I recently decided to translate a short story that I really like from English into Esperanto. I got together all my tools (Benson’s dictionary, Wells’s dictionary, PIV, and Merriam-Webster’s dictionary and thesaurus), and was about to begin, when I realized that the online translation tool Google Translate might be helpful.

According to the media, Google Translate is constantly being improved by use of something called Artificial Neural Networks and Deep Learning. So, I decided to give it a try. I ran each English paragraph through Google Translate first, and then edited the results as necessary.

As I used the translator, I realized that it shows many of the qualities of the “natural acquisition” approach to learning foreign languages. Phrases and sentences that are popular and used frequently are translated well; less popular phrases are translated poorly. Google Translate does not yet formulate grammatical rules from experience like humans (even children) do, so even simple grammatical rules are often not followed. Simple verbs are often translated with the wrong tense, and participles are almost always handled badly.

As I expected, Google Translate does not yet translate words in context, so it often returns the wrong meaning. The English word post has four or five different meanings, for example, so it is not surprising if Google Translate picks the wrong one. Even the English word upset should be translated differently, depending on whether it refers to a person (disappointed) or a container (overturned). Generally, Google Translate’s translation of any given word is not to be trusted; if you do not know for sure that it’s correct, you need to check it with your dictionaries.

[Editor’s note: What’s worse, even context may not be enough. In his essay Learning From Translation Mistakes, UN translator Claude Piron gave the example of the English word repression: in a political context it can mean subpremo, or in a psychological context, repuŝo. The right choice isn’t just a matter of knowing what field the text belongs to. In an article dealing with the Stalin era, for instance, one might find a sentence beginning with “Repression by the population of its critical reactions led to …” And even though the text deals with politics, the proper translation (repuŝo) relates to psychology. For a machine translator to be able to choose accurately in these cases, it needs an understanding of the text — something that, so far, automated tools like Google Translate don’t have.]

Nevertheless, Google Translate turned out to be a very useful tool. It can be much easier to edit an existing translation than to develop one from scratch. Also, since I did not do the original translation, I am not invested in it; I can replace large parts of it without a qualm. Finally, since I know that Google Translate will always make a literal translation, I always consider replacing that translation with one that better captures the sense of the original.

I strongly encourage you to try translating something into Esperanto using Google Translate, even if you are just beginning to learn the language. As you edit the result, all kinds of questions will come to mind, and trying to answer those questions will help you to become a better Esperantist.