THE JOURNAL OF ASIA TEFL Vol. 5, No. 4, pp.

27-47, Winter 2008

American and Japanese Listener Assessment of

Japanese EFL Speech: Pronunciation Features
Affecting Intelligibility

Atsuko Kashiwagi
Showa Women’s University, Japan

Michael Snyder
Showa Women’s University, Japan

A study was conducted to examine what pronunciation aspects of

Japanese EFL speakers cause mis-hearings for both American and
Japanese listeners. Speech samples of 20 Japanese students were
evaluated by Japanese and American judges for intelligibility and
accentedness. Intelligibility was measured unambiguously by comparing
what the subjects intended to say and what the judges transcribed. The
judges also rated the accentedness of the subjects impressionistically on
a scale of 7. Interviews were then conducted with the judges to discuss
what pronunciation features were perceived to be the primary causes of
their misunderstandings. Analysis of the data seems to indicate that
intelligibility and accentedness have a quasi-independent relationship
and that a strong accent does not necessarily lead to unintelligibility.
Data also indicate that most of the pronunciation mistakes perceived to
have caused misunderstandings are segmental, not suprasegmental. Few
suprasegmental features were perceived to have caused problems,
except for word stress irregularities which proved detrimental to
intelligibility in quite a few cases.

Key words: EIL, accentedness, intelligibility, segmentals,

suprasegmentals

27
American and Japanese Listener Assessment of Japanese EFL Speech…

INTRODUCTION

Understanding in communication is a highly complex process, but

intelligible pronunciation is a basic prerequisite to successful communication.
There is a wealth of anecdotal evidence to attest to the importance of
intelligible pronunciation (Jenkins, 2000). Research also shows that there is a
threshold level of pronunciation for nonnative speakers of English. If they
fall below this level, they will not be understood regardless of how excellent
and extensive their syntactic and semantic control might be (Hinofotis &
Bailey, 1980).
Teachers of EFL (English as a Foreign Language) and EIL (English as an
International Language) are now much more likely to set reasonable
intelligibility, rather than native-like accent, as the goal of pronunciation
instruction (Abercrombie, 1949; Celce-Murcia, Brinton & Goodwin, 1996;
Pennington & Richard, 1986). This implies that it makes more sense for
instruction to focus more on pronunciation features which contribute most to
intelligibility and de-emphasize those which are less crucial, instead of trying
to teach all the pronunciation features of native speech.

Issues in Intelligibility

Segmental vs. Suprasegmental Features

The question of what aspects of pronunciation are most crucial for

intelligibility, however, proves to be more elusive. Many teachers as well as
researchers feel that suprasegmental features are more important than
segmentals, contributing more to meaning and to listener perception of
nonnative speaker intelligibility (Avery & Ehrlich, 1992; Celce-Murcia,
Briton & Goodwin, 1996). However, empirical evidence for this assertion is
not conclusive.
A number of studies have been conducted on the relative contribution of
segmentals and suprasegmentals to intelligibility, but their findings have

28
 The Journal of Asia TEFL

often been contradictory. Anderson-Hsieh, Johnson and Koehler (1992)

conducted a study to examine how segmental accuracy and prosodic features
affected the overall impression of NNS pronunciation. Their finding that the
prosodic variable is most strongly related to the perceived intelligibility/
accentedness seemed to suggest that suprasegmental features are more
important to NNS intelligibility. Other researchers (Johansson, 1978; Palmer,
1976) also found evidence to support this view.
However, Koster and Koet (1993) reported that segmental errors made by
Dutch speakers were rated as “more annoying” than suprasegmental errors by
both NS judges and Dutch judges. Another study also concluded that
segmental errors have a more negative impact on intelligibility (Fayer &
Krasinski, 1987). Rost (1990) observed that even among native speakers,
most misunderstandings occur at a segmental level, even though higher level
schematic effects may also be at play.
Perhaps these different findings have been derived from differences in
research methodologies, as well as how intelligibility was defined and
measured, as Munro and Derwing (1995) argued. In most of the studies,
intelligibility was often equated with accentedness. Munro and Derwing,
however, made a distinction between accentedness and intelligibility, and
offered working definitions for both concepts. Intelligibility was defined as
“the extent to which a speaker’s message is actually understood by a listener”
(p. 76), and was measured by exact word matches between intended
messages and transcriptions. Accentedness was defined as “how strongly
accented a speaker is” and was measured impressionistically on a Likert scale.
They found intelligibility and accentedness were moderately correlated, but
concluded that a strong foreign accent did not necessarily lead to decreased
intelligibility. They also examined how intelligibility and accentedness were
related to variables such as phonemic errors, phonetic errors, intonation,
grammatical errors, and utterance length, but very few correlations were
found (Munro & Derwing, 1995).
In another study, the same researchers found similar general relationships
between intelligibility and accentedness, but their findings could not

