📄 Konkani Wordnet Resources

Summary

Konkani, an Indo-Aryan language of India's western coast, has few standardized computational resources. Strong dialect variation and the use of several scripts make the problem harder. The authors turn the Konkani Wordnet, built within the IndoWordNet framework, into a single cleaned dataset for direct use in NLP work. They pulled data from scattered lexical files, normalized it and reconciled inconsistent identifiers. They also checked that synsets, words, glosses and examples stay semantically coherent. The final resource has 32,370 synsets, each with a gloss, plus pronunciations, semantic relations and slightly more example sentences than synsets. To show a practical use, they built an API-based backend that generates Hindi–Konkani match-the-pairs vocabulary exercises. Both wordnets share synset identifiers, so Hindi and Konkani words attached to the same concept can be paired without a translation dictionary. The paper reports no learner evaluation. Still, the design works for any pair of IndoWordNet languages, giving reusable infrastructure for Konkani teaching tools and lexical research.

Key points

• Once consolidated, the Konkani Wordnet holds 37,719 distinct lexical entries spread across 32,370 synsets, which come with 32,370 glosses as well as 33,318 example sentences.
• The derived metrics show 100% gloss coverage, 102.9% example coverage and a lexical density of 1.17. Words per synset average about 1.72, from 55,530 word–synset links; the text also states 1.73.
• Bilingual pairing runs over two IndoWordNet-schema databases, wordnet_hindi and wordnet_konkani, using the tables wn_word and wn_synset_words. For each shared synset, the number of possible pairs is |H_s × K_s|.
• The exercise generator is a modular, UI-independent API. Swapping in other wordnet databases (e.g., Hindi–Marathi, Konkani–Marathi) adapts it to new language pairs with minimal changes.
• Planned extensions include difficulty levels based on frequency or POS, using glosses and examples in exercises, and multi-way alignment across three or more languages.