Skip to content
DSML.KZ
Publications

Talk · IEEE International Conference on Problems of Informatics, Electronics and Radio Engineering · November 15, 2024

Vector Embeddings and Signal Phrases for Word-Sense Disambiguation in Kyrgyz

Murat Jumashev, Gulnura Kabylbekovna Dzhumalieva, Aikokul Tashpulatova, Turgunbek Omurkanov, Jonathan North Washington, Aida Kasieva

Aida Kasieva
Open via DOI

Abstract

The paper introduces an original method to word-sense disambiguation in Kyrgyz which involves the use of contextual vector embeddings and signal phrases. Since Kyrgyz has an agglutinative morphological structure and it lacks digital resources, there are challenges that occur in field of natural language processing. The study's objective is to develop a word sense disambiguation system that can accurately represent Kyrgyz words' contextual semantic variants. We constructed a specialized homonym dictionary, assigning predefined signal phrases as semantic markers for each word sense. We implemented five distinct methods that use contextual embeddings and signal phrases in various granularities, from sentence-level and word-level comparisons to sliding windows of two and three words. In each of the approaches, the word sense that has the highest similarity score is selected. Future work might include automation of signal phrase generation using large language models, facilitated by pre-constructed, partially filled JSON objects for completion by a large language model to reduce human annotation.

Natural Language Processing TechniquesTopic ModelingSyntax, Semantics, Linguistic Variationword sense disambiguationKyrgyz languageagglutinative morphology

More by Aida Kasieva