Named Entity Recognition from Kazakh Speech

Bauyrzhan Kairatuly, Мадина Мансурова

2024

Closed access Metadata (CSL-JSON)
DOI 10.1109/ubmk63289.2024.10773546
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

This paper addresses the challenges of Named Entity Recognition (NER) in Kazakh speech, a critical task in Natural Language Processing (NLP). The integration of Automatic Speech Recognition (ASR) and NER technologies is explored to improve recognition accuracy. We evaluate two ASR models, Wav2vec2.0 and Whisper, fine-tuning them on the Kazakh Speech Corpus 2 (KSC2) dataset. Our study demonstrates that fine-tuning pre-trained mod-els can enhance performance, with end-to-end systems outper-forming traditional methods in speech-based NER tasks. The Transformer-based Wav2vec2.0 model showed superior NER accuracy, suggesting its potential for improving NER in low-resource languages like Kazakh.