A Delay of Principle B-Effect in Spanish Speaking Children: The

A Delay of Principle B-Effect in Spanish Speaking Children: The Role of
Lexical Feature Acquisition
Sergio Baauw, María A. Escobar & William Philip
Utrecht Institute of Linguistics - OTS,
Utrecht University
Brain and Cognitive Sciences,
Massachusetts Institute of Technology
Department of Humanities,
University of Maine at Farmington
[email protected], [email protected], [email protected]
Abstract
Many experimental studies have shown that young
children often seem to violate Principle B of the
Binding Theory, allowing pronouns to corefer with a
local c-commanding antecedent. This phenomenon,
which has come to be known as the "Delay of Principle
B-Effect" (DPBE), has been argued to be absent in the
acquisition of the Romance languages (McKee 1992;
Padilla 1990). Recent experimental evidence shows
that this is only partially true. Our findings show that,
although Spanish has no general DPBE, it does have a
DPBE in what we will call "Complex Predicate
Constructions". We tie the absence of a general DPBE
in Spanish to the underspecification of (Romance)
pronominal clitics for the feature [human]. We will
show that the presence of a DPBE in Complex
Predicate Constructions is due to the misclassification
of third person pronouns as elements that can be
referentially defective. We will argue that the results of
this experiment support Reinhart & Reuland's (1993)
"Reflexivity" approach to binding.
1. The Clitic Effect
2. Rule I and the Clitic Effect
Following Grodzinsky & Reinhart (1993) we will argue
that the DPBE in constructions such as (1a) is due to
the children’s unability to apply Rule I. Rule I is a rule
at the syntax/pragmatics interface that regulates
intrasentential coreference:
(2) Rule I: Intrasentential Coreference
(Grodzinsky & Reinhart 1993: 88)
NP A cannot corefer with NP B if replacing it
with C, C a variable A-bound by B, yields an
indistinguishable interpretation.
What this rule basically says is that if (3b) and (3c)
yield the same interpretation, (3c) should be chosen and
(3b) rejected. Stated differently: if there is no difference
in interpretation, binding (= co-indexation) is a more
economical
way
of
establishing
referential
dependencies than coreference (= counter-indexation).
(3) a. *The girli is drying heri off.
Principle B violated
English children allow her to corefer with the girl
roughly 50% of the time in structures such as (1a)
(Chien & Wexler 1990). Italian children, on the other
hand, show virtually 0% non-adult-like performance on
the Italian counterpart of (1a), (1b) (McKee 1992).
b. The girli is drying herj off.
Principle B satisfied
(1) English:
a. The girl is drying her off.
c. The girli is drying xi off.
Italian:
b. Gianni lo asciuga.
Italian weak pronouns such as lo in (1b) belong to a
separate class of pronominal "special clitics" that
Romance languages have (Zwicky 1977). This means
that the DPBE does not appear in constructions
involving "special clitics". We will call this absence of
the expected DPBE the Clitic Effect (CE). Two
questions must now be answerd: 1. what causes the
DPBE, and 2. what causes the CE?
coreference: [i] and [j] have the same
semantic value.
Since Rule I requires that two LF representations of
(1a) be compared, (3b) and (3c), it breaks down in
young children due to limitations on working memory.
When this happens, in order to determine the reference
of the pronoun children will adopt a guessing strategy,
leading to the observed 50% adult-like performance.
If the DPBE is the result of a breakdown of Rule I,
this must mean that Rule I does not apply to
pronominal clitics, such as Italian lo `him'. The
1
question is now: why does Rule I not apply to clitics?
1
Our claim that Rule I cannot apply to clitics would seem to
entail that Rule I effects with sentences like "Even Oscar likes
We will argue that Rule I does not apply to clitics
because clitics can only be bound. We believe that this
ban on coreference is tied to their underspecification
for the feature [human]. Delfitto & Corver (1993) argue
that in order for their phi-features (person, number,
gender) to be interpretable at LF, elements on D
position need some minimal lexical content. In the case
of articles, this lexical content is provided by the NP
complement. Pronouns, on the other hand, lack an overt
NP complement. For their features to be interpretable at
LF they need specification for the feature [human]. So
called "strong" pronouns, such as Italian lui, are
(positively) specified for this feature: they are
[+human] Clitics, however, are underspecified for this
feature: they are [+/-human] (see also Cardinaletti &
Starke 1995).
(4) a. Gianni lo vede. (lo = Bill/the tree)
b. Gianni vede lui. (lui = Bill/*the tree)
`G. sees him.'
