Bridging the Gap Between BabelNet and HowNet: Unsupervised Sense Alignment and Sememe Prediction

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 2789–2798 May 2-6, 2023 ©2023 Association for Computational Linguistics Bridging the Gap Between BabelNet and HowNet: Unsupervised Sense Alignment and Sememe Prediction Xiang Zhang, Ning Shi, Bradley Hauer, Grzegorz Kondrak Alberta Machine Intelligence Institute (Amii), Department of Computing Science University of Alberta, Edmonton, Canada {xzhang23,ning.shi,bmhauer,gkondrak}@ualberta.ca Abstract As the minimum semantic units of natural languages, sememes can provide interpretable representations of concepts. Despite the widespread utilization of lexical resources for semantic tasks, the use of sememes is limited by a lack of available sememe knowledge bases. Recent efforts have been made to connect Ba- belNet with HowNet by automating sememe prediction. However, these methods depend on large manually annotated datasets. Instead, we propose to use sense alignment via a novel unsupervised and explainable method. Our method consists of four stages, each relaxing predefined constraints until a complete align- ment of BabelNet synsets to HowNet senses is achieved. Experimental results demonstrate the superiority of our unsupervised method over previous supervised ones by an improvement of 12% overall F1 score, setting a new state of the art. Our work is grounded in an interpretable propagation of sememe information between lexical resources, and may benefit downstream applications which can incorporate sememe in- formation. 1 Introduction Sememes are the minimum semantic units of hu- man languages (Bloomfield, 1926). The theory of semantic primitives (Wierzbicka, 1996a) hypothe- sizes that the meaning of a word in any language can be decomposed into a finite set of language- independent sememes. For example, the English noun plant has distinct senses that correspond to the “factory” and “vegetation” concepts, respectively. The former can be represented by the sememes “in- dustrial”, “produce”, and “institute/place”, and the latter by “crop”, “tree”, and “flower/grass”. (See Figure 2 for additional examples.) Although se- memes provide a way of representing concepts, their incorporation into natural language process- ing (NLP) has been limited by a lack of available sememe resources for commonly used sense in- ventories. Tremendous efforts have been made to BabelNet Synset HowNet Senses Stage 1 Stage 2 Stage 3 Stage 4 Synset: bn00067718n 公尺 米突 metre meter 米 Senses meter 米 metre 米 meter 公尺 Figure 1: Given a BabelNet synset as input, our alignment algorithm identifies a set of corresponding HowNet senses. The algorithm consists of four stages which progressively relax alignment constrains. construct sememe knowledge bases (KBs) manu- ally. One of the most widely used is HowNet (Dong and Dong, 2003), which unfortunately is limited to only two languages: English and Chinese. A related problem is linking lexical resources to one another. As manually creating semantic knowledge bases remains a challenging and costly process, BabelNet (Navigli and Ponzetto, 2012) was instead created by combining WordNet (Miller, 1995), Wikipedia, and other resources. While Ba- belNet covers hundreds of languages, it does not include sememe information. Previous work on au- tomatically predicting sememes for BabelNet con- cepts has depended on large human labeled data (Qi et al., …

Content truncated to 500 words