Posted by: patenttranslator | January 5, 2017

Dress a Woman with a Stick or Shave a Woman with a Machine?

If you tried to attempt either of the imbecilic actions mentioned in the title of my silly post today, the chances are the woman in question would start hitting you with a stick.

And who would blame her?

And yet, “dress a woman with a stick”, and “shave a woman with a machine” are perfectly legitimate translations of the Czech sentence “Stroj ženu holí“. If you don’t believe me, take a look at this article discussing ambiguities of translation from one language into another—in this case Czech to English—that human translators must solve many times a day in their much misunderstood and often unappreciated work.

A human translator would immediately start thinking there must be something wrong with the sentence because it makes no sense. A human translator, male or female, would know that women are far too complicated beings to allow anyone to dress them, let alone with a stick. Sticks simply don’t go together well with the act of dressing a woman.

A human translator would require context to translate an ambiguous sentence such as the Czech sentence “Stroj ženu holí“. If there was no context in the source text, he or she would probably go on the internet and try to find the context to make sense of a nonsensical sentence.

If there was no context to be found anywhere, a human translator would probably translate the nonsensical sentence as “Machine is shaving a woman”, because something like that is much more plausible than a woman being dressed up with a stick. For example, I have heard that some women shave their legs, although I have never actually seen one do it.

It is understandable that most women would be very private about something like that.

At my local supermarket I have seen pink razor blades being advertised as razors for ladies, and I have stayed away from these razors based on the assumption that the features of and requirements for shaving razors for female legs and those to be applied to male beards are probably different and both types are probably not equally suited to the similar, but not identical task.

But then I found recently (from a woman) that I did the right thing by staying away from them for another reason: Guy razors are the best, much better than the one for women and there is this thing called “pink tax” on women’s razors. They are pretty much the same product (if not inferior), but they are much more expensive. The same goes also for shampoo, etc.

The is no end to the perfidy of men!

If you are interested, this is what one of these manual shavers for women, called “Venus”, looks like. It costs from 9 to about 30 dollars.

If you are in the market for an electric shaver for women, you can also chose from many models, manufactured by Panasonic, Philips, Conair, and Remington, priced from about 12 to about 20 dollars, most of which should be available at your local Walgreen’s.

Thoughts like these would probably be going through the head of a human translator who is suddenly confronted with a sentence that makes no sense and that must be translated into another language, which, incidentally, usually happens to me and probably most human translators at least once every twenty minutes.

In the case of male translators, some other thoughts might possibly be also going their heads at the same time, since male brains are particularly well adapted (thanks to evolution) for certain types of multitasking.

A good translator would thus probably translate the sentence “Stroj ženu holí”, which consists of only three short and simple Czech words, as “A machine is shaving a woman”, since it is somewhat plausible that such an action could be accomplished with an electric shaver for women’s legs (or “whatever” as Donald Trump might put it). A good human translator would most likely also leave a Translator’s Note to let the customer know that there is probably something wrong with the text in the original language.

But what is a machine translation program supposed to do with such a devilishly complicated sentence, even though it contains only three short words?

When I ran the sentence through Google Translate, the best current available machine translation program, which is somewhat disrespectfully referred to by some translators as Giggle Translate, the machine translation program translated it for some reason as “Woman shaves machine”.

It is certainly an interesting concept that might make you giggle a little because that’s what it did for me.

This is just my guess, but I think that the new “neural translation” feature of Giggle Translate, which instead of using grammatical and language rules tries to match one sentence with the closest sentence available in its gigantic database and then somehow makes everything “neural”, would most likely come up with a sentence that would say that somebody, most likely a man, is beating, or hitting, a woman with a stick.

Disgusting, revolting and unforgivable as such an act might be, since the truth is that such acts have happened, and probably on many occasions, based on the innovative approach of Google, the “neural translation” of Google Translate would probably be: “A man is hitting a woman with a stick”.

When I ran the sentence through Microsoft Translator, which is (or is supposed to be) based on grammatical rules instead of statistical probability, it came up with “Machine woman shaves”. This translation clearly shows to me that the human programmers working on the design of Microsoft Translator need to work on their knowledge of grammar, Czech grammar in this case.

The fact is that just like Google Translate, Microsoft Translator also completely ignored the rules of Czech grammar in the sentence I used to test it.

