suggest that 'mouth' is fixed, but that 'from' varies. However we have not yet considered whether the central elements of the phrase
may also vary. For example, are there any variants of 'horse', such as 'from the cow's mouth'?
You can investigate this issue by formulating a PHRASE QUERY which contains an empty slot in the place of the word 'horse'. The slot is represented
 page_89

Page 90

by the underline character_. This is an Anyword wildcard character, which will match any L-word in the word index.

Select EDIT from the QUERY menu. The Phrase Query dialogue box will be displayed.
14
15 Type in the string the _ 's mouth (be careful to follow the underline character with a space).
More than one Anyword character may be used in the same Phrase Query, but the string cannot
begin or end with one. The underline must be preceded and followed by a space.
Searching for the Anyword character is generally faster than searching for a specific but very
frequent word. For example, the query stone _ crows is very much faster than stone
the crows, because the latter involves processing all of the several million occurrences of
'the' in the corpus. However, in the current example it is not possible to avoid doing this, since
'the' is the first word of the phrase, and an Anyword character can only be used in the middle of a
Phrase Query string.
16 Click on OK to send the query to the server. The Too many solutions dialogue box will eventually be
displayed, stating that there are 146 solutions in 116 texts.

Sorting with Dual Keys

Given the relatively large number of solutions, we will download and sort only a random selection, in order to get an idea whether variants of the
idiomatic phrase are at all frequent.

Click on RANDOM, then on OK to download 50 solutions.

17
18 Sort the solutions on CENTRE with a span of 2 as Primary key, and on L EFT with a span of 2 as
Secondary key. This will group the solutions according to the variant following 'the' in the query focus,
and order the solutions within each group according to the two preceding words. Dual sort keys are a
convenient way of sorting solutions where the form of the query focus varies.
19 Scroll through the solutions to see which words other than 'horse' appear in the query focus, and whether
any of them are also used metaphorically. There appear to be no metaphorical variants of 'horse' among
the solutions, suggesting that the sequence 'the horse's mouth' is in fact fixed. If you choose to
download a Random selection, you can always check your findings by selecting Query Edit and
repeating the operation to download a different random selection.

Results
The phrase 'from the horse's mouth' appears to allow variation of its initial preposition, while the rest of the sequence is fixed. This fixed component
appears to occur only in a limited range of environments ('straight from', 'heard it from', 'go to', etc.). While these results should be treated with caution
given the small number of occurrences of the phrase in the corpus (and hence even

page_90

Page 91

smaller numbers of the different environments mentioned), it is worth noting that none of the dictionaries cited provides such detail.
4.2.2
Looking for Phrases Using Query Builder
We now look at an idiomatic phrase which appears to permit a greater range of variation, 'a bridge too far'. Taking the lead from the Guardian's 'a
fudge too far' example, we will look for occurrences of the patterns 'a _ too far' and 'an _ too far'.
With respect to the 'horse's mouth' example, the main difficulty is designing a query to include solutions where the initial article can be either 'a' or 'an'.
The only way to do this using PHRASE QUERY would be to reduce the phrase to 'too far'; however this would generate thousands of spurious solutions,
allowing anything to precede it. To allow only specific alternatives, we must resort to a more sophisticated query tool, the QUERY BUILDER.

Click on the QUERY BUILDER button on the toolbar. The Query Builder dialogue box will be displayed.
20 You can also access Query Builder from the FILE menu by selecting NEW QUERY then QUERY BUILDER.
The Query Builder allows you to construct a complex query whose components are represented by nodes. Each query consists of two kinds of nodes:
a single scope node, and one or more content nodes, the latter being Joined together by links.
The initial dialogue box contains two nodes. The black node on the left is the scope node, and indicates the scope of the query, i.e. the unit or distance
within which a solution must be found. The default scope is any one text in the corpus, or BNC document, represented by the SGML element
<BNCDOC>. The scope of the query may either be an SGML element, or a span of anything up to 99 words. It can be edited by clicking on it:
see 6.2.1 on page 115 and 7.2.2 on page 133.
The red node on the right is an empty content node, which you must fill in with a part of the query typically a word, phrase, pattern, or SGML element.
Further content nodes can be created by clicking on the branches attached to the first one: vertical branches indicate AND links, and horizontal branches
OR links. Thus, if we fill a first content node with the word 'a', we can link it with a horizontal OR branch to a further node containing the alternative 'an',
and then link it with a vertical AND branch to further nodes containing the elements '_ too far'.
Editing the Content Node
Nodes which are displayed in red must be fined in before the query can be sent to the server.

Click on the empty red node. A menu will appear by the box offering E DIT as the sole alternative which is
21 not grayed out.

page_91

Page 92

Select EDIT. A submenu will be opened up, listing W ORD, PHRASE, POS, PATTERN, SGML and A NY as
22 alternatives.

23 Select WORD. The Word Query dialogue box will be displayed. You can use this box to select a word or
list of words to form the content of the current node, in exactly the same way as you used it to formulate
Word Queries in 2.2 on page 64. The options in the Content node menus are fully described in 1.3.7
on page 207. They allow you to DELETE the node, to EDIT its content, or to CLEAR its current
content and change its type, as well as to COPY its content to the clipboard, or to PASTE from the
clipboard into it. The following types may be selected: Word, Phrase (See 4.2.1 on page 89), POS
(see 7.2.1 on page 131), Pattern (see 8.2.2 on page 150), SGML (see 5.2.2 on page 101), and Any
(the ANYWORD underline symbol).
24 Type in a, then check the PATTERN box to limit matching to this word. You must check the Pattern box
here, as otherwise LOOKUP will try to list all the words in the word index which begin with 'a',
which would exceed the permitted maximum of 200 (see 2.2.1 on page 64).
25 Click on LOOKUP to display the matching word list. The word 'a' will be displayed.
26 Click on 'a' to select it, then click on OK to insert this selection in the Query Builder node. You will be
returned to the Query Builder dialogue box. You will see that the content node is now black, showing
that it is no longer empty. It contains the string "a", which is the CQL representation of this part of the
query.

Linking Nodes Horizontally: OR

Let us now create a second content node containing "an" as an alternative to "a".

Click on the horizontal branch of the 'a' node. A new red content node will be created at the end of the
27 branch.

28 Click on this node and select EDIT then WORD to display the Word Query dialogue box once more.
29 Make sure the PATTERN box is checked, then type in the stringan and click on LOOKUP.
30 Click on 'an' in the matching words list to select it, then on OK to insert it in the Query Builder node. You
will be returned to the Query Builder dialogue box. The second content node is now also black, and
contains the string "an". The two nodes Jointly represent the disjunction 'a OR an'.

Linking Nodes Vertically: AND

We now need to add the remaining part of the query, namely '_ too far'.
 Click on the downward vertical branch of the 'a' node. A new red content node will be added to the
31 branch, and the branch will become a

page_92

Page 93

downwards arrow. The arrow indicates that the link between the two nodes has the value ONE-WAY, i.e.
that the content of the second node must follow that of the first, not necessarily directly.
32 Click on the downward vertical branch of this new node. A further empty red node will be added to the
previous one, connected by a further one-way link.
33 Click in the first red node, and select EDIT then ANY. The Anyword symbol (_) will be placed in the node,
showing that the content of this node may be any L-word. The node will remain red, even though it has
been filled, but a message will be displayed reminding you that an Anyword symbol must be placed
between two NEXT links (see below).
34 Click in the remaining red node, and select EDIT then PHRASE. The Phrase Query dialogue box will be
displayed.
35 Type in the phrase too far and click on OK to insert it in the Query Builder node. You will be
returned to the Query Builder dialogue box.

One aspect of the query remains to be specified: how exactly are the various content nodes to be linked? The ONE-WAY links inserted by default mean
that the content of the upper node must precede that of the lower, but may be at any distance from it (within the specified scope for the entire query
here, an entire BNC document). As they concern adjacent components of a single phrase, we want the three successive content nodes to be adjacent to
one another. So we must change the links between the nodes.

Click on the first One-way link, then on LINK TYPE. The Link type menu will be displayed, offering a
36 choice of NEXT, ONE-WAY, and TWO-WAY.

37 Click on NEXT. The downward arrow will be replaced by a thick vertical line representing a NEXT link,
meaning that the content of the second node must directly follow that of the first.
38 Click on the second link, and change its value to NEXT in the same way. All the nodes should now be
black. If the constructed query is syntactically correct, the message 'Query is OK' will be displayed at
the bottom of the box. Clicking on a link enables you either to change its type, or to insert an
additional node. A NEXT link, indicated by a thick line, means the content of the lower node
directly follows that of the upper node, without any intervening words or punctuation. A ONE-WAY
link, indicated by a downwards arrow, means that the content of the lower node follows that of
the upper node, but is not necessarily adjacent to it (see 5.2.4 on page 105). A TWO-WAY link,
indicated by a bidirectional arrow, means that the contents of the two nodes may occur in either
order, not necessarily adjacently (see 8.2.3 on page 152). The ANYWORD symbol can only be used in
a node between two NEXT links.

page_93

Page 94

If you INSERT a node, the link between the Upper node and the new one maintains the previous
value, while the one between the new node and the lower one is assigned the default value ONE-
WAY. However, you must ensure that all nodes are joined by links of the same type in the final
query. For a full description of links in Query Builder, see 1.3.7 on page 207.
39 Click on OK to send the query to the server. There are 36 solutions from 32 texts.

Saving the Query

The full text of the query will be displayed above the solutions in CQL format. It should read ("a" / "an") (_) ('too far'). Given the
work involved in constructing complex queries using Query Builder, it is a good idea to save such queries for future reference. Even if you do not want
to repeat exactly the same query, you may want to examine or edit it in order to construct a similar one.

Select FILE then SAVE (or click on the SAVE button on the toolbar). Assuming that the solutions are
40 displayed in a window entitled Query2, SARA will propose to save the query to your working directory
with the name query2.sqy. You are advised to choose a more memorable filename: SARA will add
the .sqy extension automatically.

Examining the Solutions

First, let us sort the solutions according to the word occupying the variable slot, so as to identify any which do not involve variants of the idiomatic
phrase 'a bridge too far'.

Select QUERY then SORT to display the Sort dialogue box.

41
42 Select CENTRE with a span of 2 as Primary key.
43 Click on OK to carry out the sort.

There are two main types of solutions which seem spurious. The first is where the variable is a unit of measure with the literal sense of physical distance
('an inch/a foot/a centimetre /a mile too far'), rather than the metaphorical one of over-ambitiousness. The second is where the noun in the variable slot
functions as object rather than as an adverbial quantifier: 'taking a hobby too far', 'drive a man too far', 'relying on a person too far away', etc.
First, thin the solutions to remove these spurious cases.

Mark all those solutions which do not have the metaphorical sense by double-clicking on them.
44
45 Select THIN and REVERSE SELECTION under the QUERY menu to remove the marked solutions from the
display.

page_94

Page 95

You should now have about 25 solutions containing 'a bridge too far' or variants of it. Near the top of the list there are two instances where 'a _ too far'
occurs in quotation marks (the collating sequence places quotation marks at the beginning of the alphabet). Both are of 'a bridge too far', in the
metaphorical sense, confirming that this is the citation Corm of the phrase. Of the remaining occurrences, several refer to the film or to the historical
event behind it, while others pun on the literal sense of 'bridge'. However there are also a number of solutions where other geographical features,
physical and even social artifacts are metaphorically invoked with the sense of overambitiousness: 'bomb', 'fair', 'party', 'peak', 'ridge', 'toll', 'treaty'. With
the exception of 'treaty', which occurs in a book title mentioned three times, all these forms occur once only in the corpus. This suggests that they are
creative rather than conventionalized variants of the basic phrase.

4.3
Discussion and Suggestions for further Work
4.3.1
Looking for Variant Phrases
This task has introduced you to a number of ways in which you can look for variant forms of a phrase using SARA.
In PHRASE QUERY or QUERY BUILDER, variants of an element within the phrase can be found by using the ANYWORD symbol _ in the relevant slot, while in
Query Builder, specific alternatives can also be expressed as alternative content nodes. In either case, the SORT option allows you to group similar
solutions.
Variants of the first or last element of the phrase are most easily identified by excluding that element from the query itself, and then sorting the solutions
by the words to the left or right of the query focus. Sorting solutions by several words to the left or to the right also highlights recurrent environments in
which a phrase is used.
These strategies are appropriate where the phrase can be defined as a fixed sequence of components as is generally the case with sayings, proverbs and
the like. To find variants of phrases where the form and sequencing of components can vary (such as active and passive forms of phrasal verbs:
compare 'break ranks' and 'ranks were broken'), you must use more complex query techniques; for some examples, see 8.1.2 on page 144, 8.1.3 on
page 145.
4.3.2
Some Similar Problems
Silver Linings
When proverbs are cited in English, they are frequently varied or curtailed with respect to their canonical form. Formulate a Phrase Query for 'silver
lining', and sort the solutions to see how often it is preceded by 'every cloud has a'. Then compare this with the collocation frequency of 'cloud' within a
span of 5, using the COLLOCATION option (see 3.2.1 on page 76). 'Silver lining'

page_95

Page 96

occurs 53 times in the corpus, and is preceded by 'every cloud has a' in only 7 of these cases. On the other hand 'cloud' occurs as a collocate
of 'silver lining' 21 tittles in a span of 5 (collocation ratio 0.40), showing that variation of the proverb is more common than exact citation.
To Be or Not to Be
How frequent are quotation and variation of Hamlet's ''to be or not to be''? What verbs can replace 'be' in this saying? Is the quotation typically
continued with the phrase 'that is the question' or some variant of it? As in the 'horse's mouth' example (see 4.2.1 on page 87), there is a trade-off
to be made between precision (minimizing the number of spurious solutions) and recall (finding all the relevant solutions). The Phrase Query
to _ or not to finds (after a considerable time) 78 solutions. Note however that this excludes any variant using phrasal verbs, such as
'to give up or not to give up'. To include these in a Phrase Query, you would have to simply use the string or not to, which would find
very many spurious solutions. (A possible alternative would be to use the SPAN scope option in QUERY BUILDER to specify that a variable number
of words can appear between 'to' and 'or not to': see 8.2.3 on page 150).
Sort the solutions using CENTRE with span 2 as Primary key, and RIGHT With span 2 as Secondary key. This will group the solutions according to
the verbs before and after 'or not to'. Remove any spurious solutions with THIN. You will find that you cannot use the COLLOCATION option (see
3.2.1 on page 76) to find how often 'question' collocates with 'to be or not to be', because if it appears it will do so more than 9 words after the
start of the query focus.
Dirty Looks
The phrase 'give someone a dirty look' would seem to offer numerous possibilities of variation. As well as allowing varying verb tenses, and any number
of recipients, it can be pluralized (you can 'give someone dirty looks') and it may be used with other verbs than 'give'. Design a query to find occurrences
of 'dirty look(s)', see what verbs occur with it, and whether these are used with both the singular and plural forms. Use QUERY BUILDER to construct a
query where a content node containing 'dirty' is joined by a NEXT link to alternative nodes containing 'look' or 'looks'. You can specify these
three nodes with either WORD QUERY or PHRASE QUERY. Sort the solutions according to the two words of the focus as Primary key and according
to the three words to the left of it as Secondary key.
As well as people giving (a) dirty look(s), their behaviour can earn them (or it).
And Finally ...
Tired of working on linguistic problems? Try using the BNC as an oracle! Type in the Phrase Query what's for dinner? (including the
question mark), and see what suggestions the corpus provides. To see the answers as well as the questions, which may occur in different
utterances or paragraphs, you should select Page mode display and expand the context to the maximum by double-clicking on

page_96

Page 97

the right mouse button (see 1.2.6 on page 51). You may have to browse extensively in the source text (see 1.2.7 on page 52) to find out what
'curried Geraldo' is, to go with all the broccoli.

page_97

Page 98

5
Do People Ever Say 'You Can Say That Again'?
5.1
The Problem: Comparing Types of Texts
5.1.1
Spoken and Written Varieties
The BNC provides a wide range of information on the origin and nature of its component texts. Most of this information is encoded in the <HEADER>
to each text, which contains, for example, the bibliographic information displayed using the SOURCE option (see 1.2.7 on page 52). Every text header
also includes a category reference or <CATREF> element, which categorizes that text according to a series of parameters such as:
type: whether written, spoken, or written-to-be-spoken (as in the case of drama or broadcasting scripts);
for written texts: author type (individual, multiple, corporate), author characteristics (age, sex, origin), region of publication, target audience, circulation,
cultural level, subject domain, medium (book, periodical, unpublished, etc.);
for spoken texts: interaction type (monologue or dialogue), speaker characteristics (age, sex, origin, class), region of recording, and, for the non-
conversational or context-governed texts (see 6.1.2 on page 112), subject domain.
SARA treats these parameters as attributes of the <CATREF> element, each of which may have a particular value. For instance, a book for children
would have the value child for the attribute WRITTEN_AUDIENCE. By searching for instances of <CATREF> elements which have certain attribute
values, you can use SARA to find all the texts in the corpus of a certain type. (A list of <CATREF> attribute values is provided in section 5.2.3 on
page 103.)
This information can also be used to investigate and compare the use of specific words or phrases in particular text-types. This task asks how far
somewhat clichéd expressions which are generally considered to belong to an informal spoken register, actually do occur in real spoken dialogue. The
first example inspects instances of 'You can say that again' in the corpus to see in what kinds of texts they appear, and then compares their frequency in
real spoken dialogue with that in the imaginary written dialogue of plays, stories and novels. The second instead uses <CATREF> attribute values to
compare the use of the phrase 'Good heavens' in spoken and written texts.
5.1.2
Highlighted Features
This task shows you:

page_98

Page 99

how to use the SGML QUERY option to count texts where the <CATREF> element has particular attribute values (for instance those texts written for
child audiences in the domain of science);
how to count numbers of solutions without downloading them, using the MAX DOWNLOADS option as a filter;
how to look for words or phrases in texts which have particular <CATREF> attribute values using ONE-WAY links in the QUERY BUILDER;
how to display SGML markup in solutions, using SGML FORMAT under QUERY OPTIONS.
It assumes you already know how to:
adjust the default download and display settings using the View menu Preferences option (see 1.2.8 on page 54);
use Word Query and Phrase Query to search for words and phrases (see 2.2.2 on page 65, 4.2.1 on page 87);
design complex queries with multiple content nodes using the Query Builder (see 4.2.2 on page 91);
adjust downloading procedure in the Too many solutions dialogue box (see 2.2.3 on page 66).
5.1.3
Before You Start
Using the VIEW menu PREFERENCES option, set the SARA defaults as follows:

400 characters
MAX DOWNLOAD LENGTH
MAX DOWNLOADS 10
FORMAT Plain
SCOPE Paragraph
VIEW: QUERY and ANNOTATION checked
CONCORDANCE checked
BROWSER: SHOW TAGS unchecked

5.2
Procedure
5.2.1
Identifying Text Type by Using the Source Option
The Longman Dictionary of Contemporary English lists 'you can say that again' as a spoken phrase, used to indicate complete agreement. But does
it in fact occur in real speech? Let us begin by using PHRASE QUERY to find all its occurrences in the corpus.

page_99

Page 100

Click on the PHRASE QUERY button on the toolbar to display the Phrase Query dialogue box.
1
2 Type in the string you can say that again and click on OK to send the query to the server.
The Too many solutions dialogue box will be displayed, showing that 'you can say that again' occurs 32
times in 26 texts.
3 As we are only interested in the kinds of texts in which 'you can say that again' appears, check O NE PER
TEXT and then click on the DOWNLOAD ALL button to change the download hits number to 26. Then click on
OK to download these solutions. You will see that the vast majority occur within quotation marks. This
suggests that they come from fictional dialogue, since quotation marks are not used in spoken texts in the
corpus.
4 Select SORT from the QUERY menu and sort the solutions by the Centre with a span of 1 as Primary key,
and by the left with a span of 1 as Secondary key. The solutions in which the focus begins with quotation
marks will be grouped near the top of the display.
5 Scroll through the remaining solutions, and double-click on those which seem to you to come from actual
speech. Then select THIN and SELECTION from the QUERY menu to delete the rest.
6 Using the SOURCE button on the toolbar, check the bibliographic data to see how many of these solutions
actually do come from spoken texts. The bibliographic data displays for written and spoken texts have
different formats, the latter including a window showing a list of the participants in the interaction.
7 Click on OK to close the Bibliographic data box.

Only 3 out of the 26 texts in which 'you can say that again' appears are actually spoken ones, while 23 are written. To interpret these figures, however,
we need to take into account the relative frequencies of the relevant written and spoken texts in the corpus. We can do this by finding out how often the
values of attributes on the <CATREF> element classify the text it heads as spoken dialogue or imaginative writing. While written dialogue is not, of course,
found exclusively in imaginative texts, a glance at the solutions would suggest that this is a reasonable approximation in this case.
5.2.2
Finding Text-type Frequencies Using SGML Query
The SGML Q UERY option allows you to find:
instances of a particular SGML element;
instances of a particular SGML element with a particular attribute or attributes having a particular value or values.

page_100

Page 101

Here we shall use it to count the number of texts in a particular category in the corpus. As was explained in 5.1.1 on page 98, we can do this by finding
all the instances of <CATREF> elements which have particular attribute values, and counting the number of solutions.
Using the Max Downloads Option to Obtain Numerical Information
Where we are only interested in finding out the number of solutions to a query, rather than in inspecting them, it is not necessary to download the
solutions from the server. Instead, we can simply use the information in the TOO MANY SOLUTIONS dialogue box, which reports the number of solutions
found. This dialogue box is displayed when the number of solutions exceeds the number specified for the MAX DOWNLOADS option under VIEW
PREFERENCES (see 1.2.8 on page 54). If this number is small (you should have set it to 10 for this task in 5.1.3 on page 99), the Too many solutions
dialogue box will appear in response to almost all queries. You can then simply read off the number of solutions from the dialogue box, and click on
CANCEL to abort the query without actually downloading any solutions from the server.
Formulating the SGML Query
Let us find out how many imaginative written texts there are in the corpus, and how many spoken dialogue ones. A comparison of the numbers of texts
in these two categories will provide a rough yardstick against which to evaluate the difference in the frequencies of 'you can say that again'. Most
written texts in the BNC are of roughly similar length (40,000 words) regardless of their type. Spoken texts from the demographically
organized part of the corpus are of a similar average length, but with more variability. Spoken texts from the context-governed part of the
corpus are on average very much shorter (8,000 words); however only 60% of these involve dialogue. The BNC Users' Reference Guide
provides full details of the numbers of texts, sentences and words in the main categories of texts in the corpus: for summary figures, see 2.1.1
on page 31.

Click on the SGML Q UERY button on the toolbar (or select NEW QUERY then SGML from the FILE menu).
8 The SGML dialogue box will be displayed.
9 Check the SHOW HEADER TAGS box. The dialogue box contains an alphabetical list of all the types of
SGML elements used in the BNC. If the SHOW HEADER TAGS box is not checked, the list will be much
shorter, containing only those elements which form part of actual texts, and excluding those
which are only found in text headers. As the <CATREF> element appears in text headers, you must
use the full list here. This handbook describes some of the most important element types; you will
find a complete list in the BNC Users' Reference Guide.
10 Scroll through the list until you find the <CATREF> element, and click on it to select it. A brief description
of the element will be displayed to the

page_101

Page 102

right of the list. The list of attributes for the element will be displayed in the bottom left-hand window.

SARA treats the <CATREF> element as having a large number of possible attributes, whose names indicate whether they are applicable to all, to
spoken, or to written texts. They are listed with their values in section 5.2.3 on the facing page.

Scroll through the list of attributes until you find SPOKEN_TYPE, and click on it to select it.
11
12 Click on the ADD button to include this attribute in the query. The ATTRIBUTE dialogue box will be
displayed, showing possible values for this attribute.
13 Click on the value dialogue to select it, then on OK to insert it in the query. You will be returned to
the SGML dialogue box. Your selected attribute value pair is now displayed in the right-hand window of
the box. Where attributes permit wide ranges of values, you must type in the required value in a
modified version of the Attribute dialogue box (see 6.3.2 on page 126, 9.2.4 on page 170, 10.2.3
on page 182 for examples).
14 Click on OK to send the query to the server. The Too many solutions dialogue box will be displayed,
showing the number of solutions to this query. Since there is only one <CATREF> element per text, the
number of texts is the same as the number of solutions.
15 Make a note of the number of solutions, and click on C ANCEL to return to the SGML dialogue box. The
box will still display your previous query, with the <CATREF> element highlighted.

Now find the number of imaginative written texts.

Scroll through the attribute list till you find the WRITTEN attributes.
16
17 Click on WRITTEN_DOMAIN to select it, then on the ADD button. The Attribute dialogue box will be
displayed, showing the values for this attribute in the BNC.
18 Click on imaginative to select it, then on OK to insert this attribute value pair in the SGML
dialogue box. Your new attribute value selection will be added to the right hand window of the box.
19 Remove the previous attribute value selection by clicking on it to select it, then clicking on REMOVE.
20 Click on OK to send the query to the server. The Too many solutions dialogue box will be displayed,
stating that there are 625 solutions in 625 texts.

page_102
 Page 103

The proportion of texts in each category containing one or more instances of 'you can say that again' is thus:
imaginative written texts 21/625
spoken dialogue texts 3/654
This suggests that texts with 'you can say that again' are more frequent in the 'imaginative written' category, though the figures are too small for reliable
generalization.
5.2.3
Displaying SGML Markup in Solutions
To see how <CATREF> attributes appear in the BNC, you may like to look at some examples containing the
written_domain="imaginative" attribute value pair.

In the Too many solutions dialogue box, click on OK to download the first 10 solutions.
21

If you look at the solutions, you will find that the <CATREF> element is not displayed. The query focus is represented only by a vertical red line. This is
because we have set the default display as PLAIN (see 5.1.3 on page 99), which only shows the words and punctuation in the text. To display SGML
markup, such as <CATREF> elements, you must choose a different display format.

Select OPTIONS under the QUERY menu. The QUERY OPTIONS dialogue box will be displayed.
22
23 Select SGML as format, then click on OK and wait for the solutions to be re-displayed. You will now
see the solutions with their complete SGML markup, shown as tags enclosed in angle brackets. The
<CATREF> element is highlighted as the query focus in each concordance line. The QUERY OPTIONS
dialogue box allows you to change the display values of the current query with respect to the
defaults selected under VIEW PREFERENCES. The options available are identical to those listed under
FORMAT and SCOPE in the User preferences dialogue box under VIEW PREFERENCES (see 1.2.8 on page
54), except that the Query Options dialogue box also allows you to edit the CUSTOM display format
using the CONFIGURE option (see 9.2.2 on page 165 for an example).

Assuming you checked the QUERY box under VIEW PREFERENCES, the Query Text will be displayed above the solutions. This shows the text of the query
in CQL (Corpus Query Language) format: <catRef target=wriDom1>.

page_103

Page 104

If you look at any solution to this query in SGML format, you will see that the < CATREF> start-tag actually contains a long string of codes. Each of these
codes is made up of a sequence of letters and a number. The letters are used to indicate such features as the text availability (allAva), its type
(allTy), its target audience (allAud), its cultural level (wriLev), its domain (wriDom), etc., while the number indicates a value for this
attribute. For example, wriDom1 represents the first of the WRITTEN-DOMAIN values ('imaginative'). A complete list of the < CATREF> codes used in the
corpus is given in 2.2 on page 234. These codes are translated into more comprehensible attributes and values by the SARA software, to make
selection easier. The following table shows the translations of these codes into attributes, and the values which can be selected for them in the
Attribute dialogue box. Each attribute can also take a value indicating 'information not available'.
Attribute Values
ALL_AVAILABILITY free; restricted (to various areas)
ALL_TYPE spoken demographic; spoken context-governed; written books and periodicals; written-to-be-
spoken; written miscellaneous; spoken; written; unclassified
SPOKEN_AGE (of demographic respondent) under 15; 15-24; 25-34; 35-44; 45-59; 60 or over
SPOKEN_CLASS (of demographic respondent) AB; C1; C2; DE
SPOKEN_DOMAIN (context-governed) educational/informative; business; public/institutional; leisure
SPOKEN_REGION south; midlands; north
SPOKEN_SEX (of demographic respondent) male; female
SPOKEN_TYPE monologue; dialogue
WRITTEN_AGE (of author) under 15; 15-24; 25-34; 35-44; 45-59; 60 or over
WRITTEN_AUDIENCE child; teenage; adult; any
WRITTEN_DOMAIN imaginative; natural and pure sciences; applied sciences; social science; world affairs; commerce
and finance; arts; belief and thought; leisure
WRITTEN_DOMICILE (of author) country or region
WRITTEN_GENDER (of target audience) male; female; mixed; unknown
WRITTEN_LEVEL (of circulation) low; medium; high
WRITTEN_MEDIUM book; periodical; misc published; misc unpublished; to-be-spoken
WRITTEN_PLACE (of publication) country or region
WRITTEN_PUBSTATUS published; unpublished
continued on next page

page_104

Page 105

Attribute Values
WRITTEN_SAMPLE whole text; beginning sample; middle sample; end sample; composite
WRITTEN_SELECTION selective; random
WRITTEN_SEX (of author) male; female; mixed; unknown
WRITTEN-STATUS (of reception) low; medium; high
WRITTEN_TIME 1960-1974; 1975-1993
WRITTEN_TYPE (of author) corporate; multiple; sole; unknown

5.2.4
Searching in Specific Text-Types: 'Good Heavens' in Real and Imagined Speech
The QUERY BUILDER option allows you to combine not only searches for different words or phrases (see 4.2.2 on page 91), but also different query
types. Thus you can join the SGML Query formulated in the last section with a Word or Phrase Query, so as to find all the cases where a word or
phrase occurs in texts which have a particular <CATREF> attribute value or values.
Like 'you can say that again', the expression 'good heavens' seems more typical of imagined than of real contemporary speech. As you can discover by
using PHRASE QUERY, it occurs 140 times in the corpus, making it too frequent to simply inspect the solutions and their sources with any comfort. We can
instead use the QUERY BUILDER to restrict the search to occurrences in imaginative written texts, and then to occurrences in spoken dialogue texts, before
comparing their frequencies.
'Good Heavens' in Imaginative Written Texts: Joining Different Query Types
Let us begin by designing a query to find the frequency of 'good heavens' in texts which fall into the imaginative written category.

Click on the QUERY BUILDER button to display the Query Builder dialogue box.
24
25 Click in the content node (displayed in red), and select EDIT then SGML. The SGML dialogue box will
be displayed.
26 Scroll through the list of elements and click on < CATREF> to select it.
27 Scroll through the attribute list and select WRITTEN_DOMAIN.
28 Click on the ADD button to display the Attribute dialogue box.
29 Select imaginative, then click on OK to insert it in the SGML Query.
30 Click on OK to return to the Query Builder. The content node will now contain the specification
<catRef target=wriDom1>. The angle brackets indicate that this part of the query regards an
SGML element.

page_105

Page 106

Click on the downwards branch to add a new empty node to the query.
31
 32 Click in this new node and select EDIT then PHRASE. The Phrase Query dialogue box will be displayed.
33 Type in the string 'good heavens' and click on OK to insert it in the Query Builder node. You will see that
the second content node now contains the string 'good heavens', representing this Phrase Query.

We now need to consider the relationship between these two nodes, and the scope within which the two nodes must be satisfied. At the moment, they
are joined by a downwards arrow, representing the link value ONE-WAY, meaning that the content of the second node follows that of the first within the
scope of the query, not necessarily directly (see 4.2.2 on page 91). The scope of the query is any <BNCDOC> element. For this query, both these
default values are correct. The <CATREF> element always occurs in the text header, and this precedes the text, in which the phrase 'good heavens' can
occur at any point. The scope must contain both the text header and the text: the only element to do so is <BNCDOC>.

Check that the 'Query is OK' message is displayed at the bottom of the box, and click on OK to send
34 the query to the server. The Too many solutions dialogue box will be displayed.

35 Read off the number of solutions and the number of texts in which they occur. There are 90 solutions in
64 texts.
36 Change the DOWNLOAD HITS number to 30 and download a random set of the solutions. You will see that
the vast majority are in quotation marks, confirming that they come from fictional dialogue.Note that the
query focus is the phrase 'good heavens': in Query Builder the query focus is always the final
content node, or the final group of nodes where these are joined by NEXT links.

'Good Heavens' in Spoken Dialogue: Using Query Edit

Let us now look at the use of 'good heavens' in spoken dialogue. The easiest way to do this is to edit the current query, specifying
spoken_type="dialogue" instead of written_domain="imaginative".

Select EDIT under the QUERY menu. You will be returned to the Query Builder. This will still show the
37 previous query.

38 Click on the <CATREF> content node and select EDIT. The SGML dialogue box will be displayed,
showing your previous <CATREF> attribute value selection.
39 Click on REMOVE ALL to remove the previous attribute value selection.

page_106

Page 107

Select SPOKEN_TYPE from the attribute list.

40
41 Click on the ADD button to display the Attribute dialogue box.
42 Select dialogue from the value list, then click on OK. You will see that the right hand window of the
box now shows the attribute value pair spoken_type="dialogue".
43 Click on OK to insert this SGML Query in the Query Builder node. The node will now contain the string
<catRef target=spoLog2>. spoLog2 corresponds to the second defined value for the
SPOKEN_TYPE attribute, that of dialogue; see 2.2 on page 234.

44 Check that the 'Query is OK' message is displayed, then click on OK to send the query to the server.
45 Read off the number of solutions and the number of texts from the T OO MANY SOLUTIONS dialogue box.
There are 32 solutions in 20 texts.
46 Click on the DOWNLOAD ALL radio button, then on OK, to download all the solutions.

5.2.5
Comparing Frequencies in Different Text-Types
We now know that there are only 32 occurrences of 'good heavens' in 20 spoken dialogue texts, while there are 90 in 64 imaginative written texts.
To compare the relative frequency of 'good heavens' in each of the above text categories, however, we need to choose an appropriate unit for
comparison. One possibility is to compare numbers of occurrences per text for each text-type, as we did in 5.2.2 on page 100. Given the variability of
the length of texts in the corpus, however, a more satisfactory way may be to compare numbers of occurrences per million words or per million
sentences.
The BNC Users' Reference Guide lists the numbers of words and sentences for the main text categories in the corpus. These are also summarized in
2.1.1 on page 28 of this Handbook. Figures for the categories of interest here are as follows:
imaginative written
1,580,771 sentences, 19,664,309 words. The 90 occurrences of 'good heavens' in this text type are thus equivalent to 56.9 per million sentences, or
4.6 per million words.
spoken dialogue
888,535 sentences, 7,760,753 words. The 32 occurrences of 'good heavens' in this text type are thus equivalent to 36.0 per million sentences, or 4.1
per million words.
By both criteria, 'good heavens' is rather more common in imaginative written texts than it is in actual speech. This difference is the more striking because
written dialogue forms only a small part of most imaginative written texts. However, for the same reason it makes little sense to test the difference

page_107

Page 108

statistically. If data for the category you are interested in are not provided in the BNC Users' Reference Guide, you can always find out the
number of sentences in a particular text category using the Query Builder. If you place a <CATREF> specification in a first content node, and a
<S> (sentence) element in a second content node, and join the two nodes with a ONE-WAY link, the query will eventually find all the sentences in
texts matching the <CATREF> specification, and you can read off their number from the Too MANY SOLUTIONS dialogue box. You cannot count
words in this way, because SARA does not allow you to formulate SGML queries using the <W > (word) element.
5.3
Discussion and Suggestions for Further Work
5.3.1
Investigating Other Explanations: Combining Attributes
There are a number of other possible explanations which come to mind for the difference in the frequency of 'good heavens' in these two text types.
One is historical change: the written component of the BNC is, on average, rather older than the spoken component. While all the spoken texts in the
corpus were collected after 1990, you may have noticed that the list of <CATREF> attributes (see 5.2.3 on page 103) includes WRITTEN_TIME, which
has 19601974 and 19751993 as values. It is thus possible that use of 'good heavens' is concentrated in the older written texts, since when it has
progressively been falling into disuse. For the same reason, it is possible that in spoken dialogue, good heavens' is only used in texts involving older
speakers.
We can check to see whether time and age are relevant variables by designing further queries using the <CATREF> element, and comparing the
number of occurrences of 'good heavens' with the number of sentences in each case. To do this, you need to repeat the procedure used in the last
section, this time designing queries where multiple attribute values are specified for the <CATREF> element, for instance wriDom1 (i.e. 'written
domain: imaginative') and wriTim1 (i.e. 'written time: 19601974').

Start a new query using QUERY BUILDER, and click on the empty content node.
47
48 Select EDIT then SGML.
49 In the SGNM dialogue box, select <CATREF> from the list of elements and WRITTEN_DOMAIN from the
list of attributes, then click on ADD to display the Attribute dialogue box.
50 Select imaginative from the list of values, and click on OK to insert the attribute value pair in the
SGML dialogue box.
51 Now select WRITTEN_TIME from the attribute list, and click on ADD to re-display the Attribute dialogue
box.

page_108

Page 109

Select 196074 from the list of values, and click on OK to add this attribute value pair to the query.
52
53 Click on OK to return to the Query Builder. You will see that the node now lists two attribute value pairs
for the <CATREF> element.
54 Create a second content node below the first containing the Phrase Query 'good heavens', and join the
 two nodes with a ONE-WAY link.
55 Check that the 'Query is OK' message is displayed, and click on OK to send the query to the server.
56 Wait for the solutions to be downloaded and read off their number from the box on the status bar. This
corresponds to the number of occurrences of 'good heavens' in older written imaginative texts.
57 Select QUERY then EDIT. You will be returned to the Query Builder.
58 Click on the bottom content node and select CLEAR then SGML. You must use the CLEAR option here,
since the EDIT option only allows you to change the content of the query specified for the node,
not its query type.
59 Select the <S> element and click on OK to insert it in the Query Builder node.
60 Click on OK to send the query to the server.
61 Read off the number of solutions from the Too many solutions dialogue box. This corresponds to the
number of sentences in older written imaginative texts.

