I’ve recently read an article about how Google Translate is still a long way off replacing human translation. The author shows how limited the software is even when faced with basic ideas any human would readily understand. When it comes to translation, it really looks like Google’s artificial intelligence is still quite dumb! The text got me thinking about the use of machine translation in foreign language learning and inspired me to write about how such tools do more harm than good when learners decide to use them instead of good dictionaries for learning vocabulary.
As an English language teacher, I’ve seen quite a few students (almost all of them, really) at some point making use of Google Translate either to get an equivalent in their mother tongue for an English word or to find out how to say something in English. While it’s undeniably fast and easy to grab your smartphone, type something into the translation app and get an apparently good-enough result, I cannot but raise against this practice, as to me it’s just one of the most counterproductive ways of trying to learn English. To show you how Google Translate spoils your vocabulary learning, let me give you two practical examples of each of the situations I mentioned above.
What does loaves mean?
So one of my students (who’s a native speaker of Portuguese) is reading the sentences from an exercise and comes across the following: “I bought five loaves of bread.” He’s unfamiliar with the word loaves and decides to look it up on Google Translate. This is what he gets:
That looks like a pretty satisfactory result. The word pães is a good Portuguese translation of loaves. The app even predicts the chunk loaves of bread. So what’s wrong? Well, as a basic learner, my student still relies on his first language for making sense of English (which, by the way, is just okay!). So when he mentally translates the sentence into Portuguese, what he gets is rather confusing: “Eu comprei cinco pães de pão.” The problem is that there are two most common English words for this kind of food: one is more general and uncountable (bread), while the other is used for referring to units of bread and is, therefore, countable (loaf). Portuguese, on its turn, has only one word (pão), which can be used to talk about both bread in general and loaves of it.
Just like any other machine translation software, all Google Translate can give you is one or a few possible translations based on statistical patterns. Because it doesn’t understand language, it has no idea about that most fundamental lack of conceptual equivalence between English and Portuguese when talking about bread. It won’t teach you about countable and uncountable nouns. It’ll have nothing to say about whatever differences there are between bread and loaves. Neither will it even tell you that the singular form of the word is loaf and not
Let’s not underestimate the smartness of my imaginary student, though. In an attempt to get to grips with this word, he might well try running the whole sentence into the software. In that case, this is what he’d get:
Again, Google Translate seems to have succeeded. Well, in fact, it has. If all the student needed was to get an idea of what the word loaves means in that sentence, then everything went well. But if what he wanted was to learn the word (which means not only what it signifies, but also which forms it has and how it is used), then Google Translate was a failure. The student already knows what bread means, so everything he can do is assume that the software has either omitted the translation of one of the words or indicated that they are synonymous, just like it seems to have done before (loaves of bread = pães). The student may have understood what loaves means, but probably won’t be able to confidently and accurately use the word on his own.
How do you say “entregar” in English?
While doing a writing task, my student is telling a story about a crime and wants to say the following “A vizinha sabia que ele era o ladrão e o entregou para a polícia” (His neighbour knew he was the thief and turned him in.). He’s not sure about how to express entregar in English (turn somebody in, in this context). He types in the Portuguese verb and gets this as a result:
Based on Google’s suggestion, the learner completes his sentence: “…and delivered him to the police.”
As you can see, Google Translate has just misled my student. Not surprisingly, it used its statistical superpowers to suggest the most probable translation (according to whatever algorithm) but ignored the fact that the Portuguese verb entregar carries a number of different but related ideas that are conveyed by different verbs in English, including hand sth in, turn sb/sth in, hand sth to sb, deliver sth, give sth to sb, give sth out, hand sth out, and so on. Each of those options has its own specific context. For example, when talking about giving something to a person in authority, you should use hand in (e.g., I have to hand in my history essay by Friday.). If you’re talking about someone giving something to each person in a group or place, the best verb is hand out (e.g., Demonstrators handed out leaflets to passers-by.). In our target sentence above, the phrasal verb turn in is the best choice, as it means to tell the police about someone, or to take them to the police, because they have committed a crime.
The latest version of Google Translate does have a feature that allows you to see other alternatives for a particular word. All you need to do is click on the blue rightwards arrow next to the first suggested translation. I did that, and as expected, I got lots of options, but no information whatsoever about which option to use in which context.
