The Old Wig, the Hanged Man and MT

J.S. Bach - The Old Wig

Professional linguists are strict about differentiating between translators and interpreters. The first deal with written, the other with spoken language.

It doesn’t sound so funny in comparison to the old joke about a jeweler and a jailer (“the first sells watches, the other watches cells”), but I’d say that the interpreter translates (orally) and a translator interprets (written text).

For the sake of argumentation, I am not going to be very discriminate about these terms. Interpretation has many meanings. I am not going to write about philosophers who interpret the world. As somebody with a musical background I am thinking of interpretation in yet another sense.

Long ago when there used to be CD stores, my father-in-law went to buy “Goldberg Variations” in Osnabrück. “By who?”, asked the salesman. “Why, certainly by J.S. Bach”, answered my father-in-law, only to learn that the salesman meant not the composer, but the pianist, that is to say an interpreter of “Goldberg Variations” by the Old Wig.

I was always wondering what makes all those interpretations so different. Pianists interpret the same music sheet, playing it note by note. Conductors use the same music score, allowing only slight variations in tempo and dynamics. But the result always varies. The salesman’s question was perfectly justified. Performing artists meticulously reproduce each written note of the original content, but it is the differences in their interpretation, however intangible they may be, that make people choose “Goldberg Variations” as performed by Glenn Gould, Murray Perahia or Martin Stadtfeld. In fact, it is the interpretation of the same familiar notes which somehow makes them different, interesting and new.

What does it have to do with translation, that is interpretation of written text? I don’t claim it to be an art (even if it is sometimes difficult to differentiate between art and skills, both in music and translation). However, differences that matter and are taken for granted in music, are looked upon askew when dealing with translation.

According to Philipp Koehn, Professor at University of Edinburgh, “translation is unpredictable. If you give a text in a foreign language to a group of translators, you can be sure that each will come up with a different translation.”

For some, this variance seems to be a predicament.

The above quotation is taken from an issue of Language and Translation, focusing on “Machine Translation” (MT). Here, the inevitable differences in translation are interpreted as a flaw. The urtext music sheet calls for various interpretations, but in translation, only one correct interpretation in each given language seems to be desirable.

The “problem” stems from the human nature, so if humans are not capable to solve it, the assumption is that technology can. MT will provide one “correct” translation and disperse all doubts about the right choice. For adepts of MT, diversity and quality are mutually exclusive. MT will set up a paragon of objective quality – translation as one true thing.

However, in practical life MT stands for the opposite of one true thing. The simplest way to produce translation errors (and have something very entertaining) is to use machine translation.

Apparently, MT adepts hope that this real problem of MT could be eventually solved by various approaches.

The long term solution is to keep on trying to make MT produce less errors. The short term solution is to make humans correct MT errors by way of post-editing.

In both cases, one reasonable approach seems to be setting up strict limits where MT, with or without human interference, i.e. interpretation, may be used.

In a recent presentation (SDL Machine Translation for Post Editing: Let’s Talk Strategy), Andy Reid, a product marketing manager at SDL, tries to narrow down the area where application of MT could help cope with large amounts of content, albeit at the cost of poor quality:

SDL - Machine Translation and Post-Editing

The picture is rather messy. I don’t know how you are going to differentiate between “websites” (human translation) and “blogs” (machine translation) or “wikis” (machine translation) and “help” (human translation).

To simplify and deconfuse, I suggest a clear-cut, four segment matrix in the tradition of the Boston Consulting Group:

Machine translation, but only for machine generated content

To answer the question “Where does Machine Translation fit in localization?” I dare say it’s best used to translate machine generated content. It is the lower left, bluish greyish corner on the picture above.

Machine translation goes in line with machine generated content in terms of communication purpose, language quality, target group orientation, copyright and other aspects that need to be considered when debating the applicability of MT. “Give Caesar what is due Caesar”: this is the one logical thing that still holds true, whenever MT is being held for a promise of one true translation.

However, the SDL presentation contains another interesting slide:

Is "mechanisation" of translation really inevitable

Word by word, note by note, everything in its place: mechanisation – industries – grow – inevitable…

This statement is typical of an attempt to present translation as an “industry”, not a “profession”. It takes for granted the idea that translation can be “industrialized” and hence has to be “mechanised” just as well.

Each translation is different and unique, every translator “will come up with a different translation”, but at the same time, translation as such is always a copy, replica or reproduction of something else. “Goldberg Variations” performed by Glenn Gould (or somebody else) are Glenn Gould’s reproduction of the original as composed by Bach. Translation, that is interpretation of written texts, follows the original. It inevitably assumes the production of written content, as a rule, by somebody else.

