Rivista di Grammatica Generativa - Lear

RIVISTA DI GRAMMATICA GENERATIVA
Volume 34, anno 2009
Valentina Bianchi A note on backward anaphora p. 3
Valentina Brunetto Syntactic diagnostics for referentiality
marking in early null objects: evidence
from Italian p. 35
Stephanie
Durrleman &
Sandrine Zufferey
The nature of syntactic impairment
in autism p. 57
Chi-Fung Lam A comparison between the Cantonese
aspect markers gan and haidou p. 87
Lidia Lonzi Caratteri della Cliticizzazione di ne p. 115
Sandra Paoli Contrastiveness and New Information:
a new view on Focus p. 137

1. Introduction
Rivista di Grammatica Generativa
34 – 2009
A NOTE ON BACKWARD ANAPHORA 1
Valentina Bianchi
In generative approaches to anaphora, a persistent view is the distinction
between binding and coreference, dating back to Reinhart (1983, 1986). 2 True
binding is syntactically constrained by a c–command requirement (Reinhart’s Bound
Anaphora Condition), and results in the binder’s argument position and the bound
pronoun being interpreted as two occurrences of the same variable bound by a
lambda operator:
(1) a. John1 loves his1 wife
b. John [λx. loves (x, x’s wife)]
On the most restrictive assumptions (see e.g. Heim & Kratzer 1998: 260–266),
semantic binding as in (1b) strictly requires syntactic binding as in (1a), and vice
versa, syntactic binding is invariably interpreted as semantic binding.
On the other hand, the kind of anaphora that obtains in non–c–command
configurations is “accidental coreference”: the unbound pronoun is a free variable
which is assigned a value by the context; the assigned value may happen to be the
1 The material in this paper was presented at the main session of the 30th GLOW Colloquium
(Tromsoe, April 12-14th 2007), at the “Workshop on Romance-Japanese: Comparative syntax
and language acquisition” (Siena, May 4-5th 2007), and in a seminar held at the Scuola
Normale Superiore, Pisa (19 October 2007). I thank the audiences at those event for various
important comments and suggestions. Thanks also to Mara Frascarelli, Ken Safir and Philippe
Schlenker for making available to me some yet unpublished papers of theirs. This work was
inspired by discussions on Weak Crossover with Michelangelo Falco, and especially by joint
work with Cristiano Chesi on his top-down minimalist grammar (Chesi 2004, 2007); my
warmest thanks to both of them. I am entirely responsible for any remaining errors.
2 See Heim (1998), Büring (2005) for general discussion. See Büring (2007) for a recent
overview of truth–conditional approaches to pronoun interpretation. An accurate and
comprehensive presentation of the empirical domain can be found in Huang (2000).
3

Valentina Bianchi
referent of some other noun phrase (presumably made salient by the linguistic
context). 3
(2) John is unhappy. [His wife] has left him.
↓ ↓
j j
The main reason for maintaining this distinction is the (not uncontroversial) 4
empirical observation that quantificational antecedents are sensitive to the c–
command requirement: a quantifier can semantically bind a pronoun in its c–
command domain (3), but not one outside its c–command domain (4).
(3) No man1 loves his1 wife (as much as she deserves).
(4) * No man1 is happy. His1 wife has left him1.
Reinhart’s original view, also reiterated in much subsequent work (e.g.
Grodzinsky & Reinhart 1993, Reinhart 2000, 2004) is that accidental coreference as
in (2) is a pragmatic phenomenon: it is not syntactically constrained, and it falls
outside the domain of semantics proper. 5
Nowadays, various discourse-oriented approaches to anaphora have been
developed, notably Grosz et al.’s (1983, 1995) Centering Theory and Ariel’s (1990)
Accessibility Theory, and also some pragmatic neo-gricean approaches (Levinson
1991, Huang 2000; see Huang 2007, 245-274 for a recent overview). The former are
generally more concerned with the problem of inter-sentential anaphora and
discourse topic management; the neo-gricean approaches seem to be more
concerned with binding configurations. All in all, there has been relatively little
interest in intra-sentential anaphora without c–command, also generally known as
“backward anaphora”: 6
3 Here I gloss over the problem of coindexing: see Heim (1998) and Reinhart (2000) for
relevant discussion. Even if the proper name and the pronoun in (2) bore the same index, this
would not be a true binding configuration for lack of c–command.
4 The c–command requirement is widely assumed, leaving aside E–type pronouns as a
separate category (Heim 1990, Elbourne 2003, Büring 2004 among others). Two important
alternatives are the Independence Principle (Safir 2004) and the Leftness Condition (see Shan
& Barker 2006, Barker & Shan 2007): see note 10 below.
5 The impossibility of coreference in c–command configurations of “Principle C violation” is
accounted for by Reinhart’s Rule I, to which I return below.
6 This is not true of psycholinguistic and acquisition research: see, among many others,
Gordon & Hendrick (1997), Kazanina (2005), Reuland & Avrutin (2005), Sorace & Filiaci
(2006), Belletti et al. (2007). Note that the backward cases are the only cases where the lack
of c–command is uncontroversial; for virtually all the forward cases, it is possible to propose
a phrase structure analysis that satisfies the Bound Anaphora Condition. I refer the reader to
4

(5) Those who know her respect Zelda (*every girl).
A note on backward anaphora
In this paper I would like to bring together four simple empirical observations
about backward anaphora, and to critically re-examine Reinhart’s view of this
phenomenon. I will suggest that backward anaphora is constrained by the interplay
of two factors: the placement of focus, on the one hand, and the intrinsic content of
the co-referring expressions, on the other; furthermore, I will argue that a proper
understanding of the phenomenon requires us to assume an incremental left-to-right
interpretation process.
2. Four empirical observations
The first empirical observation is actually fairly old and well known: backward
anaphora between a pronoun and a following R–expression is blocked when the R–
expression bears focus (cf., among others, Chomsky 1976: 343-45, Williams 1997:
587-91, Erteschik–Shir 1997: 78 ff.).
Here I will only consider instances of new information focus (focus in questionanswer
pairs), leaving open the question of whether other kinds of focus (e.g.
contrastive focus) have a similar or a different effect. 7 The relevance of new
information focus is illustrated by the contrast between (6) and (7): 8 in (6), the R–
expression is focussed and the preceding pronoun cannot corefer with it; in (7), on
the contrary, the R–expression is destressed – due to contrastive focus on the
preceding verb – and backward anaphora is possible.
(6) (Context: As for John, who does his wife really love?)
?* His wife loves JOHN.
(7) (Context: –As for John, I believe his wife hates him. –You’re wrong:)
His wife LOVES John.
Although this observation is generally accepted in the literature, its scope is
actually very restricted. When the two co-referring terms are both pronominal, focus
on the second one does not block coreference. This holds not only for first and
second person pronouns, but also for third person ones: (8c) forms a minimal pair
with (6).
Jackendoff (1990), Pesetsky (1995) and Bianchi (2001) for discussion (this issue is no longer
so hot nowadays as it used to be in the Nineties).
7
Chomsky (1976: 344) provided the following examples, with no clear indication as to the
kind of focus involved:
(i) The woman he loved BETRAYED John.
(ii) * The woman he loved betrayed JOHN.
8 In the following examples, italics indicate coreference, small caps indicates stress.
5

Valentina Bianchi
(8) (Context: Who does your/my/his wife really love?)
a. My wife loves ME.
b. Your wife loves YOU.
c. His wife loves HIM.
Similarly, focus does not block coreference when the co-referring terms are both
R–expressions. 9 (9) forms another minimal pair with (6).
(9) (Context: Who does John’s wife really love?)
John’s wife loves JOHN.
The final observation is that focus or destressing are completely irrelevant in
configurations in which the second term is an R–expression c–commanded by the
first term: (10b) is as unacceptable as (10a). 10
(10) a.* John/He claims that Mary hates JOHN.
b.* John/He claims that Mary HATES John.
These empirical observations can be schematically summarized as follows (the
square brackets represent the c–command domain of the first term):
6
(i) Focus effect in […pro]… R… configurations
(ii) Focus irrelevant in […pro]… pro… configurations
(iii) Focus irrelevant in […R]… R… configurations
(iv) Focus vs. destressing irrelevant in *[pro/R… R] configurations.
3. Reconsidering Reinhart’s account
Let us start from the focus effect in (6) vs. (7). A fairly common assumption is
that the focussed R–expression in (6) undergoes covert LF movement to some
leftward position, crossing over the pronoun: this results in a Weak Crossover
configuration.
(11) * JOHNi [hisi wife] loves ti
Whatever the exact nature of the Weak Crossover constraint, it will rule out
(11). 11
9 This is parallel to Schwarzschild’s (1999) (47), which involves contrastive focus:
(i) – John’s mother praised Bill. – No, John’s mother praised JOHN.
10 As discussed by Lasnik (1989), in languages like Thai coreference is possible when an R–
expression is c–commanded by another R–expression. This is a well known problem for
Principle C, which will also be inherited by my proposal below (see § 6.5).
11 On Reinhart’s (1983) view, a Weak Crossover configuration simply violates the Bound
Anaphora Condition: Pronoun binding can only take place from a c–commanding A–position
(Büring 2004: 24, after Reinhart 1983). In (11), the A–position occupied by the variable fails

A note on backward anaphora
This much is unobjectionable: note that Weak Crossover is a constraint on
binding, and in fact, even the grammatical examples in (8) above do not allow for a
genuinely bound interpretation (i.e., one involving two occurrences of the same
variable). For instance, (8a) does not mean that I am the (only) individual who is
loved by his own wife (12a); rather, it means that I am the (only) individual who my
wife loves (12b). 12
(12) a.* The x such that x's wife loves x = the speaker
b. The x such that the speaker’s wife loves x = the speaker
However, invoking the Weak Crossover constraint gives only half of an
explanation for (6). Even if binding is correctly ruled out, the question remains of
what rules out coreference. It seems to be often tacitly assumed that a focussed R–
expression is a “quasi–quantifier”, and as such, it cannot (co)refer. But this is too
strong an assumption, as shown by the ambiguity of (13):
(13) Only Joel voted for his proposal. (Büring 2005b: 262)
a. [λx. x voted for x’s proposal] is true of Joel and of noone else (sloppy)
b. [λx. x voted for Joel’s proposal] is true of Joel and of noone else (strict)
The focussed R–expression only Joel gives rise to a strict–sloppy ambiguity. In
Reinhart’s approach, the sloppy reading (13a) (whereby Joel is the only self-voter)
arises from syntactic (and semantic) binding, but the strict reading (13b) (whereby
Joel is Joel’s only supporter) is an instance of coreference between Joel and his. A
similar ambiguity arises in the question-answer pairs in (14):
(14) a.Whoi voted for hisi proposal? JOELi did (vote for hisi proposal).
(sloppy)
b. After the discussion of Joel’s report, who voted for his (=J’s) proposal?
JOEL (HIMSELF) did (vote for his proposal). (strict)
Since a focussed R–expression can corefer, we do need some constraint to rule
out coreference in (6). One possible candidate would seem to be Reinhart’s Rule I:
to c–command the pronoun. Safir (2004, 2007a, 2007b) rejects the c–command requirement
and proposes instead an anti–c–command requirement, the Independence Principle, whereby a
pronoun cannot depend on an antecedent if either the pronoun itself or some constituent
containing it c–commands the antecedent. Chomsky’s (1976) Leftness Condition may derive
from the basic asymmetry of left-to-right interpretation, as argued by Shan & Barker (2006).
A full development of this point exceeds the limits of the present discussion; for our purposes,
any formulation of the Weak Crossover constraint may do, and I will stick to Reinhart’s
proposal in my critical discussion.
12 This corresponds to the fact that in the context question of (8), the possessive pronoun is
not bound by the wh–phrase.
7

Valentina Bianchi
(15) Rule I (Intrasentential Coreference)
NP A cannot corefer with NP B if replacing A, at LF, with a variable bound
by the trace of B, yields an undistinguishable interpretation.
(Grodzinsky & Reinhart 1993: note 11).
However, Rule I only applies in configurations where the pronoun can be bound
by the trace of the antecedent at LF, where the trace corresponds to the argument
position. Since in (11) the trace of the focussed R–expression fails to c–command
the pronoun, (11) is not a possible binding configuration by the Bound Anaphora
Condition, and Rule I cannot even apply. 13 On the other hand, Rule I applies to the
c–command configurations in (10), and correctly rules out coreference
independently of any focus effect.
Reinhart’s (1986: 138-140) view of the contrast between (6) and (7) involves a
specific discourse constraint, which she (tentatively) formulated as follows:
(16) Topic–antecedent hypotesis: Backward anaphora is possible only if the
antecedent is in sentence–topic position. 14
Evidence in support of (16) is provided by the pair in (17a-b). According to
Reinhart, in (English) active clauses either the subject or the object can qualify as
the sentence topic, whereas in passive clauses the promoted subject is the obligatory
sentence topic, excluding the by–phrase. As predicted by (16), in (17a) either the
matrix subject or the object can be the antecedent of the backward pronoun, whereas
in (17b), the passive subject is the only available antecedent. Finally, (17c) is
inappropriate because the only plausible antecedent for the backward feminine
pronoun fails to satisfy the Topic-antecedent constraint.
(17) a. When he entered the room, √ Max greeted √ Bill.
b. When he entered the room, √ Max was greeted by * Bill.
c. # When she entered the room, Max was greeted by Kora.
13 Reinhart (2000, (26)) reformulates Rule I as follows: α and β cannot be co–valued in a
derivation D if (a) α is in a configuration to A(rgument)–bind B, and (b) α cannot
A(rgument)–bind β, and (c) the covaluation interpretation is indistinguishable from what we
would obtain if α A(rgument)–binds β. In configurations like (11), clause (a) does not hold
(cf. Reinhart’s (27), p. 16). Suppose instead that WCO is a possible (A’)–binding
configuration, although actual binding is excluded by an independent WCO constraint: then,
the (2000) formulation of Rule I would correctly exclude coreference in (11). However, this
assumption cannot account for the examples in (8)-(9): here too the rightward term is
focussed, yet coreference is not ruled out.
14 The notion of sentence-topic is that of Reinhart (1981), and it is defined in terms of an
aboutness relation. Very roughly, the sentence topic indicates which “file card” in the
conversational common ground will be updated by the propositional content of the asserted
sentence. This notion has been adopted and elaborated by Zubizarreta (1998).
8

A note on backward anaphora
The proposed constraint (16) can also account for the contrast between (6) and
(7), repeated here:
(6) His wife loves JOHN.
(7) His wife LOVES John.
A phrase bearing new information focus, being part of the comment, cannot be
the sentence topic (cf. Zubizarreta 1998: 9-10): thus, the focussed R–expression in
(6) cannot be the antecedent of the backward pronoun, because it fails to satisfy
(16). In (7), on the contrary, the destressed R–expression can qualify as the sentence
topic, hence as a possible antecedent for the backward pronoun. 15
The constraint in (16) is fairly natural, for it is well known that anaphoric
pronouns (tend to) take topics as antecedents. However, it cannot be a general
constraint on intra-sentential coreference, in view of the data in (8) and (9) above.
Two important qualifications are hidden in the term «backward anaphora»:
condition (16) only applies when (a) the coreference relation involves a pronoun and
an R–expression, and (b) the pronoun is to the left of the R–expression.
Concerning the first qualification (a), it may fall out naturally if we conceive of
(16) as a constraint on the assignment function, i.e. the context–determined function
from numerical indices to individuals which assigns a value to all free pronouns in
an LF. 16 A possible reformulation might be as follows:
(18) Topic–antecedent hypothesis (revised): An (index on an) unbound pronoun
can be assigned the referent of a following R–expression only if the latter is
the sentence topic.
This reformulation is compatible with the possibility of coreference in (8) and
(9). In (8c), for instance, the two coreferring third person pronouns will bear the
same index, and the assignment function will simply assign the same value to their
index (whether it corresponds to the sentence topic or not). 17 In (9), the two R–
expressions do not depend on the assignment function for their interpretation; hence
the condition (18) is irrelevant.
However, note that the reformulation in (18) still embodies the second
qualification (b), concerning the liner position of the pronoun and the R–expression:
15 See Erteshik–Shir (1998: 78 ff.) for an elaboration of this account.
16 Note that this does not correspond to Reinhart’s (1983) original view of unbound pronouns.
See, again, Heim (1998) for relevant discussion.
17 As for (8a-b), involving first and second person pronouns, it is controversial whether these
should be assimilated to free variables (Schlenker 2003, 2004) or not (Safir 2004). If they are
variables falling in the domain of the assignment function, they will be analysed as (8c),
modulo additional presuppositions concerning the speaker/hearer role borne by the
individuals that are assigned as values by the assignment function.
9

Valentina Bianchi
the Topic Antecedent constraint is sensitive to directionality in an apparently
irreducible way. This does not seem to follow from the view of coreference as a
«static» relation. (18) is even more puzzling from the viewpoint of incremental left–
to–right processing, in that a constraint on the assignment of a value to a free
pronoun depends on the information structure status of an R–expression following it.
Let me summarize the preceding discussion in the following points.
We have observed that the possibility of backward anaphora depends on the
interplay between two factors. The minimal pair (6) vs. (7) shows that the
information structure status of the coreferring terms is relevant (new information
focus or, alternatively, sentence topic status); on the other hand, the minimal pairs
(6) vs. (8c) and (6) vs. (9) show the relevance of the intrinsic “referential content” of
the coreferring terms (i.e. their pronominal vs. nonpronominal status). The first
factor characteristically plays a central role in discourse-oriented approaches to
anaphora, whereas the second factor is central to all the competition approaches to
anaphora. 18 The data from backward anaphora presented above strongly suggest that
both these factors are relevant, and that they interact in a very specific way.
At a purely observational level, it appears that coreference is unconstrained when
the two co–referring terms are on the same level of the referentiality hierarchy, as in
(8) and (9); it is only in case of an asymmetry that some constraint becomes
operative. However, this constraint is subject to another asymmetry concerning the
respective linear position of the less referential and the more referential term.
Anaphora is subject to the focus effect only when it is backward, namely, when the
pronoun precedes the more referential R–expression.
I believe that this linear dimension of the phenomenon is worth emphasizing. We
already know from work on inter-sentential anaphora that the linear – or rather,
temporal – order of presentation of the elements is absolutely crucial. 19 Even an
independent phenomenon like presupposition projection reveals a left–to–right
asymmetry which is plausibly due to an incremental interpretation process. The
preceding discussion suggests that this linear/temporal dimension is also crucial to
intra-sentential anaphoric relations, which are more commonly conceived of in a
static way. This point has been independently made by Shan & Barker (2006),
Barker & Shan (2007) in their analysis of quantificational binding. 20
18 The neo-gricean competition approaches reduce disjoint reference effects to the Q(uantity)–
implicatures arising from the hierarchy of referentiality between anaphoric expressions.
Burzio’s (1991, 1998) competition approach is based on a morphological hierarchy of
anaphoric expressions; Safir’s (2004) Form–to–Interpretation Principle is based on the Most
Dependent Scale. All these approaches are based on the same hierarchy (syntactic anaphor –
pronoun – name).
19 Cf. the notion of backward– vs. forward–looking center in Centering Theory.
20 And, less conclusively, by Schlenker (2005) (cp. his § 7.2).
10

4. Semantic vs. discourse anaphoricity
A note on backward anaphora
A different account of the contrast between (6) and (7) (repeated here) is found
in Williams (1997: 588), who emphasizes the role of focus vs. destressing.
(6) His wife loves JOHN.
(7) His wife LOVES John.
In his words, “if the putative antecedent does not have main sentence stress, then
the antecedent itself is anaphoric… really, John itself is dependent on something
preceding, and his is dependent on it as well.” 21 This means that there is no real
backward anaphora: in cases like (7), the pronoun is actually linked to an antecedent
in the previous linguistic context, rather than to the following R–expression. 22
(7’) John… His wife LOVES John.
In slightly different terms, this amounts to the following hypotheses. First, a
“backward” pronoun implies an antecedent in the previous discourse. Second, the
focussed R–expression in (6) must be incompatible with the presence of such an
antecedent. The first hypothesis is fully consistent with the idea, defended above,
that anaphora must be thought of in terms of a left–to–right incremental
interpretation process. From this perspective, the leftward pronoun his could not
inherit its referent from the following R–expression John, for the simple reason that
the interpretation of the pronoun precedes the interpretation of the R–expression.
21 The same observation is also found in Calabrese (1986: 14).
22 Carden (1982) explicitly argues against this view, which he dubs the “forward
[pronominalization] only” hypothesis and he defines “a theoretical garden path” (see also
Reinhart 2004, § 14.3.1) Some of Carden’s evidence, based on the analysis of a corpus of
spontaneously produced discourse, will be discussed in §6.2. One real exception to
Williams’s view is the mystery story case discussed by Safir (2007: (49a)):
(i) His back was turned to us when we came in. He swivelled in his chair to face us. The
penetrating eyes of Count Marzipan were trained upon us.
As noted by Safir, in these cases “a pronoun is introduced that we have no referent for and
wer must wait for a plausible candidate to appear that supplies a value for the pronoun”.
However, Carden (1982: 372) agrees that such cases of “withheld antecedent” are
qualitatively different from garden-variety backward pronominalization. (Similarly, in
Gordon & Hendrick’s 1997 experimental investigation of backward anaphora, the relevant
examples are introduced without any context, hence, as the authors note, “the ability to co–
referentially interpret a name and a preceding pronoun depends on a suspension of full
interpretation of the initial pronoun”.) Safir (2007) argues that backwards dependence is
possible, but it is disfavoured by the principle Preferred Covaluation, which creates a tension
until the point where the referential antecedent is introduced.
11

Valentina Bianchi
But what about the second hypothesis? Williams (1997) derives it from his view
of anaphoric destressing. The destressed R–expression in (7) is dependent on a
previous discourse antecedent; in turn, when the R–expression is focussed, as in (6),
its discourse referent cannot have been already introduced in the previous discourse;
therefore, the “backward” pronoun has no possible (preceding) antecedent.
However, Williams’s claim that the destressed proper name in (7) is anaphoric
requires some qualification. It is widely assumed that proper names are directly
referential: so how could they possibily be anaphoric? Anaphora is generally
conceived of as the property of inheriting a referent from an antecedent.
I think that it is necessary here to clearly distinguish two notions of anaphoricity:
strict semantic anaphoricity and discourse anaphoricity. The former corresponds to
the way in which a nominal expression comes to denote: pronouns are strictly
anaphoric in that they must either undergo semantic binding or be assigned a value
that they lack intrinsically (let’s say, via a context-determined assignment function);
on the contrary, R–expressions are either inherently endowed with a fixed referent
(proper names) 23 or come to denote an individual through semantic composition
(definite descriptions). From this perspective, the proper name John in (7) cannot be
strictly anaphoric in the same way that a pronoun is: 24 it is intrinsically “more
referential” than a pronoun. 25
On the other hand, the notion of discourse anaphoricity is much broader: any
noun phrase that is preceded by a coreferential term is discourse anaphoric.
Unbound pronouns (apart from deictic uses, to which I return in § 6.1 below) are
discourse anaphoric, 26 but definite R–expressions can be as well: in fact, a definite
R–expression can alternate with a pronoun when its referent has already been
linguistically introduced (this corresponds to Roberts’s (2003) strong familiarity). 27
23 Though speaking of a “fixed referent” is obviously an oversimplification, it will do for
present purposes.
24 This notion of semantic anaphoricity is not to be confused with the syntactic classification
of noun phrases by the features [±anaphoric] and [±pronominal] in standard versions of
Binding Theory.
25 Elbourne (2005) however conceives of all pronouns as covert definite descriptions.
26 Schwarzschild (1999) notes that pronouns can be focussed (not GIVEN) even though they
are in some sense necessarily “not new”. In my terms, non–deicic pronouns are strictly
anaphoric and also discourse anaphoric; hence, Schwarzschild’s GIVENness cannot be equated
with discourse anaphoricity.
27 As observed by Roberts (2003), anaphoric uses of definite descriptions are most felicitous
when there is a certain distance between the definite description and the linguistic antecedent.
See also von Heusinger (2004) for an analysis of anaphoric definite descriptions in terms of a
global choice function.
12

A note on backward anaphora
(19) There’s a doctor in our little town. He/The doctor is Welsh.
(Roberts (2003), (8))
The notion of discourse anaphoricity is thus relational, and does not exclusively
depend on the intrinsic content of the noun phrase in question.
There is an apparent tension between semantic and discourse anaphoricity:
definite R–expressions are semantically non–anaphoric, but they can be discourse
anaphoric, as in (7) or (19): they independently denote a referent, but this referent
may be already available in the previous linguistic context. This tension dissolves
once it is recognized that the two notions of anaphoricity pertain to different levels
of organization of language. An R–expression, being strictly non–anaphoric, always
«brings with it» a referent in the semantic computation of the clause containing it.
At the same time, an R–expression may or may not introduce its referent at the
discourse level, as a potential antecedent for future discourse anaphora (cf. the
forward-looking centers of Centering Theory). In the following section I will try to
make these ideas a bit more precise, and I will show how they can derive the above
observations about backward anaphora, essentially implementing a version of
Williams’s insights.
One more clarification is in order. In the following dicussion, I will continue to
describe anaphoric relations by the usual terms of introducing and retrieving/picking
up a referent. However, Safir (2004b, 2007) has argued that the type of anaphoric
relations that may link one noun phrase to another are much more varied and
complex that this. The common core of all anaphoric relations is the existence of a
dependency between an antecedent and an expression whose interpretation
somehow depends on the interpretation of that antecedent. The mechanism
informally sketched below might be conceived of as a way of implementing
dependent identity in Safir’s broader sense.
5. Sketch of a solution
5.1. Local vs. global memory registers
The preceding discussion suggested that anaphora is constrained by two levels of
organization: the local level of semantic computation of truth conditions, and a
global discourse level. I contend that the two levels, and the two relevant notions of
anaphoricity, must remain distinct: thus, an R–expression can never be strictly
anaphoric, even though it can be discourse anaphoric; conversely, a (non–deictic)
13

Valentina Bianchi
pronoun will always be strictly anaphoric, even in case of inter–sentential
«accidental coreference». 28
If both levels are relevant to anaphora, we have to assume that at both levels
some mechanism keeps track of noun phrase referents. For concreteness, I adopt an
informal version of the mechanism proposed by Schlenker (2005): in the course of
the incremental interpretation of a sentence, the referents of noun phrases are
dynamically stored in a memory register. This can be conceived as an ordered list, or
a sequence, of (discourse) referents, which can be retrieved as the value of anaphoric
pronouns. The precise details of the storage and retrieval operations are orthogonal
to the main proposal; in the Appendix, I adopt for concreteness the sequence of
evaluation and the negative indices proposed in Schlenker (2005).
In addition, I assume that, when the interpretation of the sentence is completed,
the referents are passed up to a discourse–level memory register, where they become
available for subsequent inter–sentential anaphora. I will dub these the local and the
global register, respectively. 29
The notion of strict anaphoricity defines the way in which the interpretation of an
NP affects the memory registers:
(20) a. A non–strictly anaphoric NP (= R–expression or deictic pronoun) adds its
referent to the memory register(s).
b. A strictly anaphoric (i.e. non–deictic) pronoun must retrieve a referent
contained in some accessible register.
(21) a. For any noun phrase contained in Xi, the local register of Xi is accessible.
b. For any noun phrase contained in Xi, the global register is accessible.
Note that by (21), a pronoun can retrieve a referent either from the local register
or from the global one. I return to the nature of the local registers (21a) in § 5.4. In
the following two subsections, I first discuss the role of the global register.
5.2. The role of focus
Let us go back to the notion of discourse anaphoricity, and to the crucial contrast
between new information focus and destressing in (6)-(7). I assumed above that in
(6), focus on the rightward R–expression is incompatible with a coreferential
antecedent in the previous context. Why is this the case? Recall that on Williams’s
view, the presence of an antecedent necessarily implies anaphoric destressing, and
hence, the lack of anaphoric destressing in (6) in turn implies the absence of such an
28 On this second point, I agree with Safir (2004, 2007a,b), who allows for inter–sentential
dependencies, and I differ from Schlenker (2005), who analyses all free pronouns as
demonstrative pronouns and assimilates them to R–expressions.
29 Schlenker (2005) does not distinguish local and global registers.
14

A note on backward anaphora
antecedent. Like Williams, I assume that the presence of the “backward” pronoun
implies a previous antecedent, but I suggest that the the focussed R–expression in
(6) re–introduces the same discourse referent in the rheme, as new information, and
this is an illegal move. 30
I will implement this idea by assuming that focus vs. destressing constrains the
way in which a non–strictly anaphoric NP affects the global register. As a
qualification to (20a), I propose the following condition:
(22) a. A non–strictly anaphoric NP introduces an element in the global memory
register iff it bears new information focus.
b. A non–focussed (destressed) NP does not affect the global register.
By (22), the global register only inherits from the local semantic computation of
a sentence what is contained in the rhematic part of it.
Finally, I adopt the principle of Non–Redundancy proposed by Schlenker (2005):
(23) Non Redundancy: No element may be introduced twice in one and the same
memory register.
Note that (22) and (23) jointly require destressing of discourse–anaphoric R–
expressions: after a given referent has been introduced in the global memory
register, all subsequent co–referential R–expressions are barred from bearing new
information focus; if they did, by (22) they would re–introduce their referent in the
global register, giving rise to a violation of Non Redundancy.
5.3. Global violations of Non–Redundancy
At this point, we have the basic ingredients to reconsider the backward anaphora
configurations in (6)-(9). For the time being, I exclusively concentrate on the
management of the discourse–level register: in a nutshell, I propose that the
impossibility of coreference in (6) results from a violation of Non–Redundancy at
this global level. To see how this result is derived, consider the left–to–right
interpretation of example (6):
(6) (Context: As for John, who does his wife really love?)
?* His wife loves JOHN
The backward pronoun his has no preceding antecedent in the noun phrase
containing it, nor in the matrix clause containing that noun phrase; therefore, its
referent must be retrieved from the global memory register. When the coreferential
object R–expression is processed, it introduces the same referent in the local register
30 This idea was inspired by what Levinson (1991: 114) reports as “Bolinger’s maxim”:
“Don’t reintroduce (i.e. lexically specify) the topic of the theme in the rheme”.
15

Valentina Bianchi
of the matrix clause; but since the R–expression is focussed, by (22a) it also
reintroduces its referent in the global memory register, violating Non–Redundancy.
Example (7) is minimally different:
(7) (Context: –As for John, I believe his wife hates him. –You’re wrong:)
His wife LOVES John.
Here too, the pronoun his must retrieve its referent from the global register;
however, the following R–expression is destressed and, by (22b), it only introduces
its referent in the local register of the matrix clause, but not in the global one. Thus,
Non–Redundancy is fulfilled.
Recall now that (8c), involving two coreferential pronouns, differs from (6) in
that focus does not block coreference:
(8) (Context: Who does his wife really love?)
c. His wife loves HIM.
The reason for this contrast is that pronouns are strictly anaphoric (cf. (20b)), and
never introduce a referent in the memory registers. The two pronouns in (8c) retrieve
the same referent from the global register, and the focus condition (22) has no effect.
The examples (8a-b) are entirely parallel, except for the fact that they involve
first and second person pronouns:
(8) (Context: Who does your/my wife really love?)
a. My wife loves ME.
b. Your wife loves YOU.
I adopt Schlenker’s (2005) proposal that the referents of first and second person
pronouns, i.e. the speaker and the hearer/addressee(s), are always automatically
available since they are given by the mere existence of the speech act. I differ from
Schlenker in distinguishing global and local memory registers; specifically, I assume
that these referents are contained in the global discourse register. 31 The two
coreferential pronouns in (8a) and in (8b) retrieve the same referent from the global
register (the speaker and the hearer, respectively).
Let us now reconsider (9):
(9) (Context: who does John’s wife really love?)
John’s wife loves JOHN.
The minimal difference w.r.t. (6) is the fact that the first coreferential term is a
destressed R–expression, rather than a strictly anaphoric pronoun. I propose that a
destressed R–expression does not necessarily imply the previous availability of the
relevant referent in the global register: by condition (22b), anaphoric destressing
31 In Schlenker (2005), the speaker and the hearer are introduced as the first elements of any
memory register.
16

A note on backward anaphora
simply signals the invisibility of an R–expression for the global register. From the
present perspective, in order for coreference to be allowed in (9), it must be the case
that the discourse referent denoted by John is not contained in the global register at
the point where the computation of (9) begins, although this name is indeed
mentioned in the context question. (As a matter of fact, the previous occurrence of
the name in the context question is destressed, and it cannot have introduced its
referent in the global register.) Thus, the focussed direct object in (9) introduces this
referent for the first time, in compliance with Non–Redundancy. 32
In sum, the set of assumptions in (20)–(23) provides an implementation of
Williams’s insight within the framework of an incremental left–to–right
interpretation system. 33 The advantage of this implementation is that it allows us to
see an affinity between the global discourse level of computation of anaphora and
the “local” sentence level. To show this, I now return to the local memory registers.
I will adapt Schlenker’s analysis of binding, trying to make explicit the relevant
notion of locality. (A more detailed presentation of my proposal based on
Schlenker’s formalization is deferred to the Appendix.)
5.4. Local violations of Non–Redundancy
In (21a), I have left open the definition of the local memory registers: what is the
nature of Xi? Recall that by assumption, the local registers are used for the
computation of reference and of truth conditions (e.g., the referent of John is used to
compute the reference of the complex expression John’s wife and the truth
conditions of the sentence John’s wife is pregnant).
32 This view of the problem entails a prediction: a sentence like (9) with a focussed rightward
R–expression should be impossible in a context where there is a focussed previous mention.
This prediction seems to be correct:
(i) ! I’ve had bad news about JOHN: John’s wife wants to break away from JOHN.
33
David Adger (p.c.) points out the following example, with focus on the first R-expression:
(i) JOHN’s wife loves John.
This seems to be an instance of contrastive focus. As stated in (22a), I assume that only new
information focus allows an NP to introduce its referent in the global register: contrastive
focus, instead, is not an appropriate way to introduce a new potential topic. Note that, since
the second R-expression is destressed, it would not produce a violation of Non-Redundancy
even if the first R-expression were to introduce its referent in the global register. This last
remark does not apply, however, to the following example, which I owe to another member of
the audience at the 30th Glow Colloquium:
(ii) I don’t know about Bill’s wife, but JOHN’S wife loves JOHN.
Here again, the R–expression embedded in the subject is contrastive: possibly, this is not
contrastive focus but a contrastive topic (Mara Frascarelli, p.c.); as a matter of fact, it is not
part of the rheme, as required by condition (22a).
17

Valentina Bianchi
The null assumption would seem to be that a local register is associated with
each sentence. However, we can immediately see that this assumption wrongly
predicts a violation of Non–Redundancy in the computation of (9), repeated here:
(9) John’s wife loves JOHN.
By (20a), the two R–expressions would introduce twice the same referent in the
sentence–level local register, violating Non–Redundancy; however, in (9)
coreference is perfectly acceptable.
It is possible to solve this problem by assuming that the local memory registers
are not sentence–level, but rather, they are local to smaller pieces of structure. 34 Let
us adopt the following informal definition of semantic phases:
(24) Every minimal noun phrase or clause is a semantic phase, i.e., a subpiece of a
top–down, left-to-right semantic computation.
Note that phases, as subunits of a computation, are obviously contained in one
another: in (9), the subject NP phase is contained the matrix clause phase. I will dub
a nonmatrix phase a subordinate phase, and the immediately superordinate phase,
the containing phase.
Let us assume, then, that the local memory registers are not associated with
sentences, but rather, with semantic phases. Following Chesi (2004a, 2007), I
assume that the local phase registers are linked to one another by the following
inheritance relation:
(25) a. A subordinate phase pS inherits the content of the local register of the
containing phase in the state it has reached when the computation of pS
begins.
b. A containing phase, instead, does not inherit the content of the local
register of a subordinate phase, but only the result of its overall semantic
computation.
Let us now reconsider example (9) in the light of (25) (the square brackets
indicate the phase containment relations):
(9) [[John’s wife] loves JOHN].
The subordinate subject NP phase will have a separate memory register from that
of the matrix clause, and by (25b), the latter will only receive the referent of the
whole subject NP phase, John’s wife, but not the referent of the subexpression John;
hence, the subsequent processing of the object R–expression John will not cause a
violation of Non–Redundancy in the matrix clause register. On the other hand, by
34 This is also assumed by Schlenker (2005), who, however, does not make fully explicit the
relevant notion of locality (cp. note 37 below). The definition in (24) is inspired by Chesi
(2004a, 2004b).
18

A note on backward anaphora
(25a) the subordinate subject NP phase inherits the matrix clause register in the state
it has reached when the processing of the subject begins: at this point, the matrix
clause register does not yet contain the referent of the object R–expression. Thus,
Non–Redundancy is satisfied by both the local registers. 35
(24)-(25) can also account for the Principle C effects in (10), repeated here:
(10) a. * [John claims [that Mary hates JOHN]]
b. * [John claims [that Mary HATES John]]
Here the first R–expression is contained in the matrix clause phase, and the
second R–expression is contained in the embedded clause phase. By (25a), the
embedded clause register inherits from the register of the containing matrix clause
the referent of the matrix subject R–expression: then, Non–Redundancy is violated
at the local level of the embedded clause register, when the direct object John re–
introduces the same referent which was inherited from the matrix clause phase.
Focus vs. destressing is irrelevant here, because the focus condition (22) only
concerns the introduction of a referent in the global register; however, a strictly non–
anaphoric R–expression always introduces its referent in the local register, even
when it is destressed. In other terms, a violation of Non–Redundancy at the local
level is not sensitive to the information structure status of the R–expressions.
The inheritance relation defined in (25) also allows for an implementation of
binding in terms of retrieval of a local antecedent:
(26) [John loves [his wife]].
Here the subject R–expression introduces its referent in the local register of the
matrix clause. The content of this register is inherited by the local register of the
direct object noun phrase, his wife: the pronoun his can thus retrieve its referent
directly from the local register, rather than from the global one.
One final clarification is necessary with regard to (10). As is well known, a
Principle C effect also arises when the first coreferring term is a pronoun:
(10') a. * [He claims [that Mary hates JOHN]].
b. * [ He claims [that Mary HATES John]].
35 Cf. Schlenker (2005, 16):
(i) [Bill’s teacher] likes Bill.
“The key is that the VP hates Bill is evaluated under a sequence that contains Bill’s teacher
but not Bill himself, with the result that Non–Redundancy is satisfied.” This implies that the
subject noun phrase and the predicate are evaluated with respect to different sequences: see
(44) below.
19

Valentina Bianchi
This suggests that a pronoun retrieving its referent from the global register
introduces that referent in the local register of the phase containing it: 36 the referent
of the matrix subject pronoun is inherited by the embedded clause local register and
yields a violation of Non–Redundancy in the embedded clause phase, when the
object R–expression is processed. As the reader can easily verify, this assumption
does not overgenerate in the case of (6)-(8), because the pronoun is embedded in the
subject NP phase, whose local register is distinct from that of the matrix clause
phase, which contains the coreferential object.
In sum, within a left–to–right oriented interpretation process the inheritance
relation between phases essentially captures the effects of c–command (see Chesi
2004 for general discussion). This, however, only holds for local anaphora; as (21b)
makes clear, the global register is always accessible from any phase.
5.5. Interim conclusions: local and global anaphora
The discussion of (9) and (10) has emphasized the role of the computation at the
local level, where some locality notion is relevant (here semantic phases, linked by
an inheritance relation which replaces c–command). At this point, we can rethink
Reinhart’s divide between binding and «accidental coreference» in terms of local vs.
global anaphora: a bound pronoun is a pronoun that retrieves its referent from the
phase–local register (where it may have been inherited from a previous phase); a
coreferential pronoun, instead, retrieves its referent from the global discourse–level
register. However, the mechanism of storage and retrieval are the same in the two
cases.
The local and the global registers stand in a quasi–hierarchical relation, since the
global register is superordinate to all the phase–local registers. 37 On the other hand,
despite the similarity of the storage and retrieval mechanism, these objects serve
quite different purposes, and pertain to different levels of organization of the
grammar. The local registers are essentially semantic, in that they serve the purpose
of reference and truth assessment. The global register is, instead, a repository for
potential discourse topics, and as such it is affected by discourse constraints that are
instead irrelevant to the «local» semantic computation. One such constraint, I
suggest, is the focus condition (22). This general idea is compatible with technically
different views of a (structured) global repository of referents; from the present
36 The assumption that the retrieved referent is copied from the global register into the local
one is consistent with the idea that the local registers are used for the computation of
reference and truth conditions: see the Appendix for discussion.
37 The global register is intended to model Roberts’s (2003) strong familiarity. The notion of
weak familiarity corresponds to a more general repository of referents – all those whose
existence is somehow entailed by the Common Ground.
20

A note on backward anaphora
perpective, the only necessary constraint is that this register is intrinsically dynamic
and compatible with a left–to–right intraclausal interpretation. Let me stress once
again that from this perspective, left-to-right orientation must be a property of the
grammar itself, and not merely of processing, since grammatical constraints such as
(23) and (25) are sensitive to it.
6. Further consequences and open problems
6.1. Deictic pronouns
In (20a) above, repeated here, I have assumed that deictic pronouns (those
accompanied by some sort of ostension, which depend on the physical context of the
speech act for their interpretation) differ from anaphoric pronouns in that they
introduce a referent rather than retrieving it from some register.
(20) a. A non–strictly anaphoric NP (= R–expression or deictic pronoun) adds its
referent to the memory register(s).
In this respect, deictic pronouns are assimilated to R–expressions; hence, (22)–
(23) predict that they should give rise to a focus effect in a backward anaphora
configuration. This prediction seems to be correct: an example parallel to (8c), but
with a deictic use of the pronoun, disallows backward anaphora, much as the
focussed R–expression in (6).
(27) (Context: Among these men here, could you point out one that is such that
his wife loves him?)
* His wife loves ∗HIM!
6.2. Backward anaphora with a novel antecedent
As is widely assumed, indefinite noun phrase introduce novel referents, which
are not available in the previous linguistic context. The present approach predicts the
impossibility of backward anaphora with a novel antecedent. My introspective
judgement supports this claim:
(28) a. * His wife loves a judge.
b. * His wife said that a judge was corrupt.
In (28), the non-deictic pronoun his must recover its referent from the global
register, and the following indefinite expression reintroduces the same referent in it,
violating Non–Redundancy.
21

Valentina Bianchi
Carden (1982: 367 ff.) 38 pointed out that in real text examples, backward
anaphora often has an indefinite rightward antecedent:
(29) When she was five years old, a child of my acquaintance announced a theory
that she was inhabited by rabbits (Carden’s (12)).
Here, the referent of the indefinite subject is (by hypothesis) not in the global
register at the point when the pronoun is computed, and it cannot be recovered from
the local register of the adjunct clause either. As stressed by Carden, this kind of
example cannot be accounted for by the «forward–pronominalization only»
approach, since there cannot be any antecedent preceding the pronoun.
However, note that there is something special about these examples: as stressed
by Calabrese (1986: (43)), the indefinite antecedent must be in a highly prominent
subject position (Calabrese’s Thema position), 39 which has scope over a clause–
initial adjunct phrase, though linearly following it: 40 as a matter of fact, the same
configuration also allows for quantificational binding of a pronoun within the
adjunct (Carden 1982: 368 ff.). A natural solution is to assume that the preposed
adjunct clause can be reconstructed into a position that follows the subject, i.e., its
interpretation can be «delayed» in the left-to-right interpretation. An account along
41 42
these lines remains to be worked out.
38 See also Reinhart (2004: 296) for recent discussion.
39 “(43) In an utterance U, a pronoun in U can have either the index of a referent in the
Universe of Discourse of U, or the index of the Thema in U, where the Thema of U is the
Thema of the main clause”; Calabrese (1986, (81)): “The T–domain of a Thema is given by
the c–domain of its predication” (including preposed adverbial clauses).
40 This is supported by the observation that at least in Italian, the relative position of the
Thema and of a clause–initial adjunct can be freely reversed.
41 This account predicts that in case of backward anaphora, a novel antecedent must be in
Thema position. Notice that this position typically hosts specific indefinites, rather than
nonspecific ones. Carden (1982: 366) also observes a substantial number of “first mention”
definite antecedents, corresponding to Roberts’s (2003) weakly familiar referents. I have
tentatively assumed above that the global register only contains strongly familiar referents;
weakly familiar antecedents may perhaps be included as well, if they are somehow made
salient (see Roberts 2003, § 3 for extensive discussion of the salience requirement for
pronouns’ antecedents).
42 Sorace & Filiaci (2006) and Belletti et al. (2007), in experimental studies of backward and
forward anaphora interpretation by English near-native L2 speakers of Italian, observe the
very strong tendency of backward subject pronouns to take the matrix subject as antecedent,
and attribute it to an interface processing strategy: since backward anaphora has an intrinsic
processing computational load, a processing strategy favors coreference with the first possible
referent that is encountered, i.e. the matrix subject.
22

A note on backward anaphora
We have seen above that (22) implies the necessary destressing of discourse–
anaphoric definite noun phrases. Conversely, indefinite noun phrases are constrained
in the opposite way: in order to introduce their novel referent in the global register,
they have to bear new information focus. This is consistent with the well known
observation that novel indefinites occur in the rhematic part of the clause.
6.4. Reinhart’s contrast
Another piece of data to be accounted for is Reinhart’s contrast in (17), repeated
here:
(17) a. When he entered the room, √ Max greeted √ Bill.
b. When he entered the room, √ Max was greeted by * Bill.
c. # When she entered the room, Max was greeted by Kora. 43
A relevant observation is that the by–phrase in (17c) belongs to the rhematic part
of the clause, and hence, it introduces its referent in the global memory register,
giving rise to a global violation of Non–Redundancy.
6.5. The strict/sloppy ambiguity
As is well known, Reinhart’s distinction between binding and accidental
coreference is usually invoked to account for the ambiguity between the strict and
sloppy reading of pronouns contained in elliptical VPs, as in (13) above and in (31)
below:
(31) John loves his mother, and Bill does too.
Following Schlenker (2005: 31 ff.), it is possible to derive the sloppy reading of
the elliptical VP if the pronoun contained in the latter retrieves its referent from
within the (local) memory register of the second conjunct: this register contains the
referent of the subject noun phrase Bill. 44 On the other hand, in order to derive the
strict reading, Schlenker stipulates that «an elided pronoun may optionally introduce
43 Note that in the Italian equivalent of (17a), a null subject in the adjunct clause could only
take the matrix subject as an antecedent. On the other hand, even an overt subject pronoun
would disallow coreference with a by–phrase, in the Italian eqivalent of (17b–c). The
relevance of the opposition between null and overt subjects is confirmed by the L2 acquisition
studies by Sorace & Filiaci (2006), Belletti et al. (2007); see also Carminati (2002) and
Frascarelli (2007) for relevant discussion.
44 This proposal would also subsume Reinhart’s claim that the sloppy reading is only allowed
in c–command configurations: this is because the sloppy reading requires the retrieval of an
antecedent which is contained in the (local) memory register of the clause containing the
sloppy pronoun. See however Hardt (2003) and Safir (2007b) for criticism of the c–command
condition on the sloppy reading.
23

Valentina Bianchi
in the sequence the value that its antecedent did»: namely, the elided pronoun in the
second conjunct may retrieve the same referent that the pronoun in the antecedent
VP had retrieved from the local memory register of the first conjunct, i.e. John. 45
This solution remains somewhat stipulative in Schlenker’s approach, but it is
interesting to note that it is equivalent to the analysis of strict readings recently
proposed by Safir (2007b). According to Safir, the strict reading arises when the
pronoun in the elided VP is identity–dependent on the parallel pronoun in the
antecedent VP (32b), rather than on the local subject (32a). (Dependent identity is
represented by means of arrows.)
(32) a. John loves his mother and Bill does [love his mother] too (sloppy)
24
b. John loves his mother and Bill does [love his mother] too (strict)
Safir (2007b) argues that the strict reading cannot be analysed in terms of
coreference: this is shown by the possibility of a strict reading of (33) (his (6)),
where the strict interpretation of the elided pronoun involves a bound variable (cf.
Heim 1998).
(33) Everyone thinks that he loves his pet and that St.Francis does too
a. Every x, x a person (x thinks that (St.Francis y (y loves y’s pet))) (sloppy)
b. Every x, x a person (x thinks that (St.Francis y (y loves x’s pet))) (strict)
This example is not problematic in Schlenker’s account, as he assumes that a
quantifier introduces a variable in the (local) memory register: this variable can be
exceptionally retrieved in reading (b) by the same mechanism that allows the strict
reading of (31).
It is fair to say that the exceptional retrieval of a non–local antecedent in the
strict reading is at odds with the locality of memory registers which I have assumed
(and which Schlenker also implicitly assumes). The two conjoined clauses of (31)
have two separate local registers; thus, Safir’s insight cannot be expressed by the
45 Schlenker (2005, 31 ff.) argues that this solution also yields an account of Dahl’s puzzle,
whereby in a series of pronoun c–commanding one another, a strict reading of one pronoun
blocks a sloppy reading of all the other pronouns c–commanded by it.
(i) Max said he saw his mother, and Oscar did too [say that he saw his mother].
a. Ok sloppy – sloppy: Oscar said that Oscar saw Oscar’s mother
b. Ok strict – strict: Oscar said that Max saw Max’s mother
c. Ok sloppy – strict: Oscar said that Oscar saw Max’s mother
d. *strict – sloppy: Oscar said that Max saw Oscar’s mother
The interpretation in (d) cannot be derived: if the first subject pronoun he in the elided VP of
the second conjunct has a strict interpretation, the following pronoun his (with negative index
–1) can only retrieve this same referent, and cannot refer back to Oscar.

