Assumption Grammars for Processing Natural Language
Verónica Dahl, Paul Tarau, Richard Z. Li · The MIT Press eBooks · 1997
In this paper we examine three natural language uses of a recently developed logic grammar formalism- Assumption Grammars- particularly suitable for hypothetical reasoning. They are based on intuitionistic and linear implications scoped over the current continuation, which allows us to follow given branches of the computation under hypotheses that disappear when and if backtracking takes place. We also show two results which were surprising to us, namely: a) Assumption grammars allow a direct and efficient implementation of link grammars--- a context-free like formalism developed independently from logic grammars; and b) they offer the flexibility of switching between data-driven or goal-driven reasoning, at no overhead in terms of either syntax or implementation. 1 Introduction A grammar is a finite way of specifying a language which may consist of an infinite number of sentences. A logic grammar has rules that can be represented as Horn clauses. Logic grammars can be conveniently im...