If you now calculate the frequency of 'good heavens' per million sentences in this text category, you will find that it is rather greater than the frequency
for all imaginative written texts which we calculated in 5.2.5 on page 107, suggesting that there may be a historical decline in its use.
To assess the impact of speaker age, edit the <CATREF> node to include values for both the SPOKEN_TYPE and SPOKEN_AGE attributes. The latter indicates
the age of the respondent who recorded the text in question. If you design a query to look for occurrences of 'good heavens' with different values for
this attribute, you will find that there are only two occurrences in dialogue where the respondent's age is under 35. This suggests that 'good heavens' may
be a feature of speech involving older people. While the <CATREF> element only indicates the age of the respondent (the person who recorded
the dialogue), the next task will show you how to find out the age (and sex, class, dialect, education, etc.) of the speaker of a particular
utterance (see 6.2 on page 115), and hence how often 'good heavens' is used by speakers from particular age groups (see 6.2.4 on page 123).

page_109

Page 110

5.3.2
Some Similar Problems
Talking About Politics
At the time the tapes for the spoken component of the BNC were recorded, John Major was first Chancellor of the Exchequer, and then Prime
Minister, and Nell Kinnock the Leader of the Opposition. Which of the two men is more frequently mentioned in spoken dialogue? Are they referred to
in similar terms? Use the QUERY BUILDER to find all the occurrences of 'Major'. Join an SGML Query for the <CATREF> element where the
SPOKEN_TYPE attribute has the value dialogue to a Phrase Query for 'Major', unchecking IGNORE CASE in the Phrase Query dialogue box in
order to make the search for 'Major' case-sensitive. Download all the solutions, and sort them according to the 2 words to the left in order to
check that they do in fact refer to John Major rather than Major Barbara. Mark and thin any spurious solutions, then scroll through the
remaining ones to identify adjectives used to describe Major. Then do a similar query for 'Kinnock'.
There are many more references to Major (over 120 as compared to 35), who is variously described as 'bloody', 'blooming', 'bumblinq', while
Kinnock is referred to as 'a funny' and 'a daft Welsh git'. Notwithstanding this complimentary tone, both are almost always referred to as 'Mr'
or with their first name.
Spoken Bodies and Written Ones
In their Comprehensive grammar of the English language, Quirk et al (1985: 6.4) state that pronominal forms with'one' ('someone', 'anyone', etc.)
are more commonly used than forms with 'body', citing frequencies in the LOB and Brown corpora of written English in the 1960s. Is this claim borne
out by the BNC? And is it also true of spoken English? Compare the frequencies of 'everyone' and 'everybody' in the spoken and in the written texts of
the BNC. Which of the two forms is more frequent in the category of 'written-to-be-spoken' texts? Use the QUERY BUILDER to join an SGML Query
for the text-type to a Word or Phrase Query for 'everyone' or 'everybody'. As we are only interested in numbers here, you can abort the query
from the Too many solutions dialogue box without downloading the solutions: pressing CANCEL will then return you to the Query Builder, where
you can edit the query to change the word or text-type. You can access the category WRITTEN-TO-BE-SPOKEN using either the ALL-TYPE or the
WRITTEN_MEDIUM attribute on the <CS:"small caps">CATREF> element.

The following tables show the actual frequencies found for these two words in the BNC. First, within the 3209 written texts, totalling
89,740,544 words:
form occurrence per text per million words
everybody 3317 1.03 40.0
everyone 12110 3.77 134.9
 page_110

Page 111

For the same two terms, within the 915 spoken texts, totalling 10,365,464 words:
form occurrence per text per million words
everybody 2767 3.02 266.9
everyone 1227 1.34 118.4

These figures make it very clear that the two forms are differently distributed in speech and in writing. While there are too few written-to-be-
spoken texts to permit confident generalization, the difference in the frequencies of the two forms parallels that of writing rather than of
speech (everybody' 25; 'everyone' 213).
Talking Metric
Since the early 1970s, Britain has in theory been 'going metric', with a gradual changeover to international standards of weights and measures. What
evidence is there in the BNC that people now actually talk about 'metres' rather than 'yards', about 'litres' rather than 'pints', or about 'grams' (or
'grammes') rather than 'ounces'?
Compare the frequencies of some imperial and metric measurement terms in spoken dialogue (< CATREF> attribute value
spoken_type="dialogue") and in written texts which have a high level of circulation (< CATREF> attribute value written_level=
"high"). You should include the singular and plural form of the measurement term as alternative nodes in each query (see 4.2.2 on page 91).
Note that it would be impossible to do this query sensibly for the measures foot' or 'pound', given the many other uses of these words.
The only case where a metric measure is more frequent than its imperial equivalent is 'gram', which is more frequent than 'ounce' in spoken
dialogue. Even here, however, it occurs in fewer texts. You might also like to compare the relative frequencies of these words in spoken texts
involving respondents from different age groups.

page_111

Page 112

6
Do Men Say Mauve?
6.1
The Problem: Investigating Sociolinguistic Variables
6.1.1
Comparing Categories of Speakers
In a much-cited book, Language and woman's place, Robin Lakoff (1975) argues that men and women differ in their use of many linguistic features.
Women, she claims, make use of a wider colour vocabulary, with terms like 'mauve', 'lavender', or 'beige' rarely being used by men; and of different
evaluative words, such as 'adorable', 'charming', 'sweet' and 'lovely'. Such sociolinguistic hypotheses readily lend themselves to testing against corpus
data. For instance Stenström (1992) compares men and women's use of expletives in the London-Lund spoken corpus, finding that women use
proportionally more expletives from a 'heaven' group ('heavens', 'gosh', 'blimey' etc.), while men make greater use of a 'hell' group ('bastard', 'damn' and
'devil'). The BNC provides information as to the age, sex, social and regional provenance of speakers and writers as attributes on a range of SGML
elements, which can be used in queries to investigate language use in relation to these variables. Note that not all this information may be available
for every written text, or for every utterance in speech. You can however always find out how many texts and utterances have attributes of
any one type, and compare this with the total number of texts and utterances in the corpus (See 6.2.1 on page 115).
As a first example, this task asks whether female speakers in the corpus use 'mauve' (a relatively rare word) more often than males. It looks at all the
occurrences of 'mauve' in spoken utterances, inspecting them to see if they are produced by male or female speakers.
As a second example, it considers whether female speakers use the much commoner word 'lovely' more than males. In this case we shall design
separate queries to count the number of utterances containing this word produced by each category of speaker.
Lastly, it returns to a case discussed in the previous task, to see whether use of the expression 'good heavens' is the prerogative of any particular
speaker age-group.
6.1.2
Components of BNC Texts
Each text document in the BNC, or <BNCDOC>, has the same hierarchic structure, composed of SGML elements. At the highest level, each written
text document contains a <HEADER> element, containing information about the text, and a <TEXT> element, containing the text itself. The header
contains the <CATREF> element discussed in the previous task (see 5.1.1 on page 98), along with a series of other elements containing bibliographic
information and the like.

page_112

Page 113

The structure of the <TEXT> element reflects the organization of the text itself. Thus at the highest level, it may contain <DIV1> elements, each
representing a chapter of a book, an act of a play, an article from a periodical, etc. These may in turn contain <DIV2> elements representing sections
or scenes, which may in their turn contain <DIV3> elements representing sub-sections, etc. These are further divided into paragraphs (<P> elements).
Portions of written texts which have particular functions, such as headings, lists, quotations, speeches and stage directions, notes, captions, etc., are also
marked up in particular ways.
In spoken text documents, the <HEADER> element also contains details concerning the setting and the participants, and is followed by an <STEXT>
element, containing a transcript of the original tape recordings. Each spoken text from the demographic component of the corpus consists of all the
conversations recorded by a particular respondent, and is divided into <DIV> elements, each of which corresponds to a single conversation. Each
<DIV> element is in turn divided into a series of <U> (utterance) elements corresponding to turns of the various speakers. Spoken texts from the
context-governed component of the corpus consist of public speech events of various kinds, ranging from phone-ins and meetings to sermons and
classroom interaction. They may be dialogue, in which case they consist of a series of <U> elements, or monologue, in which case they consist of a
single <U> element. In either case, various paralinguistic features are also identified (see 7.1.2 on page 129, 9.2 on page 163).
At the lower end of the hierarchy, paragraphs and utterances alike consist of sentences ( <S> elements), while sentences contain L-words and
punctuation (<W> and <C> elements).
All of these features are indicated in the same way in the BNC, using start-and end-tags placed at the beginning and end of each element. The tags
contain the name of the element, placed between angle brackets: in the case of end-tags, the name is preceded by a solidus. Thus the tags <TEXT>
and <TEXT> indicate the beginning and end of a written text, <U> and <U> indicate the start and end of a spoken utterance, and so on. As well as
allowing you to design queries to find the start or end of a particular element (see 9.3.2 on page 177), these tags enable you to restrict queries so that
solutions are found only where they occur within the scope of an element, i.e. between a start- and end-tag of that element-type. In this way you can
search for only those occurrences of a feature which appear within headings, quotations, speeches, spoken utterances, etc. The smallest elements of
text structure ( <S>, <W> and <C> elements) do not have end-tags, since the ending of any sentence is implied by the beginning of the next
sentence, just as the ending of any word or punctuation element is implied by the beginning of the next instance of one of these categories.

page_113

Page 114

All the elements composing the text may have attributes of various kinds, which are specified following the element name in the start-tag, along with
their values. For instance, all <S> (sentence) elements have an N attribute, whose value is the number of that sentence within the text in question. This is
the number which is displayed on the status bar at the bottom of the solutions window, and might be represented in the corpus by the tag <s
n=123>. All spoken utterances are similarly numbered, and also have a WHO attribute, whose value is a code identifying the speaker. The easiest way
to appreciate the internal structuring of texts in the BNC is to use the Browser display with the Show tags option checked. This can be accessed
via the Source option (see 1.2.7 on page 52), and shows a high-level SGML representation of the entire text from which the current solution is
taken. Clicking on the + sign preceding any element in the display shows its components at the next level, enabling you to progressively
expand the representation until you reach the lowest level.
This task shows you how to count the number of utterances which are produced by a particular speaker or category of speaker, by using the SGML
Query option to find occurrences of <U> elements whose WHO attributes have particular values. It then shows you how to to restrict other queries to
particular classes of speaker by specifying <U> elements with particular WHO attribute values as the query scope.
6.1.3
Highlighted Features
This task shows you:
how to search for words or phrases occurring in particular SGML elements with particular attribute values using the Query Builder SGML SCOPE option;
how to use Custom display format to show utterance or paragraph boundaries and speaker codes;
how to identify the characteristics of speakers of particular utterances using the Source option;
how to search for particular collections of speaker- or author-types by combining multiple attributes and values in an SGML Query.
It assumes you already know how to:
adjust the default download and display settings using the View Preferences option (see 1.2.8 on page 54);
use the Word Query, Phrase Query, and SGML Query options (see 2.2.1 on page 64, 4.2.1 on page 88, 5.2.2 on page 100);
 page_114

Page 115

design complex queries using Query Builder (see 4.2.2 on page 91, 5.2.4 on page 105);
adjust downloading procedure using the Too many solutions dialogue box (see 2.2.3 on page 66);
count the occurrences of particular elements and attribute configurations using the Too many solutions filter (see 2.2.3 on page 66, 5.2.2 on page 100);
modify queries using the Query Edit option (see 4.2.1 on page 88);
save and re-open queries (see 2.2.5 on page 70, 4.2.2 on page 94).
6.1.4
Before You Start
Using the VIEW menu PREFERENCES option, set the SARA defaults as follows:

2000 characters
MAX DOWNLOAD LENGTH
MAX DOWNLOADS 5
FORMAT Plain
SCOPE Automatic
VIEW: QUERY and ANNOTATION checked
CONCORDANCE checked
BROWSER: SHOW TAGS unchecked

6.2
Procedure
6.2.1
Searching in Spoken Utterances
Let us first design a query to find all the occurrences of 'mauve' in spoken utterances.

Click on the QUERY BUILDER button on the toolbar. The Query Builder will be displayed.
1
2 Click in the red content node and select EDIT then WORD to display the Word Query dialogue box.
3 Type in the string mauve, then check the PATTERN box and click on LOOKUP. The word 'mauve' will be
displayed in the matching words list.
4 Select 'mauve' from the matching words list, then click on OK to insert it in the Query Builder node. You
Will be returned to the Query Builder, where the content node now contains the string "mauve". The
double quotation marks indicate that this is a Word Query.
5 Click in the scope node (displayed in black). This currently contains the default value <BNCDOC>,
meaning a complete BNC document.

page_115

Page 116

Select SGML. The SGML dialogue box will be displayed.

6
7 Scroll through the list of elements and click on <U> to select it. As <U> is an element which occurs
within texts rather than text headers, there is no need for SHOW HEADER TAGS to be checked.
8 Click on OK to insert it in the Query Builder scope node. You will be returned to the Query Builder,
where the scope node now contains the string <u>, indicating that the entire content of the query must
be found within the scope of a single spoken utterance.
 9 Check that the 'Query is OK' message is displayed, then click on OK to send the query to the server.
The Too many solutions dialogue box will be displayed, stating that there are 14 solutions in 11 texts.
10 Click on the DOWNLOAD ALL button, then on OK to download all the solutions.

6.2.2
Using Custom Display Format
We now want to find out the sex of the speaker in each of these solutions. At first glance, many solutions may make little sense. This is because they
come from spoken dialogue, but do not show where one utterance ends and another begins. While PLAIN display format (selected as default in 6.1.4 on
the preceding page) shows only the words and punctuation of the text, CUSTOM display format additionally shows selected SGML elements and attribute
values. It is particularly useful for viewing solutions from spoken texts. We shall now select this format, using the Query Options command (see 5.2.3 on
page 103).

Select OPTIONS from the QUERY menu. The Query Options dialogue box will be displayed.
11
12 Click on CUSTOM to select it, then click on OK. You will have to wait while the solutions are downloaded
from the server once more, with the additional information required for this display format. You can
select Custom as the default display format using the VIEW PREFERENCES option (see 1.2.8 on page
55). You can also design your own custom display to visualize such features as truncations,
pauses, gaps, overlaps, voice quality and vocal and non-vocal events by using the CONFIGURE
option in the Query Options dialogue box (see 9.2.2 on page 165).

Custom format operates differently in Line and Page display modes. Assuming you are using the default version of Custom format and are in Line
display mode, you will see that the solutions now contain vertical lines. These indicate paragraph or utterance boundaries.

Use the arrow buttons to select the last solution, then click on the CONCORDANCE button on the toolbar to
13 switch to Page display mode.

page_116

Page 117

You will see that the display shows each utterance beginning on a new line, preceded by a speaker
identification code between curly brackets. Each speaker has a unique identification code, even those
whose identity is unknown. The code PS000 is used for an unidentifiable speaker. In written texts, the
default Custom format for Page mode displays shows the start of each paragraph indented on a
new line. To exploit its ability to display utterance and paragraph boundaries fully, Custom
format should be used selecting MAXIMUM scope with a high MAX DOWNLOAD LENGTH under VIEW
PREFERENCES (see 1.2.8 on page 56).
14 Make a note of the speaker identification code for the utterance containing the query focus in this
solution.
15 Click on the SOURCE button on the toolbar (or select SOURCE from the QUERY menu). The Bibliographic
data box will be displayed, showing the data for the source of the current solution. The lower window
lists the participants in this text, giving their speaker identification code, name, sex, class, age, profession
and relationship to the other speakers, where these are known.
16 Note down the sex of the speaker with this identification code, then click on OK to return to the
solutions display.
17 Use the left arrow button on the toolbar or the PGUP key to page through the remaining solutions,
looking up the sex of the speaker who says 'mauve' in each case. If the amount of context displayed is
insufficient to show the beginning of the utterance with the speaker identification code, you can always
double-click on the right mouse button to expand the context, or, if this is still insufficient, use the Browse
option (see 1.2.7 on page 52).

In 11 cases, the speaker is identified as female; in 2 as a male; in one case, the speaker's sex is unknown. While these numbers are too small for reliable
inferences, they suggest that 'mauve' may be more common in women's speech than in men's.
6.2.3
Comparing Frequencies for Different Types of Speaker: Male and Female 'Lovely'
Let us now examine a case where the word is sufficiently frequent to allow a statistical comparison in terms of speaker sex. To adopt a chi-squared test,
for instance, the observed and expected frequencies for each category should both be greater than 5. As you can discover by looking up the word index
with Word Query, 'lovely' occurs a total of 6278 times in the corpus as a whole, so it seems almost certain to meet this requirement. To find whether it is
more frequently used in utterances by male or female speakers, we shall find out:
the number of utterances containing 'lovely' produced by male speakers;
the number of utterances containing 'lovely' produced by female speakers;

page_117

Page 118

the total number of utterances produced by male speakers;

the total number of utterances produced by female speakers.
With this information we can carry out a chi-squared test to compare the observed and expected frequencies of utterances containing 'lovely' for each
speaker-type, and assess the significance of any difference. Another way of comparing male and female use of 'lovely' might be to calculate the
total number of occurrences of 'lovely', and compare that with the total number of words produced. However the current release of SARA
does not allow you to count the number of words produced by a particular speaker or speaker-type, only the number of sentences or
utterances.
Counting SGML Elements With Particular Attribute Values
Let us start by counting the total numbers of utterances by male and female speakers in the spoken texts of the corpus. The last task showed you how
to use the SGML Query option to count occurrences of a <CATREF> element with particular attribute values (see 5.2.2 on page 101). Here we shall use
a similar procedure to count occurrences of the <U> (utterance) element with particular attribute values.

Click on the SGML Q UERY button on the toolbar (or select NEW QUERY then SGML from the FILE menu).
18 The SGML dialogue box will be displayed.

19 Scroll through the list of elements and click on <U> to select it. The list of attributes for the <U>
element will be displayed in the bottom left hand window of the box.
20 Scroll through the attributes list and click on WHO.SEX to select it, then on the ADD button. The Attribute
dialogue box will be displayed, showing the list of values for the WHO.SEX attribute.
21 Click on m to select male speakers, then on OK to insert this attribute value pair in the SGML Query.

Other attributes whose values can be specified to find utterances by particular types of speaker include the following: they do not apply to any
<U> element for which the speaker is unknown.

Attribute Meaning Values

WHO.AGE speaker's age 0 (under 15); 1 (1524); 2 (2534); 3 (3544); 4 (4559); 5(60 or over); X (unknown)
WHO.DIALECT speaker's dialect (see 2.4 on page 239 for values)
continued on next page

page_118

Page 119

Attribute Meaning Values

WHO.EDUC speaker's level of education still in education;-14; 1516; 1718; 19+; unknown
WHO.FLANG speaker's first language (see 2.4 on page 239 for values)
WHO.RESP speaker who recorded the text identifier of respondent (only for utterances produced by other participants)
WHO.ROLE speaker's role in recording respondent; other
WHO.SEX speaker's sex m; f; u (unknown)
WHO.SOC speaker's social class AB; C1; C2; DE; UU (unknown)

Click on OK to send the query to the server.

22
 23 Read off the number of solutions from the T OO MANY SOLUTIONS dialogue box. This corresponds to the
number of utterances produced by male speakers.

If you are interested in examining gender as a variable, this number is a useful piece of general information, so it is a good idea to save this query for
future reference.

Click on OK to download the first 5 solutions. You cannot save a query without downloading at least
24 some solutions.

25 Click on the SAVE button on the toolbar (or select SAVE from the FILE menu). Choose a suitable mnemonic
as a name, such as USEX_M, then click on OK to save the query.

Now find out the number of utterances produced by female speakers.

Select EDIT from the QUERY menu. You will be returned to the SGML dialogue box showing the previous
26 query.

27 In the list of attributes, click on the WHO.SEX attribute to select it, then on ADD. The Attribute dialogue box
will be displayed.
28 Click on f to select female, then on OK to insert this attribute value pair in the query. The previous
selection for the same attribute win be removed automatically.
29 Click on OK to send the query to the server.

page_119

Page 120

Wait to read off the number of solutions from the T OO MANY SOLUTIONS dialogue box. This corresponds to
30 the number of utterances produced by female speakers.

31 Click on OK to download the first 5 solutions.

32 Select SAVE AS from the FILE menu, and save the query with a new mnemonic, such asUSEX_F. If you
edit a saved query, then want to save the edited version under a different name, you must use the
SAVE AS option under the FILE menu: using the SAVE button on the toolbar will automatically
overwrite the saved version.

You should also find out the total number of utterances in the corpus, in order to check what proportion of the utterances in the corpus are in fact
categorized as by male or female speakers. You can do this by simply removing the attribute value specification from the query.

Select EDIT from the QUERY menu to return to the SGML dialogue box.
33
34 Click on REMOVE ALL.
35 Click on OK to send the query to the server.
36 Wait to read off the number of solutions from the T OO MANY SOLUTIONS dialogue box. This corresponds to
the total number of utterances in the corpus.
37 Click on OK to download the first 5 solutions, then select SAVE AS from the FILE menu.
38 Save the query with a suitable mnemonic, such as UALL.

You should now have the following numbers:

female utterances: 307,539
male utterances: 304,278
total utterances: 753,395
As you can see, the BNC contains similar numbers of utterances by male and female speakers. The difference between the total for female and male
utterances (611,817), and the grand total gives the number of utterances for which either no speaker has been identified, or the speaker's sex is
unknown (141,578).
Searching in Elements With Particular Attribute Values
We now want to find out how often 'lovely' appears in male and female speech. To do this we must first limit the scope of a query for the word 'lovely'
to utterances produced by a particular speaker-type.

Click on the QUERY BUILDER button on the toolbar to display the Query Builder.
39

page_120

Page 121

Click in the scope node and select SGML. The SGML dialogue box will be displayed.
40
41 Using the same procedure as in the last section, select the <U> element and the attribute value
pairwho.sex=f.
42 Click on OK. You will see that the node now contains the string

<person ID=_0 sex="f">#<u who=_0>

meaning an utterance by a female speaker. The complexity of the query syntax here is due to the fact that strictly speaking, <U> elements in the
BNC do not have all the attributes and values listed in the SGML and Attribute dialogue boxes, but merely a WHO attribute whose value is a
speaker identification code. SARA interprets choices from the lists displayed for the <U> element in the SGML and Attribute dialogue boxes by
matching this speaker identification code against the information given in the corresponding <PERSON> element in the text header.
Now specify the content node, namely the word 'lovely'.

Click in the red content node, and select EDIT then WORD. The Word Query dialogue box will be
43 displayed.

44 Check the PATTERN box and type in the stringlovely, then click on LOOKUP.
45 Click on 'lovely' in the matching words list to select it, then on OK to insert it in the Query Builder node.
The content node should now contain the string "lovely".
46 Check that the 'Query is OK' message is displayed, and click on OK to send the query to the server.
47 Read off the total number of solutions from the Too many solutions dialogue box. This corresponds to
the number of 'lovely's produced by female speakers.

Now find out the number of 'lovely's produced by male speakers.

Click on CANCEL to abort the query. You will be returned to the Query Builder dialogue box.
48
49 Click in the scope node and select SGML. The SGML dialogue box will be displayed.
50 Select the <U> element, then select the valuem for the WHO.SEX attribute.
51 Click on OK to insert this query in the scope node.
52 Click on OK to send the revised query to the server.

page_121

Page 122

Read off the total number of solutions from the Too many solutions dialogue box. This corresponds to
53 the number of 'lovely's produced by male speakers.

54 Click on CANCEL to return to the Query Builder.

 54 Click on C to return to the Query Builder.

We have so far counted the numbers of occurrences of 'lovely' in utterances by male and female speakers respectively. These may not however be the
same as the numbers of utterances containing the word 'lovely', since it is possible that in some utterances the word 'lovely' is used more than once. To
calculate the number of utterances with 'lovely', we must count the number of times there are two occurrences of 'lovely' within a single utterance, and
subtract this figure from the totals we obtained above.

Click on the downward branch of the "lovely" content node to create a second node.
55
56 Click in the new node and select EDIT then WORD.
57 Type in the string lovely and click on LOOKUP.
58 Select 'lovely' from the list of matching words and click on OK to insert the query in the Query Builder
node. The two nodes are connected with a ONE-WAY link, meaning that the second occurrence of 'lovely'
follows the first at any distance within the specified scope, i.e. a single utterance.
59 Check that the 'Query is OK' message is displayed, and click on OK to send the query to the server.
60 Read off the number of solutions from the Too many solutions dialogue box, then click on C ANCEL to
return to the Query Builder.
There are 42 solutions, corresponding to the number of repetitions of 'lovely' within male utterances.

Now find the number of times there are two occurrences of 'lovely' in female utterances, by changing the scope of the query.

Click in the scope node and select SGML. The SGML dialogue box will be displayed.
61
62 Select the value f for the WHO.SEX attribute on the <U> element.
63 Click on OK to insert the query in the scope node, then click on OK to send the query to the server.
64 Read off the number of solutions from the Too many solutions dialogue box. There are 83 repetitions of
'lovely' within female utterances.
65 Click on OK to download the first five solutions.

page_122

Page 123

Results
The numbers of occurrences of 'lovely' and of repetitions of 'lovely' in male and female utterances respectively are thus:
female speakers: 1439, of which 83 are repetitions
male speakers: 716, of which 42 are repetitions
This means that the numbers of utterances containing 'lovely' are respectively 1356 and 674. If, for each type of speaker, you divide the number of
utterances containing 'lovely' by the total number of utterances, you can then compare their frequencies per 1000 utterances. These are:
female speakers: (1356/307,539) × 1000 = 4.4
male speakers: (674/304,278) × 1000 = 2.2
With the same data, you can also perform a chi-squared test of significance for the absolute frequencies. Given the combined number of utterances and
the combined number of utterances containing 'lovely', the expected frequencies of utterances with 'lovely' for women and men would respectively be
1020 and 1010, in comparison with the observed frequencies of 1356 and 674. This difference is significant, corresponding to a probability of p <
.0001 for one degree of freedom, and it therefore seems plausible to claim that women do use the word 'lovely' more frequently than men.
6.2.4
Investigating Other Sociolinguistic Variables: Age and 'Good Heavens'
The last query looked at the hypothesis that women use the word 'lovely' more than men. Such questions may also be posed concerning other
sociolinguistic variables encoded as utterance attributes.
Take the expression 'good heavens'. In the last task, we found that this was more common in written than in spoken dialogue, and, in the latter, more
common in data collected by older respondents (see 5.2.4 on page 105). Can we also link its use to any particular age-group of speakers?
Using the Query Builder, first design a query to find all the occurrences of this phrase in spoken utterances.

Click on the QUERY BUILDER button on the toolbar. The Query Builder dialogue box will be displayed.
66
67 Click in the scope node and select SGML.
68 In the SGML dialogue box, select <U> from the list of elements.
69 Click on OK to insert it in the Query Builder node.
70 Click in the content node and select EDIT then PHRASE to display the Phrase Query dialogue box.

page_123

Page 124

Type in the string good heavens and click on OK to insert the query in the Query Builder node.
71
72 Check that the 'Query is OK' message is displayed, then click on OK to send the query to the server.
73 Note the number of solutions and texts from the Too Many Solutions dialogue box, and then download
the first ten, checking the ONE PER TEXT option.
74 Select OPTIONS from the QUERY menu. The Query Options dialogue box will be displayed.
75 Click on CUSTOM to select it, then on OK.
76 When downloading is complete, click on the CONCORDANCE button on the toolbar to select Page display
mode.
77 Make a note of the speaker identification code for the first solution, then click on the S OURCE button on
the toolbar. The Bibliographic data box will be displayed, showing the data for the current solution.
78 Make a note of the sex and age of the speaker with that code, then click on OK to return to the
solutions display.
79 Use the right arrow button on the toolbar to page through the remaining solutions, looking up the details
of the speaker who says 'good heavens' in each case.
80 Do you notice any features shared by many of these speakers? While roughly equal numbers of
speakers of each sex seem to use 'good heavens', they tend to be relatively old most of them over 40.

Testing the Hypothesis: Alternative Attribute Values

The hypothesis just formulated was derived from the first ten solutions. Let us now see exactly how often 'good heavens' is used by older and by
younger speakers, searching first for all occurrences in utterances by speakers over 45, then for all those in utterances by speakers under 35.

Select QUERY then EDIT to return to the Query Builder dialogue box, showing the previous query.
81
82 Click in the scope node and select SGML. Select the <U> element and the attribute WHO.AGE, then
click on ADD. The Attribute dialogue box will be displayed, showing values for the WHO.AGE attribute.
Values for age are listed in bands (0 = under 15, 1 = 1524, 2 = 2534, 3 = 3544, 4 = 4559, 5 = 60
or over, unknown). You can combine two or more bands in a single query by co-selecting them.

Holding down the CTRL key, click on the value4 (4559), then on the value 5 (60+) to select them, then
83 click on OK. The selection will

page_124

Page 125
 be displayed in the right hand window of the SGML dialogue box, with the alternative values for the
attribute shown as a disjunction separated by a vertical bar.
84 Click on OK to insert this scope in the Query Builder node.
85 Check that the 'Query is OK' message is displayed, and click on OK to send the query to the server.
86 Read off the number of solutions from the Too many solutions dialogue box, then click on C ANCEL to
return to the Query Builder.
87 Now re-edit the query to consider the under-35 age group (age-band values 0, 1 and 2), and read off
the number of solutions. You will find that out of the 33 occurrences of 'good heavens' in spoken
utterances, 20 occur in utterances produced by speakers aged 45 or over, whereas only 6 are in
utterances by speakers aged under 35.
88 Click on CANCEL once to return to the Query Builder, and once more to return to the previous solutions
display.

As in the case of 'lovely', we can interpret these figures by comparing the numbers of utterances containing 'good heavens' with the total numbers of
utterances produced by speakers over 45 and under 35 respectively. As you can find out by designing appropriate queries and drinking large amounts
of coffee while waiting for the results, the spoken component of the BNC contains 210,666 utterances produced by speakers over 45, and 232,476
utterances produced by speakers under 35. Since the total numbers of utterances are similar for each of the two groups, the expected frequencies of
'good heavens' are also similar (13). Comparison of the two groups is also simplified by the fact that there are no cases where 'good heavens'
occurs twice in the same utterance.
6.3
Discussion and Suggestions for Further Work
6.3.1
Sociolinguistic Variables in Spoken and Written Texts
The design of queries concerning categories of speakers in spoken texts and categories of authors in written texts requires different approaches. In
spoken texts, the relevant unit of analysis is the individual utterance or <U> element, each of which is associated with a particular speaker. In this task
you have seen how to restrict queries to particular categories of speakers by restricting their scope to <U> elements with specific attribute values. In
written texts, on the other hand, the relevant unit is generally the entire text. As we saw in the previous task (see 5.1.1 on page 98), authorship is not
indicated as an attribute value on the <TEXT> element, but in the text header, as an attribute value on the <CATREF> element. Since the latter element
does not contain the text itself, this means that you cannot specify authorship as a scope restriction.

page_125

Page 126

Instead, to find cases where a particular feature is produced by a particular category of author, you must follow a procedure analogous to that described
in the previous task (see 5.2.4 on page 105):
design a first content node containing an SGML query for the value of the AUTHOR attribute on the <CATREF> element;
join this with a ONE-WAY link to a second content node containing a query for the desired text feature;
use the default scope of <BNCDOC> (the only element to contain both the text header and the text: see 6.1.2 on page 112).
Restricting the scope of a query to the < U> element limits the search to spoken texts for the reason that utterances only occur within spoken texts in the
BNC. Other scope selections have similar effects: <P> elements (paragraphs) only occur within written texts, and <SP> elements (speeches) only
occur within written-to-be-spoken texts. Numbered divisions (<DIV1>, <DIV2> etc.) occur only in written texts, while unnumbered divisions
(<DIV>) occur only in spoken ones. You can find out what type(s) of text an element occurs in by clicking on its name in the list of elements in the
SGML dialogue box (see 5.2.2 on page 100).
Using any of these structural elements to restrict the scope of a search to a particular text-type will also of course exclude any portions of the text which
are not contained within them. For instance, if you restrict the scope of a search to <P> or <SP>, you will also exclude headings, captions, stage
directions, lists and notes, since these do not generally appear within paragraphs or speeches. Moreover, any scope restriction based on text elements
will necessarily exclude elements contained within the text header, such as <CATREF> elements and attributes. This means, for example, that Query
Builder does not allow you to combine a restriction of scope to female utterances with a restriction of text-type to spoken monologue, nor in general to
limit queries to particular elements occurring in texts of particular types, such as headlines in periodicals, stage directions in plays, or recipes by male and
female cooks. The next task shows how you can overcome this difficulty using a CQL QUERY (see 7.2.3 on page 135).
6.3.2
Some Similar Problems
Ruddy
When the spoken component of the BNC was collected, the compilers noticed that the expletive 'ruddy' seemed to be predominantly used by older
speakers (Rundell 1995). What proportion of the spoken occurrences of 'ruddy' are in fact by speakers over the age of 45? Are similar proportions
also found for authors of written texts? Use QUERY BUILDER to restrict the scope of a Word

page_126

Page 127

Query for 'ruddy' to utterances whose speakers are aged over and under 45, following the procedure adopted for 'good heavens' (see 6.2.4 on
page 123). Count the numbers of occurrences, then carry out SGML queries to find the total number of utterances produced by speakers in
these age-bands.
While SARA allows you to treat speaker age as an attribute of utterances (see 6.2.4 on page 123), author age is only indicated as an attribute
of the <CATREF> element in the text header (see 5.2.2 on page 101). Consequently, to find figures for written texts you should use the Query
Builder default scope of <BNCDOC>, then join an SGML Query for the <CATREF> attribute values written_age=4559 and
written_age=60+ to a Word Query for ruddy with a One-way link.
There are 204 occurrences of 'ruddy' in the corpus, 78 of them in speech. In no case is 'ruddy' repeated within a single utterance. 57 of these
occurrences are produced by speakers over 45, and only 11 by speakers under 45, even though there are 50% more utterances by the younger
group. Most written occurrences of 'ruddy' are in texts which are unclassified for author age. There are only 7 occurrences produced by
authors over 45 as opposed to 14 by authors under 45: numbers of texts in the corpus produced by each age group are equal. However, a
glance at these occurrences suggests that 'ruddy' is used mainly in the sense of 'red' in writing, e.g. 'a ruddy complexion', rather than as an
expletive, as it is in speech.
Foreign Speakers
A potentially weak point of the methodology in this task is that frequencies in utterances by different categories of speakers have been compared
without considering the number of speakers in each category. While for variables such as sex and age, numbers of speakers in each category are
predictably large enough to even out the effects of individual styles, this may not be the case when categories have few members. For instance, the
WHO.FLANG and WHO.DIALECT attributes have values indicating the first language and the dialect of the speaker (for a list of languages and
dialects, see 2.4 on page 239). For some of these values, there may only be a single speaker in the BNC. Design queries to find out how many
speakers in the BNC have French as their first language, and how many of their utterances are contained in the corpus. To find the number of
speakers, use an SGML Q UERY to count occurrences of the <PERSON> element with the attribute value pair flang=FR-FRA, which
indicates French-from-France. You will have to type the value FR-FRA (in upper case) in the Attribute dialogue box. Then edit the query to
count occurrences of the <U> element where the WHO.FLANG attribute has the value FR-FRA.
There are only three French speakers from France in the BNC (2 women and 1 man), who produce a total of 873 utterances. Clearly this is too
few to allow any reliable inferences about the characteristics of utterances produced by French speakers.

page_127

Page 128

Class Consciousness
Where known, the social class of speakers in the BNC is indicated by the value of the WHO.SOC attribute on utterance elements. However, due to the
way the corpus was collected, for spoken demographic texts this information is only reliable for the social class of the respondents who made the
recordings. Bearing this in mind, does the corpus provide any evidence of class differences in ways of speaking about the establishment and its
institutions, such as universities? Use Query Builder to design queries to find occurrences of the forms 'university' and 'universities' in the
utterances of speakers who are indicated as respondents (i.e. whose WHO.ROLE attribute has the value resp) and as upper-middle class (i.e.
whose WHO.SOC attribute has the value AB). Then design a similar query for respondents from lower class ( DE) backgrounds. Note the
numbers of solutions in each case, and download them using CUSTOM display format.
Frequencies of the words 'university' and 'universities' are similar for respondents in the two social groups. It is however interesting that many
of the references in the 'DE' group occur in explanations of why the recordings are being made, suggesting that the data collection procedure
may be a source of bias here.

page_128

Page 129

7
'Madonna Hits Album' Did It Hit Back?
7.1
The Problem: Linguistic Ambiguity
7.1.1
The English of Headlines
The newspaper headline in the title to this task is mildly amusing insofar as it is syntactically and semantically ambiguous. On one reading, 'hits' is a plural
noun; on another, it is a singular verb in the present tense. As a verb, it may have its literal sense of physical aggression, or the metaphorical meaning of
criticism. While the most likely reading is perhaps the noun one an album of Madonna's hits the fact that the phrase comes from a newspaper headline,
where present tense verbs are frequent (Bell 1991), underlines the ambiguity, suggesting the alternative interpretation of 'hits' as a verb.
How justified is such an interpretation? This task examines the word 'hits' in the BNC, to see whether it is generally used in headlines as a verb or a
noun, and whether it is used literally as well as metaphorically as the joke would seem to require.
7.1.2
Particular Parts of Speech, Particular Portions of Texts
There are three main problems to be faced in carrying out the task:
how to find occurrences of a word as a particular part-of-speech (in the case of 'hits', as a singular verb rather than as a plural noun);
how to restrict a query to particular portions of texts, such as headlines;
how to further restrict a query to particular portions of particular text-types, such as headlines in newspapers and periodicals.
As you will have noticed when examining solutions in SGML format (see 5.2.3 on page 103) and when browsing texts using the Show tags option (see
1.2.7 on page 52), every L-word in the BNC is tagged as a <w> element with a Part-of-speech or POS code. Assigned by the probabilistic CLAWS
program (see 2.1.2 on page 34), this code indicates whether the word is a noun (common or proper, singular or plural), a verb (in its root, '-s', '-ing',
past, or past participle form), an adjective, a determiner, a conjunction, etc. SARA enables you to use these POS codes to:
restrict queries to cases where a word is tagged with a particular POS code, or one of a series of such codes, using the POS Query option;
display solutions with different parts-of-speech in different colours, using the POS FORMAT display option;
sort solutions by their POS codes rather than by their orthographic form, using the POS CODE collating option.

page_129

Page 130

The POS codes used in the BNC are listed in 2.1 on page 230, as well as in the HELP file.
As we saw in the last task, structural components of texts such as sections, paragraphs, utterances, sentences, etc. are indicated as SGML elements in
the BNC (see 6.1.2 on page 112). In the QUERY BUILDER, these can be used to restrict the scope of a query to occurrences within a particular type of
component (see 6.2.3 on page 117). Specialized components such as headings, captions, quotations, notes, etc. are similarly marked, so that by
restricting the scope of a query, to <HEAD> elements, for example, you can find all the occurrences in headings of a particular word or phrase.
The Query Builder does not permit you to restrict a query to a particular text-type as well as to a particular type of component. To do this you must
instead use a CQL Query.
7.1.3
Highlighted Features
This task shows you:
how to copy a string from the Word Query dialogue box to a different dialogue box using the Copy option;
how to use the POS QUERY option to search for a word as a particular part of speech;
how to use the QUERY BUILDER to restrict the scope of a query to particular components of texts;
how to use the POS FORMAT option to display different parts of speech in different colours;
how to sort solutions according to part-of-speech values using POS CODE collating;
how to use a CQL QUERY to restrict a query to particular portions of particular text-types by combining S COPE and <CATREF> specifications.
It assumes you already know how to:
adjust default settings using the View menu Preferences option (see 1.2.8 on page 54);
carry out a Word Query (see 2.2.1 on page 64);
carry out an SGML Query (see 5.2.2 on page 100);
use the Query Builder to join queries and restrict query scope (see 4.2.2 on page 91, 5.2.4 on page 105, 6.2.3 on page 117);
adjust downloading procedure in the Too many solutions dialogue box (see 2.2.3 on page 66);
 page_130

Page 131

sort solutions using Primary and Secondary keys (see 3.2.2 on page 78, 4.2.1 on page 89);
mark and thin selected solutions (see 2.2.4 on page 68).
7.1.4
Before You Start
Using the View menu Preferences option, set the default settings as follows:

400 characters
Max download length
Max downloads 10
Format Plain
Scope Paragraph
View: Query and Annotation checked
Concordance checked
Browser: Show tags unchecked

7.2
Procedure
7.2.1
Searching for Particular Parts-of-Speech with POS Query: 'Hits' as a Verb
SARA's POS Query option enables you to search for an L-word with a particular POS code or codes. This is particularly useful where the word is a
frequent one, and can act as more than one part of speech, since searching only for a particular part-of-speech value will reduce the total number of
solutions and may increase overall precision.
Let us begin by looking up the frequency of 'hits' in the word index.
Click on the Word Query button on the toolbar, and type the stringhits in the dialogue box.
1
2 Check the Pattern box so as to search the index for this word only, then click on Lookup.
3 Select 'hits' from the matching words list. You will see that 'hits' occurs 1319 times in the corpus.

