Is language subconscious?

Is language subconscious?  It's a question I hadn't given much thought to -- I'd always pretty much taken it for granted that it was.

Well anyway, apparently it's been a pretty contentious topic in linguistics and psychology for a while now, but a group of scientists from Hebrew University have found compelling evidence that it is, indeed, subconscious.  (And not only "language" language, but even symbolic mathematical language to an extent).

Discussing this elsewhere called to mind a time when I was working on learning Scottish Gaelic with the aid of a piece of software.  It flashed the word "eye" up on the screen, and I couldn't translate it.  I had a complete block.  I knew the word.  I knew I knew the word.  But it wouldn't come.

It felt to me like my brain didn't believe it was "eye", and that it wasn't sure it wasn't "I", or "aye" (or possibly even "ay").

The way I see it, a conscious language skill should have allowed me to override this confusion through conscious attention to the written form, which is (in theory) unambiguous.  The fact that I couldn't force my way through to the correct meaning suggested that language is something you have less conscious control over than you think....

But fair enough, I've always felt that contextless words are very much a first step, and that for the most part you should be practising in context.  This experience kind of confirmed it for me.

But the publishing of the findings came at the perfect time for me, having just written the draft of my last blog post about translation, and the way I automatically translated frames of reference (eg using the first person when the software says "you are..." etc), because that's the source of my dilemma:

Translation is effective precisely because native language is understood subconsciously.  You understand and internalise the meaning effortlessly.  This means you know exactly what message you want to express in the target language.  There is no other means for a teacher, book or software package to indicate so clearly to a learner what do say.  Also, there is no other means to ensure that the teacher knows exactly what the student is trying to say (I've heard many teachers "correct" students errors by leading them to say something that is valid in the target language, but completely different in meaning to what they intended to say).

Translation is ineffective if and whenever the native language is not understood subconsciously.  Too many courses present contrived, meaningless examples.  Perhaps these aren't quite as nonsensical as the examples the researchers from Hebrew University used, but as long as they are unnatural, they will interfere with subconscious processing, leaving the learner to process them consciously -- it's not "translation" that's the real problem here, it's "word juggling".

And translation's weak spot is the one I mentioned in my last blog post: our subconscious deals with relative references automatically.  When you say the word "I" to me, I understand the concept of you, the person speaking.  So dealing with first and second person pronouns is a perilous task.  In order to do it correctly, the student has to stop using their subconscious processing, and language starts to become a conscious process... which is when translation stops working as a language learning technique.

This leads me to a conclusion that I really didn't expect, and that I'm not yet fully convinced of:
the optimal mixture of native-language instruction and target-language instruction in the classroom isn't a simple function of how advanced the students are, it goes right down to the level of "I" vs "he".

Wow.

Just how do you prompt a student?

I'm a big fan of the courses recorded by Michel Thomas before his death, and I'm always happy to say so.  The biggest complaint I hear about the Thomas courses are that they "teach you to translate".  The argument goes that because the students are only ever prompted with their native language (English) then they never learn to "think in" their target language.  This bold assertion lacks any substantial evidence.  I would argue that translation is one of the best methods of prompting a student, and that avoiding it actually delays proficiency.

Last year I wrote a post entitled Translation: an unjustified scapegoat, in which I pointed out that translation is very very often blamed for errors that do not arise from native-language interference, and therefore cannot be translation errors.  What I neglected to say is that this is a real demotivator for learners.  To be told categorically that they're translating, when they don't really believe they are translating, and to be told to do something else without any instructions on how to do it... well, the teacher is essentially blaming the student.  That's not teaching, sorry.

Anyway, that's not the main point of this article, so time to put the train of thought back on the rails.

I am in favour of translation for three main reasons:

1: Translation allows simultaneous focus on meaning and form

If you perform a language class in the target language only, it is all too common for the answer to be mechanically reproducible from the question, without any real need to understand the meaning of either.

eg Do you have a flargrard? - No, I don't have a flargrard.

I've no idea what a "flargrard" is, so it's a reasonable bet I don't have one. (Note to non-native English-speaking readers: the word "flargrard" doesn't exist -- I made it up for this example.)

It gets worse if you include substitution drills:

eg
House: I have a house - I have a house
Cat - I have a cat.
Dog - I have a dog.
Flargrard - I have a flargrard

Translation, on the other hand, gives the student a prompt that can be understood unambiguously.  The student cannot fail to understand the full meaning of the sentence, a meaning which will therefore be instrinsically linked to the target sentence.

2: The so-called "form focus" of many monolingual tasks is really no such thing.

If the task can be done mechanically, as in the "answer in questions" example above, or the substitution drill, then you never have to select the appropriate form.  You never have to recall it from memory.  If you don't have to recall it from memory, you cannot learn to recall it from memory.

