Feeds:
Posts
Comments

Posts Tagged ‘Machine Translation’

You are uselessLast week the Observer published a very interesting article by Maureen Freely, who successfully translated Orhan Pamuk’s works into English. I recommend the article to everyone interested in discovering the relationship between author and translator, but it is so good that there will be something for everyone. I was particularly taken by two very good points Maureen Freely makes. Firstly, she reminds the readers about the importance of literary translation and, therefore, literary translators:

An up-and-coming Colombian novelist might be inspired not just by Borges, Conrad and Faulkner, but by contemporary novelists from Asia, Africa and Europe; his literary response to their work will go on to influence what his contemporaries on the other side of the world write next. These complex patterns of cross-fertilisation would end overnight if it were not for literary translators and the publishers who support them.

I couldn’t agree more. This notion of cross-fertilisation (I used the verb cross-pollination in one of my first posts) has always been one of my main arguments whenever translation comes up as a topic of discussion. That translations only make up less than 3% of published titles in the English-speaking world should be a cause for concern. And I am not referring to the trite rants about “cultural imperialism”, but simply to the fact that publishing and reading such a limited amount of translated literature is bad for the national literatures of English-speaking countries. Goethe believed that without outside influences national literatures rapidly stagnate. Moreover, in countries where translations constitute as much as a third of what is published, it is common for

novelists and poets to work at some point in their lives as translators. Though most will say that they did so mainly to subsidise their own writing, it is often clear, when you look at that writing, that it has been enriched by the imaginary conversations they’ve had with the poets and novelists whose words they have translated.

On a completely different note, Freely’s article also shows how machine translation is perfectly useless for literary texts. A few months ago, I compared the most used free automatic translators, in an effort to show their users how easily things can go wrong. I was quite surprised to find myself linked by Luigi Muzii who called me naïve (although he also states that Edith Grossman, Sylvia Notini and Lawrence Venuti are damaging to the profession, so, yeah, I guess a couple of pinches of salt are in order) and went on to rant about the silly literary translators’ need to feel “irreplaceable”. I never even responded to that, as my original post was pretty much enough to prove my point, and it wasn’t meant to be an in-depth technical analysis of machine translation or the work that makes it even possible, as I am anything but an expert in the field. I just analysed the results. Maureen Freely, though, gives us an even better example of how literary translator do not need to feel irreplaceable, because, apparently, they are. Here is the first sentence of Istanbul: Memories of a City as rendered by Google Translate:

A place in the streets of Istanbul, similar to ours in a different house, with everything I like, twin, or even exactly the same, starting from childhood lived another Orhan a corner of my mind I believed for many years.

Translated by Maureen Freely as:

From a very young age, I suspected there was more to my world than I could see: somewhere in the streets of Istanbul, in a house resembling ours, there lived another Orhan so much like me that he could pass for my twin, even my double.

Not the quite the same, feel, there, or am I wrong? Even more mind-blowing is Google’s translation of the first sentence of  The Black Book:

Bed-of top-from tip-to as-far-as stretched-out blue checked quilt-of rugged terrain-its, shadowy valleys-its and blue soft hills-its-with covered sweet and warm darkness-in Rüya face-down stretched-out sleeping-was.

Hmm. Let’s see what Freely did:

Rüya was lying face down on the bed, lost to the sweet, warm darkness beneath the billowing folds of the blue-checked quilt.

I don’t think I need to add much more. Nonetheless, it’s a free world, and Luigi Muzii (especially considering his experience and competence) can freely call me or anyone else “naïve” for defending the vital role of literary translation and the impossibility for a machine to replace a human being when translating literature. As for the rest of us, let’s keep smuggling words, ideas, metaphors and visions. It’s the best cure against linguistic barbarism.

Image: You are Useless, by 2493/Gavin Bobo (Flickr).

Read Full Post »

I was wandering around the ocean of links and more links, when I found a blog called Flabbertech, which announces enthusiastically that “the simultaneous translator is almost a reality”. There is a Star Trek quote and then the blogger asks us what is surely meant to be a rhetorical question:

Have you ever dreamt or even just hoped to be able to communicate with every person in the world without having to know all the languages but simply by speaking your mother tongue?

Honestly, not at all. At the most I have dreamt or hoped to be able to learn thirty or a thousand languages. Mainly because if I didn’t know a specific language, I would be lacking a tool that is essential if one wants to understand the culture speaking it. And the conversation would be somewhat beckettian, I guess. It’s perfectly fine to be a nerd, finer still dreaming about a Star Trek utopia (I had the entire classic series on tape, just for the record, and it would be awesome to have a replicator and get pappardelle with fresh porcini down here in Australia) but when we talk about reality can we please try to keep its complexity in mind?  Too many people still see completely oblivious of the fact that languages are not equivalent and interchangeable codes. I can cope with this, but on the other hand I think it’s sort of my duty to contribute my two cents and try to spread the awareness of what language is, and how the sentence “in the beginning was the word” has quite a literal meaning.