It would be immediately evident to any human translator that “Machine woman shaves” is a mistranslation because the word “Stroj” (Machine) is the subject here and based on the ending of the word “ženu” (woman), the woman in question is the object here, grammatically speaking, and thus (she) cannot be in the position of some “machine woman”, so that the word “machine” would be expressing a qualifying feature of the “woman” as if it were an adjective.

If this were the case, the noun in Czech would have a different ending, namely, it would be “žena“, not “ženu“. Although this is something that would be obvious to any Czech first grader, since Microsoft Translate does not seem to grasp even this relatively simple grammatical rule, it still has a long way to go, not only in Czech, but also in other languages.

So, dear reader, whenever you see another press release from Google Translate or from our beloved “translation industry” stating unequivocally that “machine translation is quickly approaching the level of human translation”, try to remember that machine translation has been “approaching the level of human translation” since 1947, very quickly in the last 20 years, and really very, very quickly in the last 10 years (every year, dozens of article are published stating categorically that within three to five years it will be as good or almost as good as human translation).

Or maybe you can just remember a post about the dilemma of whether to translate three simple Czech words, “Stroj ženu holí” as “A machine is shaving a woman”, “Shave a woman with a machine”, “Dress a woman with a stick”, or “A woman is shaving a machine” to put the threat of machine translation to human translators in the proper context.

Like everything else, the concept of “machine translation” is all about the proper “context” – a word so dear to the heart of every translator, but impossible to explain to a silicon brain because its origin is in human experience.

Advertisements

Responses

  1. The real issue with machine translation is not so much that in very specific cases and intricate cases it can be completely wrong, each of us has heard of these dumb translations that are simply laughable.

    The fact that makes it simply unusable except by a person able to detect where it is completely off-topic (in the old days this used to be called a translator?) is that each and every sentence implies a decision.
    Any sentence, however simple, may have multiple meanings, or may refer to previous info or implicit concept. In “truck driver” and “download driver”, “driver” is a noun but in the second case “download” is most likely a verb and “driver” is no more a human subject of the action but a file object of the download action, with a possible different translation in some languages.

    So translation implies one or more decisions at every sentence, and the percentage of right guesses is much higher for a human than than for a neural program, the statistical or algorithmic programs being last by large.

    But the cumulative effect of wrong guesses is terribly bad: 1% wrong guesses makes 3 oddities per page, which will be noticed by any average reader, but most sentences are 10, 15 or 50% ambiguous: “download driver” is an action to dowload a driver on a web page for printer support, but could be a driver to download a file in other contexts.

    The machine guesses with the ambiguity percentage of the source text, the human translators removes ambiguity, extracts the proper meaning, sorts out the implicit or hidden context (this may frustrating for the ones who discuss with them, they may always find a hidden meaning!) and does that at high throughput.

    Very simple sentences can be understood completely wrong, have you ever heard “you’re beautiful today”?
    Do you mean I was not yesterday, or do you want to date me today, or this is disparaging for the other people around, or to make me stand out to your colleagues around?

    Liked by 1 person

  2. “The machine guesses with the ambiguity percentage of the source text, the human translators removes ambiguity, extracts the proper meaning, sorts out the implicit or hidden context (this may frustrating for the ones who discuss with them, they may always find a hidden meaning!) and does that at high throughput.”

    In other words, there is no such thing as “post-editing of machine translation because to “post-edit machine translation” properly means that everything has to be restranslated by a human being possessing a human brain, preferably a highly educated person fluent in at least two language.

    The whole concept is a scam. Every association of translators that claim to be fighting for the rights of translators needs to make known its position on “post-editing” of machine translations by human translators public.

    Associations that support this scam work for the “translation industry”, against the interests of its members.

    Liked by 2 people

  3. Google has recently claimed significant improvements to Translate with the application of a neural network approach:

    Like

  4. Very good point. Have you ever tried to run the sentence “it rains cats and dogs” through Google Translate and let it translate the sentence into German?

    Liked by 1 person

  5. Thank you for your comment.

    Well, yes, but this just a funny English idiom and things like idioms and proverbs could be relatively easily programmed by MT programmers into a software package IF THEY CARED ABOUT LANGUAGES!!! But the evidence seems to indicate that they either don’t know anything about languages, or don’t give a damn because MT always butchers idioms and proverb.

    A human translator who knows both languages well and understands the context would try to find the closest idiom or proverb in the other language, which can be usually done quite easily, although not always.

    I just came across the term “child turtle” in Japanese business slang, and it did not take me very long to figure out what it means and how to translate it …. because I am a human and humans get things like that, while computers don’t.

