Artificial intelligence grounded machine translation has fundamentally changed public awareness and attitudes towards multilingual communication. Such technologies have increasingly been used to overcome language barriers not only in situations of personal use but also in high-risk environments, such as health care systems, courts, police and so on. The development of artificial intelligence and machine learning is predicted to bring machine translation technology solutions, including speech and image recognition technologies, closer to human capabilities. In some language pairs, the accuracy, quality and efficiency of machine-translated texts of certain types can be quite high.
Nevertheless, the capabilities of machine translation are sometimes overestimated. So far, the quality of the texts generated in the Lithuanian language by machine translation systems is not sufficient and may pose various risks, the nature and impact of which are not clear and have not been researched. Therefore, with the help of surveys, semi-structured interviews and the experimental eye-tracking method, this project seeks to explore and evaluate the availability, perceptions of use, quality and impact on society of machine translation technological solutions in order to overcome language barriers ensuring effective and conscious participation of various social groups in communication processes.
Project funding:
Research Council of Lithuania (RCL), National Research Programme “Modernity in Lithuania”
Project results:
The aim of the project was to investigate and assess the availability, perception, quality and impact of machine translation technologies in society. Qualitative and quantitative research methods were applied to investigate and get insights into the accessibility and acceptability of machine translation and machine translation tools that integrate speech and image recognition technologies. The research design based on surveys, semi-structured interviews and eye tracking experiments consisted of several major parts: a survey on adult respondents’ views on machine translation; semi-structured interviews on respondents’ (Lithuanians and foreigners, adults and children) views on machine translation; assessment of the perceived quality of image-to-text machine translation by means of a ranking exercise carried out during semi-structured interviews with adult respondents; an experimental study on the quality of image and speech recognition and its impact on machine translation output; an eye tracking experiment measuring the cognitive effort of the subjects (professional translators and non-professional users) needed to understand machine translated text.
The results of the project reveal that users of machine translation tools are generally aware of the shortcomings of machine translation and the overall satisfaction with machine translation is not high. However, this depends on the level of education and language skills. The finding that the higher the level of education, the less satisfied the subjects are with the quality of machine translation confirms the results of other studies carried out by foreign researchers with professional translators and language experts. The results of the project also show that when the level of language proficiency is low, the confidence in the machine translation result is higher. Those users of machine translation tools who know the language are more likely to doubt and be critical of the quality of the raw machine-translated text. The most common purposes for which machine translation is used are related to work, studies, household and leisure activities. Google Translate is the most popular machine translation tool. Other machine translation tools are used rarely by the respondents or not used at all, as the respondents are not familiar with them. The use of machine translation tools that integrate speech and image recognition functions is not common among the respondents as they are generally unaware of their availability. However, lower quality of speech-to-text and image-to-text machine translation also leads to a lack of satisfaction with such machine translation tools.
The results of the eye tracking experiments show differences in the raw machine translation acceptability and comprehension between professional users (linguists and translators) and non-professional users, i.e., people with no linguistic background who use machine translation for various purposes. Non-professional users have a lower comprehension of raw machine translated text, but demonstrate a higher level of acceptability than professional translators, and are less likely to notice errors in the machine translated text.
The project has produced 3 research articles and a monograph prepared for publication. The results of the project were presented at a scientific seminar, organised by the project team, and in 5 international conferences.
Project coordinator: Kaunas University of Technology
