From Latin to Romance: case loss and preservation in pronominal systems

M. Rita Manzini - Leonardo M. Savoia

Abstract. The evolution from Latin into Romance is marked by the loss of case in nominal
declensions. In most Romance varieties, however, pronouns, specifically in the 1st/2nd person
singular, keep case differentiations. In some varieties 1st/2nd singular pronouns present a three-way
case split, essentially the same reconstructed for proto-Romance (De Dardel and Gaeng 1992,
Zamboni 1998). We document and analyze the current situation of Romance in the first part of the
article (section 1). In the second part of the article we argue that the Dative Shifted distribution of
loro in modern Italian, accounted for by means of the category of weak pronoun in Cardinaletti and
Starke (1999), is best construed as a survival of oblique case in the 3rd person system (section 2).
This casts doubts on the weak pronoun category, as applied to Old Italian as well (Egerland and
Cardinaletti 2010).

1. From Latin to Romance varieties: Three case systems

Within the minimalist framework, Chomsky (2001, 2008) proposes that the real underlying relation
between case assigner and case assignee is agreement in phi-features. While phi-features are
interpretable on DPs, case is not; its status is that of a mere reflex of agreement on DPs. The most
obvious problem with a reduction of case to agreement is that mismatches between agreement and
case are fairly common in natural languages. First, there are instances where the EPP arguments of
finite sentences are in the accusative, for example causative constructions in infinitive-less Balkan
languages (Greek, Albanian). According to Iatridou (1993), Chomsky (2001) the relevant sentences
in Greek are untensed (i.e. they are in an invariable present form); but Manzini and Savoia (2007)
show that in Arbëresh (Italo-Albanian) varieties they involve the embedding of past tenses, as in
(1). In (1) therefore the accusative form buStrin@ agrees with TP – and in fact it alternates with the
predicted nominative.

(1) (E) bara t@ fraç@ buStrin@/buStri

(it) made-1sg Prt slept-3sg dog-acc.def/dog-nom.def
‘I made the dog sleep’
Vena di Maida
 Baker and Vinokurova (2010), in turn, study instances where accusatives are assigned in the
absence of a vP in Sakha (Turkik). Similar phenomena are described in the typological literature
under the label of 'extended accusative' (Plank 1985), documented in familiar languages like Latin.
Specifically, Late Latin has a number of accusatives corresponding to the sole argument of
unaccusative verbs, including middle-passives (M/P), as in (2a). The same pattern appears more
sporadically in Early Latin, as in (2b); other examples are discussed in Cennamo (2011), where the
data in (2) come from. Since unaccusative verbs by hypothesis lack a vP projection, the prediction
is that they exclude accusative arguments; yet ipsos ficos in (2a) and vitam in (2b) are accusatives.

(2) a. ipsos ficos imponatur

these-acc figs-acc gather-subj.M/P.3sg
‘One should gather these figs’
(Rufus of Ephesus, De Podagra 35)
b. vitam vivitur
life-acc live-indic.M/P.3sg
‘Life is lived’ (i.e. ‘One lives one’s life’)
Early Latin (Ennius, Tragoediae 241)

Given the difficulties involved in Chomsky’s conception, Manzini and Savoia (2011a) revert
to the classical idea that the category of case has an interpreted content, namely as an elementary
predicate/operator. As in standard lexicalism, entries specify a mapping between sound and
meaning (cf. Jackendoff 2002), and they do so without any distinction between the so-called
functional lexicon (including case inflections) and the substantive lexicon. Correspondingly, there is
no morphological buffer between functional nodes and the exponents that instantiate them, in the
manner of Distributed Morphology; morphology and syntax form a unified module of grammar,
projected directly from lexical items – which seems to be what the minimalist program (Chomsky
1995) intends.
In this section, we pursue Manzini and Savoia’s approach, illustrating it first with Latin, and
in particular with the –i inflection; we associate this ending with a relational content, which
accounts both for its oblique and its plural reading (section 1.1). We then analyze three-case
systems (nominative, objective, oblique) found with 1st/2nd person pronouns in Romance (sections
1.2, 1.3, 1.5, 1.6). The view that case inflections are interpreted makes it possible for us to associate
a theoretical content with the conclusion that, for instance, the –i inflection of Romance (oblique)
continues the –i inflection of Latin – namely its interpretive content remains essentially constant
(section 1.4).

1.1 The notion of case: Latin -i

Consider the Latin inflection -i seen in the genitive of the II class, as in (3b), and in the dative of the
III class, as in (3a) – i.e. generally in the oblique. For Manzini and Savoia (2010), –i, like the
prepositions to and of in English, lexicalizes an elementary predicate introducing a possession
relation between the noun (phrase) to which the case ending attaches and a local argument. This is
the first internal argument of the ditransitive predicate in the ‘dative’ (3a), and the head of the DP in
the ‘genitive’ (3b). The second internal argument of ditransitives (the dative), as in (3a), has been
connected to possession in the formal literature at least since Kayne (1984).

(3) a. mulier-i omnia dat dono

woman-dat all-pl give-3sg gift-abl
‘He gives everything to the woman as a gift’
(Plautus, Miles Gloriosus, 1137)
b. nomen domin-i
name master-gen
‘The name of the master’
(Plautus, Mostellaria, 661)

An idea put forth in similar terms by various strands of literature is that possession is in fact
a surface manifestation of the more elementary part-whole relation. Manzini and Savoia (2005,
2007) propose that the Romance clitic ne (syncretic in some varieties between genitive and dative)
denotes a superset-of some other argument of the sentence (the theme). Belvin and den Dikken
(1997:170) define the relation introduced by have as “zonal inclusion” in the following terms: “the
‘meaning’ of have … denotes a special kind of inclusion relation … dubbed ‘zonal inclusion’…
Entities have various zones associated with them, such that an object or eventuality may be included
in a zone associated with an entity without being physically contained in that entity… The type of
zones which may be associated with an entity will vary with the entity.” Manzini and Savoia
(2011a) notate the relevant relation with ‘Í’, to be understood not mathematically but as looser
zonal inclusion. This relation can be lexicalized by case endings (Latin) or it can be lexicalized by
prepositions (English). We correspondingly notate prepositions like to and of as P(Í), as a reminder
of their content. As for oblique endings like Latin -i, we label them as Q(Í), since relational content
within the nominal domain is associated with Q categories (cf. generalized quantifier theory).
In (3a), following Kayne (1984), the complement of the ditransitive verb ‘give’ is a
predication PredP. In present terms, the dative -i inflection introduces a possession predicate, Q(Í),
which takes mulier as its internal argument (the possessor) and the theme of the verb omnia as its
external argument (the possessum), as in (4).

(4) VP
wp
PredP V
wp dat
QP DP
3 omnia
N Q(Í)
mulier i

The same relation Q(Í), introduced by the same inflection –i predicates possession/inclusion
of nomen by domini in (3b). The -i inflection with Q(Í) content takes the possessor domin- as its
internal argument and the possessee nomen as its external argument, as in (5).

(5) NP
wp
N QP
nomen 3
N Q(Í)
domin i

Several issues are raised by this proposal. First, datives also occur as arguments of
unergative verbs. We have seen that with ditransitive verbs Q(Í) establishes a relation between the
argument to which it attaches and another argument present within the VP. The question is what the
Q(Í) inflection –i does in an unergative sentence like (6a). Intuitively, unergative predicates can be
paraphrased by a causative predicate associated with an eventive nominal. Thus in English answer
alternates with give an answer to. Hale and Keyser (1993), Chomsky (1995) formalize this intuition
about the complex nature of unergative predicates by assuming that they result from the
incorporation of an elementary state/event noun into a transitivizing predicate (CAUSE, or v), cf.
(6b). Within such a conceptual framework it is possible to argue that Q(Í) takes as its arguments
the noun to which it attaches (the possessor) and the elementary state/event (the possessum). Thus
(6a) can be informally rendered as ‘He caused the woman (to get) an answer’, as in (6b) – which
justifies the presence of Q(Í), i.e. the descriptive dative.

(6) a. […] mulier-i responderet

… woman-dat answer-subj.3sg
‘He would answer the woman’
(Gaius Ateius Capito, Iurisprudentia, 9.8)
b. EA [CAUSE/v [answer [Q(Í) (the) woman]

With genitives, a potential problem concerns eventive or deverbal nouns. In this instance,
the genitive lexicalizes not the possessor, with a notoriously loose relation with the head noun, but
what appears to be an internal or external argument of the noun, with a stricter relation to it, as
illustrated in (7). Despite apparent interpretive differences, we provisionally maintain the same
characterization for the genitive in (7) as we have provided for possessors. In other words,
inclusion, Q(Í), is the all-purpose attachment for complements of nouns, though its interpretation
appears to be restricted when it satisfies an argument slot of eventive/deverbal nouns.

