IndoUKC: A Concept-Centered Indian Multilingual Lexical Resource

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 2833–2840 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 2833 IndoUKC: a Concept-Centered Indian Multilingual Lexical Resource Nandu Chandran Nair∗, Rajendran S. Velayuthan†, Yamini Chandrashekar∗, Gabor Bella∗, Fausto Giunchiglia∗ ∗University of Trento, Italy †Amrita Vishwa Vidyapeetham, India nandu.chandrannair, yamini.chandrashekar, gabor.bella, fausto.giunchiglia@unitn.it rajushush@gmail.com Abstract We introduce the IndoUKC, a new multilingual lexical database comprised of eighteen Indian languages, with a focus on formally capturing words and word meanings specific to Indian languages and cultures. The IndoUKC reuses content from the existing IndoWordNet resource, while providing a new model for the cross-lingual mapping of lexical meanings that allows for a richer, diversity-aware representation. Accordingly, beyond a thorough syntactic and semantic cleaning, the IndoWordNet lexical content has been thoroughly remodelled in order to allow a more precise expression of language-specific meaning. The resulting database is made available both for browsing through a graphical web interface and for download through the LiveLanguage data catalogue. Keywords: wordnet, multlingual lexical database, Indian languages, language diversity 1. Introduction The Indian subcontinent, home to hundreds of lan- guages, presents an impressive linguistic, social, and cultural diversity. This diversity is evidently manifested in the lexicons, with many words and even entire cultural domains lacking precise equiv- alents outside of India, and often even within the subcontinent. For example, the Malayalam word ഹസ്തസൂത�ം (hasthasoothram) represents a type of ornament that only married women wear. This word cannot be found in western lexical resources since this Indian word is not that popular. Our aim is to give justice to the richness and di- versity of Indian lexicography by proposing the IndoUKC, a new lexical database for Indian lan- guages. The IndoUKC provides words and cross- lingually mapped word meanings over 18 lan- guages. The distinguishing feature of the IndoUKC is that the lexicons are mapped together in a diversity-preserving manner, meaning that both meaning equivalence and untranslatability—when a word has no equivalent in another language—are explicitly marked. Through a layer of supra-lingual lexical concepts, we have mapped the 18 Indian lexicons to Princeton WordNet and to the lexicons of thousands of other languages. A major data source of IndoUKC is IndoWordNet (IWN) (Dash et al., 2017), a multilingual word- net resource that interconnects 18 lexicons using Hindi as a hub language and that is also par- tially linked to Princeton WordNet (PWN) (Miller, 1998). IWN is used in cross-lingual computa- tional applications such as machine translation (Chakrabarti and Bhattacharyya, 2004). IWN is a rich source of lexicalisations specific to Indian lan- guages and culture(s). However, IWN does not al- ways represent such word meanings explicitly and formally enough to be exploitable in practice, due to the underspecified correspondences its mapping model is able to express. IWN also contains a cer- tain amount of syntactic noise and semantic (map- ping) mistakes. In order to solve these issues, we have carried out a thorough cleaning and re-mapping of IWN word meanings using the ‘diversity-aware’ lexical …

Content truncated to 500 words