We propose that specification for the feature [human] is
tied to nominal categorial status. As we will see in the
next section, the categorial status of an element
determines its position within the DP.
(5) A.
B.
[+N] --> [+human] or [-human] (exclusive
disjunction)
[+D] --> [+/-human]
We will argue that pronominal elements underspecified
for the feature [human], e.g. clitics, must be bound,
either in syntax or in discourse (d-linking) (6), because
binding provides a specification for the feature
[human]: the underspecified element (e.g the clitic)
inherits a value for this feature from the binder. This
also explains why clitics cannot be used deicticly:
binding excludes deixis.
In sum, we propose the following typology of
referential dependencies:
bound
unbound
syntax
discourse syntax
discourse
pronouns pronouns pronouns pronouns
(Rule I)
specified
[+human]
or [-human]
underspecified clitics
i.e.,
[+/-human]
clitics
Only when coreference is lexically possible, is Rule I
invoked. However, when coreference is lexically
possible, it blocks d-linking (= discourse binding).
Since we define binding in terms of co-indexation, it
follows that co-indexation is also involved in d-linking.
We propose that the difference between syntactic
binding and discourse binding is that the former
involves co-indexation and c-command, while the latter
only involves co-indexation (i.e. D-linking is a default
for assigning the same referent to different NPs that are
not in a c-command relation).
3. The Clitic Effect with Spanish Pronouns
Spanish, like Italian is a language with two separate
classes of pronominal elements: clitics and so called
"strong" pronouns. In (7) we give different uses of
Spanish pronouns and their interpretations:
(7) CLITICS
a. La niña la ve.
`The girl sees her.'
(la = Maria/the house)
"STRONG" PRONOUNS
Since coreference is excluded by binding, Rule I is not
invoked, hence cannot break down in young children.2
b. La niña apunta hacia ella.
`The girl is pointing at her.'
2
Interestingly, Chien & Wexler show that the DPBE is not
observed in English sentences such as (i):
(i) Every girl is pointing at her.
This is predicted, as Grodzinsky & Reinhart (1993) point out,
since coreference between a quantified subject and a pronoun
-
figure 1: referential dependencies
(6) [D/S....[+/-human]...] --> [D/S...Oj...[+/-human]j...]
(D=discourse/S=sentence)
him" cannot occur in adult Spanish or any other Romance
grammar. This is not our claim. We speculate that there is
another form of Rule I that can also apply to clitics and even
quantified DPs (cf. Heim 1993). However, this "secondlevel" Rule I seems to incur considerably greater processing
cost. This is noticable in English when the availability of a
coreference reading in (i) is compared with that in (ii):
(i) Oscar hates him.
(ii) Everybody hates him.
-
(ella = Maria/the house)
c. La niña la ve a ella.
`The girl sees her.'
(ella = Maria/*the house)
d. La niña apunta hacia ELLA.
`The girl is pointing at HER.'
is no option: quantifiers can only establish bound-variable
relations with pronouns.
(ella = Maria/*the house)
As can be seen in (7a), the clitic la `her' is [+/
-human].
(7b) shows that Spanish "strong" pronouns such as ella
do not refer exclusively to human referents, unlike
Italian lui `him'. When
ella is the object of a verb, like
in (7c), (in which case it is always doubled by a clitic),
or when it bears focal stress, like in (7d), it is
interpreted as a strong pronoun: it only has human
reference. However, when ella is the complement of a
preposition, and does not bear focal stress, like in (7b),
it has a clitic-like interpretation: both human and nonhuman reference are allowed (cf. Schroten 1992).
Recall from section 2 that we tie specification for the
feature [human] to nominal categorial status, and
underspecification for this feature to the D-position,
this position being the host of just referentiallity and
phi-features. We therefore propose the following
lexical characteristics of Spanish clitics such as la and
"strong" pronouns such as ella:
(8) la
--> [+/-human], [+D]
ella 1 --> [+/-human], [+D]
ella 2 --> [+human], [+N]
We propose the following analysis for the object DPs
of (7a-d), relating elements specified for the feature
[human] and with a [+N] categorial status to the [Spec,
DP] position, and elements underspecified for this
feature to the D position (adopting a modified version
of Uriagereka's (1995) analysis of Spanish strong
pronouns and clitics).
(9) a. lai..... DP
b.
/ \
Spec D′
/ \
D NP
ti
DP
/ \
Spec D′
/ \
D NP
ella
c. lai..... DP
d.
