Simulating the cross-linguistic development of optional infinitive errors in MOSAIC.

Abstract

The Optional Infinitive (OI) phenomenon in children’s speech has attracted a great deal of attention due to its occurrence in a variety of languages (including English, Dutch and German), and its apparent absence in other languages (such as Spanish and Italian). Wexler (1998) explains this pattern of results in terms of a Unique Checking Constraint that interacts with cross-linguistic differences in the underlying grammar to result in Optional Infinitive errors in obligatory subject languages (which require double-checking), but not in prodrop languages (which do not require double-checking). While Wexler’s account explains the cross-linguistic data, it attributes a great deal of innate linguistic knowledge to the child, and ignores the possibility that the cross-linguistic data may be equally well explained by the interaction between a simple distributional learning mechanism and the surface characteristics of the language. This paper presents simulations of the Optional Infinitive phenomenon across 4 languages (English, Dutch, German, and Spanish) using MOSAIC, a simple distributional analyser with no built-in syntactic knowledge. MOSAIC clearly simulates the different rates of Optional Infinitive errors across the languages, suggesting (a) that it is possible to explain the basic OI phenomenon without assuming large amounts of innate linguistic knowledge, and (b) that cross-linguistic differences in the OI phenomenon may be related to differences in the surface characteristics of the languages being learned.