(7) imperi-i cupiditas

command-gen desire
‘The desire for command’
(Cicero, Rhetorica ad Herennium [sp.] 2.34.1)

A different type of problem with our approach is met when we consider that –i is not just an
oblique singular in Latin, but also a plural, in particular the nominative plural of the same II
declension illustrated in (3b), as in (8). One obvious move, face to the difficulty of reducing the
plural nominative in (8) to the singular oblique in (3) is to postulate two different –I inflections, as
proposed by Halle and Vaux (1997), Calabrese (1998, 2008). The problem with this is that the
singular oblique/ plural non-oblique syncretism looks anything but accidental; Manzini and Savoia
(2011a) study it in detail in Albanian, while Johnston (1997) quotes several other examples, e.g.
Russian.

(8) unum […] domin-i fugiunt

one.thing-acc … master-pl shun-3pl
‘The masters shun a single thing’
 (Cicero, De Lege agraria, 2.70.3)

Manzini and Savoia (2010) argue that –i maintains its Q(Í) content both as the oblique in
(3) and as the plural in (8). Specifically, they propose that Q(Í) is construed as plural morphology
if its scope is restricted to the noun it attaches to. It contributes plurality to the noun as sketched in
(9) – by isolating a subset of the set (or set of sets) of all individuals that are ‘master’ – the latter
taken to be the denotation of the predicate ‘master’.1 Note that under the proposal in (9), –i
contributes only plurality to the noun it attaches to; in other words, the conventional nominative
plural in –i of the Latin II class is in reality a pure plural. We shall return to direct cases, i.e. the
conventional nominative and accusative in general, in section 1.3.

(9) a. domin [Q(Í)i]

b. $ x Q(Í) {master}
‘an x such that x is a subset of the set of individuals with the property ‘master’

Summing up so far, the main aim of this section was to introduce a conception of case as an
interpretable element, contributing to the construction of the LF of the sentence – as opposed to the
minimalist conception of case as an uninterpretable property, entering only the syntactic
computation as a reflex of agreement. This is most easily comprehended in relation to oblique cases
(essentially the inherent cases of Chomsky (1986)) – which is why we started from the oblique. At
the same time, the brief discussion of the descriptive syncretism between oblique (singular) -i and
plural (non-oblique) –i evokes a further important theoretical matter, involving the nature of the
lexicon. Two main frameworks are currently available for thinking about inflections and other
functional categories. Under the realizational view, associated in particular with Distributed
Morphology (Halle and Marantz 1993), functional nodes in the syntactic tree are represented by
clusters of abstract features; only after these are processed by the Morphological Structure
component, are actual ‘exponents’ inserted (Late Insertion). The classical lexicalist framework on
the other hand is projectionist, i.e. takes syntactic trees to be projected from lexical entries,
conceived as pairings of LF and PF properties. As far as we can tell, this latter view is held by

1
An anonymous reviewer notes that a singleton subset would satisfy the definition of plural in (9). S/he also notes that
an appropriate stipulation can be added to restrict the cardinality of the subsets. We prefer to leave the matter as is, since
it is evident that many plurals of natural languages can be satisfied by singletons. This is true for instance of generics
e.g. They are knocking at the door. (It’s Peter.) or I am peeling onions/an onion right now. It is also obviously true of
questions (How many came? Just one).
Chomsky (1995) – and implied by the minimalist principle of Inclusiveness. The projectionist view
is simpler, in the sense that it cuts out the Morphological Structure component altogether. At the
same time the argument of proponents of the realizational view is that Morphological Structure is
empirically motivated, among others by syncretisms. The discussion of Latin –i that precedes
suggests on the contrary that descriptive syncretisms can be dealt with under a projectionist view in
an advantageous way.
As before, our morphological treatment raises several issues. First, different nominal classes
(or declensions, in the terminology of traditional Latin grammars) display different syncretisms. For
instance, in (3) we have used a III class noun mulier and a II class one, dominus. In the III class, -i
is a dedicated dative singular (the genitive singular and the plural are formed by –s). In the II class,
we observe a genitive singular/nominative plural syncretism, as in (3b) and (9). In the pronominal
3rd person declension, –i is the dative singular and the plural (masculine), cf. illi (‘to him, they’). It
is only in the I class that the genitive/dative singular and the plural coincide on –i, at least if we
follow Halle and Vaux (1997) in assuming that forms like rosae ‘of/to the rose, roses’ are
underlying rosa-i. Suppose we associated Latin –i simply with the entry in (10a), reflecting the fact
that it can in principle be dative, genitive and plural, as in (10b). This is of course insufficient to
capture its varying values in the different classes.

(10) a. -i: Q(Í)

b. VP construal: ‘dative’
NP construal: ‘genitive’
N construal: ‘plural’

In fact, what we are looking for is simply a technical means by which the content of –i in
(10) can be contextually restricted to certain subclasses of N. With free lexical items/morphemes,
contextual restriction is achieved by means of selection. Therefore we assume that the ‘plural’
attachment of –i selects the subset of Ns conventionally known as II class as well the ill- base, as in
(11). The same mechanism can be used to exclude II class Ns from those that take –i in the dative
singular – and so on. Though the nominal inflection system of Albanian is somewhat simpler than
that of Latin, Manzini and Savoia (2012) argue that nothing more than selection is required to
account for the entire distribution of case endings in the various nominal classes.2

2
Since the Romance languages we concentrate on have three cases (or at best a residual comitative/instrumental fourth,
cf. section 1.2), the only oblique discussed here is genitive/dative. As is well-known, Latin also had an ablative
(instrumental etc.). In some nominal classes, –i can also take on this value. For some insights into the matter, we refer
(11) -i: Q(Í) à VP construal (‘dative’): selects for all N, except II class
N construal (‘plural’): selects for II class Ns, ill-
NP construal (‘genitive’): …………….

1.2 Romance pronouns with a three case system

An approach like the one sketched in the previous section treats case inflections, e.g. Latin –i in the
same terms as any contentive lexical entry of the language. It also allows us to speak of the
historical change they might have undergone, no differently than if we were speaking about any
other lexical category. In this section, we will review some Romance languages that preserve a
particularly rich case system, namely the three-way split (nominative, accusative, oblique)
reconstructed for proto-Romance by De Dardel and Gaeng (1992), Zamboni (1998). In sections 13-
1.5 we will turn to their analysis.
In the Romansh variety of Vella (Lumnezia Valley, Grisons), the 1st person singular pronoun
has a three case system, namely nominative, objective and oblique, as schematized in Table 1. By a
Obj we mean the form selected by the a ‘to’ preposition, i.e. the dative/oblique exemplified in
(12a). The form that is selected by other prepositions, notated P Obj in Table 1, is the same that
occurs as the object of a verb (Obj), as exemplified in (12b)-(12c); we identify it with the
accusative/objective. The 3rd person and the 1st/2nd person plural have a single form; the 2nd person
singular has two forms for nominative and oblique vs. accusative.

1sg 2sg 3sg 1pl 2pl 3pl

Nom jEu ti el/E:la nu:s vu:s els/E:las

Obj/P Obj mai tai

‘a’ Obj mi ti
Table 1. Full pronouns in Vella

(12) a. els dat@n a mi/ti/els

they give-3pl to me/you.sg/them
‘They give it to me/you/them’

the reader to the discussion of the residual ablative of Albanian (also syncretic with the oblique) in Manzini and Savoia
(2011a, 2012).
 b. els klOman mai/tai/els
they call-3pl me/you.sg/them
‘They call me/you/them’
c. els fan pEr mai/tai /els
they do-3pl for me/you.sg/them
‘They do it form me/you/them’
Vella

Similar data emerge in the pronominal systems of Southern Italian varieties, for instance
Sasso di Castalda (Lucania)3. This variety has a single pronominal form for 3rd person and for 1st
and 2nd plural. By contrast, 1st and 2nd person are associated with a three case system, again
nominative, objective (for the object of prepositions other than a) and oblique (for the object of the
a preposition), as schematized in Table 2. The processing of the relevant examples in (13) is slightly
complicated by the fact that Sasso is a Differential Object Marking (DOM) language (Aissen 2003
and references quoted there), where the definite, animate direct object is introduced by the
preposition a in (13a), exactly like the dative in (13b). Correspondingly the descriptive literature
speaks of a ‘prepositional accusative’ in (13a).