In fact, it is pretty much impossible to devise a monolingual language task that will elicit the required grammar point/structure spontaneously.  You either supply them with the structure, or you end up involved in a metalinguistic discussion that leads to one or two of the class recalling the "rule", and if we end up talking about rules, we're not connecting with spontaneous language.

3: Target-language-only normally fails to be "naturalistic".

I've discussed the issues of expository vs naturalistic language before, and in this case, I'll refer you back to the question Do you have a flargrard?  The natural response is to say simply "No," or "No, I don't," but we're generally forced to answer in (unnatural) sentences in the monolingual classroom: "No, I don't have a flargrard."  I don't know about you, but I don't like "answering in sentences" -- my brain knows it's wrong and unnecessary.  I don't like telling students off for not answering in sentences because I see this as evidence that they're actually involved in language, rather than just juggling words.


So target-language-only is potentially devoid of practice of both meaning and form, which I'd say is a pretty big problem for language learning!


Is translation the panacea then?

Well, no, because it certainly has its pitfalls.

For example, if I ask you to translate "a brown bear", am I asking you to say "a bear of colour brown" or "a bear of the species brown bear, also known as grizzly"?  And even if both translate to the same thing in the target language, there's a point of assymetry when we hit "white bear", which is not ambiguous.  Prompts for translation must be very carefully selected, then.

This limits how far we can learn a language by translation, obviously.  We cannot learn every noun and idiom by direct translation, but we don't need to -- the trick is to use translation where it's (1) obvious and easy or (2) where conscious awareness of the difference helps overcome a specific difficulty.

(1) The "obvious and easy" would include my favourite example: conditional sentences in English vs Romance languages, which translate pretty much directly -- eg  "If I was/were you, I would...", "If I'd known you were coming I'd have baked a cake" etc.  This is "advanced" material in traditional classes, but translation makes it trivially easy (to the point where Michel Thomas would be teaching it on the second or third day of his courses).

(2) An example of a specific difficulty is the difference in idiom between "to be" an age in English and "to have" an age in the Romance languages.  Not a difficult rule, but even after loooots of practice, you'll often here a learner make the mistake one way or the other.  So the practice method the teacher uses gets the student to produce the desired answer, but it doesn't build any resistance to native-language interference, so in an uncontrolled setting the original error returns.  (And the teacher blames the student for translating, and the student is confused and disheartened etc.)

One of the biggest visible benefits of translation though, is simple:

Speed, volume and throughput of practice

Because translation starts with a readily-understood prompt, you don't have to waste too much time thinking about what the prompt means or what you're being asked to do.  This means you can get through a lot more questions.  A translation-based lesson that manages to present no more questions than a target-language-only lesson is a wasted opportunity.

In an attempt to teach myself Corsican, I've written a little program that conjugates, combines and declines words and presents them to me as translation tasks, checks my answers and tells me if I got them right or wrong.  I can batter through hundreds of examples in very little time.  Kind of exhausting, yes, but pretty effective.  A couple of hours using it, spread over a couple of weeks, has hammered in some of the basics pretty solidly.

But...

Translation's biggest problem

Once you start whipping through the questions at speed, you really do start to work on autopilot, and you start to see patterns emerging in your errors.  And I noticed one specific type of mistake that I made frequently that I hadn't been too aware of before... I kept switching my "I" and my "you".

It makes perfect sense, now that I think about it, and I probably did it a lot with MT, even though I didn't pay it any mind at the time.  And heck, I've even heard the same thing from some of my students when I've asked them to translate short sentences.

Because when the computer says to me "you know it", that "you" refers to me.  It's "eio", "io", "je", "yo", "ich" or whatever.  That's what it means.  Literal direct translation is therefore something of a higher-order function, an abstraction.

And yet it seems to be quite effective.  So what do we do?

Well, personally I'll be attempting to stay away from first and second-person references as much as possible.  I'll be sticking to the minimum required to learn them individually as grammar points, but when the person is included only as part of the context for a sentence testing another grammar point, I'll favour "he/she/it/they" over "I/we/you".

But I'll certainly be paying more attention to what exactly happens when we translate.  I still think it's one of the best tools the learner has, but we've just got to work to eliminate the ambiguities....

From the mouths of bits - curiosities of machine translation

Google Translate is undeniably one of the most useful tools most of us will ever see, yet to the vast majority of people, it is a joke.

The principles behind Google Translate go completely against what we expect of language.  Our first instinct is to believe that Google used a big set of rules and tables, like in those dusty old Latin books on a shelf at the back of the university library.

But Google Translate is something very different.  It is based statistical translation techniques.  What that means is that no-one has programmed it with any rules at all, instead feeding it with gigabyte after gigabyte of text in the target language, from which it identifies patterns of words that go together, and words that don't.  It also gets some directly translated texts to compare translations, but much less than you might expect.