/ \
Spec D′
ellai / \
D NP
ti
DP
/ \
Spec D′
ella / \
D NP
We assume that focusing requires some minimal
semantic content of the pronoun (= specification for the
feature [human]), hence (7d) should be analysed as
(9d), with ella in [Spec, DP]. In (9c), which coresponds
to (7c), the [+human] specification of ella follows
without any stipulation: the pressence of a trace in D,
left by the doubling clitic leaves only [Spec, DP] to
ella.3 Crucially, (7b) is compatible with two analyses: if
the referent is non-human, only (9b) is compatible. If
the referent is human, both (9b) and (9d) are
compatible. Recall (8): ella is lexically ambiguous.
This means that when the referent is human, ella can
always be analysed as [+human]. In fact, we propose
that this is what actually happens: human reference will
block the default [+/-human] specification of ella,
leaving analysis (9b) only to instances of non-human
reference.
This allows us to make the following predictions for
the acquisition of pronominal anaphora in Spanish:
Spanish children are expected to have 0% of non-adult
like performance on (7a), (since la is [+/-human]), but
only 50% adult like performance on (7b-d) (since in all
these cases ella will be analysed as [+human], due to
the human nature of the antecedents, which excludes
analysis (9b)).
4. The Experiment
The goal of the experiment was to find out whether
there is any DPBE in the following three syntactic
contexts: simple sentences with a referential subject
(10a), simple sentences with a quantified subject (10b)
and complex predicate constructions (10c).
(10) a. La niña la seca.
(SIMPLE)
`The girl is drying her off.'
b. Cada niña la seca.
(QUANTIFIED)
`Every girl is drying her off.'
c. La niña la ve bailar.
(COMPLEX)
`The girl sees her dance.'
The experiment was carried out with 45 Spanish
speaking children (mean age 5;6) from Madrid and
Valladolid. The overall design was a Picture
Verification Task (very similar to Chien & Wexler's
(1990) 4th experiment and virtually identical in
materials and design to Philip & Coopmans (1996) as
well as Hamann & Philip (this volume). The
experiment consisted of 51 trials divided over 3 test
conditions, 8 control conditions and 15 filler items (= 3
trials per condition). The materials were
counterbalanced as to (i) whether an adult "si" or "no"
response was elicited, (ii) whether a reflexive or nonreflexive action was depicted in the picture, and (iii)
whether a pronominal or a reflexive clitic occurred in
the target input. Two sets of 21 x 29 cm colour pictures
were used, one set for each session. In figure 2 we give
one picture per test condition (SIMPLE, QUANT and
COMPLEX: adult answers “no”) (CI = context-setting
input, TI = target input).
3
Note that, although a movement analysis of clitics is
assumed here, an analysis which assumes base generation of
the clitic as the head of a functional projection (Sportiche
1992) also assumes that doubled elements, such as the
"strong" pronoun in (9c), are XPs, possibly generated in
[Spec, DP], (and moved to the Spec of the clitic head at LF).
SIMPLE (¿La niña la seca?)
Hmmm..a girl with a towell and a mom (CI)
Is the girl drying her off? (TI)
QUANT (¿Cada mamá la señala con el dedo?)
Figure 3: Percent Adult-like “No” Responses on Test
Conditions
Spanish "strong" pronouns have not been tested in this
experiment, but Hamann & Philip (this volume) show
that French children's performance on strong pronouns
was non-adult-like roughly 50% of the time. However,
they tested strong pronouns as complements of locative
prepositions, a contexts in which adults allow
coreference with a local subject.4 Clearer evidence is
provided by a pilot study by Berger (1997) on the
acquisition of pronominal anaphora in Italian. She
tested young Italian speaking children on both
sentences with object clitic pronouns, such as (11a) and
sentences where the object is a strong pronoun, such as
(11b).
(11) a. Il ragazzo lo sta indicando.
b Il ragazzo sta indicando lui.
`The boy is pointing at him.'
Hmmm...three moms and a girl (CI)
Is every mom pointing at her? (TI)
COMPLEX (¿La mamá la ve bailar?)