29
American and Japanese Listener Assessment of Japanese EFL Speech…

determine which language features (i.e., grammar scores, phonemic scores,

prosodic scores, speaking rates) contributed more to the two assessments
(Derwing & Munro, 1997).
Kashiwagi, Snyder and Craig (2006) conducted a study of speech samples
by Japanese learners of English, using the notions of intelligibility and
accentedness as defined by Munro and Derwing (1995). Their finding was
similar to that of Munro and Derwing in that accentedness was moderately
correlated with intelligibility, but a strong accent did not automatically lead
to less intelligibility. In their study, NS judges were asked to transcribe
speech samples and were interviewed later to discuss what pronunciation
features were responsible for their misunderstandings. The interview results
indicated that most misunderstandings were caused by segmental level errors.
Misplaced word stress complicated by segmental errors, and lack of sentence
stress on content words were also responsible for some misunderstandings,
but no other suprasegmental features led to major intelligibility problems.
More recent research does not directly debate the primacy of either
suprasegmentals or segmentals, but focuses more on finding out what
specific pronunciation features are responsible for the reduction of
intelligibility. Hahn (2004) used three versions of a lecture in which nuclear
stress was manipulated and found that those who listened to the lecture with
misplaced or missing nuclear stress understood the lecture significantly less.
Field (2005) manipulated lexical stress, and found that misplaced stress
resulted in decrement of intelligibility. Levis (1999), however, offers a
different viewpoint by pointing out that intonation contours may not be as
important to meaning as is sometimes believed.
In summary, past empirical evidence suggests that not all the pronunciation
features contribute equally to intelligibility, and that a strong foreign accent
does not automatically lead to unintelligibility. It does not support either the
primacy of suprasegmentals or the primacy of segmentals in contributing to
intelligibility, but shows that deviations in certain suprasegmental features do
cause misunderstandings, just as errors in the production of certain segmental
features are responsible for unintelligibility. More research is needed to find

30
 The Journal of Asia TEFL

out what specific aspects of suprasegmentals as well as segmentals interfere

most with intelligibility.

NNS Listeners’ Perspectives

The number of English speakers in the world has been estimated at 375
million first-language speakers, 375 million second-language speakers and
750 million foreign-language speakers (Graddol, 1997). As English is spoken
by more nonnative speakers than by native speakers, and instances of NNS-
NNS interaction increase, no discussion of intelligibility would be complete
without the perspectives of NNS listeners listening to NNS speech. Only a
few studies on intelligibility, however, have been conducted with NNS
listeners listening to NNS speech.
Jenkins studied NNS-NNS interaction data collected over a period of three
years in social and classroom settings. She concluded that segmental errors
were by far the most frequent causes of miscommunication in NNS-NNS
interactions. Most suprasegmentals such as features of connected speech
(elision, contractions, assimilation and weak forms), pitch movements to
signal attitude or grammatical meaning, placement of word stress, stress-
timed rhythm, were not found to cause any irreparable intelligibility problems
(Jenkins, 2000, 2002).
Riney, Takagi and Inutsuka (2005) found that American listeners and
Japanese listeners judged degrees of accent in English differently. While
American listeners primarily used segmental cues in their perception of
accent, Japanese listeners used non-segmental parameters (intonation,
fluency, and speech rate). While their research does not directly address the
issue of intelligibility, their findings have important pedagogical implications.
More empirical evidence is needed to gain insight into which pronunciation
errors cause intelligibility problems in NNS-NNS interactions, as compared
to NNS-NS interactions. NNS-NNS data will be invaluable in designing
pronunciation curricula to prepare students for a wide range of communication
situations where NSs will not be the only listeners.

