Delage et al. 2008

6 downloads 0 Views 72KB Size Report
event argument, event readings occur with stage-level predicates only. This means that ..... Romance Child's Production,” Probus 15, 47-89. Guasti, M.T. and U.

RELATIVE CLAUSES IN ATYPICAL ACQUISITION OF FRENCH HELENE DELAGE, CECILE MONJAUZE, CORNELIA HAMANN & LAURIE TULLER 1. Introduction Much current research on typical and atypical language acquisition focuses on the notion of computational complexity as an explanation for findings such as order of acquisition and/or error rates and avoidance strategies for certain constructions. In this perspective, it has been proposed that maturational constraints restrict the computations necessary for complex linguistic derivations (Rizzi, 2000; Jakubowicz, 2005; Wexler, 1998, etc.) with the result that less complex structures are used. In this context, it is commonly proposed that movement is one of the key components of computational complexity (Clark & Roberts, 1993; Jakubowicz, 2005; Hamann, 2006). On the basis of a comparative study of atypical acquisition of French relative clauses in older children (age 9 to 13), we argue that, apart from movement, other factors such as distance (Jakubowicz, 2005; Gibson, 1998; Hamann et al., 2007; Tuller et al., 2006 etc.) and, crucially, depth of embedding play a role for complexity. In earlier studies of the acquisition of relative clauses by children acquiring French it has been discussed whether early production of relatives involves WH-movement or not (Labelle, 1990; Guasti & Cardinaletti, 2003; Guasti & Shlonsky, 1995; Pérez-Leroux 1995). Although we would agree that WH-movement is part of French child grammar, we suggest that it is avoided in early stages (see Jakubowicz 2005, Hamann et al. 2007, etc.). Moreover, several studies have observed an order in the acquisition of relative clauses: besides the early production of relatives with resumptive pronouns or resumptive DPs, we find juxtaposition of root clauses, production of pseudo-relatives (clefts and presentational c’est), and production of subject relatives before object relatives (De Cat, 2002; Labelle, 1990). Other relatives (which we will refer to as “genuine” relatives), including relatives with pied-piping, are mastered later (Labelle, 1990; Guasti & Cardinaletti, 2003). Many

2

Relative Clauses in Atypical Acquisition of French

accounts have proposed in which the subject/object asymmetry has to do with antecedent-gap distance, best described by movement, which also is at the base of an explanation in terms of non-canonical argument position (Gibson, 1998). Concerning clefts, which are also predominant in French children’s early relatives, in earlier work we adopted the analysis proposed by Clech-Darbon et al. (1999) and which was discussed in the context of typical acquisition by De Cat (2002), see Tuller et al 2006, Hamann et al 2007. We pointed out that this analysis involves a flatter structure than the one involved in genuine relatives. We argued that embedding is shallower in clefts, which makes them less complex. We would like to propose here that the notion of pseudo-relatives, the term frequently used for these early relatives, encompasses not only clefts and presentational c’est, discussed by Labelle and others, but also 0-level relatives, dislocated relatives and presentational y’a. Indeed, in previous work, our study of relatives led us to describe other types of relatives as “less complex”. We were struck by the high rates of 0-level relatives and what appeared to be existential relatives, but which semantically or pragmatically seemed to be analogous to presentational c’est. We specifically turn to these and to dislocated relatives in this study. 0-level relatives, the term we will use for relative clauses which are not embedded within a matrix clause (but may be embedded in a DP), by definition, involve no embedding within IP (1); absence of embedding within IP also characterizes dislocated relatives (2). Genuine relatives involve deeper embedding: a CP inside a DP which is inside an IP (3). (1)

Little brother

(2)

who

is crying

The friend who was with Harry, he’s dead

(3) [IP I have heard [DP a boy [CP who was crying]]]. Finally, some so-called existential relatives introduced by y’a, which we had intuitively grouped together with presentational structures, have received an analysis in the literature in which there is in fact no relative clause. To our knowledge, Lambrecht (1988) was the first to talk about what we will call presentational y’a constructions in spoken French. He