Let us now find out the frequency of 'hits' as a verb.

Click on Copy to copy the stringhits to the clipboard, then click on Cancel to leave Word Query. Using the
4 Copy option in Word Query saves you from having to retype the string if you decide to switch to
another query type, Note that it copies the input string, not selections from the matching words list.

page_131

Page 132

Click on the POS Query button on the toolbar (or select New query and POS from the File menu). The POS
5 Query dialogue box will be displayed.
6 Click in the L-word window, and press Shift+Insert to paste in the stringhits from the clipboard. Only those
words which appear in the word index (L-words: see 2.2.1 on page 64) can be used in a POS Query.
To find out what L-words start with a particular string of letters, or match a particular pattern, you
must first look them up in the index using Word Query (see 2.2.1 on page 64, 8.2.1 on page 147).
7 Click in the Part-of-speech window (or press Tab). The POS codes used with the word 'hits' will be displayed,
in alphabetical order. These are: NN2, NN2-VVZ, and VVZ,.
8 Click on the NN2 code. An explanation of the code will appear to the right of the window.
9 Do the same for the VVZ and NN2-VVZ codes. The latter is an ambiguous portmanteau code, meaning
 that CLAWS was unable to decide whether a NN2 or VVZ code should be assigned i.e. whether 'hits'
was a plural noun or a singular present-tense verb. Portmanteau codes list the two most probable parts
of speech for the L-word in question. They are used for some 5% of words in the corpus. The two
alternatives are given in alphabetical order.
10 Click on VVZ to select it, then on OK to send the query to the server. The Too many solutions dialogue
box will be displayed, stating that there are 404 cases where 'hits' has a VVZ code.

While much smaller than the total frequency of 'hits', 404 is still a very large number of solutions to have to deal with. Moreover, we do not know how
many other instances of 'hits' as a verb may lurk among the cases with portmanteau NN2-VVZ codes. If the emphasis is on recall (finding all the
relevant solutions), rather than precision (minimizing the number of spurious solutions: see 1.3.1 on page 60, 4.2.1 on page 87, 4.3.2 on page 96), you
should always include relevant portmanteau instances in POS queries. We should certainly check how many occurrences of'hits' have portmanteau codes
before drawing conclusions as to its frequency as a verb.
Click on Cancel to return to the POS Query dialogue box. The list of POS codes for 'hits' should still be
11 displayed.
12 Holding down the Ctrl key, click onNN2-VVZ and on VVZ to select both. Holding down Ctrl when
clicking in the POS code display box allows you to select multiple codes.

page_132

Page 133

Click on OK to send the revised query to the server. The Too many solutions dialogue box will be
13 displayed, telling you that 780 cases of 'hits' have now been found.

This is almost double the previous number, showing that there are a great many cases where the coding of 'hits' is uncertain. It means you should
undoubtedly include portmanteau instances in a query aiming to examine occurrences of 'hits' as a verb. You will then have to inspect the solutions and
use the Thin option to delete those occurrences with portmanteau codes where 'hits' is in fact a noun (see 7.2.4 on page 137).
Inspecting 780 solutions would be no joke, however, so let us first further restrict the query to those cases of 'hits' as VVZ or NN2-VVZ which occur
in portions of texts marked as headings.
Click on Cancel in the Too many solutions dialogue box to abort the query. You will be returned to the POS
14 Query dialogue box.
15 Click on Cancel to return to the main SARA-bnc window.

7.2.2
Searching within Particular Portions of Texts with Query Builder
Specifying the Content Node
First, let us transfer the POS Query for 'hits' to a Query Builder content node.
Click on the Query Builder button on the toolbar (or select New query and Query Builder from the File menu). The
16 Query Builder will be displayed.
17 Click in the empty content node and select Edit, then POS. The POS Query dialogue box will be
displayed.
18 Click in the L-word window, then press Shift+Insert to paste in the stringhits from the clipboard.
19 Click in the Part-of-speech window (or press Tab) to display the POS codes associated with 'hits'.
20 Holding down the Ctrl key, click onVVZ and NN2-VVZ to select them, then click on OK to insert this
query into the Query Builder node. This component of the query is displayed in the node as:

"hits"=VVZ | "hits"= NN2-VVZ

In CQL, POS specifications are represented by double quotes round the L-word, followed by an equals sign and the POS code. The vertical bar
represents the disjunction between the alternative values.

page_133
 Page 134

Specifying the Scope Node

We now need to add the requirement that 'hits' should fall inside a heading. While this will potentially include not only headlines from newspapers and
periodicals, but also chapter and section titles in other written text-types, it should significantly reduce the overall number of solutions.
To do this we must change the scope of the query. The default value shown in the scope node on the left of the Query Builder, <BNCDOC>, implies
that the content of the query may occur anywhere in the text or text header. We shall change this to <HEAD>, i.e. those portions of text marked as
headings. Other elements which can be used to restrict the scope of queries to particular portions of texts include <STAGE> (stage
direction), <POEM> (verse), <CAPTION> (of a table or figure), <ITEM> (in a list), <SALUTE> (in a letter). As seen in the last task, you
can also use structural elements such as <S> (sentence), <U> (utterance), <P> (paragraph) or <SP> (speech) to restrict scope in the same
way (see 6.2.1 on page 115).
Click on the scope node and select SGML. The SGML dialogue box will be displayed.
21
22 Make sure that the Show header tags box is unchecked. This reduces the length of the list of elements in the
display, excluding those which only appear in text headers.
23 Scroll through the list of elements and click on <HEAD> to select it. The list of attributes for the
<HEAD> element will be displayed. You should not confuse <HEADER> elements, which contain
information about entire texts, and precede the text itself, with <HEAD> elements, which indicate
headings within texts.
24 Click on Type, then on the Add button. The Attribute dialogue box will be displayed, showing the list of
values for the type attribute. These include main (the main heading to a chapter, article, or section),
sub (a sub-heading following a main heading), byline (a byline, typically giving the name of the
author or source), and unspec (type unspecified).
25 Click on main, then on OK to select this attribute value pair. You will be returned to the SGML dialogue
box.
26 Click on OK to insert this selection into the Query Builder scope node. The scope is displayed as <head
type=''main''>.
27 Check that the 'Query is OK' message is displayed, then click on OK to send the query to the server. The
Too many solutions dialogue box will be displayed, showing that there are 98 solutions in 58 texts.

Since the query did not specify the text-type in which the verb 'hits' should occur, it is now advisable to check whether these solutions generally come
from headings in newspapers and periodicals or from other text-types. We can get

page_134

Page 135

some idea by looking at a sample of the 58 texts from which the solutions are taken.
Change the Download hits number to 20, select One per text, and click on the Random button. Then click on OK to
28 send the query to the server.
29 Scroll through the solutions, clicking on the Source button on the toolbar to find where each comes from.
Nearly all the solutions appear to come from newspapers and periodicals. This is also evident from the fact
that many are followed by a byline giving the name of the author a feature typical of headlines in
periodicals.

While it would be convenient to formally restrict this query to periodicals not least so as to further reduce the number of solutions we cannot do this
using the Query Builder. This is because the scope restriction in the Query Builder applies to the entire content of the query, so that restricting the scope
to <HEAD> elements within the text excludes the text header where the <CATREF> element specifying the text-type is located (see 5.2.2 on page
101).
In the current version of SARA, the only way to get round this problem is to formulate the query directly in the corpus query language, using the CQL
Query option. This requires you to write the full text of the query yourself without the aid of the Query Builder.
To formulate a CQL Query, you must first find out exactly how the various components of the query and their links need to be represented. You can do
this by carefully examining the texts of other queries you have designed using the Query Builder.
7.2.3
Searching within Particular Portions of Particular Text-Types with CQL Query
Provided that you checked the Query option in the display preferences under the View menu (see 7.1.4 on page 131), you will see the text of the current
query displayed above the solutions in the Query1 window. It should read:
 ((("hits"=VVZ | "hits"=NN2-NN2-VVZ))/<head type="main">)

This query matches any occurrence of 'hits' which has been given a POS code of either VVZ or NN2-VVZ, provided that it occurs within the scope of
a <HEAD> element whose TYPE attribute has the value main. In order to further restrict this query to periodicals, we now need to discover how to
represent a restriction to this text-type in CQL, and how to join such a restriction to the rest of the query.

page_135

Page 136

Click on the Query Builder button on the toolbar to display a new Query Builder window.
30
31 Click in the content node and select Edit then SGML. The SGML dialogue box will be displayed.
32 Check the Show header tags box. The <CATREF> element, which appears in the text header, will now be
included in the list of elements.
33 Select <CATREF> from the list of elements. The list of attributes for <CATREF> will be displayed.
34 Select WRITTEN_MEDIUM, then click on the Add button. The Attribute dialogue box will be displayed, showing
the list of values for the WRITTEN_MEDIUM attribute.
35 Select Periodical and click on OK to return to the SGML dialogue box.
36 Click on OK to insert this component of the query in the Query Builder node. You will see this component
is displayed as:

<catRef target=wriMed2>

wriMed2 is the code for periodicals, the second of the categories of written media in the BNC. For a full list, see 2.2 on page 234 .
We now need to find out how to represent the link between this component and the rest of a query. We can do so by completing the current query with
a search for any word that occurs in written periodicals.
Click on the downward branch of the content node to add a second node to the query.
37
38 Click on the new node, and select Edit then Phrase. The Phrase Query dialogue box will be displayed.
39 Type in dummy, and click on OK to insert it in the Query Builder node.

The link between the two nodes currently has the default value One-way, displayed as a downward arrow, meaning that the <CATREF> node precedes
the 'dummy' node. This is correct (see 5.2.4 on page 105).
Check that the 'Query is OK' message is displayed and click on OK to send the query to the server. The
40 Too many solutions dialogue box will be displayed.
41 Click on OK to download the first 10 solutions. The query text will be displayed above the solutions in the
Query2 window. It should read: (<catRef target=wriMed2>) * ("dummy")

The One-way link between the two components of the query is represented by a star. Each of the two components is enclosed in parentheses.

page_136

Page 137

Formulating the CQL Query

You should now be able to formulate the text of a CQL Query which combines the desired text-type and scope restrictions. It should read:
(<catRef target=wriMed2>) *
(("hits"=VVZ|"hits"=NN2-VVZ)
/ <head type="main">)

Click on the CQL Q UERY button on the toolbar (or select New query and CQL from the FILE menu). The
42 CQL dialogue box will be displayed.
43 Type in the text of the query exactly as above, taking special care over case and parentheses. Spaces
around the * and / operators are not significant. The text will automatically wrap round to the next line as
you type it.
44 Click on OK to send the query to the server. The Too many solutions dialogue box will be displayed,
stating that there are 77 solutions in 44 texts.

These numbers are only a little smaller than those for the earlier query (Query 1), which did not include a <CATREF> restriction (see 7.2.2 on page 134).
This confirms that most of the occurrences of 'hits' in headings in the corpus do indeed come from periodicals (77/98).
Download all the solutions. The solutions should be displayed in the Query3 window.
45

7.2.4
Displaying and Sorting Part-of-Speech Codes
Since we have included cases where 'hits' has the portmanteau code NN2-VVZ in the query, in some of the solutions 'hits' may be a noun rather than a
verb. We therefore need to inspect all the solutions where 'hits' has a portmanteau code in order to delete these cases. We can do this by displaying the
solutions in POS format and then using the POS CODE collating option to sort them according to their part-of-speech codes.
Select OPTIONS from the QUERY menu. The Query Options dialogue box will be displayed.
46
47 Select POS format, then click on OK, and wait while SARA down-loads the part-of-speech information
from the server. The solutions will be re-displayed, using different colours for different parts of speech.

The query focus (the word 'hits') may not always appear in the same colour. Unless you have radically modified the default colour scheme (see 1.7.4 on
page 227), cases of 'hits' with a VVZ code will be in a different colour from

page_137

Page 138

those with a portmanteau NN2-VVZ code. First, check to make sure this is the case.
Position the mouse on the word 'hits' in the first solution, and hold down the right mouse button. The POS
48 code for this word will be displayed.
49 Release the button, and go through the solutions until you find one where 'hits' is displayed in a different
colour. Repeating the above procedure, you will find it has a different POS code. In a POS format
display, positioning the mouse on any word on the screen and holding down the right mouse button
will display its POS code.

We now want to remove any solutions where 'hits' has a NN2-VVZ portmanteau code but is in fact a plural noun. We shall first sort the solutions by the
POS code for 'hits', so as to group those with portmanteau codes.
Select SORT from the QUERY menu.
50
51 Select CENTRE as Primary key with a span of 1.
52 Select POS CODE collating.
53 Click on OK to perform the sort. As the sorting of POS codes is alphabetical, solutions where 'hits' is
coded as NN2-VVZ will be grouped before those where it is coded as VVZ. In each group, the word
'hits' should be displayed in a different colour. You can only use POS code collating in a POS format
display.
54 Scroll through the NN2-VVZ group of solutions, and mark those where 'hits' is in fact a noun by double-
clicking on them (or pressing the space bar). The solutions with nouns are virtually all cases where 'hits' has
the meaning of 'successes', as in 'Beatles hits', 'video hits', etc.
55 Select THIN then REVERSE SELECTION from the QUERY menu to delete the marked solutions. The solutions
should now only contain instances where 'hits' is a verb. You can see how many of these there are by
looking at the number on the status bar.

7.2.5
Re-using the Text of One Query in Another: 'Hits' as a Noun
Having determined the frequency of 'hits' as a verb in periodical headings, it is relatively straightforward to compare it with the frequency of 'hits' as a
noun. If we design a new query to find the frequency of 'hits' in periodical headings regardless of part of speech, subtracting the number of verbs from
this latter figure will give us the number of nouns.
The easiest way to produce this new CQL query is to copy the text of the previous version to the clipboard, and then edit it.
Select EDIT from the QUERY menu to return to the CQL query. The query text will be displayed in reverse
56 video, showing that it is selected.

page_138

Page 139

Press CTRL+INSERT to copy the query text to the clipboard. Only the Word Query dialogue box has a
57 dedicated COPY button (see 7.2.1 on page 131). To copy query text from a Phrase Query, Pattern
Query or CQL dialogue box, you must select it and then use the Windows CTRL+INSERT command.
58 Click on CANCEL to return to the solutions display, then click on the CQL Q UERY button on the toolbar. A
new CQL dialogue box will be displayed.
59 Press SHIFT+INSERT to copy the text of the previous query into the dialogue box.
60 Delete the POS code specifications. The query should now read:

(<catRef target=wriMed2>)
* (("hits")/<head type="main">)

Click on OK to send the query to the server. The Too many solutions dialogue box will be displayed,
61 stating that there are 100 solutions. This suggests that the vast majority of occurrences of 'hits' in
periodical headings are in fact verbs. You can if you want verify this by displaying the solutions in POS
format and sorting them by the focus using POS code collating, then examining those tagged as nouns.

7.2.6
Investigating Colligations Using POS Collating
Literal and Metaphorical Uses
Let us now see what sorts of things hit and are hit in the headlines where 'hits' is a verb, i.e. whether there are literal as well as metaphorical uses. When
the word is being used in the literal sense, there will typically be a noun both before and after it, so we can make these cases somewhat easier to find
simply by sorting the solutions according to the parts-of-speech of the words to the left and to the right. This will highlight grammatical patterns, or
colligations, which typically precede and follow the query focus.
Download the first 10 solutions, then use the WINDOW menu to return to the window of solutions where
62 'hits' is a verb (Query3).
63 Re-sort these solutions using the two words to the right of the focus as Primary key, with POS CODE
collating. The solutions will be grouped according to the codes of the two words following 'hits'.

This grouping highlights many phrasal and prepositional verb forms, such as 'hits back' and 'hits at'. There are also a number of verb-adverb-preposition
combinations, like 'hits back at', 'hits out at/over', etc. All these seem metaphorical. We also find a range of metaphorical uses indicating increase or
decline: 'hits high', 'hits record', 'hits target', 'hits the roof', 'hits rock bottom', 'hits buffers' etc. In the group where a noun immediately follows 'hits', we
find

page_139

Page 140

a wide range of affected people, companies, places and objects, but only "Train hits getaway van" and "Baby hurt as firework hits her on head" stand
out as clearly literal. When you enlarge the context, the mysterious "Blunt instrument hits Oslo'' turns out to be a pun on the name of a sportsman whose
team beat Oslo's.
Now let us see what precedes 'hits'.
Re-sort the solutions using the two words to the left of the focus as Primary key, again with POS CODE
64 collating. The solutions will be grouped according to the part-of-speech sequence preceding the verb 'hits'
typically including some noun as its subject.

Many preceding nouns denote violent catastrophes ('attack', 'blast', 'blaze', 'blow', 'bombshell', 'hurricane'), and negative economic and political events
('letdown', 'confusion', 'scandal', 'gloom', 'crisis'). All of these involve metaphorical uses of 'hits'. Where names of people, companies and institutions
('letdown', 'confusion', 'scandal', 'gloom', 'crisis'). All of these involve metaphorical uses of 'hits'. Where names of people, companies and institutions
occur, they are generally followed by a phrasal or prepositional verb form again, as we have seen, metaphorical. In no case do we find an animate agent
acting as the subject of 'hits' in a literal sense.
Results
There is thus little evidence in the corpus to warrant a literal reading of 'hits' as a verb in the headline 'Madonna hits album'. While considerably more
frequent as a verb than as a noun, the verb 'hits' seems overwhelmingly metaphorical in headings, and where literal, not to refer to deliberate human
action. A literal reading would seem to depend on the application of a 'person-hits-object' schema which is probably associated with other types or
components of texts.
7.3
Discussion and Suggestions for Further Work
7.3.1
Using Part-of-Speech Codes
You may have noticed that even after eliminating spurious solutions to the query involving portmanteau codes, there still remain a few solutions where
'hits' is clearly a noun but is coded unambiguously as a verb! CLAWS, like all automatic part-of-speech tagging programs, is less than 100 percent
accurate, so any POS QUERY or sort with POS CODE collating is likely to involve some errors. One useful indication of the probable accuracy of the
tagging of a particular word as a given part of speech is the relative frequency with which portmanteau tags have been assigned: generally speaking, the
smaller the number of cases with a given portmanteau code as compared to the number which are assigned unambiguously to one or other of the
categories in the portmanteau, the more reliable the tagging of the word. Thus in the case of 'hits', we found

page_140

Page 141

a very high proportion of occurrences with portmanteau tags, a fact reflected in the number of tagging errors encountered.
SARA does not allow you to formulate queries in terms of parts of speech without specifying the words in question. If you are interested (say) in finding
cases where an adverb precedes a modal verb and a pronominal subject, as in 'Nor could I', you cannot simply look for this sequence of POS codes.
The only way to find particular part-of-speech sequences is as colligations: you must specify a word, word-POS pair, pattern, or phrase (with
disjunctions as necessary, such as nor | neither), and then sort the solutions to this query according to the POS codes of the surrounding
words. In this manner those solutions which match the desired pattern of codes can be grouped, and the remainder removed by thinning. The main
difficulty with this procedure is that the number of solutions initially downloaded may be very large in comparison with the number which match the
desired colligational pattern.
7.3.2
Some Similar Problems
'Blasts'
A verb which appears to be used in headlines with a similar metaphorical meaning to 'hits' is 'blasts'. How frequent is it, and does it also occur in
headlines as other parts of speech? Is it ever used literally? Use the QUERY BUILDER to design a query which looks for occurrences of 'blasts' which
have VVZ codes, in <HEAD> elements whose TYPE attribute has the value main. Then edit the query to look for occurrences which have
other POS codes.
There are 20 occurrences of 'blasts' coded as a verb, nearly all of them metaphorical. In the second version of the query, if you download all
16 solutions and display them using POS format, you will see that there are only two unambiguous noun uses, which stand out in a different
colour. All the cases with portmanteau codes appear to be verbs. The most frequent verb use is in the phrasal form 'blasts back'.
Whose 'Whom'?
The use of 'whom' is often considered formal, particularly where it is preceded by a preposition. How frequently does 'whom' appear in informal
conversation, as represented by the spoken demographic component of the BNC? Which prepositions usually precede it? And how do these figures
compare with its use in the spoken context-governed component of the corpus, which contains more 'public' speech (see 6.1.2 on page 112)? Use
QUERY BUILDER to search for occurrences of 'whom' where the <CATREF>ALL_TYPE attribute has the value spoken_demographic. Then
design a second query where it has the value spoken_context-governed. In both cases, use POS CODE collating to sort the solutions
according to the first word to the left of the query focus, in order to group those where 'whom' is preceded by PRF (the code for 'of') or PRP
(other prepositions). Mark the solutions in these groups and delete the remainder using the THIN SELECTION option. Then re-sort, again according
to the first word on the left,

page_141

Page 142

but this time using IGNORE CASE collating. This will group the solutions according to the orthographic form of the preceding preposition.
There are only 21 occurrences of 'whom' in the spoken demographic component of the BNC, in 10 of which 'whom' is preceded by a
preposition (from', 'of', 'to', or 'with). In the similarly-sized context-governed component, on the other hand, there are 250 occurrences: the
most common prepositions appear to be the same.
Exit 'Exeunt'
Are the words 'exit' and 'exeunt' still used in stage directions in contemporary English? Rather than the number of occurrences of 'exit' and 'exeunt'
in stage directions, a better indicator of their use might be the proportion of texts containing stage directions in which they appear. You can
calculate this by first performing an SGML Q UERY to find occurrences of the <STAGE> element (see 5.2.2 on page 100) and reading off the
number of texts containing solutions, and then using QUERY BUILDER to find occurrences where 'exit' or 'exeunt' appear within the scope of a
<STAGE> element, again reading off the number of texts .
This question is too specialized to be answered using the BNC. There are only 9 texts which include <STAGE> elements, and only 2 occurrences
of 'exit' or 'exeunt' in such elements, both of them in the same text. These numbers are clearly far too small to permit reliable inferences.

page_142

Page 143

8
Springing Surprises on the Armchair Linguist
8.1
The Problem: Intuitions about Grammar
8.1.1
Participant Roles and Syntactic Variants
In discussing the role of corpora in aiding those 'armchair linguists' who work with intuitive judgements, Fillmore (1992) argues that corpora cannot fully
replace native-speaker intuitions about a language. No corpus can exhaustively indicate what is and is not grammatical. However large, it will not attest
every form which is correct, and will indeed attest many forms which are incorrect in the terms of any descriptive grammar (Aarts 1991). Moreover,
corpus data may not provide appropriate instances to illustrate every kind of grammatical distinction. Noting that dictionaries do not distinguish between
'running risks' and 'taking' them, Fillmore argues that unlike 'running a risk', 'taking a risk' seems to imply that harm may occur as a result of deliberate
action by the subject. Thus, the second of the following sentences seems unacceptable because a car cannot take deliberate action, as it is not a
volitional subject:
A car parked here runs the risk of getting dented.
*A car parked here takes the risk of getting dented.
Looking at corpus examples, however, Fillmore found no cases where 'run' could not apparently be substituted by 'take', or vice-versa. Nor did he find
examples which contradicted his proposed generalization, i.e. where 'take' did not have a volitional subject. The use of corpora, he argues, still leaves a
place for invented examples in order to highlight such distinctions.
Fillmore's examples concern the semantic restrictions on the kinds of participants that can be associated with 'take a risk'. Restrictions on the use of
such expressions may also be syntactic. Discussing the phrase 'bear love', Bolinger (1976: 9) argues that "it is normal to have 'bear' in a relative clause in
an expression like 'the love that I bear them' [...]. A declarative or interrogative is all right with a negation or implied negation: 'I bear them no love'. But a
simple straightforward affirmative declaration is out of the question: '*I bear them love'." Again, corpora may not provide clear evidence to support these
intuitions. What is intuitively 'normal' may nonetheless be too rare to be adequately documented in a corpus, and what is 'out of the question' may on
occasion be found in creative (or linguistics) texts.
Evidence from corpora may nonetheless help armchair linguists to refine their hypotheses. What occurs in the corpus must be accounted for in some
way; the absence of what does not occur is at any rate matter for reflection. This task uses the BNC to look at limits on the use of the expression 'spring
a

page_143

Page 144

surprise', seeing what sorts of participants it involves, and what syntactic forms it takes. Questions to be asked include:
Does 'spring a surprise' require an animate agent as subject, or can an event also spring a surprise ('The rain sprang a surprise on us')?
What sort of syntactic variation is possible? Do we find plurals ('spring surprises'), passives ('a surprise was sprung'), continuous forms ('he was
springing a surprise'), definite uses ('they sprang the surprise on him after lunch'), nominalizations ('the springing of surprises' or 'a springer of surprises')?
In what collocations does it occur? For instance, what modifiers are used with 'surprise' (e.g. 'he sprang a big surprise', 'springer of an unexpected
surprise')?
To look at these questions poses two main problems in the design of appropriate queries:
how to include inflections and other derivates of a base or root form, such as variants in number, tense, aspect, mood, and voice, etc., nominal forms
derived from a verb, verbal forms derived from a noun or adjective, etc.;
how to include variants of phrases where components appear in different orders, or at different distances from each other.
8.1.2
Inflections and Derived Forms
While English morphology has fewer inflections than most other European languages, there are still a significant number to be considered. Most nouns
have two different forms, a singular and a plural, or four if singular and plural possessives are included (SARA counts possessives as two L-words: see
2.2.1 on page 64). Adjectives and adverbs may have a comparative and a superlative in addition to their base form, while regular verbs have an '-s' form
for the third person singular present, an '-ing' participle, and an '-ed' form used in the past tense and as past participle. Many irregular verbs also have
distinct forms for past tense and past participle. In the case of 'spring a surprise', alongside the base forms 'spring' and 'surprise', we need to consider
the inflected forms 'springs', 'springing', 'sprang', 'sprung', and 'surprises'.
English also uses affixation to derive other words from a base form. Thus adverbs, nouns and verbs may be derived from adjectives ('softly', 'softness',
and 'soften' from 'soft'), nouns and adjectives from verbs ('proof', 'proven', 'provenly', 'provable', 'provably' from 'prove'), etc. When examining possible
variants, we may also want to take such derived forms into consideration for instance, is 'springer of surprises' found in the corpus?

page_144

Page 145

We have already seen two ways in which SARA allows you to include a range of alternative forms in a single query. The L OOKUP facility in WORD QUERY
provides an alphabetical list of all the words in the corpus which begin with the same string of characters: thus typing in 'limit' generates a list including
'limitation', 'limitations', 'limited', 'limitedly', 'limiting', 'limits', etc. From this list you can select those items you wish to include in the query (see 2.2.1 on
page 64). This approach is not always adequate, however. Looking up words beginning with the letters 'go' is hardly the way to list forms derived from
the verb 'go': it will include an enormous number of spurious items (from 'goad' to 'gout' via 'Gotham'), and it will not include 'went'.
One way to avoid this difficulty is to specify each of the forms required as an alternative node in Q UERY BUILDER, using OR links (as we did for 'a' and 'an'
in the phrase 'a/an _ too far': see 4.2.2 on page 92). A more economical alternative is to specify a pattern which matches only the required forms. Thus
in the case of 'go', we can reduce the number of spurious matches by specifying that if there is a third character following the letters 'go', it must be
either an 'e' ('goes', 'goer', 'goers'), an 'i' ('going', 'goings'), or an 'n' ('gone'). 'Goad' and 'gout' will now be excluded. We will still have 'Goebbels' and
'Goethe' to contend with, but these too might be eliminated by specifying that any fourth character must be 'e', 'n', 'r' or 's'.
SARA allows you to create queries using patterns in two ways. You can use the P ATTERN option under WORD QUERY to create a list of words which
match the pattern, or else you can use the PATTERN QUERY option to include a pattern in the query directly. In this task we shall use Word Query in order
to design and test particular pattern specifications, and then apply them using the Pattern Query option.
8.1.3
Variation in Order and Distance
In the case of a phrase such as 'spring a surprise', we need to consider not only possible inflected and derived forms of its components, but also
variations in their order and distance. Syntactic variants might involve modifiers ('spring an unpleasant surprise'), passives ('a surprise was sprung on
them'), or relative constructions ('the surprise he sprang on them'). To include these we must design a query where forms of 'spring' and 'surprise' may
occur in either order, and at a variable distance from each other.
QUERY BUILDER allows you to search for different orderings of the components of complex queries by using a T WO-WAY link-type, and to vary the distance
between them by using the SPAN scope option.

page_145

Page 146

8.1.4
Highlighted Features
This task shows you:
how to list words which match a specified pattern using the PATTERN option in WORD QUERY;
how to find occurrences of words which match specified patterns using PATTERN QUERY;
how to combine query components in any order using TWO-WAY links between nodes in Query Builder;
how to restrict the scope of a query to a maximum number of words using the SPAN scope option in Query Builder;
how to save solutions to a file using the LISTING option.
It assumes you already know how to:
adjust default settings (see 1.2.8 on page 54);
carry out a Word Query look up (see 2.2.1 on page 64);
design complex queries using Query Builder (see 4.2.2 on page 91, 5.2.4 on page 105, 6.2.3 on page 117, 7.2.2 on page 133);
adjust downloading procedure in the Too many solutions dialogue box (see 2.2.3 on page 66);
sort and thin solutions (see 3.2.2 on page 78, 2.2.4 on page 68);
calculate collocates (see 3.2.1 on page 76);
save queries (see 2.2.5 on page 70).
8.1.5
Before You Start
Using the VIEW menu PREFERENCES option, set the defaults as follows:

500 characters
MAX DOWNLOAD LENGTH
MAX DOWNLOADS 10
FORMAT Plain
SCOPE Paragraph
VIEW: QUERY and ANNOTATION checked
CONCORDANCE checked
BROWSER: SHOW TAGS unchecked

page_146

Page 147

8.2
Procedure
8.2.1
Designing Patterns Using Word Query: Forms of 'Spring'
To investigate the use of 'spring a surprise', what inflections and derived forms must be considered? The only inflected form of the noun 'surprise' which
seems relevant is the plural, with its '-s' suffix. Inflections of the verb 'spring', on the other hand, include not only the suffixed forms 'springs' and
'springing', but also those with a changed stem, namely 'sprang' and 'sprung'.
If you are familiar with Unix regular expressions, you will understand how a pattern can easily be constructed to match all these forms. If you are not,
you will need to learn only a small amount of their syntax in order to understand how to construct patterns using SARA.randi=
The easiest way to master the syntax of regular expressions is through W ORD QUERY. If you check the P ATTERN option in the dialogue box, Word Query
allows you to type in a pattern and obtain a list of the words in the word index that match that pattern. This provides a good way of testing a pattern and
of seeing what the effect is before you apply it.
We will begin by using Word Query to create a pattern which matches the base, inflected and derived forms of 'spring'. If you are already familiar with
the syntax of regular expressions you may prefer to go directly to 8.2.2 on page 150.
Click on the WORD QUERY button, then check the PATTERN option in the dialogue box. Checking the
1 Pattern option indicates that the input string is to be interpreted as a pattern rather than as a
normal character sequence.
2 Type in the string spring and click on LOOKUP. Where the pattern is a string of letters without variables,
there can be at most one word in the word index that will match it in this case, the word 'spring'.

Any Character: The Dot Symbol

A variable in a regular expression is part of a pattern which can match more than one letter. For instance, to include the forms 'sprang' and 'sprung' as
well as 'spring', we can use a pattern which will match any one character in place of the vowel 'i'.
Change the input string to spr.ng The dot variable matches any single character at all, vowel or
3 consonant.
4 Click on LOOKUP. The list of matching words now contains all the six-letter words in the word index which
have 's p r n g' as their first, second, third, fifth and sixth characters respectively. This includes 'sprang',
'spring', 'sprung', but also 'spreng' and 'sprong' remember that the index includes proper names, acronyms,
foreign words, printing and spelling errors.

Alternation in Pattern Components: Square Brackets

To exclude the forms 'spreng' and 'sprong' from this list, you can either:

page_147

Page 148

specify that the second character must be 'a', 'i', or 'u';

specify that the second character must not be 'e' or 'o'.
We demonstrate both methods.
Change the input string to spr[aiu]ng and click on LOOKUP. The list of matching words now includes
5 only 'sprang', 'spring' and 'sprung'. Square brackets indicate that any one of the characters listed
between them must match for the pattern to be successful.
6 Change the input string to spr[^eo]ng and click on LOOKUP. The list of matching words should be
unchanged. A carat symbol ( <IT+>) immediately following the opening square bracket indicates that any
character other than those listed between the brackets must match for the pattern to be successful.

Whether it is more appropriate to list permitted or excluded alternatives will depend on which is more accurate and/or easy to formulate. In this case
there is little difference, so for the moment let us stick to the list of exclusions just constructed.
Repetition of Pattern Components: the ? and * Characters
The list of matching words now includes the base and irregular inflected forms of the verb 'spring'. It does not yet include the regular inflected forms, i.e.
with the suffixes '-s' and '-ing'.
Change the input string to spr[^eo]ng. * by adding a dot and a star. A character followed by a
7 star matches any number of occurrences of that character (including zero). Thus AB* matches 'A',
'AB', 'ABB' etc. The dot character followed by a star (.*) matches any number of any characters
(including zero). This sequence should not be used as the first component of a pattern.
8 Click on LOOKUP. The list of matching words is now much longer, and clearly includes many spurious items,
from 'spranger' and 'spring-cleaning' to 'springsteen' and 'spryngabedde'. The final .*sequence clearly
allows too many matches.
9 Change the input string to spr[^eo]ng.? A character followed by a question mark matches one
or zero occurrences of that character. Thus the dot character followed by a question mark matches
one or zero occurrences of any character.
10 Click on LOOKUP. The list of matching words is now shorter, being limited to six-and seven-letter words,
but it still includes 'spring-' and 'springy' as well as 'springs'.
11 Change the input string to spr[^eo]ngs? and click on LOOKUP. By substituting the dot with an 's'
followed by a question mark, you have stated that any seventh letter must be 's', and the list of matching
words now includes only

page_148

Page 149

'springs' in addition to the previous six-letter forms. The list still however lacks the '-ing' inflection.
12 Change the input string to spr[^eo]ngs?i?n?g? and click on LOOKUP. This pattern allows each of
the characters 's i n g' to appear after the base stem, provided that they occur in that order, and the
matching words list now includes all the inflections of the verb 'spring'. It still however excludes other
derived forms, notably 'springer' and 'springers'.
13 Change the input string to spr[^eo]nge?r?s?i?n?g? and click on LOOKUP. The list of matching
words now includes 'springer' and 'springers', but has also introduced 'spranger' and 'springen'. You can
 also use the question mark to include prefixed or infixed forms in a pattern. For instance, u?n?
necessary will match both 'necessary' and 'unnecessary'; orienta?t?ed will match both
'oriented' and 'orientated'.

A question worth asking at this point is whether it is really necessary for the pattern to exclude all spurious matches.
Click on 'spranger', then on 'springen', to find out their respective frequencies in the corpus. 'Springen'
14 occurs only once in the BNC, so even if it were included in the solutions to a query, it would be easy to
identify and eliminate it using the SORT and THIN options (see 3.2.2 on page 78, 2.2.3 on page 66).
'Spranger', however, occurs ten times, so it might be better to make the effort to exclude it from the
pattern. 'The problem is how to do so without excluding other similar forms, such as 'springer' and
'springers'.

Alternative Patterns: The Vertical Bar Symbol

We can exclude 'spranger' but not 'springer' by only allowing [iu] as variants for the base vowel in our pattern, specifying 'sprang' as a separate
alternative.
Change the input string to spr(ang | [iu]nge?r?s?i?n?g?) and click on LOOKUP. The
15 matching words list now includes only the forms 'sprang', 'spring', 'springen', 'springer', 'springers',
'springing', 'springs', and 'sprung'. In the next section we will use this pattern to include variants of 'spring'
in a PATTERN QUERY. The | symbol indicates a disjunction, i.e. words in the list must match either the
pattern preceding the | or that following it. The parentheses indicate the part of the string for which
alternatives are given. The | symbol is particularly useful where there is one form (in this case
'sprang') which radically differs from the others to be included, as is often the case with the past
tense form of irregular verbs.
16 To avoid having to retype the pattern you have so painstakingly designed, click on COPY to copy the input
string to the clipboard for later re-use, then on CANCEL to close the Word Query dialogue box.

page_149

Page 150

8.2.2
Using Pattern Query
Let us now construct a PATTERN QUERY which will include these inflected and derived forms of the verb 'spring'.
Why not just use WORD QUERY, selecting all the relevant forms from the word index as alternatives? The main reason is economic: Word Query allows
you to include multiple selections from the index in a single query (see 2.2.2 on page 65), but if too many are chosen, the server may not be able to
handle them all. The Pattern Query option, on the other hand, sends the server a pattern which is treated as a single alternative regardless of the number
of words which match it. It thus makes fewer demands on resources. This is particularly important where, as in this task, you want to design a complex
query involving more than one pattern. Note that Pattern Queries are very much slower where they include disjunctions at the start of the
pattern. For this reason you should always use a pattern like spr(ang|ing) rather than sprang|spring.
Other procedures which may overload the server are multiple OR disjunctions in QUERY BUILDER (see 4.2.2 on page 92), and multiple selections
from the POS list using POS QUERY (see 7.2.1 on page 131).
Click on the PATTERN QUERY button on the toolbar (or select PATTERN under NEW QUERY from the FILE
17 menu). The Pattern Query dialogue box will be displayed.
18 Press SHIFT+INSERT to paste the pattern you designed in the last section from the clipboard. The clipboard
should still contain the required string: spr(ang|[iu]nge?r?s?i?n?g?). Type it in, if it does
not. The text of the query will wrap automatically at the end of the line.
19 Click on OK to send the query to the server. The Too many solutions dialogue box will be displayed.
20 Click on OK to download the first 10 solutions. You will see that these already include two instances of
the phrase 'spring a surprise'.

8.2.3
Varying Order and Distance Between Nodes in Query Builder
We are now in a position to build a complex query combining the pattern for forms of 'spring' with one for forms of 'surprise'. We need to design a
query which:
finds cases where 'spring' and 'surprise' co-occur, including inflected and derived forms of either word (for instance 'springer of surprises');
finds cases where the two forms are not adjacent ('spring a big surprise' as well as 'springing surprises');
finds cases where these forms appear in either order (both 'spring a surprise' and 'the surprise he sprang').

page_150

Page 151

Using the QUERY BUILDER, you can satisfy these requirements by specifying two patterns as content nodes, to be joined in either order, and restricting the
scope of the query to a small number of words.
Specifying the Content Nodes
First, let us create the two content nodes in the query, one corresponding to forms of 'spring', and one to forms of 'surprise'.
Click on the QUERY BUILDER button on the toolbar to display the Query Builder.
21
22 Click on the downward branch of the empty content node to create a second content node.

