I recently heard an argument that material which talks about technical stuff is ideal for machine translation customization. This is somewhat surprising because special topics with uncommon vocabulary have often been seen unsuitable for automatic translation. Even we have suggested that machine translation produces best quality when translating text with common type of language and simple sentence structures. So why do some professionals in the translation and language technology field see it differently? There might be three reasons for that.

First of all, machine translation and customized machine translation are two different things. Raw machine translation produced by online services like Google Translate and Bing Translator utilizes the material it is able to collect form the Internet. I don’t have any statistics but I would argue that today when most online material is created by ordinary people the Internet is filled with common language and grammatically less perfect content. Thus those machine translation services are able to translate similar types of text better than highly specific materials. Whereas machine translation customization is used for special cases and with specific needs.

This brings us straight to the second reason. When someone decides to customize machine translation there is a need to translate a very large amount of material or the translation need is continuous. Either way it is a question of volumes. There is a lot of knowledge in customizing machine translation for technical material because it has been used a lot. It has been worth the effort to customize machine translation for projects with technical texts. The quality has been improving together with the gained experiences.

Thirdly, technical texts have a lot of special terms and expressions which are repeated throughout different materials. This makes it ideal for machine translation customization. Relatively small effort increases the quality remarkably. Translation quality is often evaluated to be higher when the core expressions are translated correctly. The readers of technical texts are more tolerant to minor grammatical errors than the readers of novels because technical text turns the brains towards the understanding mode more than the feeling mode. In addition technical terms seem to be quite similar in different languages, what makes them easy to translate by a machine.

Do you think machine translation is more suitable for technical or common material?


Improve your customized machine translation with MTQ: MT-Qualifier estimates the quality of machine translation automatically. Learn how it could help you.