Undersøgelse af kvaliteten af Google Oversæt, herunder kvalitet inden for maskinoversættelse

Cecilie Qvistgaard Hansen & Louise Mundt Slivsgaard

Student thesis: Master thesis


During the last decade, machine translation has become accepted as an important tool for the translation of huge amounts of text material from one language to another. Thus, companies are increasingly implementing machine translation in relation to for instance customer support just as private users are increasingly applying machine translation in connection with their activities online, among other things. One of the large suppliers of machine translation is Google, whose statistical machine translation tool Google Translate has become one of the most recognised and used ma-chine translation systems on the Internet. However, the questions are how to measure quality in machine translation and when the level of quality in machine translation can be considered acceptable. In this thesis, we look into these questions by means of a number of texts translated by Google Translate from French, Spanish and Danish on two different subjects, the declining housing mar-ket as a result of the current financial crisis and the forest fires in Australia in the begin-ning of 2009. The purpose was to examine whether the level of quality differs depend-ing on source language and subject, and whether the general level of quality of the translations made by Google Translate is acceptable. Based on statistical data on the number of incomprehensible sentences and the types of errors contained in these, we find that the quality level of the translations by Google Translate varies a great deal depending on source language and subject. This variation could be an indication of the corpus behind Google Translate being more developed for, for instance, French than for Danish, which is a relatively small language. The varying level of quality depending on subject could indicate that it is more difficult for Google Translate to translate texts containing many metaphors, LSP language etc., whereas texts containing more general terminology are more easily translated by the system. In relation to the questions of how to measure quality in machine translation, what the requirements are for machine translations, and when a low quality level can be consid-ered acceptable, the answers depend very much on the situation. Aspects such as the communication situation, including target group, text type and how much time the translator has at his/her disposal are of importance when looking at the quality level of a translation. Another thing to consider is whether to evaluate machine translation according to the same parameters that apply for human translation. Some argue that this is the correct approach while others emphasise that machine translation should be evaluated accord-ing to individual parameters. Thus, experts in the field of quality in machine translation emphasise that the most im-portant thing when looking at quality in machine translation is that the user of the trans-lated material is able to understand and apply the text. If this objective is achieved, the machine translation quality is acceptable. After having looked at the general quality level in the translations by Google Translate and at the evaluation measures in relation to machine translation, the conclusion of this thesis is that the level of quality in the translations made by Google Translate is accept-able when looking at whether the overall message is conveyed to the reader. Undoubt-edly, there are a lot of details that do no get across but after having read the translated texts, the reader had got the general message and thus, the quality level can be said to be acceptable.

EducationsMA in International Business Communication (Intercultural Marketing), (Graduate Programme) Final Thesis
Publication date2009
Number of pages214