In the first node, we need to provide a pattern which will include inflections and derived forms of the base form 'spring'.
Click on the first node, and select EDIT then PATTERN. The Pattern Query dialogue box will be displayed.
23
24 Press SHIFT+INSERT to paste the pattern you designed in the last section from the clipboard. The clipboard
should still contain the required string: [spr (ang|[iu] nge?r?s?i?n?g?). Type it in, if it
does not.
25 Click on OK to return to the Query Builder. You will see that the pattern has been inserted in the first
content node. The curly brackets surrounding the string indicate that it is to be interpreted as a pattern.

We now need to design a Pattern Query for the second node, corresponding to forms of 'surprise'. As a regular noun, this may have the forms 'surprise'
and 'surprises'. To make sure that the pattern is correct, we will again use Word Query to design and test it before copying it to the Pattern Query. You
can do this without leaving the Query Builder.
Click in the second node and select EDIT then WORD to display the Word Query dialogue box.
26
27 Tape in the string surprises? and click on LOOKUP, making sure that the PATTERN box is checked. You
will see that the matching words list includes only the desired forms.
28 Click on COPY to copy this pattern to the clipboard, then on CANCEL to return to the Query Builder.
29 Click in the second content node again, this time selecting EDIT then PATTERN.
30 Paste the pattern from the clipboard into the Pattern Query dialogue box using SHIFT+INSERT.
31 Click on OK to insert this component in the Query Builder node.

page_151

Page 152

Linking the Content Nodes

We now need to specify the link-type between the two content nodes. The downward arrow joining the nodes indicates that this currently has the default
value of ONE-WAY, meaning that the 'surprise' pattern follows the 'spring' pattern. As we also want to find cases where these nodes occur in the opposite
order, we must change the link-type to TWO-WAY.
Click on the link between the two nodes and select LINK TYPE then TWO-WAY. The link between the two
32 nodes will be displayed as a two-way arrow, indicating that the contents of the nodes in question may
occur in either order, and not necessarily adjacently (see 4.2.2 on page 92).

Scoping the Query

Lastly, we need to specify the scope of the query the limit within which all the content nodes must occur. The default scope ( <BNCDOC>) requires only
that 'spring' and 'surprise' be present in the same BNC document, which is clearly excessive.
In selecting a scope we need to find a balance between, on the one hand, maximizing recall (ensuring that all the relevant solutions are found), and on
the other hand maximizing precision (ensuring that only relevant solutions are found). For instance, if we restrict the scope to a single sentence (an <S>
element), this will exclude any cases where 'spring' and 'surprise' occur in different sentences, along such lines as 'They were unaware they were in for a
surprise. He sprang it on them after lunch.' If, on the other hand, we use a larger element as scope, such as a paragraph (a <P> element), we will
increase the risk of spurious solutions as well as excluding any occurrences in spoken texts, where <P> elements do not occur.
Where it is difficult to identify an appropriate element, you can specify the scope as a number of words, using the Query Builder S PAN option. This
allows you to indicate the maximum number of L-words within which all the content nodes must be found.
Click on the scope node and select SPAN. The Span dialogue box will be displayed.
33
34 Type in the number 10 and click on OK. You will see that the scope node now contains the number 10,
meaning a span of 10 L-words. The maximum span permitted is 99 L-words. The default value is 5
L-words.
35 Check to see that the 'Query is OK' message is displayed, then click on OK to send the query to the
server. After a while, the Too many solutions dialogue box will be displayed, telling you that there are 62
solutions.
36 Click on the DOWNLOAD ALL button, then on OK to download all the solutions. Provided that you checked
QUERY in the VIEW PREFERENCES options

page_152

Page 153

(see 8.1.5 on page 146), the query text will be displayed at the top of the solutions in its CQL form. You will see that:
each Pattern Query is surrounded by curly brackets;
the Two-way link is indicated by a sharp sign;
the scope is indicated at the end of the query by a slash followed by the number of words selected as span .
8.2.4
Checking Precision with the Collocation and Sort Options
We now need to remove any spurious solutions from the display i.e. which do not contain some form of the phrase 'spring a surprise'. Ultimately, the
only way of doing this is by inspecting the solutions, but you can facilitate the process using the COLLOCATION and SORT options.
You may remember from designing the component of the query relating to forms of 'spring' that this pattern also matched the spurious item 'springen'
(see 8.2.1 on page 148). Before going any further, let us make sure that this form is not included among the solutions to the query.
Because the second node in the query, {surprises?}, is highlighted as the query focus (see 8.3.1 on page 157), and its distance from the first
node varies, you cannot locate 'springen' by sorting the solutions. You can however find out if it occurs within a span of up to 9 L-words by using the
COLLOCATION option under the QUERY menu (see 3.2.1 on page 76). In the example here, a collocation span of 9 is equivalent to a span of 10 as a
scope in Query Builder. This is because SPAN is measured differently in the Collocation option from the way it is measured in the Query Builder.
In the Query Builder, a span is the number of L-words within which the entire query must be satisfied; that is, it must include everything from
the beginning of the first node to the end of the query focus. Thus, if the query focus is a single word, a span of 2 will include the query focus
and one word either to the left or to the right of it. In the Collocation option it is assumed that the query focus will normally be a single L-
word, and the span is calculated as the distance of the collocate from the leftmost word of the query focus, not counting the latter. Thus a
collocation span of 2 will include the two words to the left and the two words to the right of the leftmost word of the query focus a total of
five words as measured in Query Builder. In the example here, we wish to find occurrences within the 9 L-words which precede the query focus
a span of 9 in Collocation, and one of 10 in Query Builder.
Select QUERY then COLLOCATION.
37
38 Type the string springen in the COLLOCATE box and set the span to the maximum value of 9 words.
39 Click on CALCULATE. The collocation frequency of 'springen' is 0, meaning that this form does not occur as
a collocate.

page_153

Page 154

Click on CLOSE to return to the solutions display.

40
Sorting and Thinning the Solutions
A glance through the solutions suggests that active forms (where 'spring' precedes 'surprise') are much more frequent than passive or relative ones
(where 'surprise' precedes 'spring'). For this reason it would seem most appropriate to sort the solutions by the left. This will group recurrent articles and
modifiers preceding forms of 'surprise' (such as 'a' and 'big'), together with the varying forms of 'spring' occurring before the same article or pre-
modifier.
Click on QUERY then SORT, and specify the Primary key as LEFT with a span of 5 and I GNORE CASE collating.
41
42 Click on SORT to sort the solutions. Note that when sorting Plain format displays, span is counted in
terms of orthographic (blank-delimited) words rather than L-words.
43 Scroll through the solutions to identify spurious ones. You may find a number of cases where 'spring' refers
to the season, or 'springer' to a breed of dog.
44 Double-click on these solutions (or press the space bar) to mark them, then eliminate them using THIN and
REVERSE SELECTION from the QUERY menu.

Saving the Query

Given the amount of hard work you have dedicated to its design, it may be a good idea to save this query for future reference.
Click on the SAVE button on the toolbar (or select SAVE from the FILE menu), and type in an appropriate
45 mnemonic as the filename. The query will automatically be saved with the extension .sqy (see 4.2.2 on
page 94).

8.2.5
Saving Solutions with the Listing Option
You should now inspect and categorize the remaining solutions. You may find this easier using a printout. Printing directly using the PRINT button on the
toolbar will provide only one line of context for each solution (see 3.2.3 on page 81), which may not be enough to see both 'spring' and 'surprise'. It may
therefore be better to save the solutions with a larger context and then print them using another application. The LISTING option under the QUERY menu
allows you to save the solutions to a query together with the full downloaded context. You can then retrieve the Listing file into a word processor, or
SGML-aware browser, and edit or print it.

page_154

Page 155

Select QUERY then LISTING. The Listing SAVE AS dialogue box will be displayed, proposing the current
46 window title as filename, with an .sgm extension. SARA automatically assigns .sgm extensions to
Listing files.
47 Click on OK to save the solutions. You will be returned to the solutions display. Listing files save
solutions in the currently displayed order, with the currently selected format and scope (see 1.2.8 on
page 55, 5.2.3 on page 103). PLAIN format listings consist only of the words and punctuation. POS
and SGML format listings show all SGML markup, with the angle brackets of the original SGML
tags replaced by square brackets. CUSTOM format listings follow the indications in the Custom
linefmt.txt file, which you can edit using the CONFIGURE option (see 9.2.2 on page 165).
Remember that you can also save individual solutions by copying them to the clipboard using the
COPY option (see 1.2.7 on page 54). Copying to the clipboard only preserves Plain, Custom, or
SGML format, however: POS format is converted to Plain.
48 Without exiting from SARA, switch to a word processor, retrieve your .sgm file, and print it.
49 When you have finished, exit from your word processor and return to SARA.

The following extract from an .sgm listing file (saved from a Plain format display with Paragraph scope) shows the markup inserted in listing files. Each
solution is marked up as an SGML <HIT> element, whose attributes specify the text identifier and sentence number displayed on the status bar. The
query focus and the text to the left and right of it are indicated by <FOCUS>, <LEFT> and <RIGHT> tags respectively.
<hit text=A05 n=386><left>
Glasser orders his events thematically,
while also wanting to tell a story and to
spring <focus>surprises.<right> Charlie's
departure is the first of several, and this
 event is succeeded by the announcement of
a further theme when the rabbi's thunderings
pass over the heads of his congregation.
</hit>

Study your printout of the solutions (or scroll carefully through the solutions on screen) to see:
50

whether there are any more spurious solutions to be deleted;

whether the subject of 'spring' is always animate;

page_155

Page 156

how frequently 'surprise' is pre-modified, and whether these modifiers fall into any particular class;
what sorts of syntactic variants are present, e.g. passive and relative constructions, nominalization with 'springing', 'springer', etc.
Results
Inspecting the solutions shows that:
There are quite a few spurious solutions. The main reason appears to be the frequency of the noun 'spring', as in ''to her great surprise she is asked to
the Spring ball''. It would be possible to exclude these occurrences by limiting the Pattern Query and using POS QUERY to search for those forms
which need to be specified as verbs (see 7.2.1 on page 131). We would thus have three alternative content nodes:
{spr([au]ng|inging)}
"spring"=VVB|"spring"=VVI
"springs"=VVZ

By no means all the solutions have animate individual or institutional agents as their subjects. Along with Mr Kinnock, Calvin Smith, and a number of
racehorses and football teams, we find discussion, bad news, a Scottish burn, the onset of spring, the domestic financial system, and technology can all
spring surprises.
Relatively few instances use an 'on' phrase to specify the victim of the surprise ("we decided to spring a little surprise on two friends"). It is far more
common for this participant to be left unspecified.
'Spring' most frequently appears in the active singular, with 'surprise' directly preceded by the indefinite article 'a' (21 occurrences). The article is omitted
in two cases from newspaper headlines, as is typical in such contexts.
Pre-modifiers of 'surprise' generally indicate size or pleasantness: 'a little surprise', 'a major surprise', 'maximum surprise', 'a dazzling surprise', 'a This is
Your Life surprise', 'a nasty surprise', and 'the cruellest surprise (of all)'. With the plural 'surprises' we find 'some', 'any' and 'many', as well as 'one of the
biggest surprises', and 'unpleasant', 'ghastly' and 'delicious' ones.
'Surprise' occurs both as a countable and as an uncountable or mass noun, as in spring maximum surprise.
There are only a few cases where 'surprise' precedes 'spring', an involving post-modifying relative or infinitive structures: "Norwich's victory was a

page_156

Page 157

triumph for the tactical surprise sprung by manager Walker", "a new and exciting field of research which no doubt has many surprises to spring", and "if
you've got any more surprises waiting to spring on us, then warn me about them now". There are no simple passives.
There are no cases with continuous aspect: 'springing' only occurs as a gerund following a preposition.
There are no cases of nominalization with 'springer' or 'springers'.
With respect to the questions posed at the beginning of the task, there would seem to be plenty of evidence that 'spring a surprise' does not require an
animate subject. As far as syntactic restrictions are concerned, there is little evidence of passive or continuous forms, or of nominalized derivates.
8.3
Discussion and Suggestions for Further Work
8.3.1
Using Two-way Links
Where a TWO-WAY link is used to join two content nodes in the Query Builder, the query focus always corresponds to the content of the lower node,
regardless of the order in which matches for the two nodes appear in the text. This means that according to which of the two nodes is placed above,
and which below, different displays of solutions will be provided to otherwise identical queries. When using a Two-way link, you should therefore think
carefully which node you wish to be treated as the query focus, and consequently used as a basis for sorting solutions and for calculating collocates.
As well as providing a different query focus, you should also note that inverting the nodes may provide different numbers of solutions where there is
more than one match for either node within the scope specified. Roughly, SARA interprets any query involving a one-way or two-way link as "find all
the cases matching the lower node where there is also a match for the upper node". This means that if we take as our scope the part of the previous
sentence in inverted commas, and design a query with a two-way link where the first node contains the word 'the' and the second one 'node', there will
be two solutions, with 'node' as the query focus. If, on the other hand, the two nodes are inverted, there will be three solutions, with 'the' as the focus.
One case where this feature can be useful is in obtaining information from the text headers relative to a group of solutions. For instance, you may want to
know what text-types are represented in the solutions to, say, a Phrase Query for 'time immemorial'. This phrase, you may remember, occurs 45 times in
the BNC (see 3.3.2 on page 85). You could, of course, use the Browse option to examine the attributes of the <CATREF> element in the header of
each text (see 5.2.4 on page 105). However this would be a tedious process. A more

page_157

Page 158

practical option is to join the Phrase Query to an SGML query for <CATREF> with a Two-way link.
Invoke the QUERY BUILDER, and join a first node containing the SGML Query <CATREF> to a second
51 node containing the Phrase Querytime immemorial with a Two-way link.
52 Leaving the scope as <BNCDOC>, click on OK to send the query to the server. The Too many solutions
dialogue box will be displayed, stating that there are 45 solutions in 40 texts.
53 Download one solution per text. The solutions will be displayed with 'time immemorial' as the query focus.
54 Click on the QUERY BUILDER button to start a new query.
55 Design exactly the same query as before, but with the two content nodes inverted. The upper node should
now contain the Phrase Query and the lower node the SGML Query.
56 Click on OK to send the query to the server. The Too many solutions dialogue box will be displayed,
stating that there are 40 solutions in 40 texts.
57 Download all the solutions, and select SGML under Q UERY OPTIONS to display them in SGML format. The
solutions will be displayed with the <CATREF> element highlighted as the query focus.
58 Select TILE from the WINDOW menu to display both sets of solutions. As the solutions in both windows are
in the same order, you can see what text-type each of the solutions with 'time immemorial' comes from by
examining the corresponding solution in the other window. As the two queries find different numbers of
solutions, it is essential to thin the solutions to the first query to ONE PER TEXT in order to obtain an
exact correspondence between the concordance lines in the two displays.

8.3.2
Some Similar Problems
Taps
Suppose that, having an interest in plumbing or the interception of telephone calls, you want to find all the occurrences of the word 'tap' and its
inflections in the BNC. Use WORD QUERY to design an appropriate pattern. If you type in the string tapp?i?n?g?e?d?s?, you will find that as
well as 'tap', 'tapped', 'taps', and 'tapping', the list of matching words includes 'tape', 'taped', 'tapes', 'tapie', 'tapin' and 'taping'. However
relevant many of these may be to the activity in question, they are not inflections of the verb 'tap'. The problem is a common one: verbs whose
base form ends in a short vowel plus consonant usually double that consonant in their '-ing' and '-ed' inflections. One answer is to specify
separate alternatives for the forms with single and double consonants: for instance tap (s?|p[ie].*)

page_158

Page 159

Spelling Babysitting
Are forms of 'baby-sit' more commonly hyphenated or unhyphenated in the BNC? Within QUERY BUILDER, design a Pattern Query which includes
both hyphenated and unhyphenated forms of 'baby-sit', by placing a ? character after the hyphen. As this query concerns orthography, you
should limit it to written texts, by using the <TEXT> element as scope.
Sorting the solutions by the query focus shows that only about a quarter are hyphenated.
Running Risks
Is there any evidence in the BNC to support Fillmore's hypothesis that, unlike 'running risks', 'taking risks' requires a volitional subject (see 8.1.1 on
page 143)? As in the 'spring a surprise' example, you should use QUERY BUILDER to specify patterns for each of two content nodes, with a two-
way link, and a limited span as scope, since both 'run'/'take' and 'risk' may be inflected, occurring in either order, and at a variable distance
apart. For the patterns with forms of 'run'/'take', you need to provide the forms 'ran'/'took' as separate alternatives (see 8.2.1 on page 147).
There are 418 occurrences of forms of 'run' and 'risk' within a span of 5 words, and 971 of forms of 'take' and 'risk' within the same span.
Examining the first 50 solutions to each query, we find that both are fairly precise, there being relatively few cases not involving the
compounds 'take/run risks', such as "Crawl out onto the road and risk getting run over". While the subjects of forms of 'take risks' all appear
to be volitional, there are several solutions with forms of 'run risks' that have non-volitional subjects, such as "a meal that didn't run any
cholesterol risks"; "things that have run the risk of acquiring the patina of nostalgia"; "the reforms run a very high risk of being set back".
There also appear to be differences in the colligational patterns associated with the two sets of forms, the plural 'risks' being more frequent
with 'take', and the phrase 'the risk of' with 'run'. This suggests that the choice of verb may be determined by colligational as well as semantic
criteria.
Bearing Love
What evidence is there in the BNC to support Bolinger's claim that the expression 'bear love' is found in negative and relative constructions but not
simple declaratives (see 8.1.1 on page 143)? To exclude people who love bears or bores, you can use POS QUERY to specify that 'love' should be a
noun.
Using QUERY BUILDER to join a Pattern Query matching forms of 'bear' to a POS Query for the noun 'love', in either order and within the scope
of an <S> (sentence) element, finds 48 solutions. Of these, only 5 seem relevant. While 4 of them occur in relative constructions, along the
lines of 'the love they bore him', there is also one example of a simple declarative, "because she bore them so much love", suggesting that
Bolinger's claim may not be fully accurate.
Looking Up
Is it more common to 'look something up' in a dictionary or to 'consult' one? Which do you do with a corpus? First design a query to find

page_159

Page 160

out how often a form of 'look' and a form of 'dictionary' co-occur in a span of 10 words. (You need to specify the pattern
dictionar[iy]e?s? in order to include the plural 'dictionaries'.) Then formulate a second query to include forms of 'consult'. After you
have looked at the solutions, edit the two queries to replace the pattern for 'dictionary' with one for 'corpus' (bearing in mind that the latter
has both 'corpora' and 'corpuses' as plurals: see 2.1 on page 63).
There are 79 cases where a form of 'look' co-occurs with a form of 'dictionary' within a span of 10 words, some 50 of them involving the
phrasal verb 'look up'. In contrast there are only 8 where a form of 'consult' does. Neither 'look up' nor 'consult' occurs with forms of 'corpus'
so what do you do with one?
Many Happy Returns
Try looking up your birthday in the corpus, bearing in mind that dates can be written and said in a variety of ways, with the day before or after the
month, and with varying distances between them ...

page_160

Page 161

9
Returning to More Serious Matters
9.1
The Problem: Investigating Positions in Texts
9.1.1
Meaning and Position
One of the main differences between the salutations 'Hello' and 'Goodbye' is that the former tends to open conversations, while the latter closes them.
The word 'yet' generally means something different when it appears at the beginning and at the end of clauses (compare 'Yet I don't miss her' with 'I
don't miss her yet'). There are many other words and phrases whose meanings are closely linked with their position within the structural components of a
text the paragraphs, speeches, sentences, or clauses of which it is composed. This task looks at ways in which you can use SARA to examine usage in
particular positions. This will involve searches that take account of the beginnings and ends of such units as sentences, utterances, paragraphs, sections,
conversations, etc. (for an overview of the units marked up in the BNC see 6.1.2 on page 112).
The task focuses on occurrences of the adverbs 'anyway' and 'anyhow'. Most dictionaries regard these two words as virtually synonymous. With one
exception, they both have exactly the same three senses. One of these senses is roughly equivalent to 'besides', a second corresponds to 'nevertheless',
whereas a third is generally described in pragmatic terms, as a means of changing or cutting short a topic in speech, for example when returning from a
digression to the main point, cutting out minor details, or closing down the conversation. (The exception is that 'anyhow' has an additional specific sense
'in a disorderly manner'.)
One distinctive aspect of the third use described is that almost all the examples provided by dictionaries place 'anyway/anyhow' as the first word in the
sentence: "Anyhow, let's forget about that for the moment" (Oxford Advanced Learners Dictionary); "Anyway, in the end I didn't wear your jacket"
(Cambridge International Dictionary of English); "Anyway, I'd better let you have your dinner" (Collins COBUILD Dictionary). In this task we
begin by checking whether or not this third sense of 'anyway' and 'anyhow' is specific to appearances of the words at the beginning of spoken sentences
and whether the two forms are used interchangeably in this position.
9.1.2
Laughter and Topic Change
Speakers frequently digress from the main line of talk when they make a mistake, or see a possibility for teasing or joking. Such shifts of 'footing', as
Goffman (1979) calls them, often lead to laughter, leaving participants with the task of then shifting the back to more serious matters. In the second part
of this task, we look at some ways in which this is achieved, examining cases where

page_161

Page 162

sentence-initial 'anyway' or 'anyhow' are used to change the topic following laughs or laughing speech. In particular, the task investigates whether these
words are typically produced by the participant who has just laughed, or by another speaker.
9.1.3
Highlighted Features
This task shows you:
how to design an SGML QUERY to find the beginning or end of the structural units encoded as elements in the BNC;
how to design an SGML QUERY to find non-verbal and non-vocal events (such as laughs and applause) and features of voice quality (such as laughing
and whispering);
how to display utterance boundaries and speaker codes, overlapping segments, non-verbal and non-vocal events, and other features of spoken texts by
configuring the CUSTOM display format;
how to sort solutions by reference to these features.
It assumes you already know how to:
adjust default settings (see 1.2.8 on page 54);
carry out an SGML Query (see 5.2.2 on page 100);
use the Pattern option in Word Query to find all the forms which match a pattern (see 8.2.1 on page 147);
design complex queries using the Query Builder (see 4.2.2 on page 91, 5.2.4 on page 105, 6.2.3 on page 117, 7.2.2 on page 133, 8.2.3 on page 150);
adjust downloading procedure in the Too many solutions dialogue box (see 2.2.3 on page 66);
sort and thin solutions (see 3.2.2 on page 78, 2.2.4 on page 68);
save and re-open queries (see 2.2.5 on page 70);
save listings of solutions (see 8.2.5 on page 154).
9.1.4
Before You Start
Using the VIEW PREFERENCES option, set the defaults as follows:

1500 characters
MAX DOWNLOAD LENGTH
MAX DOWNLOADS 25
FORMAT Custom
SCOPE Maximum
VIEW: QUERY and ANNOTATION checked
CONCORDANCE checked
BROWSER: SHOW TAGS unchecked
 page_162

Page 163

9.2
Procedure
9.2.1
Searching in Sentence-Initial Position: 'Anyway' and 'Anyhow'
Frequencies in Speech
Let us begin by finding out how frequent these two forms are, first in the corpus in general, and then in spoken texts in particular.
Click on the WORD QUERY button on the toolbar, and check the PATTERN option in the dialogue box.
1
2 Type in any (how|way) and click on LOOKUP. The matching words list will contain the two forms 'anyhow'
and 'anyway'.
3 Click on 'anyhow' to find out its frequency in the corpus, then do the same for 'anyway'. You will see that
'anyway' is about 25 times more frequent than 'anyhow' (12,232 as opposed to 471 instances).
4 Copy the input string to the clipboard, then click on CANCEL to close the dialogue box.

If you look up the frequency of a word using Word Query, you can only find its total frequency in the entire corpus. To find its frequency in a subset of
the corpus, you must use QUERY BUILDER to restrict the scope of your query (see 5.2.4 on page 105).
Click on the QUERY BUILDER button on the toolbar. The Query Builder will be displayed.
5
6 Click in the scope node, then select SGML. The SGML dialogue box will be displayed.
7 Check that the SHOW HEADER TAGS box is unchecked, then scroll through the element list and select
<STEXT>.
8 Click on OK to insert this selection in the Query Builder scope node.
9 Click in the content node, and select EDIT then WORD. The Word Query dialogue box will be displayed.
10 Check the PATTERN box and paste in the stringany (how|way) from the clipboard, then click on
LOOKUP.
11 Select 'anyway' from the matching words list and click on OK to insert it in the Query Builder node.
12 Check that the 'Query is OK' message is displayed, then click on OK to send the query to the server. The
Too many solutions dialogue box will be displayed, stating that there are 5221 occurrences of 'anyway' in
584 texts.

Now find the frequency of 'anyhow' in spoken texts.

Click on CANCEL to return to the Query Builder.
13

page_163

Page 164

Click in the content node and select EDIT. You will be returned to the Word Query dialogue box, still
14 highlighting your selection of 'anyway' in the matching words list.
15 Change the selection to 'anyhow', then click on OK to insert it in the Query Builder node.
16 Click on OK to send the revised query to the server. The Too many solutions dialogue box will again be
displayed, stating that there are 178 occurrences of 'anyhow' in 81 texts.

In speech, the difference in the frequency of the two forms is thus even greater than in writing, with 'anyway' 30 times more frequent than 'anyhow'.
Given the size of this difference, it is perhaps surprising that some dictionaries list 'anyway' as a variant of 'anyhow', rather than vice-versa.
Sentence-Initial Position
Let us now move on to examine occurrences of 'anyway' and 'anyhow' at the beginning of sentences in spoken texts. One obvious way of doing this
might be simply to search for cases where 'anyway' or 'anyhow' are capitalised. Although this is technically possible using a PHRASE QUERY (see 10.2.4 on
page 184), it cannot be relied on to find all and only sentence-initial occurrences of these words. In the BNC spoken component, the first word of a
sentence is not always capitalised, while in the written component, it is common for all the words in headings etc. to be capitalized. We will instead use
the SGML tagging in the corpus, which unequivocally identifies the start of every sentence.
Click on CANCEL to return to the Query Builder.
17
18 Click on the upward branch of the 'anyhow' content node to create a further node above it.
19 Click in this node, and select EDIT then SGML.
20 In the SGML dialogue box, select the <S> element (which represents a new sentence), then click on OK
to insert it in the Query Builder node. The two nodes are linked by a downwards arrow, indicating the
default ONE-WAY value for the link-type. To find only cases where a new sentence tag is immediately
followed by 'anyhow', we must change the link-type to NEXT.
21 Click on the link and select LINK TYPE then NEXT. The link will be shown as a thick vertical bar.
22 Click on OK to send the query to the server.
23 Read off the number of solutions from the Too many solutions dialogue box, then click on C ANCEL to
return to the Query Builder.
24 Click in the second content node and select EDIT. You will be returned to the Word Query dialogue box.

page_164

Page 165

Select 'anyway', then click on OK to insert it in the node.

25
26 Click on OK to send the revised query to the server, and wait for the new number of solutions to be
displayed in the Too many solutions dialogue box. While there are 53 occurrences of sentence-initial
'anyhow' in spoken utterances, from 28 texts, there are 1009 occurrences of sentence-initial 'anyway',
from 279 texts.

Roughly half the occurrences in spoken utterances of these forms thus occur in sentence-initial position. The relative proportions of 'anyway' and
'anyhow' are still around 20:1, suggesting that position in the sentence makes little difference to the relative frequency of the two forms.
9.2.2
Customizing the Solution Display Format
We now want to find out if these sentence-initial uses typically mark a change in topic. Let us begin by looking at sentence-initial 'anyway'. Clearly, to
examine 1009 solutions individually would take an inordinately long time, so we will limit ourselves to the first 50.
In the Too many solutions dialogue box, change the DOWNLOAD HITS number to 50, click on the INITIAL radio
27 button, then on OK. If you have correctly selected C USTOM as the default display format (see 9.1.4 on
page 162), and have not modified the default configuration, you will see that there is a vertical line before
many of the 'anyway's in the solutions display. This line indicates a new utterance (i.e. a change of
speaker).

To see whether 'anyway' generally introduces a new topic, we must compare what occurs before and after 'anyway' in these solutions. A written
transcript of conversational data can never fully capture the information which was available to the participants, but it will be easier to understand what is
going on if you show more of the available contextual information. It may be particularly useful to display:
overlapping speech
A plain transcript where one utterance is shown following another hides those cases where speakers talk at the same time, overlapping with each other.
pausing, gaps and unclear speech
A plain transcript does not show pauses, which may indicate that the previous topic has exhausted itself or that a section has come to a close. Nor does
it show where personal references have been omitted from the transcription, or where portions of the speech were too unclear to transcribe reliably.

page_165

Page 166
non-verbal and non-vocal events
A plain transcript does not show non-verbal events, such as laughs, or non-vocal ones, such as applause, which may in fact mark the end of a topic or
section of the talk.
All of these features are explicitly tagged in the BNC, along with others. Choosing the C USTOM display format allows you to specify your own rules for
the display of such features.
Select QUERY OPTIONS then click on CONFIGURE in the FORMAT section. The LINE (linefmt.txt) and PAGE (pagefmt.txt) files will be displayed,
listing the current rules for custom display when in Line mode and Page mode respectively.
You may now edit these files, changing these rules or adding new ones.
The format of the two files is identical. Each consists of a series of rules, one per line. Each line begins with the name of an SGML element, followed in
the simplest case by just a single replacement string enclosed in double quotes. For example, if you are currently using the default versions, you will see
that the linefmt.txt file begins with the rule:
p "|"

This indicates that in Line mode, the replacement string ''|'' (vertical bar) is to appear at the beginning of every < P> element, i.e. at the start of each new
paragraph.
Since rules can only occupy a single line, a special code is needed to indicate where new lines are required within the replacement string. The convention
used is to represent new lines by the sequence \n. The sequence \t can be used similarly to insert a TAB character. For example, you will see that
several of the elements in the Page file (<DIV1>>, <DIV2>, etc.) have a replacement string consisting of the text "\n" while the <P> element in this file
has a replacement string "\n\t". This means that sections will always begin on a new line in Page mode displays, while paragraphs will begin with both
a new line and an indent.
Important information about an SGML element occurrence is often represented by its attributes. You can display the value of any SGML attribute by
including its name in the rule together with the special symbol %s in the replacement string. For example, towards the bottom of the Page file, you will
see a sequence of rules like the following:
gap desc "[%s]"

This indicates that when a <GAP> element is to be displayed in Page mode, the value of its DESC attribute should be inserted within square brackets. You

page_166

Page 167

can mix attribute values with any other text, simply by specifying where each is to appear in the replacement string by a distinct %s symbol. For
example, the default Page format file has a rule like this:
u who "\n{%s}: "

This states that in Custom format the start of each utterance (< U>) should begin on a new line, followed by whatever value its WHO attribute has between
curly braces, a colon, and a space. If you want to display more than one attribute value for a given element, you can do so simply by supplying more
than one name, and including a %s sequence for each value. For example, the following rule would display both description and duration for every
<VOCAL> element:
vocal desc dur "[%s for %s seconds]"

Finally, you can also supply a second replacement string if you want to specify some additional formatting for the end of an element, for example to
output a string at its end. An example of this procedure is given below for the <TRUNC> element.
In the PAGE file, change thepause and unclear lines to read:
28

pause dur "(.%s)"

unclear "(...)"

and add the following lines at the end of the file:

ptr "^"
trunc "" "-"
&hellip "..."

These changes will affect the Page display as follows:

each <PAUSE> element will be displayed as a dot followed by its duration, in parentheses (in the BNC, pause duration is only specified where over 5
seconds);
each <UNCLEAR> element will be displayed as three dots in parentheses;
each <PTR> element (which in spoken texts indicates the beginning or end of overlapping speech) will be displayed as a carat;
a dash will be displayed after each <TRUNC> element to indicate that the content of that element is truncated. (Note that an empty string has to be
supplied for the start-tag as well.)

page_167

Page 168

each entity reference representing a horizontal ellipsis will be displayed as three dots. (Ellipses pose a particular display problem in that there is no single
Windows character corresponding to their conventional representation as a sequence of three dots.) In the BNC all characters outside the basic
alphabet are displayed as SGML entity references. These take the form of a brief mnemonic preceded by an ampersand (&) and followed by a
semi-colon (;). Thus the string &bquo; indicates the opening of a quotation, while &equo; indicates the end of a quotation. Apart from
&hellip;, SARA converts most other entity references to appropriate screen characters automatically, other than in SGML FORMAT. A full
list of entity references used in the corpus is included in the BNC Users' Reference Guide.
Copy all the lines from u who "{%s}: " onwards to the clipboard, then paste them into the LINE file,
29 replacing the previous rule for <U> in Line mode. This will provide the same low-level information in Page
and Line mode displays.
30 Click on OK to save the edited format files. The formats you have defined will be used for all subsequent
displays in Custom format. You will be returned to the Query Options dialogue box.
31 Click on OK to return to the solutions display. The solutions will be re-displayed in your new Custom
format, with Maximum scope, as specified in the VIEW PREFERENCES options (see 9.1.4 on page 162).
MAXIMUM scope displays all the complete sentences contained within the MAX DOWNLOAD LENGTH
specified under VIEW PREFERENCES (see 1.2.8 on page 54). If you want to display conversational data
in Custom format, you should always use Maximum scope, since this is the only option to provide
more than a single utterance as context. You can change the display format and scope for the
current query at any time from the Query Options dialogue box (see 5.2.3 on page 103).
32 Click on the CONCORDANCE button on the toolbar to change from Line to Page display mode. You will now
see the current solution displayed as in a playscript, with the code for the speaker given at the beginning of
each utterance. This makes it much easier to follow the talk.
33 Using the arrow buttons on the toolbar (or the PGDN key), page through the solutions to see what happens
before and after 'anyway'. You will find that in the vast majority of cases, 'anyway' marks a topic shift. You
will also see from the carets in the display, representing <PTR> elements, that it often seems to overlap
with the end of the preceding utterance.

9.2.3
Sorting and Saving Solutions in Custom Format
We can highlight recurrent patterns in what precedes and follows 'anyway' in these examples by sorting them. In Custom format solutions are sorted as
they

page_168

Page 169

appear on the screen: that is, if you introduce new characters (for example to indicate utterance boundaries) these will be taken into account as well the
actual words of the solutions when sorting.
Click on the CONCORDANCE button to return to Line display mode.
34
35 Select SORT from the QUERY menu, and set the Primary sort key to LEFT with a span of 3 words. Then
click on OK to sort the solutions. If you scroll through the solutions you should see groups where 'anyway'
is preceded by an utterance boundary and 'Yeah' or 'Right', that is, following an expression of agreement
by another speaker which concludes a previous topic. You may also notice that 'anyway' is often preceded
by a laugh by another speaker.

Now let us look at the solutions for sentence-initial 'anyhow'. Since editing the current query would replace the current display of solutions, we shall
design a second query in a new window so as to be able to compare the two sets of solutions.
Click on the QUERY BUILDER button on the toolbar to display the Query Builder dialogue box.
36
37 Follow the same procedure as in 9.2.1 on page 163 to insert <STEXT> in the scope node, and <S> in
 the content node. Then create a second content node below the first, this time specifying the word
'anyhow', and join the nodes with a NEXT link.
38 Check that the 'Query is OK' message is displayed, and click on OK to send the query to the server.
39 In the Too many solutions dialogue box, click on the DOWNLOAD ALL radio button, then on OK. The
solutions will be displayed in the revised Custom format.
40 Click on the CONCORDANCE button to display the solutions in Page mode, and page through them using the
arrow buttons on the toolbar (or the PGDN key). You will see that 'anyhow' also generally appears to mark
a change of topic.
41 Switch back to Line display mode and sort the solutions by the left and then by the right, to see what
precedes and follows 'anyhow'.
42 Use the Window menu to switch between this and the 'anyway' query. Do you notice any differences
between the contexts of the two words? We also find agreement markers before 'anyhow', so in this
respect there appears to be no particular difference in the environments of the two words. You may find it
easier to compare these solutions if you print them. Since the PRINT option only allows you to print
one line for each solution, and you may need to see a relatively large

page_169

Page 170

context to identify topic change, you should first use the LISTING option to save the solutions to a file, which you can then print by switching to
a word processor (see 8.2.5 on page 154).
In sentence-initial position, 'anyhow' and 'anyway' both generally seem to indicate a shift of topic in speech.
9.2.4
Searching at Utterance Boundaries: Laughs and Laughing Speech
One particular use of 'anyway' and 'anyhow' appears to be to shift the topic back to more serious issues following laughter. A question which arises is
whose job this is: is it more usually the producer of the laughter who follows this with 'anyway/anyhow', or is it more often another participant? We can
answer this question by seeing what proportion of the occurrences of sentence-initial 'anyway/anyhow' subsequent to laughter follow a change of
speaker, occurring at the beginning of an utterance. If 'anyway/anyhow' is produced indiscriminately, both by producers and by recipients of laughter,
we would expect it to be utterance-initial in approximately half the cases.
Laughter and Topic Change
Let us first design a query to find all the cases where sentence-initial 'anyway' or 'anyhow' follows laughter. We shall work backwards from the final
component of the query.
Click on the QUERY BUILDER button on the toolbar to display the Query Builder.
43
44 Click in the content node and select EDIT then WORD to display the Word Query dialogue box.
45 Check the PATTERN box, paste in the patternany(way|how), and click on LOOKUP.
46 Select both 'anyhow' and 'anyway' by dragging the mouse or holding down CTRL when clicking on these
items in the matching words list.
47 Click on OK to insert this component in the Query Builder node.

Now add the requirement that 'anyway/anyhow' must be directly preceded by a sentence boundary.
Click on the upwards branch of the upper content node to add a further node to the query.
48
49 Change the link type to NEXT.
50 Click in the new content node and select EDIT then SGML to display the SGML dialogue box.
51 Select <S> from the element list, then click on OK to insert this component in the Query Builder node.

page_170

Page 171

Now add the requirement that the new sentence beginning with 'anyway/anyhow' should be directly preceded by laughter.
In the BNC, laughter is represented in two manners, which we shall consider separately. In the first, most frequent case, it is treated as a vocal event
occurring between two segments of speech, and is represented using the <VOCAL> element. The DESC attribute for this element is assigned the value
laugh. The full representation is thus <vocal desc = "laugh">. Other values of the DESC attribute on the <VOCAL> element include
cough, sigh, clears throat, etc.

Click on the upwards branch of the content node to add a further node to the query.
52
53 Change the link type to NEXT.
54 Click in the new content node and select EDIT then SGML to display the SGML dialogue box.
55 Select <VOCAL> from the list of elements. The list of attributes for the <VOCAL> element will be
displayed.
56 Select DESC from the list of attributes, then click on A DD to display the Attribute dialogue box. As the
possible values for the DESC attribute are not a predetermined set, SARA does not provide a list; you must
type in the value you require.
57 Type in the value laugh and click on OK to return to the SGML dialogue box.
58 Click on OK to insert this component in the query.
59 Check that the scope node contains the element <BNCDOC>. As <VOCAL> elements only occur in
spoken texts in the BNC, it is not necessary to restrict the query explicitly by specifying <STEXT>as
scope.
60 Check that the 'Query is OK' message is displayed and click on OK to send the query to the server. There
are 42 solutions.
61 Click on the DOWNLOAD ALL radio button and download the solutions. You will see that the query focus
begins with [laugh] (indicating non-verbal laughter), generally followed by a speaker code and
'anyway'. Where the bottom node of a query is joined to a previous node or nodes with NEXT links,
the query focus corresponds to this entire sequence of nodes.