A note on backward anaphora
mechanism of local anaphora. In the approach I have sketched here, the strict
reading would rather have to be mediated by the global register. 46 The problem
remains open (see the Appendix to Safir 2007b for extensive discussion of the strict
reading).
6.6. Exceptional violations of Principle C
Since Non–Redundancy is as categorical as Principle C is, it inherits all the
well–known problems for Principle C which were originally pointed out by Evans
(1980). One case in point is the exceptional coreference in (33):
(33) (We believed that everybody would vote for Joel, but in the end it turned out
that…) only JOEL voted for Joel.
In this particular example, we may avoid a local violation of Non–Redundancy
by stipulating that the complex phrase [only Joel] is a semantic phase distinct from
the matrix clause, and hence the two occurrences of the R–expression introduce their
referent in two distinct local registers. But the assumption is dubious, as the
focussed R-expression in (33) need not be accompanied by a focussing adverb.
In this respect, the present proposal shares the rigidity of Reinhart’s original
principle, as well as that of Schlenker’s (2005) proposal. Reinhart herself explained
away Evans-type counterexamples by observing that in contexts like (33), the
coreference option is not equivalent to the binding option (i.e., the two predicates
λx.vote(x,x) and λx.vote(x, joel) give rise to distinct truth conditions), and argued
that this is why Rule I does not apply. Along similar lines, Schlenker (2006) too has
replaced Non-Redundancy with a non–categorical principle, Minimize Restriction,
which is somewhat closer in spirit to the «competition» approaches to anaphora.
(Safir (2004, 2007a) proposes another non-categorical principle, Expected Non–
Covaluation.) Despite being highly marked, Evans-type counterexamples are
generally accepted as grammatical in the literature, and remain a counterexample for
the present Redundancy-based approach.
46 Such a move would not be unprecedented: Hardt (2003, 2004) proposed a Centering-based
account of the strict/sloppy ambiguity, whereby the elided pronoun refers to the current
Center: the strict reading results from Center retention between the first and the second
conjunct (i.e., John is retained as the current Center in the second conjunct), whereas in the
sloppy reading there is a Center shift from the first to the second conjunct (i.e.., Bill becomes
the current Center in the second conjunct in the place of John). As mentioned above, the
global register discussed here is conceptually similar to the list of backward and forward–
looking Centers. (See Hardt 2003, § 5 for an analysis of Dahl’s puzzle, mentioned in note 45.)
25

Valentina Bianchi
7. Concluding remarks
Although the plot of this paper may have seemed quite complex, it is actually
very simple. By bringing together four empirical observations about backward
anaphora, I have tried to convince the reader that a proper understanding of the
phenomenon must take into account the interplay of two factors: the information
structure status of the coreferring noun phrases, and their intrinsic «referential
content». By reconsidering Reinhart’s Topic Antecedent hypothesis, I have come to
the conclusion that anaphora must be thought of in terms of a left–to–right
interpretation process. I have then adopted Williams’s insight about the focus effect,
according to which there is no real backward anaphora; I have suggested that a
focussed rightward ‘antecedent’ illegally reintroduces a referent which was already
available at the point where the backward pronoun was processed. I have
implemented this idea within a modified and informal version of the interpretation
system proposed by Schlenker (2005).
Hardly anything in this note is entirely original; however, I hope that by
combining insights from various sources, I have succeeded in making two points.
First, it seems to me that the distinction between the semanticist’s strict anaphoricity
and the discourse theorist’s discourse anaphoricity is not always clearly stated,
which may lead to some misunderstanding of one another’s viewpoint. If my line of
reasoning is correct, in the domain of anaphora we have to recognize the coexistence
of two levels of organization: the local level of the semantic computation, which I
have argued is divided in phases, and a global discourse level, which is at the
interface between semantics and pragmatics. Both levels are dynamic and
incremental; the standard generative divide between syntactic binding and pragmatic
coreference has been restated in terms of local versus global anaphora. Secondly, I
hope I have made a convincing case for an incremental left–to–right semantic
interpretation at the intraclausal level, as well as at the discourse level. This view of
semantic interpretation supports an architecture of the grammar in which the
syntactic and semantic computation proceed in parallel and, crucially, in a top–
down, left–to–right fashion. From the syntactic viewpoint, this implies an
incremental structure building along the lines of Phillips (2003) or Chesi (2004a,b,
2007). A left–to–right architecture incorporates the central insight that the temporal
dimension is not just a fact about processing/performance, but it is intrinsic to the
constitution of natural language grammars.
26

Appendix: an implementation based on Schlenker (2005)
A1. A summary of Schlenker (2005)
A note on backward anaphora
The sequence of evaluation represents the linguistic context w.r.t. which a
constituent is evaluated. It includes
– those objects that are given by the mere existence of the speech act, i.e. the
speaker and an addressee,
– the objects that have been linguistically introduced, i.e., which are the
denotation of the terms that have been processed.
A sequence of evaluation represents a state of a memory register, which is
constructed as a sentence is processed, top down, in accordance with the following
rules:
i) R–expressions (= proper names, demonstrative pronouns and definite
descriptions): when an R–expression is processed, its denotation is added at the
end of the register. (Demonstrative pronouns bear a positive index).
ii) Non–demonstrative (= indexical and bound) pronouns: when a non–
demonstrative pronoun is processed, some element of the register is recovered
and moved to the end of the register, leaving an empty cell (#) in the original
position. Non–demonstrative pronouns bear negative indices, which indicate
“how far back” in the sequence the denotation is to be found.
iii) The speaker/author and the addressee/hearer are always the first elements of any
sequence; they are singled out by the superscripts A and H (e.g. j A^ m H )
iv) An n–place predicate is evaluated w.r.t. the last n elements of the sequence. (i.e.,
it is interpreted after all its arguments, in the lowest position of Larsonian VP
shells.)
(34) Ann runs.
[[Ann run]] w j A^ m H = 1 iff [[run]] w j A^ m H^ a =1, iff a ∈ Iw(run)
(35) # Bill likes him–1 : j A ^m H ^b j A ^ m H #^b
[[ like]] w, j A ^m H ^#^b: the interpretation of the 2–place predicate w.r.t. the
last two cells is undefined (Principle B effect)
(36) Non–Redundancy: No object may occur twice in the same sequence of
evaluation. 47
47 Apparent violations of Non-Redundancy are due to different guises, which must count as
different elements in the register (cf. Heim 1998). Schlenker (2006) also points out that his
principle Minimize Restriction (replacing Non–Redundancy) can be violated when an R–
27

Valentina Bianchi
Non–Redundancy is justified by the observation that the speaker and addressee –
which are always given at the beginning of any sequence – cannot be referred to by
an R–expression, which would reintroduce them in the sequence, by (i).
(37) a. # John is happy (said by John to Mary): j A^ m H ^j (* Non–Redundancy)
b. #Mary is happy (said by John to Mary): j A^ m H ^m (* Non–Redundancy)
(38) a. # Bill likes Bill : j A^ m H ^b^b: (* Non–Redundancy: Principle C effect)
b. # He1 likes Bill: j A ^m H ^b^b: (*Non–Redundancy: Principle C effect)
28
Schlenker’s analysis of backward anaphora configurations is as follows:
(39) [Bill’s teacher] likes Bill.
There is no violation of Non–Redundancy: «the key is that the VP hates Bill is
evaluated under a sequence that contains Bill’s teacher but not Bill himself, with the
result that Non–redundancy is satisfied (p. 16)». This implies that the subject NP
must be evaluated w.r.t. a different sequence from that of the matrix predicate.
(40) [His teacher] likes Bill.
Schlenker (note 15) argues that here his is a demonstrative pronoun, not an
anaphoric one (from what sequence could it possibly recover its antecedent?) 48
A2. A revision of Schlenker’s system
I propose the following two modifications:
i) The semantic computation must be made sensitive to phases: each phase has a
local sequence of evaluation.
ii) In addition, there is a global sequence of evaluation which keeps track of
potential antecedents (actual referents or discourse referents); a pronoun can
retrieve an element either from a local sequence (Reinhart’s A–binding) or from
the global one (Reinhart’s accidental coreference).
Let’s see in which way exactly Schlenker’s sequences must be local. Suppose we
assume a single sentence–level sequence of evaluation:
expression serves certain pragmatic purposes (e.g. adding expressive/evaluative information)
which could not be achieved by using a pronoun:
(i) John ran over a man who was trying to give the idiot directions.
48 As a matter of fact, Schlenker reproduces in different terms Reinhart’s dichotomy between
bound and unbound pronouns. But in his system A–binding no longer involves introducing a
variable in the antecedent position and co–binding of this position with the coindexed
pronoun by the same λ–operator; hence, it seems implausible that A–binding and coreference
should have such a radically different status as in Reinhart’s account. This point is
acknowledged by Schlenker (2006), who extends his analysis to anaphora across discourse.

(41) Bill likes [Ed’s teacher]
j A ^m H ^b
j A ^m H ^b^e
j A ^m H ^b^e^d (d=e’s teacher)
[[ like]] w j A ^m H ^b^e^d
A note on backward anaphora
By (iv) we obtain the wrong truth conditions (equivalent to: Ed likes Ed’s
teacher). Rather, the object DP must be evaluated w.r.t. a separate sequence s, and
returns to the matrix sequence only the result of the evaluation (=d).
(41) [[ like]] w j A ^m H ^b^d, where d= [ Ed’s teacher]] w s
In (42), in order for the pronoun his to recover its antecedent, the sequence of the
object DP must inherit what is contained in the matrix clause sequence:
(42) Bill likes [his–1 teacher].
[[ like]] w j A ^m H ^b^d, where d= [ his–1 teacher]] w j A ^m H ^b
In (43), in order for a violation of Non Redundancy to occur, the object DP must
inherit the content of the matrix clause sequence, which already includes b:
(43) * Bill likes [Bill’s teacher]
[[ like]] w j A ^m H ^b^d, with d= [[ the Bill teacher]] w j A ^m H ^b
[[ teacher]] j A ^m H ^b ^b (*Non-Redundancy)
On the contrary, in (44) the matrix clause sequence does not inherit what is
contained in the subject DP sequence, so that Non–Redundancy is not violated:
(44) [Bill’s teacher] likes Bill
[[ (17) ] w j A ^m H = [[ like Bill ] w j A ^m H ^t =
[[ like]] w j A ^m H ^t ^b
with t= [ the Bill teacher]] w j A ^m H (Schlenker 2005, 16)
In sum:
a) Sequences of evaluation must be local for any argumental constituent –
minimally, CP and NP.
b) The sequences are linked by inheritance relations which capture the effects of c–
command: the sequences of the embedded constituents inherit the content of the
matrix sequence, but only transmit back to it the result of their semantic
computation.
Let us adopt Chesi’s (2004) definition of top-down phases:
29

Valentina Bianchi
(45) A phase is the minimal part of a top–down computation in which all the
functional and selectional specifications associated to a given lexical phase
head (minimally, N or V) are satisfied (cf. Grimshaw 1991). 49
(46) i. Each CP or NP phase has a local sequence of evaluation.
ii. A subordinate phase na inherits the local sequence of the containing phase
n, in the state it has reached when the computation of na begins; na does not
transmit back its local sequence (with any elements added) to the
superordinate phase n, but only the result of the overall computation.
A global register, accessible from any phase, stores the contextually available
elements and keeps track of potential antecedents for inter–sentential anaphora:
(47) At any given point, the global register contains
i. the discourse referents for the speaker and addressee(s), singled out by the
superscripts A and H (replaces Schlenker's (iii)),
ii. Salient familiar discourse referents, in the sense of Roberts (2003, 320-
337), i.e., discourse referents pertaining to the immediate question under
discussion in the hierarchical discourse structure of Roberts (1996).
Non–salient familiar discourse referents, as well as novel referents, get to be
stored in the global register only if they satisfy the following necessary condition:
(48) Only an NP bearing new information focus can introduce an element in the
global register.
Note that an NP will introduce a referent in the global register only if it bears a
positive index. Anaphoric pronouns do not introduce anything, even when they bear
focus.
Given (48), Schlenker’s (i)-(ii) are modified as follows:
(i') An anaphoric pronoun copies an element from the global register into the last
position of the local sequence of evaluation of the current phase, or retrieves an
element from within the local sequence and moves it to the last position.
(ii') A focussed R–expression introduces an element both in the last position of the
local sequence and in the global register.
(iii') A non–focussed (destressed) R–expression introduces an element in the last
position of the local sequence, but not in the global register.
49 Chesi (2004a), Bianchi & Chesi (2006) draw a distinction between sequential and nested
phases which is irrelevant for the present purposes. I will collectively call them subordinate
phases, as opposed to the superordinate or containing phase.
30

A3. Analysis of (6)-(10)
A note on backward anaphora
In (6), the possessive pronoun copies an element g from the global register into
the local sequence of the NP phase. As this is subordinate in the matrix CP, no
sequence inheritance obtains. However, when the focussed R–expression is
processed, by (ii') it introduces g in the local sequence of CP, and also in the global
register: the latter violates Non–Redundancy.
(6) * [CP [NP His–1 wife ] loves JOHN]
GLOBAL: s A ^a H ^(m)^g (= g and possibly his wife m already introduced in
previous discourse)
NP: g (i')
CP: m(iii')^g (ii')
GLOBAL, at the end of computation: s A ^a H ^m^g^g(ii') (*Non–Redundancy)
The computation of (7) differs minimally: by (iii'), the destressed R–expression
introduces g in the local sequence of the matrix CP, but not in the global register;
hence, no violation of Non Redundancy arises.
(7) [CP [NP His–1 wife ] LOVES John]
GLOBAL: s A ^a H ^m^g (as above)
NP: g (i')
CP: m(iii')^g(iii') (global register unaffected)
In the examples (8), both the coreferring DPs are pronominal, and irrespective of
focus, they copy an element from the global register into the local sequences of the
DP and NP phases: these remain distinct, by (46ii), and Non–Redundancy is
fulfilled.
(8) a. [CP [NP My wife] loves ME].
GLOBAL: s A ^a H (^m)
NP: s A (i')
CP: m(iii')^s A (i')
(8) c. [CP [NP His–1 wife] loves HIM–1].
GLOBAL: s A ^a H ^g
NP: g (i')
CP: m(iii')^g(i')
In (9), by (iii') the first, destressed R–expression merely introduces g in the local
sequence of the subordinate DP phase, but it does not require that g be contained in
the global sequence. The second, focussed R–expression introduces g both in the
local sequence of the matrix CP and in the global sequence. At the local level, no
sequence inheritance obtains between DP and CP, by computational nesting; at the
global level, g is introduced only once. Hence, Non–Redundancy is fulfilled.
31

Valentina Bianchi
(9) [CP [NP John’s wife] loves JOHN]
GLOBAL: s A ^a H ^m
NP: g (iii')
CP: m(ii')^g (ii')
GLOBAL, at the end of computation: s A ^a H ^m^g(ii')
In (10a), the matrix subject introduces the element g in the local sequence of the
CP1 phase. This sequence is inherited by the sequential complement CP2 phase, s–
selected by the matrix V claim. When the lower R–expression is processed, it
introduces in the local sequence of CP2 the very same element g that the matrix
subject had already introduced: a violation of Non–Redundancy arises at the local
level. 50
(10) a.* [CP1 John claims [CP2 that Mary HATES John]]
GLOBAL: s A ^a H ^(g)
CP1: g (iii')
CP2: g(inherited from CP1)^m(iii')^g(iii') (* Non–Redundancy)
References
Barker, Chris & Shan, Chung–chieh. 2007. Donkey anaphora is simply binding. Ms., NYU &
Rutgers. Available from semanticsarchive.net.
Belletti, Adriana, Bennati, Elisa & Antonella Sorace. 2007. Theoretical and developmental
issues in the syntax of subjects: Evidence from near–native Italian. Natural Language
and Linguistic Theory 25, 657-689.
Bianchi, Valentina. 2001. Antisymmetry and the Leftness Condition: leftness as anti-ccommand.
Studia Linguistica 55, 1-38.
Bianchi, Valentina & Chesi, Cristiano. 2006. Phases, left branch islands, and computational
nesting. U. Penn Working Papers in Linguistics 12.1, 15-28.
Büring, Daniel. 2004. Crossover situations. Natural Language Semantics 12, 23-62.
Büring, Daniel. 2005. Binding Theory. Cambridge University Press.
Büring, Daniel. 2005b. Bound to bind. Linguistic Inquiry 36, 259-274.
Burzio, L. 1991. The morphological basis of anaphora. Journal of Linguistics 27, 81-105.
Burzio, Luigi. 1998. Anaphora and soft constraints. In Barbosa, Pilar et al. (eds.), Is the best
good enough?, Cambridge, MA, MIT Press, 93-113.
Calabrese, Andrea. 1986. Pronomina. MIT Working Papers in Linguistics 8, 1-46.
Carden, Guy. 1982. Backwards anaphora and discourse context. Journal of Linguistics 18,
361-387.
50 The same goes for (10b), where in addition, the lowest R–expression, being focussed, also
re-introduces the element g in the global register.
32

A note on backward anaphora
Carminati, Maria Nella. 2002. The processing of Italian subject pronouns. Doctoral
dissertation, University of Massachusetts at Amherst.
Chesi, Cristiano. 2004a. Phases and Cartography in Linguistic Computation. Doct. diss.,
University of Siena (http://www.ciscl.unisi.it/pubblicazioni.htm).
Chesi, Cristiano. 2004b. Phases and Complexity in Phrase Structure Building. Proceedings of
the 15th Meeting of Computational Linguistics in the Netherlands.
Chesi, Cristiano. 2007. An introduction to Phase–based Minimalist Grammars: why move is
Top–Down from Left–to–Right. In Moscati, E. (ed.), Studies in Linguistics (CISCL
Working Papers in Linguistics) 1. CISCL, University of Siena.
Chomsky, Noam. 1976. Conditions on rules of grammar. Linguistic Analysis 2.4, 303-351.
Di Eugenio, Barbara. 1990. Centering theory and the Italian pronominal system. COLING
1990, vol 2, 270-275.
Elbourne, Paul. 2001. E–Type Anaphora as NP–Deletion. Natural Language Semantics 9,
241-288.
Erteschik–Shir, Nomi. 1997. The Dynamics of Focus Structure. Cambridge University Press.
Gordon, Peter C. & Hendrick, Randall. 1997. Intuitive knowledge of linguistic co–reference.
Cognition 62, 325-370.
Grimshaw, Jane. 1991. Extended projection. Ms., Brandeis University.
Grosz, Barbara J., Joshi, Aravind K. & Weinstein, Scott. 1983. Providing a unified account of
definite noun phrases in discourse. Proceedings of the 21st annual meeting of the
ACL, 44-50.
Grosz, Barbara J., Joshi, Aravind K. & Weinstein, Scott. 1995. Centering: A framework for
modelling the local coherence of discourse. Computational Linguistics 21, 203-225.
Hardt, Daniel. 2003. Sloppy Identity, Binding, and Centering. Proceedings of SALT 13.
http://www.id.cbs.dk/~dh/publications.html (8 November 2007).
Hardt, Daniel. 2004. Dynamic Centering. Reference Resolution and its Applications, ACL–
Barcelona. http://www.id.cbs.dk/~dh/publications.html.
Heim, Irene. 1990. E–type pronouns and donkey anaphora. Linguistics and Philosophy 13,
137-178.
Heim, Irene. 1998. Anaphora and semantic interpretation: a reinterpretation of Reinhart's
approach. MIT Working Papers in Linguistics 25, 205-246.
Heim, Irene & Angelika Kratzer. 1998. Semantics in Generative Grammar. Oxford,
Blackwell Publishers.
Von Heusinger, Klaus. 2004. Choice functions and the anaphoric semantics of definite NPs.
Research on Language and Computation 2, 309-329.
Huang, Yan. 2000. Anaphora: A Cross–linguistic Study. Oxford: Oxford University Press.
Huang, Yan. 2007. Pragmatics. Oxford: Oxford University Press.
Kazanina, Nina. 2005. The acquisition and processing of backwards anaphora. Ph.D.
dissertation, University of Maryland, College Park.
Kiss, Katalin E. 1998. Identificational Focus versus Information Focus. Language 74, 245-
273.
33

Valentina Bianchi
Levinson, Stephen. 1991. Pragmatic reduction of the binding conditions revisited. Journal of
Linguistics 27, 107-161.
Reinhart, Tanya. 1981. Pragmatics and linguistics: an analysis of sentence topics.
Philosophica 27.1 (also distributed by Indiana University Linguistic Club).
Reinhart, Tanya. 1983. Anaphora and Semantic Interpretation. Chicago: The University of
Chicago Press.
Reinhart, Tanya. 1986. Center and periphery in the grammar of anaphora. In: Lust, Barbara
(ed.), Studies in the Acquisition of Anaphora, vol I, 123-150.
Reinhart, Tanya. 2000. Strategies of anaphora resolution. In: Bennis, Hans, Everaert, Martin
& Reuland, Eric (eds.), Interface Strategies, Amsterdam, North Holland.
Reinhart, T. 2004. Topics and the conceptual interface. In: Kamp, Hans & Barbara Partee
(eds.), Context Dependence in the Analysis of Linguistic Meaning, Elsevier, 275-305.
Reuland, Eric & Sergej Avrutin. 2005. Binding and Beyond: Issues in Backward Anaphora.
In A. Branco, T. McEnery and R. Mitkov (eds.), Anaphora Processing: Linguistic,
cognitive and computational modelling, John Benjamins.
Roberts, Craige. 2003. Uniqueness in definite noun phrases. Linguistics and Philosophy 26,
287-350.
Safir, Ken. 2004. The Syntax of (In)Dependence. Cambridge, MA, MIT Press.
Safir, Ken. 2005. Abandoning coreference. In J.L. Bermudez (ed.), Thought, Reference and
Experience: Themes from the Philosophy of Gareth Evans, Oxford, Oxford University
Press.
Safir, Ken. 2007b. Strict readings. Ms., Rutgers University. Available at: http://ruccs.rutgers.edu/~safir/strict.pdf
Schlenker, Philippe. 2003. A plea for monsters. Linguistics and Philosophy 26, 29-120.
Schlenker, Philippe. 2005. Non–redundancy: towards a semantic reinterpretation of binding
theory. Natural Language Semantics 13, 1-92.
Schlenker, Philippe. 2006. Minimize Restrictors! (Notes on definite descriptions, Condition C
and epithets). In Emar Maier, Corien Bary & Janneke Huitink (eds.), Proceedings of
Sinn und Bedeutung 9, 385-416. Nijmegen, NCS.
Schwarzschild, Roger. 1999. GIVENness, AvoidF and other constraints on the placement of
accent. Natural Language Semantics 7, 141-177.
Shan, Chung–chieh & Barker, Chris. 2006. Explaining crossover and superiority as left–to–
right evaluation. Linguistics and Philosophy 29, 91-134.
Sorace, Antonella & Filiaci, Francesca. 2006. Anaphora resolution in near – native speakers
of Italian. Second Language Research 22, 339-368.
Williams, Edwin. 1997. Blocking and anaphora. Linguistic Inquiry 28, 577–628.
Zubizarreta, Maria Luisa. 1998. Prosody, Focus and Word Order. Cambridge, Mass., MIT
Press.
34

Rivista di Grammatica Generativa
34 – 2009
SYNTACTIC DIAGNOSTICS FOR REFERENTIALITY
1. Introduction
MARKING IN EARLY NULL OBJECTS:
EVIDENCE FROM ITALIAN
Valentina Brunetto
Ellipsis of internal arguments is an issue which has long attracted a large number
of analyses and still continues to feed a heated debate, not only when the
phenomenon is regarded in its various attested constructions across (adult)
languages but also when its apparently deviant manifestations are looked at in child
language.
As for the first point, cross-linguistic comparison clearly shows that no such a
clear-cut generalization as a null object parameter is accountable by linguistic
analysis. Let us compare the following constructions, in which the internal object is
missing, although its properties are not amenable to a unifying account:
(1) A Maria disse que o João viu _ (Raposo, 1986)
Maria said that João saw _
(2) A painter disturbs _ much less than a writer (Goldberg, 2001)
(3) La buona musica riconcilia _ con se stessi (Rizzi, 1986)
Good music reconciles _ with oneselves
What is paramount here is that only in the example (1) the null object refers to a
specific antecedent, whose reference must be retrieved from the discourse. Lack of
specificity in the examples (2) and (3) suggests that such feature is at stake in
determining the appropriate conditions for the object to be omitted. That is, in a
language as Italian, the features on the null object are arbitrary, not inherited from
the discourse but rather under agreement with a licensing head. In such a case, we
are faced with a parametric option – which is confirmed in comparison with the
English example in (2), where the generic argument appears to semantically saturate
the verb without playing any role in syntax.
These are distinctions which we need to keep in mind if we want to give an
account of zero objects and the properties of such syntactic – if ever projected –
35

Valentina Brunetto
empty category. Narrowing the focus of our attention on null objects with a
[+specific] antecedent, we would like to maintain that discourse-anchoring is
ensured by movement.
In the spirit of recent speculations on the syntactic means by which a pronoun
comes to inherit reference from its antecedent (Kayne 2005; Belletti 2009), the idea
that the edge of the left periphery is where a pronoun must end up, in order to look
for its antecedent, is endorsed. If the null object construction precisely arises when a
highly salient topic is given in the discourse, thus allowing the recovery of the
content of the empty category, the analysis just hinted suggests that such linking can
only be obtained by means of movement to a peripheral, hence discourse-related
position.
That said, we would like to propose that the syntactic and discourse conditions
for a null object to be interpreted as referential are made available by UG and
recruited to some extent by limited-resourced populations as a case of computation
which may be less costly than cliticization. A case in point is language acquisition.
The question is not trivial as to whether children omitting a direct object are
actually marking specificity on it or rather producing a generic zero object
somewhat more freely than adults do. Although no uncontroversial uniformity as to
what to count as object omission has ever been reached in analyses on child
language 1 , it is commonly assumed that there is omission whenever a specific
object, that is to say, a pronoun, would be required to be overtly expressed.
Since the phenomenon has traditionally been reported to be significant in the
acquisition of Romance languages but basically absent in languages with weak/
strong pronouns, the special status of object clitics has generally been evoked in
most analyses of early object omissions.
We have utilised an adaptation to Italian of Costa et al.’s (2007; 2008)
experiment within the COST Action 33, aimed at investigating “cross-linguistically
robust stages of children’s linguistic performances” at age 5 (see also Varlokosta et
al. – to appear). The purpose of the present study is to try out the hypothesis that a
construction which is not found in the input – computing a variable rather than
producing an overt clitic – can be favoured nonetheless at the early stages.
If syntactic diagnostics for movement are borne out in children’s responses – as
it turned out in the last analysis, with island contexts drastically reducing the
1 The question has been recently raised by Pirvulescu (2006a) with respect to the
discrepancies which have been generally reported to hold between data coming from
spontaneous speech and elicitation studies. In actual fact, spontaneous speech analysis retains
a methodological degree of subjectivity in what to count as a transitive context, that is, the
syntactic context in which an object would be obligatory. The author individuates the “clitic
context” as the most homogeneous indicator of object omission in [+specific] discourse
conditions.
36

Syntactic diagnostics for referentiality marking in early null objects
availability of null objects – the datum can be taken to support an alternative
analysis of clitic omission, which does not need to resort to linguistic failure in
cliticization or pragmatic deficit. At least in some contexts, a discoursally-bound
variable can be arguably be discarded from deviant clitic omission.
2. Background analysis
For the ease of the descriptive analysis, we can borrow the definition of variable
from the inventory of empty categories theorised by the Binding Theory (Chomsky,
1981). Variables are A’-bound. This implies that they cannot be coindexed with any
matrix argument. Rather, they look for their antecedent at the periphery of the
matrix clause. Huang (1984) captured such derivation by postulating that Chineselike
null objects inherit their reference from topics which are salient in the discourse,
regardless of their overt or covert presence in the clause:
(4) [Top neige reni], Zhangsan shuo Lisi knanjian ei le
That man, Zhangsan says Lisi has seen
Movement to the left periphery is a crucial step of the interpretive chain
connecting the empty category and its antecedent. In other words, if the internal
object did not leave the VP, it could not properly look for its antecedent hence
failing to be interpreted as referential. We would like to maintain the following
analysis from Belletti (2009), which crucially captures the interpretive properties of
(silent) pronouns reconciling traditional binding principles with new cartographic
insights. Basically, the derivation through which a pronoun recovers its content
always involves movement of a silent pronoun to the edge of the complementizer.
The landing site must be an edge position, which avails itself of a privileged status
and functions as “hook” to the topic. This is true for clitics, and a traditional
Principle B effect as the following one finds easy explanation:
(5) Mariai dice [che tutti lai/k conoscono
Maria says that everyone her knows
Let us assume the direct object, generated in VP-complement position, is
doubled by a silent demonstrative, which reaches an unpronounced edge position
from whence it anchors to its antecedent. Then, the matrix subject Maria will act as
the antecedent if the landing site of such silent pronoun is the edge of the embedded
complementizer. However, the pronoun might well refer to an antecedent in the
discourse, salient enough to make use of the clitic felicitous. This is the case when
the silent pronoun lands to the main CP edge, looking outside.
For the present analysis of null objects, we assume that, instead of a computation
involving a topic, a (resumptive) clitic and a silent pronoun mediating the link at the
interface level, a more economical derivation allows a pronoun to reach a discourse-
37

Valentina Brunetto
related position without overt functional material inside the clause 2 . For a variable to
be discoursally-bound, it must reach the edge of a matrix CP: in fact, typical
Principle C effects ultimately follow from the same reasoning. Namely, null objects
are ruled out inside islands because they cannot move to the main edge CP to license
their referentiality; likewise, null objects display Strong Cross Over effects, due to
the fact that they need to look outside the matrix subject to recover their content.
The ambiguity reported in (5) for the interpretation of the clitic would not hold in
the following sentence in Portuguese:
(6) A Mariai disse [que o João viu ek
Maria said that João saw _
The variable must move to the main CP to recover its content, which is set by the
context. That is to say, a salient topic must be available in the discourse for a null
object construction to be appropriate.
3. The experimental rationale
Under the conditions thereof, it follows that a null category is licensed under
referential interpretation only if it is allowed to reach a discourse-related position in
syntax.
If continuity is a warranted paradigm to account for child behaviour, and the
options children select during the course of language acquisition are crucially
recruited from UG, the proposal we want to advance is the following: the early
system favours a null object construction at the first stages, as a case of option which
is effort-saving and compatible with the performance limitations of the early system.
Italian is a case in point since no such parametric option is available in the input.
Indeed, several studies have shown that object clitics emerge by the third year of life
and no deviancy is attested as to clitic placement. Rather, a certain degree of
optionality is retained over development. We would like to maintain that such
optionality is (at least partly) caused when the competitive availability of a null
object construction wins over – until positive evidence consistently indicates
otherwise and the option is abandoned.
If children’s null objects are discourse-linked, the same restrictions on the
derivation of such empty category should hold. Put it differently, if children mark
referentiality, in the discourse condition in which a salient topic is prompted as an
2 The alternation between null objects and pronouns so proposed cannot fully capture the
status of overt pronouns outside islands – where, in principle, movement would be possible
hence systematically preferred. Nevertheless, it well accounts for resumption in contexts
where pronouns “save” impossible extraction. In this sense, doubling is compatible with the
idea of movement as copy + (partial) deletion which is a tenet of the minimalist thesis.
38

Syntactic diagnostics for referentiality marking in early null objects
antecedent, precise syntactic tests can allow us to see in more clarity whether
children comply with Principle C.
Two conditions were designed in the present test. The first condition set a
felicitous discourse context for an object clitic to be elicited (after Schaeffer 2000):
(7) Experimenter: Che cosa sta facendo il cane al gatto?
What is the dog doing to the cat?
Child: ...lo lecca
…him licks
Such task was assumed as a sort of baseline, to measure children’s mastery of
clitics and level of omissions in root contexts. Data were then compared with
findings coming from the same subjects on a second task, prompting a response
inside an island, as in Costa et al. (2007; 2008) 3 . Such syntactic condition was
individuated as a reliable cut-off between variables and omissions.
(8) Experimenter: Il cane lecca il gatto e ora il gatto è contento. Perché il gatto è così
contento? Il gatto è contento [perché ...
The dog is licking the cat and now the cat is happy. Why is the cat
so happy? The cat is happy because the dog…
Child: …[perchè il cane lo lecca
…because the dog him licks
At a closer look, there are two strong constraints on the availability of a variable
in such sentence. The DP [il gatto] is given in the prompt as the unique relevant
antecedent for the expected clitic. The unavailability for an empty category to
appear inside the island stems from the impossibility to reach its antecedent – the
only way it would inherit referentiality. In fact, as the adjunct CP qualifies as an
island for extraction, no silent pronoun can be postulated which moves to the matrix
CP edge to link to the discourse. For the sake of our theoretical analysis, let us
compare the entire derivation of a null object in the two conditions 4 :
3 This experiment was conceived as an adaptation of Costa et al.’s (2007; 2008) within the
COST Action A33, aimed at investigating “cross-liguistically robust stages of children’s
linguistic performances” (see also Varlokosta et al. – to appear). The authors’ study on
European Portuguese was intended to tease out target-like null objects from clitic omissions.
For this purpose, root and island contexts were controlled in elicitation, yielding two main
findings: first of all, there is a prolonged omission stage in EP (contrary to UCC’s
predictions); secondly, children have knowledge of the syntactic conditions for referential null
objects, since they tend to “save” island constraints by producing more DPs.
4 In detailed steps, we have represented movement of the (silent) pronoun through the vP
edge. For interpretive requirements, we assume in fact that VP arguments also pass through
the right periphery to be interpreted as topics or foci (Belletti 2004).
39

Valentina Brunetto
(9) [CP Il gattoi … [CP luii [IP lecca [vP luii [VP V luii
40
[CP luii Il gatto*i è felice [luii perché il cane lecca [ vP luii [V luii
*
Not only movement is ruled out past the adjunct, but also coreference with the
matrix subject [il gatto] is impossible, since Strong Cross Over arises – as mentioned
above. In such a configuration, presence of an overt pronoun is obligatory across
languages, as it has been abundantly shown in the literature – see the classical
analyses on Chinese (Huang 1984) and European Portuguese (Raposo 1986).
In the following sections, our findings will be presented showing that children
acquiring Italian appear to reduce omissions in the second condition: a result which
only finds a coherent explanation if we assume omissions to be accountable – at
least in part – as variables exploited in the proper syntactic and pragmatic
conditions.
4. Data
4.1 Subjects
The experiment was administered to children in preschool age, who were ideally
divided into three age groups for further analysis in developmental perspective. 43
children participated in the first condition, divided as follows: 18 three-year-olds
(mean 3.6); 14 four-year-olds (mean 4.5); 11 five-year-olds (mean 5.7). 53 children
were tested in the second condition: the youngest group amounted to 19 subjects
(mean 3.7); four-year-olds to 14 (mean 4.5); five-year-olds to 20 (mean 5.6).
As said, the same children were selected when possible in order to render
comparisons between conditions as much reliable as possible.
4.2 Items
Materials consisted of 14 pictures, portraying two characters performing a
transitive action, along the lines of the examples mentioned in (7) and (8). The same
pictures served the two testing condition. In addition, 5 fillers were inserted which
prompted either a different clitic (two items represented a self-oriented action) or an
(optional) lexical DP. In the latter case, the prompt did not mention the object
directly (“what is X doing?”), such that a non-specific object was targeted.
5. Results
5.1 Results in the root condition
The following categories were controlled in children’s responses: clitics (the
expected answer); full DPs; strong pronouns; omissions. Specifically, use of clitics

Syntactic diagnostics for referentiality marking in early null objects
and full DPs and their presumed trade-off has always been a matter of debate in the
analyses on children’s spontaneous and elicited productions (Jakubowicz et al.
1998).
The present findings partly show similar rates of clitics and omissions at age
three as compared to Schaeffer (2000) 5 :
Age Clitic DP Omission Strong Other
5 year-olds 87.01% 5.84% 4.54% 0.59% 1.94%
(134/154) (9/154) (7/154) (1/154) (3/154)
4-year-olds 88.77% 3.57% 5.10% 1.53% 1.02%
(174/196) (7/196)) (10/196) (3/196) (2/196)
3-year-olds 71.82% 5.95% 15.87% 1.58% 4.76%
(181/252) (15/252) (40/252) (4/252) (12/252)
Table 1: Results in root contexts
However, omission levels depart from the datum reported by the author that the
phenomenon ceases at age 4. In actual fact, a certain optionality in clitic use is
retained in development, which is not fully target-like still at age 5. On the other
hand, it appears that clitic use is mastered at target-like levels from age 4, with
values approaching 90%. Not so in the youngest group, with a rate of clitic
production around 70%, which indicates that a developmental difficulty in mastery
of cliticization is at play. Focussing on this age span, it can be observed that a
crucial window is that before age 3.5:
25
20
15
10
5
0
22,61
12,5
5,1 4,54
3.0-3.5 ys 3.5-4.0 ys 4-5 ys 5-6 ys
Table 2: Developmental rate of omissions in root contexts
5 The following rates of omissions are reported in Schaeffer (2000): 64% at age 2; 15% at age
three; virtually disappearing from age 4.
41

Valentina Brunetto
42
100
80
60
40
20
0
60,71
77,38
88,77 87,01
3.0-3.5 ys 3.5-4.0 ys 4-5 ys 5-6 ys
Table 3: Developmental rate of clitics in root contexts
Crucially, a developmental trend is detected between the third and the fourth
year of life, with omissions almost halving after age 3.5 and clitic production
converging very soon towards the target values.
Figures in DP production do not yield a linear trend. In this respect, the
hypothesis that children might favour production of full lexical DPs in place of
deficient clitic pronouns is not consistently supported by our data. Pronominalization
of the object in such experimental setting appears as the most natural
choice even at the youngest age, and the most common inconsistencies reported are
omissions. Along the avenue pursued throughout the experiment, we asked how
many omissions among these – if any – were amenable to an analysis in terms of
topicalization without clitic resumption. Hence comparison with a different testing
condition, designed to render a variable impossible to be licensed, was the second
step of the analysis.
5.2 Results in the island condition
Prompting a clitic inside an adjunct clause, introducing a causal relation between
the characters mentioned by the experimenter, enhanced clitic production and
constrained omissions, across all age groups.
Overall, a trade-off was detected between clitics and zero objects. Rates of
lexical DPs were not influenced by the testing condition, which is a datum we will
come back to.
Age Clitic DP Omission Strong Other
5 year-olds 91.78% 3.21% 2.85% 0.71% 1.42%
(257/280) (9/280) (8/280) (2/280) (4/280)
4 year-olds 91.83% 3.06% 3.57% - 1.53%
(180/196) (6/196) (7/196)
(3/196)
3-year-olds 79.32% 4.88% 8.64% 1.87% 5.26%
(211/266) (13/266) (23/266) (5/266) (14/266)
Table 4: Results within islands

Syntactic diagnostics for referentiality marking in early null objects
The testing condition was administered in the form of a sentence completion task
and involved elicitation of the whole adjunct CP (eg. …perché lo lecca) such to
ensure that the child were computing a full clausal structure. Overall, due to the very
neutral nature of the sentential choice, children encountered no difficulties in
interpreting theta roles appropriately. However, the higher syntactic complexity
probably resulted in higher number of non-responses in the youngest group (yet,
crucially, not in higher number of omissions).
Putting the two conditions in perspective, the most striking datum is a fall-off in
omissions. If four and five-year-olds already perform close to target-like in the root
condition, inside islands omissions drop further. Rates of clitic production indicate
that mastery levels are attained at age four and thence no linear trend emerges.
Internal variability in the oldest group is to be addressed to explain the slightly lower
performance in clitic production – albeit insignificant. Instead, linearity is
undisputable in the developmental trend of null objects across age groups.
The youngest group offers the analysis new evidence that the syntactic context
can circumscribe indeed the number of omissions. Such observation amounts to
acknowledging that, to a certain extent, object drop obeys grammatical regularities.
The present findings provide a percentage of clitic omissions estimated at 8% in
island context, a datum for which there is no parallel evidence on early Italian
available at present – yet a significant 15% in root clauses, not only yielded from the
very same subjects who participated in this task, but also confirmed by Schaeffer
(2000).
Again, narrowing the three-year-old group further, a developmental shift can be
seen to occur at very fast rates. In actual fact, mean values risk to obscure that lower
performances are observed in subjects younger than 3.5, while values improve
significantly in the second half of the year eventually converging towards the nearmastery
levels observed from age 4.
20
10
0
15,71
6,12
3,57
3.0-3.5ys 3.5-4ys 4-5ys 5-6 ys
Table 5: Developmental rate of omissions inside islands
2,85
43

Valentina Brunetto
44
100
90
80
70
60
50
40
30
20
10
0
68,57
83,16
91,83 91,78
3.0-3.5 ys 3.5-4 ys 4-5 ys 5-6 ys
Table 6: Developmental rate of clitics inside islands
Omissions drop in the course of the third year of age from 15% to almost one
third reported on the oldest subjects, who start to track with the target value.
Similarly, cliticization appears to be mastered from at least age 3.5, with values well
above 80%. Comparing the two subgroups within the age span of concern, a rapid
development can be observed to be in place. However, still in the youngest subjects
clitics are produced at a rate close to 70%, which is fairly a better performance than
that observed in the same subjects on matrix clauses. With respect to what reported
in the literature, this datum adds important evidence that three-year-olds can indeed
boost clitic production under proper syntactic conditions.
6. Animacy and the status of the empty category
Over and above, a subtler analysis revealed that most zero objects were elicited
with inanimate antecedents. Three items out of fourteen in the test involved a
transitive action carried out on an inanimate object for which, in principle, the
intransitive form of the verb does not yield ungrammaticality. It is the discourse
condition designed for the task that renders an object specified for the features [animate;
+specific] obligatorily expressed: hence a clitic pronoun. It follows from
Cardinaletti and Starke’s (1999) typology of pronouns that the structural properties
of clitics are at the origin of their underspecification for animacy. Children are well
aware of the semantic properties of overt pronouns, as confirmed in the present
study by the exclusive use of strong pronouns with [+animate] antecedent and in
deictic-like responses, that is, when the object was made salient by the child by
pointing at the drawing.
Measuring children’s performances across age groups according to the animacy
feature of the objects in the items, it is rather interesting that not only inanimate
items induce zero objects at higher extent overall, but also the asymmetry was
enhanced in root clauses and more significant in the youngest group of children.

