The Integration of Syntax and Semantic Plausibility in a Wide-Coverage Model of Human Sentence Processing

Important properties of the human sentence processor are The semantic model approximates world knowledge by exploiting the link between plausibility and word co-occurrence in a corpus annotated with thematic roles. Padó et al., 2006 Given an arbitrary verb-argument pair, it predicts a preferred role relation and its plausibility. Plausibility is equated to probability of encountering the pair in the respective relation in the FrameNet corpus. Smoothing sparse data to achieve wide coverage:- Semantic generalization: Pooling observations of words from the same semantic class Nouns: WordNet synsets Verbs: classes automatically induced from FrameNet- Re-estimation smoothing: Good-Turing smoothing The predictions of this widecoverage semantic model are significantly correlated to human plausibility judgment