31
American and Japanese Listener Assessment of Japanese EFL Speech…

THE PRESENT STUDY

How intelligibility is determined is a complex process which involves a

multitude of factors. The present study is conducted in an effort to provide
more empirical evidence on the many factors contributing to intelligibility by
investigating 1) the relationships between accentedness and intelligibility, 2)
what specific pronunciation features of NNS speech cause mis-hearings, and
3) whether NSs and NNSs react differently to pronunciation errors of NNS
speech. While the present study examines some of the issues covered in the
past studies, it is different from them in that it assesses not only NS listeners’
reaction, but also NNS listeners’ reaction to NNS speech in a controlled
research setting.
In our study, the notions of intelligibility and accentedness defined by
Munro and Derwing (1995) are used. Intelligibility is broadly defined as the
extent to which a speaker’s utterance is actually understood by a listener and
unambiguously measured in percentage points by exact word matches
between intended messages and transcriptions. Accentedness is defined as the
extent to which a speaker’s pronunciation deviates from a perceived NS
version, and is measured impressionistically on a scale of seven.
The present study is different from Munro and Derwing (1995) or Derwing
and Munro (1997) in that readings of prepared passages, instead of
spontaneous speech, are used as stimuli to prevent factors other than
pronunciation, such as grammar mistakes and inappropriate word choices,
from interfering with intelligibility. It has been shown that judgments of
pronunciation are often influenced by the syntactic and semantic errors made
by the speaker (Briere, 1967; Varonis & Gass, 1982).
Recognizing that intelligibility is highly listener-dependent, the present
study emphasizes the perception of listeners, rather than the production of
speakers. Intelligibility is calculated based on listeners’ transcriptions,
accentedness is determined impressionistically by listeners, and causes of
misunderstandings are identified by listeners themselves in later interviews,
rather than the researchers making speculations based on their analysis of

32
 The Journal of Asia TEFL

pronunciation errors.
The present study investigates not only the perception of NS listeners but
also that of NNS listeners of the same stimuli. In the study, however, NS
listeners are represented only by American English native speakers and NNS
listeners only by Japanese speakers due to the difficulties of finding qualified
judges from different nationalities. The findings, therefore, may not apply
strictly to other NS or NNS groups, but they will serve to provide preliminary
data on which to design more studies.
The present study had the subjects read sensical passages of certain lengths,
rather than collections of words or sentences, in order to provide context for
listeners in their transcription tasks. Listeners are known to remedy
pronunciation errors by using contextual cues (Derwing & Munro, 2005;
Suenobu, Kanzaki & Yamane, 1992), and the purpose of the study is to find
out what pronunciation errors still interfere with intelligibility after the help
of contextual cues.
While in some of the past studies, “suprasegmental features” were used to
refer only to prosody, we define suprasegmental features more broadly as
phenomena occurring over one sound segment such as word stress, sentence
stress and rhythm as well as adjustments in connected speech, and segmental
features are defined as vowels and consonants.

Participants and Procedures

We replicated much of the research procedures of our previous research in

collecting speech samples and evaluating them (Kashiwagi, Snyder & Craig,
2006). Speech samples were collected from 20 Japanese students at a private
women’s university in Tokyo. The students’ English abilities ranged from
false-beginning to intermediate with their TOEIC scores ranging from the
high 200s to the low 600s. None of them had lived in an English-speaking
environment for more than six months.
Each student read two short passages. The total 40 passages were taken
from various lower-intermediate English textbooks to control their difficulty

33
American and Japanese Listener Assessment of Japanese EFL Speech…

and were all different. They ranged in length from 37 words to 55 words with
a mean length of 47.7 words. Before recording themselves in a college
language laboratory, the students were allowed to practice the passages aloud
and to ask us about any questions regarding pronunciation or meaning in
order to decrease the chance of simple orthographic or comprehension errors
interfering with the results.
The 40 passages were then randomized on a cassette tape, and played to
three native speaker judges and three nonnative speaker judges to be
transcribed in standard orthography. All of the native speaker judges were
American EFL instructors teaching in Japan and knowledgeable in English
phonology. All three nonnative speaker judges were Japanese EFL instructors,
who had obtained their MA degrees in TESOL in the U.S. and had an equally
good understanding of English phonology. Upon completion of the
transcription, they were given a short break and they listened to the 40
passages a second time to rate the accentedness of each utterance
impressionistically on a scale of 7 with 1 indicating “very strongly accented”
and 7 “no accent.”
Before the accentedness rating task, the judges practiced by listening to
and rating examples of readings from our 2006 study. The transcriptions and
the accentedness ratings were done separately to assure that they did not
influence each other.
A few days after the transcriptions, two researchers had an interview with
each of the judges. During the interview, the judges compared their
transcriptions and the original passages, and were asked to comment on what
they believed to be the causes of their misunderstandings while listening to
the speech samples again. Their judgments as to the causes of their
misunderstandings were noted by the researchers and later tabulated.
Intelligibility was calculated in percentage points by exact word matches
between intended messages and transcriptions just as in the studies by Munro
and Derwing (1995), and Derwing and Munro (1997). Transcription errors
which were considered to have been caused by regularizations (e.g., he walks
for he walk), word insertions, and semantic substitutions (the substitution of

34
 The Journal of Asia TEFL

one word for a semantically similar word) were ignored in calculating

intelligibility.
General American pronunciation was used as “a point of reference” in this
paper; there is no intention on our part to imply that General American
pronunciation is the only target for Japanese students to achieve. The IPA-
modified system used in Longman Dictionary of Contemporary English
(2003) was used for phonetic transcriptions.