1sg 2sg 3ps 1pl 2pl 3pp

Nom ji tu idd«/edda nuj« vuj« l•r«

P Obj me te
‘a’ Obj mi ti
Table 2. Full pronouns in Sasso

(13) a. cam@n@ a mmi/tti/jidd@

call-3pl to me/you/him
‘They call me/you/him’
b. u rain@ a mmi/tti/jidd@
it give-3pl to me/you/him
‘They give it to me/you/him’

3
Limitations of space prevent us from providing more than one example. The Sasso system is found not only in
Southern Italy (Loporcaro 2008), but also in Sardinian varieties (Manzini and Savoia 2010). In Romanian, 3rd person
and 1st/2nd plural pronouns have a two case declension, like non-pronominal Ns. 1st/2nd person singular pronouns
however distinguish dative mie/ţie from accusative mine/tine and nominative eu/tu (a three case system again).
 c. l a ffatt@ p@ mme/tte/jidd@
it has done for me/you/him
‘S/he has done it for me/you/him’
Sasso di Castalda

Actually, Sasso displays a fourth case form – originating from the combination of the
pronoun and preposition cum ‘with’ - in contexts introduced by the preposition ‘with’, as in (14a),
and other selected prepositional contexts, as in (14b), configuring an even more complex case
system (cf. fn. 2).

(14) a. ku mmik@/ttik@
with me/you.sg
‘with me/you’
b. vi@n@ addo mmik@/ttik@
comes to me/to you.sg
‘S/he comes to me/you’
Sasso di Castalda

1.3 The three case system of Vella: nominative, objective, oblique

Let us review first the system of Vella in Table 1, which presents three cases in the 1st person
singular, though it doesn’t have the added complication of the prepositional accusative. In Vella, a
morphological oblique, i.e. m-i in the 1st singular, is embedded under the preposition a ‘to’, as in
(12a). Given the account of oblique case sketched in section 1.1, we conclude that the –i inflection
in Vella lexicalizes the Q(Í) elementary predicate. The a ‘to’ preposition can itself be construed as
introducing a (Í) property, i.e. as P(Í), which doubles the dative inflection Q(Í). The predicate (Í)
is the head of a predication PredP, denoting possession/inclusion, which takes m- as its internal
argument (the possessor) and the theme of the verb (pro) is its external argument (the possessum),
as schematized in (15) (cf. (4) in section 1.1).
(15) VP
wp
V PredP
dat@n wp
DP PP
pro 3
P(Í) DP
a 3
D Q(Í)
m i

We can furthermore ask whether the framework sketched in section 1.1 and applied in (15)
allows us to provide an account of the direct cases of Vella. ‘Accusative’ corresponds to the merger
of the lexical bases m-, t- (denoting speaker, hearer) with of the inflection –ai, endowed at least with
N (nominal class) properties. Suppose –ai has no further properties. In other words, while –i in
Latin or in Vella is endowed with predicative/operator content (Q(Í)), the –ai inflection of Vella is
a pure nominal inflection N. If so, the pronominal inflection does not contribute any means by
which to attach the pronoun to the sentence spine. Rather the pronoun is attached to the sentence via
ordinary lambda conversion – which the nominal inflection N is necessary and sufficient to satisfy.
Prepositions other than a behave like verbs in that their internal argument position is simply
satisfied by the N inflection.
In the 1st person ‘nominative’ is represented by a specialized lexical base (jEu). It is

reasonable to attribute to it again minimally, and perhaps maximally, nominal N properties, besides
speaker denotation. Nevertheless all-purpose satisfaction of argumental slots (‘accusative’) is
differentiated from satisfaction of the EPP environment (‘nominative’) – which may perhaps be
conceived as a sort of specialized lambda abstract, as suggested by Butler (2004)4.
In the present account, the asymmetry between structural and inherent case of Chomsky
(1986), is reconstructed as an asymmetry between two types of argumental attachment, either via
mere lambda conversion (direct case) or via the introduction of some specialized elementary
predicate (oblique). In a very simple case system like Vella’s, this asymmetry coincides with a
lexical opposition between inflections like –i which convey relational content (here Q(Í)) and
inflections which have mere phi-features content (here for instance nominal class, -ai). In a more
complex case system like Latin, the –i inflection has both a relational construal, whereby Q(Í) is

4
The 2nd singular differs from the 1st singular in presenting a t-i form syncretic between nominative and oblique. See
section 2.1 for a comparable syncretism in the 3rd person of Old Florentine (ell-i nominative and lu-i oblique singular).
read as a syntactic-level relation between two arguments (a predicate), hence as the oblique in (4)-
(5) – or as a word-level property, i.e. plurality, in (9).5
It is worth noting that the present set of assumptions is in principle compatible with a
minimalist rendition in terms of feature checking/ evaluation – at least for oblique case. For
instance, a functional head Appl (Pylkkänen 2008) can be added to (15), as in (16) – and the dative
mi can be construed as checking it (deleting an uninterpretable feature associated with it – or
valuing such a feature). However, under the view that case inflections have interpretive content,
abstract functional heads like Appl in (16) are redundant.6

(16) els [vP [ApplP (Í) [VP dat@n [P(Í)P a mi]

Vice versa, one may wonder whether present conceptions make any contributions towards
solving the empirical difficulties represented by case/agreement mismatches, briefly reviewed at the
outset. Recall that the crucial difficulty for Chomsky (2001, 2008) is that accusative case cannot be
construed as agreement with vP. In response to this difficulty, Baker and Vinokurova (2010) adopt
Marantz’s (1991) dependent case algorithm. This amounts to treating nominative as the Elsewhere
case, while accusative is the case assigned (checked) when there is at least another DP not assigned
inherent case locally. We suggested the opposite, namely that accusative is the Elsewhere case,
since it corresponds simply to the attachment of arguments via ordinary lambda conversion (see
also Adger and Ramchand (2005) for a feature theoretic translation of lambda notation).
Nominative by contrast corresponds to the satisfaction of the specialized EPP environment (perhaps
a specialized lambda abstract). Whether our approach can do away with the dependent case
algorithm remains to be verified.
5
Furthermore, though this is beyond the scope of this article, Latin –m in the conventional accusative singular is
obviously not a phi-feature; vice versa, dative singular –o of the Latin II class is the pure nominal class ending (as
argued by Halle and Vaux 1997). See Manzini and Savoia (2010) for some discussion.
An anonymous reviewer also raises the question of the selectional properties of prepositions. In present terms
prepositions are predicates, like verbs. Verbs differ as to whether they allow their arguments to attach via simple
lambda conversion (direct case) or rather they require the extra layer of structure corresponding to oblique. The same
can be assumed for prepositions (see Manzini and Savoia 2011a, 2012 on Albanian). Exactly like different structures of
embedding can result in a shifting of the interpretive value of the predicate, so the different type of embedding under a
preposition may configure interpretive shifts (motion vs. state etc.).
6
This is the low Appl of Pylkkänen, positioned inside the VP (technically between the vP and VP projections). The
high Appl (corresponding to experiencer and benefactive datives) occupies a predicate-external position; presumably, in
present terms, it can be mimicked by a VP-external (IP) attachment of NPs/PPs endowed with the Q(Í) property.
1.4 The nature of change

