The hierarchical organization of predicate frames for interpretive mapping in natural language processing
Teruko Mitamura, Lorraine S. Levin · University of Pittsburgh eBooks · 1989
The purpose of this thesis is to develop a methodology for constructing a hierarchical organization of predicate frames for interpretive mapping in knowledge-based machine translation. Interpretive mapping refers to the mapping between predicate conceptual structures and syntactic structures, and it involves two kinds of processes. One is a mapping between grammatical functions, such as subject and object, and semantic roles in the conceptual structures, such as agent and theme. The second is a mapping between words, such as naguru 'hit' and lexical conceptual frames, such as *HIT. The thesis shows that the systematic study of lexical semantics contributes to the construction of lexical structures and interpretive mapping rules. Lexical semantics provides insight into systematic correspondences between syntax and semantics. We focus on Japanese verbs and classify them based on their common syntactic behavior, giving special attention to case alternations. Then we utilize knowledge representation techniques and inheritance mechanisms developed by artificial intelligence researchers to organize predicate frames hierarchically for interpretive mapping. The hierarchical organization of predicate frames can be incorporated into a system that does natural language processing, such as parsing a sentence into some meaning representation or generating a sentence from a meaning representation. The core set of Japanese predicate frames constructed in this thesis can be further developed by the addition of new words in the hierarchy. We expect that the lexical acquisition process can be done fairly automatically, since the mapping hierarchy has already been developed and can be used in various domains. In addition, the proposed methodology for organizing interpretive mapping rules would be helpful in the construction of predicate frames for other languages. In knowledge-based machine translation, syntactic structures and conceptual structures are combined to produce an intermediate semantic representation of the source language text. In order to combine these two structures, we need mapping rules to unite them. The main contribution of this thesis is to give a full-fledged model of interpretive mapping for knowledge-based machine translation, and to offer guidance that makes it easier to develop new translation domains.