Attaching Translations to Proper Lexical Senses in DBnary
3rd Workshop on Linked Data in Linguistics: Multilingual Knowledge Resources and Natural Language Processing, page to appear. Islande, (2014)Abstract
The DBnary project aims at providing high quality Lexical Linked Data extracted from different Wiktionary language editions. Data from 10 different languages is currently extracted for a total of over 3.16M translation links that connect lexical entries from the 10 extracted languages, to entries in more than one thousand languages. In Wiktionary, glosses are often associated with translations to help users understand to what sense they refer to, whether through a textual definition or a target sense number. In this article we aim at the extraction of as much of this information as possible and then the disambiguation of the corresponding translations for all languages available. We use an adaptation of various textual and semantic similarity techniques based on partial or fuzzy gloss overlaps to disambiguate the translation relations (To account for the lack of normalization, e.g. lemmatization and PoS tagging) and then extract some of the sense number information present to build a gold standard so as to evaluate our disambiguation as well as tune and optimize the parameters of the similarity measures. We obtain F-measures of the order of 80\\% (on par with similar work on English only), across the three languages where we could generate a gold standard (French, Portuguese, Finnish) and show that most of the disambiguation errors are due to inconsistencies in Wiktionary itself that cannot be detected at the generation of DBnary (shifted sense numbers, inconsistent glosses, etc.).