Domain Specific Iterative Word Sense Disambiguation in a Multilingual Setting
Mitesh M. Khapra, Pushpak Bhattacharyya, Shashank Chauhan, Soumya Nair, Aditya Jaikumar Sharma · 2008
We report here our work on Domain Specific Iterative Word Sense Disambiguation (WSD) for nouns, adjectives and adverbs in the trilingual setting of English, Hindi 1 and Marathi 2 . The methodology proposed relies on dominant senses of words in specified domains. Starting from monosemous words we iteratively disambiguate bi, tri and polysemous words. We combine corpus biases for senses along with information in wordnet graph structure to arrive at the sense decisions. To the best of our knowledge, this is the first attempt at a large scale multilingual WSD involving Indian languages and English. The accuracy values of approximately 65% (F1-score) for 1 Hindi is the official national language of India. The language and its close cousin Urdu are spoken by approximately 500 million people in the world. 2 Marathi is the official language of Maharashtra, a state in Western India. The language has close to 20 million speakers in the world. all the three languages compares well with the state of the art.