A Connectionist Simulation of the Form-Meaning Relationship in Japanese Business Manuals

概要

論文の詳細を見る
For improving natural/computer language conversion algorisms, this paper proposes an account of the acquisition of semantic relations (form-meaning relationship) of Japanese complex sentences using the basic concepts of connectionism and a construction-based theory of grammar (Fillmore 1988), especially those parameters extensively asserted in Ohori (2001). Firstly, this paper proposes that semantic relation (form-meaning relationship) emerges in causal chain (Croft 1991) and tries to map each complex sentence to semantic relations (i.e., cause and reason). The motivation behind the network is the notion that merely the drive to map an input sequence of sentences to a semantic relations-output is sufficient to induce the necessary internal abstractions to facilitate the mapping. To test this hypothesis, a Simple Recurrent Network (Elman 1990) has been created and tested. This paper demonstrates the emergence of the notion of "meaning" in the neural network that successfully learns to map from "sequence of sentences (form)" to "semantic relations (meaning)".
2006-03-23