Paper presented at the Fourteenth Annual CUNY Conference on Human Sentence Processing, University of Pennsylvania, Philadelphia, PA.

The Role of Phonological and Prosodic Cues in the Acquisition of Syntax: Multiple-Cue Integration in a Connectionist Model

Abstract

Children acquire the syntactic structure of their native language with remarkable speed and reliability.  This achievement is especially impressive because the child faces a difficult "bootstrapping" problem.  Discovering syntactic constraints requires knowledge about grammatical categories; but, conversely, grammatical categories are only useful for acquisition insofar as they support syntactic constraints. Recent work in developmental psycholinguistics suggests that children may bootstrap grammatical categories and basic syntactic structure by exploiting distributional, phonological, and prosodic cues. However, these cues are probabilistic, and are individually unreliable. In this talk, we present a series of simulations demonstrating that the integration of multiple probabilistic cues in a connectionist model results in significantly better, faster, and more uniform acquisition of syntax.

                A phrase-structure grammar was constructed based on previous research and independent corpus analyses (see Table 1). This fairly complex grammar was intended to capture the general syntactic trends prevalent in child-directed speech, including declarative, imperative, and interrogative sentences. Five groups of ten simple recurrent networks were trained on 10,000 sentences generated from this grammar. Each group of networks integrated a different constellation of three partially reliable cues: length (in syllables), stress, and pitch. A base network relied entirely on distributional information, with cues added to the remaining groups, including a three-cue network that took advantage of all three cues. The networks were composed of 45 lexical input/output units, and additional units for cues. The hidden/context layers contained 80 units.

                Network performance was measured in terms of the error on predicting the next set of correct grammatical categories given prior context. All networks achieved better performance than standard bigram/trigram models (p's < .0001). The nets provided with phonological/prosodic cues achieved significantly better performance than base networks (p's < .02). Using trigram performance as criterion, all multiple-cue networks surpassed this level of performance faster than the base networks (p's < .002). Moreover, the three-cue networks were significantly faster than the single-cue networks (p's < .001). The three-cue networks also exhibited significantly more uniform learning than the base networks (p < .03). Additional simulations showed a positive effect of "prenatal" exposure to the kind of gross-level prosodic information available in the womb. A final set of simulations demonstrated that learning was not negatively affected by the presence of distractor cues.

                We then applied the three-cue networks to the modeling of recent data showing that two-year-olds can integrate grammatical markers (function words) and prosodic cues in sentence comprehension [1]. The children heard sentences, such as (1), in one of three prosodic conditions depending on pause location: early natural [e], late natural [l], and unnatural [u]. Each sentence moreover involved one of three grammatical markers: grammatical (the), ungrammatical (was), and nonsense (gub). Adjusting for vocabulary differences, the networks were tested on comparable sentences, such as (2). As the children, the networks showed effects of both prosody and grammatical marker (p's < .0001), with similar differences between conditions as in the original study.

1.                   Find [e] the/was/gub [u] dog [l] for me.

2.                   Where does [e] the/is/gub [u] dog [l] eat?

This series of simulations underscores the computational feasibility of the multiple-cue approach to syntax acquisition. In concluding, we consider implications for the role of phonological and prosodic cues in adult sentence processing.

       Table 1: The Grammar Used to Generate the Corpora of Child-Directed Speech      

S ® Imperative [0.1] | Interrogative [0.3] | Declarative [0.6]

Declarative ® NP VP [0.7] | NP-ADJ [0.1] | That-NP [0.075] | You-P [0.125]

            NP-ADJ ® NP is/are adjective

            That-NP ® that/those is/are NP

            You-P ® you are NP

Imperative ® VP

Interrogative ® Wh-Question [0.65] | Aux-Question [0.35]

            Wh-Question ® where/who/what is/are NP [0.5] |  where/who/what do/does NP VP [0.5]

            Aux-Question ® do/does NP VP [0.33] | do/does NP wanna VP [0.33] | is/are NP adjective [0.34]

NP ® a/the N-sing/N-plur

VP ® V-int | V-trans NP

References:

 [1] Shady, M., & Gerken, L.A. (1999). Grammatical and caregiver cues in early sentence comprehension. Journal of Child Language, 26, 163-175.

Home | People | Research | Links | Contact | Publications | Presentations