| Abstract [eng] |
Globalization could be considered a propulsion of the modern world, since masts of information, and accordingly, translation, are growing. With the intention to ease translators’ work, in the first part of 20th century, the development of machine translation (MT) engine was started. It has been developed and currently, neural MT engines are acquiring a lot of attention. MT engines are of everyday use and has emerged in different areas, mainly technical texts, and documentation, while it is not incorporated in audiovisual translation (AVT) due to the complexity of it. While audiovisual content continued to increase, financial resources designated for translators stagnant. Taking this into account, a new approaches should be incorporated in order to satisfy the need from both sides, one of them is machine translation post-editing (MTPE). AVT in Lithuania have been started investigated in late twenties and since then, many different topics regarding subtitling, dubbing, voice-over, etc., were reviewed and analysed, although research of MT incorporation in AVT is inadequate. Thus, the object of the thesis are subtitles of a documentary. The aim is to analyse MT output and MTPE efforts of a documentary and identify their interdependence. The objectives of the thesis are the following: to overview AVT, MTPE, and an evaluation of MT from the theoretical perspective; to analyse MT output of a documentary subtitles according to MQM metrics; to analyse MTPE by measuring temporal and technical efforts; to determine the quality of machine translation output by defining correlation with temporal and technical post-editing efforts for English-Lithuanian language pair. The conducted MT output evaluation according to MQM metrics revealed that grammar and punctuation errors from language conventions branch constitute half of the total errors; literal translation and word order constitute more than one-fifth of the total amount of errors, while stylistic and terminology issues are common as well. The reason for the results could be a complex nature of AVT, context deficiency for MT engine and its insufficient incorporation in the MT engine. In terms of temporal efforts, post-editors spent more than hour to machine translation post-edit the selected subtitles. Temporal efforts are related to technical efforts, while an increase of one number conditions increment of another number. Total user events, text production, and text elimination, which comprise technical efforts could be stated to be proportional to one another. The highest amount of temporal efforts were required for literal translation, word order, and over-translation, they all belong to an accuracy branch; hence it partially corroborates MT output evaluation. Results suggest that language conventions could be post-edited in a shorter time, while accuracy errors are related to the meaning of subtitles and, due to which, more temporal efforts are required. The quality of MT output is evaluated as unsatisfactory based on measured MTPE technical and temporal efforts. |