On the basis of the discussion of Latin and Romance case in sections 1.1 and 1.3, we can
brieflyconsider the issue of change. Within the framework of Distributed Morphology, Calabrese
(1998, 2008) proposes an account for the development from Latin to Romance case systems which
treats case loss in terms of the activation of constraints disallowing certain case feature
combinations. In a language which has all possible case oppositions, none of the case constraints
applies. Languages that do not have certain cases, or have no case, activate one or more restrictions
– or all restrictions. Therefore what happens on the way from Latin to Romance is the activation of
several case constraints. The corresponding feature clusters undergo repairs that are standard under
DM (Impoverishment etc.), resulting in the insertion of syncretic exponents.
In the present model, morphosyntactic structures are projected from fully specified lexical
items; there are no abstract feature clusters and no constraints of the kind postulated by Calabrese
and therefore the change from Latin to Romance cannot be a change in these constraints. The
question is whether there is an alternative. Under the assumption that the locus of variation (and
hence of change) is the lexicon, all that can change are case inflections. We can surmise that a
certain part of the Latin case inflectional system is dropped on the way to Romance (for instance
consonantal specialized endings such as -bus for oblique plural, -m for accusative and neuter
nominative, etc.). What survives, for instance -i, survives with the same basic properties it had in
Latin. In the discussion surrounding (10)-(11) we discussed the basic mechanism, i.e. selection,
whereby –i maintains the same core content, though its distribution varies according to nominal
class. In this perspective, the –i continued by Vella is specifically the dative of singular pronouns
(mihi ‘to me’). To take another example, in standard Italian, the –i inflection is dative on the gl-i ‘to
him’ clitic, or plural on the l-i ‘them’ clitic. Again it continues the –i of 3rd person pronouns, i.e. ill-i
as dative singular and nominative plural.
In this respect, historical change raises an important question for the Distributed
Morphology model, where underlying syntactic structure involving only abstract feature clusters are
lexicalized by exponents at the PF interface. We have seen that for Calabrese (2008) change in case
systems is change in underlying feature clusters. Nothing is said about the vocabulary of exponents
that runs parallel to the underlying abstract structure, providing an externalization for it– in line
with the conceptualization in Halle and Marantz (1993), Halle and Vaux (1997). In reality, the
assumption seems to be implicit in Calabrese’s account that the vocabulary remains constant. But
why would that be? Why can’t the abstract lexicon and the vocabulary of exponents vary each on its
own?
For instance, why wouldn’t –i be found in some Romance language as an accusative
singular (this would be a Romance language with an accusative singular clitic li)? Similarly,
consider –s, which for Halle and Vaux (1997) is just a default, i.e. an empty exponent. Why
couldn’t there be a language like French, except that –s marks some the accusative singular? The
whole of historical comparative grammar is predicated on the assumption that such ‘arbitrary’
changes are impossible – essentially rematching a possible LF with a possible PF, breaking with
previously attested possible LF values for the PF form. Our point is that the existence of a
conventional lexicon predicts this basic fact about change. However if there are in fact two separate
lexicons, one for abstract contents and one for PF exponents, radical rematchings of the type
described must be blocked through additional assumptions.

1.5 Differential Object Marking: the Sasso system

The language of Sasso in (13) presents a distribution very similar to that reviewed for Vella –
except that 1st and 2nd person pronouns are preceded by the preposition a not only as datives (i.e.
goals, possessors etc.), but also as themes. In this latter instance, the traditional literature speaks of
‘prepositional accusatives’. The prepositional accusative treatment extends to animate and definite
noun phrases, as in (17a), though not to inanimates and indefinites, as in (17b). Either animacy or
definiteness may suffice in some varieties to define an appropriate context for prepositional
accusatives, but at least one of these properties is needed (Suñer 1988 on Spanish, Manzini and
Savoia 2005 on Italian varieties).

(17) a. cam@n@ a ffrat-t@

call-3pl to brother-yours
‘They call your brother’
b. ann@ piLLat@ nu/ kwiru libbr@
have-3pl taken a/ that book
‘They have taken a/that book’
Sasso di Castalda
 Prepositional accusatives fall under a large family of linguistic behaviours, whereby case is
determined not only by the argument position that a DP fills but also by its intrinsic referential
content. These behaviours are often described in terms of alignment between case and ‘animacy’.
As Dixon (1979: 85-86) comments, “though the phenomenon is often referred to under the heading
of split ergativity, it is evident that in the typological continuum it touches what we may call split
accusativity”. Similarly, using a different terminology, Aissen (2003:473) states that “the factors
that favour differential subject marking will be the mirror image of those that favour DOM”
(Differential Object Marking).
Consider then the Sasso pronominal system in Table 2. The nominative forms consist of a
specialized lexical base for the 1st person and presumably also for the 2nd person. This specialized
lexical base corresponds to the satisfaction of the EPP (perhaps just a special lambda operator, as
suggested in section 1.3). The non-nominative forms, me/te and mi/ti, consist of the lexical bases
m- and t-, inflected by -e and -i respectively. The -e ending is just a nominal class N inflection,
which we take to be sufficient to satisfy the ordinary argument-of relation (lambda conversion).
This inflection, i.e. the descriptive accusative/objective, is seen in Sasso in (13c) with embedding
under prepositions other than a.
What is directly relevant here is the –i inflection, specialized for embedding under
the a preposition. Suppose this inflection is analysed as a lexicalization of Q(Í), corresponding to
the descriptive dative/oblique. This is unproblematic for the ditransitive structure in (13b) – but if
extended to (13a) it implies that prepositional accusatives are really datives/obliques in turn. In
section 1.1, we have seen that unergative verbs such as respondo ‘answer’ in Latin (6a) can be
paraphrased by a causative elementary predicate taking an eventive nominal as its complement (cf.
give an answer to). Following Hale and Keyser (1993), Chomsky (1995) the unergative predicate
result from the incorporation of the elementary state/event into the transitivizing predicate (CAUSE,
or v). It is then possible to argue that a Q(Í) complement takes as its internal arguments the noun to
which it attaches (the possessor) and as its external argument the elementary state/event (the
possessum) – as schematized in (6b) for unergatives. We assume more or less the same for the
transitive cam@n@ ‘they call’ in (13a), with the structure in (18). In other words, ‘call’ can be
thought of as a complex predicate consisting of a causative elementary predicate (v) and an eventive
nominal (cf. give a call to somebody). If so, the two arguments of Q(Í) in (18) are the 1st person
clitic to which it attaches and the event nominal (literally ‘They caused me (to have) a call’).7

7
The main problem for the approach we are proposing is represented by the fact that prepositional accusatives, like
other accusatives, can passivize – while datives cannot. This problem is discussed in Manzini (2012), where it is
proposed that the real difference is between Q(Í) operators selected by the verb (which block passivization) and Q(Í)
(18) vP
wp
v VP
wp
cam@n@ PP
wp
P(Í) DP
a 3
N Q(Í)
mm i

In the light of (18), the gist of the prepositional accusative/DOM phenomenon is that certain
types of referents (discourse participants, animates, definites) require to be embedded via the
elementary predicate Q(Í), and are incompatible with the embedding provided by the descriptive
accusative (corresponding to simple lambda conversion). In other words, they are associated only
with certain roles in the event: agent, possessor, but not theme. Inanimate/ indefinite complements
yield a canonical transitive event structure, comprising an agent and a theme, as in (17b). In (18) the
discourse participant must on the contrary be treated as a possessor, ‘zonally including’ the event of
calling.

1.6 The Person split

The treatment of DOM in the previous section allows us to ask a crucial question, left implicit so
far, namely why 1st/2nd person referents (speaker, hearer) are associated with a richer array of cases
than other (3rd person) referents in many Romance languages (e.g. Vella, Sasso). This appears now
to be yet another facet of the alignment of case with the participant/animacy/definiteness hierarchy.
We know that splits in case alignment can occur at different points in the descriptive
animacy hierarchy. In Sasso in (17) the prepositional accusative/DOM split is between definite or
animate DPs and others – as is most often the case in Romance languages (cf. Suñer 1988 on
Spanish, Manzini and Savoia 2005 on Italian varieties). On the other hand, Manzini and Savoia
(2005: §4.9) list Central Italian dialects where prepositional accusative/DOM only affects pronouns
(including 3rd person ones) to the exclusion of other referents (Avigliano Umbro, Canosa Sannita,

operators introduced only in order to embed a participant/animate/definite DP within the VP – which can be left out
under passive, where the DP is raised out of the VP.
Torricella Peligna) – and other dialects where the split is between 1st/2nd person referents and the
rest, as illustrated in (19) with Cagnano Amiterno (cf. Colledimacine, Borbona).

(19) a. camanu a mmi/a tti

they.call to me/to you
‘They call me/you’
b. camanu issu/issi/frate-tu
they.call him/them/brother-yours
‘They call him/them/your brother’
Cagnano Amiterno

Similarly, in Vella in (12) or in Sasso in (13), 1st/ 2nd person are cut off from other referents
by the fact that they are associated with case differentiations at all. 8 In other languages, case
inflections single out pronominal DPs (including 3rd persons) from non-pronominal ones, for
instance Old Florentine, to be considered immediately below in section 2.1. Other languages have
different case systems for definite and indefinite DPs (e.g. Romanian), and the indefinite set is
typically less differentiated; for instance in Albanian nominative and accusative may be
differentiated in the definite paradigm, but not in the indefinite one (Manzini and Savoia 2011a).
Descriptively, therefore, it is clear how the participant/animacy/definitness hierarchy works.
Its theoretical status is much less well defined. What is obvious is that 1st and 2nd person referents
(speaker and hearer) are directly anchored at the universe of discourse, while 3rd referents (and also
possibly 1st/2nd plural, which involve reference to ‘others’ besides the ‘speaker’ and ‘hearer’) are
not. Seen from this perspective, human referents are also a potential set of speakers and hearers –
i.e. of potential discourse-anchored participants. In such terms, the prominence of animates does not
involve their potential agentivity (pace Dixon 1979), but rather their referential saliency (cf.
DeLancey 1981). Definiteness and indefiniteness establish a different scale of referential saliency.
What DOM suggests is that the two scales are only partially independent in the underlying ontology
of natural languages.
At the same time, the problem posed by the data in section 1.2 is not only why 1st/2nd
singular referents split away from others, but also why they have the richest case alignment. The
answer is presumably similar to the one we have suggested for the prepositional accusative/DOM
pattern – namely that DPs higher in the referential scale may require more complex embedding

