Section: New Results

Models for interoperable lexical data

Participants : Mohamed Khemakhem, Laurent Romary.

Lexical data play an essential role in computational linguistic in two complementary ways:

• They serve as basic resources with which computational linguistic process can be parameterized. Such lexical resources are usually automatically or semi-automatically produced, are highly structured and may cover various levels of linguistic description from basic morpho-syntactic content to semantic representations;

• When created manually either for the purpose of describing a language (mono- or multilingual dictionary) or as a by product other language based activities (e.g. technical writing, translation), they may serve as a primary source of observation to analyse the way the lexicon of a language is organized, is used in domain oriented content, or how languages vary across time, space and usage.

The Alpage team has a specific expertise in the domain of lexical data, having been involved in the recent years in the creation of reference resources for the French language in particular, but also as driving force in the definition of international standards for the modelling and representation of both semasiological (word to sense) and onomasiological (concept to term) lexical information:

• ISO 16642 (TMF, Terminological Markup framework) and ISO 30042 (TBX, TermBase eXchange) as reference standards for the interchange of terminological data, for instance between translators’ workbenches, but also for the modelling of dialectal information in linguistics;

• ISO 24613 (LMF, Lexical Markup Framework), a modular modelling framework for the representation of both machine and human semasiological resources;

• The Text Encoding Initiative (TEI), which since its inception has provided an XML based format for human readable dictionaries, widely used in most last scale dictionary projects worldwide.

One of the difficulties in lexical modelling is to identify the proper modelling framework for a given lexical resource but also to ensure maximal interoperability across heterogeneous lexical content. In the recent period, we have been working on the following aspects:

• Participation in the on going revision of ISO 30046, and planning of a possible integration of a TBX dialect in the TEI guidelines;

• Setting up the revision of ISO 24613 as a multi-part standard. Alpage is now involved in the provision of a reference TEI based serialisation of LMF and the part dedicated to etymological/diachronical information;

• Proposing an extension to the TEI guidelines for the representation of etymological information in dictionaries thus offering a formal basis for the study of diachronical phenomena across dictionaries [46];

• Organising a workshop in the context of the COST action eNEL that brought together the most relevant experts in the field in order to provide a set of constraints to apply the TEI guidelines in a more interoperable way across dictionary projects;

• Starting working on a machine learning based process to extract lexical content and structure automatically from digitized legacy dictionaries, This activity, base don the architecture of the Grobid library, is the basis of the PhD work by Mohamed Khemakhem.