コーパスを活用したシノニム・語法研究におけるMI-score の役割

The strategy on the statistics in corpus linguistics has been developing with the evolution of lexicography. This article is intended to show that the MI-score, which is often criticized for having a tendency to overestimate low-frequency pairs that are too specific to a particular corpus and may not generalize very well, is one of the most appropriate solutions when you try to identify interesting associations among words in corpora, especially when you would like to obtain interesting adverbials associated with verbs and adjectives.