Leveraging a Large Learner Corpus for Automatic Suggestion of Collocations for Learners of Japanese as a Second Language

Abstract

One of the challenges of learning Japanese as a Second Language (JSL) is finding the appropriate word for a particular usage. To address this challenge, we developed a collocational aid designed to suggest more appropriate collocations in Japanese. In particular, we address the problem of generating and ranking noun and verb candidates for correcting potential collocation errors in the learners’ text. Given a noun-verb construction as input, our system generates possible noun or verb correction candidates based on noun and verb corrections extracted from a large Japanese learner corpus. We use this corpus to investigate the learner's tendency to commit collocation errors, and to produce a smaller and more realistic set of candidates. After combining nouns or verbs with the generated candidates to form noun-verb pairs, the system uses the Weighted Dice coefficient as the association measure to filter out inappropriate noun-verb pairs and rank the proper collocations. We report the detailed evaluation and results on learner data. In addition, we show that our system statistically outperforms existing approaches to collocation error correction. Finally, we report a preliminary user study with JSL learners.

Technology and Vocabulary

“…Examples of innovative glossing are AR glosses (Ibrahim et al, 2018) and hyperlinks to corpus concordance lines with collocations. In the latter method, inappropriate collocations in student writing are automatically highlighted and a number of suitable collocations from linked corpora suggested (Frankenberg-Garcia et al, 2019;Pereira et al, 2016). Another emerging way to scaffold data-driven learning of collocations is through teacher-and learner-friendly corpus guides and online exercises (e.g., Poole, 2019;Vyatkina, 2020b).…”