8
In fact since it is generally assumed that 1st person referents are more prominent on the ‘animacy’ hierarchy we are not
surprised to find that in Vella, the richest case system pertains to the 1st person.
structures, making them into possessor rather than simple themes. Similarly, we suggest that less
salient referents are able to satisfy any sentential attachment in virtue of their simple nominal class
inflection N (via lambda conversion). However DPs higher in the referential scale require a more
articulated structure of embedding, which reserves pure N inflections for themes (Vella) or even just
for prepositional objects (Sasso), while specialized lexicalization are required for EPP (nominative)
and possessor (dative) embedding.
When these considerations are projected long the temporal axis, what they amount to is that
the participant/definiteness/animacy system is more resistant to the loss of specialized case
alignments – and that discourse participants are the most resistant. From this perspective, the
change from Latin into Romance is not so much characterized by the loss of case as by the
alignment of case with some (highly restrictive) cuts on the animacy/definiteness hierarchy – so that
case is preserved only by participant pronouns.

2. From Old Florentine to modern Italian: Two case systems

In section 1 we have considered the survival of a three case system (nominative, objective, oblique)
in the 1st/2nd person pronouns of Romansh and Southern Italian varieties. In this section we survey
Old Italian (Florentine) pronouns, characterized by a two case system (nominative vs.
objective/oblique) in all persons. The two case organization still characterizes the 1st/2nd person
singular of modern Italian (e.g. io ‘I’ vs. me ‘me’). Some of it survives also in the 3rd person plural,
since we argue that the special distribution for loro depends on the fact that loro oblique properties,
rather than on its weak status, as argued by Cardinaletti and Starke (1999).

2.1 Old Florentine

In late XIII and early XIV century Florentine, 1st/ 2nd singular and the 3rd person full pronouns
differentiate nominative from objective/oblique (Castellani 2009, Egerland and Cardinaletti 2010),
though 1st/2nd person plural have a single form, noi, voi. This two case system, illustrated in Table
3, recalls that of Medieval Gallo-Romance nouns (Brunot and Bruneau 1969).

1sg 2sg 3sg.m 3sg.f 3pl.m 3pl.f

Nom io tu elli/egli ella elli/eglino elle
Obj/Obl me te lui lei loro loro
 Table 3. Full pronouns in Old Florentine

To exemplify, nominative elli in (20a) alternates with lui both as direct object in (20b) and
as the object of the a preposition in (20c). Comparable examples are provided for ella vs. lei in (21).

(20) a. Elli fece uno concilio di cxiij vescovi…

He made a council of 113 bishops
(Cronica fiorentina del XIII secolo, Schiaffini 1954: 86, 24)
b. … lo Imperadore.., sì cacciò lui e elesse un altro papa…
the emperor indeed chased him away and elected another Pope
(Cronica fiorentina del XIII secolo, Schiaffini 1954: 94, 24)
c. Onde picciolo guiderdone diedi a lui di così ricco insegnamento.
whence little recompense I gave to him for such a rich teaching
(Il Novellino, VIII, Lo Nigro 1968: 81)

(21) a. … ch’ella non fusse la diricta lancia con che Cristo fu fedito…
that it(f) wasn’t the true spear with which Christ was wounded
(Cronica fiorentina del XIII secolo, Schiaffini 1954: 91, 6-7)
b. … lo quale amava anche lei…
who loved also her
(Il Novellino, XCIX, Lo Nigro 1968: 209)

In the plural, the distribution of objective/oblique loro is slightly different from that of
lui/lei, since it can occur without preposition with dative interpretation, as in (22b); its use after
preposition remains of course possible, as in (22c). In (22a) we provide the comparison with the
nominative (masculine) elli. The possibility for lui/ lei to occur as datives without preposition,
quoted in the literature and taken up by Egerland and Cardinaletti (2010) seems to characterize only
a restricted number of texts, largely poetic or erudite ones, in particular by Dante (e.g. ond’io
risposi lei ‘whence I answered her’ (Purg. 33, 91)), and by Brunetto Latini (e.g. la persona che lui
semblava rea… ‘the person that seemed (to) him guilty’ (Rettorica, 197, 13)). The prose, practical
texts that we exemplify systematically present oblique lui/lei introduced by preposition.

(22) a. Et quasi elli fecero di nuovo un’altra Tavola Ritonda

and almost they made again another Round Table
 (Cronica fiorentina del XIII secolo, Schiaffini 1954: 93, 32-33)
b. … e fece loro simigliante proposta.
and he made them a similar proposal
(Il Novellino, LXXXI, Lo Nigro 1968: 78)
c. … onde mandiamo a loro …
whereby we send to them
(Il Novellino, LXXXI,, Lo Nigro 1968: 187)

Old Florentine 1st and 2nd person forms alternate according to the modern Italian usage, and
are not exemplified here. The nominative io ‘I’ and tu ‘you’ use specialized lexical bases, while the
objective results from the bases m-, t- followed by the nominal class morphology –e (me, te). As for
the 3rd person, The ell- base and the l- base denote definiteness – or in any event the complex set of
primitives clustering around the D category (Ramchand and Svenonius 2008). In the feminine
nominative the lexical base ell- is inflected simply by the nominal class N morphology –a for the
singular and –e for the plural. The objective forms l-oro, lu-i and le-i result from the merger of the
lexical base l- (eventually inflected for nominal class (lu-, le-) with the oblique –oro (plural) and –i
inflections. 9 The same –i inflection, also turns up in the nominative ell-i both in the singular and in
the plural.
The syncretism between oblique –i (lui, lei) and plural –i (elli ‘they’) has already been
discussed in connection with Latin (1)-(7); in fact it characterizes the Latin pronoun ill-i ‘to
him/her’, ‘they’. At the same time Old Florentine extends –i to nominative singular. Manzini and
Savoia (2010, 2011b) discuss this syncretism between oblique and nominative (including the
singular) for Latin –s, which in the III class lexicalizes genitive singular (e.g. urbi-s ‘of the city’)
and nominative singular and plural (e.g. urb-s ‘the city’; urbe-s ‘the cities’). If we follow Manzini
and Savoia, Latin -s is a generalized Q inflection, compatible with oblique and with plural, i.e.
Q(Í), but also with the EPP construed as a specialized lambda abstraction closing off the
argumental structure of the sentence (see section 1.3). The same would apply to –i in Old
Florentine.
A problem raised by the analysis of lui/lei as obliques is how these forms come to be found
in the descriptively accusative position, as in (20b). It is tempting to propose that this oblique

9
In the historical literature it is standardly accepted that lui/ lei have an ‘analogical’ origin based on dative forms of the
type e-i ‘to him/her’, cu-i ‘to which’, etc. (Rohlfs 1068 [1949]: 137). In Vulgar Latin (documented in inscriptions) we
indeed find datives illui, illaei (Väänänen 1971:219). In the literature it is also generally accepted that dative illi is
directly continued by the Italian dative clitic gli, and by the corresponding clitic forms of Old French li, etc.
marking is a reflex of the person/animacy/definiteness split conditions reviewed in section 1.6. In
other words, 3rd person pronouns, (involving the l- definiteness base also found on determiners), can
only be attached as EPP arguments (nominative) or as possessors (oblique). Thus it is possible that
a particular lexical base (here l- for definiteness/D) presents an alignment not found with other
denotations. This is what we expect to find if hierarchies are at best descriptive devices – while the
underlying reality of person, definiteness and animacy alignments are discrete categories such as
definiteness.
This aspect of the Old Florentine pronominal system also poses an interesting problem for
Calabrese’s (1998, 2008) analysis. Calabrese discusses the two case system of Old French
(nominative vs. objective), where the objective continues the Latin accusative. He derives Old
French from the assumption that the oblique case filter is activated; the repair of the underlying
feature cluster leads to realization of the oblique by the accusative morphology. However this
process depends on oblique being more marked than accusative, so that it is blocked first. The
markedness hierarchy cannot be reversed to yield a system like Old Florentine similar to Old
French in opposing nominative and objective, but where it is the morphological oblique that
survives, rather than the accusative.

