评估“新译达”在翻译本地文本的质量——与谷歌翻译的对比研究 = ASSESSING THE QUALITY OF SG TRANSLATE IN TRANSLATING ITEMS WITH LOCAL COLOUR: A COMPARATIVE STUDY WITH GOOGLE TRANSLATE

Abstract

The multilingual landscape of Singapore has seen an increasing need for quality translations from English to the three other official languages, namely Chinese, Malay and Tamil. Currently available online Machine Translation engines are unable to identify the cultural contexts and nuances of the source texts, sometimes producing translations that are not fitted to the target culture. In 2019, a local Machine Translation engine, SG Translate, was developed with the aim of producing more localised translations for Singapore. Trained with government communication materials, SG Translate can translate between English and the Mother Tongue languages. This thesis aims to investigate if SG Translate can produce more localised Chinese translations when translating items with local colour, in comparison to Google Translate outputs. A total of 11 National Day Rally speeches are chosen as the texts for evaluation. An examination that is based on BLEU (Bilingual Evaluation Understudy) and a questionnaire are employed in this study to conduct both quantitative and qualitative analysis. A higher BLEU score for SG Translate shows that the translations generated by SG Translate have a higher similarity to translations done by professional human translator, as compared to Google Translate. There is also a marked preference for SG Translate outputs in the questionnaire involving the ranking of translations by Chinese Singaporeans. The strengths and weaknesses of these Machine Translation systems are analysed and discussed in this paper.