RESULTS AND DISCUSSION

Accentedness and Intelligibility

Both American judges and Japanese judges found most students in the
study more than moderately accented. They rated a majority of the speech
samples between 2 and 4, with the ratings of 2 and 3 accounting for more
than half the cases (See Figure 1). There was a significant difference between
the judges when the accentedness ratings of the six judges were subjected to
ANOVA. A Sheffe’s test found significant differences between six pairs of
judges. American Judge A was found significantly stricter in his accentedness
ratings than three of the other judges (American Judge B, Japanese Judge A
and Japanese Judge B) and Japanese Judge C was found significantly stricter
than two judges (American Judge B and Japanese Judge A). The results seem
to suggest that accentedness ratings vary considerably among the judges, but
the differences are influenced more by individual factors than the judges’
language backgrounds.

35
American and Japanese Listener Assessment of Japanese EFL Speech…

TABLE 1
Descriptive Statistics (Intelligibility and Accentedness)
Judges Minimum Maximum Mean SD
Accentedness NS Judge A 1.00 4.00 2.28 0.96
NS Judge B 2.00 6.00 3.85 1.12
NS Judge C 1.00 5.00 2.68 1.16
NS Mean 1.30 4.70 2.93 0.92
NNS Judge A 2.00 6.00 4.03 1.14
NNS Judge B 1.00 7.00 3.38 1.41
NNS Judge C 1.00 5.00 2.70 1.09
NNS Mean 1.30 5.30 3.35 1.01
Intelligibility NS Judge A 59.00 100.00 89.30 11.15
NS Judge B 70.00 100.00 92.40 8.53
NS Judge C 49.00 100.00 85.80 12.52
NS Mean 59.67 100.00 89.17 9.72
NNS Judge A 57.00 100.00 84.97 11.41
NNS Judge B 44.00 92.00 67.05 11.57
NNS Judge C 35.00 97.00 69.38 12.77
NNS Mean 47.33 95.33 73.39 9.99

FIGURE 1
Distribution of Accentedness Ratings

80
70
Number of Ratings

60
50
40
30
20
10
0
1 2 3 4 5 6 7
Accentedness

1= very strongly accented 7= no accent

The judges’ language backgrounds, however, did seem to affect their

36
 The Journal of Asia TEFL

intelligibility scores. The Japanese judges, with the exception of Japanese

Judge A, clearly had more trouble understanding the speech samples than the
American judges did. When the intelligibility scores of the six judges were
subjected to ANOVA, there was a significant difference. A Sheffe’s test
found that the intelligibility scores of Japanese Judge B and Japanese Judge C
were significantly lower than the scores of all the other judges. Both Japanese
Judge B and Japanese Judge C seemed less capable of using contextual cues
to repair pronunciation errors than the other judges, and often produced
nonsensical transcriptions. They also seemed to lose longer stretches of
speech once they had their comprehension impeded by pronunciation
deviances. Even though the three Japanese judges all had studied in the U.S.,
Japanese Judge A was the only one who had lived in the U.S. as a child. Her
exposure to English as a child could have put her phonological processing
capacity on par with that of American judges.
The results here seem to support Jenkins’ claim that pronunciation errors
tend to cause more problems for NNSs than for NSs since the latter have a
wider range of phonetic and phonological tolerance and are more able to
make use of contextual information to negotiate meaning, while NNSs “tend
to focus on the acoustic signal and direct his or her efforts to decoding what
has been heard” (Jenkins, 2002, p. 89). Jenkins based her claim on NNS-NNS
interaction data, in which the subjects did not share the same L1, but reported
that in cases where the subjects shared the same L1, pronunciation errors,
most of which were presumably caused by L1 interference, did not impede
mutual intelligibility. The results in this study, however, seem to suggest that
the Japanese judges found pronunciation errors, even those caused by
interference from their L1, more problematic to intelligibility than the
American judges did.
Pearson’s Correlation analyses between accentedness ratings and intelligibility
scores showed a sizable variation among the judges (See Table 2), with
moderate correlations found only for three judges. To examine the relationships
between intelligibility and accentedness more closely, we looked at the 23
cases which were perfectly transcribed, and examined how the accentedness

