Automatic Construction of Sememe Knowledge Bases via Dictionaries

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 4673–4686 August 1–6, 2021. ©2021 Association for Computational Linguistics 4673 Automatic Construction of Sememe Knowledge Bases via Dictionaries Fanchao Qi1,2, Yangyi Chen2,4∗, Fengyu Wang1,2, Zhiyuan Liu1,2,3, Xiao Chen5, Maosong Sun1,2,3† 1Department of Computer Science and Technology, Tsinghua University, Beijing China 2Beijing National Research Center for Information Science and Technology 3Institute for Artificial Intelligence, Tsinghua University, Beijing, China 4Huazhong University of Science and Technology 5Huawei Noah’s Ark Lab qfc17@mails.tsinghua.edu.cn Abstract A sememe is defined as the minimum seman- tic unit in linguistics. Sememe knowledge bases (SKBs), which comprise words anno- tated with sememes, enable sememes to be ap- plied to natural language processing. So far a large body of research has showcased the unique advantages and effectiveness of SKBs in various tasks. However, most languages have no SKBs, and manual construction of SKBs is time-consuming and labor-intensive. To tackle this challenge, we propose a sim- ple and fully automatic method of building an SKB via an existing dictionary. We use this method to build an English SKB and a French SKB, and conduct comprehensive evaluations from both intrinsic and extrinsic perspectives. Experimental results demonstrate that the au- tomatically built English SKB is even supe- rior to HowNet, the most widely used SKB that takes decades to build manually. And both the English and French SKBs can bring obvious performance enhancement in multiple downstream tasks. All the code and data of this paper (except the copyrighted dictionaries) can be obtained at https://github.com/ thunlp/DictSKB. 1 Introduction A word is the smallest linguistic element that can be used on its own with a particular meaning, but not the smallest semantic unit (O’Grady et al., 1997). The meaning of a word can be divided into smaller components. In linguistics, a sememe is de- fined as the minimum semantic unit of human lan- guages (Bloomfield, 1926). Some linguists believe that meanings of all words can be expressed by a limited set of predefined sememes (Goddard and Wierzbicka, 1994). For example, the basic mean- ing of “boy” can be expressed by the compositions ∗Work done during internship at Tsinghua University †Corresponding author. Email: sms@tsinghua.edu.cn husband human family male spouse economize “married man” “carefully use” word sense sememe Figure 1: Sememe annotations of the word “husband” in HowNet. of human, male and child, while the meaning of “girl” can be expressed by human, female and child, where human, male, female and child are predefined sememes. It is even deemed that this sememe-based semantic system as well as the sememe set is universal among different lan- guages, in which case sememes are also named universal semantic primitives (Wierzbicka, 1996). Sememes are implicit in words and cannot be directly used in natural language processing (NLP). Dong and Dong (2006) make a seminal contribu- tion and put the sememe-based semantic system into practice. They define a set of about 2, 000 sememes and use them to annotate senses of over 100, 000 Chinese and English words, whereupon a sememe knowledge base …

Content truncated to 500 words