Hélène Delage, Cécile Monjauze, Cornélia Hamann & Laurie Tuller

3

showed that these have an event interpretation, rather than an individual interpretation, and suggested that they do not involve a subordinate clause modifying an NP. Côté (1999) observed that the constituent following y’a has particular syntactic as well as semantic properties: in addition to the possible event reading (4), arguments may be extracted from this constituent (5) in apparent violation of subjacency, and quantifiers embedded in this constituent can receive wide scope (6): (4) (5) (6)

Y a un employé qui inspecte une maison There’s an employee inspecting a house V’là la maison qu’y a un gars qui a fait exploser There’s the house that there’s a guy who made the house explode Y a un gars qui a inspecté chaque maison There’s a guy that is inspecting every house

All of these properties are taken to motivate a structure in which the constituent following y’a is in fact a clause with the DP landing in SPEC after movement from IP (7). Since only stage-level predicates have an event argument, event readings occur with stage-level predicates only. This means that only stage-level predicates can appear in the construction. Moreover, since this analysis requires movement of the DP, non-subjects are excluded by relativized minimality. We have found that the arguments motivating this analysis also hold for the presentational c’est construction. Notice also that this analysis does not involve a relative operator (linking operator), so that the prevalence of these relatives in child language might be taken as support for analyses in which early relatives do not involve relative operators (Wexler 1991, Guasti & Shlonsky 1995). (7) CP C' IP

ya

un gars qui

t

est venu

Summarizing, the structure of pseudo-relatives differs from the structure of genuine relatives, and this difference involves depth of embedding. Whereas genuine relatives involve embedding of a subordinate clause inside a DP, which is itself inside an IP, none of the

4

Relative Clauses in Atypical Acquisition of French

constructions we have classified as pseudo-relatives contain subordinate clauses embedded in this way. If our analysis is on the right track, this means that the order of acquisition facts that we have described above can be interpreted in the following way. Children start out with shorter movement—and thus produce mostly subject relatives, and their first apparent relatives are constructions which involve less deeply embedded clauses—what we have called pseudo-relatives. In other words, in early stages, they avoid long movement (i.e. non-subject relatives), and they avoid computations which involve deeper embedding (i.e. genuine relatives). Our expectation is that an age effect will be found over TD children after age six in this regard, and that atypical children will show this avoidance pattern at higher rates and at older ages.

2. Method We looked at three different cases of atypical development (AD): Specific Language Impairment (SLI), Mild-to-Moderate Hearing Loss (MMHL) which corresponds to an average hearing loss in the range of 21 to 70 dB, and Rolandic Epilepsy (RE), the classic example of benign focal epilepsy. These three pathologies differ in interesting ways. Firstly, language impairment is very severe in SLI, but studies of children and adolescents with MMHL have found over-all moderate language impairment, with high inter-subject variability, and studies of RE have identified subtle, but real language impairment. Secondly, the source of language impairment in SLI is developmental, but not known at this time; it is sensorial in MMHL and neurological in RE, involving abnormal electrical discharges over the centro-temporal region. Lastly, SLI presumably affects children since birth and does not disappear. Children with MMHL typically have congenital hearing loss but HL is detected relatively late (around age five) and therefore these children have several crucial years of (unaided) degraded language input. RE begins in childhood between ages 3 and 13 and disappears before adulthood. Comparison of these three groups offers an interesting way to explore the nature and cause of language impairment (see also Tuller et al, to appear). AD groups consisted of 21 participants with SLI, 21 with MMHL, and 11 with RE. These were compared to 3 control groups of 12 6-yearolds, 12 8-year-olds, and 12 11-year-olds (see Table 1). Age in the AD groups ranges from 9 to 13 years and matches that of the TD-11 group.

