Knowledge Representation for Lexical Semantics: Is Standard First Order Logic Enough?

Many natural language understanding tasks require lexicons that can support extensive lexically-based inferencing. To represent lexical semantics, most existing natural language understanding systems utilize Standard First Order Logic (FOL), or a subset of it. We argue that using a representation richer than FOL would facilitate better lexically-based inferencing. Using examples drawn from actual task-oriented spoken dialogues, we motivate the addition of restricted quantiication and non-standard quantiiers, modal operators, predicate modiication, and predicate nominalization. Although these additions are not new to the computational semantics literature, their motivation from the semantics of open class words is.