37
American and Japanese Listener Assessment of Japanese EFL Speech…

ratings for those cases varied. Figure 2 shows that accentedness judgments
for those 23 cases were evenly distributed across a wide range of scores, with
five utterances rated as 1 and 2. These data indicated that these judges
understood some utterances perfectly even when they perceived the speakers
to have a strong accent. This gives further evidence that some features of
foreign accent do not necessarily reduce intelligibility.

TABLE 2
Correlation between Intelligibility and Accentedness
American American American Japanese Japanese Japanese
Judge A Judge B Judge C Judge A Judge B Judge C
.306 .561** .355* .546** .219 .257
* = p < .05 ** = p < .01

FIGURE 2
Accentedness Distributions for the 23 Cases Transcribed Perfectly

8
7
Number of Perfect

6
Transcriptions

5
4
3
2
1
0
1 2 3 4 5 6 7
Accentedness

1=very strongly accented 7=no accent

Pronunciation Errors Causing Misunderstandings

The results in the present study suggest that in the data of both American

38
 The Journal of Asia TEFL

and Japanese judges, accentedness and intelligibility are quasi-independent,

and that a strong accent does not automatically lead to misunderstandings.
However, we still need to find out which pronunciation features directly
affect intelligibility, and which features are relatively harmless ones easily
repaired by contextual cues.
Table 3 shows the types of pronunciation errors which were perceived by
the judges to be the causes of unintelligibility. A Chi-square analysis
compared the answers by the 6 judges across the three main categories of
error types (segmental errors, suprasegmental errors, and segmental +
suprasegmental errors). The result, W2(10) = 7.50, p= .67, showed there were
no significant differences between the judges.

TABLE 3
Identified Causes of Misunderstandings by Error Types
NS Judges NNS Judges
Error Types A B C A B C
Segmental errors
Consonant errors 14 10 13 13 17 27
(Incl. consonant clusters)
Vowel errors 24 16 22 18 33 26
Combination (of segmental errors) 10 8 10 10 10 11
Epenthesis (Sound insertion) 5 5 7 2 4 2
Subtotal 53 39 52 43 64 66
Suprasegmental errors
Word stress error 2 3 4 3 6 5
Stress error on noun phrase 1 2 1 3 1 2
Irregular sentence stress 3 2 2 2 2 2
Subtotal 6 7 7 8 9 9
Suprasegmental errors
+ Segmental errors
Word stress error + 3 1 5 5 3 8
segmental error
Stress error on noun 0 0 3 0 1 1
phrase + segmental error
Irregular sentence 0 0 0 0 0 0
stress + segmental error(s)
Subtotal 3 1 8 5 4 9
Overall Total 62 47 67 56 77 84
Chi-square = 36.97 (not significant at p = .92)

39
American and Japanese Listener Assessment of Japanese EFL Speech…

Segmental Errors Causing Misunderstandings

Mispronunciation of segmental sounds, especially vowels, was perceived

by both American and Japanese judges to have caused the bulk of the
problems.

Vowels

For both American and Japanese judges, vowel mispronunciation was the
most problematic. Similar vowel problems were reported by both American
and Japanese judges. The vowel /z/, substituted with a Japanese vowel /`/ in
most cases, was thought to have caused 15 instances of misunderstandings:
e.g., mad → much; manuscripts → money scripts; distracted → destructed.
The vowel /@9/ and /U/, substituted with /`/ and often with another Japanese
vowel /n/, proved problematic in five and four instances respectively: e.g.,
collars → colors; won → want. Other vowels, such as /P9/, /nT/, and /H/ also
caused misunderstandings for all the judges.
R-colored vowels (/29q/,/?q/,/@9q/,/N9q/) were perceived to be problematic in
48 cases: e.g., workmen → walkmen; letters → little; pried apart → proud
about; horns → home. Mispronunciation of diphthongs was cited as the
primary causes for 21 misunderstandings. The diphthong /dH/, substituted
with Japanese /d/, was responsible for four cases of misunderstandings. Most
diphthong errors seemed to be caused by orthographical mistakes: e.g.,
intensifies pronounced as /Hm!sdmrHeh9y/; labeling pronounced as /!k`a?kHM/.
Vowel lengths before voiced and voiceless stops also seemed to cause
misunderstandings. In casual speech where final stop consonant sounds are
often not released, the only perceptual clue in discriminating a voiced stop
and a voiceless stop is the length of a preceding vowel (Celce-Murcia,
Brinton & Goodwin, 1996). The failure to lengthen vowels before voiced
stops were perceived to have caused six misunderstandings: e.g., bag →
back; pad → path.

