コロキアムB発表

日時: 9月26日(木)4限(15:10~16:40)


会場: L2

司会: 久保尋之
WANG YIRAN D, 中間発表 自然言語処理学 松本 裕治, 中村 哲, 新保 仁, 進藤 裕之
title: Contextual Label-Aware Token Representations for Sequential Labeling
abstract: We present a novel method to construct contextual label-aware token representations for sequential labeling by utilizing a fusion layer and applying weight tying trick. Contextual label-aware token representations can selectively forget irrelevant information and thus are more informative to downstream layers. Experimental results on both named entity recognition (NER) and syntactic chunking demonstrate the performance of many existing models can be improved by applying our method.
language of the presentation: English