Hélène Delage, Cécile Monjauze, Cornélia Hamann & Laurie Tuller

Groups TD TD-6 TD-8 TD-11 AD SLI MMHL RE

N 12 12 12 21 21 11

Sex 7M 5F 6M 6F 6M 6F 15 M 6 F 12 M 9 F 7M 4F

Mean Age 6;4 8;2 11;4 11;6 11;7 11;5

Age Range 6;1 - 6,7 7;9 - 8;7 11;1 - 11;9 9;2 - 13;7 9;3 - 13;3 9;1 - 13;11

5

MLU 7.0 7.6 7.9 6.3 6.7 7.9

Table 1: General characteristics of groups with typical development (TD) and groups with atypical development (AD)

Our study is based on an analysis of spontaneous language samples transcribed in CHAT and analysed by CLAN tools (McWhinney 2000). Notice that the MLU of TD-6 matches the MLU of the deaf group and is close to that of the children with SLI. The MLU of TD-11 matches that of the RE group. Moreover, we were interested in having several ages of TD controls in order to capture the developmental aspect of relatives in French, which up to now have been systematically studied up to age six.

3. Results First of all, the rate of embedding (=N of embedded clauses/ N of verbal utterances) increases with age in TD-children, notably between ages 6 and 11. In the AD groups, the embedding rate varies according to the severity of language impairment: the RE group shows more embedding than the MMHL group, which shows more embedding than the SLI group. If we compare AD and TD groups, the SLI group and the MMHL group produce significantly less subordination than the TD-11, whereas the RE group performs like the TD-11.

TD-6 TD-8 TD-11 SLI MMHL RE

Total N Utterances Total N Relatives 685 67 723 75 722 82 1342 108 1252 101 660 58

Mean N Relatives 5.1 5.4 6.5 4.1 4.1 4.8

Table 2: Number of utterances and relatives analysed, mean number of relative clauses

Focusing now on the production of relatives, we found that their frequency, with respect to other subordinate clauses, is basically the same

6

Relative Clauses in Atypical Acquisition of French

in all groups. However, the mean number of relatives produced in the SLI (U = 257, p < .05) and the MMHL (U = 67, p < .05) groups is significantly lower than in the age-matched TD-11 group (see Table 2). Interestingly, we found very few clear resumptive pronouns (between 0% and 5% depending on the group).1 Concerning the differences between subject and non-subject relatives, we found, as expected, that the latter increase with age in the TD groups, with TD-6 and TD-8 performing significantly lower than TD-11. Each of the three AD groups also produced fewer non-subject relatives than the TD-11 group did (see Graph 1). Notice also that a sizeable proportion of SLI (43%) and deaf (33%) participants didn’t produce any non-subject relatives at all, while all of the TD-11 subjects produced at least one. Mean Number of Non-Subject Relatives 5 4 3 2 1 0

2,9 1,4

1,3

TD-6

TD-8

TD-11

1,4

1,4

1,6

SLI

MMHL

RE

Graph 1: Mean number of non-subject relatives per group

Relatives were coded according to their structure: pseudo-relatives (which included clefts, presentational c’est/y’a, 0-level relatives and dislocated relatives) and genuine relatives. The proportion of genuine relatives improves significantly with age in the TD groups. The SLI and RE groups produced significantly fewer genuine relatives than the TD-11, whereas the deaf produced as much genuine relatives as the TD-8 and -11 groups (graph 2). As for pseudo-relatives, focusing on the case of presentational y’a, we observe a clear age effect in typical acquisition. In the AD groups, we see a split based on severity of language impairment: participants with SLI produced more than participants with MMHL and with RE (see Table 3). (Only presentational y’a involving stage-level predicates and relativization of subject, as in examples (8) and (9), were included.)

1

Notice however that in spoken French, due to elision of the subject pronoun, resumptive pronouns are ambiguous with the complementizer, as in (i). (i) j'aime bien les films [qui / qu’i(ls) sont bien] (MMHL 13;2) I like movies [that are good]

Hélène Delage, Cécile Monjauze, Cornélia Hamann & Laurie Tuller

7

N Genuine Relatives / N Relatives (including root relatives) 100 80 49,8

47,2

42,1

60 % 40

20,9

17,2

23

20 0 TD-6

TD-8

TD-11

SLI

MMHL

RE

Graph 2: Proportion of genuine relatives

TD-6 TD-8 TD-11 SLI MMHL RE

N Presentational y’a / N Relatives 32.5% 22.2% 12.9% 44.3% 21.3% 25%

< SLI (U = 56, p < .05) < SLI (U = 29.5, p < .001)