Syntactic diagnostics for referentiality marking in early null objects
31,48
11,61
9,52
12,12
3,89 2,47
3-4 ys 4-5 ys 5-6 ys
[+animate]
[-animate]
Table 7: Developmental rate of omissions in root contexts: [±animate] objects
17,54
6,69
4,76
3,24
3-4 ys 4-5 ys 5-6 ys
5
2,27
[+animate]
[-animate]
Table 8: Developmental rate of omissions inside islands: [±animate] objects
As for the youngest children, considering only inanimate objects raises
percentages above 30%, which is strikingly higher but nonetheless aligns to the
figures reported by other studies (see Pérez-Leroux et al. 2007 on French 6 ; Castilla
et al. 2007 on Spanish 7 , who crucially made use of inanimate items in the tests).
Such asymmetry suggests that the [-animate] feature of the object plays a pivotal
role in inducing preference for ellipsis.
Pérez-Leroux et al. (2007) have called the attention on the far from clear-cut
distinction between argumental and pronominal omissions. When ellipsis concerns
generic inanimate objects, driving preference for an intransitive use of the verb in
the first place, what is missing is a lexically saturated argument, but not an empty
category in the sense of Chomsky (1981).
6 In Pérez-Leroux et al. (2007) French children were reported to omit the clitic at a rate of
34% at age 3, which is significantly higher than the 15.7% reported in Jakubowicz and Nash
(2001).
7 Castilla et al. (2007) reported 37% clitic omissions at age 3, dropping in development but
maintained at a level of 15% up to age 5.
45

Valentina Brunetto
When the missing object is marked in syntax as [+specific] the empty category
belongs to a different typological status, which we assumed from a long generative
tradition to be defined as a variable. In this case, animacy may not be at play, rather
the referential interpretation of the variable. Therefore, [+specific] variables should
be illicit inside islands regardless of the [±animate] feature of their antecedent. In
actual fact, the children in the experiment tend to align [±animate] null objects
inside islands from age 4. This might be taken as evidence that the plunge in
omissions in the island condition follows from them being true variables in the root
condition. In matrix clauses dropping of inanimate objects appears rather free and
less tied to age.
Three-year-olds halve omissions in the adjuncts but a considerable discrepancy
holds in both conditions between [±animate] null objects. In development, it seems
that the testing condition constrains dropping of inanimate objects from age 3.5.
Children younger than 3.5 provided lower omissions with [+animate] objects inside
islands but similar rates with inanimate items, regardless of the matrix vs. adjunct
nature of the clause.
46
27,71
21,21
33,32
6,81
3.0-3.5 ys 3.5-4.0 ys
Table 9: Three-year-olds: omissions in root contexts

Syntactic diagnostics for referentiality marking in early null objects
26,6
14,54
14,3
3,9
3.0-3.5 ys 3.5-4.0 ys
[+animate]
[-animate]
Table 10: Three-year-olds: omissions inside islands
7. What types of null objects in early grammar
The frame captured by the present analysis is far from neat. The complex
interplay of animacy and specificity in children’s zero objects suggests that
typological differences as for the status of the missing object must be taken into
consideration when interpreting child data.
As a starting point for reflection, the hypothesis was borne out that object
omissions are reduced in an illicit syntactic environment for an operator-variable
construction to arise.
This is true across all the developmental span analysed but to uneven extent and
with asymmetrical levels of preference for inanimate objects.
Moreover, development is confirmed once again to occur at incredibly fast rates.
The exploration of child phenomena at the interface is confronted with an interplay
of factors whose concomitant emergence in typical development is hard to unfold,
affecting complex aspects of language and cognition.
The early system has to cope with performance limitations and constricted
computational and memory resources. However, experimental evidence suggests
that, rather than resorting to extragrammatical explanations for ellipsis phenomena,
we should adopt an approach reconciling immaturity of the early system with
continuity in the grammatical competence children are endowed with (Rizzi 2005).
In the case of argumental ellipsis, early subject omission in non-NSLs has long
prevailed in the interest of researchers in the controversial search for evidence for
the continuity/discontinuity dichotomy in language acquisition. Among competing
approaches, the first line has been able to provide undisputable evidence that
grammatical regularities are at play in children’s omission. The option of truncating
full clausal structures, by producing “less” in the appropriate syntactic conditions,
47

Valentina Brunetto
retains more explanatory adequacy if refined from maturational arguments,
especially in the light of the data coming from special populations 8 .
The findings from the present analysis offer support to the hypothesis that
ellipsis, albeit favoured by the early system, nonetheless obeys precise regularities
which call for a grammatical option. However, several considerations need to be
added to discuss the asymmetries observed.
Inasmuch as the island condition should offer reliable rates of true omissions –
any construction which is not to be accounted for as a variable – it can be stated that
these are extremely circumscribed from age 3.5 onwards (table 10). The sheer
asymmetry between animate and inanimate null objects in root condition is reduced
in the island condition, which might be taken as evidence that children do also
interpret inanimate null objects as specific. Hence we do not expect those omissions
to be sensitive to the animacy feature of its antecedent. Still, the slightly higher level
of inanimate zero objects might appertain to a different option, namely argumental
omission intended as ellipsis of a generic internal argument s-selected by the verb.
Being interpreted as [-specific], such option is universally available and insensitive
to islands. Assuming that it might be adopted in development as well does not
amount to saying that children allow generic internal arguments to be interpreted as
referential 9 : in actual fact, data indicate that omissions follow a linear path towards
disappearance in the island condition, while inanimate objects are dropped to a
considerable extent still at age five in matrix clauses (table 2), that is to say, in the
appropriate syntactic condition to ensure discourse-linking of the silent category.
Those objects are reminiscent of deictic-like null objects, universally attested as
inanimate but crucially [+specific] – well chronicled by Sirguðsson (2004a; 2008)
with respect to special registers. As Cyrino and Lopes (2006) pointed out, those are
also the first null objects to be attested at the early stages of the acquisition of
Brazilian Portuguese, a language in which loss of third person clitic pronouns has
led to prevalent use of strong pronouns and null objects, respectively, for animate
and inanimate antecedents.
As for the Italian children’s preference for the null objects construction in those
items in which the antecedent was inanimate, input considerations might be of
interest. Cyrino, Duarte and Kato (2000) appealed to a sort of “referentiality
hierarchy” to explain the concomitant role of animacy and specificity in the types of
null objects which diachronically began to spread in Brazilian Portuguese. Arguably,
primacy of [-animate] null objects is attested in acquisition as well and first
8 Therefore, from the assumption that children do not identify the root of the clause with the
CP layer to the broadened speculation that special communication demands leave the option
open to truncate structural layers.
9 This is the position endorsed by Pérez-Leroux, Pirvulescu and Roberge (2007; 2008).
48

Syntactic diagnostics for referentiality marking in early null objects
occurrences in deictic contexts appear not to be a prerogative of null object
languages.
8. Null objects, clitic drop and alternative views
A long accepted analysis has generally paralleled object drop to a phenomenon
exclusively appertaining to the acquisition of Romance languages.
The relatively circumscribed amplitude of omissions in the domain of languages
without clitic pronouns has led to the – perhaps precipitate – conclusion that a null
object stage is not universally attested, rather failure in the costly computation of
accusative clitics is at the origin of what should be properly addressed as clitic drop.
Complexity of accusative clitics has been extensively documented, both from
theoretical analyses (Jakubowicz et al. 1996; Belletti 1999) and from genuine
linguistic evidence. The clitic system is undisputedly one of the most fragile ones in
first and second language acquisition, as well as delayed language acquisition and
linguistic breakdowns. In addition to the syntactic derivation of such peculiar
category halfway between the functional and lexical field, forced to move to check
overt case and φ-features due to its deficient D° structure, more recent
considerations have addressed the categorial status as a more explanatory account
for the pernicious difficulty of accusative clitics in the realm of clitic pronouns.
Categorial complexity in the precise sense that, as arguments merged in Vcomplement
position and bearing a theta role, they do not uniform to full object
DPs, which renders their categorization more difficult (Rizzi 2000; Hamann 2003).
Our analysis acknowledged the shared interpretive properties of cliticization and
silent topicalization without clitic resumption. Assuming that both clitics and null
objects share discourse-relatedness, we have upheld the hypothesis that a doubling
structure is at play in cliticization, where part of the computation takes place in the
inflectional field (the clitic derivation yielding, accordingly, either proclisis or
enclisis) and part at the interpretive interface, traditionally identified in the CP field
(Rizzi 1997). The trade-off we found between clitics and omissions, in the testing
condition in which referential interpretation of a null object was illicit, might be
taken as indication that two options are competing in children’s grammar, one of
which less costly hence favoured at the early stages.
Under such view, no deviancy is postulated in children’s interpretation of the
object as referential, even when this is not pronounced. Assuming this position
implies departing from those approaches that identify omissions with clitic drop.
In Schaeffer (2000) failure to licence discourse-relatedness on the clitic is
assumed to follow from lack of movement of a base-generated pro to the clitic head
in the Infl node. If we accept the movement analysis as pivotal for a descriptive
account of the syntactic means through which a clitic pronoun licences its reference,
nevertheless pursuing the hypothesis of a complete isomorphism between zero
49

Valentina Brunetto
objects and clitic drop leads to the inevitable conclusion that children never mark
specificity when omitting the clitic.
This is exactly what such accounts proposes, connecting deviancy to a non-yetdeveloped
pragmatic rule. The immature early system does not need to licence
discourse-relatedness on the object clitic because no difference is presumed in the
state of knowledge between speaker and addressee.
If the child were not producing a d-linked empty category, no difference would
have been found between matrix and adjuncts clauses, contrary to facts.
Not only pragmatic competence has been investigated in depth in child studies,
with consistent indications that children have discourse notion from very early on,
but also this factor cannot easily fit with clitic omissions in populations in which
linguistic deficit is dissociated from other cognitive modules.
Maturation arguments have been adduced in Wexler’s (1998; 2003) account of
clitic omissions as well. The Unique Checking Constraint, assumed to hold in the
non-mature child grammar against multiple feature checking, besides being uneasy
to reconcile with a continuity view of language acquisition, runs once again into the
risks mentioned above. Firstly, such a pro would not be referential; secondly,
participle agreement errors would be visible (which is a very disputed argument)
when the clitic is spelled out and the converse when the clitic is omitted.
In cross-linguistic perspective, theories advocating clitic drop must presuppose
absence of a null object stage in languages with a strong/weak pronominal system.
The Unique Checking Constraint, in addition, implies a further distinction within the
Romance languages with and without over past participle agreement. In actual fact,
its prediction has not been wholly borne out, as recent analyses on European
Portuguese showed (Costa et al. 2007; 2008). Interestingly, European Portuguese is
a language without past participial morphology. But even more interestingly,
European Portuguese is a null object language.
9. Remarks on the null object stage in cross-linguistic perspective
In recent times, Castilla et al. (2007; 2008) have questioned the radical claim –
dating back at least to Brown’s (1973) corpus – that there is no null object stage in
the acquisition of English.
Clearly, object omissions (to be intended as pronominal omissions, in a definite
condition) in this language are hardly explained by theories resorting to clitic drop.
Nevertheless, evidence is not uncontroversial. To the low percentages reported by
Bloom (1990) and Hyams and Wexler (1993) among others, Pérez-Leroux et al.
50

Syntactic diagnostics for referentiality marking in early null objects
(2008) have opposed much higher values 10 , crucially yielded with [-animate]
objects, in elicitation.
Under the proposal to assimilate object omission to a true UG option, namely an
operator-variable construction, these data do not come inexplicable.
We have proposed that a null object construction, whence a variable is A’-moved
to inherit referentiality from a topic salient in the discourse, is a UG option children
may recruit at the first stages of acquisition, when ellipsis is the most favourable
resort for the early system.
Such option crucially involves the complementizer system. As no internal
licensing is admissible under agreement with a functional head in the Inflectional
node, we must suppose that such object reaches a discourse-related position to
“hook” to its antecedent. Edge positions are dedicated sites for elements which need
to be “visible” to such topic to be properly interpreted. Such idea has been
maintained and reformulated in Phase terms (Chomsky, 2001; Kayne 2005) as a
computational device which sends to spell out the complement of a phase but leaves
its edge active in the computation, available for further operations. Therefore, the
special status of edges rests on economical grounds. We may hypothesise that what
children interpret producing a null object is a silent demonstrative. Crucially, such
derivation is not as costly as producing an overt pronoun. This is why we may
assume early object omissions to arise in those cases in which this construction wins
over, regardless of cliticization being the alternative. Given the well known
asymmetries in the acquisition of the pronominal system in languages with and
without clitic pronouns, it would be expected cross-linguistically that clitic omission
spanned a longer stage. Nevertheless, a more complex interplay of factors might be
involved in the developmental path, among which the role of the input cannot be
underestimated. In this regard, Costa et al. (in press) have noted that children
acquiring European Portuguese might undergo a more extensive phase of clitic
omission due to the complexity of alternative syntactic derivations they are exposed
to. Namely, availability of referential null objects in the input, in place of overt third
person accusative clitic pronouns, renders the latter the most fragile category in the
clitic system. In this sense, it is not so much the structural complexity of the pronoun
and the features encoded in it, as the competition of an alternative derivation – in the
author’s, referred to as Post-syntactic Complexity – that exposes accusative clitics to
a broader stage of optionality.
We should add that the complexity of cliticization is still a factor children
confront with in development. In the present experiment the island condition,
forbidding in principle a null object, actually yielded higher occurrences of clitics
from very early on – nonetheless, percentages were still far from the target in the
10 The former claimed that omissions are in the region of 8-9% at age 2, but disappear by age
3; the latter reported a plunge in omissions from 35%, at age 2, to 8% at age 3.
51

Valentina Brunetto
youngest subjects, indicating that time is needed for cliticization to be properly
mastered. Instead, the same constraint on Portuguese children was not “saved” by
higher supply of clitics, rather by a significant resort to full DPs. This datum
becomes meaningful especially on the light of the complex pattern of proclisis/
enclisis EP displays in finite clauses. Crucially, placement errors have been found to
be peculiar to this language – with a generalised preference for enclisis – not only in
acquisition but also in the registers of young generations (Duarte and Matos 2000).
Arguably, the competition between accusative pronouns and null objects belongs
to the parametric realm of UG options. This may be one added reason for the
universally attested “delay of complement clitics” (Hamann 2003) and their
pernicious optionality span.
10. Conclusions
The speculative reading we have proposed for the findings discussed is that early
null objects might be referential empty categories moved to a discourse-related
position in syntax. Children appear to “exploit” the privileged syntactic status of
root clauses to allow a null category – instead of an overt pronoun – to inherit
reference from the discourse. Even at the youngest age, omissions have been found
to drop in a syntactic condition in which an empty category would fail to be
discoursally-bound – in the testing condition, both because of intervention of an
island for extraction and of a matrix subject coreferent with it.
Children’s preference for inanimate objects deserves a multi-angled analysis
since a different typological status is at play if such object is non-referential. We
have maintained that both are universally available but only the first is sensitive to
islands. Such analysis should explain higher occurrences of inanimate null objects
overall but, at once, a certain sensitivity to islands which suggests that most
children’s inanimate null objects are deictic-like referential categories.
Aligning children’s null objects to the operator-variable construction well
documented in languages like Chinese or Portuguese meets precise theoretical
consequences. Firstly, it implies questioning that children’s null objects derive from
failure in performing clitic movement. Therefore, no pragmatic deficit is assumed if
children appear to exploit a null object option in the appropriate discourse condition
and in presence of a salient topic, which in turn implies that specificity is encoded in
such category. Finally, a complete parallel between null objects and clitic drop
would not be able to explain the – albeit reduced – phenomenon in languages
without clitic pronouns. Cross-linguistic evidence might prove relevant to bring the
debate outside the traditional domain of Romance languages. The tie between
acquisition studies and comparative syntax rests on the precious angle that child
phenomena can offer on the path of parameter setting, between positive evidence
and performance limitations.
52

References
Syntactic diagnostics for referentiality marking in early null objects
Belletti A. (1999) Italian/Romance clitics: Structure and derivation, in H. van Riemsdijk (ed)
Clitics in the Language of Europe, pp: 543-579. Mounton de Gruyter, Berlin.
Belletti, A. (2009) Pronouns and the edge of the clause. In A. Belletti, Structures and
Strategies, Routledge, Chapter 11.
Bianchi, V., M. C. Figueiredo-Silva (1994) On some properties of agreement-object in Italian
and Brazilian Portuguese. In Mazzola, M. (ed.) Issues and theory in Romance
linguistics, Washington, DC: Georgetown University Press.
Bloom, P. (1990) Subjectless sentences in child language, Linguistic Inquiry 21, pp. 491-504.
Brown, R. (1973) A First Language: the Early Stages. Cambridge, Mass.: Harvard University
Press.
Cardinaletti A. & M. Starke (1999) The typology of structural deficiency: A case study of
three classes of pronouns, in H. van Riemsdijk (ed) Clitics in the Language of Europe,
pp. 145-288. Mounton de Gruyter.
Castilla A., A. Pérez-Leroux & A. Eriks-Brophy (2007) Omission in early Spanish clitics. In
M. João Freitas and A. Gavarró Algueró (eds). Language Acquisition and
Development: Papers from GALA 2007. Cambridge: Cambridge Scholars Publishing,
112-122.
Castilla A., A. Pérez-Leroux & A. Eriks-Brophy (2008) Syntax and the Lexicon in Early
Omission of Spanish Clitics. Proceedings of the Annual Boston University Conference
on Language Development, 32, 1, 72-83.
Cecchetto, C. (2000) Doubling Structures and Reconstruction. Probus, 12, 1: 93-126.
Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht.
Chomsky, N. (2001) Derivation by phase. In M. Kenstowicz (ed.), Ken Hale. Cambridge,
Mass.: MIT Press, pp. 1-52.
Chomsky, N. (2005) On Phases, in R.Freidin, C. Otero and M.L.Zubizarreta eds., Fundational
Issues in Linguistic Theory, MIT Press Cambridge, Mass.
Costa, J. & M. Lobo (2007a) Clitic omission, null objects or both in the acquisition of
European Portuguese? In Sergio Baauw, Frank Drijkonongen & Manuela Pinto (eds.)
Romance languages and linguistic theory 2005. Amsterdam/Philadelphia: John
Benjamins, pp. 59-71.
Costa, J., M. Lobo (2007b). Complexidade e omissão de clíticos: o caso dos reflexos. In A.
Coutinho & M. Lobo (eds.) XXII Encontro Nacional da APL. APL/Colibri, Lisbon,
303-313.
Costa, J., M. Lobo, J. Carmona, C. Silva (in press) Clitic omission in European Portuguese:
correlation with null objects? In A. Gavarró and M. João Freitas, (eds.). Language
acquisition and development: Proceedings of GALA 2007. Cambridge Scholars Press,
Newcastle, 133-43.
Cummins, S., Y. Roberge (2005) A modular account of null objects in French. Syntax 8(1),
44-64.
53

Valentina Brunetto
Cyrino, S.M.L; Duarte, M.E.L. & Kato, M. (2000) Visible subjects and invisible clitics in
Brazilian Portuguese. In Kato, M. A. e Esmeralda Vailati Negrão (orgs) Brazilian
Portuguese and the Null Subject Parameter, Frankfurt & Madrid, Vervuert-
Iberoamericana, pp. 55-73.
Cyrino S., R. Lopes (2005). Animacy as a driving cue in change and acquisition in Brazilian
Cyrino, S., G. Matos (2006) Anáphora do complemento nulo: anáphora profonda ou
de superficie? Evidência do Português Brasileiro e Europeu. Letras de Hoje, Porto
Alegre, v. 41, n.1, pp. 121-141.
Duarte, I. & G. Matos (2000) Romance Clitics and the Minimalist Program. In Costa, J. (org.)
Portuguese Syntax. New Comparative Studies. Oxford: Oxford University Press, pp.
116-142.
Hamann C. (2003) Phenomena in French normal and impaired language acquisition and their
implications for hypotheses on language development, Probus 15: 91-122.
Huang, J. (1984) On the Distribution and Reference of Empty Pronouns, Linguistic Inquiry,
15: 4.
Hyams, N., K. Wexler (1993). On the grammatical basis of null subjects in child language.
Linguistic Inquiry 24: 421-459.
Hyams N., J. Schaeffer (2007) Clitic and auxiliary omissions in Italian children participle
constructions. In A. Gavarró and M. João Freitas, (eds.). Language acquisition and
development: Proceedings of GALA 2007. Cambridge Scholars Press, 292-302.
Jakubovicz C., N. Müller, O. Kang, B. Riemer, C. Rigaut (1996) On the Acquisition of the
pronominal system in French and German, in A. Stringfellow, D. Cahana-Amitay, E.
Huges and A. Zukowski (eds.) Proceedings of the 20 th Annual Boston University
Conference on Language Development, Sommerville, MA: Cascadilla Press 1, 331-
342.
Jakubowicz, C., L. Nash, C. Rigaut & G. Christophe-Loic (1998). Determiners and Clitic
Pronouns in French-Speaking Children with SLI. Language Acquisition, 7, 113-160.
Kayne, R. (2005) On Parameters and on Principles of Pronunciation, in H. Broekhuis,
N.Corver, R.Huybregts, U.Kleinhenz and K.Koster eds. Organizing Grammar.
Linguistic Studies in Honor of Henk van Riemsdijk, Mouton De Gruyter, Berlin.
Kayne, R. (2005) Pronouns and their Antecedents. In Movement and Silence, 105-135.
Oxford University Press, New York.
Lopes, R. (2007) Acquiring a null object language. When children sound target-like but really
aren’t. In: S. Baaw; J. van Kampen; M. Pinto. (Org.). The acquisition of Romance
Languages. Selected papers from the Romance Turn II. LOT Occasional Series. 1 ed.
Utrecht: Utrecht University, 2007, v., pp. 113-126.
Oliveira S. M. (2007) Objecto direto nulo, pronome tônico de 3a pessoa, SN anafórico e
clítico acusativo no português brasileiro: uma análise de textos escolares. Revista
Virtual de Estudos da Linguagem – ReVEL, Vol. 5, n. 9.
Perez-Leroux A., M. Pirvulescu, Y. Roberge (2005) “Early object omission in child French
and English”, submitted to Proceedings of LSRL 35.
54

Syntactic diagnostics for referentiality marking in early null objects
Perez-Leroux A., M. Pirvulescu, Y. Roberge (2008) Children’s interpretation of null objects
under the scope of negation. XI Congress of the International Association for the
study of Child Language, Edimburg.
Perez-Leroux A., M. Pirvulescu, Y. Roberge (2008) Null objects in child language: syntax
and the lexicon. Lingua 118, 370-398.
Pirvulescu M. (2006a) Theoretical implications of object clitic omission in early French:
spontaneous vs. elicited production. In Catalan Journal of Linguistics 5, 221-236.
Raposo, E. (1986) On the null object in European Portugues. In: Jaeggli, O., Silva-Corvalán,
C. (org.). Studies in Romance Linguistics. Dordrecht: Foris.
Rizzi, L. (1986). Null Objects in Italian and the Theory of pro. Linguistic Inquiry, vol. 17,
501-557.
Rizzi L. (1993) Some notes on Romance cliticization. Ms. University of Geneve/University of
Siena.
Rizzi, L. (1997) The Fine Structure of the Left Periphery, in L. Haegeman (ed.), Elements of
Grammar, Kluwer, Dordrecht.
Rizzi L. (2000) Remarks on early null subjects, in N. Friedemann, L. Rizzi (eds.) The
Acquisition of Syntax, Harlow Longman Editors, Chapter 10.
Rizzi, L. (2005): Grammatically-Based Target-Inconsistencies in Child Language, in K.
UdDeen, J. Nomura, B. Schulz and B.D. Schwartz, eds., The Proceedings of the
Inaugural Conference on Generative Approaches to Language Acquisition – North
America (GALANA). UCONN/MIT Working Papers in Linguistics. MIT Press,
Cambridge (MA).
Schaeffer J. (2000) The Acquisition of Direct Object Scrambling and Clitic Placement.
Amsterdam/Philadelphia: John Benjamins.
Sigurðsson, H. (2004a) Meaningful silence, meaningless sounds. In P. Pica et al (eds.)
Linguistic Variation Yearbook 2004, Volume 4, 235-259. Amsterdam /Philadelphia:
John Benjamins.
Sirguðsson H. (2008) Argument drop and the Empty Left Edge Condition (ELEC).
Submitted.
Varlokosta, S. et al. (to appear) A cross-linguistic study of pronoun and clitic production,
Language Acquisition.
Wexler, K. (1998). Very early parameter setting and the unique checking constraint: A new
explanation of the optional infinitive stage. Lingua 106: 23-79.
Wexler K., A. Gavarrò, V. Torrens (2003) Feature checking and object clitic omission in
child Catalan and Spanish, in R. Bok-Bennema, B. Hollebrandse and B. Kampers-
Mahe (eds.) Selected Papers from Going Romance 2002.
55

Introduction
Rivista di Grammatica Generativa
34 – 2009
THE NATURE OF SYNTACTIC IMPAIRMENT
IN AUTISM
Stephanie Durrleman & Sandrine Zufferey
This paper presents a first exploration of the syntactic abilities of autistic
children using the framework of truncation theory (Rizzi 1993-4, 2000). It is the first
step of an ongoing research project aiming to (a) provide a complete map of syntax
in autism, for such work has never been attempted from a sound linguistic
perspective (b) develop a targeted remediation program to enhance syntactic abilities
of autistic children. Recently, a growing body of work has emphasized the existence
of a causal relation between the acquisition of complex syntax (embedding) and the
development of a theory of mind (in normally-developing, deaf and autistic
children). Early identification and remediation of syntactic deficits is therefore
crucial not only to enhance cognitive abilites related to theory of mind, but also
because of the critical period hypothesis for language acquisition.
This paper begins with a concise literature review on language in autism, with a
particular emphasis on syntax. The possible link between the language deficit found
in autism and specific language impairment (SLI) is discussed in the light of existing
evidence both from a genetic and a linguistic perspective. We conclude that the two
populations have a linguistic profile that isn’t entirely overlapping and propose
further investigation so as to establish a more precise comparison. The syntactic
peculiarities of autistic children are then reinterpreted within the framework of
truncation theory. Several hypotheses following from this model are presented and
discussed in the light of some preliminary empirical findings gathered in naturalistic
data. Some future directions of study are also discussed.
1.1. The language abilities of autistic children
Autism is a developmental disorder found in 1 in 100 children (Kogan et al.
2009), with boys being affected 4 times more than girls. Symptoms of this disorder
become noticeable before age 3. According to the American Psychiatric Association
57

Stephanie Durrleman & Sandrine Zufferey
(2000), the three defining characteristics of autism are: (1) impaired social
interaction, (2) delays/deficits in language and communication (3) restrained,
repetitive and stereotypical behaviours, activities and interests.
Regarding the communication deficit, one of the most robust findings in the field
of autism is that verbal autistic children specifically suffer from a deficit in the
ability to use language appropriately in context (i.e. pragmatics), as a result of their
lack of theory of mind abilities (Frith 1989, Baron-Cohen 1995). The possession of
theory of mind abilities broadly designates the capacity to attribute mental states to
oneself and to others, and to reason on the basis of this information in order to
interpret and predict others’ behaviors.
To test the possession of theory of mind abilities, psychologists have developed a
method to determine if a subject has a deep comprehension of others’ minds that is
in all other respects simple enough to be understood even by an autistic child. The
method takes into account the fact that, in order to demonstrate the presence of
mindreading abilities in a subject, it is not enough to prove that the subject (A) can
predict the actions of another individual (B). As Dennett (1978) has argued, in many
cases A can perform this task without attributing mental states to B, only by using
his own world knowledge. For example, if A knows that the basket in front of him
contains biscuits, and sees B trying to find food, A can predict that B will look into
the basket, not by attributing mental states to B, but because A knows the biscuits
really are in the basket. Therefore, in order to make sure that the subjects dissociate
their own beliefs from that of others, one has to devise a situation where another
individual entertains a belief different from the subject’s belief. This reasoning has
led to the development of the well-known false belief task, which now exists in
dozens of different variants.
The most popular version of the task, originally devised to be used with autistic
children, is called the Sally-Anne task ADD (Baron-Cohen, Leslie & Frith 1985).
This task takes place in two successive phases. First, the child hears the following
story enacted with two puppets called Sally and Anne:
58
Sally has a basket and Anne has a box. Sally has a marble and she puts it into
her basket. She then goes out. Anne takes out Sally’s marble and puts it into
her box while Sally is away. Now Sally comes back and wants to play with
her marble.
After hearing the story, children have to answer the following questions:
1. Where will Sally look for the marble?
2. Where is the marble really?
3. Do you remember where Sally put the marble in the beginning?
Only children with functional theory of mind abilities should be able to answer
correctly that Anne will act according to her own (false) belief and look for her
marble in the basket rather than in the box, where it really is.

The nature of syntactic impairment in autism
Because of their lack of theory of mind, autistic children are for instance unable
to follow conversational rules, and show inappropriate turn taking and topic shift
(Chin & Bernard-Opitz 2000). They also have trouble adapting their choice of words
depending on their addressee (Frith 1989) and the register chosen often makes their
use of language seem excessively formal or even pedantic. Another form of
pragmatic impairment can be seen in autistic individuals’ tendency to interpret all
statements in an overly literal fashion, and their failure to understand all non literal
uses of language like metaphors (Happé 1995), jokes (Baron-Cohen 1997; Reddy,
Williams and Vaughan 2002) and irony (Happé 1993). More generally, autistic
individuals are also unable to understand indirect speech acts (Aarons & Gittens
1992) and to perform most kinds of inferential reasoning (Dennis et al. 2001);
Frazier Norbury & Bishop (2002). Despite its obvious relation to language, the
communication deficit in autism can already be observed during the preverbal
period, by autistic children’s lack of proto-declarative pointing (proto-imperative
pointing is however present).
As can be seen from the above review, the discovery of a link between pragmatic
deficits and the theory of mind hypothesis in autism has led to an important number
of works during the past twenty years. Autism has therefore been used as an
example of dissociation between language competence and performance, because of
autistic subjects’ deficient language use in the presence of what was considered to
be a relatively spared linguistic system (Mundy & Marcus 1997). In the words of
Frith (1989/2003: 118): “ whatever the level of syntactic or semantic skill in autism,
the level of pragmatic skill will be lower”.
However, Tager-Flusberg (Tager-Flusberg 2000: 4) observes that “this emphasis
on the pragmatic language impairment in autism has led to a relative neglect of
other linguistic deficits that are found in most individuals with this disorder” (our
emphasis). In a recent study involving verbal children with autism, Eigsti et al.
(2007) have also found that autistic children have “clear language difficulties that go
beyond what would be expected based on developmental level”.
Moreover, it is worth noticing that language delay is one of the main diagnostic
criteria for autism (see point 2. of the characteristics above). When children display
autistic features such as repetitive behaviours without language delay, they are
diagnosed with Asperger syndrome (Howlin 2003). The criteria listed in the DSM-
IV (Diagnostic and Statistical Manual of Mental Disorders 4th edition, American
Psychiatric Association 2000) regarding the language impairments of autistics are
(1) late or lack of development of language without attempt to compensate with
gestures, (2) impairment in the ability to initiate or sustain a conversation, and (3)
stereotyped, repetitive, and idiosyncratic language.
Again, this description mostly emphasizes the pragmatic component of language.
However, in the accompanying descriptors, the DSM-IV also mentions aberrant
prosody and immature syntax. The reason for this emphasis on pragmatics is that
59

Stephanie Durrleman & Sandrine Zufferey
this deficit is found in all individuals suffering from autism (it is therefore called a
core feature of autism). On the other hand, the linguistic abilities of autistics display
great variability: about 50 to 75% of autistic individuals do not develop any form of
language during their lifespan (Rapin 1991) and at the other end of the spectrum, at
least some individuals do not suffer from any residual impairment in adulthood
(Eigsti & Bennetto 2009). In sum, even though syntactic and lexical abilities can be
considered as relative strengths in autism with respect to pragmatic abilities, they
are by no means unimpaired in most individuals.
As far as the lexicon is concerned, it has often been observed that autistic
children have a tendency to make idiosyncratic uses of words and to create
neologisms (Happé 1996; Frith 1989). To make a case in point, Attwood (2003)
gives the example of an autistic child who coined the word snook, to refer to the
chocolate bits found in some ice creams. Again, this deficit with the lexicon has
been attributed to autistic subjects’ lack of theory of mind, since the role of theory of
mind abilities in lexical acquisition has been recently highlighted for typicallydeveloping
children (Bloom 2000). More precisely, the ability to share one’s
attention with another speaker, one of the major precursors of a functional ToM
typically acquired around 18 months (Baron-Cohen 1995), has been proposed as one
possible cause of the sudden vocabulary spurt, often documented in the literature
(Baldwin 1991; 1993). Contrary to typical children, autistic children are unable to
use this cue in order to attribute a new name to an unknown object (Baron-Cohen,
Baldwin & Crowson 1997).
In the field of phonology, one area of consensus is that autistic children have an
aberrant prosody and often speak with inappropriate volume. Again this deficit has
often been explained in the light of the theory of mind deficit hypothesis (Baron-
Cohen 1995), the idea being that autistic children are unable to adapt to the context
of communication and the physical distance of their addressee. The development of
basic phonology has not been the subject of much enquiry, since autism cannot be
diagnosed during the first months of life. Besides, the exact phonological abilities of
autistic subjects is a matter of controversy 1 . According to Tager-Flusberg (2001),
verbal autistic children reach a mature phonological system during childhood,
comparable to that of normally-developing children. This finding was again taken as
an example of dissociation between a spared linguistic system in the face of an
impaired use of language. However, Rapin & Dunn (2003: 166) have presented new
arguments that tend to show that even basic phonology might be impaired in autism.
These authors have conducted electrophysiological studies showing “auditory input
abnormalities erase in lateral temporal cortex even in verbal individuals on the
autistic spectrum”.
1 The DSM-IV does not make any mention of phonology in its despcription of language
impairment in autism.
60

The nature of syntactic impairment in autism
To sum up, in addition to a generalized theory of mind deficit leading to
communication difficulties, it is also the case that most of the autistic population
suffers from some kind of deficit in the functional domains of language such as
phonology and, as we will show in more details below, syntax.
One of the major unanswered questions regarding language and autism is to
know if the patterns of language development found in autism parallels that of
normally-developing children with a substantial delay, or if language is disrupted in
a more fundamental way. There is at least one important known difference between
autistics and normally-developing children’s acquisition of language: namely the
role of echolalia. Echolalia implies the reiteration of a word or phrase stated by
others, as if echoing them. It may be immediate or delayed. A classic example of
immediate echolalia is a child echoing back a question that an individual is asking
him, with the same prosody. If a lapse of time has passed between the occurrence of
the verbal utterance and its repetition, this is referred to as delayed echolalia.
Examples of delayed echolalia are the echoing of television commercials or parental
reprimands some time after hearing these, and just as these were pronounced. Even
though all children make use of echolalia, in autistic children this phenomenon lasts
longer and is much more frequent. According to Prizant & Duchan (1981), a large
proportion of autistic children’s early speech is echolalic. The exact roles of
echolalia in autistics’ speech remain unclear but Eigsti et al. (2007) suggest that it
might serve several purposes like covering the inability to provide an answer, as a
form of ritual activity or as a way to retain information in memory. The frequency of
echolalia is therefore an important factor to take into account for studies targeting
natural productions by autistic children.
1.2. The acquisition of syntax by autistics
As observed for phonological abilities, the syntactic abilities of autistic children
is a matter of controversy. Eigsti et al. (2007) also observe that “few studies have
examined grammatical abilities in autism, with mixed findings”. Some studies (Fein
& Waterhouse 1979; Howlin, 1984a, 1984b; Tager-Flusberg et al., 1990) conclude
that autistic children’s syntactic development is delayed but follows a normal
trajectory with respect to normally-developing children, while others (Bartolucci,
1976; Bartolucci & Albers, 1974; Bartolucci et al 1980; Dalgleish 1975) conclude
that syntax is deviant in a more fundamental way. We will review these findings
below but it is already worth noting that one probable reason for this lack of
consensus is that many studies were conducted before strict, reliable measures of
syntactic development were developed in the 1990s 2 and were solely based on basic
2 For instance the Autism Diagnostic Interview-Revised (Cox et al., 1999; Lord, Rutter &
LeCouteur, 1994) and the Autism Diagnostic Observation Schedule (Lord et al., 1989; Lord,
Rutter, & DiLavore, 2000).
61

Stephanie Durrleman & Sandrine Zufferey
measurements of production like the MLU. In the more recent studies, the general
conclusion seems to be that syntax is indeed deviant and not only delayed
(Condouris et al. 2003; Rapin & Dunn, 2003; Kjelgaard & Tager-Flusberg, 2001).
However, as Eigsti et al. (2007) conclude: “a comprehensive investigation of
language in a well-controlled sample of children with autism, compared with both
typically developing and developmentally delayed groups, is needed”.
One point that seems uncontroversial is that autistic children’s syntax is usually
more simple than that of their age and IQ matched peers. More specifically, even
though basic sentences are usually mastered by verbal autistic subjects, Paul and
Cohen (1984) note that grammatical development “eventually reaches a plateau, at
least in some individuals”. Moreover, Kjelgaard and Tager-Flusberg (1999) have
found that 25% of autistic children with a normal nonverbal IQ score more than two
standard deviations below the mean on standard measures of grammatical ability.
Here we present an overview of the literature on the emergence of syntax in
autistics. We will start from what we know about the status of the highest functional
projections and work our way down the tree to the lexical projection.
1.2.1. CP (Complementizer Phrase)
1.2.1.1. Embedding
The fact that the spontaneous speech of autistics includes less sentential
complements and relative clauses than MA matched peers has been noted in
numerous studies (Bartolucci et al. 1980; Scarborough et al. 1991).
While it is widely accepted that Theory of Mind is delayed in autism, the
emergence of embedding in autistic individuals has been found to correlate with the
ability to pass false belief tasks: those individuals who make use of CP perform
better on ToM tasks. Tager-Flusberg (2000) studies older children and adolescents
with autism and concludes that there are ‘close connections between linguistic
knowledge of complement constructions and a representational understanding of
mind (…). The data presented here indicate that (individuals with autism) may
depend especially on complements for verbs of communication as their
developmental entrance into the ability to represent false beliefs’. The investigation
in Tager-Flusberg and Joseph (2005) corroborates this claim that the specific syntax
of complement structures is a significant predictor of successful false-belief
reasoning.
This link has also been found in a study by Fisher et al. (2005) who conclude
that: “language, in particular grammar, and theory of mind appear to be more
strongly related in children with ASD (Autism Spectrum Disorder) than in those
with moderate learning difficulties. We speculate that this relationship may be
causal, with some grammatical understanding being a precursor of theory of
mind”. This finding is crucial because it implies that a targeted remediation
62

