(Beginning to) Model Semantic
Processing
Ulrike Baldewein
Probabilistic models of language
processing to date have mostly focused on syntactic
effects. However, human sentence processing has the goal of
understanding the meaning of an utterance and therefore,
semantics should not be neglected in modelling human language
processing. I propose a model that takes both syntactic and
semantic processing into account by adding a semantic module on
top of a standard model of syntactic processing. Early stages of
semantic processing will be modelled by the assignment of thematic
roles to a verb's arguments based on co-occurrence in a semantically
annotated corpus. I also outline a strategy for testing the
semantic model on its own, including approaches to dealing
with (sometimes crippling) data sparseness.