Table 3: Proportion of presentational y’a relatives clauses per groups

(8) Y a une roue qui se crève (SLI 11;11) There’s a tire that went flat (9) Y a des bonhommes qui nous parlent (RE 11;2) There are guys who are talking to us (10) Le dernier qu’on a fait (SLI 11;8) The last one we have made We hypothesized above that 0-level relatives, as in (10), be included in non-genuine relatives, given their structure. As predicted, the three AD groups produced more of this type of unembedded relative clauses than the TD-11 group did. Besides pseudo-relatives, we also looked at cases in which a relative seems to have been avoided, either via juxtaposition of root clauses, as in (11) or via self-interruption of the relative as in (12). We found both of these in AD groups (SLI: 21 occurrences, 7.3%2; MMHL: 15 occurrences, 6.1%; RE: 14 occurrences, 7.7%), but much less frequently in the TD groups (TD-6: 6 occurrences, 1.2%; TD-8: 3 occurrences, 1.5%; TD-11: 6 occurrences, 2.2%). 2

Rate = N of occurrences of juxtaposition and self-interruptions / N of embedded clauses.

8

Relative Clauses in Atypical Acquisition of French

(11) Y a un loup garou il a mangé la fille (MMHL 12;6) There is a werewolf it ate the girl (12) Et c’est une personne qui dit euh … (SLI 11;6) And it’s a person who says uh… Not surprisingly, relative clauses produced in AD groups occur more frequently in erroneous utterances than those produced in TD groups (Graph 3).3 The MMHL group is particularly noteworthy in this regard: although their rate of genuine relatives is high (Graph 2), their error rate is also very high. % Relatives in Erroneous Utterrances / N Relatives

100 80 %

38,4

60

28,8

40 20

6,2

17,8

9,2 1,2

0 TD-6

TD-8

TD-11

SLI

MMHL

RE

Graph 3: Proportion of relatives in erroneous utterances

4. Conclusion Three principle results emerge from this study. The first point we would like to underline concerns the age effect observed in TD children. Not only do they continue to produce more relatives of any kind with age, but they produce more non-subject and genuine relatives with age,4 and thus, for example, the predominant use of presentational y’a decreases. This is particularly noticeable in the 11-year-olds. Next, as expected, rates of avoidance as defined by use of less complex constructions (such as juxtaposition, self-interruption, 0-level relatives, presentational constructions, clefts and dislocated relatives) are higher in the AD groups. Finally, in the AD groups, the general trend is that rates of avoidance mirror severity of language impairment: this is especially noticeable for presentational y’a and for 0-level relatives. Indeed, the SLI group displays both more avoidance of all complex relatives and, at the same time, the highest error rates. The RE group, which is known to have subtle language 3

All differences are significant. For example, SLI > TD-8 (U = 58.5, p < .05), MMHL > TD-8 (U = 65, p < .05), RE > TD-8 (U = 35.5, p < .05). 4 Gayraud & Martinie (2004) found similar results in a study of spontaneous language of TD children (9-year-olds, 12-year-olds, 16-year-olds) and adults.

Hélène Delage, Cécile Monjauze, Cornélia Hamann & Laurie Tuller

9

impairment, performs like TD-11 in their over-all use of subordination, but like the AD groups on measures of avoidance and between the TD groups and AD groups for rate of error. The deaf group appears to resemble the SLI group on many measures, notably the error rate, but differs in their rate of genuine relatives, which is similar to TD-11. This inter-pathology comparison raises interesting perspectives about the nature of language impairment in these three groups (see Monjauze, 2007, for RE and Delage & Tuller, 2007, for MMHL). In conclusion, we wish to emphasize the link between computational complexity and the acquisition of relatives in typical and atypical contexts. We consider complexity to be a linguistic concept which refers both to syntactic movement and to depth of embedding. We hope to have shown that surface relatives whose structure can be argued to involve less embedding, (compared to ordinary relative clauses) are predominant in early stages of typical language development and in some stages of AD language development. We have also shown that these structures do involve movement, but are flatter, and in particular do not involve embedding of a CP inside a DP, which is itself inside an IP. If computational complexity has an effect on syntactic working memory, flatter structures of this type would mean that computation is easier in that determining which constituents are connected to a particular IP can proceed faster.

