17 September, 2015
Machine Translation has been in place for more than 50 years now. As any other technology, it has evolved and it is now being widely used in many different scenarios.
However, machine translation is not just entering the text in an online service like Google, clicking a button and waiting for the translation to come up. That is just the primitive, raw translation. Machine translation can be more useful and efficient (and much more correct) if some basic requirements are met, both before and after translation.
Translate neutral texts
Texts with lots of cultural references are difficult to translate for humans. Imagine how it is for a machine that does not have context to take into account! Therefore, machine translation will work better with texts that do not make any reference to culture, traditions, religion, politics, TV or plays on words. For us, it works better with technical texts.
Although you may get some decent results with general MT systems, like Google Translate, which mix translations from all types of domains (legal, medical, food, tourism, IT…), MT will work best if you work with a system that has been made with texts belonging to the same topic (domain). This way, the terminology used will be that of the topic.
Although some rules are applied to MT systems, data is still the most important source of content. The more already translated texts for that topic, the better results you will get. Very often, 100% human translations are used for the engine to be created, in order to populate it with enough quality content that will help the system to do the translations automatically.
Machine Translation will not work if you do not fix the errors. Otherwise, you will not be able to retrain the system and improve the quality of the translation. Depending on your needs, you can make a bigger or a smaller effort on improving those translations, but you need to do a minimum so that MT fits its purpose, which is delivering better and better translations in less time.
As said before, the MT system lacks most of the context you have as a translator, and it will translate what it sees. Therefore, if there is a wrong word, it will translate it and it will not make any sense in the translation. Actually, this would be the same in the source language, but a human eye will easily notice the error and infer what the correct word is. Having said this, do always run the spell checker, regardless of whether you are using machine translation or not.
The same happens with punctuation. A comma in a wrong place can change the meaning of a sentence, and therefore, the translation will also be different.
Use short sentences. And here is why we think that MT works better with technical texts than with marketing texts. Long sentences full of subordinated clauses may be difficult for the MT system to translate.
Keep your terminology consistent. Again, something that you will find more often on technical texts. If you change the terminology continually, the system will not know which term should be used on each instance, and you will have to do more post-editing work to fix those errors.
At Jensen Localization we have MT systems in place that can be tailored to your needs. If you take all these tips into account when ordering your MT project, it will be easier and faster to build an MT system for you, so that you can start managing your translations more efficiently.
You can also contact us to let us know about your translation needs.
Machine Translation, Tools, Translation