Semantic predicate-argument relations reflected in syntax

Question

I am a computational linguist and work on event extraction. One of the subtasks is very similar to finding predicate-argument relations, e.g. the agent and patient of a predicate. I want to build a system that heavily relies on syntactic dependencies for this subtask.

However, some papers I read conclude that predicate-argument relations cannot be reliable accessed via syntax because, e.g., the agent is not always the subject and the patient not always the object. Unfortunately, these papers only scrape the surface of the subject.

Are syntactic dependencies the best way to look for predicate-argument relations? Are there any books or papers dedicate to the interaction of predicate-argument relations and syntax, maybe investigating interactions of verb semantics and syntax?

Answer 1

To achieve this you need dependencies at deep structure (stripped of all the function words) and lexical mapping. You need a good parser and a valency lexicon with mapping to semantic/θ roles, which is sufficient to produce argument structures and semantic representations.