의미 관계의 교차 언어 반교사 학습

Abstract

Relation extraction is the task of identifying semantic relationships between named entities in natural language documents. Many researchers have conducted extensive studies on relation extraction in the last decade

however, statistical systems based on supervised learning are still limited because they require large amounts of training data to achieve a high performance level.This dissertation proposes cross-lingual annotation projection approaches for relation extraction. The main idea of this approach is to obtain training examples in a resource-poor language without significant annotation efforts by utilizing parallel corpora. The annotations in the resource-rich source language sentences are propagated to the corresponding target language sentences through the results of word alignment.To make our method more reliable, we introduce two projection approaches: direct projection and graph-based projection. Direct projection is presented with noise reduction strategies to improve robustness against errors generated by preprocessing. The graph-based projection approach adopts a label propagation algorithm to generate quality projections. We demonstrate the merit of our methods using a Korean relation extraction system trained on projected examples from an English-Korean parallel corpus. Experiments show the feasibility of our approach through a comparison to other systems based on monolingual resources.