Anyway, let’s not delve into the nature of language and the way it’s inextricably linked to culture, and let’s see: does this thing work? Are we really in Star Trek, and I hadn’t noticed? There’s a video explaining how the widget works:

Images of Captain Kirk aside, it doesn’t really seem that mind-blowing. Watching the video we find out that the program is being tested  in 25 situations American soldiers commonly find themselves in Afghanistan. In the comments, we find out that you need 4-5 seconds to translate 10 words. Basically, it’s nothing amazing. Especially when one thinks of the cost of these projects. Very advanced technology, don’t get me wrong, but – considering the results – was it really necessary? Brian Weiss, part of the evaluation team, says (2 mins 20 secs)

unfortunately there is a shortage of interpreters, a shortage of very reliable interpreters, and machine translation offers a unique tool in the sense that machines don’t get tired, people do.

Meet Captain Obvious. To reply in style, do we dare say that people interpret – in the fullest sense of the word, meaning that they are able to evaluate and give meaning to, say, non-verbal messages – and machines don’t? A flesh-and-bone interpreter will be quicker and much more accurate. Secondly, they could count on their knowledge and understanding of the “other” culture. Finally, they will be able to interact with people using not only these cultural tools, but also that 80% of communication that is non-verbal. Big shots from an intelligence agency should know these are not small details. Or at least, an average citizen whose security is supposedly in their hands would hope so.

Weiss might be right when he says that there is shortage of interpreters, and I guess that very few professionals are ready to move to Afghanistan or Iraq and work on the field. Why not look to the military, then, and invest in training? Many people choose to join the army to be able to study, why not offer incentives to whoever chooses the path of language learning and interpreting?

A little research, and I find out, on Cellular News, that the project, named TRANSTAC, is currently being tested on Pashto, Dari and Iraqi Arabic. One would expect some major breakthrough, but the gist is old news:

All new TRANSTAC systems all work much the same way, says project manager Craig Schlenoff. An English speaker talks into the phone. Automatic speech recognition distinguishes what is said and generates a text file that software translates to the target language. Text-to-speech technology converts the resulting text file into an oral response in the foreign language. This process is reversed for the foreign language speaker.

Are you kidding me? Firstly, there’s the problem of how reliable speech recognition might be (think of noise, and above all dialects and idiolects). Secondly, judging by the shortcomings of machine translation, I wouldn’t be to sure of that text-to-text part, either. Synthesis is probably the only segment that, bar major disasters, seems to be reliable. Is this stuff really seen as preferrable to the training of flesh-and-bones “war interpreters”?

Apparently so. The problem of bad translation in the theater of war was analysed brilliantly by Emily Apter in The Translation Zone. Where we can see that similar technologies have been around for a while and so far

the results proved to be unreliable, and in the worst cases fatally flawed

Emily Apter’s essay also includes a quote of a New York Times article from 2003, where Edward Luttwak hints to the linguistic side of intelligence gathering:

”To be a case officer you have to be a poet,” he continued. ”You need to romance and seduce. You need to be able to learn Urdu in six months.” Woefully short of language skills, many American intelligence officials, ”can’t even ask for a cup of coffee.”

Great. It’s not hard to imagine the kind of information gathered by an organisation which does not seem to have any idea of how important linguistic competence is. To the point that, instead of implementing new programs and train operatives with an in-depth knowledge of the language and culture of the war zone, they prefer to play Spock & Kirk with unlikely and dreadfully expensive gadgets which will hardly ever replace human intelligence. Looks like a bunch of fluff to me.

PHOTO: Communicator vs. iPhone, by Lee Bennett (Flickr)

Read Full Post »

A lot of people keep telling me that online machine translators “give you an idea anyway”.  A while ago, I found a sentence that is quite useful to test these widgets. It is worth noting that it’s a very simple sentence, with a purely communicative function, but it contains four terms – fan, club, match and cup – all of which have two very different meaning and can be quite ambiguous for a machine to translate. Let me clarify why, for the non-bilingual. The Italian word for fan as in the device that produces air current is ventilatore, while a supporter is tifoso or fan; club as in what Fred Flintstone carried around is clava or mazza while a group of people with a common interest is called a club in Italian too. Match as in what you use to light your cigarette is fiammifero or cerino but it’s partita if you mean a game, and cup is translated as tazza if it’s what your drink your tea in, but as coppa if you mean a trophy. So, the sentence is

“the Italian fan club was ready for the last match of the World Cup”

The experiment is a simple one. Translating the sentence from English to Italian, then back to English. I used it to test Babel Fish, WorldLingo, Reverso, Bing Translator, Systran and Google Translate, with surprising outcomes.