40
 The Journal of Asia TEFL

Consonants

Mispronunciation of a consonant was the second most often cited reason

by both American and Japanese judges, and the consonants perceived to be
problematic were quite similar across the two groups of judges. The most
problematic consonants for the two groups of judges were /k/ (both word-
initial and word-final positions) and /q/, which are often confused with each
other, or substituted with Japanese /3/ (Avery & Ehrlich, 1992). Each
consonant was cited as the primary reason for misunderstanding in 14 cases:
e.g., laws → in order; light → xxx (i.e., nothing transcribed); frames →
flames; role → law. Consonant clusters including /k/ and /q/ were also
perceived to be problematic in 18 cases; e.g., /rk/, /oq/, /sq/, /fq/, /eq/, /cq/.
The consonant /e/ was thought to have caused misunderstandings in eight
cases, of which six were reported by Japanese judges. No cases of
misunderstanding involving /u/ were reported in this study.
The two consonants /S/ and /C/ were responsible for two and six cases
respectively: e.g., thousands → someone’s; they → take. Most of the
misunderstandings involving /S/ and /C/ were also reported by Japanese
judges.
Final /s/ and /c/ were also thought to be problematic for both American and
Japanese judges. When /s/ and /c/ in word-final positions are followed by
another consonant, they often assimilate at the same place of articulation to
the following consonant and are not released: e.g., that pen → /Czo´oDm/. As
a result, /s/ and /c/ become inaudible signaled only by a pause which
psychologically accounts for the deletion. The failure to put in a sufficient
pause was thought to have caused ten misunderstandings in this study: e.g.,
can’t move → can move; movie stunt work → movie star work; old property
→ all property.

Combination of Segmental Errors

The problematic segmentals described above, both vowels and consonants,

41
American and Japanese Listener Assessment of Japanese EFL Speech…

also caused many misunderstandings when combined. The phrase “not so”
was misunderstood as “also” by five judges when there was not enough pause
to account for the unreleased final /s/, and / @9 / in “not” is replaced with a
Japanese vowel /n/. The word “thought” was misunderstood as “solved”
when /S/ is substituted with /r/, and /P9/ was replaced with /nT/.
Mispronunciation of the consonant /v/, while it caused few problems alone,
was responsible for five misunderstandings when compounded by vowel
errors. The word “worn” wasn’t understood when the initial /v/ was
pronounced with not enough rounding of the lips and the vowel /N9q/ was
substituted with Japanese /n/. Another word “warm” was misunderstood as
“bone” when the mispronunciation of /v/ is added to by a vowel error. Other
consonants, such as /u/ and word-initial /o/, were also perceived to be
problematic when combined with other segmental errors.

Epenthesis

Japanese speakers’ tendency to insert a vowel after a consonant, either on

its own or coupled with other pronunciation errors, was cited as the primary
reason for misunderstanding in 25 cases, of which 17 were reported by
American judges. For example, the word “drills” was misunderstood as
“riddles” when a Japanese vowel /L/ was inserted after /c/. Most epentheses
occurred in consonant clusters including /k/ and /q/, before a plural suffix or
before a past tense suffix, suggesting that students tended to insert a sound
before what they considered a difficult pronunciation task.

Suprasegmental Errors as Causes of Misunderstandings

Intonation, rhythm patterns and features of connected speech were not

cited to have caused misunderstandings by either of the two groups of judges,
even though some suprasegmental features such as irregular word and
sentence stress apparently led to some mis-hearings. This could be because
the subjects are reading passages as opposed to that of using spontaneous

42
 The Journal of Asia TEFL

speech, a limitation we acknowledge in our conclusions.

