2018 Sat Session C 1415

Saturday, November 3, 2018 | Session C, Terrace Lounge | 2:15pm

Beyond linear non-adjacent or adjacent dependencies: Infants track hierarchical syntactic dependencies
C. Legrand, R. Shi

Previous studies suggest that toddlers are sensitive to non-adjacent dependencies. English-learning infants discriminate sentences containing correct non-adjacent dependencies (e.g., is …_ing) from those containing dependency violations (e.g., can …_ing)(Santelmann & Jusczyk, 1998). French-learning infants track non-adjacent number dependencies between Determiner of the subject-NP and Auxiliary of the VP, e.g., “le-Det-singular … va-Aux- singular” (Van Heugten & Shi, 2010). Infants track non-adjacent dependencies when the variability of the intervening elements is high (Gomez, 2002), and the length of the intervening elements affects their performance (Santelmann & Jusczyk, 1998). Moreover, Höhle, et al (2006) found that toddlers’ success in detecting discontinuous verbal dependencies relied on the kinds of intervening structure. One important question remains: Do infants process non-adjacent dependencies at the hierarchical grammatical level, or simply sequentially at the linear surface level? The present study directly tested this question.

We created French sentences involving grammatical gender agreement between a left- tropicalized subject-NP and the subject-pronoun, following feature-percolation principles. In structure 1, the subject-NP contained a modifying prepositional phrase, e.g., La bananeFEM dans le chapeauMASC, elleFEM demeure au fleuve (“The banana in the hat, it stays in the river”):

[[Det N1 [Prep [Det N2]NP1]PP]NP2 [Pron]NP [V …]VP]S. Here the higher NP2 inherits the feminine gender of the head-noun banane. N2 chapeau under PP has no impact on the gender of NP2. NP2 is the category that matters for the feature dependency with the subject-pronoun.

Structure 2 contained conjoined NPs dominated by the higher NP3, e.g., La bananeFEM et le chapeauMASC, ilsMASC demeurent au fleuve (“The banana and the hat, they

…”): [[[Det N1]NP1 Conj [Det N2]NP2]NP3 [Pron]NP4 [V …]VP]S.

If N1 and N2 have the same gender, the higher NP3 inherits that gender. But here, N1 banane and N2 chapeau have different genders; the higher NP3 thus resorts to the default masculine feature. The subject-pronoun ils agrees with NP3 in gender.

In a preferential looking experiment French-learning 30-month-olds heard sentences in these two structures (Table 1). One group heard correct-agreement sentences containing a PP-modifier and incorrect-agreement sentences with conjoined subject-NPs, e.g.,

[La bananeFEM dans le chapeauMASC]FEM, elleFEM VP

*[La bananeFEM et le chapeauMASC]MASC, ellesFEM VP

Another group heard correct-agreement sentences containing conjoined subject-NPs and incorrect sentences with a PP-modifier, e.g.,

[La bananeFEM et le chapeauMASC]MASC, ilMASC VP

*[La bananeFEM dans le chapeauMASC]FEM, ilMASC VP

For both groups, dependency operates at the higher phrasal level, between subject- Pron and the preceding subject-NP. If infants processed gender agreement hierarchically, both groups should discriminate grammatical versus ungrammatical sentences, with a uniform looking preference. However, if they simply analyzed linear statistics, either adjacently (between subject-Pron and chapeau) or non-adjacently (between subj-Pron and banane), both groups should yield no discrimination.

Results: an ANOVA showed a significant discrimination of grammatical versus ungrammatical trials (F(1,22)=10223, p=.004), with no effect of group, nor group x grammaticality interaction. Thus, infants did not simply perform linear statistical computations of adjacent or non-adjacent dependencies. The results are a striking demonstration of infants’ knowledge of hierarchical syntactic structures and their understanding of dependencies as defined in the grammar.