Introducing Lexical Masks: a New Representation of Lexical Entries for Better Evaluation and Exchange of Lexicons

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 3046–3052 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 3046 Introducing Lexical Masks: a New Representation of Lexical Entries for Better Evaluation and Exchange of Lexicons Bruno Cartoni, Denny Vrandeˇci´c, Daniel Calvelo Aros, Saran Lertpradit Google {brunocartoni, vrandecic, dcalvelo, slertpradit}@google.com Abstract The evaluation and exchange of large lexicon databases remains a challenge in many NLP applications. Despite the existence of commonly accepted standards for the format and the features used in a lexicon, there is still a lack of precise and interoperable requirement specifications about how lexical entries of a particular language should look like, both in terms of the numbers of forms and in terms of features associated with these forms. This paper presents the notion of “lexical masks”, a powerful tool used to evaluate and exchange lexicon databases in many languages. Keywords: lexicon, interoperability, evaluation 1. Introduction Lexicon databases are at the core of many NLP systems. Yet their maintenance, evaluation and exchange between different systems can be cumbersome. Aside from the eval- uation of the quality of individual entries, the evaluation of the consistency of the structure is a problem that is hard to tackle manually. From an interoperability perspective, sev- eral attempts have been made to homogenize the morpho- syntactic features used to represent the lexicon, but few ini- tiatives try to provide minimal requirements concerning the internal structure of a lexicon entry and its minimal specifi- cation. In this paper, we introduce the concept of lexical masks which aims at representing, in a consistent way, the ex- pected internal structure of lexical entries. Masks are de- fined for each language and each part-of-speech in that language. We first describe the basic principles of lexical masks, how these masks have been designed for several languages, and how they have been used to automatically evaluate an existing lexicon. We also introduce different levels of specifications that account for the defectiveness of some lexical entries, but also allow some flexibility in the verification process. Finally, we describe how the masks are used in Wikidata to validate existing entries and help contributors to add entries more easily. 2. State of the Art The interoperability of language resources has always been a challenge for the NLP community. The notion of inter- operability covers two dimensions (Ide and Pustejovsky, 2010): syntactic interoperability and semantic interoper- ability. The first one involves the data format and com- munication protocol while the second one focuses on the meaningful interpretation of one system’s data by another. In the domain of machine-readable lexicons, different projects have tackled the difficult issue of unifying features to describe lexical entries, allowing a meaningful semantic interoperability. Specifically, projects have aimed to unify the descriptors used to label specific forms in a lexical en- try. Examples include the General Ontology of Linguistic Description GOLD 20101 (Farrar S., 2010) and the Univer- sal Dependency Tree-bank2 (Zeman et al., 2017) that inher- ently has to …

Content truncated to 500 words