The nature of syntactic impairment in autism
programme focusing on complex syntax may be beneficial for the development of a
functional theory of mind in autistic children.
One indication that this hypothesis is on the right track comes from a study of the
deaf community conducted by Schick et al. (2007). The study of this population
indeed suggests that the acquisition of grammar, and in particular of syntactic
(complement) structures, is fundamental to the development of ToM rather than the
other way around since deaf children, unlike autistics, do not suffer from other
social or cognitive delays which could be the underlying cause of an immature ToM.
Nevertheless a delay in ToM was apparent in part of this population: More
specifically, the findings revealed that “There was a significant delay on ToM tasks
in deaf children of hearing parents, who typically demonstrate language delays (…).
In contrast, deaf children from deaf families performed identically as same-aged
hearing controls. Both vocabulary and understanding syntactic complements were
significant independent predictors of success on verbal and low-verbal ToM tasks”.
The link between the level of performance on syntactic tests and the ability to
pass ToM tasks has also been investigated for normally-developing children.
Astington and Jenkins (1999) conducted a seven-month longitudinal study of
normally developing three-year old children and found that once age and previous
performance were controlled, their performance on standardized tests of language
ability predicted later performance on ToM tasks, but not vice versa. Another crucial
finding from this study is that syntactic ability was a more powerful predictor of
later ToM ability than was semantic ability.
It should be noted however that complement clauses are not specifically related
to ToM abilities. In a study by Smith et al. (2003), relative clauses were also found
to be a good predictor of future ToM abilities. Therefore, it seems that the general
syntactic principle of embedding is related to ToM rather than complement
structures only. Note that these are the very structures that were erase noted earlier
as being less present in autistics’ speech as compared to that of MA matched peers
(Bartolucci et al. 1980; Scarborough et al. 1991).
1.2.1.2. Questions
In typically developing children syntactic transformations such as subjectauxiliary
inversion begin around the age of 3 to 3½: e.g. ‘Is Daddy mad?’, ‘Where is
he going?’ (Crain & Lillo-Martin 1999). Sometimes an intermediate form of
interrogative is also used, as in ‘Where he is going’ which involves fronting of a
Wh-element without subject-auxiliary inversion. This is particularly frequent with
‘why’ as in ‘Why they can’t turn?’
Eigsti, Bennetto & Dadlani (2007) investigate various areas of morphosyntactic
development in 5 year-old children with autism and compare these with 3 year-old
typically developing (TD) children matched on non-verbal mental age. For
questions, they conclude that ‘The children with autism had response patterns that
63

Stephanie Durrleman & Sandrine Zufferey
were significantly different than in the TD group (…)”. In fact for this aspect of
grammar, they noticed a significant difference not only in comparison with the
typically developing group, but also with the group of children with non-specific
developmental delays. From a syntactic point of view, it would be interesting to
know exactly what sort of properties were found to be different in the interrogative
productions of ASD children. This would allow us to assess to what extent these
properties are similar to those found in instances of typical acquisition and would
then simply be persisting later in autistics (e.g. lack of subject-auxiliary inversion).
Alternatively, the patterns may represent errors that are particular to this population
and then more indicative of an impairment rather than a delay. It would also be
enlightening to know if certain questions are more affected than others, for example
‘why’ questions as opposed to other wh-questions, since according to recent
accounts of the CP (Rizzi 2001) the projection where ‘why’ may first merge,
Int(errogative)P, is arguably higher in the left-periphery than the position targeted by
other wh-elements 3 .
That interrogatives are affected in autism is also suggested in Condouris et al.
(2003). These authors applied various measures of linguistic ability to autistic
children, in order to compare the scores obtained with production measures like the
MLU and the Index of Productive Syntax (IPSyn) with comprehension tests like
Clinical Evaluation of Language Fundamentals and Peabody Vocabulary Test. The
results revealed that all measures expect for the IPSyn were strongly correlated.
More specifically, Condouris et al. (2003) argue that the IPSyn score artificially
lowers the syntactic abilities of the autistic population because of the coding
procedure of the IPSyn. This test divides syntactic abilities in four dimensions: NP,
VP, Questions and Negation, and sentence structure. According to Condouris et al.
“the low scores obtained on the IPSyn were because the children with autism used a
narrow range of grammatical constructions and had especially low scores on the
question/negation subscale of the IPSyn”. This finding suggests that the syntactic
deficit in autism seems to be located on the higher projections of the clause.
1.2.2. AgrP (Agreement Phrase)
Third person -s in normally developing 5 year olds has been found to be used
correctly 95% of the time (Rice et al 1995), while children with autism with a mean
age of 8-9 years have difficulty with using this third person singular morpheme
(Roberts 2004). In a study with SLI (Specific Language Impairment) children, Rice
et al. (1995) found that 5-year-old children with SLI got 26% of correct answers. At
8 years of age, the SLI group was at 96% (similar to normally developing 5 year-
3 Although this proposal was initially made based on Italian data, its significance in
accounting for properties of the acquisition of other languages such as English is evident (see
e.g. Thornton 2007).
64

The nature of syntactic impairment in autism
olds). Using similar scoring criteria, Roberts et al. (2004) have investigated three
groups of children with autism divided on the basis of their performance on the
Peabody Picture Vocabulary Test III (PPVT; Dunn & Dunn, 1997), using the criteria
that Kjelgaard and Tager-Flusberg (2001) adopted for defining language subgroups
in autism. Group 1 (Normal) included children whose PPVT standard scores were
within the normal range, group 2 (Borderline) included children whose PPVT
standard scores were between 1 and 2 standard deviations below the mean and group
3 (Impaired) included children whose PPVT standard scores were two or more
standard deviations below the mean. There were no significant age differences
between these three groups. They found that children in the Normal group were
correct on 81% of trials, the Borderline group on 69%, and the Impaired group on
65% of trials. Thus, the language impaired children with autism (with a mean age of 8-
9 years) performed somewhat better than 5-year-old children with SLI but worse than
normally developing 5-year-olds and worse than SLI children in the same age range.
1.2.3. TP (Tense Phrase)
Roberts et. al (2004) discuss tense acquisition in autistics ranging from 5 to 15
years old and conclude that a subgroup of these (language impaired) children
showed high rates of past tense marking omission, scoring correctly on 68% of the
trials, which is less than the scores of typically developing 5 year olds, revealed to
attain 92% accuracy on past tense marking (Rice et al. 1995). In fact, autistic
children have been found to mark past tense significantly less than even matched
controls with Down syndrome (Tager-Flusberg 1989). Such difficulties with
handling tense have been attributed by some authors (e.g. Bartolucci et al. 1980) to
semantic-pragmatic deficits rather than morpho-syntax.
1.2.4. AspP (Aspectual Phrase)
A study by Bartolucci and Alberts (1974) elicited productions of present
progressive -ing and past tense -ed in autistic subjects and revealed that the children
with autism performed well on the present progressive form, although they were less
successful on the past tense trials. In addition, Roberts et al. (2004) when probing
for third-person singular marking and past tense in this population, noted the
presence of responses which ‘involved a high proportion of present progressive -ing
responses’ (Roberts et al 2004: 441). An example given is “He fixing teeth” in reply
to a question asking what a dentist does.
1.2.5. VP (Verb Phrase)
As observed for aspect, the lexical abilities of autistic subjects also appears to be
relatively spared. This is not to say that there are no problems at all: a study by
Howlin (2003) on the lexicon of autistic children shows that subjects suffering from
Asperger syndrome, even though they did not have any language delay during
childhood, have a lexicon significantly smaller than normal once they reach
65

Stephanie Durrleman & Sandrine Zufferey
adulthood. This deficit is found both in comprehension and production: the average
delay in comprehension is 12 years and 10 years in production. More interestingly
still, this delay was similar for subjects who had been diagnosed with highfunctioning
autism and with Asperger syndrome during their childhood. This is a
remarkable result, since the diagnosis criterion for Asperger syndrome is precisely
the absence of language delay during childhood. Howlin’s study therefore tends to
indicate that the differences between high-functioning autism and Asperger
syndrome are irrelevant for adults.
Still, Eigsti et al. (2007) note a sharp contrast between the handling of
grammatical and lexical categories, and conclude that “In contrast to syntactic
development, the present study demonstrated that lexical knowledge was an area of
relative strength”. This study illustrates that the use of VP is almost equivalent to
that found in typically developing children matched for non-verbal IQ, while
sentences and questions/negation were significantly worse. In this respect, the
lexical core is intact in comparison to e.g. that reported for Questions (and
Negations), which are far below typically developing performance. The finding that
the lexicon is relatively unimpaired is not new, as the authors note: “Consistent with
previous studies, the children with autism seemed to comprehend and produce as
many words as their peers. While some aspects of lexical knowledge in autism are
atypical, including the use of idiosyncratic meanings for words and neologisms
(Rumsey et al. 1985; Rutter 1970; Volden & Lord 1991), adults with autism often
have larger vocabularies than would be predicted from their other language abilities
(Lord & Paul 1997)”. As a result of these considerations, Eigsti & al. (2007) suggest
a need to match autism groups to appropriate control groups through use of other
forms of measures to receptive vocabulary, since “receptive vocabulary abilities,
such as those measured by the PPVT-III, are likely an inadequate marker of
‘language ability,’ and studies that match groups based only on lexical measures are
likely to overestimate the comprehension of children in the autism group”.
That the lexical abilities of autistics are relatively spared is also claimed by
Walenski et al. (2008) whose study reveals “better-than-normal performance” for
lexical processing in the set of autistic boys studied, since these performed faster
than the matched group of male controls on picture naming. They conclude
cautiously that although the source of this ‘enhancement’ is not clear, it may well be
linked to the ‘seasaw effect’ with declarative memory being compensatorily overdeveloped
because of a possible abnormality in procedural memory. It is worth
noting here that procedural memory is linked, amongst other things, to syntax.
1.2.6. The nominal group: DP (Determiner Phrase) and NP (Noun Phrase)
To our knowledge, there is no study of autistics that systematically draws a clear
division between the acquisition of the lexical versus grammatical sections of the
nominal domain. The sort of errors reported often involve what appears to be
66

The nature of syntactic impairment in autism
omission or misuse of articles, but a precise assessment of the various elements that
occur in the extended nominal projection is yet to be provided.
1.2.7. A-chains
It should be noted that in addition to the difficulties associated with sections of
the syntax as sketched above, certain syntactic transformations also appear
challenging for autistics: Movement operations such as those involved in Passives
are revealed by Perovic et al (2007) to be problematic. This same study indicates
that children with autism perform poorer than controls on binding, suggesting
together with the finding on passives that this population has a general difficulty
with A-chains. In sum, the difficulties in building clause structure outlined in the
previous sections appear to be accompanied with difficulties in establishing relations
within this stucture.
2. A manifestation of SLI?
In the literature on autism, a recent controversial proposal (Bishop 2003)
suggests that the language deficit in autism is similar to that observed with SLI
children. Although most of the evidence comes from genetics, some studies on
language have found similar patterns in both populations: notably an inability to
repeat nonwords and problems with tense marking (Kjelgaard & Tager-Flusberg
2001).
According to this explanation, there is no language deficit in autism proper
(other than those related to pragmatics), but some autistics have core impairment in
linguistic ability similar to that seen in SLI. From a genetic point of view, Tomblin,
Hafeman, and O’Brien (2003) found that the prevalence of autism in siblings of
children with SLI (1%) was significantly higher than general population estimates.
However, there are reasons to doubt that linguistic difficulties found in autism are
simply a manifestation of an associated SLI. As Whitehouse et al. (2008) observe,
“the sample size [of Hafeman et al.’s study] was very small for an epidemiological
study, and the rate of autism cases in siblings of children with SLI did not
significantly differ from that of the control group (0.4%)”. Several studies also
indicate that relatives of autistic children often report language difficulties. However
these difficulties were not apparent when they were tested with standardized
language assessments (Bishop et al. 2004; Whitehouse et al. 2007).
In sum, most of the behavioral evidence for an SLI sub-group of autistics comes
from their inability to repeat nonwords. We argue that this test, while revealing a
common symptom in both populations, very probably has different causes in the two
cases. A similar problem has arisen in relation to the false-belief task. Some studies
(Frith & Corcoran 1996) show that schizophrenic patients as well as autistics
typically fail at this task. However, it has been repeatedly pointed out (see Bloom &
67

Stephanie Durrleman & Sandrine Zufferey
German 2000 for a review) that the false-belief task involves more than just mental
state attribution – like a good working memory and linguistic abilities – and most
probably, these additional demands are responsible for schizophrenics’ failure rather
than an inability to reason about mental states, as it is the case for autistics.
As Whitehouse et al. (2008) argue, it is likely that autistic and SLI children’s
failure at non-word repetition tasks is due to different causes. In their words:
“nonword repetition deficits in SLI result from genetic factors affecting neuroanatomical
pathways responsible for language, nonword repetition deficits in the
autistic population are associated with broader autistic symptoms”. Among them we
can mention poor attention to others’ speech and a difficulty to imitate.
Similarly, some recent studies (e.g. Miller 2001; Gillott et al. 2004) seem to
indicate that SLI children also fail at the false-belief task 4 . But once again the reason
for failing seems quite different in the two populations. For SLI children, the more
linguistically demanding the task, the higher the rate of failure (as could be expected
in the light of their linguistic impairment). On the contrary, for autistic children,
nonverbal ToM tasks are in fact harder than their verbal equivalents, revealing that
their difficulties do not come from linguistic demands but from a real impairment
with mental state attribution. Indeed, it seems that autistic children use linguistic
cues (like their knowledge of completive structures) to compensate for their lack of
ToM abilities. This difference is further confirmed when comparing the kinds of
explanations given by both populations in response to ToM tasks. While autistic
children provide mostly “physical state answers” (i.e. answers that do not resort to
inferences about mental states), SLI children provide as many mental state answers
as control children. However, the linguistic complexity of the task results in an
inappropriate inference being made.
The other grammatical point of convergence between autistics and SLI children
is their problem with grammatical marking of tense. But once again, the two groups’
performance is not comparable. In the study by Roberts et al. (2004), the language
impaired children with autism (with a mean age of 8-9 years) performed somewhat
better than 5-year-old children with SLI but worse than SLI children in the same
age range.
Some other linguistic markers of SLI have been identified in the literature:
Hamann et al. (2003) point out that SLI children acquiring French show difficulty
with clitic pronouns, and in particular with pronominal complement clitics. Hamann
(2006) underlines that French speaking SLI children further differ from their TD
peers in their net preference for in-situ question formation, producing almost twice
as many of these than the controls (40.4% vs. 24.2%), as well as in their omission of
4 These findings do not however replicate earlier studies (Leslie and Frith 1988; Perner et al.
1989) who found that SLI children consistently pass simple (first order) ToM tasks like the
Sally-Ann task.
68

The nature of syntactic impairment in autism
subjects in all types of fronted Wh-questions. The occurrence of missing subjects in
Wh-questions was also noted for SLI children acquiring German (Hamann 1998)
although was absent from TD German children (Clahsen et al. 1996). It is therefore
worth exploring the characteristics of autistic children’s speech with respect to these
properties as well, in order to get a complete comparison of the two groups’
overlapping linguistic problems.
3. A new look at syntactic impairment in autism
Looking over the areas of grammatical delays and impairments in ASD
individuals, an observation that can be made is that the higher up one goes in the
tree, the more likely it is that autistics encounter difficulty. These difficulties are
summarized in the tree below. In light of the area of impairment, the question arises
to know whether this population is having a harder time building these higher layers
of functional projections.
impaired complement clauses and questions
impaired use of -s
impaired use of -ed, high rate
of omission
69

Stephanie Durrleman & Sandrine Zufferey
One possible explanation for the data can be provided in terms of truncation.
Truncation theory (Rizzi 1993-4, 2000) has been instrumental in accounting for
various syntactic properties of typical language acquisition. This theory essentially
proposes that children pass through a phase where they have the option of omitting
the initial layer(s) of structure, thereby using the most economical amount of
structure necessary to accommodate the overt material. More specifically, the theory
proposes that an immature system initially does not systematically project to the CP
layer, but if and when it does, it must contain the intermediate projections as well.
This analysis of ‘peeling off’ layers from the top explains various phenomena found
in early systems such as root null subjects (1) and optional infinitives (2):
(1a) _ was a green one (1;10) (Brown 1973)
(1b) _ have a drink grape juice first (1;10)
(1c) _ falled in the briefcase (1;10)
(2a) Dormir tout nu (1;10) (Pierce 1989)
Sleepinf all naked
(2b) Ich der Fos hab’n (2;1) (Poeppel & Wexler 1993)
I the frog haveinf
(2c) Eerst kleine boekje lezen (2;6) (Haegeman 1995)
First small book reading
Truncation accounts for root null-subjects (1) by stating that in a clause stripped
of the left periphery, subjects occupy the highest position of the syntactic tree (i.e.
the specifier of the root) which is a position that is exempted from the identification
requirement normally applying to null elements. It therefore follows from the
truncation approach that these early null subjects are not commonly attested in
interrogatives, since these obligatorily contain a CP and therefore the specifier of the
clause would no longer qualify as specifier of the root. The other phenomenon
addressed by the truncation approach is that of root infinitives (2): an early system
can also apply truncation to lower layers than CP, e.g. TP. This accounts for the
frequent appearance of root infinitives in child language, and for their
incompatibility not only with wh-constructions but also with functional verbs such
as auxiliaries and modals. Once the system matures and truncation ceases to be an
option, clauses project up to CP and both null subjects and root infinitives no longer
occur. This has been attested for various languages (see Haegeman 1995 for Dutch,
Rasetti 2000 for French, Hamann and Plunkett 1998 for Danish). The requirement
that the root be a CP emerges in the grammar sometime before the child’s third
birthday.
70

The nature of syntactic impairment in autism
Coming back to the autism data, if higher layers persist in being optionally
omitted even in older autistics, then we expect to find other properties related to the
relevant structural zone to also be affected in this population, with properties related
to lower layers being potentially preserved. Moreover if truncation indeed continues
later in ASD than in the TD population, it becomes conceivable that this will have a
significant impact on the development not only of syntactic but also cognitive
abilities, since it has been found that certain cognitive developments are related to
the emergence of particular syntactic constructions, as discussed earlier for ToM
abilities and the mastery of CP. Indeed it is now a widely accepted view that there is
a critical period for language acquisition in early childhood, and that once this
window of opportunity is passed then the task of acquiring a language becomes
increasingly more difficult, and eventually ceases to be possible (see e.g. Lenneberg
1967, Curtiss 1977, Kegl et al 1999). This underlines the importance of implementing
early intervention for triggering language development in such cases of pathology.
4. A corpus study
In order to assess the nature of syntactic impairment in autism, we have searched
for relevant examples in natural productions of autistic children. Contrary to SLI,
not many resources are available for autism, and some data cannot be used for a
syntactic analysis because the children recorded are barely verbal. Overall, only the
corpus gathered by Helen Tager-Flusberg (Tager-Flusberg 1990), and made
available on the CHILDES database, could be used for our purposes. This corpus
contains recordings of 6 autistic and 6 Down’s syndrome children. The duration of
the study was variable across children, with an average of 22 months. The number of
recordings also vary from 8 to 13. Contrary to most naturalistic data, children in this
corpus are not recorded during unconstrained interactions, but during structured play
sessions with their mothers, aiming at eliciting as many linguistic productions as
possible. The reason for this specific context is that, contrary to normallydeveloping
children, autistic children would produce very little verbal interactions if
not pressured to do so.
In order to assess the syntactic abilities of autistic children, we have looked at
grammatical markings for all layers of the tree. Since most data had to be collected
manually, we have circumscribed our search space in the following way. We have
selected recordings from two children situated at different points in the autistic
spectrum in order to have a comparison of their grammatical difficulties. At the
higher end, we chose to concentrate on Brett, the most verbal child in the corpus,
who had an MLU of 3.74 at the beginning of the study. We have excluded from our
comparison data all the children with an MLU inferior to 2 (3 in total), because we
could not reasonably expect them to produce many multi-word utterances. The
second child we chose for comparison was Roger, who had an MLU of 2.31 at the
71

Stephanie Durrleman & Sandrine Zufferey
beginning of the study. Even though Brett, 5.8 years at the beginning of the study, is
almost two years older than Roger who is 3.9 years, they have a very similar IQ: 108
for Brett vs. 105 for Roger. These two children are therefore very well suited for our
comparative purposes.
From a methodological point of view, it is also worth mentioning that cases of
echolalia have been eliminated from the data. Multiple repetitions of a same
sentence have been credited only once. Productions resulting from direct prompts
have also been disregarded.
4.1. CP
The first observation we made for the CP layer is that there is an almost
complete absence of relative and complement clauses in the corpus, for both
children studied 5 .
For questions, we note that there are few of them in the corpus, despite the
constant prompts provided by the caregivers. In our data, most of the questions were
asked by the most verbal child, Brett. Traditionally, this lack of interrogatives was
explained by autistics’ reluctance to initiate communication. However true this
explanation may be, we note nevertheless that when the pragmatic intention of
asking a question is there, access to the relevant structure appears challenging, as
illustrated by this exchange with Brett:
(3a) CHI(LD): have a diet coke at the refrigerator?
(b) CHI: a milk ball?
(c) MOT(HER): no.
(d) CHI: have milk ball please?
(e) CHI: I want milk ball.
(f) CHI: a milk ball please.
(g) CHI: I want milk ball.
(h) CHI: I want milk ball.
(i) MOT: I want a milk ball.
(j) CHI: I want a milk ball.
(h) MOT: good asking.
As far as WH-questions are concerned, the proportion of all WH-words is
described in table 1 below.
5 There is a single occurrence of a (subject) relative clause found in Brett’s entire corpus,
towards the end of his 2 years of recordings: I saw an elephant then a doggy that goes woof.
There was also one instance of complementizer ‘that’, produced by Roger with a verb of
perception: People see that I use all the soap.
72

The nature of syntactic impairment in autism
wh word Brett Roger Total nb Total percentage
why 5 0 5 9%
what 29 3 32 55%
who 0 0 0 0%
when 6 0 6 10%
how 1 0 1 2%
where 9 5 14 24%
Moreover, many questions contained some kind of syntactic problem, like absence
of AUX/T (example 4), or absence of subject-auxiliary inversion (SAI) (example 5).
Null subjects in Wh contexts (6) were completely absent for Brett and quite rare for
Roger 6 (1%), unlike that attested for SLI children by Hamann 7 (1998, 2006).
(4) Brett: Where yellow box?
(5) Brett: No, why I not do it on the trunk?
(6) Roger: did you cry when xxx got a brown booboo?
Table 2 contains a summary of the syntactic problems of all WH-questions in the
corpus. Overall, 25% of the WH-questions were ungrammatical. As can been seen,
the major grammatical problem is due to an omission of AUX/T. Wh-movement as
such seems however preserved 8 .
no AUX/T marking no SAI null subject
why 2 1
what 2 1
when 2
where 6 1
total 12 (21%) 2 (3%) 1 (2%)
6 It is worth noting that even the examples that are transcribed showing an absent subject in an
embedded context such as that given here are somewhat problematic: here xxx seems to imply
that something was mumbled but remained unclear. The other examples of an unspecified
subject in an interrogative were all similar to: ‘do want some fruit bar?’, where the intended
sequence appears to be ‘do you want some fruit bar?’. In such examples, one is tempted to
think of a slight phonological erosion of the pronoun ‘you’ which might have yielded the
resulting transcription, rather than the total omission of the subject. In any event, such
occurrences are extremely rare.
7 Also contrary to what was found for SLI children (Hamann 2003), all questions but one had
correct wh-movement with only one instance of WH in-situ found, produced by another child
from the corpus: CHI: Smacked who, snoopy?
8 Some preliminary findings with negation also point to the same conclusion. While negation
proper seems to be preserved, negative polarity items appear to be erroneously used, and
AUX/T is almost systematically omitted in these constructions.
73

Stephanie Durrleman & Sandrine Zufferey
4.2. AgrP
Our data also confirms problems with AgrP. As far as this projection can be
considered responsible or partially responsible for impaired use of the third person
singular -s, the corpus reveals that this agreement marking is missing in 21% of the
productions at the higher end of the spectrum (Brett’s data), and in 71% of the
production in Roger’s data.
(7) CHI: A cow make woof woof.
(8) CHI: Brett hold my square.
(9) CHI: He say that.
(10) CHI: It feel soaking.
For both children we observe a marked tendency to omit the subject position in
declaratives:
(11a) MOT: What did you do in school today?
(b) CHI: ∅ did work with Peggy at school.
(12) CHI: ∅ need to turn the fan on.
(13a) MOT: What can the knife do?
(13b) CHI: ∅ can cut you badly.
Over the 2 year period, 23% of Brett’s productions exhibit null subjects, and
40% of Roger’s. It is worth underlining that the corpus reveals the children’s
mothers providing a good deal of explicit prompting and modeling for subjects
throughout, and that this can be thought to have played an important role in the
children’s significant improvement over the 2 years with respect to this property:
Brett’s first recording exhibited 53% null subjects, while two years later his
productions of null subjects had dropped to 20%; and Roger commenced with 62%
null subjects and by the final recording his productions contained only 11% null
subjects.
Brett omitted subjects only when these would occupy the specifier of the root, as
predicted by a truncation account, as did Roger for 99% of his null subjects 9 , with
non-root subjects being systematically realized. In some instances, we can observe
this trend in the very same sentence, with a null subject in the main clause and an
overt one in the embedded clause:
9 Note again that the 1% of cases where Roger allowed null subjects in an embedded context
do not constitute entirely clear examples of embedded null subjects. As previously indicated,
data such as the following were attested: did you cry when xxx got a brown booboo? Where
the presence of (xxx) suggests nevertheless that the child is aware of the necessity to specify a
subject in such a context, but possibly simply doesn’t articulate it precisely in such instances.
74

(14) CHI: ∅ happens if [I’m a bad boy].
The nature of syntactic impairment in autism
Hamann (2006) revealed that French SLI children differed from their typically
developing peers in allowing null subjects in non-root contexts, namely in Whquestions.
The findings here suggest then that the autistic and SLI population differ
for this property.
As noted in section 3, null subjects are a characteristic of immature systems,
although typically these cease to appear once truncation ceases to be an option,
normally occurring before the ages of the autistic children here. Therefore there is at
least a visible ‘delay’ in this respect, suggesting that truncation persists longer in this
population 10 .
4.3. TP
As far as TP is concerned, we can confirm that regular past tense (-ed) is used far
less frequently than aspectual (-ing). We only count for Brett 38 correct uses and for
Roger 9 correct uses of past tense, whereas the -ing form is very frequently used
instead (see below). We also find a clear example below, taken from Brett’s data
that illustrates autistics’ tendency to struggle with past tense:
(15a) MOT: No what did you just do?
(b) CHI: Work on the puzzle.
(c) CHI: I work on the puzzle.
(d) MOT: What were you doing?
(e) CHI: I am working on the puzzle.
(f) MOT: What were you doing?
(g) CHI: I was working on the puzzle.
(h) MOT: That’s right.
10 Aside from what looks like a marked delay in realizing overt subjects linked to a
prolongation of truncation, there seem to be other issues with subject position. Consider the
following:
CHI: can I have to me?
MOT: what do you want Brett?
CHI: I want the big play dough to me?
CHI: I wanna give it.
Here, we can derive from the general gist of the exchanges that the child is trying to get his
mother to give him ‘the big play dough’, but is having a hard time using language to do so.
Striking in the child’s last reply is the instance of ‘wanna’, where the understood subject ‘you’
should be present structurally if not phonologically, as in ‘I want (you) to give it (to me)’. As
such, the presence of the embedded subject ‘you’ should block ‘wanna’-contraction. As we
can see, this is not the case. Further investigation is necessary to elucidate what is happening
in such instances.
75

Stephanie Durrleman & Sandrine Zufferey
The auxiliary is sometimes also missing, both in interrogatives and in
declaratives, as shown in the examples below 11 . Here again, we note an important
difference between the two children under investigation. While Brett has a very low
rate of omission of 4.5%, Roger has a rate of omission of 20%. Moreover, in 25% of
the cases, Roger omits the auxiliary in a question context. Note that these omissions
cannot at first sight be attributed to truncation, since in a wh context, the auxiliary
does not occupy the specifier position of the root. We will come back to this issue in
section 5.
(16) CHI: Where ∅ yellow box?
(17a) MOT: Who’s sitting down?
(b) CHI: We sitting down.
(18a) MOT: Where is the moon?
(b) CHI: Moon in sky.
(19a) MOT: Who’s this?
(b) CHI: That Bert.
4.4. AspP
Regarding progressive aspect, this seems to be commonly overtly marked on the
verb stem, although often in the absence of auxiliaries as noted earlier:
(20) CHI: why you crying Paul?
(21) CHI: why it not snowing?
(22) CHI: I am working on the puzzle.
11
In Roger’s data, we occasionally find examples such as the following:
‘Roger don’t touch’
However it is not entirely clear whether this is an imperative repeated to himself (i.e. here an
instance of delayed echolalia) or a declarative containing an auxiliary that is unmarked for
agreement. In any event, this sort of structure is only ever found in root contexts. If indeed we
have an auxiliary lacking in agreement appearing only in root contexts, then this is a crucial
observation since truncation of the higher layers would be a straightforward way to account
for the phenomenon: In root contexts, as opposed to interrogative contexts, stripping away of
CP and even AgrP, is possible, and would yield an auxiliary unmarked for agreement. A
similar pattern for auxiliaries was observed in instances of typical acquisition by Guasti &
Rizzi (2002). These authors conclude that this state of affairs upholds that Agreement and
Tense features are licensed in distinct syntactic positions, with the former being located
higher than the latter in the clausal architecture. Given the uncertainty as to the exact intention
of the child studied here, we leave the issue open.
76

(23) CHI: I was working on the puzzle.
(24) CHI: we sitting down.
The nature of syntactic impairment in autism
The most verbal child Brett produced 115 occurrences of progressives through
an -ing form, which corresponds to three times the amount of correct regular -ed
past tense used by this child. This comparison provides even more striking results
lower down the spectrum: Roger produced 85 occurrences of -ing forms, which
amounts to 10 times the number of correct regular past tense forms used by this
child.
4.5. DP
The data on nominals suggest that articles are commonly dropped: in the case of
Brett the rate of omission is 17% and this number goes to up 32% for Roger. On the
other hand, the two children make frequent uses of adjectives, as illustrated in (25),
(26) and (27) below. Adjectivals are arguably lower in the universal nominal hierarchy
than other functional material (Cinque 2000) and so at first sight, indeed the higher
projections of the DP are less morphologically realized than the lower ones:
(25) MOT: I want ∅ milk ball
(26) CHI: I want ∅ amusement park.
(27) CHI: ∅ star going around.
(28) CHI: Where ∅ yellow box?
Below we see that numerals are also used, but in the absence of plural -s on the
noun. The rate of plural omission is 15% for Brett. This number is harder to
compute for Roger, who uses many words in isolation, which makes it often difficult
to identify where a plural would have been required.
(32) CHI: Two fireplace.
4.6. Reflexives
Some of the productions in this corpus suggest a difficulty with reflexives:
(29a) MOT: Who dressed you this morning?
(b) CHI: I dressed me.
(30) CHI: Wanna hold it to me.
This is reminiscent of the observation in Perovic et al (2006) that autistic
children struggle with binding.
77

Stephanie Durrleman & Sandrine Zufferey
5. Summary of findings and a first analysis
As already noted, a review of the literature indicates that the higher sections of
the clause tend to be more visibly omissible than the lower sections in productions
of this population, and this corpus study upholds that initial impression. Also, once
the higher layers are present, we observe that the child does not omit the lower ones.
For example, although subjects in root contexts were often omitted, once the child
produced an interrogative, subjects were overtly realized as they no longer occupy
the specifier of the root. All these observations uphold a truncation account, where
the children are then exploring the option to project reduced structures. However,
while this option is typically found in immature systems (and abbreviated adult
registers), ceasing in developing systems around the age of 3, we note that in this
instance of pathology the omission of higher material appears to persist.
In light of the critical period hypothesis, it becomes conceivable that this
pronounced delay may result in a considerable perturbation of the system on other
levels. For example, although these children produce structures that include Whfronting,
therefore suggesting that they exploit the CP, we nevertheless find
auxiliary omission in these productions, unlike in the target language. At first sight
this seems difficult to account for in terms of truncation. However in light of the
fine-grained maps which have been recently developed for the C-system (Rizzi
1997, 2004), the full left-periphery projects all the way up to ForceP, the projection
responsible for encoding illocutionary force. A recent study of interrogatives in a
phase of English acquisition (Guasti 2000) proposes that what some children peel
off is this last layer, which implies that they form interrogatives where Focus° has
become the head of the root, unlike in adult language where the head of the root is
always Force°. As a result of this application of truncation to a more articulate map,
we are able to understand why the children here may also omit the auxiliary in these
contexts: in such reduced structures building only up to Foc(us)P, the auxiliary
occupies the initial head position, a position which is not subject to the Principle of
Clause Internal Identification. This approach in terms of truncation then explains the
particularity of null auxiliaries found in interrogatives of this corpus as well.
Moreover, the peeling off of ForceP has other repercussions on the language of
the children under investigation: This projection is the one responsible for hosting
the complementizer that which encodes declarative force. If children with autism
truncate ForceP, they are not projecting all the way up to the position typically used
in embedding. It is interesting to note that subordinate constructions were extremely
rare in this corpus, with only two occurrences of complementizer that found in the
20 recordings over a two year period. To summarize these findings, consider the
following tree:
78

absence of complementizer that
WH-element can be fronted
AUX/T is the head of the root i.e.omissible
The nature of syntactic impairment in autism
In light of the research discussed in section 1.2.1 which upholds that complement
structures go hand in hand with theory of mind abilities (Tager-Flusberg 2000,
Tager-Flusberg & Joseph 2005, Schick et al. 2007), it becomes increasingly
important to conceive of a way to build the syntactic apparatus in autistics all the
way up to ForceP, with a remedial program working its way through consolidating
the intermediate projections along the way. This is because, on the one hand,
cognitive abilities such as a delay in ToM may crucially rely on the development of
this part of the syntax and therefore never fully develop in its absence. On the other
hand, if this section of the clausal architecture is not consolidated because of the
extended delay in the language acquisition of autistics, characteristics of the
grammar which build upon this full architecture, arising later in typical acquisition,
may never be entirely mastered by autistics. This view has been applied to other
populations to explain their achieving a certain level of grammatical ability and then
no more and has been named ‘Specific Grammatical Impairment’ (Perovic &
Wexler 2007). These authors conclude that certain aspects of grammar that arise
earlier in the typically-developing population are intact in the Williams syndrome
subjects but those that arise later “seem exceptionally delayed, and even
unattainable”. As a result of these observations, they propose that:
“it is misleading to speak about grammatical development in Williams
syndrome as either ‘normal’ or ‘abnormal’. Rather, it is both, if the terms can
be applied at all. Grammatical development in Williams syndrome is ‘normal’
in that children with Williams syndrome have Universal Grammar (UG) and
the properties of UG mature in the order in which they mature in typical
development. In this way, the language of children with Williams syndrome
looks like that of typically-developing children. But certain grammatical
operations never develop in children with Williams syndrome. In this way, the
79

Stephanie Durrleman & Sandrine Zufferey
80
language of children with Williams syndrome looks quite different from the
language of typically-developing children”.
We claim that a similar scenario appears to hold for children with autism, but
frame this in terms of a prolonged application of truncation, delaying, and in some
cases prohibiting, the acquisition of the higher layers of the clause.
6. Conclusion
This preliminary investigation of syntax in autism suggests the relevance of
truncation theory in accounting not only for phenomena of typical language
acquisition, but also for the instance of pathology examined here. We have seen that
the distribution and restriction on the occurrence of null elements follow from a
view where what is stripped away is every projection higher than the truncation site,
but all projections dominated by it are present. As such we account, for example, for
subject omission being a readily available option once it occurs in the specifier of
the root. We also explain the dramatic increase in overt subjects in Wh-contexts,
where the projection of FocusP implies that the subject no longer benefits from the
privilege of the root, yielding the almost complete absence of null subjects in such
configurations.
Moreover this approach to syntax in autism provides support for the hypothesis
that when language develops too slowly, the higher layers of the tree are the ones
most likely to remain affected since they are the ones systematically vulnerable to
the truncation mechanism: We speculate that persistence in truncation of these
layers, induced by the pronounced language delay beyond the critical period, may
well result in the top structural levels never being solidly acquired. Given that theory
of mind abilities have been precisely linked to the development of structures that
make use of the highest layer of the clausal architecture (namely complement
structures and relative clauses which are articulated by means of the CP layer), the
impaired theory of mind in autism may plausibly stem not only from a pragmatic
deficiency, but also from a syntactic one.
Admittedly much more empirical work needs to be conducted before this
hypothesis can be proven: for now, although complement clauses have been noted to
be impaired in many autistics, relative clauses have yet to be investigated. However
if future research continues to confirm the view that complex syntax associated with
higher clausal layers are affected in autism, then such findings will ultimately be
enlightening with respect to remediation programs. Notably, programs aiming at
developing theory of mind abilites would therefore need to include exercises
targeting syntax of the higher layer of the left-periphery. More generally, speech
therapists could consider detecting which layer the child has been able to build to,
and then develop exercises to trigger subsequent layers from that point in a
systematic fashion, as this may well begin much lower than the CP layer for some.

