[4Pin1] インタラクティブ(2)

[4Pin1-52] Dataset Construction Method for Word Reading Disambiguation

〇Koki Nishiyama¹, Kazuhide Yamamoto¹, Hideharu Nakajima² (1. Nagaoka University of Technology, 2. NTT Media Intelligence Laboratories)

Keywords:Word Reading Disambiguation, Data Construction Method, Crowdsourcing

We propose a data construction method for word reading disambiguation. The method gives unique reading word to each reading of reading ambiguous word, collects sentences including the unique word, replaces the unique word in sentences to the original ambiguous word and tags readings of reading ambiguous words to the reading corresponding to the unique word. Through experiments, we confirmed the method collects data numerically balanced between readings.