Word Stress

While Jenkins (2000, 2002) as well as Kashiwagi, Snyder and Craig (2006)
reported that word stress deviations did not lead to many unintelligibility
problems unless they occur in tandem with segmental errors, the present
study found that word stress error alone seemed to be powerful enough to
cause misunderstandings for both American and Japanese judges. A total of
23 cases was reported: e.g., communities (no clear stress) → comments;
bystanders (no clear stress) → xxx; proc!ess → possess; as!pects → xxx.
When word stress irregularities were coupled with other problems, they were
responsible for an additional 25 cases.
What is important to note is that many cases of word stress errors were
caused when the speaker did not make a clear enough distinction between
stressed and unstressed syllables; the speaker pronounced the stressed
syllables a little more loudly than the unstressed syllables, but it was not
accompanied by a higher pitch and longer duration as the English stress rules
require, and apparently was not perceived as the correct word stress pattern
by the listener.

Stress in Noun Compounds vs. Adjective-Noun Phrases

English stress rules dictate that in noun compounds the main stress be
placed on the first component, but that in adjective-noun phrases, it be put on
the noun: e.g, !blÂack$board, but $black !board. Failure to observe these rules
caused ten misunderstandings; it was responsible for an additional 5
misunderstandings when coupled with segmental errors. The adjective-noun
phrase “Dead Sea,” when the main stress was placed on “Dead,” caused
problems for five of the six judges. Other compounds that caused problems
were “grape growers,” “slow waltz,” “clay pot,” and “precious way.”

43
American and Japanese Listener Assessment of Japanese EFL Speech…

Sentence Stress

While failures to reduce function words in sentences did not cause any
problems, insufficient sentence stress on content words was reported to have
caused 13 cases of miscommunication. For example, the failure to place
sufficient sentence stress on “think” in “some think” resulted in four of the
six judges misunderstanding it as “something.”

CONCLUSION

The present study suggests that in the data of both American and Japanese
judges, a stronger accent does not automatically lead to less intelligibility,
and that not all the pronunciation errors detract from intelligibility.
Closer examination of each instance of mis-hearing showed that American
and Japanese judges tended to experience difficulties with similar pronunciation
errors. Mispronunciation of segmental sounds, especially vowels, was
perceived by both American and Japanese judges to have caused the bulk of
their problems. Deviations in suprasegmental features were thought to be less
problematic by both groups of judges. Even though irregular word and
sentence stress did seem to lead to some problems, other suprasegmental
features such as intonation, rhythm patterns and features of connected speech
(elision, contractions, assimilation and weak forms), while they may have
contributed to foreign accent, were not found to be detrimental to intelligibility.
The data obtained in the study resemble those of Jenkins (2000, 2002) and
Kashiwagi, Snyder and Craig (2006) in the relative contributions of
segmental and suprasegmental features to intelligibility, while there are some
differences. The earlier two studies did not find irregular word stress causing
unintelligibility, but the present study seemed to show that word stress
affected intelligibility in quite a few cases. In addition, while the present
study suggested that vowels were more problematic than consonants, Jenkins
consistently found that consonant errors caused more problems than vowels.

44
 The Journal of Asia TEFL

There is another important difference with earlier research. While Jenkins

(2002) reported that in cases where the subjects shared the same L1,
pronunciation errors did not impede mutual intelligibility, the present study
showed that the Japanese judges did not find Japanese student EFL speech
samples easier to understand than American judges did. Even though most of
the pronunciation errors seemed to be caused by L1 transfer, the Japanese
judges were not always able to draw on their L1 knowledge to compensate
for pronunciation deviations. Presumably hampered by processing overload,
they seemed to rely more on acoustic information and less on contextual cues.
Further research is needed to answer why some of the findings of the present
study did not agree with those of previous studies.
The present study, while providing some important pedagogical implications,
is limited in several ways. First, it is limited by using readings of prepared
passages as opposed to spontaneous conversations in which prosodic features
may be more important in conveying intentions and emotions. Secondly,
more research is needed to obtain perspectives of NS judges who are
unfamiliar with Japanese speakers of English as well as non-Japanese NNS
judges’ perspectives to understand a more complete picture. Finally, though
the six judges in the present study are all experienced EFL teachers with
some training in English phonology, it is still possible that they failed to
associate their misunderstandings with suprasegmental errors as suprasegmental
features are more subtle in their impact. Further research which examines the
influence of specific suprasegmental features on intelligibility will help
provide valuable information.
Teachers are constantly faced with the difficult task of choosing how to
best spend their class time in increasingly crowded English curricula. More
empirical evidence on what specific aspects of L2 speech have the greatest
impact on intelligibility for both NS and NNS listeners is essential for
prioritizing pronunciation instruction, and defining a more realistic goal.
Further studies are needed in order to refine our understanding of what
constitutes basic intelligible pronunciation and help teachers make good
choices for the pronunciation part of their EFL/EIL classes.