The nature of syntactic impairment in autism
It is worth noting that the findings here further suggest that language difficulties
in autism differ somewhat from those in SLI: Certain properties in SLI do not follow
directly from a truncation approach, such as omission of subjects in Wh questions
(Hamann 2006). If future investigation of autistic children continues to uphold that
they generally avoid null subjects in such contexts, as the children studied here
appear to, then this would be one way in which the two populations do not overlap
and their speech therapy would then diverge in such respects.
Hopefully this first attempt to examine language in autism from a syntactic
perspective will add another dimension to the current trend in research whose focus
is mainly the pragmatic deficiency evident in this condition. As researchers look
more precisely at syntax, and disentangle its role from pragmatics as much as
possible, we should finally arrive at a more balanced view of the linguistic
challenges of autistic individuals and develop more complete remediation programs
for them, fine-tuned to their needs.
Bibliography
Aarons, M. and T. Gittens (1992). The handbook of autism: a guide for parents and
professionals. London, Routledge.
American Psychiatric Association (2000). Diagnostic and statistical manual of mental
disorders, text revision DSM IV-TR, 4 th ed. Washington DC, American Psychiatric
Association.
Astington, J. and J. Jenkins (1999). "A longitudinal study of the relation between language
and theory-of-mind development." Developmental psychology 35(5): 1311-1320.
Attwood, T. (2003). Le syndrome d'Asperger et l'autisme de haut niveau. Paris, Dunod.
Baldwin, D. A. (1991). "Infant's contribution to the achievement of joint reference." Child
development 62(5): 875-890.
Baldwin, D. A. (1993). "Infant's ability to consult the speaker for clues to word reference."
Journal of child language 20(2): 395-418.
Baron-Cohen, S. (1995). Mindblindness: an essay on autism and theory of mind. Cambridge,
The MIT Press.
Baron-Cohen, S. (1997). "Hey! It was just a joke! Understanding propositions and
propositional attitudes by normally developing children, and children with autism."
Israel journal of psychiatry 37: 174-178.
Baron-Cohen, S., D. A. Baldwin, et al. (1997). "Do children with autism use the Speaker's
Direction of Gaze (SDG) strategy to crack the code of language ?" Child development
68: 48-57.
Baron-Cohen, Simon, Leslie, Alan and Frith, Uta 1985. "Does the autistic child have a
"theory of mind"?" Cognition 21: 37-46.
Bartolucci, G. (1976). Formal aspects of language in childhood autism. Autism and severe
psychopathology: advances in child behavior analysis and therapy. J. Steffen and P.
Karoly. Lexington, MA, Lexington Books: 159-185.
81

Stephanie Durrleman & Sandrine Zufferey
Bartolucci, G. and R. J. Albers (1974). "Deictic categories in the language of autistic
children." Journal of autism and childhood schizophrenia 19: 131-141.
Bartolucci, G., S. J. Pierce, et al. (1980). "Cross-sectional studies of grammatical morphemes
in autistic and mentally retarded children." Journal of Autism and Developmental
Disorders 10(1): 39-50.
Bishop, D. (2003). Children’s communication checklist-2. London, Harcourt Assessment.
Bishop, D., M. Maybery, et al. (2004) "Are phonological processing deficits part of the broad
autism phenotype?" American Journal of Medical Genetics 128B.: 54-60.
Bloom, P. (2000). How children learn the meanings of words. Cambridge, MA, The MIT
Press.
Bloom, P. and T. German (2000). "Two reasons to abandon the false belief task as a test of
theory of mind." Cognition 77: B25-B31.
Brown, R. (1973). A First Language. The Early Stages. Harvard University Press, Cambridge,
Mass.
Chin, Y. H. and V. Bernard-Opitz (2000). "Teaching conversational skills to children with
autism: effect on the development of a theory of mind." Journal of autism and
developmental disorders 30(6): 569-583.
Cinque, G. (2000). "On Greenberg's Universal 20 and the Semitic DP". University of Venice
Working Papers in Linguistics, 10(2): 45-61.
Clahsen, H, Kursawe, C. Penke, M. (1996). "Introducing CP: Wh-questions and subordinate
clauses in German child language". Essex Research Reports in Linguistics 7: 1-28.
Condouris, K., E. Meyer, et al. (2003). "The relationship between standardized measures of
language and measures of spontaneous speech in children With autism." American
journal of speech-language pathology 12(3): 349-358.
Cox et al. (1999). "Autism Spectrum Disorders at 20 and 42 months of Age: Stability of
Clinical and ADI_R Diagnosis." Journal of Child Psychology and Psychiatry, 40:5,
719-732.
Crain, S. & Lillo-Martin, D. (1999). An Introduction to Linguistic Theory and Language
Acquisition. Oxford: Blackwell.
Curtiss, S. (1977). Genie: a psycholinguistic study of a modern-day "wild child". Boston:
Academic Press.
Dalgleish, B. (1975). "Cognitive processing and linguistic reference in autistic children."
Journal of Autism and Developmental Disorders 5(4): 353-361.
Dennett, D. C. (1978). “Beliefs about beliefs”. In Behavioural and Brain Sciences, 4, 568-
570.
Dennis, M., A. Lazenby, et al. (2001). "Inferential language in high-function children with
autism." Journal of autism and developmental disorders 31(1): 47-54.
Eigsti, I.-M. and L. Bennetto (2009). "Grammaticality judgments in autism: deviance or
delay." Journal of child language 36(5): 999-1021.
Eigsti, I.-M., L. Bennetto, et al. (2007). "Beyond pragmatics: morphosyntactic development in
autism." Journal of autism and developemental disorders 37: 107-1023.
82

The nature of syntactic impairment in autism
Fein, D., & Waterhouse, L. (1979). "Autism is not a disorder of language." Boston, MA:
Paper presented at the New England Child Language Association.
Fisher, N., Happé, F., Dunn, J., "The relationship between vocabulary, grammar, and false
belief task performance in children with autistic spectrum disorders and children with
moderate learning difficulties". Journal of Child Psychology and Psychiatry and
Allied Disciplines, Vol. 46(4), 409-419. Blackwell Publishing.
Frazier Norbury, C. and D. Bishop (2002). "Inferential processing and story recall in children
with communication problems: a comparison of specific language impairment,
pragmatic language impairment and high-functioning autism." International journal
of language and communication disorders 37(3): 227-251.
Frith, C. and R. Corcoran (1996). "Exploring theory of mind in people with schizophrenia."
Psychological medicine 26: 521-530.
Frith, U. (1989). Autism: explaining the enigma. Oxford, Blackwell.
Gillott, A., F. Furniss, et al. (2004). "Theory of mind ability in children with specific language
impairment." Child language teaching and therapy 20(1): 1-11.
Guasti, T. (2000) “An excursion into interrogatives in early English and Italian”. In M. A.
Friedeman & L. Rizzi (Eds.) The acquisition of syntax, pp. 105-128, Singapore:
Pearson Education Limited.
Guasti, T and Rizzi, L. (2002), 'Agreement and Tense as Distinct Syntactic Positions:
Evidence from Acquisition'. In G. Cinque (Ed.) Functional Structure in DP and IP:
The Cartography of Syntactic Structures, Vol 1, 167-194. Oxford University Press.
Haegeman, L. (1995). “Root Inifinitives, Tense and Truncated Structures in Dutch”,
Language Acquisition, 4: 205-255.
Hamann, C. & K. Plunkett (1998). "Subjectless Sentences in Child Danish". Cognition 69:
35-72.
Hamann, C., Ohayon, S., Dubé, S., Frauenfelder, U.H., Rizzi, L., Starke, M., et al. (2003)
“Aspects of grammatical development in young French children with SLI”. In
Developmental Science, 2003, 6, 151-160.
Hamann, C. (2006) “Speculations about early syntax: The production of Wh-questions by
normally developing French children and French children with SLI”. In Catalan
Journal of Linguistics, 5, 143-189.
Happé, F. (1993). "Communicative competence and theory of mind in autism: a test for
relevance theory." Cognition 48: 101-119.
Happé, F. (1995). "Understanding minds and metaphors: insights from the study of figurative
language in autism." Metaphor and symbolic activity 10(4): 275-295.
Happé, F. (1996). Autism: an introduction to psychological theory. Cambridge, Harvard
University Press.
Howlin, P. (1984a). "The acquisition of grammatical morphemes in autistic children: A
critique and replication of the findings of Bartolucci, Pierce, and Streiner, 1980."
Journal of Autism and Developmental Disorders, 14, 127-136.
83

Stephanie Durrleman & Sandrine Zufferey
Howlin, P. (1984b). "The acquisition of grammatical morphemes in autistic children: a
critique and replication of the findings of Bartolucci, Pierce, and Streiner, 1980."
Journal of Autism and Developmental Disorders, 14(2), 127-136.
Howlin, P. (2003). "Outcome in high-functioning adults with autism with and without
language delays: implications for the differenciation between autism and Asperger
syndrome." Journal of autism and developmental disorders 33(1): 3-13.
Kegl, J., Senghas, A. & Coppola, M. (1999). "Creation through Contact: Sign Language
Emergence and Sing Language Change in Nicaragua", in M. Degraff (ed.) Language
Creation and Language Change: Creolization, Diachrony, and Development, MIT
Press, Cambridge (Mass).
Kjelgaard, M. and H. Tager-Flusberg (1999). "A psychological approach to understanding the
social and language impairments in autism." International review of psychiatry 11(4):
325-334.
Kjelgaard, M. and H. Tager-Flusberg (2001). "An investigation of language impairment in
autism: implications for genetic subgroups." Language and cognitive processes 16:
287-308.
Kogan MD, Blumberg SJ, Schieve LA, et al. (2009). "Prevalence of parent-reported diagnosis
of autism spectrum disorder among children in the United States", 2007. Pediatrics.
2009;124(5): 1395-1403.
Lenneberg, E. H. (1967). Biological Foundations of Language. Wiley.
Lord, C., Risi, S., Lambrecht, L. Cook, Jr., E. H., Leventhal, B. L., DiLavore, P. Pickles, A. &
Rutter, M. (2000). "The Autism Diagnostic Observation Schedule - Generic: A
standard measure of social and communication deficits associated with the spectrum
of autism." Journal of Autism and Developmental Disorders 30, 205-223.
Miller, C. (2001). "False belief understanding in children with specific language impairment."
Journal of communication disorders 34: 73-86.
Mundy, P., & Markus, J. (1997). "On the nature of communication and language impairment
in autism". Mental Retardation & Developmental Disabilities Research Reviews, 3,
343-349.
Paul, R. & Cohen, D.J. (1984). "Outcomes of severe disorders of language acquisition."
Journal of Autism and Developmental Disorders 14, 405-422.
Perovic, A., Modyanova, N., Wexler, K. September (2007). "Knowledge of c-command and
A-movement in children and adolescents with autism and with Asperger syndrome".
GALA 2007: Generative Approaches to Language Acquisition. Barcelona, Spain.
Talk.
Pierce, A. (1989). On the Emergence of Syntax: A Crosslinguistic Study. PhD Dissertation,
MIT.
Poeppel, D. & Wexler, K. (1993). "The full competence hypothesis of clause structure in
early German". Language 69, 1-33.
Prizant, B & Ducan, J (1981). "The functions of immediate echolalia in autistic children."
Journal of Speech and Hearing Disorders 46. 241-249.
84

The nature of syntactic impairment in autism
Perovic, A. and K. Wexler (2007). "Complex grammar in Williams syndrome." Clinical
linguistics and phonetics 21(9): 729-745.
Rapin, I. (1991). "Autistic Children: Diagnosis and Clinical Features." Paediatrics 87(5): 751-
760.
Rapin, I. (2003). "Update on the language disorders of individuals on the autistic spectrum."
Brain and development 25: 166-172.
Rasetti, L., (2000). Null subjects and root infinitives in the child grammar of French. In
Friedemann & Rizzi, eds., 236-268.
Reddy, V., E. Williams, et al. (2002). "Sharing humour and laughter in autism and Down's
syndrome." British journal of psychology 93(2): 219-242.
Rice, M.L., Wexler, K. and Cleave, P.L. (1995). "Specific language impairment as a period of
extended optional infinitive." JSLHR 38, pp. 850-863.
Rizzi, L. (1993/1994). "Some Notes on Linguistic Theory and Language Development: The
Case of Root Infinitives". Language Acquisition, Vol. 3, No. 4, pp. 371-393.
Rizzi, L. (1997) .“The Fine Structure of the Left Periphery”, in L. Haegeman (ed.), Elements
of Grammar. Kluwer, Dordrecht.
Rizzi, Luigi, (2001). "On the position 'Int(errogative)' in the left periphery of the clause." In:
Cinque, G. and Salvi, G., Editors, (2001). Current Studies in Italian Syntax. Essays
Offered to Lorenzo Renzi. Elsevier, Amsterdam, 287-296.
Rizzi L. (2000). "Remarks on Early Null Subjects." In: M.-A. Friedemann & L. Rizzi (eds.)
The Acquisition of Syntax: Issues in Developmental Comparative Linguistics. London:
Longman, 269-292.
Rizzi, L. (2004). The Structure of CP and IP: The Cartography of Syntactic Structures, Vol 2.
Oxford University Press, Oxford-New York.
Roberts, J., M. Rice, et al. (2004). "Tense marking in children with autism." Applied
psycholinguistics 25: 429-448.
Rumsey JM, Duara R, Grady C, Rapoport JL, Margolin RA, Rapoport SI, et al. (1985). Brain
metabolism in autism. Resting cerebral glucose utilization rates as measured with
positron emission tomography. Arch Gen Psychiatry 42: 448-455.
Rutter, M. (1970). "Autistic children: Infancy to adulthood". Seminars in Psychiatry, 42 (5),
448-455.
Scarborough, H., Rescorla, L., Tager-Flusberg, H., Fowler, A., & Sudhalter, V. (1991).
"Relation of utterance length to grammatical complexity in normal or languagedisordered
groups". Applied Psycholinguistics, 12, 23-45.
Schick, B.; de Villiers, P.; de Villiers, J.; Hoffmeister, R. (2007). "Language and Theory of
Mind: A Study of Deaf Children". Child Development, Vol. 78 (2), 376-396(21)
Blackwell Publishing.
Smith, N., B. Hermelin, et al. (2003). "Dissociation of social affect and theory of mind in a
case of Asperger syndrome." UCL Working Papers in Linguistics 15: 303-322.
Tager-Flusberg, H. (1989). A psycholinguistic perspective on language development in the
autistic child. Autism: New directions on diagnosis, nature, and treatment. G.
Dawson. New-York, Guilford Press: 92-115.
85

Stephanie Durrleman & Sandrine Zufferey
Tager-Flusberg, H. (2000). Language and understanding minds: connections in autism.
Understanding other minds. Perspectives from developmental cognitive neuroscience.
S. Baron-Cohen, H. Tager-Flusberg and D. Cohen. Oxford, Oxford University Press:
124-149.
Tager-Flusberg, H. (2001). "Understanding the language and communicative impairments in
autism." International Review of Research in Mental Retardation 23: 185-205.
Tager-Flusberg, H., S. Calkins, et al. (1990). "A longitudinal study of language acquisition in
autistic and Downs syndrome children." Journal of Autism and Developmental
Disorders 20: 1-21.
Tager-Flusberg, H. and R. Joseph (2005). How language facilitates the acquisition of falsebelief
understanding in children with autism. Why Language Matters for Theory of
Mind. J. Astington and J. Baird. Oxford, Oxford University Press: 298-318.
Tomblin, J., L. Hafeman, et al. (2003). "Autism and autism risk in siblings of children with
specific language impairment." International Journal of Language and
Communication Disorders 38: 235-250.
Volden, J. and C. Lord (1991). "Neologisms and idiosyncratic language in autistic speakers."
Journal of Autism and Developmental Disorders 21(2): 109-130.
Walenski, M., Stewart, Mostofsky, Gidley-Larson, Ullman (2008). "Brief Report: Enhanced
Picture Naming in Autism". Journal of Autism and Developmental Disorders, 1395-
1399.
Whitehouse, A., J. Barry, et al. (2007). "The broader language phenotype of autism: A
comparison with specific language impairment." Journal of Child Psychology and
Psychiatry 48: 822-830.
Whitehouse, A., J. Barry, et al. (2008). "Further defining the language impairment of autism:
is there a specific language impairment subtype?" Journal of communication
disorders 41: 319-336.
86

Rivista di Grammatica Generativa
34 – 2009
A COMPARISON BETWEEN THE CANTONESE
ASPECT MARKERS GAN AND HAIDOU
Chi-Fung Lam
A comparison between the Cantonese aspect markers gan and haidou
In past research on Cantonese, the progressive aspect is typically stated as being
marked by the postverbal particle gan 1 (Cheung 1972, Gao 1980, Li, et al 1995, Yip
& Matthews 1995). Few people recognize that the preverbal locative adverb haidou
can also be used as a progressive aspect marker (Cheung 1972, Gao 1980). For those
discussions where the progressive marker haidou is mentioned, only a very brief
description is devoted to haidou when compared to gan. Moreover, no attention is
paid to the difference between them and haidou is usually considered simply as an
alternative to gan or used to emphasize the ongoing meaning when it co-occurs with
gan in the way that haidou is preverbal and gan is postverbal (Li et al 1995, Yip &
Matthews 1995).
This paper intends to discuss the difference between gan and haidou and to
explain it by considering notions originating from the two-component theory of
aspect advocated by Smith (1997). In her theory, both the viewpoint aspect and
situation aspect of the verb constellation contribute to the aspectuality of a sentence.
It is well known that the progressive aspect can occur in sentences with Activity and
Accomplishment (e.g. Gu 1999:193-4). Thus, it is not surprising that both haidou
and gan can occur in these two types of events. However, I will show that gan is
also compatible with a certain subset of Stative sentences. This hypothesis builds on
the observation that the situation type Stative, as in Smith’s terminology, may
further be divided into two types which are quite different in nature, namely
Property and State, I claim that the State is compatible with gan while the Property
1 Throughout this paper, jyut ping, the Hong Kong Cantonese romanization system designed
by the Linguistic Society of Hong Kong (http://www.lshk.org/cantonese.php), is used to
represent the Cantonese data. Cantonese is a tone language but for the sake of convenience
the lexical tone is left out unless it is significant in the discussion.
87

Chi-Fung Lam
is not. This additional classification helps us to establish a syntactic and semantic
distinction between gan and haidou. It provides an effective criterion to predict the
distribution of the two aspect markers. In a nutshell, the goal of this paper is to show:
(G1) Haidou is a typical progressive aspect marker which is used in Activities and
Accomplishments.
(G2) Gan should be treated as a general imperfective aspect marker. While it can
be used with Activities and Accomplishments, it can also be employed with
[-permanent] States.
This paper is set out as follows. First of all, the imperfectiveness of gan and
haidou is justified. Examples are given to show that they determine imperfective
aspect where no absolute tense is indicated; however, by default, they assume that
the reference time is just the same as the speech time, in other words, the present
tense. Second, I provide some examples of sentences which lead to my main
concern in this paper. Some of them admit only haidou and some admit only gan. I
then try to explain these distributions by considering the situation aspect and discuss
their implications on our understanding of the aspectuality of these two markers.
Finally, some apparent exceptions to my claim are mentioned, which I will attribute
to syntactic, semantic and pragmatic factors.
1. Gan and haidou as aspect markers in Cantonese
Cantonese uses grammaticalized morphemes to express the viewpoint aspect.
Most of these morphemes still have their concrete lexical meanings when they are
not used as aspect markers. Among them, postverbal particles like gan, zyu, hoi, lok
express imperfective aspect. Previous research has very often neglected haidou or
simply described it as an appendix to gan. This is motivated by the observation that
preverbal haidou can replace gan or co-occur with it to emphasize progressiveness.
Such research does not seem to accept haidou as a progressive aspect marker like
the postverbal particles. It has even been suggested that haidou is just like an adverb
(Li, et al 1995). In this section, I will illustrate situations in Cantonese where the
postverbal gan and preverbal haidou in Cantonese do act as imperfective markers in
that the reading of the present tense is assumed when no other temporal constituent
is used. Discussion of the progressiveness of each of these markers will be left to
Section 2. Before considering gan and haidou, it is necessary to mention the system
of aspectual classification I adopt in this paper.
1.1 Classification of imperfectives
There are two kinds of aspect, namely viewpoint aspect and situation aspect.
Both aspectual systems are relevant to my argument and will also be discussed in
this paper. This section however concentrates solely on the first system.
88

A comparison between the Cantonese aspect markers gan and haidou
Imperfective aspect can basically be divided into two kinds, namely the perfective
and imperfective. The perfective presents a situation as a whole, including its initial
and final endpoints while the imperfective only focuses on some part of the whole
situation and thus excludes both endpoints. Smith (1997) classifies those situations
lacking marked aspectual morphemes as neutral aspect, which is flexible, including
the initial endpoint of a situation as well as at least one of the internal stages.
Throughout this paper, only the imperfective will be discussed.
Comrie (1976:25) presents the following tree diagram to illustrate the
classification of the viewpoint aspectual system:
Viewpoint aspect
Perfective Imperfective
Habitual Continuous
Non-progressive Progressive
The table is constructed by pairs of oppositions, aiming at covering various
languages. It should be noted that a language may not have corresponding
grammatical marking on all viewpoint aspects. For English, the construction be + Ving
acts as the whole subclass of continuous aspect. Equally the table may not be
adequate for some languages. According to previous discussion (Cheung 1972, Gao
1980, Li, et al 1995, Yip & Matthews 1995), Cantonese has quite a rich aspectual
system. However, these linguists fail to agree on how to classify and interpret some
of the particles. For instance, the most diversely interpreted one is hoi, which was
classified as continuous in Cheung (1972), progressive in Gao (1980), inchoative in
Li, et al (1995) and habitual in Yip and Matthews (1995). The studies concurred
mainly in classing gan as progressive, and zyu as durative or continuous (and thus
non-progressive continuous). An exemption to this last point is Gao (1980), who
classified zyu as progressive just like gan and hoi.
I do not intend to go through, in detail, all of these aspect markers and their
various classifications. In the following sections, gan and haidou will be discussed
in more depth, and particularly focus will be given to their role in marking tense and
aspect.
1.2 Gan as a continuous aspect marker in Cantonese
The postverbal particle gan has been treated as a typical progressive marker in
Cantonese (Cheung 1972, Gao 1980, Li, et al 1995, Yip & Matthews 1995). I argue
that although this is mainly the case, it may also serve as a continuous aspect marker.
Its progressiveness, and its interaction with tense, will be discussed in Section 2.
89

Chi-Fung Lam
Before dealing with the temporal semantics of gan, we will take a look at its
distribution. Firstly, gan immediately follows the simple verb, for example (Yip &
Matthews 1995: 202),
(1) Go-zan-si ni dung lau zung hei -gan.
that CL time this CL flat still build GAN
“At that time this block of flats was still being built.”
Second, it is an infix in a verb-complement compound, following the verb and
preceding the complement. For example (Yip & Matthews 1995: 202), in Cantonese,
gan can be inserted into the verb-complement compound, paakto (to date):
(2) Keoidei loeng-go paak-gan-to. 2
they two CL date GAN
“The two of them are dating.”
For the syntax of gan, it should be noted that gan also functions as an adjective –
meaning tight – as well as the complement in a verb compound where it represents
the result or aim of the action on the object. So, V + gan + O forms a resultative
predicate (Cheung 1972:107-8) 3 . Since there is no difference between these two
gan’s in this construction, sentences like the following one are ambiguous:
(3) Ngo jiga bong-gan tiu sing.
now tie GAN CL rope
(i) “I now tie the rope tight.”
(ii) “I am now tying up the rope.”
In the above example, if we interpret gan as an adjective, the reading (3i) will be
obtained. Otherwise, we have to interpret gan as an imperfective aspect marker (the
imperfectiveness of gan will be justified below; for the moment I will take it for
granted), as in (3ii). We can use a syntactic test to clarify which interpretation is
being adopted. For instance, a verb phrase with adjectival complement gan is
compatible with the perfective aspect marker zo, that is, bong-gan-zo (have tied tight)
while a verb phrase with imperfective aspect marker gan is not. However, such
ambiguity is quite rare because the adjective tight, due to its specific meaning, can
be a complement to only a few verbs. In contrast, the imperfective aspect marker
gan can be applied to a large set of verbs, at least to all verbs indicating action.
I am now about to discuss the tense and aspect semantics of gan. Consider the
following examples:
2 CL refers to numeral classifier.
3 The verb paakto can be analyzed as a complex one in that paak and to are both verbs, with
the latter being nominalized, hence its similarity to a VO compound. In Cantonese, the aspect
suffix is attached to the V in VO compounds.
90

A comparison between the Cantonese aspect markers gan and haidou
(4) Camjat saam-dim-zong go zan, ngo zong zou-gan je
yesterday three o’clock that time I still do GAN thing
“At three o’clock yesterday, I was still working.”
(5) Ngo jiga aamaam sik-gan je
I now just eat GAN thing
“I am just eating.”
(6) Tingjat saam-dim-zong go zan, ngo zong zou-gan je
tomorrow three o’clock that time I still do GAN thing
“At three o’clock tomorrow, I will still be working.”
(7) Ngo sik-gan je
I eat GAN thing
“I am eating.”
These examples demonstrate that gan can be used in the past as in (4), present as
in (5) and future as in (6). Note that in each of these three examples, a certain
reference time is given, namely cam-jat saam-dim-zong (at three o’clock yesterday)
in (4), ji-ga (now) in (5) and ting-jat saam-dim-zong (at three o’clock tomorrow) in
(6). The verb phrases with gan anchor these time adverbials; in other words, the
event time (E) is equal to the reference time (R). The tenses of (4)-(6) are then
obtained by comparing the time of the adverbials, that is, the reference time (R) to
the speech time (S). Example (7) does not have any time adverbial as the reference
time. By default, it should be present. However, since the verb phrases in (4)-(8) are
all of the same structure on the surface, that is, V + gan + O, it means that this ganform
has no tense in nature and thus no absolute tense but relative tense, that is, the
tense relative to the context or a given reference time in the discourse. Suppose (7) is
in a context which provides a clear past or future tense, then (7) can still be
interpreted in past or future tense.
Now I turn to aspect. Suppose in a sentence, the postverbal gan is not or cannot
be interpreted as a resultative complement. Is it really an imperfective marker rather
than a perfective one, or does it bear no relevance to aspect? Certain tests provide
pertinent insights. Since the imperfective form differs from the perfective in that it
excludes the endpoints of the situation, we may expect that predication with gan is
incompatible with the meaning of completion:
(8) a. #Camjat saam-dim-zong go zan, ngo zong zou-gan je, jau aamaam zou-yun.
yesterday three o’clock that time I still do GAN thing and just do finish
“#At three o’clock yesterday, I was still working and had just finished.”
b. Camjat saam-dim-zong go zan, ngo zong zou-gan je. Ji-ga aamaam zou-yun.
yesterday three o’clock that time I still do GAN thing now just do finish
“At three o’clock yesterday, I was still working. Now I’ve just finished.”
91

Chi-Fung Lam
(8) is obtained by adding a clause with meaning of completion to the end of (4).
(8a) turns out to be semantically ill-formed (denoted by #). The only possible
reading is (8b) where the added clause is interpreted as a new utterance which is
structurally independent of the previous clause and conveys a present reading. Yun,
literally meaning “finish”, is a verbal suffix. Based on its clear literal meaning, the
predication with yun has to be perfective. Since the truth values of V-gan and V-yun
are opposite, a predication with gan has to be interpreted in the imperfective sense.
Recall the hierarchical classification of the viewpoint aspectual system suggested
by Comrie (1976:25) mentioned in Section 1.1. After confirming that gan is
imperfective, I would like to ask whether it can express the habitual. Let us consider
the following examples:
(9) Ngo gau-nin fung laibaailuk paau (*-gan) -bou
I last year every Saturday jog
“Last year I jogged every Saturday.”
(10) Ngo jiga fung laibaailuk paau (*-gan) -bou
I now every Saturday jog
“Now I jog every Saturday.”
(11) Ngo ha-go jyut hoici fung laibaailuk paau (*-gan) -bou
I next CL month start every Saturday jog
“Starting from next month, I will jog every Saturday.”
From the above examples we know that gan cannot express habitual aspect
whether the sentence is in the past, present or future tense. In Comrie’s hierarchy,
gan should be classified as continuous in aspect.
What I have shown in this section is that the verbal suffix gan is a continuous
aspect marker and the predication with gan anchors the time adverbial if there is any,
and produces a relative tense; however, it has nothing to do with absolute tense but
is present by default. Moreover, there is a syntactic test to distinguish the aspect
marker gan from the resultative complement gan.
1.3 Haidou as a continuous marker in Cantonese
In this section, I discuss haidou in the order of its morphology, syntax and finally
semantics of tense and aspect. Haidou consists of two morphemes, hai and dou. Dou
is not a free morpheme but a nominal suffix. 4 It represents a variable location
depending on the noun phrase preceding it and can be translated in English as
4 Syntactically the Cantonese dou can be comparable to the Mandarin li, though the latter
literally means “inside” and the former “place”. The morpheme li is called “localizer” in Chao
(1968), and is argued to be a clitic that is a subtype of noun in Huang, Li & Li (2009).
92

A comparison between the Cantonese aspect markers gan and haidou
“place”. 5 The noun phrase preceding it can be any spatial object and so the
construction NP-dou is very productive, for instance:
(12) a. Zeong toi -dou
CL table place
“the place near/on the table”
b. Peter-dou
Peter place
“the place near Peter/Peter’s home”
Other than NPs, dou can also follow the deitics ni “this” and go “that”, the
question word bin “which” to form constructions meaning “here”, “there” and
“where” respectively.
(13) a. Ni-dou “this place/here”
b. Go-dou “that place/there”
c. Bin-dou “which place/where”
Hai is a locative preposition, roughly equivalent to the English “at/in” or “be
at/in” in English. Then we have the construction hai X-dou which is actually a
prepositional phrase. However, the X in-between can be empty and in this case the
prepositional phrase haidou can mean hai ni-dou “(be) here” or hai go-dou “(be)
there”. This can be illustrated by the following conversation:
(14) A: Haidou a.
Be here/there SFP 6
“It’s here/there”
B: Hai bindou a?
at where SFP
“Where is it?”
A: Hai zeong toi -dou.
at CL table place
“It’s on/near the table.”
In the above, the first utterance of person A regards a variable location that
depends on their own location or on the direction they are pointing in. Person B
seeks for a clearer location and person A replies with a concrete one.
Now I turn to syntax. The question I would like to address whether there exists a
syntactic test which can distinguish the aspect marker haidou from the prepositional
5 For recent discussion on the element PLACE, see Cinque (2010).
6 SFP refers to sentence final particle, which is frequently used to realize mood in Cantonese.
93

Chi-Fung Lam
phrase haidou. 7 This test may help to distinguish the temporal meaning from the
spatial one. In Cantonese, prepositions can be classified into two categories, namely
the “pure” prepositions and the “verb-like” ones, as I will call them. This division is
established by two syntactic tests which aim at distinguishing prepositions from
verbs. The first test is whether the morpheme A can appear in the A-not-A
construction to form yes/no questions. Some prepositions cannot appear in A-not-A
construction:
(15) *Jau-m-jau ni- tiu lou heoi a?
from not from this CL way go SFP
“Go from this way?”
94
In such cases, the appropriate structure should be as follows:
(16) Hai-m-hai jau ni-tiu lou heoi a?
Be not be from this CL way go SFP
“Go from this way?”
Hai in the above example is the copula “be” and is a different morpheme from
the locative preposition hai; the preposition is in the second tone of Cantonese,
namely hai2 and the copula in the sixth tone, hai6. However, some other
prepositions can appear in the A-not-A construction, such as the locative preposition
hai:
(17) Nei hai2-m-hai2 li-dou zyu a?
You at not at here live SFP
“Do you live here?”
But at the same time, it can also be achieved by the copula:
(18) Nei hai6-m-hai6 hai2 lidou zyu a?
You be not be at here live SFP
“Do you live here?”
The examples of yau and hai respectively illustrate the so-called pure
prepositions and the verb-like prepositions. To form the yes/no question, pure
prepositions have to use the copula, as shown in (15) and (16) while the use of the
copula is not obligatory in verb-like prepositions. The question could be asked as to
why the latter categories are not simply classified as verbs? The answer to this lies
in the second syntactic test which claims that, to be a verb, the morpheme must be
able to be suffixed by at least one aspectual marker. Hai does not fulfill this criterion;
7 An important note to be emphasized here is that throughout this paper only the preverbal but
not postverbal haidou is discussed. In Cantonese, there is actually a postverbal haidou which
can also be interpreted as a prepositional phrase and as an aspect marker. Nevertheless, it is
outside the interests of this paper and so will not be discussed.

A comparison between the Cantonese aspect markers gan and haidou
in the following example, hai cannot be followed by the imperfectives gan, zyu, hoi
and lok, perfect zo and experiential gwo:
(19) Nei hai (*gan/*zyu/*hoi/*lok/*zo/*gwo) li-dou zyu a?
You at ASP here live SFP
“Do you live here?”
The debate on the classification of prepositions and verbs is not my concern here.
The reason I mention the above tests is their use in determining haidou as a
prepositional phrase or as an aspect marker. When haidou is interpreted as a
prepositional phrase before a verb phrase, both constructions of hai6-m-hai6 dou
and hai6-m-hai6 haidou can be used to form interrogatives. In contrast, when haidou
is interpreted as an aspect marker, only hai6-m-hai6 haidou can be used to form
interrogatives. So in the following, (20) has only one reading while (21) has two:
(20) Keoi hai-m-haidou sik faan a?
He at not at place eat rice SFP
“Will he have a meal here?”
(21) Keoi hai-m-hai haidou sik faan a?
He be not be at here eat rice SFP
(i) “Will he have a meal here?”
(ii) “Is he having a meal?”
Therefore, to determine whether haidou in the sentence Keoi haidou sik faan
refers to a prepositional phrase or an aspect marker, we may adopt the syntactic test
suggested above. If both constructions are permitted to form interrogatives, then
haidou refers to a prepositional phrase. If only the “be-not-be” construction is
permitted, haidou refers to an aspect marker.
After establishing a syntactic means of testing the existence of the aspect marker
haidou, I shall now move on to discuss the relation of haidou to tense and aspect. I
claim that haidou is a continuous aspect without the absolute tense. Since the
argument is exactly the same as that mentioned in Section 1.2, I will not repeat it but
will simply give the examples and brief notes as reference. After replacing the
postverbal gan in examples (4)-(11) by the preverbal haidou, we have the following
examples.
(22) Camjat cat-dim-zong go-zan, ngo zong haidou zou je
yesterday seven o’clock that time I still at here do thing
“I was still working at seven o’clock yesterday.”
(23) Ngo jiga aamaam haidou sik je
I now just at here eat thing
“I am just eating”
95

Chi-Fung Lam
(24) Tingjat cat-dim-zong go-zan, ngo zong haidou zou je
tomorrow seven o’clock that time I still at here do thing
“I will still be working at seven o’clock tomorrow.”
(25) Ngo haidou sik je
I at here eat thing
“I am eating”
The above examples aim to show that haidou is independent of any time location
and so can be used in the past, present and future. By default however, the present
tense will be employed.
(26) a. #Camjat saam-dim-zong go-zan, ngo zong haidou zou je, jau aamaam zou-yun
yesterday three o’clock that time I still at here do thing and just do finish
“#At three o’clock yesterday, I was still working and had just finished.”
96
b. Camjat saam-dim-zong go-zan, ngo zong haidou zou je. Ji-ga aamaam zou-yun.
yesterday three o’clock that time I still at here do thing just now do finish
“At three o’clock yesterday, I was still working. Now I’ve just finished.”
The above examples aim to demonstrate that haidou has to be interpreted in the
imperfective sense.
(27) Ngo gau-nin fung laibaailok (*haidou) paaubou
I last year every Saturday jog
“Last year I jogged every Saturday.”
(28) Ngo jiga fung laibaailok (*haidou) paaubou
I now every Saturday jog
“Now I jog every Saturday.”
(29) Ngo ha -go jyut hoici fung laibaailok (*haidou) paaubou
I next CL month start every Saturday jog
“Starting from next month, I will jog every Saturday.”
Examples (27)-(29) aim to show that the imperfective haidou should be
classified as a continuous aspect since it is not habitual.
What I have shown in this section is that the preverbal haidou is a continuous
aspect marker in which the time adverbial, if there is one, is anchored by the
predication it specifies. In this way a relative tense is produced which has nothing to
do with absolute tense but is set to the present tense by default. Additionally there is
a syntactic test to distinguish the aspect marker haidou from the prepositional phrase
haidou.

A comparison between the Cantonese aspect markers gan and haidou
2. The progressiveness of gan and haidou
This section investigates the main concern of this paper, namely the problem of
whether gan and haidou are progressive aspect markers. For the time being, I will
just assume that they both can act as a continuous imperfective aspect marker. As
the past literature mentioned, gan can convey the progressive sense, i.e. that of
ongoing actions. Though less attention was paid to haidou, some linguists did take it
as conveying progressive aspect and as an alternative to gan or as co-occurring with
gan to emphasize the progressive sense (Cheung 1972, Gao 1980, Li, et al 1995, Yip
& Matthews 1995). I agree that both gan and haidou can convey progressive sense,
as illustrated in the following examples:
(30) a. Keoi jiga {siu-gan / tiu-gan-mou / tai-gan go-bun syu / hau-gan dou mun}.
He now {laugh GAN / dance GAN / see GAN that CL book / knock GAN
CL door}
b. Keoi jiga haidou {siu / tiu-mou / tai go-bun syu / hau dou mun}.
He now HAIDOU {laugh / dance / see that CL book / knock CL door}
“He is now laughing / dancing / reading that book / knocking at the door.”
The verb constellations which can freely adopt gan and haidou are so common
that people may have the illusion that the two aspect markers have the same
distribution. Nevertheless, there are some other examples which indicate certain
subtle differences. I name the following differences by (D1), (D2) and (D3). Let us
look at (D1):
(31) a. Keoi jiga {faannou-gan / soengsam-gan / beng-gan / soen-gan Jesou}.
He now {worry IMPFV / sad IMPFV / sick IMPFV / believe IMPFV Jesus}
“He is now being worried / being sad / being sick / believing in Jesus.”
b. *Keoi jiga haidou {faannou / soengsam / beng / soen Jesou}.
In (31a), the use of gan produces grammatical sentences, but in (31b), despite the
same predicates being employed as those in (31a), haidou cannot be used. Let us
pass on to (D2):
(32) a. Keoi sik-gan faan dou haidou tai syu.
He eat IMPFV rice also IMPFV see book
“While having a meal, he is still reading a book.”
b. *Keoi haidou sik-faan dou tai-gan syu.
In (32a), gan is used in the adverbial clause and haidou in the main clause
without affecting grammaticality. However, in (32b), with the same verb
constellations as those in (32a), swapping gan and haidou makes the sentence
ungrammatical. Now let us look at (D3):
97

Chi-Fung Lam
(33) a. Keoidei paak-gan-to.
They date IMPFV
(i) “They are dating (right now)”
(ii) “They are (in the state of) dating. / They are a pair of lovers.”
98
b. Keoidei haidou paakto.
They date IMPFV
“They are dating (right now)”
(34) a. Keoi duk-gan syu.
He read IMPFV book
(i) “He is studying (right now).”
(ii) “He is (in the state of) studying. / He is a student”
b. Keoi haidou duk syu.
He IMPFV read book
“He is studying (right now).”
A lot of verb constellations incorporating gan are ambiguous between two
readings, as is the case in (33a) and (34a). However, only the first set of readings
survives when gan is replaced by haidou, as in (33b) and (34b). Such a subset of
verb constellations is so large that we cannot simply treat this phenomenon as
exceptional or lay it down to certain pragmatic reasons. This subset includes wan
gung (search a job), hok faatman (learn French), haauleoi (consider), zou je (work),
zeoi go go neoizai (woo the lady) and so on. How could the differences (D1), (D2)
and (D3) between gan and haidou be explained? I will later provide one reason
which is general enough to cover all three differences. Indeed this is the main goal
of the paper. Since the explanation involves situation aspect however, we should
first take a look at the classification of situation aspect with remarks on its
application in Cantonese. After that, I will mention and discuss my explanation of
(D1), (D2) and (D3).
2.1 Classification of situation aspect
As mentioned at the beginning of Section 1, both kinds of aspect, namely
viewpoint aspect and situation aspect, are relevant to my argument. In Section 1, I
have only focused on the viewpoint aspect of gan and haidou. In this section, I will
first briefly review situation aspect and then discuss its significance for the main
goal of this paper.

A comparison between the Cantonese aspect markers gan and haidou
2.1.1 Vendler’s classical classification
Vendler (1957) classified the English verbs into Statives 8 , Activities,
Accomplishments and Achievements. He adopted various tools, some syntactic and
some semantic, to implement this classification. First, the group of Statives and
Achievements are different from the group of Activities and Accomplishments in
that the latter admit the continuous tense but the former do not:
(35) I am running. (Activity)
(36) I am drawing a circle. (Accomplishment)
(37) *I am knowing. (Stative)
(38) *I am reaching the top. (Achievement)
For the group which admits the continuous tense, Vendler suggested a semantic
criterion to make a further distinction. Suppose a verb in this group takes the
continuous tense and it is acting at some moment. If it is still true at the moment
after it ends, it is an Activity; otherwise, it is an Accomplishment. For example, even
if the event that someone was running is suddenly terminated, the fact that he did
run will never be denied; however, someone suddenly stops drawing a circle, it is
not true to say that he did draw a circle since the circle has not been completed.
Finally, we come to the group which does not admit the continuous tense. To
distinguish Statives from Achievements, Vendler used the “how long” question.
This type of question is compatible with Statives but not with Achievements. For the
latter, instead, the “at what time” question is appropriate:
(39) For how long did you love her? (Stative)
For three years.
(40) *For how long did you reach the top? (Achievement)
(41) At what time did you reach the top? (Achievement)
At noon sharp.
After a brief introduction to Vendler’s system of situation aspect, I will consider
Smith’s system, which constitutes another milestone on the studies of the topic.
2.1.2 Smith’s two-component theory
Smith (1997) comprehensively investigated viewpoint aspect and situation
aspect. I will mention five points extracted from Smith (1997) which are relevant to
the goal of this paper. To be clear, I will discuss them one by one.
8 Vendler’s (1957) original term is ‘State’, and Smith (1997) seems to use ‘State’ and
‘Stative’ without any distinction. To be consistent, this paper uses ‘Stative’ for ‘State’ in
Vendler (1957) and reserves ‘State’ for a subset of ‘Stative’ which will be defined in section
2.1.3.
99

Chi-Fung Lam
First of all, Smith emphasized that situation aspect is conveyed by the verb
constellation which is defined by a main verb and its arguments, including the
subject, whereas viewpoint aspect is conveyed by a grammatical morpheme, usually
verbal. Therefore, when determining the situation aspect of a sentence, she considers
not only the verb and object, just as Vendler did, but also other arguments, including
adjuncts. This criterion allows the following classification:
(42) a. Mary walked to school. (Accomplishment)
b. Mary walked in the park. (Activity)
The main difference of the above two sentences is that (a) offers a natural
endpoint while (b) does not since Mary might be wandering in the park without any
goal.
Secondly, Smith noticed that adverbials are significant in determining the
situation type:
(43) a. Bill knew the truth. (Stative)
b. Suddenly Bill knew the truth. (Achievement)
In the above, adding the adverb “suddenly” changes the Stative to an
Achievement. Nevertheless, Smith insisted that every verb constellation still has its
own primitive situation type, which is called the basic-level categorization. And
when there is a type shift triggered by adverbials or other information from context,
it is called the derived-level categorization. In this paper, the former one is mainly
considered.
Thirdly, other than the four types suggested by Vendler, Smith considered one
more type, called Semelfactive, which includes verb constellations such as “knock
on the door” and “cough”. To a certain extent, a Semelfactive is similar to an
Achievement in that it refers to an instantaneous action and so is not supposed to be
in the continuous aspect:
(44) a. ?Mary knocked for five minutes. (Semelfactive)
b. ?Mary was coughing. (Semelfactive)
A knock or a cough is instantaneous and should have no duration. However, (44)
are not ungrammatical as they can be interpreted as a continuous series of many
single actions. A repetitive Semelfactive however should be interpreted as a
multiple-event Activity. What then is the main difference between Semelfactive and
Achievement? In the latter, the result is achieved by the action, while in the former,
no necessary result remains after the action. For example, once someone has
achieved the result of “reaching the top”, they should stay at the top, at least for a
while, which is a change of state when compared to the situation before the action.
In contrast, after someone has completed the action of “knocking on the door” once,
there should be no change of state for either the person or the door. Thus a
Semelfactive can be, and usually is expected to be repeated many times to derive an
100

A comparison between the Cantonese aspect markers gan and haidou
activity. In contrast, once someone has reached the top, it is impossible to reach it
again within a short period. In short, Semelfactives and Activities are [-telic]
whereas Achievements are [+telic].
Fourth, Smith regarded aspect as a parameter which is realized differently in the
languages of the world and intended to provide an adequate theory to account for the
similarities and differences in aspectual systems. She regarded the two-component
theory as abstract enough to achieve the above goal. The theory asserted that the
aspectual meaning of a sentence is a composite of the viewpoint aspect and situation
aspect. Accordingly, the following two Accomplishments can be distinguished by
their different viewpoint aspects:
(45) a. Mary walked to school. (Accomplishment + Perfective)
b. Mary was walking to school. (Accomplishment + Imperfective)
Smith claimed that these two aspectual systems are independent (Smith 1997:81-
86). One of the arguments, which is relevant to my goal in this paper, is whether the
imperfective and the stative belong to the same category. As a result of stative
imperfectives existing in languages like Russian, French and Mandarin Chinese, she
concluded that imperfective and stative are two different categories.
Finally, she attributed the five situation aspects to three distinct features, namely
[static] vs. [dynamic], [durative] vs. [instantaneous] and [telic] vs. [atelic]. This
provides a foundation of the situation aspect as a parameter which is realized in
different ways in different languages. The following table summarizes this work:
(46) a.
[+static]
Stative
b.
Situation aspect
[+telic]
Accomplishment
[-static]
Event
[+durative] [-durative]
[-telic]
Activity
Static Durative Telic
Stative [+] [+] [-]
Activity [-] [+] [-]
Accomplishment [-] [+] [+]
Semelfactive [-] [-] [-]
Achievement [-] [-] [+]
[+telic]
Achievement
[-telic]
Semelfactive
101

Chi-Fung Lam
These five situation aspects should be able to cover all verb constellations in
every language. Theoretically, three binary features should generate eight different
combinations. Smith (1997) recognized only five combinations however. This is due
to the nature of [+static], which intrinsically implies [+durative] and [-telic], so that
three other combinations are excluded. The tree diagram in (46a) is therefore just a
convenient way to construct a hierarchical classification which facilitates the
discussion of my argument in this paper as [±static] is my primitive concern. As a
matter of fact, we can also start with [±durative] in the first branch. Thus, in
principle, (46b) is clearer and more comprehensive.
2.1.3 Comments on Smith’s theory
Among the five points extracted from Smith (1997) which I have just mentioned
above, I have some remarks on the last two which may affect the main goal of this
paper. In the fourth point, it is doubtful whether the two kinds of aspects are really
independent, as claimed by Smith (1997:81-86). In her discussion on the
relationship between statives and imperfectives, it seemed that the term
“progressive” is equated to “imperfective”. She shifted between the terms freely in
the discussion. Although I agree with her conclusion that the imperfective and
Stative are two different categories as the imperfective Stative exists in some
languages, it should be recalled that the progressive and Stative are still mutually
exclusive in distribution. This then involves the nature of the progressive and the
Stative. For the former, which refers only to events, it appears more precise to
concede a stricter and narrower sense of progressiveness. For the latter, I prefer to
divide the Stative into two subsets, namely ‘State’ and ‘Property’. In the following
sections I will clarify these two stands.
2.1.4 Nature of progressive and Stative
In this section, I will show in detail the nature of progressive and Stative since
this helps us reach the main goal of this paper, that is to clarify the differences (D1),
(D2) and (D3) of gan and haidou, raised in the beginning of Section 2. In (47), (48)
and (50), the data comes from discussing English aspects. To ensure that the
implication is still applicable to Cantonese, a Cantonese translation is given
following the English data.
Vlach (1981) believed that the progressive is actually a Stative. In his opinion,
the progressive operator, such as the “be + V-ing” construction in English, makes
events stative and never takes a stative verb constellation since they are already
stative. He justified this claim by comparing these two sentences:
(47) a. Max was here when I arrived.
Ngo lei dou go zan, aa Max hai dou.
b. Max was running when I arrived.
Ngo lei dou go zan, aa Max paau-gan bou.
102