2.2 Italian loro as a weak pronoun

Modern Italian, like Old Florentine, has io/me, tu/te contrasts in the 1st/ 2nd person (see Table 3).
However the 3rd person forms lui, lei, loro cover the entire spectrum of argumental positions,
including the nominative. In other words, 1st and 2nd singular display once again a more robust
association with case than 3rd person, as studied in section 1. We argue that an exception to this
state of affairs is represented by the 3rd person plural modern Italian loro, which maintains the
distribution seen in Old Florentine (22b), imputed here to oblique case. On the contrary,
Cardinaletti and Starke (1999), Cardinaletti (1998) argue that the special distribution of loro in
modern Italian is to be captured through the category weak pronoun. The same category is argued to
account for the distribution of lui/lei/ loro in Old Italian/ Florentine by Egerland and Cardinaletti
(2010). In particular weak loro would account for (22b), while its strong counterpart would be
responsible for the objective distribution in (22c), exactly as for their modern Italian counterparts.
Let us briefly review Cardinaletti and Starke’s (1999) evidence. They observe that loro
‘they/them/to them’ in modern Italian has two different distributions. In one distribution, it fills the
same positions as any ordinary noun phrase, patterning together with lui ‘he/him’ and lei ‘she/her’
as in (23b). In the other distribution, loro is associated with a Dative Shift position which is
unavailable to lui/lei, as in (23a) – and similarly for the genitive position in (23c) (Cardinaletti
1998). Modern Italian (23) closely parallels the examples from XIII-XIV century Florentine prose
texts in (20)-(22).

(23) a. Ho offerto loro/*lui/*lei il mio aiuto

I.have offered them/him/her my help
b. Ho offerto il mio aiuto a lui/ lei/ loro
I.have offered my help to him/her/them
c. Il loro/*lui/*lei libro
The their/his/her book

From the distributional facts in (23), Cardinaletti and Starke conclude that there are two loro
– namely a strong loro and a weak loro. They support this categorization by correlating the
distributions observed with independent criteria, also adopted by Egerland and Cardinaletti (2010:
416). First, they argue that strong loro is interpreted as human, while weak loro (the Dative Shift/
genitive one) can have any reference. However the intuitions of the speakers we consulted are that
loro generally admits of inanimate reference also in the strong distribution. This is supported by
corpus data like (24) from the national newspaper La Repubblica (Baroni et als. 2004).10

(24) a. Consideriamo, per un attimo, l'automobile […] Eccola lì. Anzi, eccole lì […] proprio
Consider, for a moment, the car … There it is. Or better, there they are … just
nel momento in cui l' uomo non può più servirsi di loro […] per qualche giorno,
at the time when a man can no longer use (of) them … for a few days
pensa che, ecco, in città si può, forse, vivere senza di loro
he thinks that, yes, in the city one can, perhaps, live without (of) them
b. Quando poi […] arriva a dipingere il fondo, tutto si fa indistinto, sciolto
When next…he gets to painting the background, all becomes indistinct, melting
nella luce […] Sarà la fila lunga delle colline che si fanno tutte rosa sotto il pallido
into the light … It may be the long line of hills that become all pink under the pale
azzurro del cielo sopra di loro
blue of the sky above (of) them

10
This is not a corpus study, and we interrogated the database in a completely unsophisticated way, simply asking for
expressions which ought to be acceptable in the relevant readings, here “senza di loro”, “sopra di loro”.
 Furthermore, for Cardinaletti and Starke, weak pronouns cannot be coordinated. It is
difficult to have an intuition about examples of coordination for weak loro, since they involve
coordination of loro with itself in the Object Shift position, as in (25). We therefore suggest that
judgement should be suspended on these examples and that coordination should be tested instead on
the genitive, where loro can in principle be coordinated with any possessive pronoun. In the
judgements we collected and in the corpus data from La Repubblica in (26) this coordination yields
wellformed results. 11

(25) ¹Ho dato loro e loro tutti i miei soldi

I.have given them and them all the my money
‘I gave all my money to them and them’
(26) a. (com' era) malsano quel vento dell'Est che, loro, respirarono gioiosamente a pieni
How unhealthy that wind from the East was that they breathed joyously with full
polmoni. E che oggi, per fortuna loro e nostra, non soffia più.
lungs. And that today, for (ggod) fortune their and ours, no longer blows.
b. Santagata e Morganti, che negli anni hanno […] ricondotto alla loro e nostra
S. and M., who throughout the years have brought back to their and our
quotidianità, […] anche i mondi degli autori volta a volta visitati: Dostoevskji […]
daily life, also the worlds of the authors in turn visited: D…

Next, in Cardinaletti and Starke’s judgement weak loro cannot be modified by adverbs,
specifically by only and also. Data like (27) are instead acceptable for our speakers, and they are
indeed sourced from the La Repubblica corpus. Note that even if the postnominal position is
involved for the possessive in (27b), the weak form loro seems to be used, not the strong form
preceded by the ‘of’ preposition. If vice versa postnominal loro is claimed to be strong, this means
that the absence of the prepositional layer is no longer a predictor of weak status – undermining a
different generalization.

(27) a. La diagnosi ha dato anche loro la certezza che erano sane pure le loro figlie
the diagnosis has given also them the certainty that were healthy their daughters too

11
An anonymous reviewer questions the relevance of expressions like per fortuna loro in (26a), given that they seem to
have special properties such as the lack of a definite determiner. However (26b) presents none of these special
properties.
 b. al Marsiglia mancavano cinque giocatori per squalifica (colpa anche loro).
Marseille lacked five players because of disqualification (fault also theirs)

Cardinaletti and Starke’s idea is that weak pronouns are structurally smaller than strong
pronouns, though they are bigger than clitics. Specifically, clitics are IP-like constituents. Weak
pronouns correspond to a projection ΣP (in the sense of Laka (1990)), which contributes prosodic
properties to them. Strong pronouns have a CP-like structure, where the preposition that introduces
them (e.g. a ‘to’ in a loro ‘to them’) is assimilated to a C head. However, introducing a C layer or a
Σ layer in a sentence implies introducing LF-relevant properties. Therefore introducing such a layer
in the structure of a pronoun ought to yield LF-relevant distinctions between weak pronouns, strong
pronouns and clitics. In reality, they all refer in the same way (i.e. deictically, anaphorically and as
bound variables). Morphology concurs with semantics in supporting a similar structuring for all 3rd
person pronouns in Romance. For instance, supposedly weak and strong loro are morphologically
identical. More to the point, even clitics are at least as complex as full pronouns, corresponding to
the merger of two separate morphemes, namely an l- base, introducing definite reference, and
inflectional endings introducing nominal class and case. The only way to avoid the obvious
conclusion that full pronouns and clitics are equal in size (internal constituency) is to embrace a
realizational model of the lexicon (see section 1.4), as Cardinaletti and Starke do.
In the next section we will propose that loro is a just full pronoun (like lui or lei) except that
its special oblique case properties allow it to occur in the Dative Shift and possessor positions not
available to lui/lei in (23). In other words, loro provides no evidence for the strong vs. weak
categorization. The category clitic is not questioned here. For it, we adopt the standardly accepted
definition suggested by Sportiche’s (1996) analysis – namely that clitic pronouns correspond to
specialized functional heads on the sentential spine.

2.3 Italian loro as an oblique

Briefly, Cardinaletti and Starke (1999) account for the distribution of loro in (23) on the basis of the
assumption that a loro is a strong pronoun and a-less loro a weak pronoun. Being strong, a loro has
a CP-like layer introduced by the preposition a, which is characterized by case properties. Weak
loro lacks this layer and therefore must occur in a position where it can get case via agreement,
namely a [Spec, Agr] position, identified with the Dative Shift position. Here we argue on the
contrary that Dative Shift loro has case, namely oblique case. Conversely, other occurrences of loro
depend on lack of oblique case.
Consider first loro with the ordinary DP distribution (no Dative Shift), as in (23b).
Following Kayne (1984), and as discussed above in connection with Latin (4) and Vella’s (15), the
complement of a ditransitive verb like ‘offer’ is a predication denoting possession. In present terms,
a is the predicate denoting possession, P(Í), taking loro as its internal argument (the possessor) and
the theme of the verb is its external argument (the possessum), as in (28). Since the Í relation is
introduced by P, no oblique Q(Í) property is required on loro.12

(28) VP
wp
V PredP
offerto wp
DP PP
3 wp
il mio aiuto P(Í) DP
a loro

Consider then Dative Shift loro, as in (23a). The present hypothesis is that this position
depends on loro being associated with an inflectional oblique. In other words, the –oro ending is a
lexicalization of the Q(Í) relation taking as its internal argument the l- pronominal base to which it
attaches and as its external argument the theme of the verb. The resulting surface constituency is as
in (29). We will return to whether (29) is a base structure (hence a VP) or a derived structure below.