Yahoo! Babel Fish has – incomprehensibly – quite a reputation, but that might be due more to its old age than to the results it achieves:

EN > IT

il randello di ventilatore italiano era pronto per l’ultimo fiammifero della tazza di mondo”

IT > EN

___ randello of Italian fan the ready era for l’ last match of the world cup”

Right away, translating into Italian, Babel Fish fails on all our ambiguous words, and misses a preposizione articolata (that’s when the prepositions merge with the articles), so that, despite maintaining the sentence’s structure, the result is incomprehensible. Going back to English, it loses two articles, leaves one in Italian, mistakes a verb for a noun and doesn’t translate the noun randello. The resulting English sentence is not just incomprehensible, it’s not English at all. My rating: 1/10

WorldLingo suffers from similar problems:

EN > IT

“il randello di ventilatore italiano era aspetta per l’ultimo fiammifero della tazza del mondo”

IT > EN

the randello of Italian fan it was waits for for the last match of the cup of the world

All four ambiguous words are misunderstood by the software, that – inexplicably – can’t even translate ready with pronto, coming up with a surprising aspetta. Going back towards English, once again randello is not translated, and while era is recognised as a verb, the translator adds it in front of it, indicating that it doesn’t realise that the preceding noun is the subject of that verb. Moreover, that aspetta which popped out of the blue becomes – quite logically – waits for, followed by another for which translates per. As was the case for Babel Fish, it was really hard to do worse than this. My rating: 1/10

Let’s try Reverso:

EN > IT

il bastone di ventilatore italiano era pronto per l’ultimo fiammifero della Coppa del Mondo”

IT > EN

the baton of Italian fan was ready for the last match of the World Cup”

A little step forwards. Finally World Cup is recognised as Coppa del Mondo, but nonetheless we’re still talking about an ultimo fiammifero. Three out of fours ambiguous words are not translated correctly. Club here becomes bastone (stick) instead of randello, but nothing changes. Era pronto is correctly translated into was ready. Nonetheless, the sentence we get is still not understandable. Going back to English, the first half of the sentence is definitely absurd, bastone becomes baton and the word order doesn’t make much sense. My rating: 4.5/10

Note how these first three softwares used randello and bastone to translate club, which actually corresponds in most cases to clava (which is also a cognate), or mazza (e.g. golf club = mazza da golf).

Let’s move to Microsoft‘s Bing Translator :

EN > IT

il fan club italiano era pronto per l’ultima partita di Coppa del mondo”

IT > EN

the fan club Italian was ready for the latest batch of ___ World Cup”

Hell! The English to Italian translation is impeccable, except for a capital. Maybe Bill Gates will prove himself once more?  Unfortunately, going back to English things unravel. There is a problem with the word order (fan club Italian instead of Italian fan club). Those four ambiguous words go back to what they were, but – unexpectedly – that unfortunate latest batch comes up, an ultima partita in a commercial sense, and that really kills the sentence. Then, to be painfully strict, an article is lost in translation. It would be hard to get the original meaning out of the final sentence, if you weren’t reading this article. To be fair, though, compared to the competitors above, we start seeing some logic in the translation process. My rating: 5/10.

Now, let’s move to the free translator on Systran‘s website, the first translation software ever and the leading supplier of translation software.

EN > IT

il fan club italiano era pronto per l’ultima partita della coppa del Mondo”

IT > EN

the fan Italian club ready era for the last game of the World Cup”

Once again, four out of four ambiguous words are translated correctly. Except for a capital, the English to Italian translation is fine. Revert to English, and we grimace in disappointment (or is it relief?)  because, inexplicably, a verb is once again mistaken for a noun. A fundamental mistake that affects the comprehension. Then, there is a word order problem (the fan Italian club). Shame about that era translated as a noun, which really kills the final English sentence. My rating: 5/10

Finally, let’s see how Google Translate performs.

EN > IT

Il fan club italiano era pronto per l’ultima partita della Coppa del Mondo”

IT > EN

___ Italian fan club was ready for the final match of ___ World Cup”

Bloody hell, doesn’t Google always do things properly? The translation from English to Italian is perfect. Going back to English, we even see a lexical correction, where the machine uses  final instead of last – very appropriately, given the context. Strangely, though, Google Translate fails with the article il and the preposizione articolata della, ignoring them completely, and producing a sentence which is perfectly understandable but still sounds funny. In any case, it is pretty clear that this new widget comes off as the prodigy here – but we’re talking about Google, should we really act surprised? My rating: 8.5/10

To sum up, if you need to grasp the general meaning of a sentence, Google Translate (and Google Translate only) seems to guarantee a minimum of reliability. Nevertheless, let’s make clear that as soon as we move from simple informative sentences to more complex ones, even Google Translate is not to be trusted. I will soon analyse a recent case to demonstrate this. In the meantime, if it is a translation that you need, I warmly advise you to contact a flesh-and-blood professional translator.

Read Full Post »