Occasionally, this statistical approach throws up some very odd results.

For example, on How-To-Learn-Any-Language someone recently gave the example of a Finnish band who sing some of their songs in English and some of them in Finnish.  When he translated a piece of Finnish with the song title Kuolema Tekee Taiteilijan in it, it spat out the Siren, which is another of their song, but one they sing in English.  The correspondent on HTLAL blames that on human correction, but that is highly unlikely.  Instead, Google's algorithm will have correctly identified the first song title as Finnish and the second as English, even when in a document in the other language, and therefore it won't add the Finnish song's title to the English database or the English song's title to the Finnish database.  And because both co-occur with the band name, the software ends up associating them.

In fact, if you look at the bands list of singles, you'll find that Kuolema Tekee Taiteilijan was released directly before the Siren, so it could be that the Google algorithm is actively looking for a translation directly after an embedded foreign word.  So if I talk about the clàrsach (Gaelic harp) or about an t-Eilean Sgitheanach (the Isle of Skye), you get the picture.  And it's quite right that Google Translate should do that, it just so happens that while it means that it makes less mistakes, the mistakes it does make are mistakes that look particularly weird to us.

Creoles - the same story once again

So I was directed this morning to a news story on the BBC about the translation of the Bible to Jamaican Patois.  It's a move that's long overdue -- whatever you think about religion, you have to accept that the place of worship is vitally important in the survival of language wherever a large percentage of the population are religious.  The lack of a Bible translation and the use of the dominant language in religious services have been cited in the decline of many languages, including Scottish Gaelic.

It has been welcomed by some:

Several women rise to testify, in patois, to what it means to hear the Bible in their mother tongue.

"It's almost as if you are seeing it," says a woman, referring to the moment when Jesus is tempted by the Devil.

"In the blink of an eye, you get the whole notion. It's as though you are watching a movie… it brings excitement to the word of God."

Unfortunately, not everyone is so happy.

But some traditionalist Christians say the patois Bible dilutes the word of God, and insist that creole is no substitute for English.

You know what?  There was a time when people would insist that English is no substitute for Latin.  And even that was bigotted, because the Latin Bible was just another translation of the Greek, and it wasn't even that accurate!
What we have here is proof, if proof were needed, that a great many objections to minority language are a simple case of resistance to change.

Creole in primary education

The article doesn't restrict itself to the Bible, but follows on to a topic that is a matter of active debate in most creole-speaking countries: the place of Creole in the primary sector.

The story is always the same: the "big" language is of major economic importance, and therefore should be the focus of education.  As a political statement, it's appealling, and it doesn't take much thought to agree with it.  Which is just as well, because it doesn't really hold up to much scrutiny.

One thing that has been fairly well proven across the world is that kids do better in school if they are given "initial literacy" (their first experience of reading and writing) in their own language.  On the other hand, gaining their initial literacy in a new language actually hampers their ability to pick up the language accurately.
Worse, in some creole-speaking countries, the teachers are really only creole-speakers themselves.  Education in Haiti, for example, is very heavily orientated towards French, but the teachers really don't speak the language properly.  What you end up with is kids who aren't competent in either their own language or the "important" language.

All the figures show that the best thing to do is to start school in the kids' own language (and the teachers'!), and that the new language is best introduced in a spoken form, and by a native speaker.

Which isn't quite the same as what we do in Scotland with Gaelic-medium education, sadly....

Translation: an unjustified scapegoat

I couldn't count the number of times that I've heard a teacher respond to an error by saying "that's because you're translating! You need to think in the language!"  This is all well and goo... no, there's nothing good about it.  I found it particularly frustrating when I found myself incapable of saying something specific in Gaelic, and my "friend" refused to let me simply say it in English.  I then said it wrong and he then let loose with the old "because you're translating" line (except in Gaelic, just for variety).

Well no, the problem wasn't that I "was translating", it was that I had never learned how to say it.  I hadn't learned it, I couldn't say it -- simple as that.

Translation has become something of a bogeyman.  If you make an error caused by native-language-interference, the witchfinder in front of you will cry "translation!" and insist you must "learn to think in the language".  Except that quite often these days, the witchfinder will be someone who doesn't actually speak your language and therefore blames translation when the converse is actually true.

My favourite example is the English and Spanish conditional constructions.  Spanish-speakers regularly get the English wrong, and teachers are wont to issue the usual battlecry of "think in English!" followed by a lecture on "2nd and 3rd conditionals" in abstract grammatical terms.  But in fact, Spanish conditionals translate almost verbatim into their English equivalents, so if the Spanish folk were simply encouraged to translate, they'd master the English forms in about half-an-hour.

