Development of Evidence-Based Grammars for Terminology Extraction in OneClick Terms

Abstract

This paper presents a new generation of terminology extraction grammars for the OneClick Terms system. Unlike previous grammars built using linguistic judgment, the new grammars use rules inspired by patterns frequently observed in existing terminology databases. This evidence-based approach leads to a more accurate coverage of term candidates of lexical structures typical for authentic terms. The internal variety and maximum length of recognized terms have also increased. Due to the use of techniques known from corpus linguistics in their design, the resulting grammars maximize the coverage of terms while keeping a manageable size.
In the paper, we first describe how term grammars are used in OneClick Terms (Baisa et al., 2017) to enable terminology extraction for individual languages. Then we explain the procedure which we use to design next-generation term grammars for seven languages (English, Estonian, French, German, Italian, Portuguese, Spanish). This includes studying the IATE term base (Zorrilla-Agut & Fontenelle, 2019) to derive information on the typical structure of terms in each language. Eventually, we provide figures concerning the new term grammars and their recall of the IATE terms, and we discuss directions for further development.