Multilingualization of Medical Terminology: Semantic and Structural Embedding Approaches

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4157–4166 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4157 Multilingualization of Medical Terminology: Semantic and Structural Embedding Approaches Long-Huei Chen, Kyo Kageura The University of Tokyo Tokyo, Japan longhuei@g.ecc.u-tokyo.ac.jp, kyo@p.u-tokyo.ac.jp Abstract The multilingualization of terminology is an essential step in the translation pipeline, to ensure the correct transfer of domain-specific concepts. Many institutions and language service providers construct and maintain multilingual terminologies, which constitute important assets. However, the curation of such multilingual resources requires significant human effort; though automatic mul- tilingual term extraction methods have been proposed so far, they are of limited success as term translation cannot be satisfied by simply conveying meaning, but requires the terminologists and domain experts’ knowledge to fit the term within the existing terminology. Here we propose a method to encode the structural properties of terms by aligning their embeddings using graph convo- lutional networks trained from separate languages. The results show that the structural information can augment the standard bilingual lexicon induction methods, and that taking into account the structural nature of terminologies allows our method to produce better results. Keywords: Terminology, Multilinguality, Machine Learning, MultiWord Expressions, Translation Resources 1. Introduction Terminology occurs along the line between vocabularies and knowledge systems. While many regard terminolo- gies as specialized dictionaries that contain domain-specific terms, terminology exhibits characteristics one does not typically expect from a list of words. Specifically, we con- sider the prescriptive aspects of terminologies one cannot easily observe in the descriptive process of dictionary cre- ation (Felber, 1984; Kockaert and Steurs, 2015). As termi- nologies in any language are the depiction of the underlying knowledge system, the designation of a term must reflect the concept it represents in a consistent and accepted man- ner to facilitate expert communications (Sager, 1990). While many terminology processing techniques conflate text processing with terminology processing (Pazienza et al., 2005; Chung, 2003; Pe˜nas et al., 2001), we recognize that the unique nature of terminologies provides essential information that can contribute to more effective model- ing. We draw inspiration from the work of human termi- nologists, as they consult both the text and the terminol- ogy when conducting actions such as term identification or multilingualization. Domain experts and terminologists do not simply create terms based on their occurrence in the text, but understand that its designation must reflect its posi- tion within the domain- and language-specific terminology. This is related to the theoretical understanding that, linguis- tically, the formation of term candidates is not syntactic but lexical, and that term candidates turn into terms with their incorporation into the terminological system through a con- scious, social process (Kockaert and Steurs, 2015). We therefore hypothesize that domain-specific terminolog- ical information can improve the term multilingualization process. Specifically, terminologies exhibit a structural na- ture not typically expected in general vocabulary (Sager, 1990). We propose a method towards multilingualizing ter- minologies by taking into account the specialized concep- tual structure reflected in …

Content truncated to 500 words