Incremental Grammar Induction from Child-Directed Dialogue Utterances

摘要

We describe a method for learning an incremental semantic grammar from data in which utterances are paired with logical forms representing their meaning. Working in an inherently incremental framework, Dynamic Syntax, we show how words can be associated with probabilistic procedures for the incremental projection of meaning, providing a grammar which can be used directly in incremental probabilistic parsing and generation. We test this on child-directed utterances from the CHILDES corpus, and show that it results in good coverage and semantic accuracy, without requiring annotation at the word level or any independent notion of syntax.