Chinese Lexical Sememe Prediction Using CilinE knowledge

摘要

Sememes are the smallest semantic units of human languages,the composition of which can represent the meaning of words. Sememeshave been successfully applied to many downstream applications innatural language processing (NLP) field. Annotation of a word’s sememesdepends on language experts, which is both time-consuming and laborconsuming,limiting the large-scale application of sememe. Researchershave proposed some sememe prediction methods to automatically predictsememes for words. However, existing sememe prediction methods focuson information of the word itself, ignoring the expert-annotated knowledgebases which indicate the relations between words and should value in sememepredication. Therefore, we aim at incorporating the expert-annotatedknowledge bases into sememe prediction process. To achieve that, wepropose a CilinE-guided sememe prediction model which employs an existingword knowledge base CilinE to remodel the sememe prediction fromrelational perspective. Experiments on HowNet, a widely used Chinesesememe knowledge base, have shown that CilinE has an obvious positiveeffect on sememe prediction. Furthermore, our proposed method can beintegrated into existing methods and significantly improves the predictionperformance. We will release the data and code to the public.