Compound Words Analysis and Inferences of Japanese Clinical Texts

Abstract

To realize a medical case retrieval using ccg2lambda, which is an integrated system that maps a sentence to a higher-order logical formula by syntactic analysis and semantic composition based on Combinatory Categorial Grammar (CCG), we should consider how to obtain semantic representations of compound words in clinical texts. In this paper, we modify the compound word analysis module of ccg2lambda and provide a logical inference system for a medical case retrieval. The compound word analysis module assigns semantic tags to the constituents of compound words using sequence labeling models. Based on this information, we derive the CCG parse trees of compound words and obtain their semantic representations that allow logical inference. In addition, we create an inference test set involving compound words with critical texts. The experiments showed that our inference system outperformed the BERT NLI model on the inference test set involving compound words.