(29)
wp
V wp
offerto DP DP
3 3
D Q(Í) il mio aiuto
l oro

The alternation between genitive loro structures, and structures where the possessor is
introduced by the di ‘of’ preposition, reproduces the alternation in (28)-(29) – as schematized in
(30). The di preposition with P(Í) content in (30a) takes the possessor as its object to the right and

12
The representation in (28) is of course simplified. Non-oblique loro has the same internal structure as oblique loro in
(29) below. Following the discussion of Latin and Old Florentine –i, furthermore, the plural reading of loro depends
simply on a different construal of the Q(Í) property associated with the –oro inflection.
the possessee as its subject to the left. Genitive loro yields the inverse order of possessor and
possessee, as in (30b).

(30) a. DP
wp
D NP
il wp
N PP
libro 3
P(Í) DP
di loro

b. DP
wp
D wp
il DP N
3 libro
D Q(Í)
l oro

In general, the possessor to the right implies a preposition; the possessor to the left implies
some oblique case properties. Right-left reordering of arguments obviously invites a treatment in
terms of movement. The question is whether movement is actually involved (leaving an interpreted
variable in the extraction site) or what are observed are simply two different linearizations for the
same basic merger in PredP. Here we leave the question open, noting however that the present
model is entirely neutral with respect to it. In other words, adoption of the present proposal does not
interfere with further theoretical choices. Specifically, in section 1.3, we indicated that nothing
prevents us from assuming that there is a dedicated functional head corresponding to the (Í)
content of oblique case inflections, presumably to be identified with the low Appl head of
Pylkkänen (2008). In terms of a (Í) functional head, we can model the right-left reorderings in
(28)-(39) by movement, along the lines (31). The idea is that the case properties of loro in (29) and
(30b) require it to be positioned in the Q(Í) position, so that the relevant structures are to be refined
as in (31a) and (31b) respectively.

(31) a. … offerto [(Í) loro [VP offerto [il mio aiuto loro]
b. [DP il [(Í) loro [NP libro loro]

Interestingly, (31a) closely matches Cardinaletti and Starke’s schema of derivation for
Dative Shift loro (moved leftward to an AGR functional head). However, for Cardinaletti and
Starke, loro moves to a functional position because it is smaller than a loro; this is the essence of
the strong/weak distinction. In the present approach, loro and a loro have the same properties,
though differently lexicalized, by Q(Í) and P(Í) respectively. The Dative Shift/possessive loro is
accounted for on the basis of the category Q(Í) (oblique case). However loro can also have the
same non-oblique distribution as lui/ lei. Therefore we assume that the –oro inflection can carry
both plural and oblique properties, or just plural properties, as in (32). This correctly yields the
alternation between oblique loro in (29) and (30b) and non-oblique loro in (28) and (30a).13

(32) l-: definite

-oro: plural, (oblique)

In short, Q(Í) oblique case is sufficient to predict the distribution of Dative Shift/possessive
loro.14 Vice versa, at least for modern Italian pronouns, the notions of weak and strong pronoun are
at best redundant. Note that our stance here does not consist in denying that there may be several
pronominal series in the languages we are considering. Descriptively, loro cannot be entirely
reduced either to other full pronouns (which do not share the Dative Shift/possessive distribution) –
nor to clitics. What we are calling into question is that this has anything to do with the weak
pronoun category of the theoretical literature, as opposed to independently needed categories (here
oblique case).
Going back to Old Florentine, we can now analyze examples like (22b) with the Dative Shift
distribution of loro in the same way as their modern Italian counterpart (29). Evidence for a Dative
Shift distribution of lui/lei is however restricted to a few authors and there is no evidence for it in
the prose practical texts exemplified in (20)-(22). Therefore we would have to assume that the –i
inflection of lui/lei no longer has an oblique value, even at this stage of the development of the
language. We of course assume the same to hold of modern Italian.

13
Technically, oblique and plural in (32) are just two different construals of the Q(Í) predicate.
14
One further distributional piece of data concerning dative loro has not been discussed here, namely the fact that it can
be found between the auxiliary and the participle, as in (i).
(i) Ho loro promesso il mio aiuto
I have them promised my help
‘I promised my help to them’
This positioning of loro is an independent issue. For instance it could be dealt with by assuming that the Dative
Shift position can be higher than the participle position, given a movement analysis of the type in (31).
2.4 Concluding remarks: the form of the lexicon.

Theoretically, the raison d’être of a category like that of weak pronoun is upholding a certain
conception of the organization of the lexicon, hence of grammar. Cardinaletti and Starke (1999)
motivate it on the basis of a classical criterion, crossing morphology and distribution. In general,
given a morphology M specialized for distribution D, one says that M + D individuate category C.
For instance the morphology loro (M) with Dative Shift distribution (D), as in (23a) individuates
the category weak pronoun (M+D). This differs both from the form (a) loro with ordinary (P) DP
distribution (strong pronoun) and from the form gli with clitic distribution.
If such a categorization is generalized, it yields essentially the same paradigms as in a
descriptive or normative grammar, with a proliferation of syncretisms and homophonies. What
appears to be important in such lexicons is the underlying regularity of abstract categories; the
amount of opacity present at the PF interface (in the form of homophony or
syncretism/neutralization) is irrelevant. In other words, a (near) invariant syntactic-semantic
structure combines with (near) arbitrary variation at the PF interface, leading to the adoption of
realizational models of the lexicon, as opposed to projection from the lexicon to the syntax.
For instance, consider Romansh varieties, where non-clitic pronouns have both a distribution
available to Italian lui/lei, for instance in left dislocated position, as in (33b) – and a distribution
unavailable to lui/lei (and non-oblique loro) in Italian, for instance in V-adjacent position, as in
(33a). In such a language the strong-weak categorization would lead to the postulation of two
completely homophonous series of pronouns, one with the ‘strong’ distribution in (33b) and the
other with the ‘weak’ distribution in (33a). This obscures the fact that forms like el/Ela ‘he/she’
have exactly the same overall distribution as any lexical DP in the language (e.g. a name like John).
Similarly, Egerland and Cardinaletti (2010) classify the occurrences of non-clitic pronouns in Old
Florentine according to the categories weak and strong (or “free”) pronouns. For instance an
occurrence like (21b) would be ‘strong’, while an occurrence like (22a) would be weak
(incompatible with lui/lei/loro in modern Italian). It is immaterial to them that the strong and weak
series are lexically identical.

(33) a. els klOman mai/tai/el/Ela

they call me/you/him/her
‘They call me/you/him/her’
b. mai jEu diErm@l
me, I sleep
 ‘(As for) me, I sleep’
Vella

In this section, we have argued that Italian loro can be accounted for (without loss of
empirical adequacy or theoretical generality) by ignoring such abstract schemas of organization as
the strong vs. weak opposition. Under the lexicalist conception of the architecture of grammar that
we adopt, the mapping between LF content and PF content, with its potential for variation, is
carried out by the lexicon – and the computational component operates on lexical items and not on
abstract properties. Under such a view, nothing leads us to expect that categories are represented
uniformly throughout a given language – or across languages. Thus loro is best accounted for as an
optional survivial of oblique case.15
More generally, in this article we have found considerable evidence as to the survival of a
case system into Romance, specifically in the full pronouns system. As a side result of our main line
of investigation, we have been led to doubt that Romance languages have a weak pronouns series,
besides full and clitic ones. 16

References

Adger, David & Gillian Ramchand. 2005. Merge and Move: Wh-dependencies revisited. Linguistic
Inquiry 36: 161-193.
Aissen, Judith. 2003. Differential object marking: iconicity vs. economy. Natural Language &
Linguistic Theory 21: 435-483.
Baker, Mark & Natasha Vinokurova. 2010. Two Modalities of Case Assignment: Case in Sakha.
Natural Language and Linguistic Theory 28: 593-642

