Automatic Discovery of Nativelike Selections through Topic Modeling: an Experiment Based on Online Reviews

Abstract

Nativelike selections are linguistic expressions favored by native speakers over other grammatically correct alternatives to express a certain concept. They are important items for language teaching and lexicography. This study proposes an automatic method for discovering them through the application of the topic modeling technique of BERTopic which discovers the frequent linguistic expressions that express the same topic. The method’s effectiveness and efficiency are demonstrated through an experiment based on online reviews of a dish soap product in English, Chinese, and Japanese.