45
American and Japanese Listener Assessment of Japanese EFL Speech…

THE AUTHORS

Atsuko Kashiwagi is a professor in the Department of English Communication

at Showa Women’s University in Japan. Her interests are in curriculum
development, learner differences, and the teaching of listening and pronunciation
skills.
Email: akowagi@swu.ac.jp

Michael Snyder is also a professor in the Department of English

Communication at Showa Women’s University. Besides pronunciation, he is
interested in the development of vocabulary and reading ability through
content-based language teaching.
Email: mksnyder@swu.ac.jp

REFERENCES

Abercrombie, D. (1949). Teaching pronunciation. English Languae Teaching, 3, 113-122.

Anderson-Hsieh, J., Johnson, R., & Koehler, K. (1992). The relationship between
native speaker judgments of nonnative pronunciation and deviance in
segmentals, prosody, and syllable structure. Language Learning, 42(4), 529-
555.
Avery, P., & Ehrlich, S. (1992). Teaching American English pronunciation. Oxford;
Oxford University Press.
Briere, E. J. (1967) Phonological testing reconsidered. Language Learning, 17, 163-171.
Celce-Murcia, M., Brinton, D. M., & Goodwin, J. M. (1996) Teaching pronunciation:
A reference for teachers of English to speakers of other languages. Cambridge:
Cambridge University Press.
Derwing, T. M., & Munro, M. J. (1997). Accent, intelligibility, and comprehensibility:
evidence from four L1s. Studies in Second Language Acquisition, 19, 1-16.
Derwin, T. M., & Munro, M. J. (2005). Second language accent and pronunciation
teaching: A research-based approach. TESOL Quarterly, 39(3), 379-397
Fayer, J. M., & Krasinski, E. (1987). Native and nonnative judgments of intelligibility
and irritation. Language Learning, 37(3), 313-326.
Field, J. (2005). Intelligibility and the listener: The role of lexical stress. TESOL

46
 The Journal of Asia TEFL

Quarterly, 39(3), 399-423

Graddol, D. (1997). The future of English? London: British Council.
Hahn, L., D. (2004). Primary stress and intelligibility: Research to motivate the
teaching of suprasegmentals. TESOL Quarterly, 38(2), 201-223.
Hinofotis, F., & Bailey, K. (1980). American undergraduate reactions to the
communication skills for foreign teaching assistants. In J. Fisher, M. Clarke, &
J. Schacter (Eds.), On TESOL ’80: Building bridges (pp. 120-133). Washington,
DC: TESOL.
Jenkins, J. (2000). The phonology of English as an international language. Oxford:
Oxford University Press.
Jenkins, J. (2002). A sociolinguistically based, empirically researched pronunciation
syllabus for English as an international language. Applied Linguistics, 23(1),
83-103.
Johansson, S. (1978). Studies in error gravity: Native reactions to errors produced by
Swedish learners of English. Gothenburg Studies in English, 44, 1-17.
Kashiwagi, A., Snyder, M., & Craig, J. (2006). Suprasegmentals vs. segmentals: NNS
phonological errors leading to actual miscommunication. JACET Bulletin, 43,
43-57
Koster, C. J., & Koet, T. (1993). The evaluation of accent in the English of Dutchmen.
Language Learning, 43(1), 69-92.
Levis, J.M. (1999) Intonation in theory and in practice, revisited. TESOL Quarterly,
33(1), 37-63.
Munro, M. J., & Derwing, T. M. (1995). Foreign accent, comprehensibility, and
intelligibility in the speech of second language learners. Language Learning,
45(1), 73-97
Palmer, J. (1976). Linguistic accuracy and intelligibility. In Proceedings of the 4th
International Congress of Applied Linguistics (pp. 505-513). Stuttgart, Germany:
Hockschul Verlag.
Riney, T. J., Takagi, N., & Inutsuka, K. (2005). Phonetic parameters and perceptual
judgments of accent in English by American and Japanese listeners. TESOL
Quarterly 39(3), 441-466.
Rost, M. (1990). Listening in language learning. London: Longman.
Suenobu, J., Kanzaki, K., & Yamane, S. (1992). An experimental study of intelligibility of
Japanese English. International Review of Applied Linguistics in Language
Teaching, 30, 146-156.
Varonis, E. M., & Gass, S. (1982). The comprehensibility of non-native speech.
Studies in Second Language Acquisition, 4(2), 114-136.

47
Reproduced with permission of copyright owner. Further reproduction
prohibited without permission.