Some of you might think that the software hasn’t been provided any context, and that’s why it failed to suggest an accurate translation. Shall we give Google Translate another chance? Ok, I typed in the whole original sentence in Portuguese. What I got in return was an even more confusing outcome:
Quite a lot went wrong this time. Precisely, the translation machine has failed thrice and, as a result, the English version of the sentence is not only confusing, but it actually goes in somewhat the opposite direction to the original one.
Firstly, the software’s decision to translate the definite article literally into English has created an ambiguity, as the reader may now think that the neighbour was the thief himself. This happens because, unlike Portuguese, English has no gender mark in nouns. In the Portuguese sentence, the word for thief is masculine while the word for neighbour is feminine. This makes it clear for any reader that we mean different people. A solution, in this case, would be to replace the definite article with a possessive adjective, therefore assuming that the thief lived next door and was eventually turned in by his neighbour.
Secondly, for mysterious reasons, Google Translate decided that the direct object of the verb entregar was neither the thief nor even a human being and, therefore, should be translated into “it”. This happened because the software wasn’t created to understand language, but to “decode” it. It was designed to see texts as strings of sentences, sentences as strings of words and words as strings of letters, and all of that being merely lifeless symbols to be decoded into another inert set of symbols. It doesn’t know words actually represent real-world things and abstract feelings and notions. It has no idea that the symbol ladrão represents a man who steals things and that pronouns usually refer to previously mentioned elements. Consequently, it makes no difference for the machine to translate the Portuguese unstressed object pronoun “-lo” as “it” or “him” or “her”, as the only thing it should (or was programmed to) know is that, from a statistical point of view, “it” is the most probable choice among those three, God only knows why!
Lastly, the previous error led the artificially (un)intelligent translation machine into a word-choice trap, as the selection of the verb hand seems to have been triggered by the assumption that the object of the verb was not a person, but a thing. The verb hand actually means to give something to someone by holding it in your hand and offering it to them, not to blow the whistle on someone. As a consequence, what Google translate offers the learner is quite a bit of nonsense, as the translated sentence means that the neighbour, who acknowledged himself as a thief, gave something – we don’t know what – to the police. Excuse me, what?!
Why shouldn’t I use Google Translate for learning vocabulary?
That is a most simple question to answer: it won’t teach you much; actually, it may even misteach you! The thing is, the goals of Google Translate don’t really match the needs of foreign language learners. While machine translation is quite useful as some sort of emergency resource for those who don’t speak (and aren’t learning) the target language, it will get in the way of those aiming at proficiency.
Good language learners don’t want to simply get the gist of a word and just guess how to use it. They want to acquire it, that is, to make it part of their own vocabulary and be able to use it appropriately whenever they like. It’s true that this deeper kind of knowledge of the lexicon comes with experience and contact with the language, but it’s just as true that instant translations given by a limited computer programme, which doesn’t even know what meaning is, will not be that helpful.
Google Translate will give us – language learners – options A, B or C, but it won’t tell us whether option B is better than option A, and it won’t explain to us why or when or how to use options A, B or C. It’s unable to give us exactly what we need the most for achieving a high level of proficiency.
There’s a better tool for learning vocabulary, and it’s been around for centuries. I’m going to talk about it in my next post.
How can I use Google Translate to learn a foreign language?
By now, you must be thinking that I’m some sort of anti-AI activist that wants translation machines to be banned. Not quite. In fact, I use Google Translate pretty often, both to teach and to learn languages. My whole point in this post is that using machine translation isn’t a very smart way to learn vocabulary. However, this kind of tools has some very interesting applications in foreign language learning and teaching.
One of the tasks I often have my intermediate and advanced students do is what I call Proofreading Google Translate. It consists of mistranslated sentences that the students have to correct. One way of doing that is to give them only the incorrect sentences. In such case, their task is to find all the possibilities of corrections and different senses the “author” might have intended. Another option is to give them both original and translated sentences and focus on how to correct the latter so that it both sounds natural in English and keeps the same meaning as the former. I don’t have such tasks in my classes on a regular basis, though. I usually use them as extra exercises for raising awareness of grammar and collocation patterns.
For more ideas on how to use mother tongue techniques (including translation and machine translation) in the foreign language classroom, I recommend checking out Philip Kerr’s handbook “Translation and Own-language Activities.”
If you liked this post, and if you’d like to read more about foreign language learning and teaching, multilingualism and linguistics in general, make sure to follow Multilingual Paths. In my next post, I’m going to tell you about the best alternative to machine translation you could possibly get.
See you then!