A comparison between the Cantonese aspect markers gan and haidou
Vlach claims that be here and be running are both stative since these two
predicates have to be true for a non-empty interval before and at the moment of the
happening of the event I arrived. With a little modification, I present his definition
of Stative as follows:
(48) A verb constellation VP is defined as a Stative if the truth of the sentence
“PAST(VP) when I arrived” requires that VP was true for some period
leading up to the time of my arrival. 9
The usual understanding of the term “progressive” refers to ongoing actions.
Therefore, it is more reasonable to reserve the [dynamic] feature for progressive. If
the verb constellation conveys no internal stage or the internal stage cannot be
detected in the physical world, it should not have the quality of a progressive. Thus,
I reject considering the progressive as a Stative. However, this intuition has to be
justified by a linguistic test. Smith (1997:84-86) noticed that by using this kind of
when-clause test, Statives can still be distinguished from progressives:
(49) a. Mary was angry when John broke the glass.
Aa John da lan zek bui go zan, aa Mary hou nau.
b. Mary was singing when John broke the glass.
Aa John da lan zek bui go zan, aa Mary coeng-gan go.
In the examples above, Mary’s singing has to happen before John’s breaking the
glass. I call this the “before” reading. However, Mary’s being angry can happen
before or just after John’s breaking the glass. It then has two readings, namely the
“before” reading and “after” reading. But there is another problem for this criterion
to be workable, that is, for many typical Statives, the whole clause sounds very odd:
(50) a. #Max (already) understood Kant when I arrived.
#Ngo lei dou go zan, aa Max (jigging) ming baak zo Hongdak
b. ?Max was (already) tall when I arrived.
?Ngo lei dou go zan, aa Max (jigging) hou gou.
By adding the adverb “already”, (b) is rendered distinctly more grammatical, but
it still remains quite stilted. In fact, Vlach noticed this oddity but he accepted that
the adding of “already” is sufficient to resolve it, and so he still adopted (50) as the
definition of Stative. Thus, this broad definition of Stative not only includes all the
statives usually accepted, but also progressives. I would like to point out two
problems here. First, in (47a), “be here” is special as it is logically implied by “I
arrive” and cannot be treated as a general case of Statives. It is therefore not an
appropriate demonstration of the definition. Second, it is clear that “be angry” is
9 This VP is only a convenient symbol for verb constellation and is not restricted to the
traditional structure of the verb phrase; rather it may include arguments and light verbs.
103

Chi-Fung Lam
semantically different from “be tall”. It is this distinction which leads to the oddity
mentioned above. We should follow this intuition to split elements that are usually
classed as Statives into two subclasses. “Be angry” may change easily and swiftly,
and is to a great extent controllable by the agent but “be tall” may not. The former
can be classified as Property while the latter is better classed as a State. Then we
may consider Stative as consisting of Property and State. “Understand” is more like
a Property but not so solid as “be tall”. So in (50a), “understand” can be interpreted
as happening after or just before the when-event, when “already” is absent, just like
the two interpretations of “be angry” in (49a). Furthermore, I propose that the
progressive be reserved for events, like “run”. Here I use the term “event” so as to
distinguish it from “stative” but it should be understood that only the Activity and
Accomplishment typically take the progressive. If was running in (47b) is replaced
by ran which is not progressive but neutral in viewpoint aspect, the action of Max’s
running should be interpreted as starting just after my arrival. Therefore, by using
such when-clause test, we have actually four exclusive subsets as follows:
(51)
“before” reading “after” reading
Property No (or not available) No (or not available)
State Yes Yes
Event (in progressive) Yes No
Event (in neutral viewpoint) No Yes
The last case does not exist in Cantonese since an explicit, rather than a neutral,
viewpoint aspect is obligatory there. The subclass Property has been suggested in
Francis and Matthews (2005). They classified the verbs in Cantonese in three classes,
namely Property, State and Action. They adopt the semantic feature [permanent] as
the defining characteristic of Property, which is equivalent to the idea of individuallevel
predication, as opposed to stage-level predicate (Francis and Matthews 2005:
294-6, Smith 1997:32-5, Carlson 1977). Lehmann (1999) thought that a State is
transitory, contingent, more temporary and implies endpoints. A Property on the
other hand may characterize an entity. He gave an example which highlights this
distinction. Spanish uses two copulas, ser and estar, with adjectival predicates to
express respectively Property and State:
(52) a. Juanita es guapa.
“Juanita is pretty.”
104
b. Juanita está guapa.
“Juanita is looking pretty.”
I claim that in Cantonese, the verbal suffix gan is a general imperfective aspect
marker which can follow the State while the preverbal haidou is a progressive aspect
marker which cannot select the State. I will justify this claim in the next section. But

A comparison between the Cantonese aspect markers gan and haidou
to show that it is significant to classify Stative into Property and State in my
discussion, I point out that gan is compatible with State but not with Property. See
the following examples:
(53) a. Keoi hou go.
He very tall
“He is (very) tall.” 10
b. *Keoi go -gan.
He tall IMPFV
“He is getting taller and taller.”
(Intended meaning: “He is tall”)
(54) a. Keoi hou soengsam.
He very sad
“He is (very) sad.”
b. Keoi soengsam-gan.
He sad IMPFV
“He is sad.”
Both (53) and (54) are Statives but (53) is a Property, and so is [+permanent],
while (54) is a State, and is instead [-permanent]. (54b) is perfectly compatible with
gan but in (53b), gan changes the reading of “be tall” to “become tall”. The reason
of this shift is that “be tall” is a Property, which is in nature incompatible with gan.
Finally, let me conclude my classification of situation aspect by re-drawing the
table in (46) as follows:
(55) a.
[+permanent]
Property
[+static]
Stative
[-permanent]
State
Situation aspect
[+telic]
Accomplish
ment
[-static]
Event
[+durative] [-durative]
[-telic]
Activity
[+telic]
Achieve-
ment
Compatible with the general imperfective aspect marker gan
Incompatible with the progressive aspect marker haidou
[-telic]
Semel-
factive
10 By default, the Cantonese adjectival predicates are introduced by hou, an adverb of degree
meaning “very”. So in (53a), the “very” sense is possible but not obligatory.
105

Chi-Fung Lam
b.
Static Durative Telic Permanent
State [+] [+] [-] [-]
Property [+] [+] [-] [+]
Activity [-] [+] [-] [-]
Accomplishment [-] [+] [+] [-]
Semelfactive [-] [-] [-] [-]
Achievement [-] [-] [+] [-]
2.2 Difference between gan and haidou
In this section, I will justify the above claim that the verbal suffix gan is a
general imperfective aspect marker and the preverbal haidou is a progressive aspect
marker. This claim is supported by its ability to explain the differences (D1), (D2)
and (D3), mentioned in the end of Section 1, where examples (31)-(34) are provided.
Let me go through these examples (renumbered) and demonstrate how positing a
difference between the two verbal elements allows these three differences to be
accounted for.
(D1):
(56) a. Keoi jiga {faannou-gan / soengsam-gan / beng-gan / soen-gan Jesou}
He now worry IMPFV / sad IMPFV / sick IMPFV / believe IMPFV Jesus
“He is now being worried / being sad / being sick / believing in Jesus.”
b. *Keoi jiga haidou {faannou / soengsam / beng / soen Jesou}.
Usually, we classify the verb constellations faannou (being worried), soengsam
(being sad), beng (being sick) and soen Jesou (believing in Jesus) as Statives.
However, in my approach, they should be classified as States, as distinct from
Properties since these verbs generally refer to a temporary situation which is
consequently subject to change. It should be noted that the last one soen Jesou
(believing in Jesus), is more ambiguous as someone’s religious belief is supposed to
last for a long amount of time, possibly their whole lifetime. The imperfective soengan
Jesou can therefore be used only in some contexts. For example, it may describe
a believer who is not so serious about their religion and has changed it several times,
or the message of their belief is new and surprising to the listener. In short, all these
verb constellations involve no physical action or dynamicity, meaning that they are
incompatible with the progressive aspect marker haidou but compatible with the
general imperfective aspect marker gan.
(D2):
(57) a. Keoi sik-gan faan dou haidou tai syu.
He eat IMPFV rice also IMPFV see book
“While having a meal, he is still reading a book.”
106
b. *Keoi haidou sik-faan dou tai-gan syu.

A comparison between the Cantonese aspect markers gan and haidou
In (57a), gan is used in the adverbial clause and haidou in the main clause.
However, the sentence will be ungrammatical if the two aspect markers swap, as in
(57b). Of course, there are still two other cases:
(57) c. Keoi sik-gan faan dou tai-gan syu.
d. Keoi haidou sik-faan dou haidou tai syu.
“He is having a meal and reading a book.”
Though (57c) and (57d) are grammatical, their interpretations are slightly
different. They no longer take sik faan (have a meal) as the background and tai syu
(read a book) as the focus. Rather, they juxtapose the two ongoing actions where
dou means no longer “still” but “and”. My claim can account for this phenomenon.
Since gan is a general imperfective which can convey a stative meaning, it stativizes
the action of sik faan so that sik-gan faan can be an adverbial. However, haidou is a
progressive aspect marker, which has to emphasize onging action and thus have no
function of stativization. It could be pointed out that the item dou is ambiguous
between the two meanings of “also” and “and”, something that renders the example
sentences not very reliable. The same outcome can be derived from replacing dou by
go zan, that means only “at that moment”. The contrast between gan and haidou
would then become more prominent:
(58) a. Keoi sik-gan faan go zan haidou tai syu.
b. *Keoi haidou sik-faan go zan tai-gan syu.
c. *Keoi sik-gan faan go zan tai-gan syu.
d. *Keoi haidou sik-faan go zan haidou tai syu.
Notice that both gan and haidou can convey progressive meaning for Activities
and they can go with verb constellations like sik faan and tai syu well when there is
no adverbial:
(59) a. Keoi sik-gan faan / tai-gan syu.
b. Keoi haidou sik-faan / haidou tai syu.
Therefore, the stativization of gan is an adequate account for (57) and (58). Now
we go to the last difference of gan and haidou.
(D3):
(60) a. Keoidei paak-gan-to.
They date IMPFV
(i) “They are dating (right now)”
(ii) “They are (in the state of) dating. / They are a pair of lovers.”
b. Keoidei haidou paakto.
They IMPFV date
“They are dating (right now)”
107

Chi-Fung Lam
(61) a. Keoi duk-gan syu.
He read IMPFV book
(i) “He is studying (right now).”
(ii) “He is (in the state of) studying. / He is a student”
108
b. Keoi haidou duk syu.
He IMPFV read book
“He is studying (right now).”
A lot of verb constellations can be interpreted in two ways when they are with
gan, as in (60a) and (61a). However, only the first readings survive when gan is
replaced by haidou, as in (60b) and (61b). I call the reading in (i) “concrete” and that
in (ii) “abstract”. The main characteristic of the concrete reading is that it refers to
an ongoing action in the physical world and so can be recognized through sensory
media. The defining feature of the abstract reading on the other hand is that it refers
to a temporary state which is usually known by information, such as someone’s
occupation, relationship with others or recent hobbies, which are supposed to have
started not very soon before, exist now and last for a definite period of time. These
two readings have two sources. One source is the different interpretation on the verb
constellation. For example, in (60) paakto could refer to an activity carried out by
two lovers, such as going shopping together and acting intimate, as in (a)(i), but it
can also refer to the relationship of the pair of lovers, as in (a)(ii). Note that in each
of the two readings, paakto has the same lexical meaning. The other source of
ambiguity is from the different lexical meanings. For example, in (61) duk syu can
simply mean “read a book”, as in (a)(i), but it can also be abstracted as a state of
studying in school or equivalently being a student as in (a)(ii). Nevertheless, not all
Activities or Accomplishments permit the dual reading with gan, as illustrated in
(62):
(62) a. Keoidei sik-gan je / haidou sik je.
“They are eating”
b. Keoi teng-gan jamngok / haidou teng jamngok.
“He is listening to music.”
c. Keoi tai-gan syu / haidou tai syu..
“He is reading a book.”
Many verb constellations such as sik je (eat) and teng jamngok (listen to music),
cannot be interpreted as a state. (61) and (62c) make a sharp contrast. Though both
duk syu and tai syu mean “read a book”, only duk syu can imply “be a student”. So
for this kind of verb constellations gan-sentences do not convey the other reading. In
short, the abstract reading can be obtained by States only when the verb
constellations are potential; in this case, gan can appear while haidou cannot.

(63)
A comparison between the Cantonese aspect markers gan and haidou
I conclude my discussion on (D3) with the following table:
Verb
constellation
Reading
“Concrete”
(with gan / haidou)
“Abstract”
(with gan)
With only one
reading
Sik je Eat Not available
Teng jamngok Listen to music Not available
Tai syu Read a book Not available
With two readings;
Ambiguity lies in
interpretation
Paakto Date Relationship of lovers
Wan gung search a job
(e.g. from newspaper or
online right now)
Hok faatman learn French
(e.g. reading a French
textbook right now)
In the state of searching a job
(being unemployed)
In the state of learning French
(e.g. having a 2-hour French
lesson every Monday )
Haauleoi Consider and about to reply In the process of consideration
With two readings;
Ambiguity lies in
lexical meaning
Zou je Do something
(e.g. doing a task)
Work
(as an occupation)
Duk syu Read a book Be a student
Zeoi go go neoizai Run after the lady Woo the lady
2.3 A brief conclusion of Section 2
The following conclusions can be derived from Section 2. First, based on the
system of viewpoint and situation aspects advocated by Smith (1997) and on some
data in English, Cantonese and Spanish, I proposed to treat the two subclasses of the
Stative, Property and State, as being on the same level as the other four situation
aspects, namely Activity, Accomplishment, Achievement and Semelfactive. This
109

Chi-Fung Lam
splitting is supported by the when-clause test, which indicates a syntactic and
semantic distinction among Properties, States and progressives. Significantly, this
division can adequately and consistently account for the differences (D1), (D2) and
(D3) between the verbal suffix gan and preverbal haidou in Cantonese. Simply put,
the general imperfective aspect marker gan has the ability to stativize and can go
with States (but not Properties) whereas the progressive aspect marker haidou
cannot co-occur with any Stative.
3. Non-aspectual factors of the difference between gan and haidou
Though aspect is quite effective in distinguishing the distribution and meaning of
gan and haidou, there are some other factors which may restrict their distribution.
These constraints may be of a syntactic, semantic or pragmatic nature. In (64) and
(65), I give one example for each of these three types of restrictions.
The first is a syntactic constraint. Let us consider the following data:
(64) Peter hai tousyugun japbin haidou hang lei hang heui.
Peter at library inside PROG walk come walk go
“Peter is wandering inside the library.”
(65) a. *Peter hang lei hang heui-gan.
b. *Peter hang-gan lei hang heui.
c. *Peter hang-gan lei hang-gan heui.
In Cantonese, the construction V-lei-V-heui has an iterative meaning, that is, the
action denoted by the verb V repeats continuously. Though lei (come) and heui (go),
as such, are verbs, they act as directional particles in the construction. I would like to
demonstrate that haidou can precede this construction to impose a progressive
meaning. In (64), to avoid haidou being interpreted as locative adverb, Peter should
be considered as being inside the library. He is therefore very likely not in the same
place as the speaker at the moment of speech, and thus force haidou to be interpreted
as a progressive aspect marker. The grammaticality of (64) shows that this is really
the case. Therefore, in principle, gan should also be able to make the sentence
progressive. However, the syntactic property of the construction V-lei-V-heui
prohibits the appearance of any other postverbal particles, including gan. No matter
the position in which gan is put, the sentence is ungrammatical, as shown in (65a)-
(65c) (only three of the possible positions are shown). These examples therefore
demonstrate one exceptional case where the verb constellation is compatible with
haidou but not gan.
The second is a semantic constraint. Let us consider the following data:
(66) Peter hang-gan gwolai / gwoheui.
Peter walk PROG come here / go there
“Peter is walking towards here / towards there.”
110

A comparison between the Cantonese aspect markers gan and haidou
(67) a. *Peter haidou hang gwolai.
b. Peter haidou hang gwoheui.
Peter (from) here walk go there
“Peter is walking towards there from here”
Again, gwolai (come here) and gwoheui (go there) are directional particles. The
literal meaning of haidou is “at here, be here”, which is frequently used as a
preverbal locative adverb. If the action denoted by the verb constellation contains an
obvious shift in location, then it will conflict with the semantics of haidou and result
in ungrammaticality. In (67a), haidou is not compatible with gwolai (come here)
which implies that Peter is not in the same place as the speaker. Additionally,
despite the grammaticality of (67b), haidou can only be interpreted as a locative
adverb, meaning “be here, at here” but sometimes understood as “from here”. This is
illustrated by the sentences in (68), which are compatible with gwoheui (go there).
These examples demonstrate an exception where a typical Activity is compatible
with gan but not haidou.
The last is a pragmatic constraint. Let us consider the following example:
(68) a. Ngodei cidou la, nei zong haidou sik je!
we late SFP you still PROG eat thing
“We are late. You’re still eating!”
b. #Ngodei cidou la, zong sik-gan je!
we late SFP still eat PROG thing
“We are late. You’re still eating!”
If someone intends to show (usually negative) surprise about an ongoing action
being done by someone else, haidou is more appropriate, as the contrast between
(68a) and (68b) shows. Since the use of the progressive in this way, aims at pointing
out the fact that the action should have been completed and not be ongoing, the
temporal adverb zong “still” is often used to form constructions zong haidou V and
zong V-gan. On the other hand if a person intends to prevent someone else from
getting a bad shock following an ongoing action of theirs, gan is more appropriate.
This is shown in the contrast between in (69a) and (69b) below:
(69) a. m-hou-yisi, ngo zidou jiging gwo sai zong,
danhai ngo zong sik-gan je, jiu ci-di sin dou.
sorry I know already over all clock
but I still eat PROG thing need late.little first arrive
b. #m-hou-yisi, ngo zidou jiging gwo sai zong,
danhai ngo zong haidou sik je, jiu ci-di sin dou.
sorry I know already over all clock
but I still PROG eat thing need late.little first arrive
111

Chi-Fung Lam
112
“Sorry, I know that I’ve been late.
But I’m still eating and I’ll be a bit late.”
The latter utterance shows that the person who is late does not intend to speed up
and is not regretful at all. Hence, it sounds odd, especially if there is “sorry” at the
beginning. However, it is not easy to explain why there is such a difference between
gan and haidou.
4. Conclusion
The verbal suffix gan in Cantonese is a progressive aspect marker which cooccurs
with Activities and Accomplishments, while the preverbal haidou is a
locative adverb that can sometimes substitute gan to convey progressive meaning.
Our findings show that:
(G3) Haidou is a typical progressive aspect marker which is used in Activities and
Accomplishments.
(G4) Gan should be treated as a general imperfective aspect marker. While it can
be used with Activities and Accomplishments, it can also be employed with
[-permanent] States.
This claim not only clarifies the aspectual nature of gan and haidou, but also
provides a consistent explanation to account for some syntactic and semantic
differences between them, as named (D1), (D2) and (D3) in this paper. Moreover, it
supports the splitting of Statives into States and Properties in Smith’s classification
of the five situation aspects. In order to defend my claim, it was necessary to
enumerate some apparent exceptions. I attributed them to non-aspectual factors
which include syntactic, semantic and pragmatic characteristics of gan and haidou.
References
Carlson, Gregory N. (1977) Reference to kinds in English. Doctoral dissertation, University
of Massachusetts, Amherst.
Chao, Yuen-Ren (1968) A Grammar of Spoken Chinese. Berkeley: University of California
Press.
Cheung, Samuel Hung-Nin (1972) Xianggang Yue yu yu fa de yan jiu [Cantonese as spoken
in Hong Kong]. Hong Kong: The Chinese University of Hong Kong.
Cinque, Guglielmo (2010) Mapping Spatial PPs: an introduction. In Guglielmo Cinque and
Luigi Rizzi, eds., Mapping Spatial PPs. The Cartography of Syntactic Structures,
Volume 6: 3-25. New York: Oxford University Press.
Francis, Elaine J. and Stephen Matthews (2005) A multi-dimensional approach to the
category ‘verb’ in Cantonese, Journal of Linguistics 41: 269-305.
Gao, Huanian (1980) Guangzhou fang yan yan jiu [A Study of Guangzhou Cantonese]. Hong
Kong: Commercial Press.

A comparison between the Cantonese aspect markers gan and haidou
Gu, Yang (1999) Dongci de ti ji titai [Aspect and aspectuality of verbs]. In Xu Liejiong ed,
Gong xing yu ge xing: Han yu yu yan xue zhong de zheng yi [Universals and
Particulars: Controversies in Chinese Linguistics]. Beijing: Beijing Language and
Culture University Press.
Huang, C.-T. James, Y.-H. Audrey Li, and Yafei Li (2009) The Syntax of Chinese. New York:
Cambridge University Press.
Lehmann, Christian (1999) Aspectual type(s). Keith Brown, Jim Miller eds. Concise
Encyclopedia of Grammatical Categories. Amsterdam: Elsevier.
Li, Xinkui, et al (1995) Guangzhou fang yan yan jiu [A Study of Guangzhou Cantonese]
Guangzhou: Guangdong ren min chu ban she.
Smith, Carlota S. (1997) The Parameter of Aspect (second edition). Dordrecht: Kluwer
Academic Publishers.
Vendler, Zeno (1957) Verbs and times, The Philosophical Review, Vol. 66, No. 2. (Apr.,
1957), pp. 143-160. Also in Zeno Vendler ed. (1967), Linguistics and Philosophy.
Ithaca: Cornell University Press.
Vlach, Frank (1981) The semantics of progressive. In Philip Tedeschi and Annie Zaenen eds.
Syntax and Semantics 14: Tense and Aspect. New York: Academic Press.
Yip, Virginia, and Stephen Matthews (1994) Cantonese: a Comprehensive Grammar. London:
Routledge.
113

Rivista di Grammatica Generativa
34 – 2009
CARATTERI DELLA CLITICIZZAZIONE DI NE
Lidia Lonzi
In quest’analisi della ne-cl(iticization) l’opposizione pertinente coinvolge coppie
minime contenenti non verbi inaccusativi vs. inergativi, bensì verbi inergativi
in forma semplice (accettabile) vs. perifrastica (inaccettabile). Questo dato viene
interpretato all’interno dell’analisi standard (Belletti & Rizzi, 1981, Burzio, 1986)
e successivi sviluppi, in particolare Kayne (1993) sulla selezione dell’ausiliare. Esso
rifletterebbe un sistematico slittamento dei verbi interessati dallo status di inergativi
a quello di inaccusativi, permesso nel caso della forma semplice ma bloccato
nel caso della forma perifrastica dall’interazione della struttura sintattica di Avere
con il clitico partitivo.
La caratterizzazione sintattico-semantica e di struttura informativa della ne-cl
che ne deriva permette una prima verifica dell’ipotesi di Levin & Rappaport Hovav
(1995) sull’analogia di condizioni discorsivo-funzionali tra ne-cl e locative inversion
dell’inglese.
1. Premessa
– Dal punto di vista sintattico, la spiegazione standard della cliticizzazione di
ne (ne-cl) è quella fornita dall’analisi di Belletti & Rizzi (1981), Burzio
((1981)1986), Chomsky (1981), ed elaborata successivamente in Cinque (1990),
Cardinaletti & Giusti (1992), tra gli altri. Essenzialmente, nelle frasi con verbo inaccusativo,
la traccia dell’NP quantificato può essere legata da un ne partitivo cliticizzato
al verbo. Poiché solo la posizione oggetto risulta c-comandata dal clitico,
frasi con ne-cl sono possibili con i verbi inaccusativi, il cui soggetto è appunto generato
in posizione oggetto, ma non con gli inergativi il cui soggetto è un argomento
esterno. Quest’analisi ha subito un aggiornamento sostanziale in Belletti (2002),
che vedremo (2.2.-2.3).
– Dal punto di vista lessico-semantico, sono da rilevare due dati per quanto concerne
la sua applicabilità:
(i) ne-cl è applicabile in maniera generalizzata solo a cose, a nomi col tratto
[-umano], perché la nozione stessa di “ne partitivo” implica materialità, perti-
115

Lidia Lonzi
nente per la quantificabilità dell’NP. In particolare, restringendo l’attenzione ai
nomi numerabili, occorre un referente riducibile a una somma di entità uguali e, in
quanto materiale, visibile, udibile, mobile, occupante spazio. I soggetti umani devono
poter subire una riduzione a un denominatore comune: una vera e propria
“oggettualizzazione” – in parte comportata dalla stessa struttura del verbo inaccusativo,
che assegna ruolo-θ tema al suo argomento (v. 2.2.1). Di qui l’inaccettabilità
di una frase come (1a) vs. (2a). In (1a), infatti, è bloccato il senso metaforico presente
nella frase predicativa (1b) (v. anche (1c) vs. (2c)):
(1) a *Ne sono maturati tanti, di genitori, in questa occasione
b Tanti genitori sono maturati in questa occasione
c(?)?Sono maturati tanti genitori, in questa occasione
(2) a Ne sono scesi tanti, di passeggeri, alla prima fermata
b Tanti passeggeri sono scesi alla prima fermata
c Sono scesi tanti passeggeri, alla prima fermata
(ii) ne-cl è possibile anche con i verbi inergativi, purché le condizioni specificate
in (i) siano rispettate. V. l’accettabilità di (4a) vs. (3a), inaccettabile per le note
ragioni, soprattutto a livello stilistico:
(3) a *Ne meditano tante, di persone, qui
b TANTE PERSONE meditano, qui
c Meditano tante persone, qui
(4) a Ne passeggiano tante, di persone, qui
b TANTE PERSONE passeggiano, qui
c Passeggiano tante persone, qui
– Infine, dal punto di vista della struttura informativa, si può sostenere che nella
frase con ne-cl il predicato è presupposto, e il quantificatore è in focus. Questo aspetto
si collega in maniera interessante con la questione sollevata in Levin & Rappaport
Hovav (1995) riguardo alle caratteristiche comuni alle due costruzioni a cui
è generalmente attribuita una capacità diagnostica di inaccusatività: ne-cl, appunto,
e la locative inversion dell’inglese, in cui il focus sarebbe costituito dal soggetto.
Nelle sezioni che seguono riprenderò in considerazione tutti e tre gli aspetti appena
visti: sintattico, § 2, lessico-semantico, § 3, e di struttura informativa, § 4. La
distinzione, come risulterà dalla stessa discussione, risponde a fini espositivi. Lo
scopo è quello di vedere più da vicino questa costruzione – chiarendo, in particolare,
la questione del suo uso con i verbi inergativi – e, insieme, dare un primo contributo
alla comprensione dei tratti in comune con la locative inversion.
116

Caratteri della cliticizzazione di ne
2. Specializzazione di ne per il cognate object dell’inergativo
2.1 Il fenomeno dell’accordo
È noto che l’opposizione pertinente nell’analisi della ne-cl coinvolge coppie
minime contenenti verbi inaccusativi (ausiliare Essere) vs. inergativi (ausiliare Avere).
Meno noto, o almeno accettato, è che possono formare coppie minime contenenti
verbi inergativi nella forma semplice vs. composta. Si veda (5a) (= (4a)), col
verbo in forma non perifrastica, vs. (5b); oppure (6a) vs. (6b):
(5) a Ne passeggiano tante, di persone, qui
= “sono tante le persone che passeggiano, qui”
b *Ne hanno passeggiato tante, di persone, qui
c Hanno passeggiato tante persone, qui
(6) a Ne dormono tanti, di bambini, qui fuori
= “sono tanti i bambini che dormono, qui fuori”
b *Ne hanno dormito tanti, di bambini, qui fuori
c Hanno dormito tanti bambini, qui fuori
Le due coppie minime (5a-b) e (6a-b), entrambe con verbo inergativo, mostrano
che il fenomeno stabilito sopra, al punto (ii), vale solo quando il verbo è nella forma
semplice. L’ulteriore contrasto tra le frasi (b) vs. (c) mostra che il fattore determinante
per l’inaccettabilità delle frasi (b) è la ne-cl. V. Bentley (2004). 1,2
1 Una spiegazione della devianza di ne-cl con i verbi inergativi è avanzata in Bentley (2004),
in un quadro di Role and Reference Grammar. Secondo Bentley, le eccezioni osservabili in
alcuni casi con le forme non perifrastiche (segnalate in Lonzi, 1986) sarebbero comunque
devianti perché implicherebbero “una disparità fra la sintassi e la semantica”, fra il predicato
d’azione, e, rispettivamente, il predicato stativo associato alla lettura esistenziale. La mia
analisi tende ad analizzare tali “eccezioni” come sistematici slittamenti dei verbi interessati
dallo status di inergativi a quello di inaccusativi, per cui tale disparità risulterebbe annullata.
Infatti, se tale slittamento non è permesso, per la natura stessa del predicato d’azione, v.
(3a), e/o per i particolari tratti del nome soggetto, v. (1a), la frase risulta inaccettabile.
2 Si noti che la lettura preferita di una frase come (i): “ci sono tanti bambini dormienti nell’unico
letto a castello”, ne spiega direttamente l’inaccettabilità su basi pragmatiche. In (i),
non è possibile la lettura intesa con valore distributivo (“sono tanti i bambini che dormono
abitualmente ognuno nel suo letto a castello”) derivante da un’interpretazione non “materiale”
del verbo, lettura che è possibile nella frase predicativa (ii):
(i) *Ne dormono tanti, di bambini, nel letto a castello
= “sono tanti i bambini che dormono nel letto a castello”
(ii) Tanti bambini dormono nel letto a castello.
117

Lidia Lonzi
L’inaccettabilità di (5b)-(6b) è spiegabile con la struttura transitiva associata all’ausiliare
Avere, e, crucialmente, con una specializzazione del clitico ne per il cognate
object (l’“oggetto interno”) dell’inergativo (Lonzi, 1986: 111sgg.). Quando
l’ausiliare perfettivo è presente, la specializzazione di ne per il cognate object
dell’intransitivo emerge con un effetto di inaccettabilità se deve legare la traccia del
soggetto, v. (6b), e di grammaticalità se lega la traccia dell’oggetto interno, v. (7):
(7) Ne ha dormiti tanti t, di sonni agitati, quel bambino
La vera e propria spia dell’illegalità di (6b) è il mancato accordo del participio.
La non disponibilità di una posizione argomentale per l’NP quantificato, richiesta
dalla ne-cl, è rivelata dal non-accordo tra il quantificatore e il participio del verbo
inergativo nella forma debitamente non marcata (v. la discussione in 3.2).
Ne, in quanto clitico dell’oggetto, dovrebbe infatti determinare l’accordo del
participio (Burzio, 1986: 55) col DP pertinente 3 , si tratti dell’oggetto di verbi transitivi
o di inergativi, v. (7) 4 .
2.2 La posizione “oggetto”
2.2.1 La posizione del soggetto postverbale nelle frasi a broad focus
La posizione di base dell’NP quantificato è una posizione oggetto. Ma quanto è
pertinente il requisito del c-comando da parte di ne per la sua legittimazione?
Il problema è stato avanzato da Levin & Rappaport Hovav (1995), che si chiedono
se, adottando la teoria del “soggetto nel VP”, sia ancora sostenibile la spiegazione
data da Burzio (1986) della grammaticalità di ne-cl con i verbi ergativi e della
sua incompatibilità con i verbi inergativi. La struttura pertinente di una frase con
verbo inergativo consentirebbe ancora il c-comando da parte di ne nei confronti
della traccia del soggetto in Spec,VP, e quindi la spiegazione di Burzio sarebbe superata.
Belletti (2002), aggiorna la spiegazione di Burzio e propria (Belletti, 1988;
3 “Un participio passato si accorderà in genere e numero con un elemento avente una relazione
di binding con l’oggetto diretto” (Burzio, 1986, 55). Quanto al meccanismo coinvolto,
v. D’Alessandro e Roberts (2007), e riferimenti.
4 Oltre ai casi di cognate object, la forma ad accordo regolare del participio con l’oggetto,
illustrata in (i), dovrebbe suggerire un valore transitivo causativo, secondo il modello analizzato
in Levin & Rappaport Hovav (1995: 110-111) che, forse con sfumature dialettali, si applica
anche in italiano. Anche se questo valore non è disponibile per passeggiare, lo è, per
es., per l’inergativo dondolare.
(i)*Ne hanno passeggiate tante di persone, qui (, le nostre infermiere)
(ii) Ne hanno dondolate tante di culle, qui (, le nostre infermiere).
118

Caratteri della cliticizzazione di ne
Belletti & Rizzi, 1981) assegnando al soggetto postverbale dell’inergativo una posizione
A’ che non favorisce l’estrazione.
Belletti non prende in esame il contrasto associato alla presenza/assenza
dell’ausiliare. Nella sua analisi, la marginalità delle frasi con ne-cl e verbi inergativi
dipende dal fatto che l’estrazione pertinente non avviene, come con gli inaccusativi,
dalla posizione oggetto, ma dallo Spec,FocP – la posizione nella periferia sinistra
di VP in cui si trova il soggetto postverbale con questo tipo di verbi: 5 la decisa
marginalità di (8) va dunque spiegata con l’estrazione di ne da una posizione A’
(Spec,FocP):
(8) ??Ne ha telefonato uno al giornale (= es. (43b) in Belletti 2002)
Sempre secondo Belletti, la leggera marginalità osservabile in (10), che strutturalmente
non porrebbe problemi come non ne pone (9) – con il soggetto in
Spec,FocP – trova invece spiegazione nella struttura informativa a broad focus, che
non è canonica perché il costituente in FocP non è in posizione finale di frase. 6
(9) Ha telefonato uno studente
(10) (?)Ha telefonato uno studente al giornale (= es. (43a) in Belletti, 2002)
In (10), come osserva Belletti, al giornale fa parte del focus della frase (altrimenti,
e in maniera non pertinente, si tratterebbe di una frase con marginalizzazione
del complemento, che dovrebbe quindi assumere l’intonazione tipica del costituente
destressed. V. Cardinaletti, 2002). Tuttavia, la sua ipotesi che (10) sia:
(i) a intonazione non marcata, e, insieme,
(ii) col soggetto in Spec,FocP anziché in posizione oggetto,
5 Per semplificare l’esposizione non prendo in considerazione altre analisi recenti delle frasi
a soggetto “postverbale” – Longobardi (2000) e riferimenti, Cardinaletti (2002) – secondo
cui il soggetto in focus si troverebbe nella posizione canonica di soggetto, e sarebbe il predicato
a essersi spostato a sinistra in posizione periferica. I diversi approcci non dovrebbero
avere conseguenze per le mie conclusioni sul riordino dei complementi. Queste analisi si
aggiungono agli aggiornamenti relativi alla teoria del “soggetto nel VP” che devono affrontare
una nuova spiegazione del comportamento di ne-cl. V. Alexiadou, Anagnostopolou, Everaert
(2003).
6 La stessa marginalità che si riscontra in (9), con ordine V S PP e V inergativo, Belletti la
ritrova in frasi con lo stesso ordine e V inaccusativo ma con soggetto definito, perché in violazione
dell’indefinite requirement che vige con questi verbi, v. (i):
(i) (?)È arrivato lo studente al giornale (= es. (41a) in Belletti)
La spiegazione della marginalità di (i) sarebbe la stessa fornita per (9) (struttura informativa
non canonica), dato che il soggetto occuperebbe la posizione FocP anche in questo caso. Qui
non prendo in esame l’indefinite requirement, che merita un’indagine a sé stante.
119

Lidia Lonzi
non trova un supporto intonativo.
I contorni intonativi di una frase con soggetto postverbale e a broad focus come
(10), in termini tradizionali frase “eventiva”, sembrano infatti gli stessi di una frase
inaccusativa con l’argomento interno in situ. La spiegazione di Belletti della “decisa
marginalità” di (8), che sarebbe dovuta a estrazione da Spec,FocP, non rende
conto del dato intonativo di (10), di cui tuttavia prende atto. Anche i dati relativi ai
contesti diagnostici delle frasi con soggetto in FocP, come (11b), vs frasi a broad
focus, come (11a) – per es. la compatibilità con gli avverbi orientati sul soggetto,
osservabile solo nelle prime (v. Lonzi, 1986) – suggeriscono di prendere in considerazione
la possibilità che in frasi come (10) il soggetto sia un argomento interno.
(11) a *Imprudentemente, ha telefonato uno studente al giornale
b Imprudentemente, ha telefonato uno studente, al giornale
Nella mia analisi, (10) è accettabile perché la frase a broad focus con soggetto
postverbale non è incompatibile con la presenza di Avere, solo ne-cl lo è. L’accettabilità
di (10) è la stessa di (14c) più avanti, con un verbo inergativo come ballare
degradato a inaccusativo. Date certe condizioni, questo slittamento, già osservato in
(4) e (6) con ne-cl e inergativo nella forma semplice, nelle frasi eventive è compatibile
anche con l’espressione di Avere, non essendoci il conflitto, descritto più sopra
(§2.1), con la specializzazione di ne per il cognate object. Avere diventerebbe un
ausiliare perfettivo neutralizzato rispetto a Essere.
A fini espositivi, vorrei però modificare (8) come in (12), che, col mancato accordo,
fa emergere l’illegalità di entrambi gli esempi (8) e (12a), in contrasto con
(13a):
(12) (Di quelle persone,)
a *ne ha telefonato una al giornale
b ne è venuta una al giornale
= “c’è stata una di quelle persone che ha telefonato/è venuta al giornale”
(13) (Di quelle persone,)
a ?ne telefonerà una al giornale
b ne verrà una al giornale
= “ci sarà una di quelle persone che telefonerà/verrà al giornale”
In (13a), l’estrazione avverrebbe dalla stessa posizione da cui viene estratto
l’oggetto dell’inaccusativo nelle frasi (b), e questo spiegherebbe la sua accettabilità
(sia nel quadro standard che in quello di Belletti). Anche intonativamente, (13a)
non crea problemi. Il focus è l’intero predicato (“telefonerà una al giornale”) e la
presupposizione è la semplice presupposizione di esistenza associata al tema implicito
“quelle persone”. L’interpretazione è: “ci sarà una persona che telefonerà al
120

Caratteri della cliticizzazione di ne
giornale”, mentre non è ammessa quella consueta col solo quantificatore in focus:
“sarà una la persona che telefonerà al giornale”, ottenibile solo con “al giornale”
marginalizzato 7 .
Telefonare, in (13), si comporta come un verbo inaccusativo di directed motion,
col senso di “arrivare telefonicamente” (v. 3.2). L’analoga interpretazione di (12)
“c’è stata una persona che ha telefonato al giornale”, seppur accessibile, è illegale.
Gli ess. (14)-(17) mostrano che i verbi inergativi non permettono indistintamente
frasi eventive come (10)-(11):
(14) a *Ha cavalcato un ragazzo fino alla cima
a’ Un ragazzo ha cavalcato fino alla cima
b *Hanno studiato alcuni ragazzi in biblioteca
b’ Alcuni ragazzi hanno studiato in biblioteca
c Hanno ballato due spagnoli sulla spiaggia
c’ Due spagnoli hanno ballato sulla spiaggia
(15) a *Cavalcherà un ragazzo fino alla cima
a’ Un ragazzo cavalcherà fino alla cima
b *Studieranno alcuni ragazzi in biblioteca
b’ Alcuni ragazzi studieranno in biblioteca
c Balleranno due spagnoli sulla spiaggia
c’ Due spagnoli balleranno sulla spiaggia
(16) a *Ne ha cavalcato uno fino alla cima (di ragazzi)
b *Ne hanno studiato alcuni in biblioteca (di ragazzi)
c ?*Ne hanno ballato tante sulla spiaggia (di ragazze)
(17) a *Ne cavalcherà uno fino alla cima (di ragazzi)
b *Ne studieranno alcuni in biblioteca (di ragazzi)
c (?)Ne balleranno tante sulla spiaggia (di ragazze)
= ”ci saranno tante ragazze che balleranno sulla spiaggia”
Solo ballare tra gli inergativi in (14)-(17), frasi (c), permette la forma con soggetto
postverbale a broad focus (14c) e (15c), e, coerentemente con questo dato, la
ne-cl nella forma non perifrastica (17c) vs. (16c).
7 La parafrasi di (13) è una frase esistenziale (13b), che riflette l’articolazione P/A della frase,
vale a dire la sua struttura a broad focus con tema presupposto ripreso da ne. Come dimostra
la stessa discussione di Belletti (2002), non si tratta dei casi più comuni di articolazione
P/A delle frasi a ne-cl.
121

Lidia Lonzi
Vi è dunque un complesso coerente di restrizioni lessico-semantiche, da indagare
in altra sede, che limita la costruzione illustrata in (11), e, in seconda battuta,
l’estrazione di ne illustrata in (13).
In (11) come in (13) vale il template inaccusativo, nel senso che la posizione
oggetto viene occupata dall’argomento, spogliato di agentività, di un verbo ergativo
che assegna il tipico θ-ruolo multiplo agente/tema (v. Chomsky, 1981: 139). Tale
argomento presenta pertanto, correttamente, il semplice θ-ruolo tema.
2.2.2 La posizione del soggetto postcomplemento
Nel caso di riordino dei complementi, Belletti (2002) prevede la marginalità di
ne-cl anche per frasi con verbi inaccusativi (v. ancora Belletti, 2002), dato che il
riordino dei complementi comporta focalizzazione del soggetto e di conseguenza
l’estrazione di ne avverrebbe da un DP in Spec, FocP. Bisogna però osservare che
la marginalità predetta è molto meno marcata, e quasi scompare, come in (18a),
rispetto ai casi in cui il PP pertinente non è sottocategorizzato, come in (18b):
(18) a Ne arriverà al giornale (ancora) uno, di articoli, ormai
= “sarà (ancora) uno l’articolo che arriverà al giornale”
b *Ne svennero allo spettacolo (ancora) due, di studenti, ricordo
Anche l’accettabilità delle frasi a broad focus con “riordino dei complementi” si
divide secondo lo stesso pattern di marginalità marcata se il PP non è sottocategorizzato
– e ciò dimostra che qui l’ostacolo all’estrazione è a monte: (18b) è marginale
perché è marginale (19b):
(19) a Arriverà al giornale (ancora) un articolo
b ?*Svennero allo spettacolo (ancora) due studenti
Il locativo, in (19b), non è propriamente un complemento. Nel quadro di Belletti
(2002), l’NP quantificato potrebbe sì occupare la posizione in FocP ma non sarebbe
possibile il movimento in TopP di quello che non è un genuino remnant VP,
movimento che genererebbe l’ordine in (19b).
Una suggestiva spiegazione del contrasto in (18a) vs. (18b), è che in (18a) l’estrazione
di ne da (19a) sia in realtà ammessa perché il riordino dei complementi
che si osserva in (19a) è ottenibile con una semplice derivazione incrementale
(Phillips, 2003) e un predicibile effetto di focalizzazione del DP in posizione finale
in una struttura intonativa non marcata (Nespor & Guasti, 2002). La stessa derivazione
non è possibile per (19b), dove, come si è visto, il locativo non è propriamente
un complemento. 8
8 In proposito, si tenga conto che i PP locativi, diversamente da altri dipendenti del predicato
che sono nettamente distinguibili in argomenti obbligatori e non, non hanno un comporta-
122

