A Comparative Empirical Evaluation of Neural Language Models for Thai Question-Answering
Fangyi Zhu, Nasith Laosen, Kanjana Laosen, Kannikar Paripremkul, Aziz Nanthaamornphong, See-Kiong Ng, Stéphane Bressan · 2022
Despite engineers and researchers' significant and continuing efforts in developing natural language processing tools for the Thai language, the Thai language is, alongside many others, a de facto low-resource language. Can unsupervisedly trained neural language models come to the rescue? The remarkable success of transformer-based language models in most natural language processing tasks promises the advent of a much needed polyglot panacea. It seems, unfortunately, that powerful enough models are not yet available for most other-than-English languages. To assess the situation, we propose to empirically and comparatively evaluate the performance of existing neural language models for the task of extractive question-answering for the Thai language.