References Clark, R. and I. Roberts (1993) “A Computational Model of Language Learnability and Language Change,” Linguistic Inquiry 24, 299-345. Clech-Darbon, A., G. Rebuschi and A. Rialland (1999) “Are there cleft sentences in French?,” in: G. Rebuschi and L. Tuller (Eds), The Grammar of Focus, 83-118. Amsterdam-Philadelphia: John Benjamins. Côté, M. H. (1999) “Quantification existentielle sur des événements et structure des pseudorelatives,” in: F. Corblin, C. Dobrovie-Sorin and J.-M. Marandin (Eds), Empirical Issues in Formal Syntax and Semantics 2, 169-190. The Hague: Thesus. De Cat, C. (2002) French Dislocation, PhD Dissertation, University of York. Delage, H. and L. Tuller (2007) “Language Development and Mild-toModerate Hearing Loss: Does Language Normalize with Age?” Journal of Speech, Language, and Hearing Research 50, 1300-1313. Gayraud, F. and B. Martinie (2004) Complexité syntaxique et difficulté de traitement : l’exemple des relatives. In : A. Rabatel (Ed). Interactions orales en contexte didactique. Lyon : Presses universitaires de Lyon.

10

Relative Clauses in Atypical Acquisition of French

Gibson, E. (1998) “Linguistic Complexity: Locality of Syntactic Dependencies,” Cognition 68, 1-76. Guasti , M. T. and A. Cardinaletti (2003) “Relative Clause Formation in Romance Child’s Production,” Probus 15, 47-89. Guasti, M.T. and U. Shlonsky (1995) “The Acquisition of French Relative Clauses Reconsidered,” Language Acquisition 4, 257-276. Hamann, C. (2006) “Speculations about Early Syntax: The production of Wh-questions by normally developing French children and French children with SLI,” Catalan Journal of Linguistics 5, 143-189. Hamann, C., L. Tuller, H. Delage, C. Monjauze and C. Henry (2007) “(Un)successful subordination in French-speaking children and adolescents with SLI” in Proceedings of BUCLD 1, 286-297. Jakubowicz, C. (2005) “The Language Faculty: (Ab)normal Development and Interface Constraints,” Paper presented at the GALA conference, University of Sienna, September 8-10. Labelle, M. (1990) “Predication, WH-Movement, and the Development of Relative Clauses,” Language Acquisition 1, 95-119. Lambrecht, K. (1988) “Presentational cleft constructions in spoken French,” in: J. Haiman and S. Thompson (Eds), Clause Combining in Grammar and Discourse, Typological Studies in Language, 18, 135179, Amsterdam-Philadelphia. MacWhinney, B. (2000). The Childes project: Tools for Analyzing Talk, 3e éd., Mahwah, NJ, Lawrence Erlbaum. Monjauze, C. (2007) Language and Epilepsy with centro-temporal spikes: deficits and long-term effects. PhD dissertation, University of Tours, France. Pérez-Leroux, A. T. (1995) “Resumptives in the acquisition of relative clauses,” Language Acquisition, 4, 105-138. Rizzi, L. (2000) “Remark on Early Null Subjects,” in: M.A. Friedeman and L. Rizzi (Eds.), The acquisition of syntax, 269-292. London: Addison-Wesley Longman. Tuller, L., H. Delage and C. Monjauze (2006) “Avoiding Complexity in AD Development of French,” Paper presented at the Latsis Colloquium, University of Geneva, January 26-28. Tuller, L., H. Delage and C. Monjauze (to appear) “Clitic pronoun production as a measure of atypical language development in French”. Lingua. Wexler, K. (1991) “Some issues in the growth of Control” Occasional Paper #44, Department of Brain and Cognitive Science, MIT. Wexler, K. (1998) “Very Early Parameter Setting and the Unique Checking Constraint: A New Explanation for the Optional Infinitive Stage,” Lingua, 106, 23-79.