| Abstract [eng] |
This paper presents a research on human acceptability of machine translation from English to Lithuanian. The novelty of this research is that it provides an insight on professionals as well as basic users’ (experts and non-experts’) acceptability of a random machine translation system’s output for the English–Lithuanian language pair. The relevance of the research is that it analyses the aspects that make a translation acceptable and acknowledges the issues that have the most influence on acceptability as perceived by machine translation users. Therefore, the object of the research is human acceptability of machine translation from English to Lithuanian. The aim is to analyse acceptability of machine translation from English to Lithuanian and its influential factors as assessed by machine translation users. The set-out objectives to meet the aim are as follow: 1. to discuss human acceptability in translation studies; 2. to discuss translation quality assessment metrics; 3. to analyse experts’ and non-experts’ acceptability of machine translation; 4. to indicate the most problematic machine translation aspects by assessing the machine translation quality; 5. to evaluate the factors that have the most impact on the machine translation quality and acceptability as assessed by its users. The research includes the descriptive, comparative, qualitative and quantitative statistical analysis methods. The main tool to analyse acceptability of machine translation is the acceptability model proposed by Castilho (2016) who proposed that acceptability is measured by considering three elements: usability, quality and satisfaction. Moreover, Multidimensional Quality Metrics (MQM) introduced by Lommel, Burchardt and Uszkoreit (2015b) are employed to assess machine translation quality by considering accuracy, fluency, style and locale convention dimensions. Acceptability is assessed via two different surveys by 20 experts and 70 non-experts. In this research, experts are individuals who have experience in translation and/or degree in translation or languages, whilst non-experts are basic machine translation users whose background is unknown. The results of the analysis show that machine translation from English to Lithuanian is deemed unacceptable by all surveys’ respondents. Even though non-experts share more positive views towards satisfaction and usability of machine translation, they agree with experts that the low quality of a translation is the main factor that makes it unacceptable. Further analysis also shows that machine translation users are mostly affected by critical accuracy errors present in a translation, whereas fluency issues are identified as disturbing yet not major, and issues of style and locale convention are not perceived as having any influence over the comprehensibility of machine translation by both experts and non-experts. Taking everything into account, machine translation from English to Lithuanian is deemed unacceptable by its users, the main reason of which is low quality of machine translation. The low quality and incomprehensibility of machine translation is mostly influenced by critical accuracy errors as perceived by both experts and non-experts. The research is comprised of two parts. The theoretical part of the work discusses theoretical aspects of human acceptability in general, human acceptability of machine translation, and methods to measure acceptability. It also reviews translation quality assessment methods. The analytical part of the work presents an analysis of experts and non-experts’ acceptability of machine translation from English to Lithuanian and the factors that have an impact on it as perceived by its users with a focus on the quality element. |
