Article Insects, us & the future

From Babylon to Babelfish – the evolution of translation aids

Over 130 years after the publication of the celebrated ‘English As She Is Spoke’, are language aids still capable of producing unintentional hilarity? Max Mueller puts online translation engines to the test.

When Pedro Carolino published his seminal phrase book English As She Is Spoke, the English-speaking world took note. Mark Twain was moved to write: “Nobody can imitate this book successfully, nobody can hope to produce its fellow; it is perfect.” Its 60 pages contained linguistic pearls of “Idiotisms and Proverbs” such as “To craunch a marmoset” and “A horse baared don’t look him the tooth.” Categories like “fishes and shell-fishes” listed examples of British marine life as “…hedge hog, muscles, snail, wolf, torpedo and sea-calf.”

Carolino had translated an earlier French-Portuguese phrasebook with the help of a dictionary but without the faintest knowledge of spoken English. This approach will be familiar to anyone who has tried to translate a larger passage of text into another language using internet translation engines. From unintelligible furniture assembly instructions to nonsensical safety advice in exotic holiday resorts, the creators of Google Language Tools et al have a lot to answer for.

Nevertheless, for the last 18 months, computer programmers have claimed huge advances in accuracy. Competitors Google, Microsoft and IBM are at the forefront of new approaches such as ‘rules-based’, ‘statistics-based’ or ‘parser-based’ programmes which, they hope, will finally inject some sense into the gobbledygook, double-dutch and mumbo-jumbo of online translation.

A good starting point to put these claims to the test is Babelfish, one of the earliest programmes of its kind on the internet. Carolina’s famous “A horse baared don’t look him the tooth” (from the Portuguese ‘A cavalo dado não se ihe olha para o dente’ or ‘Don’t look a gift horse in the mouth’) produced “The given horse if does not look at to it for the tooth.” The wrong parser-base, perhaps?

Google Translate fared a little better. “Entre a cruz e a caldeira” (Portuguese for “Between the crucifix and the kettle” meaning “Between a rock and a hard place”) came out as “Between the devil and the boiler”, hinting at a smidgen of understanding and the prospect that one day the programme might make the leap from ‘boiler’ to ‘deep blue sea’.

Despite its grammatically unconvincing strapline, ‘an evolution of language translator’, Ackuna did better still. The German proverb “Ehrlich währt am längsten” (literally “honesty lasts the longest”) was rendered “Honesty is the best”, missing a genuinely idiomatic translation by just one - albeit crucial - word.

Next up was industry giant Microsoft. For its ‘linguistically informed statistical MT system’ Bing Translator, we chose, perhaps a little unfairly, Friedrich Hölderlin’s 1801 ode “Brot und Wein”. Spelling and punctuation updated, the first stanza of the German poet’s Romantic masterpiece was put through the Bing mill, with astonishing results. Apart from being free of even unintentional humour, the results were amazingly accurate. The first two lines translated as:

All around lies the city, is still the lighted street and, decorated with lanterns, rush away the cars

Compare this to Michael Hamburger’s definitive 1966 translation, a work said to accurately render Hölderlin’s ‘lexical and rhythmic choices’:

Round us the town is at rest; the street, in pale lamplight, grows quiet
and, their torches ablaze, coaches rush through and away

Admittedly, not all of Hölderlin’s other lines survived the Bing treatment intact. It is also likely that the near-poetic tone of the Microsoft version actually stems from the programme’s shortcomings. But the mere fact that the translation is even intelligible shows that online programmes have come a long way since the early incarnations of Babelfish. If further proof were needed, a final attempt to outfox the technology produced remarkable (some would even say, sinister) results. automatically translates phrases from English into 56 other languages in sequence and back again. We thought that this ultimate version of ‘Chinese whispers’ would strip even the most straightforward sentence of its meaning. However, the sites ‘top ten’ translations were:

“To be or not to be, that is the question” into “Ask”;
“Will you translate this?” into “Payment?”;
“Stalker!” into “Hi!” and
“If I had a choice, I would quit my job today” into “Back to work today.”

One wonders what Mark Twain would have made of it all.

How this article was made

  • 295 points
  • 7 backers
  • 6 drafts
  • 1 comment
Creative Commons License

Also in this issue