Machine translation quality has improved remarkably over the years. Automatic translations are becoming more and more usable in different contexts. Heavy users might have noticed that the development, however, hasn’t been linear. Machine translation has moved up some kind of quality steps.

The reason for jumps to higher quality steps is technical. Machine translation technologies are getting much attention and resources. The idea of how machine translation works has changed. The steps from rule-based to statistical machine translation and then to the use of artificial intelligence (or AI) have all improved the quality. The technologies are getting closer to the way humans adapt their first languages.

Little babies learn a language by being surrounded by it in their everyday life. They hear it over and over again. They try to say a word or two. They will first say it wrong but improve by repeating it many times. They learn new words and what those words mean. They do it even without having someone explicitly explaining these things to them. They hear, their brains work very hard and they learn.

Using a lot of data

AI and neural nets learn languages a little like babies. Data is the most important element. For technologies, it is crucial to have enough data. Just like for babies it is crucial to hear people speak and read aloud. The more is better. There is plenty of research saying parents should read to their children. When parents read to their children, they will succeed in life. They will have better skills in communication and linguistic. No human nor machine can learn without enough data coming in.

Using good quality data

Likewise, the quality of the data is important. One-sided interactions with the language limit the learning process. Varied and rich exposure to different uses of language brings more versatile results. But, too wide data isn’t going to work either. There should be enough repetition to ensure remembering. Babies learn first those words which they hear most often. They hear them in different sentences and situations. The more they hear those words the stronger the meaning and usage of them become in their minds.

Getting support for the learning

All language learners need some outside support as well. Control by someone who knows more. Parents tend to correct if a child says something in the wrong way. Or at least they repeat the idea in other words back to the child. They often use phrases like ’did you mean that’ or ’what do you try to say’ or ’did I understand you correctly’. Even the most modern technologies need that kind of guidance and feedback as well.

The difference is that AI won’t really understand. Technologies know words and can link meanings and synonyms. They can produce spoken or written texts. They can even maintain a conversation through chat or customer service. But they don’t understand. Babies and children can. Human brains are just amazing. We are clever enough to use machines and technologies in those tasks that can be outsourced. This way we can concentrate on more interesting and creative tasks, and challenge our brains to keep on improving.

(Final note: AI and neural nets are very advanced technologies. For many they are just abstract and unclear terms. This text simplifies them and tries to give an overall idea of how they work. If you want to tell your own description, we’ll be more than happy to publish it here on our blog. Just leave a comment below or email us.)