Thinning the Solutions

Even though we joined the various nodes of the query with N EXT links, you may notice that in some solutions, there are features occurring between a
laugh and 'anyway/anyhow'. A NEXT link simply means that no words may occur between the content of the two nodes in question (see 4.2.2 on page
92). Other elements, however, may intervene. For instance, there might be:

page_171

Page 172

<UNCLEAR> or <GAP> elements indicating intervening speech which could not be transcribed;
<PAUSE>, <EVENT> or <VOCAL> elements of significant length (these elements are assigned a DUR (duration) attribute value where they exceed 5
seconds);
<PTR> elements, indicating that parts of two utterances were spoken simultaneously rather than consecutively.
Most significantly for our purposes, the current query also allows an utterance boundary to occur between a laugh and the sentence beginning with
'anyway/anyhow'. You should therefore inspect the solutions to make sure none of them are spurious in these respects.
First, let us sort the solutions by the query focus. This will help group cases where particular features occur between laughs and 'anyway/anyhow'.
Select SORT from the QUERY menu, and set the Primary sort key to CENTRE with a span of 4 words. Then
62 click on OK to sort the solutions.
63 Scroll through the solutions looking for any feature displayed between a laugh and 'anyway/anyhow' which
is not an indication of a new speaker, and double-click on those cases where this implies that
'anyway/anyhow' does not immediately follow the laughter.

There are a couple of cases where there is intervening unclear speech, clearly implying a space between laughter and 'anyway', and which should
therefore be marked. On the other hand, there are cases where there is an intervening laugh by another participant, or where 'anyway' is preceded by a
shift to a laughing voice quality (see 9.2.4 on the next page), which should be maintained. Finally, towards the top of the display you will find cases with
intervening <PTR> elements (represented by carets), which indicate overlaps. To assess the implications of these for the sequencing of the talk, they will
need to be examined more closely.
Overlaps in Spoken Dialogue
In spoken dialogue texts in the BNC, <PTR> elements are placed at the beginning and end of those portions of utterances which overlap with others, for
example where one speaker begins to speak before another has finished. For each of the overlapping segments, the <PTR> elements marking the
beginning of overlap have the same value on their T attribute. Similarly, the elements marking the end of overlap also have the same value on this
attribute. By comparing the attribute values of the various <PTR> elements, you can thus identify what overlaps with what. As our Custom format display

page_172

Page 173

does not show these attribute values, we shall examine the solutions in SGML format.
Delete the solutions you have already marked as spurious using THIN and REVERSE SELECTION under the
64 QUERY menu.
65 Again under the QUERY menu, select OPTIONS then SGML.
66 Examine all the solutions where <PTR> elements occur between the <VOCAL> element and
'anyway/anyhow'. These should be grouped at the top of the display.
67 Double-click on any cases where the overlapping implies that there is a gap between the laughter and
'anyway/anyhow'. In most cases, you will find that the <PTR> elements indicate either that the laughter
overlaps with the previous utterance, or that it overlaps with 'anyway' or with further laughter produced
prior to it. None of these implies a gap between laughter and 'anyway/anyhow'.
68 Delete any solutions you have marked using THIN and REVERSE SELECTION from the QUERY menu.

All the remaining solutions to the query should now be cases where 'anyway/anyhow' directly follows laughter, with or without a change of speaker.
Scrolling through them, you will see that in the vast majority there is a change of speaker, and that there are no occurrences of 'anyhow'.
Laughing Speech and Topic Change
Not all laughter in the BNC is encoded using the <VOCAL> element. Speakers may also utter words while laughing, or speak in a laughing tone. In
these cases, laughter is considered to be a feature of the voice quality with which a particular piece of speech is produced. Such changes in voice
quality are represented by <SHIFT> element, whose NEW attribute describes the quality. The end of the piece of speech with this voice quality is
indicated by a second <SHIFT> element, this time without attributes. For example, if the word 'Yeah' is spoken with a laughing voice, the full
representation is:
<shift new="laughing">Yeah<shift>

To find instances of laughing voice quality immediately before 'anyway' or 'anyhow', we must identify cases where the latter is preceded by a <SHIFT>
element indicating the end of laughing voice quality. However, the particular type of voice quality is only indicated at the point where it begins (by the
value of the NEW attribute), not where it ends. Consequently, you cannot formulate a query to find the end of a specific kind of changed voice quality, but
only one which will find a return to unmarked voice quality. The

page_173

Page 174

solutions to such a query will thus include cases where other voice qualities are involved, and these spurious solutions will have to be identified by
inspection and removed by thinning. Other values of the NEW attribute on the <SHIFT> element include reading, shouting,
singing, whispering, imitating Italian accent, mimicking baby voice etc. You can generate a more
complete list of such values by designing an SGML Q UERY to find occurrences of the <SHIFT> element, and sorting the solutions in SGML
format.
First, let us save the solutions to the last query, involving non-verbal laughs.
Select SAVE from the FILE menu, and save the query aslaugh. The window title will change to
69 LAUGH.SQY.
70 Select EDIT from the QUERY menu to display the Query Builder.
71 Click in the <vocal desc=laugh> node and select EDIT to display the SGML Query dialogue box.
72 Select <SHIFT> from the list of elements, and click on OK to insert it into the Query Builder node.
73 Click on OK to send the query to the server. There are 7 solutions.

We now need to remove any solutions where the shift in voice quality preceding 'anyway/anyhow' does not involve laughing. These will be easier to
identify if we view the solutions in Custom format.
Select OPTIONS from the QUERY menu and display the solutions in CUSTOM format. Assuming you have
74
74 adopted the format specified in 9.2.2 on page 165, the <SHIFT> element indicating the end of marked
voice quality Will appear as a box-shape actually, an empty pair of square brackets.
75 Scroll through the solutions, using the buttons on the horizontal scroll bar to examine the text to the left of
the query focus. Note the NEW attribute value on the previous <SHIFT> element indicating the beginning of
marked voice quality, which should be displayed between square brackets. You will find cases of
[singing], [screaming] and [whispering] as well as [laughing].
76 Mark all the solutions where the marked voice quality prior to the query focus does not involve laughing,
then use REVERSE SELECTION from the THIN menu to remove them from the display.

As in the previous query, we now need to examine the remaining solutions and remove any where other SGML elements between laughing speech and
'anyway/anyhow' indicate that these are not consecutive.
Select OPTIONS then SGML from the Q UERY menu to re-display the solutions in SGML format. You will see
77 that there is one solution

page_174

Page 175

where laughing speech is followed by a laugh from the same speaker prior to an utterance boundary and
'anyway'. As this instance should already have been included in the previous query as a case of laughter
preceding 'anyway/anyhow', we can delete it here.
78 Mark this solution and remove it from the display using the THIN REVERSE SELECTION option. You will be left
with 3 solutions. In all of these, the next sentence following laughing speech begins with 'anyway'. There
are no occurrences with 'anyhow'.
79 Select SAVE AS from the FILE menu and save the query aslaughing. You must use SAVE AS rather than
SAVE here, in order not to overwrite the previous query (laugh.sqy).

Results
Having removed all the spurious solutions, we can now examine both queries to see whether, following laughter or laughing speech, 'anyway/anyhow' is
typically produced by the same or by a different speaker. This involves establishing the proportion of solutions in which there is an utterance boundary
between laughter or laughing speech and 'anyway/anyhow'. We can do this by sorting them in Custom format. As new utterances are indicated by the
speaker code between curly brackets, sorting by the centre (the query focus) will group those cases where shifts in voice quality or laughs are followed
by an utterance boundary.
Select OPEN from the FILE menu and re-open the querylaugh.sqy in a new window.
80
81 Download all the solutions, and display them in Custom format.
82 Sort the solutions by the Centre, with a span of 4.
83 Scroll through the solutions to see if there are any cases where there is no utterance boundary between
laughter and 'anyway/anyhow'. While there are some cases where more than one speaker laughs prior to
'anyway', there are none where laughter is only produced by the same speaker who says 'anyway'.
84 Use the WINDOW menu to switch to your second query (LAUGHING.SQY).
85 Display the solutions in CUSTOM format, and see if there are any of these where 'anyway' is produced by
the same speaker as the preceding laughing speech. In contrast with the previous query, in two of the three
occurrences there is no intervening utterance boundary, the speaker who laughs going on to produce
'anyway'.

page_175

Page 176

We are now in a position to provide answers to some of the questions posed at the beginning of this section concerning the use of 'anyway' and
'anyhow' following laughter.
As far as their relative frequencies are concerned, if we take the two queries together, there are 43 occurrences of 'anyway' following a laugh or
laughing speech, and none of 'anyhow'. However, given the overall proportions of these two forms in speech in the BNC (30:1), we would not have
expected to find more than a few occurrences of 'anyhow', and its absence here can hardly be judged significant.
'Anyway' tends to follow laughter by another rather than by the same speaker. There are only two cases where laughter is followed by 'anyway' with no
intervening utterance boundary, both involving laughing speech. Since they appear to be exceptions, it may be of interest to look at these two cases in
greater detail.
Click in the LAUGHING.SQY window, then click on the CONCORDANCE button on the toolbar to change to
86 Page display mode, and examine the two solutions where laughing speech is followed by an 'anyway' from
the same speaker.

You will see that in both these solutions 'anyway' marks a switch from dialogue to monologue. If you look up their sources using the S OURCE option, you
will find that both in fact come from television programmes. This provides further support for the hypothesis that after laughter, the job of changing topic
with 'anyway' is preferably performed by a different participant from the one who laughed. As often happens in corpus-based work, it is this exception
which seems to confirm the rule.
9.3
Discussion and Suggestions for Further Work
9.3.1
Searching Near Particular Positions
We have now confirmed our hypothesis that most sentence-initial occurrences of 'anyway' and 'anyhow' in speech appear to indicate a topic change. We
have not however attempted to see if occurrences which are not in sentence-initial position can have this function. We can investigate this issue in a
number of ways, most simply by formulating a query to search for occurrences of 'anyway/anyhow' as the second word of the sentence in speech.
Using the Query Builder, we can use the ANY option under EDIT to insert a one-word slot between an <S> node and an 'anyway/anyhow' node, within
the scope of an <STEXT> element. This shows that the most common words to preface 'anyway/anyhow' are 'and', 'but' and 'so'. In most of these
solutions, 'anyway/anyhow' again seems to indicate a topic change. A wider range of positions

page_176

Page 177

can be investigated by using the SPAN scope option (see 8.2.3 on page 150). For instance, to find cases where 'anyway' or 'anyhow' occurs
within the first four words of a sentence, we can join an SGML Query for <S> to a Pattern Query for {any(way|how)} with a One-way
link, specifying the scope of the query as a span of 4 words. Such a query will not, however, be restricted to spoken texts: Query Builder will
only allow us to define a single scope node, which we have used to specify the span of 4 words. To supply the additional constraint that this
must be done within a spoken text, it is necessary to use a CQL Q UERY (see 7.2.3 on page 135), with a formulation such as the following:
<stext>* ((<s>* ("anyway"|"anyhow"))/5)

This specifies that the scoped part of the query is to be joined to an SGML Query for an < STEXT > element, where the latter (whose start-tag
occurs at the beginning of each spoken text in the corpus) may occur outside the span of 5 words in which the rest of the query is to be
solved.
9.3.2
Some Similar Problems
And so to Bed
Prescriptive grammars traditionally deplore the use of sentence-initial 'And' (such as Pepys' ''And so to bed''). Design a query to find sentence-initial
occurrences of 'And so to' in written texts in the BNC. Use the <TEXT> element to restrict the scope of the query to written texts.
Bed may be a good place to go at this point, as you will have a long wait for solutions, there being an enormous number of occurrences of
'and', of 'so', and of 'to' in the BNC for SARA to examine, even if there are in fact only 47 solutions where they occur together as a phrase, 4
of them preceding 'bed'.
How Are You?
Faced with the everyday question 'How are you?', it seems that everybody has to lie for much of the time if they are to get on to other topics of
conversation (Sacks 1975, 1992). How do people actually answer this question in the corpus? Design a query to find occurrences of 'how are you?' at
the end of a spoken utterance, then select Plain display format and sort the solutions by the right to identify the most frequent answers. Use PHRASE
QUERY to specify 'how are you?' (including the question mark) as the content of a first node in the Query Builder. Then attach it with a NEXT
link to a second node containing a </U> end-tag. You can select the end-tag by clicking on the END radio button in the SGML dialogue box.
There are 128 occurrences of utterance-final 'how are you?' in the BNC. Sorting these by the right, you will find that the most frequent
responses are '(I'm) fine' and 'All right'/'Alright'. 'Very well' and 'Not too bad' are less common. Negative responses are very rare indeed.

page_177
 Page 178

Goodbye
When we say 'goodbye', do we mean it? Is that really the end of the conversation? In the BNC, each recorded conversation is treated as a distinct
<DIV> element (only used in spoken texts: see 6.1.2 on page 112). Find out how often 'Bye', 'Good-bye' and 'Goodbye' occur within the scope of a
<DIV> element. Then, design a query to find out how often they occur near the close of that element, i.e. within (say) the 10 words before a </ DIV>
end-tag. While it is easy to find the number of occurrences of these forms within <DIV> elements (1260), it is less straightforward to find how
often they occur near the end of such elements. In Query Builder, when two or more content nodes are joined with One- or Two-way links, the
query focus corresponds to the bottom node, and the number of solutions is the number of times that this bottom node is found. Thus, if you
use a one-way link to join an upper node containing forms of 'goodbye' to a lower node containing a </DIV> end-tag, you will find the
number of times a </DIV> tag is preceded by 'goodbye', rather than the number of times that a form of 'goodbye' occurs before a </DIV> tag.
So, for example, if sottieone says 'Goodbye, goodbye' at the end of a conversation, this will only count once.
There are two ways of getting round this problem.
You can place the </DIV> node before the 'Goodbye' node, joining them with a Two-way link. This will count all occurrences of 'Goodbye'
whether they immediately precede or follow the end of a conversation; you will then have to run a separate query to find the occurrences
immediately following the end of a conversation and subtract them from the total.
You can simply do an SGML query for the </DIV> tag and then use the COLLOCATION option to count the number of times a form of 'Goodbye'
is found within the required span. Again, however, this will also include occurrences which follow the end of the conversation, which will have
to be separately counted and subtracted from the total.
In either case, counting the number of 'false positives' is easily done: use Query Builder to search for a </DIV> tag followed by a form of
'Goodbye' within the required span using a One-way link.
Only 17% (211/1260) of occurrences of forms of 'goodbye' occur within the last ten words of a <DIV> element in speech. Overall, it appears
that saying "Goodbye" in conversation may be anything but final. However, we need to bear in mind that strictly speaking, a </DIV> tag
only marks the end of a recording, so in many cases we cannot be sure that this was in fact the end of the conversation.

page_178

Page 179

10
What Does 'SARA' Mean?
10.1
The Problem: Studying Pragmatic Features
10.1.1
How Are Terms Defined?
Rather than a particular language form, this final task investigates a particular pragmatic function of language: the definition of terms. Obviously, there is
no way we can design a query or queries to locate all of the ways definitions are provided in the BNC, so we must follow an ad hoc procedure. We will
start by examining some contexts where definitions seem likely to be found, in the hope that by progressively accumulating examples, we may find
recurrent features which can then be investigated more systematically. The procedure will illustrate how different options and query-types may be used to
complement each other in a wider research perspective.
10.1.2
Highlighted Features
This task shows you:
how to classify and retrieve particular solutions from different queries using the BOOKMARK command;
how to search for solutions in specific texts in the corpus by using the ID attribute on the <BNCDOC> element;
how to design queries which are case-sensitive, or which include punctuation, using the PHRASE QUERY option;
how to group solutions according to case using the ASCII collating option;
how to combine a WORD QUERY and a PHRASE QUERY to find occurrences of compounds which may be written as one word, two words, or with a
hyphen;
how to use the COLLOCATION option as a heuristic to evaluate potential queries.
It assumes you already know how to:
design Word, Phrase, POS, Pattern, and SGML queries, and combine these using the Query Builder (see 5.2.4 on page 105, 6.2.3 on page 117, 7.2.2
on page 133);
design CQL queries (see 7.2.3 on page 137);
copy a selected solution to the Windows clipboard using the Copy option (see 1.2.7 on page 54);
sort and thin solutions (see 3.2.2 on page 78, 2.2.4 on page 68);
save and re-open queries (see 2.2.5 on page 70);

page_179

Page 180

look up collocation frequencies using the Collocation option (see 3.2.1 on page 76);
browse through individual texts (see 1.2.7 on page 52).
10.1.3
Before You Start
Set the following defaults using the VIEW PREFERENCES option:

2000 characters
MAX DOWNLOAD LENGTH
MAX DOWNLOADS 15
FORMAT Plain
SCOPE Maximum
VIEW: QUERY and ANNOTATION unchecked
CONCORDANCE checked
BROWSER: SHOW TAGS unchecked

10.2
Procedure
10.2.1
Looking for Acronyms with POS Query
Let us start by choosing an easily-recognisable class of expressions which are likely to be defined in texts: acronyms. We shall attempt to make a
collection of acronym definitions, arbitrarily beginning with definitions of 'SARA'. Since the BNC was collected some time before the baptism of the
SARA described in this handbook, we are hardly likely to find references to an 'SGML Aware Retrieval Application'. However, it is quite probable that
the same series of letters will be used as an acronym with other senses in the corpus.
As acronyms are generally proper nouns, we shall use the POS QUERY option (see 7.2.1 on page 131).
Click on the POS QUERY button on the toolbar. The POS Query dialogue box will be displayed.
1
2 Type in the string sara, and click in the Part-of-speech display window (or press TAB) to display the POS
codes associated with 'sara' in the corpus. SelectNP0 (proper noun) and click on OK to send the query to
the server. The Too many solutions dialogue box will be displayed, stating that there are 1266 solutions in
149 texts.
3 Check the ONE PER TEXT box and download all the solutions. It seems reasonable to assume that repeated
references to 'sara' in any one text will probably have the same meaning, and that the first one in each text is
as likely as any to be flanked by a definition or explanation.

page_180

Page 181

Scroll through the solutions to see if you can find any acronyms. While most of the solutions involve
4 reference to women called Sara, you may notice some cases where 'SARA' is in upper case throughout,
suggesting that it may be an acronym.
Let us start by looking at these cases.
Sort the solutions with Centre and span 1 as Primary key, and Right and span 2 as Secondary key, using
5 ASCII collating. ASCII collating distinguishes upper and lower case, and will group solutions where 'SARA'
is in upper case throughout at the top of the display. Several of these appear to be acronyms 'SARA title',
'SARA rules', 'SARA the Severn Auxiliary Rescue Association', and 'super SARA', for example.
6 Mark these solutions and remove the rest by thinning; then click on the CONCORDANCE button to switch to
Page display mode.
7 Page through the remaining solutions to see which, if any, provide definitions of 'SARA'. Of the cases listed
above, only 'SARA, the Severn Auxiliary Rescue Association' is defined in these solutions.

10.2.2
Classifying Solutions with Bookmarks
The one definition of 'SARA' we have found takes the form of a parenthetical explanation immediately following the acronym, placed between commas.
Let us mark and classify it so that we can find it again without difficulty. We can do this using the BOOKMARK option.
Click on the CONCORDANCE button to return to Line display mode.
8
9 Select the solution referring to the 'Severn Auxiliary Rescue Association' by clicking on it.
10 Select BOOKMARK from the EDIT menu. The BOOKMARK NAME dialogue box will be displayed. The name of
the current query is shown in the top left of the box: bookmark names are stored together with the name
of the query.
11 Click in the input window at the top of the box and type in a suitable name for this acronym definition,
such as ACR, DEF. The name is a mnemonic indicating that the acronym is followed by a comma and
the definition. Unlike Query names, Bookmark names can contain spaces and punctuation, with a
total of up to 12 characters. You can if you want give more than one name to a solution, but each
name can only refer to one solution in a given query.
12 Click on ADD to register the bookmark. The Bookmark name dialogue box disappears.
13 To check that your bookmark has been registered, select BOOKMARK from the EDIT menu again. You
will see that the bookmark name now

page_181

Page 182

appears in the lower window of the dialogue box. The DELETE option in the dialogue box allows you to
remove a selected bookmark from the list in the lower window.
14 Click on CANCEL to close the dialogue box.
15 Save the query as sara.sqy. The bookmark will be saved along with the query.

10.2.3
Searching in Specified Texts
The solutions in the window SARA.SQY all come from different texts, and all appear to involve different meanings of 'SARA'. Only some of them
include definitions, however. It seems worthwhile to look in each text to see whether the acronym is defined at some other point. To do this, we need to
restrict our search to the specific texts identified during the last query.
Scroll through the solutions in the SARA.SQY window using the arrow buttons or the cursor keys, and
16 make a note of the text identifier code (displayed on the Status bar) for each of the solutions in which no
definition is given.
17 Invoke the QUERY BUILDER and insert the POS QUERY"sara"=NP0 in the content node.
18 Click on the upward branch of this node to create a new content node.
19 Click in the new node and select EDIT then SGML. The SGML dialogue box will be displayed.
20 Select <BNCDOC> from the list of elements. The list of attributes for this element will be displayed.
21 Select ID from the list of attributes, and click on A DD to display the Attribute dialogue box. The text
identifier of each document is indicated as a value of the ID attribute on the <BNCDOC> element. It takes
the form <bncDoc id=BDXXX>, where 'XXX' corresponds to the text identifier code displayed on
the status bar.
22 Type in as the value the letters BD followed by the first text identifier code (all in upper case). The first text
identifier code should be B7L.
23 Click on OK to insert this attribute value pair in the right hand window of the SGML dialogue box. You
can only type in a single explicit value as the target for an attribute search like this one: patterns or
wildcards are not allowed.
24 Click on OK to insert this component in the Query Builder node. It will be displayed as <bncDoc
id="BDB7L">.
25 Click on the right-hand branch of this content node to create an alternative content node, then click in the
new node and select EDIT.

page_182

Page 183

Select SGML and repeat the previous procedure, specifying the next text identifier code in your list
26 (K5A).
27 Create a third alternative content node corresponding to the next text in your list (K9E).
28 Create a fourth alternative content node corresponding to text B75, the final text in your list. If you were
looking for all the solutions in only one text, you could specify this text as the scope of the query
rather than as a content node.
29 Check that the 'Query is OK' message is displayed, and click on OK to send the query to the server.
There are 6 solutions.
30 Save the query using a suitable mnemonic, such as sara2.sqy.
31 Use the SOURCE option to browse through the text preceding each hit. You will find that one of the things
SARA stands for is the 'Scottish Amateur Rowing Association'.

In this example, the acronym 'SARA' is used some way after the full name, and it is assumed that the reader will associate the acronym to the referent
without this link being specifically marked. We will treat this as a new category, of implicit definition by anaphora.
Return to the solutions display and bookmark this solution with the name ACR ANA Remember to save
32 the query again in order to keep this new bookmark. In this text (a collection of newspaper articles
from The Scotsman), you may have noticed that the full name precedes what is in fact the second
use of 'SARA'. Here we would hardly want to say that the third use of the acronym is explained
subsequently, since the explanation appears in a different article. BNC texts from periodicals often
consist of more than one text in the usual sense of the word, just as those from the spoken
demographic component consist of more than one conversation (see 6.1.2 on page 112). Where
such texts are involved, you should carefully consider the implications of selecting One-per-text
thinning, or of using One-way or Two-way links to join content nodes in the scope of a single
<BNCDOC> element.

There remain three texts in which 'SARA' appears not to be defined. One is a reference to 'SARA title 111 in text K9E a specialised report, where it
appears to be assumed that readers are already familiar with the acronym. The others are references to 'Super SARA' and 'Super- SARA' in texts B7L
and B75. In both these texts, however, you may have noticed that there is also a hyphenated one-word version of the acronym ('Super-SARA'), which
is defined as a nuclear reactor safety project. We will return to the problem of dealing with such variants in 10.2.5 on page 186.

page_183

Page 184

10.2.4
Serendipitous Searching: Varying the Query Type
Solutions to queries often contain unexpected or curious features, which may or may not be related to the original purpose of the query. It is tempting
and often rewarding to go off at a tangent to investigate casually-encountered features, following a 'serendipity principle'. For instance, a glance through
the solutions in the SARA2.SQY window shows that acronyms may flock together. The first solution, which refers to 'Super SARA', also contains
references to 'LOFT' and to 'Super-SARA' a hyphenated version of the acronym found by our original query. We can design queries using these other
acronyms to look for further definitions, and then look through the solutions to these queries for further acronyms to use in further queries, in an ever-
widening cycle. In each case, however, it is important to select the type of query which will best maximise recall and precision.
 Design a POS QUERY for 'loft' as a proper noun (NN1-NP0 or NP0). There are 15 solutions, three of
33 them involving acronyms rather than attics. One of these follows the acronym with an explanation in
parentheses ('Line Oriented Flight Training'). It also contains a reference to something called 'OATS'.
34 Thin the solutions to remove the attics, and bookmark the solution containing an explanation as ACR
(DEF).
35 Save the query with a suitable mnemonic, then design a further query to find occurrences of 'OATS'. If
you use POS Query, you will find that oats' is not tagged as a proper noun in the corpus. To avoid
references to porridge or horse-breeding, you might either use the Query Builder to restrict a Word Query
to the text which contained the acronym, as in 10.2.3 on page 182, or else design a case-sensitive Phrase
Query. The fact that 'Oats' is not tagged as a proper noun here is a reminder that the automatic
POS annotation applied to the BNC is not one hundred percent reliable (see 7.3.1 on page 140).
36 Click on the PHRASE QUERY button on the toolbar. The Phrase Query dialogue box will be displayed.
37 Uncheck the IGNORE CASE box and type in the stringOATS (in upper case). Then click on OK to send the
query to the server.
38 Page through the 10 solutions to see if any provides a definition. You will discover that 'OATS' stands for
'Oxford Air Training School' and for 'Office Automation Technology and Services'.
39 Bookmark these definitions and save the query with suitable mnemonics.
40 Page through the solutions to look for any further acronyms. You will find, inter alia, a reference to
'CPL/IR'.

page_184

Page 185

Now design a query to find occurrences of 'CPL' and its compounds, such as 'CPL/IR'.
Invoke the QUERY BUILDER, click in the content node, and select EDIT then WORD. The Word Query
41 dialogue box will be displayed.
42 Type in the string cpl and click on LOOKUP to display the forms in the index beginning with this string. You
will see that as well as 'cpl', the list contains a number of compound forms.
43 Select all the forms in the list to see their total frequency in the corpus. There are 63 occurrences of forms
beginning 'cpl'.

'Cpl' is of course not merely an acronym in English, but also an abbreviation for 'Corporal'. In this latter sense, it is likely to be written with its first letter
in upper case. Acronyms, on the other hand, are likely to be completely in upper case. One way to exclude occurrences of the military sense is therefore
to make the query case-sensitive, specifying that the string 'CPL' must be in upper case throughout. To make a query case-sensitive, however, you must
use Phrase Query, which, unlike Word Query, does not allow you to specify a list of alternative forms. Here we shall use the Query Builder in order to
combine both query types to maximise the advantages of each.
In the Word Query dialogue box, select all the forms in the list other than 'cpl', then click on OK to insert
44 this query into the Query Builder. This part of the query does not need to be case-sensitive, since there is
little risk of confusion with abbreviations except for 'Cpl', which we have excluded.
45 Click on the right-hand branch of the content node to create an alternative content node.
46 Click in the new node and select EDIT then PHRASE. The Phrase Query dialogue box will be displayed.
47 Uncheck the IGNORE CASE box and type in the stringCPL (in upper case). Then click on OK to insert this
component in the Query Builder.
48 Check that the 'Query is OK' message is displayed, and click on OK to send the query to the server. The
Too many solutions dialogue box will be displayed, stating that there are 35 solutions in 12 texts.
49 Download all the solutions, then select QUERY TEXT under QUERY OPTIONS to display the query text. You will
see that the text of the query is: ("cpl/ir"|"cpl/irs"|"cpls"|$('CPL')) The dollar
sign indicates that the final component of the query is case-sensitive. You can also use the $ symbol to
make all or part of a CQL Q UERY case-sensitive (see 7.2.3 on page 137).

page_185

Page 186
 Turn off the Query Text display, then page through the solutions looking for acronym definitions. You will
50 discover that 'CPL' has a variety of meanings, including 'Command Processing Language', 'Communist
Party of Lithuania', and 'Cats' Protection League'.
51 Create bookmarks for those solutions which provide definitions, using appropriate mnemonics. If the
same pattern occurs more than once, add a number to the bookmark name. The bookmarks you create
should include a number of versions of ACR (DEF) and of DEF (ACR). Within any single query,
each bookmark must have a different name, providing an unambiguous indication of the solution
it refers to. If, as in this task, you are using bookmarks to categorize solutions, this means that
you must give solutions in the same category slightly different names, for instance by adding a
different number in each case.
52 Save the query as cpl.sqy.

We could go on with the process of browsing through solutions for acronyms, and then designing appropriate queries to find their definitions, more or
less ad infinitum. Such lateral movement through the corpus can be extremely productive: here we have in a matter of minutes come up with a fair
number of definitions. As we have categorized them using bookmark names, we can now retrieve and compare the solutions in particular categories.
Before doing so, however, let us return briefly to one of the acronyms in the original query, 'Super SARA'.
10.2.5
Finding Compound Forms: Combining Phrase and Word Queries
The spelling of compound forms in English is notoriously varied. We have found, for instance, both 'Super SARA' and 'Super-SARA' in one text, as well
as 'Super- SARA' in another, and it would not be surprising to also find 'Supersara' as a single unhyphenated form, nor other combinations of upper and
lower case. One-word forms, whether unhyphenated or hyphenated, are listed in the SARA word index, and can therefore both be included in a Word
(or Pattern) Query by specifying a pattern such as super-? sara: the pattern will also match any case variants (see 8.3.2 on page 159). Two-
word realisations, on the other hand, can only be found by a Phrase Query (the Ignore case box must be checked to match case variants). To include all
these possibilities in a single query, we must therefore use the Query Builder, specifying the Word or Pattern Query super-?sara and the Phrase
Queries super sara and super- sara in alternative nodes.

Invoke the QUERY BUILDER and insert a Pattern Query to find occurrences of the patternsuper-?
53 sara in the content node.

page_186

Page 187

Add an alternative content node containing a Phrase Query to find occurrences of the two-word form
54 super sara. IGNORE CASE should be checked.

55 Add a further alternative node containing a Phrase Query for the two-word form super- sara.
56 Check that the 'Query is OK' message is displayed, and click on OK to send the query to the server.
There are 48 solutions in 9 texts.
57 Download all the solutions, then page through them to look for any definitions.

These solutions illustrate further ways in which acronyms may be defined. For instance, the acronym may pre-modify a general noun such as 'project'
or 'experiment' ('Europe's Super-SARA experiment into nuclear safety', 'the Super-sara nuclear safety experiment', 'Europe's super-SARA project'). Or
it may post-modify a general noun followed by 'known as' (''the project, known as Super-SARA [...]'').

Bookmark these solutions with suitable names, such as ACR GN and GN, KA ACR, adding numbers
58 if the same pattern occurs more than once. Then save the query with a suitable mnemonic.

10.2.6
Viewing Bookmarked Solutions
We have now identified several categories of acronym definitions. Let us try and summarise the results of the investigation so far by examining the
bookmarked solutions in the various queries.

Check that all the queries you have designed in this task are still open; re-open any that have been
59 closed. You can only work with bookmarks in queries that are currently open.

60 Under the EDIT menu, select GOTO. The GOTO dialogue box will be displayed. The list should include all
 the bookmarks you have assigned in alphabetical order: each bookmark name is followed by the name
of the query in which it is found.
61 Scroll through the bookmarks and make a list of the various categories to which you have given different
names. These categories should include ones where:

the acronym is followed by a definition between commas or parentheses: ACR, DEF and ACR (DEF);
a definition is followed by the acronym between commas or parentheses: DEF (ACR) and DEF, ACR;

page_187

Page 188

the acronym pre-modifies a general noun which partly explains it, such as 'project' or 'experiment': ACR GN;
a definition using a general noun is followed by a comma and 'known as', plus the acronym: DEF, KA ACR;
the acronym refers anaphorically to a definition earlier in the text: ACR ANA.
Commas vs Parentheses in Definitions
Using the Windows TILE or CASCADE commands in order to show multiple windows, we can now select and simultaneously display all the solutions from
a particular category or categories. Let us start by examining all the examples of the following:
ACR, DEF
ACR (DEF)
This will enable us to see whether there are differences between the use of commas and parentheses to introduce definitions of acronyms.

Scroll through the bookmarks and make a list of the names of all the queries which contain solutions in
62 these categories.

63 Click on CANCEL to close the Goto dialogue box, then close any queries which are not included in the list.
64 Select GOTO from the EDIT menu once more, and scroll through the bookmarks to find the first from the
categoryACR, DEF.
65 Click on OK to display the solution corresponding to the bookmark. The window containing this
solution will become the active window.
66 Click on the CONCORDANCE button to display the solution in Page mode.
67 Select GOTO from the EDIT menu once more, and find the next bookmark in this category. If this
bookmark indicates a solution in a different query to the previous one, select and click on OK to display
it.
68 If this bookmark indicates a solution in the same query as the previous one, click on C ANCEL to close the
Goto dialogue box, and select the NEW WINDOW command from the Window menu. Then return to the
Goto dialogue box and select this bookmark. Click on OK to display it. The corresponding solution will
be displayed in the new (duplicate) window. You must create additional windows in order to
simultaneously display more than one bookmarked solution from a given query.
69 Return to the GOTO dialogue box and repeat the procedure for each bookmark from the same category.

page_188

Page 189

Select TILE from the WINDOW menu to view all the bookmarked solutions.
70
71 Repeat the procedure with the bookmarks ACR (DEF).There appears to be little difference between
the examples with parentheses and commas.

Now do the same for the bookmarks where these components are in the reverse order:
DEF, ACR
DEF (ACR)

Use the CLOSE ALL option from the WINDOW menu to close the current array of windows.
72
73 Re-open all the original queries and select GOTO to display the full list of saved bookmarks.
74 Repeat the procedure described in this section for these new categories.

10.2.7
Including Punctuation in a Query
Clearly, we have not found anything like all the ways acronyms are defined in the corpus. We have however identified some patterns such definitions
take, and may now be able to design queries to find all the occurrences of particular patterns in order to see how widely they are used, and whether
they have other functions. There are limits to this procedure, since it is only feasible to look for patterns containing specific words. It would, for example,
be absurd to attempt a query looking for the pattern ACR, DEF, as this would involve finding all the commas in the corpus. We could at most
download a sample of the solutions, sort them using POS CODE collating to group those where the comma was preceded by a proper noun (see 7.2.4 on
page 137), and then look to see how many of this group involved acronyms probably a miniscule proportion. Similarly, were we to try to investigate
acronyms which pre-modify a general noun, we could at most look for specific general nouns, such as 'project' and 'scheme', and sort the solutions to
group the relatively few cases where the general noun was preceded by a proper noun.
Where, on the other hand, the acronym and its definition are linked by a recurrent lexical form, we can design a query which has much greater precision.
The pattern DEF, known as ACR contains a fixed lexical sequence which we can incorporate in a Phrase Query. To increase precision, this query
can include not only the words 'known as', but also the comma which precedes them in the pattern.

page_189

Page 190

Click on the PHRASE QUERY button on the toolbar to display the Phrase Query dialogue box. Then type in
75 the string, known as and click on OK to send the query to the server. The Too many solutions
dialogue box will be displayed, stating that there are 767 solutions in 487 texts.
76 Check the ONE PER TEXT box and download the first 250 solutions. We are once more assuming that the
first occurrence in any text is as likely as any to involve a definition.
77 Sort the solutions using the first three words to the right as Primary key, with ASCII collating. This will
group the solutions where 'known as' is followed by a word beginning with an upper case letter, such as
an acronym, at the top of the display.

You will see that many of the solutions at the beginning of the list have the next word in inverted commas generally a nickname. Following these,
however, you should find some acronyms in upper case.

Save the query as known-as.sqy.

78
79 Scroll through the solutions and mark those which correspond to acronym definitions. While the majority
of acronyms are in upper case, there may be some in which only the first letter is capitalised.
80 Delete the unmarked solutions using the THIN option. You should be left with 510% of the original
solution set.

