Understanding Implicit Language Structures

Abstract

We present a logic programming parsing methodology which is especially interesting for understanding implicit human language structures. It records parsing state constituents through linear assumptions to be consumed as the corresponding constituents materialize throughout the computation. Parsing state symbols corresponding to implicit structures remain as undischarged assumptions, rather than blocking the computation as they would if they were subgoals in a query. They can then be used to glean the meaning of elided structures, with the aid of parallel structures. Word ordering inferences are made not from symbol contiguity as in DCGs, but from invisibly handling numbered edges as parameters of each symbol. We illustrate these ideas through a metagrammatical treatment of coordination, and contrast them with constraint-based approaches, both within and outside Chomskyan-like frameworks of grammar.