I found a quote on-line the other day:

"Disillusionment regarding the relevance and usefulness of learning theory for educational practice has been responsible, in part, for the emergence of the theories of teaching that are avowedly independent of the theories of learning."

Ausubel, David. Educational Psychology: A Cognitive View, 1968

1968, but this still seems to hold today.  When we discuss learning, there's one word that is perhaps more important than any other: generalisation.   Sadly, there's very little discussion of this in teaching literature.  Our job is to teach, your job is to learn, so generalisation is dismissed as the student's responsibility.

The problems that are often blamed on translation are better blamed on generalisation.

Native-language interference is inappropriate generalisation of known patterns (the fact that they are in the native language is incidental).

On top of that, generalisation even accounts for errors such as the conditionals where native-language interference should provide correct results.  How so?  Well, many intermediate and advanced learners have a tendency to try to construct any new target language structure out of known language structures -- ie they assume the language they have been taught is the whole language and try to generalise the known structures to cover any new case, and they fail to innovate.

For a specific example of generalisation gone wrong regardless of medium, I have to go back a good few years to my first experience of the language classroom: high-school French.

We started off with phrases, albeit with translation.  It was all "what is your name?" "I am 12 years old" "I'm fine, and you?"

Now our teacher told us that "j'ai" was "I have" and that it only meant "I am" when discussing ages, but several of the class got themselves in a right guddle over this.  Some would say "j'ai" instead of "je", and some would say "j'ai" instead of "je suis".  And then they would try using "je suis" instead of "j'ai".

The problem here wasn't translation, because we weren't building up from grammar rules -- we were substituting words in fixed phrases in the hopes of learning "by induction from examples".

So the problem needs to be described in terms of generalisation.  Regardless of what the teacher said, certain pupils automatically generalised telling their age to "I am".  A change of medium of instruction couldn't have altered that.

So what could have changed that?
I think the main problem was that this was our first encounter of the verb "to have".  Quite quickly we moved onto how many brothers and sisters we had, but by this point the confusion had set in.

Ausubel proposed something called progressive differentiation.  Under this framework you teach a core, high-level overview of a concept, and then refine all the particulars and special variations after.  The core use of "avoir" in French is possession of a physical thing.  If that had been taught and thoroughly learned before a specialist idiomatic form was encountered, inappropriate generalisation would have been impossible.

This is how Michel Thomas does it.  In his French, Spanish and Italian courses, you play around with "I have it", "I don't have it", "I don't have it but I want it" etc before coming anywhere near idiomatic constructions such as "to have hunger" for "be hungry".  It is thus impossible to generalise "to have" incorrectly as "to be".  Yes, you can generalise incorrectly the other way, and talk about being hungry (and then get a bollocking off your teacher for translating) but this is a far smaller error.  Not only that, this is an error of "not having learned yet", rather than an error of "learning wrong".

So please, don't simply shout down translation indiscriminately -- it's not the nasty beast you think it is.

Passive vs Active (cont) -- Assimil.

In my previous post, active vs passive, I mentioned the Assimil series of books.

I've was listening to one of their courses in the bath this morning and it strikes me that they are on the right track here -- in some languages.

Each lesson in the Assimil method teaches by dialogues, and you are expected to listen to each dialogue multiple times.  This is backed up by a translation exercise that in the course I was listening to (Le Catalan) was actually quite effective in rearranging the language taught into a new and fairly different form, but one that is still within the reader's comprehension.  Unfortunately the other book I've tried (Iniciación al Euskera) presents tasks that are impossible given the previously taught knowledge.

With all the listening, reading and relistening of the main dialogue, you're possibly only getting 5-10% of your time exposed to new comprehensible material.  The study I quoted said that even a 50-50 split of active vs passive was as effective as 100% active study.  Does this suggest that Assimil is not optimally efficient or that the fundamental differences between music and language are enough that this doesn't apply?

The perils of "community" translations.
Recently Facebook, one of the highest valued websites since the inception of the internet, decided to become multilingual. Despite having raised a phenomenal amount of money from Microsoft, they didn't hire any professionals -- they opened the project to "the community".
Why? Because they wanted to expand their user base -- without investing a penny.
All well and good, but you get what you pay for, and that's what Facebook got.
I've already spotted one error in one of their translations, and it has a serious effect on the meaning of user profiles. The relationship status "it's complicated" has been translated to Spanish as literally "in a complicated relationship", which is a very different thing.
Why did this happen?
They asked specifically for native speakers of the target language only. Unfortunately that meant the team, such as it was, lacked sufficient source language knowledge. The Emperor's Nose fallacy holds whether you have a dozen, a hundred or a million people: the average of all guesses is not necessarily the right answer.