Incorporating defaulting effects into MOSAIC: Building a two-factor model of the Optional Infinitive stage

Daniel Freudenthal, Julian M. Pine and Fernand Gobet.

In several languages, children go through a stage during which they produce Optional Infinitive (OI) errors: non-finite verb forms in contexts in which a finite verb form is required. MOSAIC is a computational model of grammatical development that simulates the cross-linguistic patterning of OI errors by learning reduced compound finite structures from the input. However the current version of MOSAIC is unable to simulate the very high rate of OI errors in early child English.

In this talk, we investigate the possibility that the high rate of OI errors in English reflects the operation of two processes: 1) the learning of reduced compound finite structures and 2) defaulting to the most common form of the verb in the input (for most English verbs, the bare stem, but for most Spanish verbs, the 3sg present tense). We describe a version of MOSAIC in which the original mechanism for producing reduced modal structures is complemented by a mechanism that defaults to the most frequent form of the verb in utterance-final strings of increasing length in the input. We show that this version of the model provides a better fit to data from both English and Dutch without affecting the model’s previously good fit to Spanish. This result reflects the tendency of the model to show frequent and persistent defaulting to the bare form in English, a small amount of defaulting to the 3sg in Spanish, and defaulting to the infinitive at low MLUs, but to the bare stem at higher MLUs in Dutch. The model thus provides a plausible two-factor model of the OI stage, which captures cross-linguistic differences in the patterning of both OI and defaulting errors. It also has the potential to serve as the basis for a model of verb-marking errors in children with SLI.