2.3 Posizione del quantificatore nelle frasi con ne-cl
Caratteri della cliticizzazione di ne
Levin & Rappaport Hovav (1995) suggeriscono che una frase con ne-cl come
(20) potrebbe avere la struttura (21a), col DP in posizione di aggiunto. Qui, ovviamente,
occorre pensare che la traccia di ne soddisfi il requisito del c-comando per
un qualche principio di non esclusione (Chomsky, 1986). La struttura pertinente
più generalmente accettata è quella in (21b), dove l’NP quantificato, con la traccia
di ne 9 , non si è mosso dalla posizione di base.
(20) pro ne fiorirono (solo) due, di gerani rossi, l’anno scorso
(21) a IP (pro I(nej (fiorironoy))VP(VP(ty ti))) DPi(solo due NP(tj))
b IP (pro(I(nej (fiorironoy))VP(ty DPi(solo due NP(tj)))
In (20), il quantificatore rappresenta il nuclear focus. Tutto il resto, tra cui il PP
partitivo, fa parte della presupposizione. In proposito, occorre precisare che l’NP
coreferente col clitico costituisce sempre il tema – implicito o esplicito – in questo
tipo di frasi.
A riprova, se si adotta il criterio usato da Levin & Rappaport Hovav (1995) per
la locative inversion, il quantificatore può trovarsi a destra del complemento strettamente
sottocategorizzato, come nel caso di riordino dei complementi in (18a). Si
vedano gli ess. (22a)-(24a) più sotto.
(22) a In realtà, ne è fuggito dal parco (solo) uno, di grizzley
b In realtà, ne è fuggito (solo) uno, dal parco, di grizzley
(23) a ?In realtà, ne sono scampati al ciclone (solo) tre, di alberi
b In realtà, ne sono scampati (solo) tre, al ciclone, di alberi
(24) a *In realtà, ne è svenuto all’Omnimax (solo) uno, di spettatori
b In realtà, ne è svenuto (solo) uno, all’Omnimax, di spettatori
Poiché l’ordine dei costituenti e l’articolazione P-A sono strettamente connessi
con la struttura intonativa, gli esempi (a), con l’ordine pertinente V PP Q, in cui il
quantificatore rappresenta il nuclear focus, sono tutti da leggere con intonazione
non marcata, quindi con accento prominente sul costituente più a destra prima della
pausa, diversamente dagli esempi (b), in cui il PP pertinente è marginalizzato. La
presupposizione, in (22), è che sia fuggito un numero X di grizzley dal parco, e
mento altrettanto consistente. Per esempio, anche se facoltativi, danno risultati vagamente
strani se stranded in forme di VP Preposing parziale, comportandosi in questo come veri argomenti.
9 Per lo statuto di ne come NP v. Cinque 1990; Cardinaletti e Giusti 1992.
123

Lidia Lonzi
pertanto (22) può rispondere, per es., alla domanda: È fuggito qualche grizzley? è
preferibile non includere il PP locativo nel contesto di domanda, perché in tal caso
non entrerebbe in modo naturale nella risposta con la forma in (a), ma solo con la
forma in (b). La messa a punto della questione è irrilevante ai nostri fini. Le stesse
considerazioni valgono per (23). In (23a) la frase presenta una struttura assertiva
compatibile con una domanda sullo stato degli alberi e al ciclone ha lo stesso status
informativo di dal parco più sopra. In (22) e (23), dove il PP (dal parco e al ciclone)
è sottocategorizzato, si hanno i risultati desiderati.
Anche in (24) le conoscenze condivise permetterebbero di includere all’Omnimax
nella presupposizione dell’enunciato, ma il “riordino dei complementi” non è
possibile. In (24a), il quantificatore non può seguire il PP locativo non sottocategorizzato
senza che si rompa la linea intonativa con il locativo come inciso (In realtà
ne è svenuto, all’Omnimax, solo uno, di spettatori), secondo una sorta di scrambling
(Belletti e Shlonsky, 1995) (trascuro il fatto che le frasi (b), in un contesto diverso,
potrebbero tutte esser prive di pausa prima del PP pertinente in funzione di
focus. V. più avanti (22c)-(24c)). Conforme all’analisi di Levin & Rappaport Hovav,
sarebbe comunque il valore di focus del sintagma quantificato a permettere
l’ordine dei complementi in (22a) e (23a), spiegabile con un aggiornamento
dell’aggiunzione a VP nei termini indicati sopra, § 2.2.
3. I verbi coinvolti nella ne-cl: struttura argomentale e selezione dell’ausiliare
3.1 Ne-cl come contesto diagnostico di inaccusatività?
Ne-cl è possibile quando l’NP quantificato ha ruolo tematico di tema in senso
standard, quindi interpretabile come un oggetto di cui si predica la collocazione, o
un cambiamento di collocazione, o anche un cambiamento di stato. Ne-cl è pertanto
in stretta relazione con la non-agentività del “soggetto”, quindi con una causalità
esterna ad esso. I verbi che permettono regolarmente ne-cl sono un sottoinsieme dei
verbi inaccusativi: verbi ad ausiliare Essere, che hanno una precisa rappresentazione
sintattica e, come hanno mostrato Levin & Rappaport Hovav (1995), rientrano
in precise classi semantiche. Dai verbi inaccusativi nel loro insieme vanno infatti
esclusi quei verbi che possono implicare volontà e/o selezionare un DP soggetto
[+umano], come andare o inginocchiarsi, detti di directed motion e, rispettivamente,
spatial configuration in Levin & Rappaport Hovav (1995) (v. anche Levin,
1989), oppure verbi come commuoversi o maturare (nel senso morale visto in 1).
Con questi verbi ne-cl è da stilisticamente trascurato (25a) a inaccettabile su base
124

Caratteri della cliticizzazione di ne
semantico-pragmatica (26a) (in (25b) la sostituzione di pochi con alcuni è intesa a
conservare l’intonazione non marcata):
(25) a %Se ne sono inginocchiati pochi, di sportivi, durante la cerimonia
b Alcuni sportivi si sono inginocchiati durante la cerimonia
c Si sono inginocchiati pochi sportivi, durante la cerimonia
(26) a *Se ne sono commossi tanti, di medici, per la vicenda
b Tanti medici si sono commossi per la vicenda
c Si sono commossi tanti medici, per la vicenda
D’altro lato, verbi che denotano un aspetto materiale del referente, come spargersi
o pullulare, in Levin & Rappaport Hovav (1995) detti di appearing, per ammettere
ne-cl richiedono che il soggetto [+umano] appartenga a una collettività indifferenziata,
non di singoli individui, v. l’inaccettabilità di (27a), che, come è facile
verificare, può migliorare con una sostituzione appropriata del nome soggetto:
(27) a *Nei dintorni, se ne erano sparsi molti, di congressisti
b Molti congressisti si erano sparsi nei dintorni
c ?Si erano sparsi molti congressisti, nei dintorni
Se ne-cl si applica con i verbi intransitivi a causa esterna – cioè gli inaccusativi
– obbedendo a restrizioni semantico-pragmatiche per quanto riguarda il soggetto
[+umano], occorre sottolineare che si applica anche, per quanto stabilito in 1.2,
punto (ii), con i verbi intransitivi a causa interna – cioè gli inergativi – purché, in
prima istanza, non agentivi: per es. i “verbi di emissione” nella terminologia di Levin
& Rappaport Hovav (1995) (di emissione visiva, uditiva, ecc., come luccicare o
trillare), oppure i verbi indicanti movimenti istintivi (come tossire, starnutire, trasalire,
sussultare). L’interessante di questi ultimi è che presentano una peculiare
associazione di: (a) causalità interna (normale caratteristica dei verbi inergativi), e:
(b) natura involontaria (normale caratteristica dei verbi inaccusativi). La selezione
di Avere riflette la caratteristica in (a), mentre l’ammissibilità di ne-cl riflette quella
in (b).
Si potrebbe dire che l’ammissibilità di ne-cl mette in evidenza un aspetto concettuale
del verbo (natura involontaria) che non è compatibile con (o, semplicemente,
pertinente per) la selezione dell’ausiliare Avere. Viceversa, la selezione di Avere
mette in evidenza un altro aspetto concettuale (causalità interna), che non è compatibile
con (o, semplicemente, pertinente per) l’applicazione di ne-cl. Ho già mostrato
che, con i verbi inergativi elencati sopra, se l’ausiliare è presente emerge netta-
125

Lidia Lonzi
mente l’incompatibilità con ne-cl 10 . In assenza dell’ausiliare, invece, il risultato è
accettabile anche in italiano standard. 11
Ora, secondo Levin & Rappaport Hovav (1995) la prova pertinente di inaccusatività
è data dalla selezione dell’ausiliare Essere, non dalla compatibilità con ne-cl.
Alla luce delle condizioni semantico-pragmatiche o di ordine stilistico esposte più
sopra, che limitano il campo di applicazione di tale procedura, tale proposta risulta
appropriata. Ancor più alla luce del fenomeno appena descritto, relativo ai verbi
inergativi che sono a cavallo fra le due classi di verbi e che viceversa lo estendono,
permettendo ne-cl in modo parziale.
Qui, il meccanismo implicato rifletterebbe l’affinità concettuale tra i verbi inergativi
denotanti reazioni involontarie e i verbi tipicamente inaccusativi. I verbi in
questione, infatti – dato che ricevono Avere, legato eminentemente alla causalità
interna del partecipante principale – non potrebbero essere identificati come inaccusativi,
e trattati di conseguenza, se non per il prevalere del tratto semantico che
favorisce ne-cl: la non agentività.
E proprio la scissione del loro comportamento rispetto a ne-cl (secondo la presenza
o assenza di Avere) può essere vista come una conferma del fatto che, sintatticamente,
i verbi monoargomentali si dividono rigidamente in due classi: gli inaccusativi,
con soggetto originato in posizione oggetto, e gli inergativi con soggetto
più esterno 12 . Certamente evidenzia la pertinenza, nell’italiano standard, dei due
10 Non tutti concordano su questo punto, v. per es. Saccon (1993). Oltre a presentare i dati di
un dialetto veneto – il dialetto di Conegliano – che con gli inergativi accetta ne-cl anche in
presenza dell’ausiliare, Saccon arriva a conclusioni analoghe per l’italiano (cf. l’uso recente
di attenuare i giudizi di agrammaticalità per ne-cl con gli inergativi, indipendentemente dalla
presenza o meno dell’ausiliare: Brunetti, 2003, e riferimenti).
11 Per quanto abbiamo visto, i verbi di emissione e quelli indicanti processi istintivi, limitatamente
alla forma non perifrastica, potrebbero forse essere trattati alla pari dei verbi inaccusativi
per una sorta di Default Linking Rule generalizzata (Levin & Rappaport Hovav
(1995); v. anche Perlmutter, 1978, che li classifica addirittura come tali).
(i) Default Linking Rule (Levin & Rappaport Hovav (1995):154)
An argument of a verb that does not fall under the scope of any of the other linking rules
is its direct internal argument.
Per Linking Rule, Levin & Rappaport Hovav (1995) intendono un tipo di regola che identifica
quei componenti di significato del verbo che danno luogo a una classificazione inaccusativa
o inergativa.
12 Si potrebbe proporre una doppia struttura argomentale associata a certi verbi inaccusativi,
dipendente ovviamente dall’agentività o non-agentività del soggetto (v. Perlmutter, 1978) e
spiegare così il loro duplice comportamento rispetto a ne-cl. Ma tale proposta solleva problemi
complessi e non la seguirò.
126

Caratteri della cliticizzazione di ne
ausiliari per la posizione del soggetto nella struttura sintattica coinvolta (Burzio,
1986; Kayne, 1993; tra gli altri).
Per concludere, ne-cl, quando risulta applicabile, è sì un inequivocabile contesto
di inaccusatività, cioè di struttura con soggetto in posizione oggetto, ma non è un
contesto diagnostico valido per determinare l’appartenenza di un verbo a una delle
due classi definite più stabilmente in base alla selezione dell’ausiliare (per non parlare
del contesto cruciale a livello interlinguistico, cioè dell’accordo vs non accordo
di soggetto e participio).
Qui, mi sono limitata a osservare un intreccio di fattori. L’ausiliare Essere denota
che il soggetto è generato in posizione oggetto, e si può assumere, per semplicità,
che questo sia vero anche per alcuni verbi che, con soggetti col tratto
[+umano], non favoriscono ne-cl (né, del resto, la costruzione V S in frasi a broad
focus). Di contro, ciò che è semanticamente pertinente per ne-cl include anche alcuni
verbi indicanti processi involontari (starnutire, tossire) o processi “che colpiscono
i sensi” nonostante che selezionino l’ausiliare Avere. Assumo qui che selezionino
Avere perché indicano processi a causalità interna. Nelle forme perfettive
la costruzione con ne-cl è a sua volta esclusa per l’interpretazione indesiderata di
ne che ne deriva come clitico del cognate object (che questo sia lessicalizzato e
ammesso, o che non lo sia). Pertanto, il fatto che non si abbia ne-cl:
i) né con verbi ad ausiliare Essere nei casi in cui il soggetto è dotato dei tratti
inerenti al tratto [+umano];
ii) né in presenza dell’ausiliare Avere con verbi che pur ammettono ne-cl nella
loro forma semplice (perché a soggetto [-agentivo]),
dipende da ragioni diverse tra loro. Nel caso (i) si tratta di un’incompatibilità
semantica – o stilistica – tra il tratto [+umano] e la costruzione partitiva in sé. Nel
caso (ii) si tratta di un’incompatibilità sintattica, o configurazionale, indotta dalla
struttura dell’ausiliare. Infatti l’incompatibilità non è con i verbi “ad ausiliare Avere”
ma con l’effettiva espressione di questo ausiliare, e con la struttura transitiva
che esso comporta e che si scontra con quella richiesta dal clitico.
Se ne-cl non ha valore diagnostico di inaccusatività è per un complesso di restrizioni
specifiche, presumibilmente estranee alla struttura sintattica coinvolta nell’estrazione
(posizione A dell’NP quantificato, nei termini di Belletti 2002). Ne-cl
richiede proprietà sintattiche cruciali dell’inaccusatività, ma ha una semantica che
ne restringe l’applicazione anche all’interno dei verbi inaccusativi.
3.2 Selezione dell’ausiliare (Kayne, 1993)
Secondo Kayne (1993), i fenomeni di accordo e la selezione dell’ausiliare dipendono
dalla sottostruttura participiale, a sua volta dipendente da proprietà strutturali
sia della lingua interessata – che permette o no di attivare determinate proie-
127

Lidia Lonzi
zioni participiali – sia, in particolare, del verbo, che contiene o no determinate posizioni
argomentali. Si vedano frasi come (28) e (29), con litigare, un regolare verbo
inergativo a causalità interna, quindi agentivo, degradato da ne-cl a livello dei
verbi denotanti un’azione involontaria:
(28) Ne litigano troppi t , alle feste (, di ragazzi)
(29) *Ne hanno litigato troppi t , alla festa (, di ragazzi)
(30) Hanno litigato troppi ragazzi, alla festa
In (28) il verbo si comporta come un inaccusativo che assegna ruolo-theta tema
al suo argomento la cui traccia è legata da ne (v. più avanti). L’analisi di Kayne ci
permette di spiegare l’inaccettabilità di (29). In (30), che è la frase di partenza, la
struttura participiale è il DP più grande in (31):
(31) Essere D/P° DP(AGRs T AGRo VP (V litigato DPtroppi ragazzi)))
Se fosse generato in posizione oggetto, il DP troppi ragazzi passerebbe attraverso
AGRo inducendo accordo del participio (ne hanno litigati troppi t, alla festa),
contrariamente a quanto si osserva. Il mancato accordo è dovuto a skipping di A-
GRo da parte del DP, fenomeno a cui consegue la selezione dell’ausiliare Avere,
come effetto dell’incorporazione di D/P° in Essere (richiesta da specifiche proprietà
dell’AGRs dell’italiano, v. Kayne, 1993:19). Questo fenomeno di skipping trova
immediata spiegazione se pensiamo che il DP pertinente sia generato come soggetto.
Per ipotesi, infatti, Avere è associato a verbi con una posizione oggetto occupata
da una categoria nulla che rappresenta l’oggetto accusativo (e che preclude il legamento
desiderato dell’oggetto nominativo da parte di ne). Anche il mancato accordo
del participio in (30) trova così un’immediata spiegazione.
Se il verbo fosse inaccusativo, con il DP generato in posizione oggetto (come si
potrebbe dedurre da (28)), in (29), oltre all’ausiliare Essere dovremmo avere accordo
con l’oggetto, come nell’inaccettabile (32):
(32) **Ne sono litigati troppi t, alla festa (, di ragazzi)
Mentre (32) è inaccettabile perché il verbo litigare è inergativo e il DP soggetto
non può quindi passare per AGRo, (28) è accettabile perché lo stesso verbo nella
forma semplice può comportarsi come un inaccusativo, con un soggetto con ruolo
tematico di tema in posizione oggetto.
Va da sé che il ne potrebbe legare propriamente la posizione oggetto in una costruzione
a cognate object esplicito del verbo inergativo, ma tale costruzione è inedita
con un verbo come litigare. V. (33):
(33) % I ragazzi ne hanno litigati troppi t (, di litigi furibondi), alla festa
128

Caratteri della cliticizzazione di ne
Ancora una volta, dobbiamo osservare che è il mancato accordo tra soggetto e
participio a “rivelare” la non grammaticalità di una frase come (29). Non inaccusativa,
a differenza di (28) – non è nemmeno inergativa/transitiva, a differenza di
(33). 13
L’analisi di Kayne rende operativa l’ipotesi dell’esistenza di un oggetto
dell’inergativo (Hale & Keyser, 1993), di cui forse non sono state sufficientemente
considerate le implicazioni negli studi più recenti sul “puzzle” inaccusativo (Alexiadou,
Anagnostopoulou, Everaert, 2003).
4. Ne-cl e locative inversion
4.1 Struttura informativa della frase con ne-cl
Dal punto di vista della struttura informativa, nella frase con ne-cl a intonazione
non marcata il predicato è presupposto, e il quantificatore è in focus (tipicamente,
nuclear focus). Questo perché l’NP quantificato costituisce il tema della frase,
indipendentemente dal fatto che sia o no espresso. Quando è espresso è regolarmente
dislocato.
La stessa sottoclasse di verbi inaccusativi che permettono regolarmente ne-cl
può entrare nelle frasi a broad focus con soggetto postverbale (o frasi “eventive”
con soggetto non agentivo in situ). Abbiamo visto che le frasi eventive hanno una
struttura informativa diversa da quella delle frasi con ne-cl, e mi sembra utile un
rapido confronto fra le due costruzioni proprio alla luce dei dati in (22a,b)-(24a,b)
in 2.3 sopra, che qui ripeto con l’aggiunta delle frasi (c), dove anche il PP complemento
è in focus:
(22) a In realtà, ne è fuggito dal parco (solo) uno, di grizzley
b In realtà, ne è fuggito (solo) uno, dal parco, di grizzley
c ?In realtà, ne è fuggito uno dal parco, di grizzley
13 Nei casi in cui c’è accordo apparente tra un DP maschile e il participio passato nella forma
non marcata richiesta da Avere, l’effetto di offesa si riduce/scompare, e la frase è marginalmente
accettabile. V. (ii) vs.(i):
(i) ?*Non ne ha litigato una, alla festa (di ragazze)
(ii) ?(?)Non ne ha litigato uno, alla festa (di ragazzi)
Si direbbe che l’analisi delineata da Kayne (1993) sia confermata dalla maggiore accettabilità
di frasi come (ii), quasi entrassero ingannevolmente in uno stampo a cognate object. E,
naturalmente, anche dal fenomeno osservato in dialetti che selezionano unicamente
l’ausiliare Avere, come il siciliano, dove è richiesto l’accordo del participio con il soggetto
dei verbi inaccusativi. V. Bentley, 2005.
129

Lidia Lonzi
(23) a ?In realtà, ne sono scampati al ciclone (solo) tre, di alberi
b In realtà, ne sono scampati (solo) tre, al ciclone, di alberi
c ?In realtà, ne sono scampati tre al ciclone, di alberi
(24) a *In realtà, ne è svenuto all’Omnimax (solo) uno, di spettatori
b In realtà, ne è svenuto (solo) uno, all’Omnimax, di spettatori
c In realtà, ne è svenuto uno all’Omnimax, di spettatori
Il quantificatore può figurare solo a sinistra di un PP non sottocategorizzato
(24b,c) vs. (24a), mentre, oltre che a sinistra (22b,c)-(23b,c), può figurare anche a
destra (22a) e (23a), quando il PP è sottocategorizzato. Ora, nella costruzione a BF,
con l’adeguata intonazione richiesta, rigorosamente non marcata, il soggetto dovrebbe
figurare sempre a sinistra del PP in quanto non ci sarebbero le condizioni di
struttura informativa per il riordino dei complementi. E questo sembrerebbe confermato
(ovviamente, l’esito negativo di (24e) è ridondante, dato quello di (24a)):
(22) d (Oh no!) è fuggito un grizzley dal parco
e ??(Oh no!) è fuggito dal parco un grizzley
(23) d (Guardate:) sono scampati tre alberi al ciclone
e ??(Guardate:) sono scampati al ciclone tre alberi
(24) d (Incredibile!) è svenuto uno spettatore all’Omnimax
e *(Incredibile!) è svenuto all’Omnimax uno spettatore
Nelle frasi (22d,e)-(24d,e), con interpretazione a broad focus e intonazione non
marcata, i giudizi si riferiscono esclusivamente all’appropriatezza pragmatica del
riordino dei complementi nelle frasi (e) rispetto a tale interpretazione. In effetti,
nell’analisi di Belletti (2002), che interpreta sistematicamente il riordino dei complementi
V PP O come focalizzazione del soggetto in Spec,FocP, (22e)-(23e) avrebbero
un’interpretazione con remnant VP in TopP. Le stesse forme sono infatti
accettabili, pur con la ripetizione innaturale del PP, in risposta a qualcosa come:
Che animali sono fuggiti dal parco?, Cos’è scampato al ciclone?
L’opposizione pertinente è tra (22a)-(23a), più sopra, dove il soggetto (in focus),
può figurare a destra del PP, e (22e)-(23e), a broad focus, dove la stessa posizione è
pragmaticamente inappropriata. L’idea è che nelle frasi a broad focus l’ordine “derivato”
dei complementi (PP-DP) – in cui si potrebbe pur argomentare, contra Belletti,
che la gerarchia strutturale resti immutata (Larson, 1990; Phillips, 2003) – sia
ingiustificato informativamente. Il dato costituito da questa sottile opposizione rinforza
l’analisi di ne-cl come costruzione essenzialmente a narrow focus, qualunque
sia il trattamento sintattico prescelto per quest’ultimo (v. 2.3).
130

Caratteri della cliticizzazione di ne
4.2 Aspetti discorsivo-funzionali della frase con locative inversion
Un’ipotesi avanzata da Levin & Rappaport Hovav (1995) è che nelle frasi con
ne-cl compaiano verbi analoghi a quelli che permettono la locative inversion dell’inglese,
e che tra queste due costruzioni delle due lingue vi siano in comune fattori
di tipo comunicativo. Poiché Levin & Rappaport Hovav (1995: 228-232) argomentano
che anche nella locative inversion il focus è costituito dal soggetto, almeno
un aspetto di quest’ipotesi sembra preliminarmente confermato dai dati in 4.1,
mentre resta interamente da affrontare l’aspetto relativo alle proprietà comunicativo-funzionali
dei verbi in comune tra queste costruzioni.
Levin & Rappaport Hovav (1995) osservano, in particolare, che il verbo dev’essere
light. In altre parole, nella locative inversion il verbo ha una funzione introduttiva
rispetto al focus, pur senza intaccare la funzione di tema discorsivo del PP
locativo (per un’analisi recente, v. Partee & Borschev, 2000, e riferimenti).
Il fatto che, come mostra l’abbondante corpus di Levin & Rappaport Hovav
(1995) relativo alla locative inversion, si tratti sempre di predicati ragionevolmente
prevedibili, coinvolge una generale conoscenza del mondo e quindi, almeno a prima
vista, non sembrerebbe dover essere giustificato in termini di vera e propria presupposizione
di frase.
Qui, forse, per una convenzione meramente letteraria, questa qualità “ragionevolmente
prevedibile” del predicato si giustifica all’interno di uno scambio di domande/risposte
che il narratore deve rispettare (dagli esempi delle autrici dobbiamo
infatti dedurre che si tratti di un artificio di tipo eminentemente narrativo). La locative
inversion è strutturata come una risposta a una potenziale domanda del lettore.
Del resto, la convenzionalità di questa costruzione è rappresentata dallo stesso
tema discorsivo, il locativo, che qui ha caratteristiche informative molto diverse da
quelle che si ritrovano per es. nel tema discorsivo vero e proprio di una conversazione
quotidiana, di uno scambio di battute, o di un articolo di cronaca. In questo
caso il tema locativo, come è stato più volte osservato per il tema temporale, non
costituisce la base conoscitiva comune di lettore e narratore, bensì, sotto la finzione
di questa conoscenza comune, non fa che introdurre nuove conoscenze (per alcuni
esempi in proposito, v. Lonzi, 1989). La domanda abbastanza artificiosa a cui risponderebbe
per es. (36), adattata da Levin & Rappaport Hovav (1995: p. 221, 8b),
con un verbo di appearance sostituito a un verbo di spatial configuration, sarebbe,
per convenzione: – Che cosa spiccava sul coperchio del piano?:
(36) Sul lucido coperchio nero del piano spiccavano tre fotografie in cornici
d’argento
On the black lacquer top of the piano PERCHED three brass framed
pictures…
131

Lidia Lonzi
In questo senso preciso mi sembra da condividere la proposta di trattare il soggetto
della locative inversion come un narrow focus analogo al nome quantificato
nella ne-cl, suscettibile del trattamento pertinente. 14 Quanto ai verbi in comune tra
le due costruzioni, Levin & Rappaport Hovav (1995) osservano che i verbi usati
nella locative inversion sono tipici verbi inaccusativi: di apparenza, di esistenza, e
di directed motion, e verbi al passivo. Ne sarebbero esclusi, oltre ai verbi transitivi,
i verbi inergativi (v.*On the corner smoked a woman, da Bresnan (1994), che però
propone questo esempio per l’incompatibilità che dimostrerebbe tra locative inversion
e locativo non argomentale: smoke, infatti, non sottocategorizza alcun locativo).
15 In realtà, sembrano possibili anche certi verbi inergativi – come per es. dance
– (v. Around the fire danced the women, Bresnan (1994), citato in Levin & Rappaport
Hovav (1995): 232), purché abbiano una componente rappresentativa, quindi
concreta e materiale, come abbiamo visto per i casi di ne-cl con verbo non perifrastico,
per es. in (14c)-(17c). Suggerirei che nelle frasi a locative inversion questo
specifico requisito figurativo derivi dalla convenzionalità di tipo retorico-letterario
della costruzione, cioè dalla sua funzione propriamente narrativa, mentre in quelle
con ne-cl deriverebbe dalla semantica del partitivo.
Infine, per trovare rispondenza nell’osservazione di Levin & Rappaport Hovav
(1995) secondo cui i verbi inergativi nelle frasi con ne-cl descriverebbero
un’attività caratteristica dell’entità di cui sono predicati, quindi prevedibile, come
avviene nella locative inversion, mi sembra utile tornare all’idea che nel concetto
14 La proposta di Levin & Rappaport Hovav (1995), secondo cui il soggetto in quanto focus,
sarebbe aggiunto a VP, a parte gli aggiornamenti richiesti (v. sopra 2.3), si accorda con quella
fatta in letteratura per le frasi con there “presentativo” – dove il soggetto sarebbe obbligatoriamente
in focus (Schütze, 1999, e riferimenti).
15 È tuttavia da domandarsi se nella locative inversion il locativo rappresenti necessariamente
un argomento del verbo, contrariamente a quanto sostenuto in Bresnan (1994, riportato in
Levin & Rappaport Hovav (1995): 228). Da un lato, trattandosi di un locativo circostanziale,
o con funzione di cornice, non ha alcuna relazione con la posizione di argomento o di aggiunto
a VP (Lonzi, 1991: 381 sg.). In (i) – corrispettivo italiano di una costruzione a locative
inversion – vs. (ii) o (iii) – il locativo non può essere ripreso da un clitico:
(i) Sulla mensola del caminetto, (*ci) spiccava il dagherrotipo di un’antenata.
(ii) Sulla mensola del caminetto, (ci) mise tutto quello che non era riuscita a mettere sul
tavolo
(iii) Sulla mensola del caminetto, non (ci) aveva mai mangiato
D’altro lato, come è suggerito in Levin & Rappaport Hovav (1995), lo stretto rapporto che
sembra sussistere tra locativo e verbo è piuttosto di natura comunicativa: dato un certo locativo,
il verbo deve essere prevedibile per poter entrare nella parte presupposta della frase.
132

Caratteri della cliticizzazione di ne
di quantità numerabile e materiale propria di ne-cl non vi è spazio per comportamenti
individuali. 16
5. Osservazioni conclusive
Concludendo, il fatto che verbi ad ausiliare Avere ammettano ne-cl nella forma
non perifrastica deve significare che dispongono – forse come fenomeno addirittura
paragrammaticale – della stessa struttura argomentale degli inaccusativi, in alternanza
con la struttura argomentale degli inergativi associata obbligatoriamente
all’ausiliare Avere. La posizione riservata al cognate object diventerebbe disponibile
per l’oggetto nominativo. In proposito, ho fatto tesoro dell’analisi di Kayne
(1993), secondo cui la selezione dell’ausiliare tiene conto della sottostruttura participiale,
in ultima analisi della struttura argomentale del verbo pertinente.
In tal modo il verbo inergativo permette in subordine due strutture, come avviene,
regolarmente, per i verbi inergativi a doppio ausiliare (correre, volare, atterrare…).
Nella forma semplice si applica la struttura ergativa: il ne lega la traccia in
posizione oggetto dell’NP quantificato dotato delle caratteristiche lessicosemantiche
richieste; nella forma perifrastica si applica la struttura inergativa: la
stessa posizione è occupata da una categoria nulla che rappresenta l’oggetto accusativo,
quindi non si può avere il legamento desiderato del soggetto da parte di ne.
La spiegazione di Belletti della non estraibilità dalla posizione A’ non spiega la
differenza evidenziata tra presenza e assenza dell’ausiliare, salvo non si ammetta
un’interdipendenza fra l’ausiliare Avere e l’attivazione di tale posizione. Come
conseguenza occorrerebbe comunque pensare che nella forma (eventiva) semplice il
DP sia nella posizione dell’oggetto inaccusativo.
La questione del rapporto tra ne-cl e locative inversion sollevata da Levin &
Rappaport Hovav (1995), a cui è dedicata la sezione finale di questo articolo, richiederebbe
una rassegna dettagliata dei verbi coinvolti sul modello di quella compiuta
da Levin & Rappaport Hovav (1995) per l’inglese. Qui, spero di aver dato un
primo contributo al suo approfondimento, delineando certe caratteristiche semantico-lessicali
e di struttura informativa che sembrano comuni a entrambe.
16 È opportuno sottolineare che, se si analizzasse la locative inversion lungo le linee tracciate
in Bresnan 1994 (v. Levin & Rappaport Hovav (1995): 6.6.1), con la sua ipotesi che il
“soggetto” sia generato in posizione oggetto e il PP locativo sia il vero soggetto, la limitazione
nell’uso degli inergativi troverebbe immediata spiegazione.
133

Lidia Lonzi
Bibliografia
Alexiadou, A., E. Anagnostopoulou, M. Everaert, 2004. Studies on Unaccusativity: Explorations
of the Syntax-Lexicon Interface, Oxford University Press.
Belletti, A. 1988. The case of unaccusatives. Linguistic Inquiry, 19, 1-34.
Belletti, A. 2002. Inversion as Focalization. In A. Hulk and J.Y. Pollock (eds.) Inversion in
Romance and the Theory of Universal Grammar, Oxford University Press.
Belletti, A. & L. Rizzi. 1981. The Syntax of ne: Some Theoretical Implications, The Linguistic
Review, 1, 117-154.
Belletti, A. & U. Shlonsky. 1995. The Order of Verbal Complements: A Comparative Study.
Natural Language & Linguistic Theory, 13, 3, 489-526.
Benincà, P. e C. Poletto. 2001. Topic, Focus and V2: defining the CP sublayers. In A. Belletti
and L. Rizzi (eds.) Proceedings of the Workshop on the Cartography of Syntactic
Poitions and Semantic Types, Certosa di Pontignano, November 1999.
Bentley, D. 2004. Ne-cliticisation and split intransitivity. Journal of Linguistics, 40, 2, 219-
262.
Bentley, D. 2005. Sur la force d’une approche non-dérivationnelle de l’analyse linguistique:
quelques données de l’italo-roman. Ms. Salford University.
Bresnan, J. 1994. Locative Inversion and the Architecture of Universal Grammar. Language,
70: 72-131.
Brunetti, Lisa (2003). A Unification of Focus. Tesi di Dottorato. Università di Firenze.
Burzio, L. 1986. Italian Syntax. Dordrecht, Reidel.
Cardinaletti, A. 2002. Against Optional and null clitics. Right dislocation vs. marginalization.
Studia Linguistica 56, 29-57.
Cardinaletti, A. & G. Giusti. 1992. Partitive ne and the QP-hypothesis. A Case study. In
Elisabetta Fava (ed.) Proceedings of the XVII Meeting Of Generative Grammar.
Torino, Rosenberg & Sellier.
Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht. Foris.
Cinque, G. 1990. Lo statuto categoriale del ne partitivo. In Miscellanea di studi in memoria
di Paolo Zolli. Università di Venezia.
D’Alessandro, R. & Ian Roberts (2007), Movement and Agreement in Italian Past Participles
and defective Phases, XXXIII Incontro di Grammatica Generativa, Università di
Bologna.
Hale, K. & S. J. Keyser. 1993. On Argument Structure and the Lexical Expression of Syntactic
Relations. In K. Hale & S. J. Keyser (eds) The View from Building 20. Essays
in Linguistics in Honor of Sylvain Bromberger. Cambridge, Mass. MIT Press.
Kayne, R.S. 1993. Toward a Modular Theory of Auxiliary Selection, Studia Linguistica, 47,
3-31.
Larson, R.K. 1990. Double Objects Revisited: Reply to Jackendoff. Linguistic Inquiry, 21,
589-632.
134

Caratteri della cliticizzazione di ne
Levin, B. 1989. English Verb Classes and Alternations. Chicago. University of Chicago
Press.
Levin, B. & M. Rappaport Hovav. 1995. Unaccusativity. At the Syntax-Lexical Semantics
Interface. Cambridge, Mass. MIT Press.
Longobardi, G. 2000. “Postverbal” Subjects and the Mapping Hypothesis. Linguistic Inquiry,
31, 691-702.
Lonzi, L. 1986. Pertinenza della struttura Tema-Rema per l’analisi sintattica. In: H. Stammerjohann,
Theme-Rheme in Italian. Tübingen. Narr. Ripubblicato in L. Lonzi,
1998, Avverbi e altre costruzioni a controllo. Bologna. il Mulino.
Lonzi, L. 1989. On certain peculiarities of narrative cohesion. In: M.-E. Conte, J.S. Petöfi,
E. Sözer. Text and Discourse Connectedness. Amsterdam. Benjamins.
Lonzi, L. 1991. Il sintagma avverbiale. In: L. Renzi e G. Salvi, Grande Grammatica Italiana
di Consultazione. Bologna. il Mulino.
Nespor, M. & M.T. Guasti. 2002. Focus-stress Alignment and its Consequences for Acquisition,
Lingue e linguaggio, 1, 79-106.
Partee, B. & V. Borschev. 2000. The Russian Genitive of Negation in Existential Sentences
and the Notion of Existential Sentence. IATL, Tel Aviv, June 17-18.
Perlmutter, D. 1978. Impersonal Passives and the Unaccusative Hypothesis. Proceedings of
the Fourth Annual Meeting of the Berkeley Linguistic Society, Los Angeles, University
of California, pp. 157-189.
Phillips, C. 2003. Linear Order and Constituency. Linguistic Inquiry. 34, 37-90.
Saccon, G. 1993. Post-verbal Subjects. A study based on Italian and its Dialects. Doctoral
Dissertation, Harvard University.
Schütze, C.T. 1999. English Expletive Constructions are not Infected. Linguistic Inquiry. 30,
467-484.
Zubizarreta, M.L. 1998. Prosody, Focus, and Word Order. Cambridge, Mass. MIT Press.
135

Rivista di Grammatica Generativa
34 – 2009
CONTRASTIVENESS AND NEW INFORMATION:
Introduction
A NEW VIEW ON FOCUS 1
Sandra Paoli
It is a fairly accepted view that word ordering phenomena are, to a certain extent,
determined by the informational import of constituents, loosely their contextually
known or new status. Since the introduction of information structure in the inventory
of projecting categories in transformational theories of syntax (cf. Horvarth, 1986;
Kiss, 1987; Brody, 1990), the label Focus has been applied to phrases performing
two discourse functions, roughly the introduction of new information and the
introduction of a contrast 2 . A number of linguists have argued, more or less
explicitly, in favour of a clear-cut distinction between the two types on the basis of
syntactic, phonetic, phonological and pragmatic factors (cf. Kiss, 1998; Zubizarreta,
1998; Nespor and Guasti, 2002; Donati and Nespor, 2003; Benincà and Poletto,
2004). Others (cf. Frascarelli, 2000; Brunetti, 2004; Lonzi, 2006; Stoyanova, 2008)
have claimed that the two are not syntactically distinct. As far as contrastiveness is
concerned, it has been analysed in binary [+/-] terms, much on the lines of newness
of information, reflecting the traditional dichotomy ‘focus-topic’, ‘new-old’.
Investigating cases of ‘hybrid’ focal elements that simultaneously express new
information and a contrast, this article addresses a neglected area within the
1 The implementation of the ideas presented here has greatly benefited from conversations
and discussions with Louise Mycock and from comments on an earlier version by Mary
Dalrymple. Special thanks go to Wallace Chafe, for taking the time to read this piece and to
highlight any points at which his ideas had been misunderstood, and for his willingness to
have the conversation in spite of the theoretical divide. A heart-felt ‘Grazie’ to Paola Benincà
and Nicola Munaro for their support. The full responsibility for any shortcomings rests solely
with the author.
2 The term focus has been used widely, in different and contradictory ways, to the extent that
often it is not clear what it is meant by it. Although it is vital to address these discrepancies
and bring some much needed clarity and transparency, it lies outside the immediate scope of
this article to undertake such a task.
137

Sandra Paoli
transformational tradition of grammar: the pragmatic and psychological import of
discourse salient phrases. It is clear that an interface phenomenon such as
information structure simply cannot be captured fully relying solely on its syntactic
properties. Given that contrastiveness is a relational property, its specification must
be expressed taking into account the status of the element with which the contrast is
established: by this we mean the level of consciousness at which it is held in the
mind of speaker and hearer 3 .
Drawing on the notion of ‘activeness’ of a referent introduced by Chafe (1987)
and elaborated by Lambrecht (1994), this article argues for the following:
i. there is more to contrastiveness than a simple [±] distinction;
ii. degrees of contrastiveness obtain from the treatment of a piece of discourse
‘in terms of cognitive processes dynamically unfolding through time’ (Chafe,
1987:48);
iii. newness of information and contrastiveness are not mutually exclusive
properties of phrases: we view contrastiveness as orthogonal to both new and old
information, and as such able to combine with both;
vi. whether the further distinctions between contrastive phrases that we advocate
for here are expressed at the syntactic level is subject to great cross-linguistic
variation.
The observation that the two types of focus, new information and contrastive, are
not mutually exclusive nor incompatible, is by no means new (cf. Bolinger, 1961:87;
and, more recently, Frascarelli, 2000:91): the novelty of our approach is in the way
this idea is implemented, with the inclusion of properties pertaining to nonlinguistic
4 dimensions in the make-up of constituents and their potential expression
at the syntactic level.
After a brief overview of the syntactic representation of Focus in section 1, data
from a North-eastern Italian variety are analysed and discussed in section 2. This
leads, in section 3, to the identification of some basic features which, combined,
provide a way of capturing the breakdown of the properties of the two types of
contrastive focus; section 4 investigates whether these features find expression at the
syntactic level, and addresses some remaining issues. Some of the prosodic
properties of the two types of contrastive focus are examined in section 5, and in
section 6 we present a few concluding remarks 5 .
3 Cf. section 3.2 for a specification of whose consciousness is relevant to the discussion.
4 Intended here in its lose meaning of ‘grammar-driven’.
5 Part of the data presented in this article was first discussed in Paoli (2010), which adduces to
a more complex system of Focus projections by referring to degrees of contrast expressed in
terms of explicitness and implicitness. The current contribution is an elaboration of the idea of
degrees of contrastiveness, and it offers a more in-depth and expanded analysis, allowing for
further-reaching theoretical considerations.
138

Contrastiveness and New Information: a new view on Focus
1. New Information and Contrastive: different types of Focus?
The notion of focus has been extensively discussed and defined from various
perspectives, pragmatic, semantic, phonological and syntactic. On a pragmatic level,
the information in focus is what is perceived by the speaker to be the most important
part of their utterance, and the one that provides the hearer with essential elements
of knowledge (cf. Dik, 1989; among others). The ‘focal’ status is assigned to at least
one syntactic constituent, be it a smaller part or the whole clause, and it generally
expresses new information. Semantically, focus is defined as the non-presupposed
(i.e. not known, not shared by both speaker and hearer) information in a sentence
(cf. Bolinger 1954:152; Halliday, 1967:204ff; Jackendoff, 1972:230, among some of
the earlier scholars), which has scope over a set of possible alternatives. At the
phonological level focus is generally fore-grounded by means of stress. Syntactically,
as a result of an increased interest in the syntactic encoding of information structure
(cf. Brody, 1990; Kiss, 1995, 1998; Puskás, 2000, among others) combined with the
formulation of an articulated CP (cf. Rizzi, 1997), focus has come to be interpreted
in the transformational tradition of grammar as a syntactic projection, Foc(us)
P(hrase), to which focalised phrases move and where they receive the relevant
interpretation.
Within the Cartographic approach, the further distinction between N(ew)
I(nformation) F(ocus) and C(ontrastive) F(ocus) also finds expression at the
syntactic level, with the introduction of two separate projections, NIF P(hrase) and
CF P(hrase) respectively, which are either assumed to be located in the same
syntactic space (cf. Benincà and Poletto, 2004) or not (cf. Rizzi, 1997; Belletti,
2004).
It is significant that NIF and CF are analysed as projecting distinct phrases: in a
system such as the one proposed by Benincà and Poletto (2004), which, crucially,
does not admit recursion of syntactically and semantically identical material, the two
are implicitly assumed to be distinct, and the existence of ‘hybrid’ types,
simultaneously expressing new information and a contrast, is excluded a priori.
Interestingly, the original reasons behind the syntactic distinction between CF
and NIF proposed by Kiss (1995, 1998) for Hungarian, have been shown not to hold
cross-linguistically (cf. Frascarelli, 2000). While Kiss distinguishes between the
operator-like Identificational Focus (which loosely corresponds to what we call CF
here 6 ) and Information Focus (our NIF) which does not involve movement,
Frascarelli (2000) points out that in Italian, for example, NIF, just like CF, is subject
to weak cross over as well as parasitic gaps effects. The two seem to be, therefore,
syntactically indistinguishable; yet, they are undeniably different. It therefore seems
6 An investigation of the relation between Kiss’s Identificational Focus and CF is, however,
not relevant to the current discussion.
139

Sandra Paoli
that such differences must find expression in dimensions other than syntax. What
remains to be established is the exact nature of the information related to CF and
NIF that is (or could potentially be) encoded at the syntactic level, and a way to
capture its variation across languages.
2. Focus in Triestino
One of the so-called Italian dialects, Triestino, a variety spoken in the northeastern
corner in the city of Triest, offers an interesting term of comparison with the
standard language. While in Italian (cf. 1) a phrase expressing the narrow focus of a
sentence needs to be placed post-verbally when the VP is repeated in the answer,
Triestino (cf. 2) admits its fronting to a sentence-initial position 7 :
1 a. Sp. A: Cosa comprano a Marina i tuoi genitori? Ita
‘What are your parents buying Marina?’
140
b. Sp. B: (Le comprano) UNA MOTO.
‘They are buying her a motorbike’
b’. Sp. B: *UNA MOTO le comprano.
2 a. Sp. A: Cossa i ghe ciol a Marina tua Ts
what scl to-her they-buy to Marina your
mama e tuo papà?
mum and your dad
‘What are your parents buying Marina?’
b. Sp. B: (I ghe ciol) UNA MOTO.
scl to-her they-buy a motorbike
‘They are buying her a motorbike’
b’. Sp. B: UNA MOTO i ghe ciol.
What is of immediate interest here is the nature and status of the fronted element
UNA MOTO ‘a motorbike’ in (2b’). Clearly, it is not an instance of CF as it is usually
conceived: firstly, it simply provides the value for the x sought in the preceding turn,
‘which x are your parents buying for Marina’; secondly, its pitch accent is audibly
different from the typical L+H* that marks contrastive elements, followed by a
noticeable rise and fall on the element negated. Yet, given the grammaticality of
7 The following abbreviations are used throughout the examples: scl (subject clitic, a weak
pronominal form found in Northern Italian and Tuscan dialects); Ita (Italian), Ts (Triestino),
Sard (Sardinian), Sic (Sicilian). If no specification is made, the languages in the examples
should be clear from the previous text. The narrow focus of the sentence is indicated in SMALL
CAPITALS and purely CF in BOLD CAPITALS. As a simplification, the Triestino fronting
cases are written in SMALL CAPITALS.

