Enhancing Translation of Myanmar Sign Language by Transfer Learning and Self-Training

Proceedings of Machine Translation Summit XIX, Vol. 1: Research Track, pages 111–122 September 4–8, 2023, Macau SAR, China. ©2023 The authors. This article is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). 111 Enhancing Translation of Myanmar Sign Language by Transfer Learning and Self-Training Hlaing Myat Nwe hlaingmyatnwe@jaist.ac.jp Kiyoaki Shirai kshirai@jaist.ac.jp Natthawut Kertkeidkachorn natt@jaist.ac.jp Japan Advanced Institute of Science and Technology, Nomi, Ishikawa, Japan Thanaruk Theeramunkong tthanaruk@gmail.com Sirindhorn International Institute of Technology, Thammasat University, Thailand Ye Kyaw Thu yekyaw.thu@nectec.or.th Thepchai Supnithi thepchai@nectec.or.th National Electronic & Computer Technology Center (NECTEC), Thailand Natsuda Kaothanthong natsuda@siit.tu.ac.th Sirindhorn International Institute of Technology, Thammasat University, Thailand Abstract This paper proposes a method to develop a machine translation (MT) system from Myanmar Sign Language (MSL) to Myanmar Written Language (MWL) and vice versa for the deaf community. Translation of MSL is a difficult task since only a small amount of a parallel corpus between MSL and MWL is available. To address the challenge for MT of its being a low-resource language, transfer learning is applied. An MT model is trained first for a high- resource language pair, American Sign Language (ASL) and English, then it is used as an initial model to train an MT model between MSL and MWL. The mT5 model is used as a base MT model in this transfer learning. Additionally, a self-training technique is applied to generate synthetic translation pairs of MSL and MWL from a large monolingual MWL corpus. Furthermore, since the segmentation of a sentence is required as preprocessing of MT for the Myanmar language, several segmentation schemes are empirically compared. Experiments show that both transfer learning and self-training can enhance the performance of the translation between MSL and MWL compared with a baseline model fine-tuned from a small MSL–MWL parallel corpus only. 1 Introduction In Myanmar, approximately 1.1 M of the population is deaf or has a hearing impairment.1 Hard-of-hearing people have difficulty comprehending spoken languages because they cannot distinguish sounds. They mostly rely on Myanmar Sign Language (MSL) for communication instead of voice. Since the structure of the grammar, syntax, and lexicon of MSL are different from Myanmar Written Language (MWL), both deaf and hearing people find it rather difficult 1https://themimu.info/disabilities-dashboard 112 to learn. In 2010, the Myanmar government launched a project to establish a standard sign language with the aid of the Japanese Federation of the Deaf (Swe, 2010). This highlights the importance of supporting and promoting MSL to ensure the deaf community has equal access to education and opportunities in Myanmar. Currently, a relatively small number, 0.006% of deaf people have a university education. This percentage is significantly smaller than that for the general population of Myanmar. However, there are few (and limited) assistive technolo- gies available for them. Therefore, deaf people require appropriate ways or tools to enhance communication with hearing people as well as to support their education. Nowadays, machine translation (MT) plays a role in breaking down language barriers and improving communication between people from various cultures and …

Content truncated to 500 words