15
The restricted distribution of lui/lei/loro in Italian in direct case positions (nominative, accusative) is explained by
Cardinaletti and Starke (1999) in terms of an Economy of representations principle, which they formulate simply as
“Minimize structure (up to crash)”. By it, cliticization/null pronouns are obligatory if possible. The issue whether there
are bona fide optimization processes in grammar is again too complex to be addressed here. In any event, the
complementary distribution (if real) is between clitics/null pronouns and full pronouns and does not require the category
weak pronoun.
16
An anonymous reviewer reminds us that the discussion of Romance and other languages by Cardinaletti and Starke
(1999) is based on the notion of weak pronoun originally formulated by Holmberg (1986) for Germanic languages.
Whether such a category can be given theoretical and empirical coherence in Germanic languages (and what its relation
is with the category clitic of Romance) is outside the scope of the present work.
Baroni, Marco, Silvia Bernardini, Federica Comastri, Lorenzo Piccioni, Alessandra Volpi, Guy
Aston & Marco Mazzoleni. 2004. Introducing the "la Repubblica" corpus: A large,
annotated, TEI(XML)-compliant corpus of newspaper Italian. In Proceedings of LREC
2004, 1771-1774. Lisbon: ELDA.
Belvin, Robert & Marcel Den Dikken. 1997. There, happens, to, be, have. Lingua 101: 151-183.
Brunot, Ferdinand & Charles Bruneau 1969. Précis de grammaire historique de la langue
française. Paris: Masson et Cie.
Butler, Johnnie. 2004. On having arguments and agreeing: semantic EPP. York Papers in
Linguistics Series 2 1: 1-27
Calabrese, Andrea. 1998. Some remarks on the Latin case system and its development in Romance.
In José Lema & Esthela Trevino (eds.), Theoretical Advances on Romance Languages, 71-
126. Amsterdam: John Benjamins.
Calabrese, Andrea. 2008. On Absolute and Contextual Syncretism. In Andrew Nevins & Asaf
Bachrach (eds.), The bases of inflectional identity,156-205. Oxford: Oxford University
Press.
Cardinaletti, Anna. 1998. On the deficient/ strong opposition in possessive systems. University of
Venice Working Paper in Linguistics 8: 65-111
Cardinaletti, Anna. & Michal Starke 1999. The typology of structural deficiency: A case study of
three classes of pronouns. In Henk van Riemsdijk (ed.), Clitics in the Languages of Europe,
145-233. Berlin: Mouton de Gruyter.
Castellani, Arrigo. 2009. Nuovi saggi di linguistica e filologia italiana e romanza, 1976-2004.
Roma: Salerno.
Cennamo, Michela. 2011. Impersonal constructions and accusative subjects in Late Latin. In Andrej
Malchukov & Anna Siewierska (eds.), Impersonal Constructions, 169-188. Amsterdam:
John Benjamins.
Chomsky, Noam. 1986. Knowledge of Language. Its nature, Origin, and Use. Praeger, New York.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge Mass.: The MIT Press.
Chomsky, Noam. 2001. Derivation by phase. In Michael Kenstowicz (ed.), Ken Hale: A life in
language, 1-52. Cambridge, Mass: The MIT Press.
Chomsky, Noam. 2008. On Phases. In Robert Freidin, Carlos Otero & Maria-Luisa Zubizarreta
(eds.), Foundational Issues in Linguistic Theory. Essays in Honor of Jean-Roger Vergnaud,
133-166, Cambridge, Mass: The MIT Press.
De Dardel, Robert & Paul Gaeng. 1992. La déclinaison nominale du latin non classique: essai d’une
méthode de synthèse. Probus 4: 91-125.
DeLancey, Scott. 1981. An interpretation of split ergativity and related patterns. Language 57: 626-
657.
Dixon, Robert M. W. 1979. Ergativity. Language 55: 59-138.
Egerland, Verner & Anna Cardinaletti. 2010. I pronomi personali e riflessivi. In Giampaolo Salvi &
Lorenzo Renzi (eds.), Grammatica dell’italiano antico, vol. 1, 401-467. Bologna: Il Mulino.
Hale, Ken & Samuel J. Keyser. 1993. On argument structure and the lexical expression of syntactic
relations. In Ken Hale & Samuel J. Keyser (eds.), The view from Building 20, 53-108.
Cambridge, MA: The MIT Press.
Halle, Morris & Marantz, Alec. 1993. Distributed morphology and the pieces of inflection. In Ken
Hale & Samuel J. Keyser (eds.), The view from Building 20, 111-176. Cambridge, Mass.:
The MIT Press.
Halle, Morris & Bert Vaux. 1997. Theoretical aspects of Indo-European nominal morphology: The
nominal declension of Latin and Armenian. In Jay Jasanoff, Craig Melchert & Lisi Olivier
(eds.), Mir Curad. A Festschrift in honor of Calvert Watkins, Universität Innsbruck.
Holmberg, Anders. 1986. Word order and syntactic features in the Scandinavian languages and
English. Ph. D. Dissertation,University of Stockholm.
Iatridou, Sabine. 1993. On nominative Case and a few related things. Papers on case and
Agreement II. MIT Working Papers in Linguistics, 19: 175-196
Jackendoff, Ray. 2002. Foundations of Language. Oxford, Oxford University Press
Johnston, Jason. 1997. Systematic homonymy and the structure of morphological categories: some
lessons from paradigm geometry. Ph. D. Dissertation, University of Sydney.
Kayne, Richard. 1984. Connectedness and binary branching. Foris, Dordrecht.
Laka, Itziar. 1990. Negation in Syntax: On the Nature of Functional Categories and Projections.
Ph.D. Dissertation, MIT, Cambridge, Mass.
Lo Nigro, Sebastiano (ed.). 1968. Novellino e Conti del Duecento, Torino: Utet.
Loporcaro, Michele. 2008. Opposizioni di caso nel pronome personale: i dialetti del mezzogiorno in
prospettiva romanza. In Alessandro De Angelis (ed.), I dialetti meridionali tra arcaismo e
interferenza, 207-235. Palermo: Centro di Studi filologici e linguistici siciliani.
Manzini, M. Rita. 2012. From Romance clitics to case: Split accusativity and the Person Case
Constraint. In Irene Franco, Sara Lusini & Andrés Saab (eds.), Romance languages and
linguistics 2010. Selected papers from ‘Going Romance’ Leiden 2010, 1-20. Amsterdam:
John Benjamins.
Manzini, M. Rita & Leonardo M. Savoia. 2005. I dialetti italiani e romanci. Morfosintassi
generativa. Alessandria: Edizioni dell’Orso, 3 vols.
Manzini, M. Rita & Leonardo M. Savoia. 2007. A unification of morphology and symtax. Studies in
Romance and Albanian varieties, London, Routledge.
Manzini, M. Rita & Leonardo M. Savoia. 2010. Case as denotation: variation in Romance. Studi
Italiani di Linguistica Teorica e Applicata, XXXIX: 409-438.
Manzini, M. Rita & Leonardo M. Savoia. 2011a. Grammatical categories: variation in Romance
languages. Cambridge, Cambridge University Press.
Manzini, M. Rita & Leonardo M. Savoia. 2011b. (Definite) denotation and case in Romance. In
Janine Berns, Haike Jacobs and Tobias Scheer (eds.), Selected papers from ‘Going
Romance’ Nice 2009, 149-165. Amsterdam: Benjamins.
Manzini, M. Rita & Leonardo M. Savoia. 2012. ‘Case’ categories in the Geg Albanian variety of
Shkodër. Res Albanicae 1: 23-42.
Marantz, Alec. 2000 [1991]. Case and licensing. In Eric Reuland (ed.), Arguments and Case:
Explaining Burzio’s Generalization, 11–30. Amsterdam: John Benjamins.
Plank, Frans. 1985. The extended accusative/restricted nominative in perspective. In Frans Plank
(ed.), Relational Typology, 269-310. Berlin: Mouton de Gruyter.
Pylkkänen, Liina. 2008. Introducing arguments. Cambridge, MA: MIT Press.
Ramchand, Gillian & Peter Svenonius. 2008. Mapping a Parochial Lexicon onto a Universal
Semantics. In Theresa Biberauer (ed.), The Limits of Syntactic Variation, 219-245.
Amsterdam: John Benjamins.
Rohlfs, Gerhard. 1968 [1949]. Grammatica storica della lingua italiana e dei suoi dialetti.
Morfologia. Torino: Einaudi.
Schiaffini, Alfredo. 1954. Testi fiorentini del dugento e dei primi del trecento. Firenze: Sansoni.
Sportiche, Dominique. 1996. Clitic constructions. In Johan Rooryck & Laurie Zaring (eds.), Phrase
structure and the lexicon, 213-276. Dordrecht: Kluwer.
Suñer, Margarita. 1988. The role of agreement in clitic doubled constructions. Natural Language
and Linguistic Theory 6: 391-434.
Väänänen, Veikko. 1971. Introduzione al latino volgare, Bologna: Pàtron.
Zamboni, Alberto. 1998. Cambiamento di lingua o cambiamento di sistema? Per un bilancio
cronologico della transizione. In József Herman (ed.), La transizione dal latino alle lingue
romanze, 99-127. Tübingen: Niemeyer.