Contrastiveness and New Information: a new view on Focus
(2b), and assuming that direct objects are generated post-verbally, UNA MOTO in (2b’)
is visibly not in its canonical position.
The organisation of the elements in main, declarative clauses in Triestino and
Italian is not different: the unmarked position for a verbal complement is postverbal,
and the overt subject of transitive verbs occurs pre-verbally. In thetic
sentences (answers to the question ‘What is happening?’), both Italian (cf. 3a) and
Triestino (cf. 3b) require SVO word order, and disallow other variations (cf. 4):
3 a. Quel tizio dice sciocchezze. Ita
S V O
b. Quel mato (el) dizi monade. Ts
S V O
that guy scl he-says idiocies
‘That guy is talking rubbish’
4 a. *Sciocchezze dice quel tizio. Ita
O V S
b. *Monade el dizi quel mato. Ts
O V S
idiociesscl he-says that guy
The ungrammaticality of (4b) clearly suggests that the fronted element in (2b’) is
not in its canonical position; it is therefore plausible to imagine that it occupies a
pre-verbal position dedicated to discourse-prominent elements. The near ungrammaticality
of (5b) in Triestino, showing the ill effects of trying to co-index the fronted
element Marina with a post-verbal possessive adjective, sua / suo ‘her’, confirms
that the phrase is interpreted as an operator:
5 a. Sp. A: ?*Chii i vedi suai mama e suoi papà? Ts
who scl they-see her mum and her dad
‘Whoi do heri parents see?’
b. Sp. B: ?*[MARINA]i i vedi suai mama e suoi papà.
Marina scl they-see her mum and her dad
‘Heri parents see Marinai’
The evidence afforded by the licensing of parasitic gaps also points in the same
direction: the fronted phrase in Triestino displays operator-like properties,
suggesting that it occupies a scope position at LF.
6 a. Sp. A: Cossai la ga ciolto senza pagar [ei]?
what scl she-has taken without to-pay
‘Whati did she take without paying [ei]?’
141

Sandra Paoli
142
b. Sp. B: [I POMIDORI]i la ga ciolto senza pagar [ei].
the tomatoes scl she-has taken without to-pay
‘She has taken tomatoes without paying for them’
In view that the fronted element in (2b’)
i. does not contrast with any other element present in the overt discourse,
ii. expresses the narrow focus of the sentence,
iii. is subject to weak cross-over effects (cf. 5),
iv. and licenses parasitic gaps (cf. 6),
it would seem natural to conclude that the constituent is an instance of NIF.
Indeed, it is a well-attested fact that other varieties, of which Sicilian and Sardinian
are primary examples, license NIF in a sentence-initial position:
7 a. Sp. A: Chi cci ricisti? Sic
what to-them you-said
‘What did you say to them?’
b. Sp. B: A VIRITÀ cci rissi.
the truth to-them I-said
‘I told them the truth’ (adapted from Bentley, 2007: 53)
c. Sp. A: Ue l’ as postu? Sard
where it you-have put
‘Where have you put it?’
d. Sp. B: SUPRA SA MESA l’ appo postu.
over the table it I-have put
‘I have put it on top of the table’ (adapted from Jones, 1993: 18)
In both (7b) and (7d) the pre-verbal constituent is analysed as occupying a
position in the leftmost portion of the clause, NIFP: it represents the narrow focus of
the sentence and it provides Speaker A with the information they are seeking.
Analysing the cases of fronting in Triestino in an analogous way to the Sicilian
and Sardinian examples would, nevertheless, fail to capture and account for some
important facts.
3. An interpretation
The lack of contrast with another element previously introduced in the discourse
does not necessarily mean that the cases of fronting in Triestino are instances of
pure NIF, defined as a referent 8 newly introduced into the discourse. There are
subtle, yet significant differences between the responses in (2b) and (2b’). The
8 The term ‘referent’ here is used to indicate the element that corresponds to the specific
linguistic expression under examination, be it an individual or a state of affairs.

Contrastiveness and New Information: a new view on Focus
response with the canonical SVO order is the most neutral (communicatively
speaking) and natural one, in which UNA MOTO ‘a motorbike’ simply provides the
new information sought by Speaker A. By instead fronting the object in (2b’), UNA
MOTO i ghe ciol ‘A motorbike they are buying her’, Speaker B performs two
communicative tasks: they provide the new information and, based on their
knowledge of A’s expectations, they signal that such information will be
unexpected. This could be due to a variety of reasons: perhaps Speaker A knows that
Marina’s parents do not like motorbikes, or they usually do not buy such expensive
gifts as birthday presents, etc.
Cross-linguistic evidence shows that, indeed, factors such as surprise for an
unexpected event can be encoded morphologically by languages: Cinque (1999:84-
85, 201 note 21) mentions the Korean suffix -kwun-, the Menomini verbal suffix
-asah- and the sentential particles in the Tibeto-Burman language Akha. These are
all analysed by Cinque as expressions of Evaluative Mood, one of the highest
projections in his hierarchy, placed in the Mood field between the highest Speech
Act and the lower Evidential. The Turkish suffix -mIş also marks unexpectedness;
from a very preliminary investigation it appears that the suffix can express equally
speaker- and hearer-oriented surprise, something that has gone unnoticed so far (cf.
Slobin and Aksu, 1982; DeLancey, 1997 among others). Compare the following,
from Slobin and Aksu (1982:187):
8 a. Kemal gel-di.
Kemal come-past
‘Kemal came’
b. Kemal gel-mIş.
Kemal come-mIş
‘Kemal came (imagine!)’
(8a) simply communicates 9 the event that Kemal, some time prior to the
utterance time, came. In its mirative use, (8b) expresses the same event but using the
affix -mIş it adds that such an event was unexpected. Interestingly, the same
sentence could also be used as an answer to the question ‘Who came?’ when the
speaker knows that the information they are about to provide will be unexpected by
the hearer: the context offered was that Kemal had not been to visit for a long time 10 .
9 Sirin Tufan, p.c.
10 A recent article by Zimmermann (2008) independently makes a similar link. He makes a
case for analysing elusive phenomena such as contrastivity and emphasis in terms of hearer
expectation or discourse expectability.
143

Sandra Paoli
It is the notion of lack of explicit contrast, and more generally of contrastiveness
as a gradable concept, that we set to investigate in order to account for the Triestino
cases.
3.1 Contrastiveness as a scalar notion
That contrastiveness may not be a matter of a simple binarity is not a new idea.
Bolinger (1961:87) for example, sees CF and NIF as the end points of a continuum:
when for the value assigned to a variable x there is an unlimited number of
alternatives, the reading is not contrastive; the narrower the number of alternatives,
the more likely the contrastive interpretation 11 . This view of contrastiveness does not
assume a discrete step between [-contrastive] and [+contrastive]; moreover, it relies
on its semantic properties, more precisely on the scope of the variable over a set of
possible alternatives. Our position is different with respect to two crucial points:
firstly, we view contrastiveness in discrete terms, assuming that there is a clear
separation between a non-contrastive and a contrastive reading. What we suggest is
that within the contrastive interpretation there are varying degrees of strength of
contrastiveness. Secondly, in order to express these degrees of strength, we draw on
its pragmatic rather than semantic characteristics: the necessity of separating the
pragmatic from the semantic-quantificational level in the understanding of the
notion of contrastiveness has been convincingly argued for in Molnár (2002).
There seems to be solid cross-linguistic evidence to motivate further refinement
within the category CF: a comparison of Finnish and Hungarian data (cf. Molnár,
2002:152, Molnár and Järventausta, 2003), suggests the existence of two distinct
types of CF. Both involve movement, but while one (the Hungarian type) requires
both movement to the left-periphery and adjacency to the verb, the other (the
Finnish type) simply requires movement to the left-periphery. A similar distinction
is also witnessed in Basque (Etxepare, 1997; cited in Molnár, 2002:152), which
reserves the verb-adjacent position in the left periphery to cases of ‘emphatic focus’,
and allows non-verbal-adjacent movement to instances of ‘contrastive focus’ 12 . The
clear differences in word order and adjacency requirements of the Finnish,
Hungarian and Basque data make a strong case for identifying further types of CF at
the syntactic level. This may not be, however, a property shared cross-linguistically,
and other languages may not express syntactically the difference between different
types of CF.
11 A number of other linguists (cf. Halliday, 1967; Chafe, 1976; Rooth, 1992) claim that the
contrastive reading only obtains when the variable operates on a closed set of entities as
opposed to an open one. For arguments against such a distinction, see, for example, Brunetti
(2004).
12 No further information is given by the author about the distinction between these two types
of Focus. What is of interest here is the fact that they have a syntactically different behaviour.
144

3.2 Pragmatic scalarity
Contrastiveness and New Information: a new view on Focus
Let us focus on the pragmatic nature of the degrees of contrastiveness.
Contrastiveness is clearly a relational property, being the result of the interaction of
two elements: no individual item may be contrastive on its own. It therefore makes
sense, and it is indeed necessary, to take into consideration the status of the element
with which the contrast is established. This is precisely what the meaning of explicit
contrastiveness introduced in the previous section refers to. CF traditionally assigns
a new value to the variable whose (incorrect) value has already been assigned, in the
form of ‘It’s X, not Y’. Thus, crucially, the element with which the contrast is
established is already present in the discourse; and not simply present, but active.
We refer here to the notion of activeness used by Chafe (1987:22ff) and reinterpreted
by Lambrecht (1994:99ff). According to Chafe any event of transmitting
information does not only involve knowledge, but consciousness, too: given the size
of the knowledge and information that our minds hold, it is only natural that at any
given point, only a small amount can be focused on or is, in Chafe’s words, active.
Within the state of activeness, Chafe identifies three possible levels:
9 a. the referent is active: it is the focus of consciousness at a given moment;
b. the referent is semi-active (accessible): it is in a person’s awareness, in their
peripheral consciousness, but not directly focused on;
c. the referent is inactive: it is either not present at all or held in a person’s long-
term memory, neither focally nor peripherally active.
Within these three levels there are also a number of further specifications that
could be identified. In the case of an active referent, it may be textually (i.e. present
in the discourse) or situationally (i.e. deictically present) active; an inactive element
may be so by virtue of not being present in the person’s knowledge at all or by
having been demoted from a previously accessible status. In the case of a semiactive
referent Lambrecht (1994:100) adds that the accessible state could be due to
one of three factors:
10 i. deactivation from an earlier active state (textually accessible);
ii. inference from another active or accessible element (inferentially accessible,
either from the linguistic or extra-linguistic contexts) 13 ;
iii. deictically present in the text-external world (situationally accessible).
The levels of activeness, although clearly boxed as discrete values in the
following summarizing table, are to be interpreted as points along a continuum:
13 A referent that is inferentially accessible on an extra-linguistic basis is possibly one that is
part of shared knowledge between the speaker and hearer, be it specific to their own
environment, or culture, or universal, in other words the shared world knowledge of speaker
and addressee.
145

Sandra Paoli
146
Level of
activeness
1. Active
2. Semiactive
(accessible)
3. Inactive
Characteristics Referent is
Focus of
consciousness
In peripheral
consciousness,
speaker has
background
awareness of it.
Neither focally nor
peripherally active
Table 1: Levels of activeness
a. Linguistically active, hence
explicitly mentioned in the
conversation;
b. Extra-linguistically active,
and not (necessarily) mentioned
in the conversation;
a. Demoted from an earlier
active state;
b. Inferrable (either
linguistically or extralinguistically);
c. Deictically present, hence
situationally accessible;
a. Demoted from an earlier
accessible state, in long-term
memory;
b. Never present, hence
completely new.
On the basis of this classification, we propose that the varying degrees of
contrastiveness obtain by establishing a contrast with referents that are at different
levels on the activeness scale. In order to implement this further distinction in a
manageable way, we propose an array of ‘proto- 14 , cross-dimensional’ features,
expressing properties that encompass the psychological, the pragmatic, the semantic,
the prosodic as well as the syntactic domains. These are the building blocks into
which constituents of interface phenomena such as discourse saliency can be broken
down. Identifying an [active] feature which can combine with [+contrastive] enables
us to derive explicit and implicit contrast: in section 4 we discuss whether such a
feature can find expression at the syntactic level.
Before moving on to applying these distinctions to the Triestino cases, a few
more words of explanation of the ‘active’ status of a referent are in order. As Chafe
points out (1994:73ff), if language is to perform its communicative function
successfully, the speaker needs to take into account the addressee’s mind: clearly, if
14 Further research may reveal that the atomic features identified here are not atomic after all.
Their introduction in the system is an attempt to incorporate extra-grammatical properties in
the feature specification of these categories: their atomic nature is not essential, but a starting
point.

Contrastiveness and New Information: a new view on Focus
the speaker thinks that the referent Mary is active in the consciousness of their
interlocutor, they will be able to refer to her with the pronoun she. If the speaker’s
judgement were only based on the state Mary held in their own mind, the use of
such a pronoun may cause a breakdown in communication. In a similar fashion, the
treatment of a referent as active or inactive is a consequence of the speaker’s
monitoring of what is happening in the addressee’s mind.
With specific reference to Triestino, and with a view in mind to identify feature
combinations that could readily, if needed, be expressed at the syntactic level, we
simplify the ternary distinction proposed by Chafe to the binary [+active] and [active],
collapsing the semi-active and the inactive ones into the single [-active] 15 :
Feature Level of activeness
[+active] 1. Active
[-active]
2. Semi-active (accessible)
3. Inactive
Table 2: Mapping of levels of activeness onto features
Although our primary interest here is on the permutations of [± active] and
[+contrastive], the combination of [active] and [new] is also possible 16 , hence
offering a way of further distinguishing within the NIF category (cf. the distinction
suggested within the inactive status, between (3a) and (3b)). In spite of obvious
scope for overlap between [±active] and [±new], their existence as independent
features is justified: a newly introduced element is necessarily inactive in the sense
15 The need for Chafe’s three-way distinction may indeed prove necessary when extending the
investigation of Focus to other languages. The simplification to two suggested here does not
theoretically preclude it.
16 Not all permutations would be pragmatically possible: an element could not simultaneously
be [+active] and [+new]; the combination of [-active] and [-new] would correspond to the
case of (3a) in Table 1, representing an inactive element that has been demoted from an earlier
active state, and is not, therefore, new. In the two remaining possibilities, [+active], [-new],
and [-active], [+new], the [active] feature does not add significantly to the status of newness
of an element: in the former case the referent is the focus of consciousness, and hence not
new, and in the latter, being new, it is not. This highlights a degree of overlap between
[±active] and [±new]. A significant difference in the way [±active] interacts with
contrastiveness and newness of information falls out of the relational nature of contrast
mentioned above: while combined with [+contrastive], [± active] can only refer to the level of
consciousness pertaining to the element with which a contrast is established, when combined
with [±new], [± active] refers to the degree of consciousness of the referent itself.
147

Sandra Paoli
that being introduced at that very precise moment it cannot already be ‘lit up’ in the
listener’s mind, but so could be an element that has been introduced at some earlier
point in the discourse, and therefore is not new, but has been pushed in the long term
memory and is, therefore, inactive. Thus, while [±new] refers to the properties of a
referent with respect to its status in discourse, [±active] reflects the level of
consciousness at which it is held in the mind.
3.3 Application of these distinctions
The proto-, cross-dimensional features suggested here as well as their combination,
are by no means exhaustive nor definitive: this is an attempt at including in a
manageable fashion (hence the choice of binary values for the features) important
properties that refer to the mental status of referents which need to be taken into
account when discussing contrastiveness in a comprehensive and exhaustive way.
Although transformational approaches to grammar allow for discourse properties to
have a determinant role in accounting for word order phenomena and have adopted
them as syntactic projections, their integration and use in the system has been done
in a selective way. The concepts of contrastive focus, new information focus and
even topic have been included in a rather sterilised fashion, that is to say without
taking into account the complexity of their properties, especially of their psychological
and pragmatic import. The identification and introduction of these crossdimensional
features as the building blocks of different types of CF is a first step
towards allowing the incorporation of the whole range of properties of information
structure into the various computations.
Focusing on the interaction of [±active] and [+contrastive], we obtain the
following two types of CF:
148
[+active] [-active]
[+contr] Explicitly Contrastive Focus Implicitly Contrastive Focus
Table 3: Degrees of contrastiveness
E(xplicitly) CF is the well-known type that corresponds to the syntactic structure
‘It’s X, not Y’: Y has been introduced in the discourse and is still active in the mind
of the participants. I(mplicitly) CF corresponds to a ‘new’ type, a contrast
established with an element that is not in the active consciousness of the speaker but
is part of the peripheral focus of knowledge of the participants. Imagine A and B are
talking about a recent visit of a friend of theirs, John. Both A and B know that John
has been a vegetarian for a long time; yet, there is no linguistic mention in the
current discourse of this. In this context, the utterance ‘John had A CHUNKY STEAK
for dinner (imagine!)’ would express something unexpected on the basis of the state

Contrastiveness and New Information: a new view on Focus
of affairs that can be inferred when talking about John, i.e. among other things, that
he is a vegetarian. The contrast is implicit because the fact that John is a vegetarian,
has not been explicitly mentioned in the conversation: it is, nevertheless, accessible
when talking about John 17 .
As well as exemplifying degrees of contrastiveness, the cases of Triestino
fronting also point to the possibility of an element expressing simultaneously new
information and a contrast. If we allow for the combination of the two types of CF
and the feature [+new], we obtain further specifications, shown in Table 4:
[+new] Explicitly Contrastive,
New Info, focus
[+contr, +active] [+contr, -active]
Table 4: Newness and contrastiveness
Implicitly Contrastive,
New Info, focus
Although the feature [active] can indeed combine with [new] as discussed in
note 16, in Table 4 it is a property of contrastiveness, in other words a secondary
feature for which contrastiveness is further specified. This needs to be indicated with
the introduction of further bracketing:
Type of focus Feature specification
Explicitly Contrastive, New Info,
focus
{[+new], [+contr, +active]}
Implicitly Contrastive, New Info,
focus
{[+new], [+contr, -active]}
Table 5: Formal representation
The type {[+new], [+contr, -active]} is the representation of the cases of fronting
in Triestino: the fronted element is not only an instance of ICF but also expresses
new information. The other type, {[+new], [+contr, +active]}, refers to an ECF
phrase which also expresses new information. Not immediately obvious, which
perhaps questions its legitimacy, this would be the case of a conversation carried out
by three people. A makes a statement, B questions it, C replies correcting the
information provided by A:
17 This concept of inferrability is very similar to the idea of file-cards (Heim, 1982, Vallduví,
1992): a knowledge store in which every referent is associated to a card with the relevant
information about it.
149

Sandra Paoli
11 a. A: Mary had steak for dinner.
b. B: What did she have for dinner (sorry I didn’t hear you)?
c. C: She had FISH.
Let us now turn to the discussion of another aspect of contrastiveness. We have
seen how contrastiveness can derive from an unexpected state of affairs.
Interestingly, a similar effect is obtained when a given situation is so totally obvious
that its explicit mentioning would be unnecessary (possibly violating Grice’s, 1975
Maxim of Relation ‘Be relevant’), and hence, contrary to expectation. Consider the
following:
12 a. Sp. A: What are you fishing for?
Sp. B: I am fishing for FISH!! (what do you think, duh!)
150
b. Sp. A: What are you doing with that vase?
Sp. B: I’m PUTTING FLOWERS in it!! (what do you expect??)
These examples are particularly effective in Italian, especially (12a). The
reaction to a completely superfluous question is a contrast, given that the
information can be implicitly inferred from the shared world knowledge. The
contrast can therefore derive both from an assertion that goes against what is
expected, but also from an assertion that is exactly what is expected and hence not in
need of being spelled out. In terms of features, this case would fall within the ECF,
with the further specification that the referent is extra-linguistically active 18 .
4. Further reflections
4.1 Cross-linguistic variation
Intuitively, it makes sense to recognise the cross-dimensional nature of interface
phenomena and allow its expression through a range of cross-dimensional features.
Whether these features are ‘translatable’ onto the syntactic level is not immediately
apparent, and as mentioned in section 3.1, the degree of cross-linguistic variation is
remarkable. The view of what it means to be expressed at the syntactic level that we
take here is rather simplistic: a feature is syntactic if it causes word re-ordering
phenomena.
The reason for including [±active] as a further specification of [+contrastive] is
clearly motivated by the data at hand: the difference it marks, though, need not be
interpretable at the syntactic level. In the Triestino cases the specific type of
contrastiveness does not seem to have a correlate at the syntactic level: the
18 This distinction may make a case for allowing the further specifications of the referent
described in Table 1 to be also included in the array of features: here we simply add this
observation to make the point of the complexity of contrastiveness.

Contrastiveness and New Information: a new view on Focus
triggering force behind these cases of fronting is the [+contrastive] specification,
irrespective of its implicitness or explicitness. An object that expresses NIF can only
be found post-verbally: it can be licensed sentence-initially only when it also
expresses a contrast. It therefore seems that in Triestino [+contrastive] is a syntactic
feature in that it causes re-ordering phenomena 19 , while [+new], on its own, is not.
This is certainly not the case cross-linguistically: indeed, the syntactic expression of
[+new] and [+contrastive] features is subject to considerable variation.
It was mentioned above that a syntactic position dedicated to NIF in the left edge
of the clause is endorsed by Benincà and Poletto (2004), who on the basis of
examples such as (13) from Italian claim that NIF can indeed target a position within
the left periphery, but this becomes available only in the presence of a CF phrase:
13 A GIORGIO, QUESTO LIBRO, devi dare.
to Giorgio this book you-must give
‘It is to Giorgio that you must give this book’
Since the direct object questo libro ‘this book’ cannot be resumed by an object
clitic, i.e. it is not left-dislocated, they conclude that it must therefore be an instance
of Focus, more specifically, NIF. It is not clear that (13) is equally acceptable
throughout the Italian peninsula. It must be stressed that Italian is subject to strong
regional variation with the consequence that the ‘standard’ language yields widely
differing grammaticality judgments depending on the geographical origin of the
speaker. Given the interpretation of (13), the possible question that could elicit it,
Che cosa devo dare a Mario? ‘What do I need to give to Mario?’, does not seem to
produce it: a less awkward answer 20 , which is by no means completely felicitous,
would be (14), in which the direct object expressing new information necessarily
appears post-verbally.
14 ?A GIORGIO, devi dare QUESTO LIBRO.
to Giorgio you-must give this book
‘It’s (actually) to Giorgio, (and) you must give him this book’
The ability of Italian to license phrases expressing NIF at the left edge of a
clause is at least questioned by the evidence in (14) 21 : it seems that such position is
not uniformly available across the peninsula. As mentioned above, the acceptability
19 Cf. Lonzi (2007) for an interesting discussion of the syntactic status of contrastiveness:
cross-linguistic considerations clearly point to a great degree of variation.
20 This is based on the intuition of a small group of speakers including both North-eastern and
North-western Italian informants.
21 Cf. Stoyanova (2008:40ff) for a discussion of the availability of more than one focus
position to Italian, and its link with the formation of multiple wh-questions.
151

Sandra Paoli
of (14) may be a matter of regional variation, depending, in other words, on the
underlying influence of the individual dialects that the various speakers have.
It was mentioned earlier that Sicilian and in Sardinian (cf. Bentley, 2007; Jones,
1993), allow constituents expressing new information sentence-initially (cf. 15a,
Sardinian from Jones, 1993:18; 15b, Sicilian) without the need for the Focus field to
have been activated by a CF phrase. In such varieties, a functional projection
associated with NIF, NIFP, has been assumed:
15 a. CUSSU LIBRU appo lessu. Sard
this book I-have read
‘I have read this book’
152
b. CARNI mangiai. Sic
meat I-ate
‘I ate meat’
Although CF and NIF cannot co-occur, Bentley (2007: 53, from which the
following examples are taken) notices that in Sicilian CF (cf. 16c), unlike NIF (cf.
16b), can be separated from the verb phrase, on the lines of the distinction encoded
in the Finnish, Hungarian and Basque data mentioned earlier. This suggests that the
element expressing contrastive focus occupies a higher position than the one
encoding new information:
16 a. Sp. A: Chi cci ricisti a tò niputi?
what to-them you-said to your nephews
‘What did you tell your nephews?’
b. Sp. B: A VIRITÀ (*a mè niputi) cci rissi.
the truth (*to my nephews) to-them I-said
‘I told them the truth’
c. NA LITTRA, a Pina, cci scrissi (no nu pizzinu)
a letter to Pina to-her I-wrote (not a card)
‘It’s a letter that I wrote to Pina, not a card’
The new information direct object A VIRITÀ ‘the truth’ cannot be separated from
the verb: the same restriction does not apply to the contrastive NA LITTRA ‘a letter’
in (16c).
The conclusion that can be drawn from this, is that, clearly, [+contrastive] and
[+new] do not have the same status cross-linguistically: while in some varieties they
are both responsible for re-ordering phenomena (as in Sicilian), in others only one of
them triggers overt movement (cf. [+contrastive] in Triestino). The encoding of such
variation may be subject to parametrisation and linked to other properties of the
language. Furthermore, the discussion in Bentley (2007) seems to suggest that in
those languages in which both [+new] and [+contrastive] are syntactic features, there

Contrastiveness and New Information: a new view on Focus
are distinct and ordered functional projections associated with them: in Sicilian the
position dedicated to CF appears to be higher than the one occupied by NIF.
We therefore conclude that both [+contrastive] and [+new] may be syntactic
features, but certainly need not be; the same can be said of the feature introduced in
this article, [±active], and the evidence to motivate the choice is to be found in each
individual language. If examples such as (13) are indeed acceptable with the
interpretation of the higher phrase as contrastive and the lower one as new
information, the uniqueness of Focus (cf. Rizzi, 1997) needs to be restated in less
constrained terms.
With respect to the debate on whether CF and NIF are syntactically distinct, the
evidence discussed here points to great variation. Our understanding of Focus as a
syntactic space follows Benincà and Poletto’s (2004) idea of a field, which could
indeed be internally articulated in two projections, CFP and NIFP, in those
languages in which both [+contrastive] and [+new] features are expressed
syntactically. This would include Sicilian and (possibly) Sardinian. In Triestino and
perhaps Italian, it would be a space hosting only elements carrying the feature
[+contrastive] and displaying the other properties associated with quantificational
elements.
Summarising the evidence discussed so far, this is the distribution and nature of
the features [+new] and [+contrastive]:
Italian Triestino Sicilian Sardinian
Features [+new] [+contr] [+new] [+contr] [+new] [+contr] [+new] [+contr]
Syntactic
status
?
Table 6: Cross-Linguistic variation
4.2 Are [±active] features syntactic at all?
Let us now return to the question of whether [+active] is a syntactic feature or
not. Consider the following asymmetry in Triestino: with a factive verb such as
preocuparse ‘to worry’, [+contrastive], [+active] phrases (in bold capitals) can
equally target the main or the embedded left periphery (cf. 17 a and b), while only
the left periphery of the main clause is readily available to [+contrastive], [-active]
elements (cf. 18 a and b).
17 a. [UNA MOTO me preocupa [che i ghe cioghi a
a motorbike to-me it-worries that scl to-her buy to
Marina, no una bici]].
Marina not a bike
‘I am worried that they buy a motorbike for Marina, not a bike’
153

Sandra Paoli
154
b. [Me preocupa [che UNA MOTO i ghe cioghi a Marina, no una bici]].
18 a. Sp. A: Cossa te preocupa che i ghe cioghi a Marina?
what to-you it-worries that scl to-her buy to Marina
‘What are you worried that they buy Marina?’
b. Sp. B: [UNA MOTO me preocupa [che i ghe cioghi]].
a motorbike to-me it-worries that scl to-her buy
‘I am worried that they buy her a motorbike’
b’. Sp. B: [?Me preocupa [che UNA MOTO i ghe cioghi]].
Interestingly, this asymmetry disappears when the selecting predicate is an
epistemic verb such as creder ‘to believe’, and both the main and embedded left
peripheries are readily available to both [+contrastive], [+active] and [+contrastive],
[-active] phrases (cf. 20b and b’):
19 a. [UNA MOTO credo [che i ghe cioghi a Marina, no
a motorbike I-believe that scl to-her they-buy to Marina not
una bici]].
a bike
‘It’s a motorbike that I think they are buying Marina, not a bike’
b. [Credo che [UNA MOTO i ghe cioghi a Marina, no una bici]].
20 a. Sp. A: Cossa te credi che i ghe cioghi a Marina?
what scl you-believe that scl to-her they-buy to Marina
‘What do you think they will buy Marina?’
b. Sp. B: [UNA MOTO credo [che i ghe cioghi]].
a motorbike I-think that scl to-her they-buy
‘I think they are buying her a motorbike’
b’. Sp. B: Credo [che UNA MOTO i ghe cioghi].
Summarising the distribution of the two types of CF:
Factive Epistemic
ECF ICF ECF ICF
Main
Embedded
Table 7: Summary of main vs embedded left periphery available
This asymmetry in the licensing of elements in the embedded left periphery has
been analysed by Poletto (2000:121ff) as a reflection of the different selecting
properties of the two classes of verbs. Epistemic verbs (which she terms ‘bridge
verbs’) select a ‘fuller’ CP than factive verbs (her ‘non-bridge verbs’): the structure

Contrastiveness and New Information: a new view on Focus
embedded under the latter appears ‘poorer’ (see also, Benincà and Poletto, 2004:61),
in the sense that there are fewer positions available.
The fact that there does not seem to be a position available for the [+contrastive],
[-active] element in ‘reduced CPs’ selected by factive verbs may suggest that a
position for ICF elements is not constantly accessible: it is only available in a
‘larger’ CP. We re-interpret Benincà and Poletto ‘reduced’ CP to mean a CP in
which only the primary features are realised, but not their further specification. In
this case, [+contrastive] would be expressed, but its further distinction [-active]
would not: so an ECF but not an ICF phrase would find expression in a reduced CP.
5. Some evidence from prosody
Let us now turn to a brief investigation of the prosodic properties of the two
types of CF identified. Consider the following minimal pairs (21b) and (22b), ECF
and ICF respectively:
21. a. Sp. A: La polizia ga fermà Marina.
the police it-has stopped Marina
‘The police have stopped Marina’
b. Sp. B: MANUELA i ga fermà (no Marina).
Manuela scl they-have stopped not Marina
‘It is Manuela that they have stopped (not Marina)’
22. a. Sp. A: Chi ga fermà la polizia?
who it-has stopped the police
‘Who have the police stopped?’
b. Sp. B: MANUELA i ga fermà!
Manuela scl they-have stopped
‘They have stopped Manuela!’
The context of (22b) is that Manuela is a very careful and observant driver, and it
is hence unexpected that she should be stopped by the police. These are the pitch
tracks of two token utterings obtained using Praat: each track also indicates the Hz
values of the beginning, the peak(s) and the after-peak. As they are, these figures are
not significant in absolute terms, but are useful in their relative values.
155

Sandra Paoli
156
Figure 1 – Pitch track for (21b)
Figure 2 – Pitch track for (22b)
In addition to the sharp rise and fall (which do not figure in the pitch track in
figure 1) on the second half of (21b), … no Marina, ‘… not Marina’, the two
prosodic contours also differ in the initial and maximum values. ICF starts at a
higher value than ECF, and reaches a higher peak on the stressed syllable of
Manuela. Figure 3, superimposing pitch tracks for ECF (dashed line) and ICF
(dotted line) to one extracted from an instance of a neutral uttering of Manuela (in a
continuous line) shows this clearly:

Contrastiveness and New Information: a new view on Focus
Figure 3 – Comparing pitch tracks
Not only are both (21b) and (22b) clearly not uttered with neutral intonation, but
there are noticeable differences between their respective tracks: the curve associated
with ICF is much more pronounced than the one produced by ECF. This feels
somehow counter-intuitive: explicit and implicit should parallel with a higher and a
lower contour, respectively. Nevertheless, this may find a plausible explanation in
cognitive terms.
Notice that it is more ‘costly’, cognitively speaking, to activate something that is
not in the focus of consciousness than it is to keep active something that is already
active. In order words, something that is already present in the discourse (i.e. given)
requires less mental effort to become active than something that is not. This is what
Chafe (1994:73) defines as ‘activation cost’. As ECF establishes a contrast with a
referent that is given and active, it will be less costly than ICF, which, expressing a
contrast with a semi-active referent, will on the other hand involve more cognitive
energy. There may well be a link between the extra cost and the higher intensity of
the pitch tracks. A similar observation can be made on the syntactic level. While an
answer to a simple question naturally excludes the VP, in the cases of Triestino
fronting that we have discussed here, the VP is preferably repeated (cf. 23b).
Furthermore, the fronting of the constituent is needed to obtain the unexpected
reading: the same reading is not available when the ICF phrase occurs post-verbally
cf. 23c), differently from ECF (cf. 23d). Compare the following to (22):
23 a. Sp. A: Chi ga fermà la polizia?
who it-has stopped the police
‘Who have the police stopped?’
b. Sp. B: ?Manuela!
157

Sandra Paoli
158
c. Sp. B: ?I ga fermà Manuela!
scl they-have stopped Manuela
d. I ga fermà MANUELA (no Marina).
scl they-have stopped Manuela not Marina
‘It is Manuela that they have stopped (not Marina)’
This idea that the higher cognitive cost may also be expressed prosodically (in
terms of higher hertz values) and syntactically (in terms of more syntactic material
and necessary movement) is purely speculative at this stage, however appealing it
may be. It clearly needs substantial further investigation, and we propose to do this
in future work.
Concluding with an evaluation of the evidence discussed in this section, ICF
appears to be uttered with a non-neutral intonation: the sharp rise and fall resemble
the contour produced by ECF, possibly linked to the presence of a contrast. Yet, the
pitch track produced by ICF starts and peaks at higher values than ECF, suggesting
that the two are different at the prosodic level. This lends further support to our
claim that the two are pragmatically, and to a certain degree syntactically, distinct.
6. Conclusions
The information structure of a sentence is clearly contributed by a number of
different dimensions: psychological, pragmatic, syntactic, semantic, prosodic. If we
are to take it seriously and express its import accurately, it is necessary to identify its
complexity and to allow the expression of such complexity within grammar. As a
first step in this direction, we have introduced in the form of a feature the notion of
‘activeness’. By conceiving a feature specification breakdown applicable to Focus
that aims at expressing its multi-dimensionality, we have introduced the concept of
‘cross-dimensional features’, traits that may also find expression at the syntactic
level.
In this article we have suggested that:
• contrastiveness is a scalar notion: there is more to it than a simple [±] value;
• the varying degrees of contrastiveness can be obtained through establishing a
contrast with a referent that is held at different levels of the activeness scale
in the listener’s consciousness;
• in order to do this, the feature [±active] has been introduced;
• [±active] can combine with both [±contrastive] and [±new];
• [new] and [contrastive] can combine with each other;
• the features [+contrastive] and [+new] can be syntactic: we assume that this is
the case in those languages in which they trigger word re-ordering
phenomena; in those languages in which they do not, the phrases that carry
them are simple instances of the syntactic category of Focus;

Contrastiveness and New Information: a new view on Focus
• when both [+contrastive] and [+new] are expressed at the syntactic level, the
two are hosted in distinct and ordered projections, with CF being higher than
NIF;
• only the primary values of the syntactic features [+contrastive] and [+new]
can find expression in ‘reduced’ CPs; full CPs are able to host both primary
and secondary;
• whether the cross-dimensional features identified in this article are expressed
at the syntactic level is subject to great cross-linguistic variation, and it
depends on the strategies employed by a language to express information
structure.
The system suggested here, and only roughly implemented, is not the final
product, but a first step towards trying to solve the tension between syntax and
pragmatics witnessed in the understanding of information structure.
References
Belletti, A. (2004) Aspects of the low IP area. In: L. Rizzi (ed.), The Structure of CP and IP.
The Cartography of Syntactic Structures. Volume 2. Oxford/New York: Oxford
University Press: 16-51.
Benincà, P. and C. Poletto (2004) Topic, Focus and V2: Defining the CP Sublayers. In: L.
Rizzi (ed.), The Structure of CP and IP. The Cartography of Syntactic Structures.
Volume 2. Oxford/New York: Oxford University Press: 52-75.
Bentley, D. (2007) Relazioni grammaticali e ruoli pragmatici: siciliano e italiano a confronto.
In: D. Bentley and A. Ledgeway (eds.), Sui dialetti italoromanzi: Saggi in onore di
Nigel B Vincent (the Italianist 27, Special supplement 1). Norfolk: Biddles: 48-62.
Bolinger, D. (1954) English prosodic stress and Spanish sentence order. Hispania 37: 152-
156.
Bolinger, D. (1961) Contrastive accent and contrastive stress. Language 37:1: 83-96.
Brody, M. (1990) Some Remarks on the Focus Field in Hungarian. UCL Working Papers 2:
201-225.
Brunetti, L. (2004) A unification of focus. Padova: Unipress.
Chafe, W. (1976) Giveness, contrastiveness, definiteness, subjects, topics and point of view.
In: C. N. Li (ed.), Subject and Topic. New York: Academic Press: 25-55.
Chafe, W. (1987) Cognitive constraints and information flow. In: R. Tomlin (ed.), Coherence
and grounding in discourse: outcome of a Symposium. Amsterdam/Philadelphia: John
Benjamins Publishing Company: 21-51.
Chafe, W. (1994) Discourse, Consciousness and Time: The Flow and Displacement of
Conscious Experience in Speaking and Writing. Chicago & London: The University
of Chicago Press.
Cinque, G. (1999) Adverbs and Functional Heads: A Cross-linguistic Perspective. Oxford/
New York: Oxford University Press.
159

Sandra Paoli
DeLancey, S (1997) Mirativity: The grammatical marking of unexpected information,
Linguistic Typology 1: 33-52.
Dik, S. C. (1989) The Theory of Functional Grammar, Part I: The Structure of the Clause.
Berlin: Mouton de Gruyter.
Donati, C. and M. Nespor (2003) From focus to syntax. Lingua 113:11: 1119-1142.
Extepare, R. (1997) Two types of focus in Basque. In: B. Agbayani and S. W. Tang (eds.),
Proceedings of the 15 th West Coast Conference on Formal Linguistics. Stanford:
CSLI: 113-127.
Frascarelli, M. (2000) The syntax-phonology interface in focus and topic constructions in
Italian. Studies in Natural Language and Linguistic Theory 50. Dordrecht: Kluwer.
Grice, H. P. (1975) Logic and conversation. In: P. Cole and J.L. Morgan (eds.), Syntax and
semantics: Vol, 3. Speech acts. New York: Academic Press: 41-58.
Halliday, M. A. K. (1967) Notes on transitivity and theme in English part II. Journal of
Linguistics 3: 199-244.
Heim, I. (1982) The semantics of definite and indefinite noun phrases. PhD thesis, University
of Massachusetts, Amherst.
Horvarth, J. (1986) Focus in the Theory of Grammar and the Syntax of Hungarian. Dodrecht:
Foris.
Jackendoff, R. (1972) Semantic representation in generative grammar. Cambridge, Mass.:
MIT Press.
Jones, M. (1993) Sardinian Syntax. London: Routledge.
Järventausta, M. and V. Molnár (2003) Discourse configurationality in Finnish and
Hungarian. In: J. Hetland & V. Molnár (eds.), Structures of focus and grammatical
relations, Tübingen: Niemeyer: 111-148.
Kiss, K. É. (1987) Configurationality in Hungarian. Dordrecht: Reidel.
Kiss, K. É. (1995) Discourse Configurational Languages Introduction. In: K. É. Kiss (ed.),
Discourse Configurational Languages. Oxford/New York: Oxford University Press:
3-27.
Kiss, K. É. (1998) Identificational Focus Versus Information Focus. Language 74: 245-273.
Lambrecht, K. (1994) Information structure and sentence form. Topic, focus and the mental
representations of discourse referents. Cambridge: Cambridge University Press.
Lonzi, L. (2006) Intonazione contrastiva e strutture di base. In Annali Online di Ferrara –
Lettere, Vol. 1: 53-74.
Lonzi, L. (2007) Contrastive intonation and its disguises of syntactic structure in Italian. Ms.
Molnár, V. (2002) Contrast – from a contrastive perspective. In: H. Hasselgård et al (eds.),
Information Structure in a Cross-Linguistic Perspective. Amsterdam/New York:
Rodopi: 147-161.
Nespor, M. and M.T. Guasti (2002) Focus to stress alignment and its consequences for
acquisition. Lingue e Linguaggio 1: 79-106.
160

Contrastiveness and New Information: a new view on Focus
Paoli, S. (2010) In focus: an investigation of information and contrastive constructions. In: R.
D’Alessandro, A. Ledgeway and I. Roberts (eds.), Syntactic variation: The dialects of
Italy. Cambridge: Cambridge University Press: 277-291.
Poletto, C. (2000) The Higher Functional Field. Oxford/New York: Oxford University Press.
Puskás, G. (2000) Word order in Hungarian: the Syntax of A-bar Positions. Amsterdam/
Philadelphia: John Benjamins Publishing Company.
Rizzi, L. (1997) The Fine Structure of the Left Periphery. In: L. Haegeman (ed.), Elements of
Grammar. Handbook of Generative Syntax. Dordrecht: Kluwer: 281-337.
Rooth, M. E. (1992) A theory of focus interpretation. Natural Language Semantics 1: 75-116.
Slobin, D. and A. Aksu (1982) Tense, aspect, and modality in the use of the Turkish
evidential. In: P. Hopper (ed.), Tense-Aspect: Between Semantics and Pragmatics.
Amsterdam: Benjamins: 185-200.
Stoyanova, M. (2008) Unique Focus: Languages without multiple wh-questions. Amsterdam/
Philadelphia: John Benjamins Publishing Company.
Vallduví, E. (1992) The Informational Component. New York: Garland Press.
Zimmermann, M. (2008) Contrastive Focus and Emphasis. Acta Linguistica Hungarica, 55:3-
4: 347-360.
Zubizarreta, M. L. (1998) Prosody, Focus, and Word Order. Cambridge, Mass.: MIT Press.
161

RIVISTA DI GRAMMATICA GENERATIVA
Volume 34, anno 2009
A note on backward anaphora Valentina Bianchi
Syntactic diagnostics for referentiality
marking in early null objects: evidence
from Italian
The nature of syntactic impairment
in autism
A comparison between the Cantonese
aspect markers gan and haidou
Valentina Brunetto
Stephanie Durrleman
& Sandrine Zufferey
Chi-Fung Lam
Caratteri della Cliticizzazione di ne Lidia Lonzi
Contrastiveness and New Information:
a new view on Focus
Sandra Paoli