Do we all agree that production of written texts has become an industry? Do we take it for granted that mechanisation of writing “is almost inevitable as industries grow”?

There used to be times when writing as well as science, ethics, culture and art, actually everything else besides working in a factory or on a field, were declared to be in need of industrialisation. For somebody like me, with a Russian and German background, the word “mechanisation” rings bells. It has its firm place in a series of associations like collectivisation, industrialisation, mechanisation… German Agitprop, Russian Proletkult, Mayakovsky and Bertolt Brecht, cultural Bolshevism of all sorts… Thank God we are through with electrification, at least.

Kasimir Malevich - Head of a Peasant

Approximately at the same time, in the beginning of the 20th century, there appeared the so called Mechanical Pianos, also known as reproducing pianos, player pianos or autopianos.

These were self-playing, automated music instruments, with keys activated not by a human hand, but by the so called “piano roll”. Player pianos were used mostly as recording devices that provided the most accurate and consistent “live” interpretation of a musical piece by an absent human pianist at that time.

The mechanical piano was a cutting edge technology, far more superior than the low-tech phonograph or gramophone. It didn’t pretend to be one true thing, but it certainly was something special and unique. Nevertheless, it didn’t last long.

The crash and the Great Depression “wiped out production” (Wikipedia). However, mechanisation of music interpretation didn’t stop with the virtual death of mechanical pianos in 1929. Nowadays, MIDI files and music notation software like Sibelius or PriMus allow to completely mechanise the performance of music. With notation software you can play back your music automatically, without human assistance.

Characteristically, the area of application is rather limited. The software is great for composers to get a realistic idea how the composition would sound when performed on music instruments, it is the composer’s equivalent of a WYSIWIG editor for a graphic artist.

But however consistent, accurate and natural sounding the machine interpretation of a music piece might be, I have never encountered a recording with “Sibelius” as the performing artist.

I don’t know whose interpretation of “Goldberg Variations” my father-in-law chose among those recommended by the salesman in Osnabrück. It was certainly not a recording of the mechanical piano or some notation software, because there were none available. Never recommended and probably never asked for.

Cassandra jean - Tarot cards - the Hanging Man

Adepts of Machine Translation have been trying to present MT as the cutting-edge technology in the translation “industry”, as the future of translation par excellence. Arguments to the contrary are usually ignored or dismissed as prejudiced opinions of the soon to be extinct species of slowcoach, technology-averse translators. If you as a professional translator keep rejecting MT, you are either stupid, or simply scared to lose your last miserable, pathetic quality translation jobs, that is what the arguments of all those MT gurus boil down to. If they had read Tom Wolfe’s latest novel (“Back to Blood”, 2012), they could quote a character from his book: “You cannot be cutting edge, if your generation is dead or dying”.

The sad thing is it is only a matter of time until you belong either to one generation or the other. Humans are cursed to hang in between.

It is impossible to predict the future of technology. Despite massive efforts to promote Machine Translation as something cutting edge, the reputation of MT is so low that at the time the term itself is being replaced by newer, nicer euphemisms.

Translators who use machine or automatic, as it is increasingly more often called, translation (e.g. through SDL or MemoQ plugins) try to hide this fact. Clients are getting nervous if something reeks of MT or other “tools”, as automatic translation tools are getting ever more available.

Because no matter how you call it – automatic, semi-automatic, PEMT or even HAT (human-assisted translation) – the old problem remains. It is the problem of a mechanical piano, which nobody wants to listen to.

In the words of Philipp Koehn, Professor at University of Edinburgh, human translation is unpredictable. The future is unpredictable too. But a look into history may be helpful. Because history repeats itself. It simply has to. Nobody listens.

“The Hanging Man” and “Page of Blades” (used for the Machine Translation matrix) are two Shadowhunter Tarot cards by the freelance illustrator and graphic artist Cassandra Jean from Florida.

The picture by Kasimir Malevich is called “Head of a Peasant” (1929, Russian Museum, St. Petersburg). Malevich’s paintings are also mentioned in Tom Wolfe’s new novel “Back to Blood”(2012) which takes place in Florida. A pure coincidence, I guess.

  1. John White’s avatar

    Consider, though, that your father bought a recording of Gould, or whomever — your father did not hire Gould to play for him whenever he wanted to hear Goldberg Variations.  Thus the "performance" was the same every time, hums and all, and, while acceptably entertaining, certainly not the experience that hearing Gould live in his parlor would have been.



Your email address will not be published. Required fields are marked *