I have been using Google Translate for translating from several languages to English for well over a year now.
I put it to use recently on several Russian patents dealing with electronics and chemistry, Russian blogs about new inventions, and a Russian university transcript of a medical doctor.
I also tried it out among other things on several French patents about electronics, forklifts and industrial vehicles, several chemical patents in Slovak and Czech, two long financial statement of a major Slovak producer of chemical products, and dozens of Japanese and German patents covering a whole range of fields. I can see that Google Translate (GT) has a number of obvious strengths, as well as quite a few weaknesses.
The strengths and weaknesses could be summarized as follows: people who don’t know much about translation and who expect that GT will replace human translators will be mostly disappointed. Translators who can use it instead of their many dictionaries on paper will be mostly delighted.
First about the wonderful strengths of Google Translate:
One big advantage of GT for translators is that when an extremely obscure term in a foreign language has only one possible translation, GT will quickly fetch it for us like an obedient doggie.
For example, it used to take me forever to find the names of plants, algae and different types of seaweeds which are used for Japanese food and thus often featured in Japanese patents dealing with foodstuffs and food processing technology. Instead of using (originally Chinese) meaningful characters for these words, the Japanese names are often written in a Japanese alphabet called katakana, which makes it difficult for me to figure out what these words mean. It used to take me forever to find English equivalents on Internet. But these days I just type them in katakana into GT and the correct Latin name appears on the right side.
Medical terminology transliterated (originally from Greek) into Russian so that the original spelling is often hidden in Cyrillic, such as othorhinolaryngology or phthisiopulmonology, used to be difficult for me when I could clearly see only a part of the word. But all I have to do these days is type it in Russian in GT to find the correct English equivalent.
Because I made a conscious decision to specialize in technical translation and patents in particular a long time ago, I am not at all up to speed on financial terminology. But since every now and then companies sending me mostly patents will also send financial reports, for example describing sales, profits, loans and investments in product development in their subsidiaries in Europe, I either have to find somebody who can translate this kind of material, or do it myself.
How do I find a translator who knows the correct terms in a language such as Slovak for things like “available liquidity coefficient”, “credit liquidity coefficient”, or “depreciation for fixed assets”?
Well, I don’t have to do that anymore. I can just go to Google Translate and specify the source language as Slovak and the target language as English because my Slovak is pretty good, I just don’t know financial terms in Slovak. But since these terms usually have only one equivalent in English, GT will again translate them correctly.
GT thus removes or at least reduces the knowledge barrier that used to prevent me from translating documents in languages that I know quite well if they were in fields that I have been avoiding for decades.
And now about some of its many weaknesses:
As described above, GT works very well if you use it mostly as a dictionary, not so well if you think that it can be really used the same way one would use a human translator to translate big chunks of text.
GT is often unable to make determinations that would be quite easy to make for a human translator. It also works much better with languages that are quite similar to English, such as French or German, than with languages that are quite different from English, such as Japanese or Czech.
One of many problems that GT is often struggling with in Japanese is that there are no spaces between words in Japanese, by which I mean that Japanese text looks like this: “oneofthemanyproblemsthatGTisoftenstrugglingwithinJapaneseisthattherearenospacesbetwenwords”. GT sometime breaks up the Japanese continuum in the wrong place, thus creating wrong words with the wrong meaning.
I also found that GT often used to misidentify the pronunciation of Japanese characters, which often depends on context, when I would click on the speaker icon to test the “pronounce” feature. The speaker icon is still there, but the sound mysteriously disappeared. Did GT developers realize that they need to put more work into this feature?
GT usually “understands” the meaning of a sentence in French or German much better than in Japanese.
The meaning of what is produced by GT after it has processed a complicated Japanese sentence will be often comprehensible to me only if I compare it to the original sentence. I think the main reason for this is that a human translator understands, or should understand instinctively, that he or she must remember the “topic” of the sentence, called “wadai” in Japanese, in order to translate the text correctly. I am not sure that it is possible to even program something like this into machine translation.
Unlike European languages such as English, French, or German, Japanese is incredibly context-dependent. If I fail to keep constantly reminding myself what is the “topic” of a convoluted sentence that I am translating from Japanese, I am bound to mistranslate it because I will misunderstand it. Again, I doubt that this is something that can be programmed into any machine translation program.
I was sort of surprised to realize that GT is also very bad at things like idioms, proverbs, or what one could call “cultural references”, not only in languages like Japanese or Czech, but even French or German. Try to think of a well known proverb in another language and type it into GT. The chances are that if the proverb is longer than about 5 or 6 words, it will be mistranslated to the point of ridiculousness.
Since there is a relatively small, finite number of well known proverbs in any language, and an even smaller number of well known quotes, for example from Shakespeare or the Bible, one would think that it would be relatively simple for GT to include such “cultural references”. But one would be wrong, at least at this point.
To my surprise, because GT is sometime unable to identify correctly grammatical features such as endings indicating the case of a noun in Czech, it will often mistranslate something because of that.
Google Translate is slowly replacing my dictionaries
Whenever I translate something even a little bit complicated, I usually surround myself with a protective wall built from several specialized dictionaries. I used to do that out of necessity, but now I probably mostly do it out of a habit, or perhaps to ward off evil spirits, although once in a while I still reach for one of my favorite chemical or technical dictionaries when I don’t like what GT suggests.
Just like search engines made it possible to quickly find just about anything online in the last decade or so, programs such as Google Translate, Microsoft Translate, and many specialized online dictionaries, including terms suggested by translators on Proz, have made the work of this specialized translator much easier now compared to the situation only a few years ago.
Search engines did not put specialized research services out of business. On the contrary, they made it possible for these businesses to expand the range of services that they offer to customers and to charge higher rates for value added through specialized knowledge by leaving simple grunt work to search engines, as well as to reach more customers.
As I already wrote in another post about Google Translate in February of this year, I think that instead of replacing human translators, this is similar to what Google Translate will do for experienced translators, although people who don’t know much about translation, and even some translators who are now scared silly when Google Translate spits out a perfectly translated sentence, often postulate that it is only a matter of time before this patent translator is replaced by his favorite dictionary.