    Liked by 1 person

  6. Machine translation is the least reliable of all terminological sources…

    Like

    • Yes, it is, but it is still very useful, although much more so when you are a translator who can tell the correct parts from the incorrect ones.

      Like

      • Of ALL terminological sources available to a translator, machine translation is the LEAST reliable.

        The many non-translators who infest the translation market as translation resellers (at least 99% of the resellers) do not realise that we translators need reliable terminological sources.

        They do not realise that translators need good terminological databases, very fast to encode, much much much more than automatic translation. Because they do not understand the product or service they are reselling, nor how it is produced.

        An automatic translator will never tell you in which SOURCE SENTENCE it found what it is suggesting nor which ORGANISATION this sentence comes from – information which translators need for 2 important things: CONTEXT and RELIABILITY.

        With machine translation, the translator has to check all the parts that he or she is unsure of. Or take a chance if he or she is in a hurry. So automatic translation is NOT ONLY UNRELIABLE, but – much worse – ALSO MISLEADING.

        Non-translating intermediaries and other non-translators had the development of “translation software programs” sub-sub-contracted out (!) with the sole and only (!) purpose of extorting (!) a substantial (!) part of the remuneration of a translation profession that has always been reputed for being poor (!).

        The next step, as per their own online conversations, “since translators have been so gullible as to mentally accept the fuzzy match concept”, was to extort even more remuneration by having them “mentally accept the concept that now the machine is translating and the bilingual person is only the reviser of machine translation”.

        Fortunately, in great part thanks to online social networks (I mean, the free ones like LinkedIn etc, not the online gulags like proz.com where any act of rebellion is instantly deleted by teams of “moderators” (henchmen), and the recidivists kicked away even if their whole business depended on the platform), this time freelance translators rebelled against yet another form of extortion and even crooks number one SDL had to admit that machine translation SLOWED DOWN THE WHOLE TRANSLATION PROCESS.

        Instead of providing whole sentences, SDL’s extortion machine started providing bits and pieces of sentences, which the translator paid at the price of a reviser could then accept or reject one by one.

        But from the ads I am seeing, very few intermediaries are asking for PEMT (post-editing machine translation) (or rather PEMpT: post-editing machine pseudo-translation).

        Of course, since “trying to produce the smallest amount of effort possible to obtain the same gain” is a law of Nature, and since non-translators trying to profile themselves as employers, not only in their propaganda towards end-customers, but also towards freelance translators who pay their own taxes, gave us to understand that THEY know better and that we should CONFORM OURSELVES TO THE MODERN WAY OF TRANSLATING conceived by non-translators, we end up using machine translation, in the hope of finding some treasures among the piles of shit, which we have to analyse one by one, and check if we have time.

        I have realised how much machine translation tires me, even if the first reflex is to use it because it COULD be a useable draft, and that it CONSIDERABLY SLOWS ME DOWN because not only I have to analyse piles of shit, but I inevitably end up CORRECTING THE SOURCE CORPUS, which takes forever.

        Also, automatic translation drafts KILL CREATIVITY (not so much for technical and patent translation of course).

        And as to the style, it tends to produce WORD FOR WORD “translations”.

        So, yes, it would be better if machine translation of a better quality were produced.

        But right now it slows translators down considerably, it is not only unreliable but also misleading, it is tiring, it produces word for word “translations” and it kills creativity

        – which is the art itself of translating: producing beautiful sentences, as if they had been originally produced in the target language, while respecting the meaning of the source text.

        This is normally a profound intellectual effort, which disappears, but which is replaced by analysing piles of shit, checking them if we have time and correcting & completing the source corpus, so as not to get the same mistakes again and again (so we make the quick calculation that losing time correcting the corpus now might save us time in the next 20 years).

        Automatic translation is a “false friend” in every sense of the term.

        Liked by 1 person

  7. Well, you are right, of course. But machine pseudo-translation is useful to technical translators who can take a quick look at the terms and accept those that they know are correct, instead of having to keep looking for them in specialized databases online. Not every term is instantly on the tip of my tongue, far from it.

    But if I had to edit machine-translated materials, it would be completely counterproductive, of course.

    Liked by 1 person

  8. Well, technical translation is a very distinct part of the translation business.

    And it is – sorry, but – “not really translation” in the sense that most other translations require a lot of creativity in transposing the same idea into another language… and culture.

    In technical translation there is no (or little) cultural difference to take into account. A machine is a machine, wherever it is.