Results: The Collocation Option as a Query Heuristic

While ',known as' clearly has a wide range of other uses (particularly in introducing nicknames), one of its regular functions would appear to be that of
defining or explaining acronyms. There appear to be two main patterns in the data: in the first, the full name is spelt out, followed by ',known as
[acronym]'; in the second, an expression including a general noun such as 'project', 'method', or 'scheme' is used to describe the referent, followed by
',known as [acronym]'.
As we have not downloaded all the solutions (or even a random set), we do not know exactly how frequent these two patterns are. There is no way to
design a query to investigate the first pattern, but we can do so for the latter, for instance in the form:
("project" | "scheme" | "method")(', known as)

To decide whether it is worthwhile doing so (i.e. whether recall will be enhanced, with many more solutions than those already noted being found), you
can use the COLLOCATION option in the current query in order to find how frequently these three words occur close to ', known as'. Not all these
occurrences will necessarily be definitions, but their number may

page_190

Page 191

indicate whether this idea is worth following up. When calculating collocations, remember to use all the available solutions, not just the ones downloaded
in the current query results.
10.3
Discussion and Suggestions for Further Work
10.3.1
Using Bookmarks
While the limited investigation in this task does not allow any generalised conclusions as to ways in which definitions are provided in the corpus, we have
seen how SARA allows you to accumulate instances of particular features which have been located, memorising them under the form of bookmarks, so
as to return to them when the range of instances is sufficient to warrant tentative conclusions, or to formulate testable hypotheses. Bookmarks allow you
to
record particular solutions so you can return to them easily;
annotate particular solutions by assigning them distinctive names;
record and annotate solutions from several queries, so that similarly classified solutions can be retrieved and displayed simultaneously in different
windows.
These features are particularly useful when you are
working simultaneously on various features of a solution set, since different types of bookmark names can be assigned to solutions displaying different
features, and more than one bookmark can be assigned to a particular solution;
working with provisional categories of solutions, since bookmarks do not affect the composition of the solution set and can be deleted, relocated or
renamed as required (unlike the THIN option, which deletes marked or unmarked solutions irrevocably: see 2.2.4 on page 68).
10.3.2
Punctuation in Different Query Types
SARA only allows you to search for sentence punctuation using PHRASE QUERY. In a WORD QUERY, you can only use punctuation where this forms part of
an L-word, typically in abbreviations such as 'Dr.' or 'M.P.' You cannot include such forms in a PHRASE QUERY, as the punctuation will not be interpreted
as being part of the word. PATTERN QUERY functions in the same way as Word Query, always providing that dots, parentheses, inverted commas or
question marks are preceded by a backslash to indicate that the character is to be interpreted literally, not as a variable or special character. In a CQL
QUERY the interpretation of any punctuation will depend on whether it is included within a wordi.e. whether it is included in the double inverted commas
surrounding the wordor within a phrase, i.e. within single inverted commas.

page_191

Page 192

10.3.3
Some Similar Problems
More Serendipity
The solutions to the final query in the last section (see 10.2.7 on page 189) included some very strange nicknames. How not to be intrigued by 'Bridget
Bostock, known as the Cheshire Pythoness' or 'Zen Sharpik, known as Nostrils'? Reopen the unthinned version of the query (KNOWN_AS.SQY) and allow
yourself a few minutes to browse through the solutions, switching to Page display mode, checking sources and browsing in the original text as and when
your curiosity is aroused.
If nothing else takes your fancy, find out what else comes from Cheshire in the corpus, or whether any other people called Zen are mentioned.
Download a random 30 occurrences of 'Cheshire' to see what collocates appear to recur with Cheshire in the solutions. Then use the
COLLOCATION option to investigate their frequency in the entire set of solutions on the server. Use POS QUERY to find occurrences of the name
'Zen', downloading one per text.
There are many more Cheshire cats in the BNC than there is cheese. As well as Zen philosophy, you will find a number of restaurants, a 5-year-
old twin, an Italian detective, and an elkhound.
Looking for Misunderstandings
Not all pragmatic phenomena are so easily identified as definitions. How might you attempt to find instances of misunderstandings in spoken dialogue in
the BNC? As, presumably, anything can be misunderstood in talk, we can hardly predict particular features which are likely to give rise to
misunderstandings in the way we could predict that acronyms might be defined. But we can think of expressions which are likely to be used
when a misunderstanding is suspected or discovered: 'What I meant was'; 'I thought you said'; 'I didn't mean'; 'Ah, I see' and the like. Use the
Query Builder to design queries to find these and similar phrases in spoken dialogue, and browse the preceding context of their solutions to
see if they index misunderstandings. Bookmark those which appear to do so, and see which queries appear most effective as a means of
locating misunderstandings.
Several of these phrases appear to regularly occur in contexts of misunderstanding, including 'what I meant was' (all 4 occurrences), and 'I
didn't mean' (about half the 46 occurrences). Many of the solutions to the latter query occur in apologies another pragmatic function which it
might be interesting to investigate using the BNC.
Not Ms(.) Thatcher
What proportion of the uses of 'Mrs' or 'Mrs.' in the BNC are in reference to Margaret Thatcher? Is she ever referred to as 'Ms Thatcher' or 'Ms.
Thatcher'? Use WORD QUERY to look up the joint frequency of 'Mrs' and 'Mrs.' in the word index, then use the QUERY BUILDER to find all the
occurrences of either form together with 'Thatcher' as a proper noun, in a span of 3. Read off the number of solutions from the Too many
solutions dialogue box, then edit the query to find the frequency of 'Ms Thatcher' or 'Ms. Thatcher'.

page_192

Page 193

Mrs Thatcher is referred to 1759 times in 336 texts, out of a total of 22,150 occurrences of 'Mrs' or 'Mrs.'. She is never referred to as 'Ms' or
'Ms.'. Interestingly, the opposite is true of Ms. Bhutto.
Other SARA Acronyms
The acronym 'SARA' has a variety of meanings in the BNC. What about some of the other acronyms used in this handbook, such as 'POS' or 'CQL'?
A search for 'pos' as a proper noun yields 7 solutions, mainly referring to postal orders and point-of-sale debiting equipment. There are no
solutions for 'cql', which is absent from the word index.
The British National Pumpkin
Acronyms are often defined by spelling out the corresponding name immediately before or afterwards, as in the various 'CPL' examples (see 10.2.4 on
page 184). This suggests that a further strategy we might use to find acronyms is to look for common name components and see if they have acronyms
associated with them. For instance, this handbook regularly uses the acronym 'BNC' to refer to the British National Corpus. Design a query to find out if
the words 'British National' form part of other names, and whether any of these define acronyms. To look for occurrences of 'British' and 'National'
you cannot productively use POS queries, since both these forms are tagged as adjectives, regardless of whether they form part of a name.
Instead you should use the IGNORE CASE option in Phrase Query to restrict the query to cases where 'British' and 'National' are both capitalised.
Sort the solutions by the right with a span of 1, using ASCII collating, so as to group at the head of the display those solutions where the word
following 'British National' begins with a capital letter.
As well as the BNC, you will encounter the BNB ('British National Bibliography') and the BNP ('British National Party') a rather different
organization from the 'British National Pumpkin society', which, alas, has no acronym cited.

page_193

Page 195

III:
REFERENCE GUIDE

page_195

Page 196

1
Quick Reference Guide to the SARA Client
This section summarizes the facilities provided by the SARA client program. In the previous part of the BNC Handbook, these facilities were introduced
in the context of specific tasks, and as a means of exploring the BNC. In this section, for ease of reference we review them in the same order as they
are presented on the initial menu bar. In a number of cases, references are also provided to the tasks in which particular functions listed here are used.
You will also find extensive cross references between the descriptions, and in the on-line help system.
We begin by describing the process of logging on to the server. Once logged on you will see a menu bar across the top of the screen, from which all
the functions of the client program are accessible. The remainder of this section of the BNC Handbook describes these functions.
1.1
Logging on to the SARA System
Once the SARA client program has been correctly installed on your system, you launch it in the same way as any other Windows application, for
example by double-clicking on an icon. A box will appear, displaying version information for the release of SARA you are using. Click on OK or press
ENTER and the LOGON dialogue box will appear. You must have a username and a password before you can access a networked SARA server, for
licensing reasons. These will normally be allocated by the person responsible for managing the BNC server to which you are connecting. Note that the
username may be quite different from that you use for other network services such as e-mail. SARA can also be run in local mode (i.e. with both the
server and the client running on the same machine). In this case, you may not need to log on.
When you have typed in your username, press the TAB key to move to the password box, and type in your password. It will not appear on the screen,
but will be validated by the server when you press the ENTER key or click on the OK button. If you have typed a valid password and username, the box
will be replaced by the message of the day, identifying the server to which you are connected. Press ENTER again, or click on OK, and you will see the
main SARA screen. If you make a mistake entering your password, the system will let you try again. If you want to give up, click on the CANCEL button.
You can change your password but only once you have successfully logged on! (see further section 1.7.5 on page 227).
If your system is not configured correctly, or if the server which you are trying to reach is not available, you will be offered the chance to select a
different server address, or to run SARA in local mode, as further discussed

page_196

Page 197

in section 1.7.5 on page 227. Do not change the server address unless you need to: any changes made here will apply to all subsequent use of SARA
on your machine. Press on the CANCEL button to close the Communications dialogue box if you simply want to try the same server again later.
See section 1.2.1 on page 49 for a more detailed description of the log on procedure.
1.2
The Main SARA Window
All use of SARA is done within the main SARA window. There is a menu bar across the top of this window, which can be used to select the various
SARA commands available to you, and a tool bar which can be used to select particular commands rapidly. The buttons on this bar are reproduced on
the inside cover of this book: see section 1.7.1 on page 225 for a summary of the commands accessible from the tool bar. At the bottom of the screen,
there is an iconified representation of the corpus which you are currently searching (in the current release, this is always the whole of the BNC).
The SARA client can operate in either of two modes. In query mode (the default, and usual mode of operation) the client accepts queries, acts on
them, and displays their solutions in one or more query windows. In browse mode, the client opens a browse window showing the whole of a particular
text, which you can then read. In browse mode, the Query menu is replaced by a Browser menu.
We use the terms query and solution throughout this handbook to refer to two distinct aspects of your interaction with the system. When you use
SARA, you will usually be asking it to find examples of the occurrences of particular words, phrases, patterns, etc. within the BNC. We refer to the
request you make of the system as a query; the set of examples or other response which this request produces we refer to as the solutions. For example
the query dog _ cat will find occurrences of phrases such as 'dog and cat', 'dog or cat', etc., which may be sorted or thinned in a variety of ways;
these are called solutions to the query. The distinction is important, because SARA allows you to save and manipulate queries and their solutions
independently.
To use SARA you give various commands, selected from the menu bar or the tool bar. Some commands affect the behaviour of the client or the server,
for example by setting limits for the amount of text to be downloaded in response to a query, or to change the format of text being displayed by the
client. Other commands create new queries and open new windows in which to display their solutions.

page_197

Page 198

The commands available are logically grouped together according to Microsoft interface guidelines. The order in which they are discussed follows their
order on the menu bar.
To select an item from the menu bar, click on the appropriate word with the mouse, or type the appropriate keyboard shortcut. A menu will open up,
from which further options can be selected with the mouse, possibly with further sub-menus in some cases.
The following options are available from the top level menu structure.
FILE
Most choices on this menu manipulate SARA query files. Select it to create, re-use, or save a query, as well as to print the solutions to a query, or to exit
from the program, as further described in section 1.3 on the next page.
EDIT
Choices on this menu manipulate the current SARA query. Use it to copy selected solutions to the Windows clipboard, or to attach named bookmarks
to particular solutions, for later recovery, as further described in section 1.4 on page 213.
BROWSER
This menu option appears instead of the QUERY menu option if no query is currently open, or when SARA is in browse mode; choices on this menu
control how texts are displayed when you are browsing through them. Use it to turn on or off the display of low level SGML tags in browse mode, as
further described in section 1.5 on page 215.
QUERY
This menu option appears only when you have sent a query to the server to be processed, when it replaces the BROWSER menu option. Most choices on
this menu modify the solutions to the current SARA query. Use it to re-submit a modified version of the current query, to thin, save, or sort the solutions
to the current query, or change the way they are displayed. It can also be used to calculate collocations and to view the source of a particular solution,
as further described in section 1.6 on page 216.
VIEW
Choices on this menu determine how solutions are displayed by SARA. Use it to limit the length of downloaded solutions, and their number, to choose a
default format for displaying them, to set the colours and format of the display, and to configure various options for this and subsequent SARA sessions,
as further described in section 1.7 on page 225.
WINDOW
Choices on this menu allow you to manipulate the windows on the screen. Use it to open a new main SARA window, to tidy the way that existing
windows are displayed, etc., as further described in section 1.8 on page 228.

page_198

Page 199

HELP
Choices on this menu allow you to read the built-in SARA help file. Use it to browse the on-line SARA documentation, as further described in section
1.9 on page 229.
1.3
The File Menu
Most of the commands on this menu manipulate queries, as opposed to the solutions which they return from the corpus: the exceptions are P RINT and
PRINT PREVIEW, both of which relate to the solutions returned by the current query.
The following commands are provided on the File menu.
NEW QUERY
Open a submenu, from which you can select the kind of query you wish to define. A new query window is then opened for you to define that kind of
query. See 1.3.1 on the following page for information about the types of query that may be defined.
OPEN
Open a previously defined query.
CLOSE
Close the current query and its associated window.
SAVE
Save the current query as a file, using the name specified in the title bar of the query window.
SAVE AS
Save the current query as a file, giving an option to change its name from that specified in the title bar of the query window.
PRINT
Print the solutions to the current query.
PRINT PREVIEW
Display on the screen the format in which the current solutions will be printed.
RECENT FILE
Open a recently accessed query (a list of filenames is displayed in the menu at this point).
EXIT
Exit from the Client program.
By default, the first query defined during your SARA session is named Query1, the second Query2, and so on. The name of a query appears in the title
bar of the window containing its solutions, and is also used to identify the file in which the query is saved. This implies that a query name must also be a
valid MS-DOS filename. You can rename a query by means of the SAVE AS command, provided that the name you supply contains only characters
which are legal in filenames under MS-DOS, and does not exceed eight characters in length. The file extension .sqy is assigned by default.
Queries are opened or saved using the normal Windows dialogue boxes for file manipulation, which allow you to change drives, specify filenames, etc. If
you do not know how to use these, consult any introductory text on using Microsoft Windows.
 page_199

Page 200

1.3.1
Defining a Query
The NEW QUERY option on the File menu opens a submenu from which you can select which type of query you want to perform. SARA allows you to
define the following different kinds of query:
WORD
searches the SARA word index, and then optionally also searches the BNC for a word or words selected from those found (see section 1.3.2);
PHRASE
searches the BNC for a phrase (see section 1.3.3 on page 202);
POS
searches the BNC for a word with a specific part of speech (POS) code or codes (see section 1.3.4 on page 203);
PATTERN
searches the BNC for words matching a pattern or regular expression (see section 1.3.5 on page 204);
SGML
searches the BNC for SGML tags (see section 1.3.6 on page 206);
QUERY BUILDER
combines queries of different or the same kinds into a single complex query using a visual interface (see section 1.3.7 on page 207);
CQL
searches the BNC using a query defined in the Corpus Query Language (CQL), SARA's own internal command language (see section 1.3.8 on page
210).
More detail about each kind of query is given in the appropriate section. There is a button on the tool bar for each kind of query: it is generally quicker
to press the button than to select it from the menu. All the buttons on the toolbar are reproduced on the inside of the cover of this book, along with the
names used to refer to them.
1.3.2
Defining a Word Query
Example Word Queries are discussed in sections 2.2.1 on page 64 and 2.3.1 on page 72. A Word Query may be defined in any of the following ways:
select WORD from the submenu of the N EW QUERY option on the FILE menu;
press the WORD QUERY button on the tool bar;
within Query Builder, select WORD from the EDIT submenu..
Any of the above will cause the Word Query dialogue box to be displayed, containing a window into which you can type a word, or part of a word, to
be searched for in the SARA index. If the PATTERN checkbox to the right of the window is checked, whatever you type will be interpreted as a pattern.
If it is not checked, whatever you type will be interpreted as a word stem. (Strictly

page_200

Page 201

speaking, a word stem is also a kind of pattern: the word stem XXX is exactly equivalent to the pattern XXX. *)
The LOOKUP button carries out a search of the SARA index. Every form found in the index which starts with the same letters as the word or part of a
word you typed in will be displayed in the lower window. If the PATTERN checkbox was checked, every word matching the pattern you typed in will be
displayed.
For example, typing in colour with the PATTERN checkbox unchecked will produce a list of words beginning with the letters 'colour', ('colour',
'colours', 'colouring', etc.) If the box is checked, only the word 'colour' will be produced, since this is the only word which matches that pattern.
(Patterns are described in section 1.3.5 on page 204.) Typing in colou?r.* with the pattern box checked will produce a list of all words beginning
with the letters 'color' and 'colour'. Note that, in this case, if the box is not checked, no words will be returned, since there is no word beginning 'colou?
r.*' in the BNC.
A pattern expression which begins with anything other than a literal will usually involve a search through the whole BNC index, which will take a very
long time indeed, and should be avoided. This implies that searches for word-endings are not easily done.
Hyphenated words and words followed by some punctuation characters may not always be indexed in the way you expect. Note that not every item in
the index is a conventional orthographic word: as further discussed in section 2.1.2 on page 34, the index uses L-words which may be parts of
conventional orthographic words, such as 'n't, or orthographic phrases, such as 'in spite of'.
The lower window will not display more than 200 items: a warning message will appear if the word or word part you typed was not specific enough,
perhaps because it was too short. If the word you wish to look up is also a very common prefix, check the PATTERN box to select only the word, rather
than all words beginning with that string of characters.
You can click on one or more of the word forms displayed in the lower window to select them. As is usual with Windows application, clicking on one or
more items with the CTRL key depressed will select each of them; clicking one and then another with the S HIFT key depressed will select both those two
and all the other items between them in the list.
When an item is selected in this way it is highlighted on the screen, and a count is displayed in the box indicating the frequency and z-score for the
selected word forms within the texts (but not the headers) making up the BNC. (For fiurther discussion of these statistics, see section 2.2.1 on page 64.)

page_201

Page 202

When items are selected, the QUERY button can be pressed to carry out a search for these word forms within the BNC. Section 1.3.9 on page 212 gives
further details of the process of downloading the solutions found.
The other buttons in the Word Query dialogue box have the following effects:
COPY
copies the input string to the Windows clipboard;
CLEAR
deletes any previous input and selections from the dialogue box;
CANCEL
leaves the dialogue box without starting a query.
When a Word Query is carried out as a part of a Query Builder query (see 1.3.7 on page 207 below), the Q UERY button is labelled OK and clicking on
it simply adds the selected word or words into the query being constructed.
1.3.3
Defining a Phrase Query
Example Phrase Queries are discussed in sections 1.2.5 on page 51 and 4.2.1 on page 87. A Phrase Query may be defined in any of the following
ways:
select PHRASE from the submenu of the N EW QUERY option on the FILE menu;
press the PHRASE QUERY button on the tool bar;
within Query Builder, select PHRASE from the EDIT submenu.
Any of the above will cause the Phrase Query dialogue box to be displayed. This dialogue box contains a window into which you can type a word or
phrase, a checkbox labelled IGNORE CASE, and a checkbox labelled SEARCH HEADERS.
You can type any sequence of words, or a single word, into the window. Press the OK button (or the E NTER key) and a search is carried out for the
specified phrase within the BNC.
If the S EARCH HEADERS checkbox is checked, then the search is carried out within the text headers as well as the texts. Otherwise, only the texts are
searched.
If the I GNORE CASE checkbox is checked, the search treats upper and lower case forms of a letter as identical. For example, with the box checked, a
search for SARA will recover occurrences of 'SARA', 'Sara', or 'sara'; otherwise, only the first of these will be found.
These two check boxes are the only ways SARA provides for searching in a case sensitive way, or for searching within the headers, other than by using
a CQL query (on which, see section 1.3.8 on page 210).
A Phrase Query can contain punctuation characters as well as words. For example, the query, whereas will find occurrences of 'whereas' only where
they are preceded by a comma. When searching for a match, new lines between

page_202

Page 203

components of a Phrase Query are not significant: for example, it makes no difference if the comma is at the end of one line and the 'whereas' at the
start of the next.
A special punctuation character known as the Anyword character _ can be used within a Phrase Query (but not at the start or end of one). It will match
any L-word, that is, any item in the index. For example, the Phrase Query home _ centre will recover phrases such as 'home loan centre', 'home
improvement centre', 'home planning centre', etc. As noted above, an L-word may be a clitic (such as 'n't') or a phrase (such as 'matter of fact'): this
should be borne in mind when defining queries containing the Anyword character.
Each part of a Phrase Query is searched for separately, and the results are then combined. Consequently, if a Phrase Query contains any very common
words (for example, 'to', 'the', etc.) it may take a very long time to execute: in such cases it is usually better to replace the very high frequency word with
an Anyword character. For example, to find the phrase 'die the death', type die _ death and discard the (fairly small) number of false positives
such as 'die a death', using the THIN command described in section 1.6.3 on page 218.
There is no limit on the number of words a Phrase Query may contain, but the total length of the string may not exceed 200 characters.
Click on the OK button to send the query to the server, or click on the CANCEL button to cancel it; see further 1.3.9 on page 212.
1.3.4
Defining a POS Query
An example POS Query is discussed in section 7.2.1 on page 131. A POS, or part of speech query behaves in the same way as a Word Query, except
that it can only search for a single word, which can be further restricted according to its part of speech (POS) code. It may be defined in any of the
following ways:
select POS from the submenu of the NEW QUERY option on the FILE menu;
press the POS QUERY button on the tool bar;
within Query Builder, select POS from the EDIT submenu.
All of the above will cause the POS Query dialogue box to be displayed, containing two display windows. When the word to be searched for is typed
into the upper window, and the mouse is clicked in the lower window, the lower window is filled with a list of the different parts of speech that the word
in question has been assigned within the corpus. The same effect can be obtained by typing in a word and pressing the TAB key.
For example, the word 'snore' appears in the corpus as the base form of a verb ( VVB), as a singular common noun (NN1), and as a portmanteau
(NN1-VVB),

page_203

Page 204

as well as with various other less frequent codes. All these possibilities appear in the lower box. (The POS codes used in the BNC are further discussed
in section 2.1.2 on page 34.)
To search for the nominal senses only, highlight the NN1 in the lower window, and press OK. To search for both nominal and portmanteau cases, hold
down the control key while highlighting the NN1 and NN1-VVB entries, and then press OK.
Note that it is not possible to search for a particular part of speech without specifying the word to which it is attached. This implies that you cannot use
SARA to search for such things as sequences of three or more adjectives, nor for occurrences of a specific word preceded by any word with a
particular part of speech. You can however group together solutions which have particular POS codes on either side of a given word form by using the
SORT option (see further section 1.6.2 on page 217).
The Help system contains an annotated list of POS codes used in the current version of the corpus: this list also appears in appendix 2.1 on page 230.
A brief explanation of each POS code is also displayed when you select it from the upper box in the POS Query dialogue box.
Click on the OK button to send the query to the server, or click on the CANCEL button to cancel it; see further 1.3.9 on page 212.
1.3.5
Defining a Pattern Query
An example Pattern Query is discussed in section 8.2.2 on page 150. A Pattern Query may be defined in any of the following ways:
select PATTERN from the submenu of the N EW QUERY option on the FILE menu;
press the PATTERN QUERY button on the tool bar;
within Query Builder, select PATTERN from the EDIT submenu.
All of the above will cause the Pattern Query dialogue box to be displayed. This dialogue box contains a window into which you can type a pattern. The
pattern is validated, and a search is carried out for all the words which match it. See further 1.3.9 on page 212.
As noted above in section 1.3.2 on page 200, a pattern can also be typed as part of a Word Query in order to produce a list of matching words. This is
a very useful way of checking the result of a Pattern Query without actually carrying it out by searching the BNC: for an example, see 8.2.1 on page
147.
A pattern is a string of characters which is used as a template to match words in the SARA index. The characters making up a pattern can be:
 page_204

Page 205

literal characters, such as A, B or C, which simply match occurrences of the same character; pattern-matching is never case-sensitive, so a and A are
equivalent;
special characters which behave in a special way within patterns; if a special character is to be used within a pattern but interpreted as if it were a
literal, it must be preceded by the backslash character.
The special characters recognized by SARA are:
.-[]()^?*+|\
The dot special character matches any single character. For example the pattern f ... matches any four letter word beginning with F.
A bracketed sequence matches any one of the characters contained within it. For example, the pattern [ aeiou] matches any vowel.
A sequence can contain a hyphen to express a range. For example, the patterns [09] and [0123456789] are equivalent: either one will match any
digit.
The caret special character can appear at the start of a sequence, to indicate that any character not in the sequence should be matched. For example,
the pattern [^aeiou] will match any which is not a vowel; the pattern [0]9 will match anything which is not a digit.
Single characters or bracketed sequences can be repeated as often as necessary to make up a complete pattern. For example, the pattern [09]
[09] [09] will match all three-digit numbers; the pattern m [09] [09] will match an M followed by two digits.
The question mark special character can follow either a single character or a bracketed sequence of characters, to indicate that the character is
optional. For example, the pattern colou?r will match either 'colour' or 'color'; the pattern [09] [09][09] ? will match all two- or three- digit
numbers, e.g. 99 or 42 or 123 or 912.
The star special character can follow either a single character or a bracketed sequence of characters, to indicate that the character is optional and may
be repeated. For example, the pattern hm [hm]* will match words beginning with HM and containing only those two letters, no matter how long they
are, for example 'hm' or 'hmmmm' or 'hmmhmhmmmm'; the pattern sorrow.* will match any word beginning with the letters 'sorrow', including
'sorrow' itself.
The plus special character can follow either a single character or a bracketed sequence, to indicate that the character appears at least once. For
example, the pattern sorrow.+ will match any word beginning with the letters 'sorrow', except for 'sorrow' itself; the pattern m [09] + will match
all words composed

page_205

Page 206

of the letters M followed by at least one digit, and nothing but digits, e.g. 'M1', 'M2345'; similarly, the pattern e+k will find 'ek', 'eek', 'eeeeek', etc.
The plus or star characters can be used to indicate repetition at any point in a pattern. However, matching of patterns beginning with such sequences
(for example, *ing to recover all words ending with 'ing') is likely to be unacceptably slow, since it requires a scan through the entire word index. In
general, it is best to make the first component of any pattern a literal. Repetition can however be effectively used in the middle of a pattern: for example
effec.*ly will match 'effectively' or 'effectually'.
Two or more patterns can be combined as alternatives using the disjunction special character. For example, the pattern seek¦sought will match
either the word 'seek' or the word 'sought'. Parentheses can be used to group parts of a pattern together: for example, the same effect could be
obtained by the pattern s(eek¦ought).
Any character preceded by the backslash special character will be treated as a literal even if it is a special character. For example, the pattern Mr?s?
\. will match any of 'M.', 'Mr.', 'Mrs.' or 'Ms.'. Without the backslash, the final dot would be interpreted as a special character, matching any character
at all. A backslash is unnecessary within square brackets: the pattern M [rs.]* would have a similar effect to the above, except that it would also
match forms lacking a final dot (plus a number of probably unintended matches, such as 'mss.').
1.3.6
Defining an SGML Query
Example SGML Queries are discussed in sections 5.2.2 on page 100 and 6.2.1 on page 115. An SGML Query searches for an SGML start- or end-
tag, optionally further qualified by attribute values. (SGML the ISO Standard Generalized Markup Language is briefly described above at section 2.1.2
on page 33; see also chapter 5 of the BNC Users' Reference Guide.) An SGML Query may be defined in any of the following ways:
select SGML from the submenu of the NEW QUERY option on the FILE menu;
press the SGML QUERY button on the tool bar;
within Query Builder, select SGML from the EDIT submenu.
All of the above will cause the SGML dialogue box to be displayed.
As well as information about words and their parts of speech, the BNC index searched by SARA contains details of where the SGML elements of
which the corpus is composed begin and end. The start of an SGML element is indicated by a start-tag; its end is indicated by an end-tag. Start-tags
may additionally carry

page_206

Page 207

named attributes, with particular values, to convey additional information about the element occurrences they delimit.
You can use this information to restrict searches to particular types of text (the categorization of a text is indicated by attributes of a <CATREF>
element within its header), or to find particular types of text component for example newspaper headlines, which are mostly tagged <HEAD
TYPE=MAIN> in the BNC, or pauses (<PAUSE>) in spoken texts.
The SGML dialogue box contains a scrollable list of the element names defined for the corpus. For an explanation of the way these elements are used in
the corpus, refer to the BNC Users' Reference Guide. If the S HOW HEADER TAGS checkbox is checked, the names of all elements used in the corpus will
appear; if it is not, then those which are used only in the headers will be excluded. To search the corpus for an SGML start- or end-tag, you select the
name of the element concerned from this list by clicking on it. A brief description of the way this element is used is then displayed.
Provided that the START radio button is selected, a list of any attributes defined for this element will be displayed in the lower left hand window. You can
restrict the search to occurrences of this element having particular values for some combination of these attributes by selecting attribute names from the
list, one at a time, and adding them into the query. Alternatively, if you do not select any attribute name fom the list, the query will select occurrences of
this SGML element whatever attribute values it may have.
When you select an attribute name from the list, clicking on the ADD button will open a further dialogue box in which you can specify the desired value for
the attribute you have selected. For some attributes, the range of possible values is defined in advance, and SARA will therefore present you with a list
from which you can select one or more particular values; for other attributes, the range of possible values is not defined, and SARA will therefore
present you with a window into which you must type the required value explicitly. In either case, when you have specified the value, press OK to close
this dialogue box. Several attribute value constraints may be added in this way. You can also remove a particular constraint by selecting it from the right
hand window in the SGML dialogue box, and then clicking on the REMOVE button, or remove all of them by clicking on the R EMOVE ALL button.
Click on the OK button to send the query to the server, or click on the CANCEL button to cancel it; see further 1.3.9 on page 212.
1.3.7
Defining a query with Query Builder
Query Builder is a special purpose tool which allows you to create complex queries using a visual interface. Some sample uses of Query Builder are

page_207

Page 208

described in sections 4.2.2 on page 91, 4.3.2 on page 95, and 5.2.4 on page 105. The Query Builder command can be used in either of the following
ways:
select QUERY BUILDER from the submenu of the N EW QUERY option on the FILE menu;
press the QUERY BUILDER button on the tool bar.
Either of these will cause the Query Builder dialogue box to be displayed. This dialogue box is used to define a Query Builder query as further described
in this section.
Parts of a complex query are represented in the Query Builder dialogue box by nodes of various types. A Query Builder query always has at least two
nodes: one, the scope node, defines the scope (that is, context) within which a complex query is to be evaluated. The other nodes, which may be linked
in various ways, are known as content nodes. These define the various things which are to be found within this scope. Any form of query can be used in
a content node (except for a CQL or Query Builder query).
For example, you might use the Query Builder to search for the word 'fork' followed or preceded by the word 'knife' within the scope of a single <S>
(sentence) element. Alternatively, you might specify the same search but define its scope as a number of L-words. The default scope for all Query
Builder queries is a <BNCDOC> element, i.e. any one of the 4124 distinct text samples making up the BNC.
The scope of a query is represented in the scope node which appears on the left of the dialogue box. To the right of this is a single empty content node.
Clicking with the mouse inside a content node opens a submenu, from which you can select either EDIT, CLEAR, or (for nodes other than the first one)
DELETE. Selecting EDIT opens a further submenu, from which you select the type of query you wish to define for that node, or, if you have already
defined a query for the node, to edit it. Selecting CLEAR cancels any previous choice, allowing you to select a new query type for the node. DELETE
removes the content node, but leaves the rest of the query unchanged. When a node has been defined, its content can be copied to the clipboard by
selecting COPY from the submenu; this content can then be retrieved into another node by selecting PASTE.
Further content nodes can be added to the right of, above, or below the initial node, simply by clicking the mouse on the branch in that direction. Nodes
added to the right of a node represent alternatives to it. For example, the Query Builder representation of a query to find either the word 'fork' or the
word 'knife' within the scope of a single <S> element would have two content nodes linked horizontally, one searching for 'knife' and the other for

page_208

Page 209

'fork'. (An alternation of this kind could also be represented within a single content node by using a Pattern Query, or a Word Query with alternatives.)
The scope node for this query would indicate the SGML element <S>.
Nodes added above or a content node represent additional constraints. If a content node searching for 'fork' is placed below one searching for 'knife',
then both terms must be found within the scope defined by the scope node, rather than just one of them. The vertical line linking the two content nodes
indicates the order and proximity required. Clicking on the line opens a submenu from which you can select one of the following possibilities:
NEXT
(represented by a thick line): no words or punctuation can appear between the query term indicated above the line and the term below the line;
ONE-WAY
(represented by a downwards pointing arrow): the query term indicated above the line must precede the term below the line within the scope indicated
by the scope node. This is the default link type;
TWO-WAY
(represented by a double-headed arrow): the query terms above and below the line may appear in any order within the scope indicated by the scope
node.
In the current version of SARA, you must use the same kind of link (Next, One-way, or Two-way) between all the content nodes of a single query. If
the content nodes in a query are joined by different kinds of link, no solutions are found.
To change the scope, click on the scope node. A submenu opens, from which you can choose either SGML or S PAN. Choosing SGML opens the
SGML dialogue box, from which you can select an SGML element, possibly modified by attribute values, as in an SGML Query (see further section
1.3.6 on page 206). Choosing SPAN opens a dialogue box in which you can enter the number of L-words within which the whole query must be
satisfied. With a span of n words, the total number of L-words matched (including both content nodes and any intervening words) may not exceed n.
For example, searching for 'knife' and 'fork' within a span of 4 will find 'knife and sharp fork' as well as 'knife and fork' but not 'knife and very sharp
fork'. The maximum span is 99 L-words.
All the content nodes making up a Query Builder query must be satisfied one way or another. Where several content nodes are specified as alternatives,
at least one of them must be satisfied in addition to any constraints expressed by content nodes specified or above the set of alternatives, and vice versa.
Thus, a query in which 'fork' or 'spoon' are specified as alternatives, with an additional

page_209

Page 210

node below specifying 'knife' with a one-way link, will find 'fork and knife' and spoon and knife' but will not find either 'spoon' on its own or 'fork and
spoon'.
A content node can contain any kind of query (other than a CQL or a Query Builder query) one or more alternatives chosen from the Word Query
dialogue box; a Phrase Query; a POS Query; a Pattern Query; or an SGML Query. The Anyword character can also be entered as a content node in its
own right.
Once you have completed defining the query, press the OK button to carry out a search, or press CANCEL to cancel it; see further 1.3.9 on page 212.
1.3.8
Defining a CQL Query
CQL is short for the Corpus Query Language. It is the command language which the SARA client program uses to communicate with the SARA
server. Usually expressions in CQL are generated for you by the client program, but there is no reason why you should not type them in directly as well.
There are also a few features of the command language which cannot be easily (or at all) expressed by the current client except in this way.
Some example CQL Queries are discussed in sections 7.2.3 on page 135 and 9.3.1 on page 176. A CQL Query may be defined in either of the
following ways:
select CQL from the submenu of the NEW QUERY option on the FILE menu;
press the CQL QUERY button on the tool bar.
Either of the above causes the CQL dialogue box to be displayed. This dialogue box contains a window into which you can type a CQL Query. The
query is then validated, and a search is carried out (see further 1.3.9 on page 212).
The syntax of CQL is defined informally here. More detailed information about the language is provided with the SARA server documentation. The
CQL form of any query can always be viewed by switching on the QUERY TEXT option on the QUERY menu (see 1.6.4 on page 218).
Atomic queries
A CQL Query is made up of one or more atomic queries. An atomic query may be one of the following:
a word, punctuation mark, or delimited string, e.g. jam, ?, "Mrs. ";
a word-and-POS pair, e.g. "can" =NN1;
a phrase, e.g. 'not on your life';

page_210

Page 211

a pattern, e.g. {reali?e};

an SGML query, that is, a search for a start- or end-tag: attribute values may also be searched for, e.g. <head type="main">;
the ANYWORD character _, which will match any single L-word.
Unary operators
The following unary operators are defined in CQL:
CASE
The $ operator makes the query which is its operand case-sensitive.

HEADER
The @ operator makes the query which is its operand search within headers as well as in the bodies of texts.

Binary operators
A CQL expression containing more than one atomic query may use the following binary operators:
SEQUENCE
One or more blanks between two queries matches cases where solutions to the first immediately precede solutions to the second.

DISJUNCTION
The | operator between two queries matches cases where either query is satisfied.

JOIN
The * operator between two queries matches cases where both queries are satisfied in the order specified; the # operator between two queries matches cases
where both queries are satisfied in either order.

Query scope
When queries are joined, the scope of the expression may be defined in one of the following ways:
SGML
A joined query followed by a / operator and an SGML start-tag query matches cases where the joined query is satisfied within the scope of the specified SGML
element.

NUMBER
A joined query followed by a / operator and a number matches cases where the joined query is satisfied within the number of L-words specified.

Some simple examples follow:

cat _ dog finds three-word phrases of which the first word is 'cat' and the last is 'dog'
cat*dog finds occurrences of 'dog' preceded anywhere within the same document by 'cat'

page_211

Page 212

cat#dog finds occurrences of 'dog' preceded or followed by 'cat' anywhere within the same document
cat*dog/10 finds occurrences of 'dog' preceded by 'cat' within a span of 10, i.e. with fewer than eight L-words intervening
cat*dog/<head> finds occurrences of 'dog' preceded by 'cat' within a single < HEAD> element
1.3.9
Execution of SARA Queries
Whichever type of SARA query you define, the process of executing it is the same, and proceeds as follows:
press the OK button to send the query to the server;
the red BUSY light on the status bar at the bottom of the main window will be lit, indicating that the server is processing your query;
the server returns a count of the number of hits found to the client;
if this number is less than the Max Downloads figure set in the User Preferences dialogue box (see 1.7.5 on page 227), solutions will start to appear in a
new window, named Query N (where N is the number of this query in the session);
if the number of hits is greater than the Max Downloads figure, the TOO MANY SOLUTIONS dialogue box will appear.
The Too many solutions dialogue box allows you to reset the download limit temporarily, and also to specify which of the available solutions should be
displayed. For an example of its use, see 2.2.3 on page 66. The number of solutions to be downloaded can be re-set manually, by typing a new number
into the box at the bottom, or automatically, by clicking on either or both of the DOWNLOAD ALL and ONE PER TEXT buttons.
In either case, when solutions are downloaded, they appear in order, starting from the beginning of the corpus. If the R ANDOM checkbox is selected,
solutions are chosen at random until the specified number has been reached; if it is not, then either all solutions are chosen, or (if ONE PER TEXT is chosen)
the first in each text, until the limit has been reached.
When downloading is complete, the red BUSY light will go out.
You can scroll, sort, thin, save, see the sources of, or otherwise manipulate the solutions using options from the Query menu, as described in section 1.6
on page 216.
You can interrupt execution of a query at any time before downloading of solutions begins by pressing the E SC key. This will abort processing of the
query as soon as possible.

page_212

Page 213

1.3.10
Printing Solutions to a Query
You can print the solutions to a query in three different ways:
using the PRINT command on the FILE menu (or the PRINT button on the toolbar), you can print the currently displayed set of solutions one-per-line;
using the COPY command on the EDIT menu (or the COPY button on the toolbar), you can save a single solution on the Windows clipboard, and then
import it to a word processor for later printing;
using the LISTING command on the QUERY menu, you can save the whole of a set of solutions to a file in SGML format, and then import it to a word
processor for later printing.
Only the first of these is discussed in this section; for an example of its use, see section 3.2.3 on page 81. The other two are discussed in sections 1.4
and 1.6.6 on page 222 respectively.
Choosing the PRINT command will open a standard Windows Print dialogue box. You can select whether printing should be done in landscape or portrait
mode, the printer to be used, and configure the printer in the normal Windows manner, either from the PRINT dialogue box or directly from the PRINT
SETUP command on the FILE menu.
The current version of SARA does not allow you to change the page layout of the report printed: it contains a running title, derived from the query, and
page numbering. References for each solution are printed down the left margin, indicating the text and the sentence number from which it comes. As
much of each solution as will fit on a single line is included.
You can use the PRINT PREVIEW command on the FILE menu to see a rough indication on the screen of how the solutions will look when printed.
For more flexible formatting of query solutions, use the L ISTING command on the QUERY menu to save the solutions in SGML format, as described in
section 1.6.6 on page 222. This file can then be formatted in any way appropriate using the word processor of your choice.

1.4
The Edit Menu
The Edit menu allows you to save or manipulate one of the solutions to a query, known as the current solution. The menu has three commands:
COPY
copies the current solution to the Windows clipboard;
BOOKMARK
creates a bookmark, that is, a named pointer to the current solution;

page_213
 Page 214

GOTO
moves to a previously defined bookmark, and makes the solution to which it points the current solution.
At any time one of the set of solutions being displayed is known as the current solution: in Page display mode, this is the solution which is visible on the
screen; in Line display mode, it is the solution which has a broken line above and below it (see 1.6.4 on page 218). In Line mode you make a solution
current by scrolling to it, and clicking on it with the mouse. In either mode, you can move forward or backward through the solutions by using the arrow
keys on the tool bar or the cursor keys on the keyboard. For examples of their use, see 1.2.6 on page 51.
Choosing the COPY command from the Edit menu copies the current solution to the Windows clipboard, in whatever format it has on the screen. For an
example of its use, see 7.2.1 on page 131. If you now open a Windows application such as Notepad or some other word processor, and select the
PASTE command from its Edit menu, the current solution will be available to that application. This is a simple way of copying information from SARA to
other programs.
A bookmark is a name which you can attach to the current solution so that you can refer to it again, after it has ceased to be current. Bookmarks are
specific to particular queries, and are saved and retrieved along with queries. For an example of their use, see 10.2.2 on page 181.
To create a bookmark for the current solution, select the BOOKMARK command from the Edit menu. The BOOKMARK NAME dialogue box will appear, into
which you can type any short name for the bookmark. If the name is already in use, you will be offered the choice of overwriting the current bookmark
of that name. A list of the names you are currently using appears in the lower part of the window.
To make a different solution current, you can choose one of your named bookmarks by means of the G OTO command on the Edit menu. Choosing this
command opens the GOTO dialogue box.
Clicking on a bookmark in this window will make the solution to which it points the current one, assuming that the solutions concerned are still available.
If you delete a solution, for example by thinning, any associated bookmark will also be deleted. Sets of solutions remain available for as long as they are
either visible in an open window or present as minimized icons on the desk top. If you save a query (using the SAVE or SAVE AS command on the FILE
menu), its associated bookmarks are saved along with it. If you subsequently close the query (using the CLOSE command on the FILE menu, or the
window controls),

page_214

Page 215

its associated bookmarks disappear from the Goto dialogue box, but will reappear if you re-open the query.
1.5
The Browser Menu
The Browser menu is available only when SARA is in browse mode. In this mode, you are able to browse through the whole text of any part of the
corpus in a special purpose window. For an example of its use, see 1.2.7 on page 52.
When SARA starts, it is initially in browse mode. As soon as you open a query, or create a new one, SARA switches out of browse mode, and the
Browser menu is replaced by the Query menu, discussed in section 1.6 on the following page. To open the browse window, click on the BNC icon at
the bottom left of the SARA main screen, or click on the BROWSE button which appears on the bottom right of the S OURCE window described in section
1.6.7 on page 223.
In browse mode, you can move from one text in the corpus to the next, simply by using the arrow keys on the button bar, or their equivalent keyboard
shortcuts. The texts appear in alphabetical order, according to their three character identifiers.
Each text sample of the BNC has a similar structure, represented by SGML elements. Each text is represented by a <BNCDOC> element, which is
composed of a <HEADER> element and either a <TEXT> or an <STEXT> element, depending on whether the text is written or spoken. These
elements are all further subdivided into elements of other named kinds. <HEADER> elements have a rather complex substructure, following
international standards for bibliographic description. Both <TEXT> and <STEXT> elements are composed fundamentally of <S> (sentence)
elements, which contain a mixture of <W> (word) elements and <C> (punctuation) elements. In written texts, these are grouped into elements such as
<P> (paragraph) or <HEAD> (heading); in spoken texts, they are grouped into <U> (utterance) elements. (For a more detailed description, see the
BNC Users' Reference Guide.)
In browse mode, this structure is presented visually in the form of a list of container elements, each of which can be selectively expanded. When a text
is first displayed, only the outermost <BNCDOC> element containing it is visible. It appears in the Browse window, with a plus sign to the left of the
SGML start-tag, which indicates that this element is not yet fully expanded. Click on the plus sign to see the SGML elements of which it is composed: a
<HEADER> element, and a <TEXT> or <STEXT> element. Click on the plus sign before one of these elements to expand this in its turn.
When an element is expanded, the plus sign in front of its start-tag turns into a minus sign, indicating that that element has been expanded. You can

page_215
 Page 216

continue in this way, expanding elements down to the lowest level <W> and <C> elements for any text. If you click on a minus sign, the expansion of
the element will be removed.
If you entered browse mode by clicking on the B ROWSE button during display of the solutions to a SARA query, a red horizontal line will also appear in
the Browse window. This line marks the place in the text where the current solution was found; you can move directly to this point by clicking on the box
at the left end of the red line. Since this requires that the whole of the text must be downloaded from the server to the client, there may be some delay
between your clicking on the box, and the display of the element containing the solution. Once the text is available, the display will automatically scroll to
it.
You can now inspect the content of any elements before or after the hit by clicking on the plus signs, as before.
You use the TAGS command on the BROWSER menu to determine which of the SGML tags around parts of the text are to be displayed. By default, all
tags are displayed in the Browse window (the TAGS command on the BROWSER menu is checked); click on this command to switch off display of the
low-level tags <W>, <C> and <S>.

1.6
The Query Menu
The Query menu allows you to manipulate the solutions to a query in various ways. You can edit a query using the EDIT command; sort the solutions in
various ways using the SORT command; thin the solutions using the THIN command; set various options about the appearance of the solutions using the
Concordance, Options, QUERY TEXT or ANNOTATION commands; save the solutions to a file in SGML format using the LISTING command; or display bibliographic
information about a particular solution using the SOURCE command. You can also calculate collocational information for the solutions to a query using the
COLLOCATION command.
1.6.1
Editing a Query
For an example of the use of the E DIT command, see section 4.2.1 on page 88. Selecting the EDIT command from the Query menu will re-display
whichever dialogue box it was that launched the query whose solutions are currently displayed. The command is grayed out and unavailable if no
solutions are being displayed.
The query dialogue box will be displayed as it was when the query was sent to the server by pressing the OK button. You can change any part of the
dialogue box, and resubmit it by pressing the OK button again, or press CANCEL to close the dialogue box and return to the previous display of solutions.

page_216

Page 217

1.6.2
Sorting Solutions
By default, the solutions to a query are displayed in their order of appearance within the corpus, alphabetically ordered by its three character filenames.
This is rarely of any particular significance, except to group solutions from the same text, and so it is generally desirable to re-order a line mode
(concordance) display. This is done by selecting the SORT command from the Query menu, which displays the SORT dialogue box. For an example of its
use, see section 3.2.2 on page 78.
You can use the radio buttons in this dialogue box to specify either one or two keys for the sort, and a single collating sequence, applicable to both
keys. The keys determine which part of each hit is to be used to sort the solutions; the collating sequence determines how these keys are to be
compared when deciding on their relative order.
The PRIMARY keys for all the context lines are compared first, according to the collating sequence indicated. If any duplicates are found, the S ECONDARY
keys are used to order them. Note that the same collating method must be used for both keys.
The SPAN box indicates how many words make up the key in each case. The LEFT, CENTRE or RIGHT radio buttons indicate the position of the key
relative to the query focus (i.e. the hit word, phrase, or SGML element in the context). If the L EFT radio button is selected, and the SPAN is 1, the key
will be the word to the left of each query focus. If the CENTRE radio button is selected, and the SPAN is 1, the key will be the first word of the query
focus itself. If the RIGHT radio button is selected and the SPAN is 1, the key will be the first word following the query focus.
The ASCENDING and DESCENDING radio buttons indicate whether the keys are to be sorted into ascending or descending alphabetical order.
The collating method used for both keys is indicated by the radio buttons to the right of the dialogue box. With the ASCII radio button selected, keys
are compared according to the ASCII character sequence, in which all uppercase letters precede all lower case ones: 'Zebra' precedes 'antelope'. With
the IGNORE CASE button selected, case distinctions are ignored, so that 'Zebra' and 'zebra' are regarded as the same key. With the IGNORE ACCENTS button
selected, accented letters are treated as if they were unaccented, so that 'élève' and 'élevé' are regarded as the same key.
In Plain and Custom format displays, span is calculated and sorting carried out by orthographic (i.e. blank-delimited) words, as displayed on the screen.
If the solutions being sorted are being displayed in either POS or SGML format (see further section 1.6.4 on the following page), then span is calculated
and
 page_217

Page 218

sorting carried out by L-words; additionally, the POS CODE button is available for selection. Selecting it causes keys to be sorted not by their
orthographic form but the alphabetical order of their part of speech code. This has the effect of grouping together keys with the same POS code. You
can use it, for example, to sort a set of solutions by the POS code of the word following the query focus.
1.6.3
Thinning Solutions
Selecting the THIN command from the QUERY menu opens up a sub-menu from which four selections are available, each of which allows you to reduce
the number of solutions for the current query. For an example of its use, see section 2.2.4 on page 68. The commands available are:
SELECTION
discards from the solutions all those which have not previously been selected (i.e. all those solutions which do not appear on the screen in reverse video
are discarded);
REVERSE SELECTION
discards from the solutions all those which have previously been selected (i.e. all those solutions which appear on the screen in reverse video are
discarded);
RANDOM
solutions are discarded at random until the number of solutions matches the number you specify in a sub-window;
ONE PER TEXT
discards all but the first solution from any one text.
The current item in a displayed list can be selected either by double-clicking on it, or by pressing the space bar.
Each time you request a random selection from a given set of solutions, you will get a different random sequence. The only way to get the same random
selection more than once is to save the query after thinning it. When a thinned query is saved, any thinning is saved at the same time.
1.6.4
Options for Displaying Solutions
The solutions to a query can be displayed in one of two modes and in one of four different formats. You can also vary the amount of context or scope
displayed for each solution. Which options are in effect for a particular query will depend on the initial settings specified by the USER PREFERENCES dialogue
box (see 1.7.5 on page 227).
Display mode
In Line mode, each occurrence of the item searched for is displayed as a single line on the screen; in Page mode, each occurrence is displayed in full on
the screen, taking as many lines as necessary. For a detailed discussion, see section 1.2.6 on page 51.
The CONCORDANCE button is used to switch between one mode and the other. The initial mode is set by the CONCORDANCE checkbox in the USER

page_218

Page 219

PREFERENCES dialogue box (see 1.7.5 on page 227): if this is checked, Line mode is used; otherwise Page mode is used. Clicking on the C ONCORDANCE
button or selecting the CONCORDANCE command from the QUERY menu enables you to switch modes for a particular set of solutions.
The usual Windows controls are available to enable you to display different parts of a large set of solutions. In Line mode, you can use the vertical scroll
bar to the right of the window to scroll up and down the solutions; in either mode, you can use the arrow buttons in the tool bar to step through the
solutions one at a time. You can also use the cursor keys, PGUP and PGDN, HOME and END, to move through the solution set in the usual way.
Display format
For further discussion of the four display formats, see section 1.2.8 on page 55. Select the O PTIONS command from the QUERY menu to display the
QUERY OPTIONS dialogue box. The radio buttons selected here determine the format used to display the current solutions and the amount of context (or
scope) visible to either side of the query focus, as further discussed in section 1.6.4.
The following four display formats are available:
PLAIN
only the words and punctuation of each hit are displayed, optionally with the query focus in a different colour or typeface (as determined by the fonts
and colour selected: see 1.7.4 on page 227);
POS
part of speech information for any word on the screen can be displayed by clicking on it with the right mouse button; in addition, words may be
displayed in different colours depending on their part of speech, as determined by the colour scheme in use (see 1.7.4 on page 227); selecting this
format also makes it possible to sort the solutions by their part of speech code (see 1.6.2 on page 217);
SGML
each hit is displayed with its full SGML markup;
CUSTOM
each hit is displayed according to a user-defined format, as further discussed in section 1.6.4 on the following page; the CONFIGURE button can be used to
change this format if the default is inappropriate.
Changing any of these options will affect the display of the current query only. To change the display of all subsequent queries, changes must be made in
the USER PREFERENCES dialogue box (see 1.7.5 on page 227). Note also that changing the format of the display will usually require that the solutions be
downloaded again.
Display Scope
The maximum amount of context which can be displayed for each hit is set by the M AX DOWNLOAD LENGTH specified in the USER PREFERENCES dialogue box
(see 1.7.5 on page 227). This sets an upper limit, as

page_219

Page 220

a number of characters. Setting it very high will result in long download times; setting it too low will limit the usefulness of what can be displayed on the
screen.
Within this overall limit, there are four options for determining the amount of context displayed on the screen by default:
AUTOMATIC
the whole of the smallest unit (larger than a <W>) within which the query focus appears;
SENTENCE
the whole of the <S> element within which the query focus appears;
PARAGRAPH
the whole of the <P> or <U> element within which the query focus appears;
MAXIMUM
as many <S> elements as possible on either side of the query focus, up to the limit imposed by the Max download length.
If the scope setting results in less than the Max download length being displayed, you can always expand what is displayed up to that maximum by
double-clicking on the display with the right mouse button. This will expand the context up to what would have been obtained if the Maximum scope
setting were in force, for the current hit only.
The query focus is that part of a downloaded hit which is normally highlighted within the display. In a simple Word, Pattern, or Phrase Query, it is the
whole of the word or phrase found which matched the query. In an SGML Query, it is the SGML start- or end-tag which matched the query. In a
Query Builder query, it is either the part of the text which was matched by the last content node, i.e. that nearest the bottom of the screen, or the part
matched by a group of content nodes joined by NEXT links.
Custom display format
In CUSTOM format, hits are displayed according to a format which you can tailor to your own liking. You can specify whether or not particular SGML
elements should be displayed starting on a new line, whether or not their associated attributes should be displayed, and also specify additional characters
to be displayed in association with them.
Two such specifications can be supplied; one, held in the file linefmt.txt determines how hits should be displayed in Line mode displays; the
other, held in a file called pagefmt.txt, determines how hits should be displayed in Page mode displays. These are ordinary ASCII files which can
be edited and displayed by any editor (such as Notepad), or by pressing the CONFIGURE button in the QUERY OPTIONS dialogue box. The files must be held
in the working directory used by the SARA client on your system, and must be writable. (See further section 1.10 on page 229.)

page_220

Page 221

The syntax of these files is fairly self explanatory. Each line specifies how a particular element type is to be displayed: if no line is supplied for any
element, no special action is taken for it. A line begins with the name of an element (optionally followed by one or more attribute names) or an entity.
This is followed by a quoted string, which gives the replacement value for the named entity or for the element's start-tag. A second quoted string can also
be supplied to provide a replacement for an element's end-tag. Within replacement strings, the string %s is used to represent the value of the attribute
whose name was specified. Formats intended for use in Page mode displays can also use the string \n to indicate a new line and \t to indicate a tab
indent.
For example, the default page format file contains the following lines:
div1 "\n"

pause "..."

event desc "[%s]"

u who "\n{%s}: "

The first line indicates that the display should start a new line at the start of each new <DIV1> element. The second line indicates that any <PAUSE>
element should be displayed as three dots. The third line indicates that any <EVENT> element should be displayed as whatever value has been supplied
for its DESC attribute, enclosed in square brackets. Finally, the last line indicates that the content of every <U> element should be prefixed by the start of
a new line, the value of its WHO attribute within curly brackets, followed by a colon and a space.
For a more detailed example, see section 9.2.2 on page 165. Care should be taken in preparing custom format files, as no syntax checking is currently
performed; if you plan to modify them extensively, you are recommended to make back-up copies of the files linefmt.txt and pagefmt.txt
before you begin.
1.6.5
Additional Components of the Query Window
In addition to the display of solutions, the query window can contain two other components, each in a separate pane.
QUERY TEXT
The QUERY TEXT command from the QUERY menu opens or closes a pane in which the CQL text of the current query appears. This cannot be changed,
but is useful for documentary purposes. For the syntax of CQL, see section 1.3.8 on page 210. Any thinning options applied to the query are also
displayed

page_221

Page 222

ANNOTATION
The ANNOTATION command from the QUERY menu opens or closes a pane in which you can write any comment or annotation you wish. Such
documentation may be useful for future reference when re-running a query.
Both query text and annotation are saved together with the query, along with any valid bookmarks you defined for it.
1.6.6
Saving Solutions to a File
Selecting the LISTING command from the QUERY menu opens a standard file dialogue box in which you can specify a name for the file in which the current
solutions are to be saved. For an example of its use, see section 8.2.5 on page 154. The solutions are saved in SGML format in a file with the same
name as the query itself and with the suffix .sgm.
Here is the start of a sample listing file, showing the results of a search for the word 'corpuses'.
<!DOCTYPE bncXtract
PUBLIC "-//BNC//DTD BNC extract 1.0//EN">
<bncXtract>
<hdr date='10-Nov-1996 00:03:29' user=lou
server='163.1.32.247' format=untagged>
<source>This data is extracted from the
British National Corpus. All rights in the
texts cited are reserved. This data may not
be reproduced or redistributed in any form,
other than as provided for by the Fair Use
provisions of the Copyright Act</source>
<query><![CDATA["corpuses"] ]></query> </hdr>
<hit text=EWA n=531><left> Where an absolute norm
for English cannot be relied on, the next best
thing is to compare the corpus whose style is
under scrutiny with one or more comparable
<focus>corpuses<right>, thus establishing a
relative norm. </hit>
<hit text=FRG n=1222><left> These methodological
difficulties are associated with a more general
problem of deriving generalizations
 page_222

Page 223

from <focus>corpuses<right,>. </hit>

</bncXtract>

Housekeeping information about the query itself is saved in a <HDR> element at the start of the file, giving the date the query was solved, the name of
the user, and the server machine, as well as the actual text of the query. Each solution to the query is saved as a separate <HIT> element. The TEXT
attribute gives the three character identifier of the text in which the hit was found; the N attribute gives its sentence number. The query focus of the hit is
represented as a <FOCUS> element; its left context is represented as a <LEFT> element, and its right context as a <RIGHT> element.
Solutions are saved in a listing file in the format in which they are displayed. Thus, if the solutions include SGML tags (i.e. solutions are being displayed
in POS or SGML format), these tags will also appear in the listing file, which may make it difficult to process by other SGML-aware software. To make
this less problematic, any angle brackets appearing as content of a <HIT> element are converted to square brackets before the listing file is produced.
For example, the second <HIT> element above would appear as follows if the same query were saved in SGML mode:
<hit text=FRG n=1222><left>[s n=1222] [w DT0]These
[w AJ0]methodological (w NN2]difficulties [w VBB]are
[w VVN]associated [w PRP)with [w AT0]a [w AV0]more
(w AJ0]general [w NN1]problem [w PRF] of
[w AJ0-VVG]deriving [w NN2] generalizations
[w PRP]from [w NN2]<focus>corpuses<right>(c PUN].
</hit>

Note that both the above examples have been reformatted to fit on the printed page: in an actual listing file, no extra line breaks are introduced within the
body of a <HIT> element. A full specification of the listing file format is included in 3 on page 240.
1.6.7
Displaying Bibliographic Information and Browsing
Selecting the SOURCE command from the QUERY menu or clicking on the SOURCE button on the tool bar will display a BIBLIOGRAPHIC DATA window
containing information about the text in which the currently selected solution appears. It also gives an indication of the size of the text in words and s-
units. The information presented is the same as that available from the reference list included in the BNC Users' Reference Guide. Further information
about a

page_223

Page 224

text, for example its classification, is available only by inspecting elements in its header.
The Bibliographic data for a written text will generally specify its author, title, date, and publisher. The Bibliographic data for a spoken text will identify
the situation in which it was recorded, and will also supply demographic or descriptive details for each person speaking in a lower window. This window
can be scrolled left to right or up and down as needed.
Click on the OK button to close the Bibliographic data window. Click on the BROWSE button to switch to browse mode, enabling you to inspect the
whole of this text, as discussed above in section 1.5 on page 215.
1.6.8
The Collocation Command
The COLLOCATION command allows you to calculate how frequently words collocate, i.e. appear together within the current results. For an example of its
use, see section 3.2.1 on page 76. For example, if your current query solutions show occurrences of the word 'death', you might wish to see how often
the word 'die' appears within a certain number of words of the query focus.
Selecting the COLLOCATION command from the QUERY menu opens the COLLOCATION dialogue box. The name of the current query is displayed, together
with the number of hits. Enter the L-word, punctuation mark, or SGML start-tag for which a collocation score is required (the collocate) in the box
labelled COLLOCATE and press the CALCULATE button. Two counts appear in the box, indicating how often this collocate appears within a specified span,
and what proportion of the hits this represents. You can repeat this process as often as you like, with each new collocate appearing in the same results
box. If the collocate appears more frequently than the query focus itself, it is displayed in the highlight colour.
Collocation scores are calculated within the SPAN (i.e. number of L-words) indicated in the box at the bottom left of the dialogue box, by default one
word to either side of the first word in the query focus. The span is always counted from the leftmost end of the query focus. Changing the span causes
the scores for all words to be recalculated. The maximum span is 9 L-words to either side of the first word in the query focus.
You can calculate collocation scores either with respect to the number of hits actually downloaded, or with respect to the number of hits present in the
corpus, depending on the setting of the USE DOWNLOADED HITS ONLY checkbox in the top left corner of the dialogue box.
Note that it is not possible to find out which words collocate strongly with a given word other than by trial and error: you must specify the words for
which
 page_224

Page 225

a collocation score is required. It is also impossible to specify a pattern or phrase as a collocate.

You can print the contents of the Collocation dialogue box at any time, by clicking on the P RINT button. This is the only way of saving the results of a
Collocation analysis in the current release of the software.
1.7
The View Menu
This menu contains commands which allow you to customize the appearance of the main SARA window.
1.7.1
Tool Bar Command
Choose this command to display or hide the tool bar at the top of the window. A check mark appears next to this menu item when the tool bar is
displayed.
The tool bar contains a row of buttons, each of which provides rapid access to one of the following commonly used SARA functions. The buttons on
the tool bar are reproduced inside the cover of this handbook. Here they are listed in left to right order, together with a brief description of what each
one does.
OPEN
Opens an existing query. Equivalent to selecting OPEN from the FILE menu (see 1.3 on page 199).
SAVE
Saves the current query. Equivalent to selecting SAVE from the FILE menu (see 1.3 on page 199).
COPY
Copies the current solution to the clipboard. Equivalent to selecting COPY from the EDIT menu (see 1.4 on page 213).
PRINT
Prints the solutions to the current query. Equivalent to selecting PRINT from the FILE menu (see 1.3 on page 199).
WORD QUERY
Opens the Word Query dialogue box (see 1.3.2 on page 200).
PHRASE QUERY
Opens the Phrase Query dialogue box (see 1.3.3 on page 202).
POS QUERY
Opens the POS Query dialogue box (see 1.3.4 on page 203).
PATTERN QUERY
Opens the Pattern Query dialogue box (see 1.3.5 on page 204).
SGML Q UERY
Opens the SGML dialogue box (see 1.3.6 on page 206).
QUERY BUILDER
Opens the Query Builder dialogue box (see 1.3.7 on page 207).
CQL Q UERY
Opens the CQL dialogue box (see 1.3.8 on page 210).
CONCORDANCE
Toggles between Line and Page mode for display of solutions. Equivalent to selecting CONCORDANCE from the QUERY menu (see 1.6.4 on page 218).
ALIGN
Toggles between left, centre, and right alignment of the highlighted query focus in a Line mode display (no menu equivalent).

page_225

Page 226

FIRST SOLUTION
Selects the first of a set of solutions (no menu equivalent).
PREVIOUS SOLUTION
Selects the solution preceding the current one (no menu equivalent).
NEXT SOLUTION
Selects the solution following the current one (no menu equivalent).
LAST SOLUTION
Selects the last of a set of solutions (no menu equivalent).
SOURCE
Opens a Bibliographic data window for the currently selected solution. Equivalent to selecting SOURCE from the QUERY menu (see 1.6.7 on page 223).
HELP
Opens the on-line Windows help file. Equivalent to pressing F1 (see 1.9 on page 229).
CONTEXT HELP
Switches to Context Help mode (no menu equivalent). In this mode, the cursor changes to a large question mark. Move the cursor to any button or box
and click on it to display a brief explanation of its function.
1.7.2
Status Bar Command
Choose this command to display or hide the status bar at the bottom of the main SARA window. A check mark appears next to the menu item when
the status bar is displayed.
The status bar has three areas: the leftmost part is used to display messages describing the action to be executed by the currently selected menu item or
tool bar button; the central part provides information about the currently selected solution; the rightmost part displays information about the current state
of the keyboard.
The left area of the status bar describes actions of menu items as you use the arrow keys to navigate through menus. This area similarly shows messages
that describe the actions of tool bar buttons as you depress them, before releasing them. If after viewing the description of the tool bar button command
you wish not to execute the command, then release the mouse button while the pointer is off the tool bar button.
The centre area of the status bar identifies the corpus (BNC) and then the currently selected solution. The first pane shows the number of this solution,
the total number of solutions and the number of texts from which they are taken; the next shows the three letter identifier of the text in which the current
solution occurs; and the last gives the number of this <S> element within this text. For example, if the currently selected solution occurs in the 145th s-
unit of text ABC, and is the third of 12 solutions taken from 8 texts, the display will read 3:12 (8) ABC 145.

page_226

Page 227

The status bar display may not be updated until a set of solutions has been completely downloaded.
The rightmost part of the status bar indicates which if any of the three lockable keys on the keyboard (C AP, NUM, or SCRL) is currently latched down.
1.7.3
Font
Selecting FONT from the VIEW menu causes the standard Windows Font dialogue box to appear. This dialogue box lists the fonts available on your
system. You can use it to set the font in which solutions are to be displayed.
1.7.4
Colours
Selecting COLOURS from the VIEW menu causes the COLOURS dialogue box to appear. This is used to specify the colours used to display parts of the
solutions. Different colours may be specified for each part of speech code, for the query focus (i.e. the actual hit word) and for the default. You can also
specify that any of these items is to be displayed in a bold face, in italic, or both.
To change the colour used for words of a particular type, first highlight the relevant item or items from the list at the left of the dialogue box by clicking
on them with the mouse, in the same way as words are selected from the Word Query box. Next press the COLOUR button, which causes the Windows
Color dialogue box to open. Click on the desired colour. The dialogue box disappears, and the SAMPLE window in the COLOURS dialogue box changes to
show the effect of the choice just made. If this is unsatisfactory, click the RESET button to revert to the original colour. The BOLD and ITALIC check boxes
can be selected to change the weight and slant of the selected items independently of their colour.
A set of such specifications is known as a colour scheme. Each colour scheme is saved in a file with extension .col. The name of the colour scheme
currently in force is displayed at the top right of the Colours dialogue box: to select a new scheme, press the OPEN SCHEME button. This opens a file
dialogue box, in which the scheme may be named. The SAVE SCHEME button saves the set of colours currently defined. The M ERGE SCHEME button opens an
existing colour scheme and allows you to modify it further.
Click the OK button to close the dialogue box and save all changes. Click the CANCEL button to leave without changing the colour scheme.
1.7.5
User Preferences
Selecting PREFERENCES from the VIEW menu causes the USER PREFERENCES dialogue box to be displayed.
This dialogue box is used to set the default behaviour of the SARA client. Changes made here affect all subsequent queries in this and following
sessions,

page_227

Page 228

but not any currently displayed solutions. In addition to changing the defaults for the way that results are displayed (as discussed above in section 1.6.4
on page 219) you can reset
the maximum download length;
the maximum number of hits to be downloaded;
the address of the server to which the client will attempt to connect;
your personal password.
Pressing the COMMS button displays the COMMUNICATIONS dialogue box. To connect to a different server, you will need to know two numbers: the host
name or IP address of the computer concerned, and the port on which it listens for calls from a SARA client. Consult your local administrator for more
information on these.
Pressing the PASSWORD button displays the PASSWORD dialogue box. To change your password, you must supply your current password, and type the
new password twice. The new password will take effect from the next time you try to log in.
1.8
The Window Menu
The commands on this menu allow you to move and manipulate the windows on the screen, in the same way as most other Microsoft Windows
applications. The following commands are available:
NEW WINDOW
Opens a new window with the same contents as the currently active window. You can open multiple query windows to display different parts or views
of a query at the same time. If you edit the query or thin the set of solutions, all other windows containing the same query reflect those changes. When
you open a new window, it becomes the active window and is displayed on top of all other open windows.
CASCADE
This command arranges the open windows so that they overlap each other, with the title bar of each window staying visible.
TILE
This command re-arranges the open windows so that none of the windows overlaps any other.
ARRANGE ICONS
Choose this command to arrange the icons for minimized windows at the bottom of the main window. If there is an open query window at the bottom
of the main window, then some or all of the icons may not be visible because they will be underneath this query window.
WINDOW 1, 2, ...
At the bottom of the Window menu SARA displays a list of all currently open query windows, with a check mark in front of

page_228

Page 229

the one which is currently active. You can choose a different query name from this list to change the currently active window.
1.9
The Help Menu
The commands on this menu allow you to consult SARA's built-in help system, in the same way as most other Microsoft Windows applications. The
following commands are available:
INDEX
Displays the opening screen of the built-in help file, which contains an index of its contents.
USING HELP
Displays general instructions on using the Windows help system.
ABOUT SARA
Displays the copyright notice and version number of your copy of SARA.
1.10
Installing and Configuring the SARA Client
Before trying to install the SARA client, you should find out the following information:
the host name or IP address of the computer system on which the server you plan to use is running;
the port on this computer system where the server you plan to use listens for incoming calls.
You should also check that the computer on which you plan to install the client has the following characteristics:
direct connection to a TCP/IP network from which the server you plan to use is accessible;
about 1 Mb free disk space;
runs either Microsoft Windows 3.1 or above, or Windows95.
The installation process creates a SARA directory to hold client executables and a number of parameter files. If you are installing the software on a local
network, you should take care that these parameter files are installed in a writable directory, though the executable itself need not be.
Please read the README file supplied with the software for details of any changes and for specific details of the installation process. Up to date
information about the current state of the SARA software is also regularly announced on the Bnc-discuss mailing list, and on the BNC's own web site.
For further details, see http://info.ox.ac.uk/bnc/sara.html

page_229

Page 230

2
Code tables
2.1
POS codes in the CLAWS5 tagset
The following tables list the codes used for part of speech. These codes appear as attribute values within <W> or <C> elements as appropriate. Note
that some punctuation marks (notably long dashes and ellipses) are not tagged as such in the corpus, but appear simply as entity references.
Code Usage
AJ0 adjective (general or positive), e.g. 'good', 'old'.
AJC comparative adjective, e.g. 'better', 'older'.
AJS superlative adjective, e.g. 'best', 'oldest'.
AT0 article, e.g. 'the', 'a', 'an', 'no'. Note the inclusion of 'no': articles are defined as determiners which typically begin a
noun phrase but cannot appear as its head.
AV0 adverb (general, not sub-classified as AVP or AVQ), e.g. 'often', 'well', 'longer', 'furthest'. Note that adverbs, unlike
adjectives, are not tagged as positive, comparative, or superlative. This is because of the relative rarity of
comparative or superlative forms.
AVP adverb particle, e.g. 'up', 'off', 'out'. This tag is used for all 'prepositional adverbs', whether or not they are used
idiomatically in phrasal verbs such as ''Come out here'', or "I can't hold out any longer".
AVQ wh-adverb, e.g. 'when', 'how', 'why'. The same tag is used whether the word is used interrogatively or to introduce a
relative clause.
CJC coordinating conjunction, e.g. 'and', 'or', 'but'.
CJS subordinating conjunction, e.g. 'although', 'when'.
CJT the subordinating conjunction 'that', when introducing a relative clause, as in "the day that follows Christmas". Note
that this is treated as a conjunction rather than as a relative pronoun.
CRD cardinal numeral, e.g. 'one', '3', 'fifty-five', '6609'.
DPS possessive determiner form, e.g. 'your', 'their', 'his'.
DT0 general determiner: a determiner which is not a DTQ, e.g. 'this' both in "This is my house" and "This house is mine".
A 'determiner' is defined as a word which typically occurs either as the first word in a noun phrase, or as the head of a
noun phrase.
 page_230

Page 231

Code Usage
DTQ wh-determiner, e.g. 'which', 'what', 'whose', 'which'. The same tag is used whether the word is used interrogatively or to
introduce a relative clause.
EX0 existential 'there', the word 'there' appearing in the constructions "there is ...", "there are ...".
ITJ interjection or other isolate, e.g. 'oh', 'yes', 'mhm', 'wow'.
NN0 common noun, neutral for number, e.g. 'aircraft', 'data', 'committee'. Singular collective nouns such as 'committee' take
this tag on the grounds that they can be followed by either a singular or a plural verb.
NN1 singular common noun, e.g. 'pencil', 'goose', 'time', 'revelation'.
NN2 plural common noun, e.g. 'pencils', 'geese', 'times', 'revelations'.
NP0 proper noun, e.g. 'London', 'Michael', 'Mars', 'IBM'. Note that no distinction is made for number in the case of proper
nouns, since plural proper names are a comparative rarity.
ORD ordinal numeral, e.g. 'first', 'sixth', '77th', 'next', 'last'. No distinction is made between ordinals used in nominal and
adverbial roles. 'next' and 'last' are included in this category, as 'general ordinals'.
PNI indefinite pronoun, e.g. 'none', 'everything', 'one' (pronoun), 'nobody'. This tag is applied to words which always
function as heads of noun phrases. Words like 'some' and 'these', which can also occur before a noun head in an
article-like function, are tagged as determiners, DT0 or AT0.
PNP personal pronoun, e.g. 'I', 'you', 'them', 'ours'. Note that possessive pronouns such as 'ours' and 'theirs' are included in
this category.
PNQ wh-pronoun, e.g. 'who', 'whoever', 'whom'. The same tag is used whether the word is used interrogatively or to
introduce a relative clause.
PNX reflexive pronoun, e.g. 'myself', 'yourself', 'itself', 'ourselves'.

continued on next page

page_231

Page 232

Code Usage
POS the possessive or genitive marker ''s' or ' ' '. Note that this marker is tagged as a distinct L-word. For example, "Kids'
work or someone else's" is tagged <lt>w NP0>Kids<w POS>' <w NN1>work <w CJC>or
<w PNI>someone <w AV0>else<w POS>'s</CE>
PRF the preposition 'of'. This word has a special tag of its own, because of its high frequency and its almost exclusively
postnominal function.
PRP preposition other than 'of', e.g. 'about', 'at', 'in', 'on behalf of', 'with'. Note that prepositional phrases like 'on behalf of'
or 'in spite of' are treated as single words.
PUL left bracket (i.e. (or [ )
PUN any mark of separation (. ! , : ; - ? )
PUQ quotation mark (' ' " ")
PUR right bracket (i.e.) or])
TO0 the infinitive marker 'to'.
UNC unclassified items which are not appropriately classified as items of the English lexicon. Examples include foreign
(non-English) words; special typographical symbols; formulae; hesitation fillers such as 'errm' in spoken language.
VBB the present tense forms of the verb 'be', except for 'is' or ''s', i.e. 'am', ''m', 'are', ''re', 'be' (subjuntive or imperative), 'ai' (as
in 'ain't').
VBD the past tense forms of the verb 'be', i.e. 'was', 'were'.
VBG the -ing form of the verb 'be', i.e. 'being'.
VBI the infinitive form of the verb 'be', i.e. 'be'.
VBN the past participle form of the verb 'be', i.e. 'been'.
VBZ the -s form of the verb 'be', i.e. 'is', ''s'.
VDB the finite base form of the verb 'do', i.e. 'do'.
VDD the past tense form of the verb 'do', i.e. 'did'.
VDG the -ing form of the verb 'do', i.e. 'doing'.
VDI the infinitive form of the verb 'do', i.e. 'do'.
VDN the past participle form of the verb 'do', i.e. 'done'.
VDZ the -s form of the verb 'do', i.e. 'does'.
VHB the finite base form of the verb 'have', i.e. 'have', ''ve'.
VHD the past tense form of the verb 'have', i.e. 'had', ''d'.
VHG the -ing form of the verb 'have', i.e. 'having'.
VHI the infinitive form of the verb 'have', i.e. 'have'.

continued on next page

page_232

Page 233

Code Usage
VHN the past participle form of the verb 'have', i.e. 'had'.
VHZ the -s form of the verb 'have', i.e. 'has', ''s'.
VM0 modal auxiliary verb, e.g, 'can', 'could', 'will', ''ll', ''d', 'wo' (as in 'won't').
VVB the finite base form of lexical verbs, e.g. 'forget', 'send', 'live', 'return'. This tag is used for imperatives and the present
subjunctive forms, but not for the infinitive (VVI).
VVD the past tense form of lexical verbs, e.g. 'forgot', 'sent', 'lived', 'returned'.
VVG the -ing form of lexical verbs, e.g. 'forgetting', 'sending', 'living', 'returning'.
VVI the infinitive form of lexical verbs, e.g. 'forget', 'send', 'live', 'return'.
VVN the past participle form of lexical verbs, e.g. 'forgotten', 'sent', 'lived', 'returned'.
VVZ the -s form of lexical verbs, e.g. 'forgets', 'sends', 'lives', 'returns'.
XX0 the negative particle 'not' or 'n't'.
ZZ0 alphabetical symbols, e.g. 'A', 'a', 'B', 'b', 'c', 'd'.

The following portmanteau codes are used to indicate where the CLAWS system has indicated an uncertainty between two possible analyses:
Code Usage
AJ0-AV0 adjective or adverb
AJ0-NN1 adjective or singular common noun
AJ0-VVD adjective or past tense verb
AJ0-VVG adjective or -ing form of verb
AJ0-VVN adjective or past participle
AVP-PRP adverb particle or preposition
AVQ-CJS wh-adverb or subordinating conjunction
CJS-PRP subordinating conjunction or preposition
CJT-DT0 'that' as conjunction or determiner
CRD-PNI 'one' as number or pronoun
NN1-NP0 singular common noun or proper noun
NN1-VVB singular common noun or base verb form
NN1-VVG singular common noun or -ing form of verb
NN2-VVZ plural noun or -s form of lexical verb

continued on next page

 page_233

Page 234

Code Usage
VVD-VVN past tense verb or past participle

2.2
Text Classification Codes
We list here all classification codes that may appear within the <CATREF> element in the header of each text. Not all of the values defined here are
actually used within the BNC. Furthermore, not all of them are searchable using SARA, although they may appear in solutions to SGML queries. A list
of the searchable codes is given in section 5.2.3 on page 103.
The following table shows the codes which can be used to classify all kinds of text, according to their availability, or their type.
Code Usage
allAva Text availability
allAva1 free, world: Freely available worldwide
allAva2 restricted, world: Available worldwide
allAva3 restricted, Not-NA: Not available in North America
allAva4 restricted, Not-US: Not available in U.S.A.
allAva5 restricted, EU: Not available outside the European Union
allAva6 restricted, Not-USP: Not available in U.S.A. & Philippines
allAva7 restricted, Not-NAP: Not available in North America & Philippines
allTyp Text type
allTyp1 Spoken demographic
allTyp2 Spoken context-governed
allTyp3 Written books and periodicals
allTyp4 Written-to-be-spoken
allTyp5 Written miscellaneous

The following table lists the classification codes which can be specified for spoken texts (either demographic or context-governed) only. Note that the
classifications for demographically sampled texts apply to the respondent only, not necessarily to all speakers transcribed.
Code Usage
scgDom Domain for context-governed material
scgDom1 Educational
scgDom2 Business
scgDom3 Institutional

continued on next page

page_234

Page 235

Code Usage
scgDom4 Leisure
sdeAge Age band for demographic respondent
sdeAge1 014
sdeAge2 1524
sdeAge3 2534
sdeAge4 3544
sdeAge5 4559
sdeAge6 60+
adeCla Social class for demographic repondent
sdeCla1 AB
sdeCla2 C1
sdeCla3 C2
sdeCla4 DE
sdeSex Sex of demographic respondent
sdeSex1 Male
sdeSex2 Female
spoLog Interaction type
spoLog1 Monologue
spoLog2 Dialogue
spoReg Region where text captured
spoReg1 South
spoReg2 Midlands
spoReg3 North

The following table lists all classification codes which may be specified for any written text.
Code Usage
wbpSel Books & periodicals: selection method
wbpSel1 Selective
wbpSel2 Random
wmiPub Miscellaneous materials: publication status
wmiPub1 Published
wmiPub2 Unpublished
wriAAg Author age band
wriAAg1 014
wriAAg2 1524
wriAAg3 2534

continued on next page

page_235

Page 236

Code Usage
wriAAg4 3544
wriAAg5 4559
wriAAg6 60+
wriADo Author domicile
wriAD036 Australia
wriAD124 Canada
wriAD250 France
wriAD276 Germany
wriAD372 Ireland
wriAD380 Italy
wriAD422 Lebanon
wriAD492 Monaco
wriAD554 New Zealand
wriAD620 Portugal
wriAD702 Singapore
wriAD756 Switzerland
wriAD826 United Kingdom
wriAD840 United States
wriAD920 UK North (north of Mersey-Humber line)
wriAD921 UK Midlands (north of Bristol Channel-Wash line)
wriAD922 UK South (south of Bristol Channel-Wash line)
wriASe Sex of author
wriASe1 Male
wriASe2 Female
wriASe3 Mixed
wriASe4 Unknown
wriATy Type of author
wriATy1 Corporate
wriATy2 Multiple
wriATy3 Sole
wriATy4 Unknown
wriAud Intended age of audience
wriAudl Child
wriAud2 Teenager
wriAud3 Adult
wriAud4 Any
wriDom Domain
wriDom1 Imaginative

continued on next page

page_236

Page 237

Code Usage
wriDom2 Informative: natural & pure science
wriDom3 Informative: applied science
wriDom4 Informative: social science
wriDom5 Informative: world affairs
wriDom6 Informative: commerce & finance
wriDom7 Informative: arts
wriDom8 Informative: belief & thought
wriDom9 Informative: leisure
wriLev Circulation level
wriLev1 Low
wriLev2 Medium
wriLev3 High
wriMed Medium
wriMed1 Book
wriMed2 Periodical
wriMed3 Miscellaneous published
wriMed4 Miscellaneous unpublished
wriMed5 To-be-spoken
wriPP1 Place of publication
wriPP372 Ireland
wriPP826 United Kingdom
wriPP840 United States
wriPP920 UK North (north of Mersey-Humber line)
wriPP921 UK Midlands (north of Bristol Channel-Wash line)
wriPP922 UK South (south of Bristol Channel-Wash line)
wriSam Type of sample
wriSam1 Whole text
wriSam2 Beginning sample
wriSam3 Middle sample
wriSam4 End sample
wriSam5 Composite
wriSta Reception status
wriSta1 Low
wriSta2 Medium
wriSta3 High
wriTas Target audience sex
wriTas1 Male
wriTas2 Female

continued on next page

page_237

Page 238

Code Usage
wriTas3 Mixed
wriTas4 Unknown
wriTim Time period
wriTim1 19601974
wriTim2 19751993

For all the classifications listed above, the absence of of information may be indicated either by the absence of any code, or by the presence of a code
ending with a zero instead of a number. For example, written texts for which type of author is unknown may be indicated either by the absence of any
value beginning wriAty or by the presence of the specific value wriAty0.
2.3
Dialect Codes
A single set of codes, derived from the International Standard for language and country identification, is used to identify regional origins, first language,
and dialects spoken by participants, as specified in the <PERSON> element in the text header. Speakers for whom such information was recorded will use
one or more of the following codes as values for the WHO.FLANG or WHO.DIALECT attributes.
Code Usage
CAN Canada
CHN China
DEU Germany
FRA France
GBR United Kingdom
IND India
IRL Ireland
USA United States
XXX Unknown
ZZG Europe
XDE accent: German
XEA accent: East Anglia
XFR accent: French
XHC accent: Home Counties
XHM accent: Humberside
XIR accent: Irish
XIS accent: Indian subcontinent
XLC accent: Lancashire

continued on next page

page_238

Page 239

Code Usage
XLO accent: London
XMC accent: central Midlands
XMD accent: Merseyside
XME accent: north-east Midlands
XMI accent: Midlands
XMS accent: south Midlands
XMW accent: north-west Midlands
XNC accent: central northern England
XNE accent: north-east England
XNO accent: northern England
XOT accent: unidentifiable
XSD accent: Scottish
XSL accent: lower south-west England
XSS accent: central south-west England
XSU accent: upper south-west England
XUR accent: European
XUS accent: U.S.A
XWA accent: Welsh
XWE accent: West Indian

2.4
Other Codes Used
Chapter nine of the BNC Users' Reference Guide includes exhaustive tables for the following:
all SGML elements used in the corpus, with a brief description of each;
all SGML entity references used in the corpus, with a brief description of each;
all values used in the corpus for the TYPE attribute on division elements (<DIV1>, <DIV2> etc.);
all values used in the corpus for the R (rendition) attribute, chiefly on <HI> elements, to indicate typographic rendering of the source;
all values used in the corpus for the NEW attribute on the <SHIFT> element, to indicate changes in voice quality for spoken texts;
codes used to identify relationships documented between participants, as specified in the <RELATION> element in the text header;
all non-orthographic words ('L-words') recognized by the CLAWS system in the current version of the corpus.

page_239

Page 240

3
SGML Listing Format
This section defines the format in which the LISTING command on the QUERY menu saves its results. The files produced are valid SGML documents,
conforming to a simple document type description specified, and distributed with the SARA client. The purpose of this format is to preserve contextual
information about the source of each solution in the Listing file, and to make possible subsequent re-processing of the results file by software which may
or may not be SGML-aware.
Each file begins with the following DOCTYPE statement:
<!DOCTYPE bncXtract PUBLIC
"-//BNC//DTD BNC extract 1.0//EN">

Any SGML tags present in the solutions are retained but made invisible by changing their delimiters from < and > to [and]
Each file contains a single <BNCXTRACT> element, conforming to the following document type definition:
<!ELEMENT bncXtract - - (hdr,hit+)>
<!ATTLIST hdr date CDATA #required
user CDATA #REQUIRED
server CDATA #REQUIRED
format (plain|tagged) plain>
<!ELEMENT hdr - - (source,query,note?)>
<!ELEMENT (source|note) - o (#PCDATA)>
<!ELEMENT query - - (CDATA)>
<!ELEMENT hit - - (left, focus, right)>
<!ATTLIST hit text CDATA #REQUIRED
n NUMBER #REQUIRED>
<!ELEMENT (left|focus|right) - o (#PCDATA)>

A more detailed description of the elements used by this DTD, specifying their possible contents and attributes follows. Elements are described in the
order in which they appear within a <BNCXTRACT>.
<HDR>
contains only <SOURCE>, <QUERY>, and <NOTE> elements, which must be supplied in that order. The first two are mandatory; the last can be
omitted if no comment is attached to the query. The following attributes must be supplied:
DATE
date the file was saved, in form DD-MMM-YYYY HH:MM:SS;

page_240

Page 241

USER
SARA user name under which the query was performed;
SERVER
IP address (name or number) of the server from which the client obtained this set of solutions;
FORMAT
value either tagged (solutions were displayed in POS, Custom, or SGML format) or plain (solutions were displayed in Plain format).
<SOURCE>
contains the following (fixed) text:
This data is extracted from the British National Corpus. All rights in the texts cited are reserved. This data
may not be reproduced or redistributed in any form, other than as provided for by the Fair Use provisions of
the Copyright Act.

<QUERY>
text of the CQL query as displayed by SARA (including any thinning information). Enclosed within a CDATA marked section.
<NOTE>
any user-supplied annotation may appear here. The element may be omitted if no annotation was supplied.
<HIT>
a single solution from the solution set. It always contains three elements <LEFT>, <FOCUS> and <RIGHT>, described below, and always bears the following
attributes:
TEXT
three-character identifier of the text in which this solution appears;
N
sequence number of the <S> element within which the query focus of this solution begins, taken from its N attribute.
<LEFT>
the left context for the solution, i.e. everything preceding the query focus.
<FOCUS>
the focus of the solution, i.e. the part which is highlighted in Line display mode.
<RIGHT>
the right context for the solution, i.e. everything following the query focus.
For an example Listing file, see 1.6.6 on page 222.

page_241

Page 242

4
Bibliography
Aarts, J. 1991. 'Intuition-based and observation-based grammars' in Aijmer and Altenberg 1991. 4462.
Aarts, J., de Haan, P. and Oostdijk, N. (eds.) 1993. English language corpora: design, analysis and exploitation. Amsterdam: Rodopi. 5771.
Aijmer, K. and Altenberg, B. (eds.) 1991. English corpus linguistics. Harlow: Longman.
Aijmer, K., Altenberg, B. and Johansson, M. (eds.) 1996. Languages in contrast. Lund: Lund University Press.
Alatis, J. (ed.) 1991. Georgetown round table on language and linguistics. Lingustics and language pedagogy: the state of the art . Washington
DC: Georgetown University Press.
Altenberg, B. 1991. 'A bibliography of publications relating to English computer corpora' in Johansson and Stenström 1991. 355396.
Altenberg, B. 19901994. ICAME Bibliography, parts 1 to 3. Bergen: Computing Centre for the Humanities (files available from
ftp://nora.hd.uib.no/ICAME)
Anderson, A.H., Badger, M., Bard, E.G., Boyle, E., Doherty, G., Garrod, S., Isard, S., Kowtko, J., McAllister, J., Miller, J., Sotillo, C., Thompson,
H.S. and Weinert, R. 1991. 'The HCRC map task corpus' Language and speech 34: 351366.
Armstrong, S. (ed.) 1994. Using large corpora. Cambridge MA: MIT Press.
Armstrong-Warwick, S., Thompson, S., McKelvie, D. and Petitpierre, D. 1994. 'Data in your language: the ECI multilingual corpus 1' in Matsumoto
1994.
Aston, G. 1995. 'Corpora in language pedagogy: matching theory and practice' in Cook and Seidlhofer 1995. 257270.
Aston, G. in press. 'Enriching the learning environment: corpora as resources for ELT' in Wichmann et al in press.
Atkins, S., Clear, J. and Ostler, N. 1992. 'Corpus design criteria' Literary and linguistic computing 7: 116.
Atwell, E. 1996. 'Machine learning from corpus resources for speech and handwriting recognition' in Thomas and Short 1996. 151166.
Baker, M., Francis, G. and Tognini-Bonelli, E. (eds.) 1993. Text and technology: in honour of John Sinclair. Amsterdam: Benjamins.
Bauer, L. 1993. 'Progress with a corpus of New Zealand English and some early results' in Souter and Atwell 1993. 110.
Bell, R. 1991. The language of news media. Oxford: Blackwell.

page_242

Page 243

Biber, D. 1988. Variation across speech and writing. Cambridge: Cambridge University Press.
Biber, D. 1993. 'Using register-diversified corpora for general language studies' in Armstrong 1994. 219241.
Biber, D. and Finegan, E. 1989. 'Styles of stance in English: lexical and grammatical marking of evidentiality and affect' Text 9: 93124.
Biber, D. and Finegan, E. 1991. 'On the exploration of computerized corpora in variation studies' in Aijmer and Altenberg 1991. 204220.
Biber, D. and Finegan, E. 1994. 'Intratextual variation within medical research articles' in Oostdijk and De Haan 1994. 201221.
Biber, D., Conrad, S. and Reppen, R. 1994. 'Corpus-based approaches to issues in applied linguistics' Applied linguistics 15: 169189.
Biber, D., Conrad, S. and Reppen, R. 1996. 'Corpus-based investigations of language use' Annual review of applied linguistics 16: 115136.
Biber, D., Finegan, E. and Atkinson, D. 1993. 'ARCHER and its challenges: compiling and exploring a representative corpus of historical English
registers' in Fries et al. 1993. 113.
Blount, B. and Sanches, M. (eds.) 1975. Sociocultural dimensions of language use. New York: Academic Press.
Bolinger, D. 1976. 'Meaning and memory' Forum linguisticum 1: 114.
Burnard, L. (ed.) 1995. Users' reference guide to the British National Corpus. Oxford: Oxford University Computing Services.
Butler, C. (ed.) 1992. Computers and written texts. Oxford: Blackwell.
Chafe, W., Du Bois, J.W. and Thompson, S.A. 1991. 'Towards a new corpus of spoken American English' in Aijmer and Altenberg 1991. 6482.
Chomsky, N. 1965. Aspects of the theory of syntax . Cambridge MA: MIT Press.
Church, K.W. and Hanks, P. 1990. 'Word collocation norms, mutual information, and lexicography' Computational linguistics 16: 2229.
Church, K.W. and Mercier, R.L. 1993. 'Introduction to the special issue on computational linguistics using large corpora' in Armstrong 1994. 122.
Cobuild 1995. COBUILD collocations on CD-ROM. London: HarperCollins.
Collins, P.C. and Peters, P. 1988. 'The Australian corpus project' in Kytö et al 1988. 103121.
Cook, G. 1995. 'Theoretical issues: transcribing the untranscribable' in Leech et al. 1995. 3553.
Cook, G. and Seidlhofer, B. (eds.) 1995. Principle and practice in applied linguistics. Oxford: Oxford University Press.

page_243

Page 244

Coulthard, M. 1993. 'On beginning the study of forensic texts: corpus concordance collocation' in Hoey 1993. 8697.
Coulthard, M. (ed.) 1994. Advances in written text analysis. London: Routledge.
Crowdy, S. 1995. 'The BNC spoken corpus' in Leech et al. 1995. 224235.
Davison, R. 1992. 'Building a million-word computer science corpus in Hong Kong' ICAME journal 16: 123124.
Dunning, T. 1993. 'Accurate methods for the statistics of surprise and coincidence' in Armstrong 1994. 6174.
Edwards, J. 1993. 'Survey of electronic corpora and related resources for language researchers' in Edwards and Lampert 1993. 263310.
Edwards, J. 1995. 'Principles and alternative systems in the transcription, coding and mark-up of spoken discourse' in Leech et al. 1995. 1934.
Edwards, J. and Lampert, M.D. (eds.) 1993. Talking data: transcription and coding in discourse research. Hillsdale NJ: Erlbaum.
Ellegård, A. 1978. The syntactic structure of English texts: a computer-based study of four kinds of text in the Brown University Corpus.
Gothenberg Studies in English 43. Gothenberg: Acta Universitatis Gothoburgensis.
Fillmore, C.J. 1992. 'Corpus linguistics or computer-aided armchair linguistics' in Svartvik 1992. 3560.
Firth, J.R. 1957. Papers in linguistics 19341951. London: Oxford University Press.
Fligelstone, S. 1993. 'Some reflections on the question of teaching, from a corpus linguistic perspective' ICAME journal 17: 97109.
Fligelstone, S. 1992. 'Developing a scheme for annotating text to show anaphoric relations' in Leitner 1992. 153170.
Francis, G. 1993. 'A corpus-driven approach to grammar. Principles, methods and examples' in Baker et al. 1993. 137156.
French, J.P. 1992. 'Transcription proposals: multi-level system' NERC-WP4-50. Working paper for NERC.
Fries, U., Tottie, G. and Schneider, P. (eds.) 1993. Creating and using English language corpora. Amsterdam: Rodopi.
Garside, R. 1987. 'The CLAWS word-tagging system' in Garside et al. 1987. 3041.
Garside, R, 1993. 'The marking of cohesive relationships: tools for the construction of a large bank of anaphoric data' ICAME journal 17: 527.
Garside, R. 1996. 'The robust tagging of unrestricted text: the BNC experience' in Thomas and Short 1996. 167180.

page_244

Page 245

Garside, R., Leech, G. and Sampson, G. (eds.) 1987. The computational analysis of English: a corpus-based approach. Harlow: Longman.
Goldfarb, C. 1990. The SGML handbook. Oxford: Clarendon Press.
Goffman, E. 1979. 'Footing' in Goffman 1981. 124159.
Goffman, E. 1981. Forms of talk . Oxford: Blackwell.
Granger, S. 1993. 'International corpus of learner English' in Aarts et al. 1993. 5771.
Grabowski, E. and Mindt, D. 1995. 'A corpus-based learning list of irregular verbs in English' ICAME journal 19: 522.
Greenbaum, S. 1991. 'The development of the International Corpus of English' in Aijmer and Altenberg 1991. 8391.
Greenbaum, S. 1992. 'A new corpus of English: ICE' in Svartvik 1992. 171179.
Guthrie, L., Guthrie, J. and Cowie, J. 1994. 'Resolving lexical ambiguity' in Oostdijk and De Haan 1994. 7993.
Halliday, M.A.K. 1992. 'Language as system and language as instance: the corpus as a theoretical construct' in Svartvik 1992. 6177.
Hoey, M. (ed.) 1993. Data, description, discourse: papers on the English language in honour of John McH. Sinclair. London: HarperCollins.
Hofland, K. and Johansson, S. 1982. Word frequencies in British and American English. Bergen: Norwegian Computing Centre for the
Humanities/London: Longman.
Holmes, J. 1994. 'Inferring language change from computer corpora: some methodological problems' ICAME journal 18: 2740.
Johansson, S. 1980. 'The LOB corpus of British English texts: presentation and comments' ALLC journal 1: 2536.
Johansson, S., Atwell, E., Garside, R. and Leech, G. 1986. The tagged LOB corpus: user's manual. Bergen: Norwegian Computing Centre for the
Humanities.
Johansson, S. and Ebeling, J. 1996. 'Exploring the English-Norwegian parallel corpus' to appear in the Proceedings of the Sixteenth ICAME
Conference, Toronto, May 1995.
Johansson, S and Hofland, K. 1989. Frequency analysis of English vocabulary and grammar based on the LOB corpus. Oxford: Clarendon
Press.
Johansson, S. and Hofland, K. 1993. 'Towards an English-Norwegian parallel corpus' in Fries et al. 1993. 2537.

page_245

Page 246

Johansson, S. and Stenström, A.-B. (eds.) 1991. English computer corpora: selected papers and research guide. Berlin: Mouton de Gruyter.
Jones, S. and Sinclair, J.McH. 1974. 'English lexical collocations' Cahiers de lexicologie 24: 1561.
Karlsson, F. 1994. 'Robust parsing of unconstrained text' in Oostdijk and De Haan 1994. 122142.
Kennedy, G. 1992. 'Preferred ways of putting things with implications for language teaching' in Svartvik 1992. 335373.
Kjellmer, G. 1991. 'A mint of phrases' in Aijmer and Altenberg 1991. 111127.
Knowles, G. 1996. 'Corpora, databases and the organization of linguistic data' in Thomas and Short 1996. 3653.
Kucera, H. and Francis, W.N. 1967. Computational analysis of present-day American English. Providence RI: Brown University Press.
Kytö, M. 1993. Manual to the diachronic part of the Helsinki corpus of English texts . Helsinki: University of Helsinki, Department of English.
Kytö, M., Ihalainen, O. and Rissanen, M. (eds.) 1988. Corpus linguistics hard and soft. Amsterdam: Rodopi.
Kytö, M. and Rissanen, M. 1996. 'English historical corpora: report on developments in 1995' ICAME journal 20: 11732.
Lakoff, R. 1975. Language and woman's place. New York: Harper and Row.
Leech, G. in press. 'Teaching and language corpora: a convergence' in Wichmann et al. in press
Leech, G. and Falton, R. 1992. 'Computer corpora what do they tell us about culture?' ICAME journal 16: 2950.
Leech, G. and Fligelstone, S. 1992.'Computers and corpus analysis' in Butler 1992. 11541.
Leech, G. and Garside, R. 1991. 'Running a grammar factory: the production of syntactically analysed corpora or treebanks' in Johansson and
Stenström 1991. 1532.
Leech, G., Garside, R. and Bryant, M. 1994. 'The large-scale grammatical tagging of text: experience with the British National Corpus' in Oostdijk and
De Haan 1994. 4763.
Leech, G., Myers, G. and Thomas, J. (eds.) 1995. Spoken English on computer: transcription, mark-up and application. Harlow: Longman.
Leitner, G. (ed) 1992. New directions in English language corpora. Berlin: Mouton de Gruyter.

page_246

Page 247

Ljung, H. 1991. 'Swedish TEFL meets reality' in Johansson and Stenström 1991. 245256.
Louw, B. 1993. 'Irony in the text or insincerity in the writer? The diagnostic potential of semantic prosodies' in Baker et al. 1993. 157176.
Mair, C. 1993. 'Is see becoming a conjunction? The study of grammaticalisation as a meeting ground for corpus linguistics and grammatical theory' in
Fries et al. 1993. 127137.
McEnery, A. and Wilson, A. 1996. Corpus linguistics. Edinburgh: Edinburgh University Press.
Matsumoto, Y. (ed). 1994. International workshop on shareable natural language resources. (Proceedings of the 15th International Conference
on Computational Linguistics.) Nara: Institute of Science of Technology.
Miller, G. 1990. 'Wordnet: an on-line lexical database' International journal of lexicography 3: 235312.
Mindt, D. 1996. 'English corpus linguistics and the foreign language teaching syllabus' in Thomas and Short 1996. 232247.
Murison-Bowie, S. 1996. 'Linguistic corpora and language teaching' Annual review of applied linguistics 16: 182199.
O'Donoghue, T. 1991. 'Taking a parsed corpus to the cleaners: the EPOW corpus' ICAME journal 15: 5562.
Oostdijk, N. and De Haan, P. (eds.) 1994. Corpus-based research into language. Amsterdam: Rodopi.
Quirk, R. 1974. The linguist and the English language. London: Arnold.
Quirk, R., Greenbaum, S., Leech, G. and Svartvik, J. 1985. A comprehensive grammar of the English language. Harlow: Longman.
Rundell, M. 1995. 'The word on the street' English today 43: 2935.
Sacks, H. 1975. 'Everyone has to lie' in Blount and Sanches 1975. 5780.
Sacks, H. 1992. Lectures on conversation. Oxford: Blackwell.
Sampson, G. 1994. English for the computer. Oxford: Oxford University Press.
Sampson, G. 1996. 'From central embedding to corpus linguistics' in Thomas and Short 1996. 1426.
Shastri, S.V. 1988. 'The Kolhapur corpus of Indian English and work done on its basis so far' ICAME journal 12: 1526.
Sinclair, J.McH. (ed.) 1987. Looking up. London: Collins.
Sinclair, J.McH. 1991. Corpus, concordance, collocation. Oxford: Oxford University Press.

page_247
 Page 248

Sinclair, J.McH. 1992. 'The automatic analysis of corpora' in Svartvik 1992. 379397.
Sinclair, J.McH. 1996. 'Preliminary recommendations on corpus typology' EAGLES Document TCWG-CTYP/P (available from http:
//www.ilc.pi.cnr.it/ EAGLES /corpustyp /corpustyp.html).
Souter, C. and Atwell, E. (eds.) 1993. Corpus-based computational linguistics. Amsterdam: Rodopi.
Sperberg-McQueen, C.M. and Burnard, L. 1994. Guidelines for electronic text encoding and interchange (TEI P3). Chicago and Oxford: Text
Encoding Initiative.
Stenström, A.-B. 1991. 'Expletives in the London-Lund corpus' in Aijmer and Altenberg 1991. 230253.
Stenström, A.-B. and Breivik, L.E. 1993. 'The Bergen corpus of London teenager language (COLT)' ICAME journal 17: 128.
Stubbs, M. 1995. 'Collocations and semantic profiles: on the cause of the trouble with quantitative studies' Functions of language 2: 2355.
Stubbs, M. 1996. Text and corpus analysis. Oxford: Blackwell.
Svartvik, J. (ed.) 1990. The London-Lund corpus of spoken English. Lund: Lund University Press.
Svartvik, J. (ed.) 1992. Directions in corpus linguistics. Berlin: Mouton de Gruyter.
Svartvik, J. and Quirk, R. 1980. A corpus of English conversation. Lund: Gleerup.
Thomas, J. and Short, M. (eds.) 1996. Using corpora for language research: studies in the honour of Geoffrey Leech. Harlow: Longman.
van Halteren, H. and Oostdijk, N. 1993. 'Towards a syntactic database: the TOSCA analysis system' in Aarts et al. 1993. 14561.
Wichmann, A., Fligelstone, S., McEnery, A. and Knowles, G. (eds.) in press. Teaching and language corpora. Harlow: Addison-Wesley-Longman.
Widdowson, H.G. 1991. 'The description and prescription of language' in Alatis 1991. 1124.
Willis, D. 1990. The lexical syllabus. Glasgow: Collins.
Zipf, G. 1935. The psychobiology of language. Boston: Houghton-Mifflin.

page_248

Page 249

INDEX
NOTE In the following index, italics are used for technical terms, and typewriter font for SGML element names. Attribute names are given in
lower case, while Commands, menu options, etc. in the SARA program are capitalised.
A
About SARA, 49, 229
Add, 102, 105, 107, 108, 118, 119, 124, 134, 136, 171, 181, 182, 207
Align, 68, 79, 225
aligned, 16
All, 67
all_availability, 104
all_type, 104, 110, 141
Alt, 50, 60
AND, 91, 92
Annotation, 56, 64, 67, 68, 71, 75, 87, 99, 115, 131, 146, 162, 180, 216, 221
Any, 92, 93, 176
Anyword, 86, 89, 92, 93, 95, 210
anyword, 90, 203
archives, 5
Arrange icons, 228
Arrow, 59
Ascending, 79, 217
ASCII, 79, 81, 179, 181, 193, 217
atomic queries, 210
attribute, 34, 98, 114, 207
Attribute, 102
author, 126
Automatic, 55, 219
B
backslash, 206
balanced, 5
Bibliographic data, 53, 223
BNC, 49, 50
BNC document, 39, 53, 91
<bncDoc>, 53, 91, 106, 112, 115, 126, 127, 134, 152, 158, 171, 179, 182, 183, 208, 215
<bncXtract>, 240
Bold, 227
bookmark, 213, 214
Bookmark, 179, 181, 191, 213, 214
Bookmark name, 181, 214
bracketed sequence, 205
browse mode, 53, 60, 197, 214
Browse, 48, 53, 54, 117, 215, 224
Browser, 53, 56, 64, 75, 87, 99, 115, 131, 146, 162, 180, 198, 216
Busy, 51, 58, 212
C
<c>, 34, 53, 56, 113, 215, 216, 230
Calculate, 76-78, 82, 153, 224
Cap, 226
<caption>, 134
caret, 205
Cascade, 188, 228
Case, 211
case-sensitive, 72

page_249

Page 250

category reference, 98
<catRef>, 39, 98-106, 108-112, 118, 125-127, 130, 135-137, 141, 157, 158, 207, 234
Centre, 79, 90, 94, 96, 138, 172, 217
chi-square, 40
Clear, 66, 92, 109, 202, 208
client, 49
Close, 71, 78, 80, 81, 154, 199, 214
Close all, 60, 189
collating, 217
Collating, 79
collections, 5
colligate, 8
colligation, 14, 81, 139
collocate, 8, 86, 224
Collocate, 76, 77, 153, 224
Collocation, 75, 76, 78, 80, 82, 83, 95, 96, 153, 178, 179, 190, 192, 216, 224
collocation frequency, 13, 76
collocation ratio, 77
Colour, 227
colour scheme, 227
Colours, 55, 56, 227
Comms, 56, 228
Communications, 49, 50, 228
complex transitive, 74
concordance, 7
Concordance, 48, 52, 56, 57, 59, 60, 64, 68-70, 75, 87, 99, 115, 116, 124, 131, 146, 162, 168, 169, 176, 180, 181, 188, 216, 218, 225
Configure, 103, 116, 155, 166, 219, 220
content, 34
content nodes, 91, 92, 208
context, 7
Context Help, 50, 226
context-governed, 31, 32, 98, 113
contrastive, 13
copular, 74
Copy, 489, 54, 66, 92, 130, 131, 139, 149, 151, 155, 202, 213, 214, 225
corpus, 2
Corpus Data Interchange Format, 33
corpus header, 33
Corpus Query Language, 67, 210
CQL, 56, 67, 137, 200, 210
CQL Query, 126, 130, 135, 137, 139, 177, 185, 191, 210, 225
Ctrl, 60, 65, 66, 124, 132, 133, 139, 170, 201
current solution, 213
Custom, 52, 55, 103, 114, 116, 124, 128, 155, 162, 165, 166, 168, 174, 175, 219, 220
D
date, 240
Delete, 88, 92, 182, 208
demographic, 31, 113
desc, 36, 166, 171, 221
Descending, 79, 217
descriptive features, 30
diachronic corpora, 16
dialect-specific, 5
dialogue, 32
dialogue box, 47
Direction, 79
disjunction, 92, 206, 211
dispersion, 58, 60, 67
<div>, 113, 126, 178
</div>, 178

page_250

Page 251

<div1>, 54, 113, 126, 166, 221, 239

<div2>, 113, 126, 166, 239
<div3>, 113
document type definition, 34
domain, 29
dot, 205
double-click, 47
Download all, 67, 100, 107, 116, 152, 169, 171, 212
Download hits, 66, 67, 82, 106, 135, 165
dur, 172
E
elements, 34, 112
encoded, 24
End, 177, 219
end-tag, 34, 113, 206
entity references, 34, 52, 168
Esc, 51, 212
<event>, 36, 172, 221
Exit, 50, 71, 199
external, 5
F
F1, 50, 226
F4, 50, 60
field, 23
File, 50, 51, 64, 68, 70, 71, 81, 91, 94, 101, 118-120, 132, 133, 137, 150, 154, 174, 175, 198, 200, 202-204, 206, 208, 210, 212-214, 225
File Open, 71
First solution, 225
focus, 7
<focus>, 155, 223, 241
Font, 55, 56, 227
foreign language teaching, 13
Format, 64, 75, 87, 99, 103, 115, 131, 146, 162, 166, 180, 241
frequency counts, 7
G
<gap>, 36, 166, 172
general noun, 187
Goto, 187-189, 213, 214

H
<hdr>, 222, 240
<head>, 34, 54, 130, 134, 135, 141, 207, 211, 215
</head>, 34
header, 40
Header, 211
<header>, 98, 112, 113, 134, 215
Help, 50, 130, 199, 226
<hi>, 239
hit, 7
<hit>, 155, 222, 223, 241
Home, 219
hyphen, 205
I
iconify, 47
id, 179, 182
idiom principle, 14
Ignore accents, 79, 217
Ignore case, 51, 61, 79, 81, 110, 142, 154, 184, 185, 187, 193, 202, 217
independent, 41
Index, 229
Initial, 66, 67, 165
Insert, 68, 94, 132, 133, 139, 150, 151
Italic, 227
<item>, 134

J
Join, 211
K
keys, 79, 217

page_251

Page 252

keywords, 40
KWIC, 7
L
Landscape, 81
Last solution, 226
<left>, 155, 223, 241
Left, 79, 90, 154, 169, 217
lemmatization, 25
line, 52
Line, 166, 168, 218
linguistic annotation, 25
link, 91
Link type, 93, 152, 164
listing, 70
Listing, 54, 146, 154, 155, 170, 213, 216, 222, 240
literal characters, 205
logging on, 196
Logon, 49, 196
Lookup, 64, 65, 76, 83, 92, 115, 121, 122, 131, 145, 147-149, 151, 163, 170, 185, 201
L-word, 35, 64, 132, 144, 201
M
markup, 9, 33
Max download length, 51, 55, 64, 75, 87, 99, 115, 117, 131, 146, 162, 168, 180, 219
Max downloads, 55, 64, 75, 87, 99, 101, 115, 131, 146, 162, 180
Maximum, 56, 117, 168, 220
medium, 29
menu bar, 196
Merge scheme, 227
mixed, 5
mode, 23
monitor corpus, 21
monologue, 32
mutual information, 83

N
n, 114, 212, 223, 241
natural language processing, 13, 18
new, 173, 174, 239
New query, 51, 64, 71, 91, 101, 118, 132, 133, 137, 150, 199, 200, 202-204, 206, 208, 210
New Window, 188, 228
Next, 93, 96, 106, 164, 169-171, 177, 209, 220
Next solution, 225
nodes, 91, 208
non-verbal, 24
normally distributed, 41
<note>, 240, 241
Num, 226
Number, 211
O
One per text, 67, 69, 72, 100, 124, 135, 158, 180, 190, 212, 218
One-way, 93, 94, 99, 106, 108, 109, 122, 126, 136, 152, 164, 209
Open, 71, 175, 199, 225
Open scheme, 227
open-choice, 14
Options, 56, 99, 103, 116, 124, 137, 158, 166, 173, 174, 185, 216, 219, 220
OR, 91, 92, 145, 150
Order, 79
orthographic transcription, 36
overlap, 172
P
<p>, 39, 53, 54, 113, 126, 134, 152, 166, 215, 220
page, 51, 218
Page, 166, 167

page_252

Page 253

Paragraph, 55, 220

paralinguistic, 24
parallel corpus, 16
parentheses, 206
parsing, 25
part-of-speech, 129, 132
part-of-speech tagging, 25
Password, 56, 228
Paste, 92, 208, 214
pattern, 145, 200, 204
Pattern, 66, 75, 83, 92, 115, 121, 131, 145-147, 150, 151, 163, 170, 200, 201, 204
Pattern Query, 145, 146, 149, 150, 191, 204, 225
<pause>, 36, 167, 172, 207, 221
<person>, 121, 127, 238
PgDn, 57, 60, 168, 169, 219
PgUp, 60, 117, 219
Phrase, 51, 92, 93, 106, 123, 136, 185, 200, 202
Phrase Query, 48, 51, 57, 58, 72, 82, 86-89, 91, 95, 96, 99, 100, 105, 164, 177, 179, 184, 186, 190, 191, 202, 225
Plain, 55, 103, 116, 155, 219
plain text, 24
plus, 205
<poem>, 134
port, 228
portmanteau code, 35, 132, 233
POS, 55, 79, 81, 92, 132, 133, 137, 155, 200, 203, 219
POS code, 129
POS code, 129, 130, 137-141, 189, 217
POS format, 129, 130
POS Query, 129-132, 140, 150, 156, 159, 180, 182, 184, 192, 203, 225
pragmatic annotation, 26
precision, 60, 87, 96, 152
Preferences, 48, 50, 54, 64, 66-68, 75, 87, 88, 99, 103, 115-117, 131, 146, 152, 162, 168, 180, 227
Previous solution, 225
Primary, 86, 217
Print, 54, 71, 75, 78, 81, 154, 169, 199, 212, 213, 224, 225
Print preview, 81, 199, 213
Print Setup, 71, 81, 213
probabilistic models, 18
production, 22
<ptr>, 167, 168, 172, 173

Q
<q>, 39
query, 49, 50, 70 197
<query>, 240, 241
Query Builder, 86, 91, 95, 96, 99, 105, 108, 110, 115, 120, 123, 126, 130, 133, 136, 141, 142, 145, 150, 151, 158, 159, 163, 169, 170, 182, 185,
186, 192, 200, 208, 225
query focus, 52, 217, 219, 220
query mode, 197
Query Text, 56, 67, 68, 71, 103, 185, 210, 216, 221
Query Text, 56, 67, 68, 71, 103, 185, 210, 216, 221
question mark, 205
R
r, 239
Random, 67, 69, 90, 135, 212, 218
recall, 60, 87, 96, 152
Recent File, 199
reception, 22

page_253

Page 254

register, 12
register-specific, 5
regular expressions, 147
<relation>, 239
Remove, 102, 207
Remove All, 106, 120, 207
replacement string, 166
representativeness, 63
Reset, 227
respondent, 109
Retry, 49
Reverse selection, 69-71, 89, 94, 138, 154, 173-175, 218
re-size, 47
<right>, 155, 223, 241
Right, 79, 96, 217
S
<s>, 34, 53, 56, 108, 109, 113, 114, 134, 152, 159, 164, 169, 170, 176, 177, 208, 215, 216, 220, 226, 241
<salute>, 134
Sample, 227
sample corpus, 21
Save, 68, 70, 94, 119, 120, 154, 174, 175, 199, 214, 225
Save As, 70, 120, 155, 175, 199, 214
Save scheme, 227
scope, 113, 208, 219
Scope, 55, 64, 75, 87, 99, 103, 115, 130, 131, 146, 162, 180
scope node, 91, 208
Scrl, 226
Search headers, 51, 202
Secondary, 86, 217
Selection, 69-71, 100, 141, 218
selection features, 29
semantic prosody, 14
Sentence, 55, 220
Sequence, 211
server, 49, 241
Setup, 81
SGML, 33
SGML, 55, 92, 101-103, 105, 108, 109, 116, 118-124, 126, 134, 136, 155, 158, 163, 164, 170, 171, 173, 174, 182, 183, 200, 206, 209, 211,
219
SGML elements, 206
SGML format, 99, 168
SGML Query, 99-101, 114, 118, 127, 142, 162, 174, 206, 225
SGML scope, 114
<shift>, 36, 173, 174, 239
Shift, 66, 68, 132, 133, 139, 150, 151, 201
Show header tags, 101, 116, 134, 136, 163, 207
Show tags, 53, 64, 75, 87, 99, 114, 115, 129, 131, 146, 162, 180
solutions, 49, 50, 197
Sort, 75, 78-81, 89, 94, 95, 100, 138, 149, 153, 154, 169, 172, 204, 216, 217
<source>, 240, 241
Source, 48, 52, 53, 57, 61, 73, 98-100, 114, 117, 124, 135, 176, 183, 215, 216, 223, 226
<sp>, 126, 134
span, 13
Span, 76, 77, 79, 96, 145, 146, 152, 153, 177, 209, 217, 224
special characters, 205
special corpora, 11

page_254

Page 255

spoken demographic, 28
spoken text, 113
spoken_age, 104, 109
spoken_class, 104
spoken_domain, 104
spoken_region, 104
spoken_sex, 104
spoken_type, 102, 104, 107, 109, 110
<stage>, 134,142
star, 205
Start, 207
start-tag, 34, 113, 206
Status bar, 50, 64, 226
<stext>, 113, 163, 169, 171, 176, 177, 215

T
t, 172
Tab, 49, 132, 133, 166, 180, 196, 203
tag, 25, 34, 53
Tags, 53, 216
tenor, 23
<text>, 112, 113, 125, 159, 177, 215
</text>, 113
text, 222, 241
text header, 33, 51, 72
text identifier, 52
Thin, 63, 68-71, 73, 89, 94, 96, 100, 133, 138, 141, 149, 154, 173-175, 190, 191, 203, 216, 218
thinning, 56
Tile, 71, 158, 188, 189, 228
time, 29
Too many solutions, 63, 66, 67, 69, 101, 107, 108, 119, 120, 212
Toolbar, 50, 64
<trunc>, 36, 167
Two-way, 93, 145, 146, 152, 157, 209
type, 34, 134, 135, 141, 239
t-test, 40

U
<u>, 36, 113, 114, 116, 118, 121-127, 134, 167, 168, 215, 220, 221
</u>, 113, 177
<unclear>, 36, 167, 172
under-represented, 60
upwards collocation, 77
Use downloaded hits only, 76, 82, 85, 224
user, 241
User preferences, 53, 54, 56, 218, 219, 227
Using Help, 229
V
value, 98, 114, 207
variable, 147
View, 48, 50, 52, 54-56, 64, 66-68, 75, 87, 88, 99, 101, 103, 115-117, 131, 135, 146, 152, 162, 180, 198, 227
<vocal>, 36, 167, 171-173
vocal event, 171
voice quality, 173
W
W
<w>, 34, 53, 56, 108, 113, 129, 215, 216, 219, 230
who, 36, 114, 121, 167, 221
who.age, 118, 124
who.dialect, 118, 127, 238
who.educ, 119
who.flang, 119, 127, 238
who.resp, 119
who.role, 119, 128
who.sex, 118, 119, 121, 122
who.soc, 119, 127, 128

page_255