An Integrated Formal Representation for Terminological and Lexical Data included in Classification Schemes

An Integrated Formal Representation for Terminological and Lexical Data included in Classification Schemes Thierry Declerck1,2, Kseniya Egorova3, Eileen Schnur1 1DFKI GmbH, Stuhlsatzenhausweg 3, D-66123 Saarbr¨ucken, Germany 2Austrian Academy of Sciences, Sonnenfelsgasse 19,1010 Vienna, Austria 3Saint Petersburg State University of Architecture and Civil Engineering, 2-nd Krasnoarmeiskaya St. 4, 190005 St. Petersburg, Russia declerck@dfki.de, kseniya.a.egorova@gmail.com, eileen.schnur@dfki.de Abstract This paper presents our work dealing with a potential application in e-lexicography: the automatized creation of specialized multilingual dictionaries from structured data, which are available in the form of comparable multilingual classification schemes or taxonomies. As starting examples, we use comparable industry classification schemes, which frequently occur in the context of stock exchanges and business reports. Initially, we planned to follow an approach based on cross-taxonomies and cross-languages string mapping to automatically detect candidate multilingual dictionary entries for this specific domain. However, the need to first transform the comparable classification schemes into a shared formal representation language in order to be able to properly align their components before implementing the algorithms for the multilingual lexicon extraction soon became apparent. We opted for the SKOS-XL vocabulary for modelling the multilingual terminological part of the comparable taxonomies and for OntoLex-Lemon for modelling the multilingual lexical entries which can be extracted from the original data. In this paper, we present the suggested modelling architecture, which demonstrates how terminological elements and lexical items can be formally integrated and explicitly cross-linked in the context of the Linguistic Linked Open Data (LLOD). Keywords: SKOS-XL, OntoLex-Lemon, Terminology, Lexicography 1. Introduction The topic of extracting dictionaries from raw data was dis- cussed in the context of the recently terminated European Network of e-Lexicography (ENeL) COST Action1, and it was also the motto of the fifth biennial eLex conference2 called “eLex 2017: Lexicography from scratch”. The work presented in this paper takes its source within this context, meaning in detail that we aim to investigate the automated extraction of domain-specific multilingual lexicons from comparable classification schemes or taxonomies. As a dataset for our investigation, we selected two compa- rable industry classification schemes that are used in var- ious stock exchanges like Euronext, the New York Stock Exchange, the Toronto Stock Exchange, etc., or within business reports of companies. The decision to use these sources was also inspired by former ontology mapping work applied to this kind of data, as described in (Gro- mann and Declerck, 2014), building thus on related pre- vious work. We planned to follow an approach based on cross-taxonomies and cross-languages string mappings to automatically detect candidate multilingual dictionary en- tries for this specific domain. After an in-depth analysis of both data sources, which are available in Excel files, it soon became obvious that there is a necessity to proceed from a formally identical struc- ture derived from the distinct sources. As a consequence of that, we decided to first transform the comparable clas- sification schemes into a shared formal representation lan- guage in order to be able to properly align their components tails.1See http://www.elexicography.eu/ for more de- 2See https://elex.link/elex2017/. before …

Content truncated to 500 words