Simulating Optional Infinitive Errors in Child Speech through the Omission of Sentence-Internal Elements

A new version of the MOSAIC model of syntax acquisition is presented. The modifications to the model aim to address two weaknesses in its earlier simulations of the Optional Infinitive phenomenon: an over-reliance on questions in the input as the source for Optional Infinitive errors, and the use of an utterance-final bias in learning (recency effect), without a corresponding utterance-initial bias (primacy effect). Where the old version only produced utterance-final phrases, the new version of MOSAIC learns from both the left and right edge of the utterance, and associates utterance-initial and utterancefinal phrases. The new model produces both utterance-final phrases and concatenations of utterance-final and utteranceinitial phrases. MOSAIC now also differentiates between phrases learned from declarative and interrogative input. It will be shown that the new version is capable of simulating the Optional Infinitive phenomenon in English and Dutch without relying on interrogative input. Unlike the previous version of MOSAIC, the new version is also capable of simulating cross-linguistic variation in the occurrence of Optional Infinitive errors in Wh-questions. The Characteristics of Early Child Speech Early child speech is often telegraphic, and (in many languages) lacks inflections that are required in the adult grammar. For example, English-speaking children produce utterances such as Play car and He go and Dutch-speaking children produce utterances such as Pappa eten (Daddy eat) and Trein spelen (Train play). As children grow older, the length of their utterances increases, their speech becomes less telegraphic, and they provide the appropriate inflections more frequently. However, there is a period in which children use verbs in both their correct (inflected) and incorrect (uninflected) forms in contexts in which inflected forms are required. The apparent lack of inflection in child speech has been the subject of considerable linguistic and Nativist theorizing. Wexler (1994) proposes the Optional Infinitive hypothesis, which states that young children know the full grammar of their language but optionally use nonfinite forms where the adult grammar requires a finite form. Wexler’s hypothesis explains the data from a variety of languages. However, there are two main weaknesses associated with the account. First, it fails to provide any quantitative predictions regarding the rate at which children will use nonfinite forms in finite contexts, and, second, it ignores the possibility that children’s early language use may reflect the operation of an input-driven learning mechanism as opposed to rich innate linguistic knowledge. Simulating Child Language in MOSAIC MOSAIC is an attempt to investigate the extent to which children’s early language use can be explained by an inputdriven learning mechanism. MOSAIC learns from ChildDirected speech and produces output that can be directly compared to children’s speech. MOSAIC has already been used to simulate the basic Optional Infinitive phenomenon in English and Dutch (Freudenthal, Pine & Gobet, 2002a, submitted), as well as phenomena related to Subject Omission (Freudenthal, Pine & Gobet, 2002b) and the Modal Reference effect (Freudenthal, Pine & Gobet, 2004). MOSAIC is a simple discrimination network that incrementally learns and stores utterances that are presented to it. An important restriction on MOSAIC’s learning mechanism is that it builds up its representation of an utterance by starting at the end of the utterance and slowly working its way to the beginning. MOSAIC is therefore capable of producing an utterance such as He go by producing the final phrase of Does he go? Similarly, it can produce the Dutch utterance Trein spelen by producing the ending of the phrase Ik wil met de trein spelen (I want with the train play). As MOSAIC sees more and more input, it learns to produce progressively longer utterances. As utterances become longer, they are more likely to contain finite verb forms. (Both in Dutch and English finite verb forms tend to occur near the beginning of the utterance. A model that produces utterance-final phrases will therefore 1 Data from languages like Dutch, which has an infinitival morpheme, suggest that, rather than dropping inflections, children are using non-finite verb forms in finite contexts.