    I am sorry for the “not really translation” and it is true that technical translation, contrary to all the other types of translation, requires a very good knowledge of a specific terminology… and of the treated subject. So in this sense, it is reserved to an elite.

    To prove what I am saying (as if it were necessary, but at least to illustrate what I have just said):

    CAT tools (“(computer-aided) translation software programs”) have been created by non-translators who only based themselves on technical translation to develop their product.

    Because technical translation is something that untrained bilinguals can more or less well do if they are a specialist of the treated subject.

    Just like the guy who created the proz.com platform: he was just only a mechanical engineer who had been doing technical translations for only 2 years.

    After that he thought he was a translator (!), he created this online translation platform and it was a disaster to customers during the first few years.

    Until his team gradually learned from its own mistakes – and after most good agencies had fled away, leaving this bidding place to the low-end crooks, former commercial people on the dole trying to make a buck on the back of mostly bilingual amateurs (untrained “translators”) and selling the whole stuff as “translations” to gullible end-customers…

    This type of untrained, amateur “technical translators” ‘s work is now being sold by non-translating intermediaries to gullible end-customers, who totally ignore how much they should pay for a translation, who are too lazy to find their own technical translators online in an era when each and every translator has his or her own website plus several online profiles detailing their language pairs, specialisation areas, training and work experience, with online translation samples and references – which are exactly the same elements that intermediaries base themselves on to choose translators and which is thus a thing that customers could do too, and if they can afford the luxury of using an intermediary, they should be ready to pay a surcharge instead of pushing rates down because then it means that the freelancer works for peanuts until he or she decides to change professions or goes on social security or early retirement at the expense of all taxpayers, i.e. all customers.

    A translator (a real one, not an untrained amateur who thinks he or she should work faster, without ever looking things up in terminological sources, because he or she wants to compensate for the near-gratuity rates he quoted) can only translate, say, depending on whether he translates this type of subject in this language pair very regularly or not, 200 words per hour (this is my top speed in my top language pair in my top specialisation area, using just MS Word + Autocorrect as a typing accelerator and terminological database, very quick to encode) (I have already produced translations of 500 source words per hour but at the expense of quality, taking some risks (recently using machine translation heavily), but at the same time the agency knew that a translator cannot produce quality in so little time and should have educated the customer, i.e. the customer should have known that quality was at stake with such a short deadline).

    With a CAT tool + machine translation, my average speed is 50 source words per hour, which is below profitability level, at the expense of my health and social life, and which means missing deadlines from time to time and losing customers.

    Anyway, if end-customers knew how many words per hour a translator can really translate (and so if they would stop believing the LIES of NON-TRANSLATORS who act as translation resellers), they would know how much they should pay for a translation.

    If end-customers knew how little value the intermediary brings to an already excellent translation, they would realise that intermediaries, who take at least half of the customer’s money, are grossly overpaid and how the people who do the actual word are grossly underpaid – and how they end up paying for it via their taxes sooner or later, so this means they pay for translations twice: once to the intermediary and once via their taxes.

    Free tests are less and less accepted by freelance translators, especially when they have some experience and can produce samples of previous translations and references. Free tests are for junior translators and bilingual amateurs only. But which customer wants his translations to be taken care of by unexperienced people or by amateurs?

    So, if any end-customer reads this (and I hope they read Steve Vitek’s blog posts, because he has over 30 years of experience in the art of translating now), please find your own translators and stop feeding purely commercial people who do not have the faintest idea of what translation is all about, yet will disinform you with their propaganda.

    If end-customers keep on feeding low-end crooks who underpay translators, they have to know that they will pay for the price difference via their taxes sooner or later anyway.

    Like

  9. Every type of translation requires creativity. Technical translation requires a somewhat different type of creativity than the one you need, Isabelle, as well as precision and specialized knowledge in technical fields, which is exactly what you need in your field as well.

    I could have decided to specialize in your field 30 years ago, but I decided instead to translate patents because there was a lot of work in that field.

    To call any type of translation “non-creative” is in my opinion misinformed and somewhat arrogant.

    And it’s not the customers who are too lazy to look for individual translators, it’s the individual translators who are too lazy and too dumb to be able to figure out how to make it possible for direct customers to find them, and instead just create online profiles on Proz and TranslatorsCafe, or LinkedIn or Facebook, where they are not likely to be found, and if they are found, compete with many other translators offering very low rates.

    https://patenttranslator.wordpress.com/2015/05/09/the-difference-between-creative-and-non-creative-translation/

    Like


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Categories

%d bloggers like this: