Evaluation of a method for automatic acquisition of semantic rules from a bracketed corpus using inductive learning

摘要

In this paper, we propose a method for automatic acquisition of semantic rules from a bracketed corpus using inductive learning, and evaluate a system based on this method. It is necessary for advanced natural language processing to use semantic information. In previous methods for automatic acquisition of semantic information, they require many patterns for detecting and deciding information, or need other semantic information such as deep cases, thesaurus, and so on. Our method acquires semantic rules, which include deep cases and thesaurus, from only sentences analyzed syntactically by unsupervised inductive learning. Our system learned from 200 sentences in the EDR corpus. As a result, a 35.3% of accuracy rate was achieved in 50 sentences as open data using the acquired semantic rules.