She found that children allowed corefrence much more
often with constructions with strong pronouns such as
(11b) than constructions like (11a) with a clitic
pronoun. However, Italian (and French) strong
pronouns are not optionally specified for the feature
[human], unlike Spanish strong pronouns; they are
always [+human]. In this respect Dutch might be
interesting. A Dutch pronoun such as hem `him' could
be analysed on a par with Spanish "strong" pronouns
such as ella, if we assume hem and the weak form `m to
be somehow morphologically related (by means of
some sort of PF contraction rule): in that case hem
could be considered optionally specified for the feature
[human].5
4
Hmmm...a big mirror, a mom, and a girl (CI)
Does the mom see her dance? (TI)
Figure 2: Test Conditions
The results show that our predictions with respect to
Spanish children's interpretation of clitics such asla are
borne out: virtually adult like performance was
obtained on sentences such as (10a) (SIMPLE
condition). Children's performance on control
conditions containing the anaphor se was also virtually
100% adult-like.
SIMPLE
| 90%
QUANTIFIED
90%
COMPLEX
64%
|
It is not clear why roughly 50% of the time French
children reject coreference between the locative complement
and the subject in constructions such as (i).
(i) Le garçon a mis le boite derrière lui.
`The boy has put the box behind him.'
Perhaps the child has to learn that locative complements can
be optionally treated as a different binding domain, allowing
pronouns to be bound by the subject of the sentence. If the
child is unaware of this, coreference (instead of binding) is
the only option left to establish referential dependency
between the pronoun and the subject. If coreference is
established, Rule I is invoked and subsequently breaks down.
This explanation is rendered some plausiblity if we consider
the fact that not in all languages the same prepositions create
their own binding domain. In Spanish, for example, not only
locatives seem to (optionally) define their own binding
domain:
(ii) a. Juani sólo piensa en éli.
John only thinks about him
b. Juan sólo piensa en sí mismo.
John only thinks about himself
We will leave this matter for future research.
5
To consider `m as a contracted form of hem is not so
obvious as it might seem. Zwart (1992) shows that Dutch
(12) a. Gisteren zagen we 'm.
('m = John/the car)
b. Gisteren zagen we HEM.
(hem = John/*the car)
`Yesterday we saw him.'
As predicted, Dutch has a DPBE in SIMPLE sentences
(Philip & Coopmans 1996). Future research should
show whether these predictions hold for Spanish strong
pronouns.
5. A DPBE in Spanish
As the results indicate, the DPBE is not totally absent
in Spanish. Spanish children exhibit 36% non-adult like
performance on the COMPLEX condition such as
(10c). We will argue that these results can easily be
accommodated in Reinhart & Reuland's (1993)
"Reflexivity" approach to binding.
According to R&R, the distribution of pronouns is
regulated by two different modules of UG, the revised
Binding Theory and a well-formedness condition on AChains.
(13) Reinhart & Reuland's Binding Theory
Principle A: A reflexive-marked syntactic
predicate must be interpreted reflexively
Principle B: A reflexively interpreted semantic
predicate must be reflexive-marked
(14) General Condition on A-Chains
A maximal A-chain (a1.....an) contains exactly one
link --a1-- that is both [+R] and structurally Case
marked.
R&R argue that the bound reading of la niña and la in
(10c) is ruled out by the A-Chain Condition, not by
principle B, which only applies to co-arguments of a
single predicate: la and la niña do not belong to the
same predicate, since la receives its thematic role from
bailar `dance', although its accusative Case is assigned
by the root predicate ve `sees'. An -Chain
A
consisting
of la niña and la violates the A-Chain Condition, since
weak pronouns are derived from strong forms from a
diachronic point of view. Yet, from a synchronic point of
view, Zwart argues, they must be independently stored in the
lexicon, since there is no productive phonological rule that
derives `m from hem or 'r from haar `her'.
This obviuosly
weakens our proposal with respect to Dutch pronouns. On the
other hand, Zwart also notes that Dutch weak pronouns are
more like strong pronouns in the sense that their movement
parallels object movement (scrambling) to a larger extent than
is the case with the Romance languages and Norwegian:
Dutch weak pronoun movement licences parasitic gaps,
(unlike Norwegian and Romance clitic movement) which
means, according to Zwart, that they first move as XPs to an
A'
-position (like scrambled DPs). Future research should
shed more light on the exact status of Dutch weak pronouns.
both are [+R], i.e. elements that are able to refer
independently to an object in the discourse.
(15) a. [La niñai sei vió [ ti bailar]].
+R -R
The girl saw herself dance.'
b. *[La niñai lai vió [ ti bailar]].
+R +R
`The girl saw her dance.'
The question we have to answer now is: why do
Spanish children allow a reflexive reading of (10c)?
We will argue that they can analyse la as both [+R] and
[-R] (= referentially defective). That is because they
overgeneralizing the [+R]/[-R] nature of first and
second person pronouns, which are [-R] when they are
in the tail of an A-Chain and [+R] otherwise.
