Meter based omission of function words in MOSAIC

Abstract

MOSAIC (Model of Syntax Acquisition in Children) is augmented with a new mechanism that allows for the omission of unstressed function words based on the prosodic structure of the utterance in which they occur. The mechanism allows MOSAIC to omit elements from multiple locations in a target utterance, which it was previously unable to do. It is shown that, although the new mechanism results in Optional Infinitive errors when run on children’s input, it is insufficient to simulate the high rate OI errors in children’s speech unless combined with MOSAIC’s edge-first learning mechanism. It is also shown that the addition of the new mechanism does not adversely affect MOSAIC’s fit to the Optional Infinitive phenomenon. The mechanism does, however, make MOSAIC’s output more child-like, both in terms of the range of utterances it can simulate, and the level and type of determiner omission that the model displays.