Trial of Sentence Alignment on Professionally Simplified Articles for Japanese Text Simplification

Abstract

Text simplification is the task that transforms complex text into simpler one while preserving its meaning. Previous parallel corpora in Japanese for this task are constructed from simple text written by students or crowd workers, and there are no parallel corpora written by experts. We are constructing a parallel corpus by manually aligning sentences in articles that have been simplified by experts. This paper reports on the analysis of a part of our corpus that has been constructed at present. The human evaluation showed that our corpus contains a variety of simplification operations and consists of simpler sentences than the existing ones.