(16) a Juan me ha visto.
(me = [+R])
`J. has seen me.'
b. Me he secado.
(me = [-R])
`I dried myself off.'
c. Los hombres te han visto. (te = [+R])
`The men have seen you.'
d. ¿Te has secado?
(te = [-R])
`Did you dry yourself off?'
The differences between the child system and the adult
system could be schematized as in (17).
(17) Adult Spanish
Child Spanish
me --> [+R] & [-R] [+R] & [-R] 1st p.
te --> [+R] & [-R] [+R] & [-R] 2nd p.
la --> [+R]
[+R] & [-R] 3rd p. fem.
This account is supported by cross-linguistic evidence.
French and Norwegian have a DPBE only in the
COMPLEX condition, and as expected 1st and 2nd
pers. pronouns are [+R]/[-R] in these languages
(Hamann & Philip, this volume; Philip & Hestvik, to
appear). Dutch has a DPBE in the SIMPLE condition,
but a much stronger DPBE in the COMPLEX
condition. As expected, Dutch 1st and 2nd pers.
pronouns are [+R]/[-R]. English, on the other hand, has
the same DPBE in the SIMPLE and COMPLEX
condition (Philip & Coopmans 1996). As expected,
English 1st and 2nd pers. pronouns are only [+R],
which means that no overgeneralization can take place.
(18) a. *I am drying me off.
b. *Did you dry you off?
The breakdown of Rule I is the only source of the
DPBE in English.
Ackowledgements
The research reported here has been supported partly
by the Spanish Ministery of Education (F.P.I.
Program), and partly by the International Office of the
Utrecht University. We wish to thank the children who
participated in this study, their parents and their
teachers, at the Colegio Aldeafuente in Madrid, and the
Colegio Pinoalbar in Valladolid, Spain. We also want
to thank Margarita España Villasante for her assistance
in running the experiments, and Peter Coopmans and
Denis Delfitto for their comments.
References
Berger, C. (1997) research paper `Acquisition of
Syntax,' Utrecht University.
Cardinaletti, A. & M. Starke (1995) `The Tripartition
of Pronouns and its Acquisition: Principle B Puzzles
are Ambiguity Problems,'NELS 25, 1-12.
Chien, Y.-C. & K. Wexler (1990) `Children's
Knowledge of Locality Conditions in Binding as
Evidence for the Modularity of Syntax and
Pragmatix,'Language Acquisition, 1, 225-295.
Delfitto, D. & N. Corver (1993) `Feature Asymmetry
and the Nature of Pronoun Movement', ms. Utrecht
University.
Hamann, C. & W. Philip (this volume) `The French
Delay of Principle B Effect', ms. Utrecht University.
Heim, I. (1993) `Anaphora and Semantic
Interpretation: A Reinterpretation of Reinhart's
Approach,' Seminar für Sprachwissenschaft,
Eberhard-Karls-Universität, Tübingen.
Grodzinsky, J. & T. Reinhart (1993) `The Innateness of
Binding and Coreference,'Linguistic Inquiry, 24, 69102.
McKee, C. (1992) `A Comparison of Pronouns and
Anaphors in Italian and English Acquisition,'
Language Acquisition, 2, 21-54.
Padilla, J. A. (1990) On the Definition of Binding
Domains in Spanish, Kluwer Academic Publishers,
Dordrecht
Philip, W. & P. Coopmans (1996) `The Role ofLexical
Feature Acquisition in the Development of
Pronominal Anaphora,'Amsterdam Series on Child
Language Development, vol. 5.
Philip W. & A. Hestvik (to appear) `Reflexivity, AntiSubject Orientation and Language Acquisition,
NELS 27.
Reinhart, T. & E. Reuland (1993) `Reflexivity,'
Linguistic Inquiry 24, 657-720.
Schroten, J. (1992) `On Spanish Definite Determiners:
Personal Pronouns and Definite Articles,'Recherches
de Linguistique Romane d'UtrechtXI.
Sportiche, D. (1992) `Clitic Constructions' ms. UCLA,
Los Angeles, Calif.
Uriagereka, J. (1995) `Aspects of the Syntax of Clitic
Placement in Western Romance,'Linguistic Inquiry,
26, 79-123.
Zwart, J-W (1992) `Notes on Clitics in Dutch,' paper
presented at the ESF Workshop on Clitics, Lund
May 22, Groningen University.
Zwicky, A.N. (1977) `On Clitics,' ms., Indiana
University.