The False Quest: Eradicating Informational Entropy

georgetownexperiment1954In her 8 November presentation “Crypto-Monolingualism: Machine Translation and the Poetics of Automation,” Avery Slater challenges us to consider the ways machine translation has relied on efforts to locate, nay create, a singular language that eradicates impediments to communication caused by linguistic diversity. In pursuit of this goal, proponents of machine translation ignore the fact that linguistic messiness starts with individual languages. Although that messiness complicates translation, it is not caused by translation.

In its earliest iterations, machine translation sought a way to account for, reign in, and even take advantage of the paradoxes of informational entropy.  On the one hand, an unambiguous event can be easily conveyed because it contains a limited amount of information. On the other hand, an ambiguous event, while more difficult to convey because it contains more information, can also be a compact and efficient vehicle for conveying that load of information, a realization which poets have long relied.

In order to circumvent and circumscribe informational entropy, machine translation initially sought a mono-language identified as a “new Tower of Anti-Babel,” whose desideratum seems closer to an Ante-Babel, that mythical period before linguistic plurality.  Reaching this desired mono-language meant more than a simple unified linguistic experience; it also promised to eliminate equivocation and ambiguity. This natural generative language (NLG) becomes the code into which all languages can be re-encrypted.  Moreover, each utterance in this language would point to a single, predictive meaning.

By integrating statistical models into their algorithms, more recent forms of machine translation would seem to have abandoned the formulaic substitutions based on dictionary definitions and syntactic rule books. And yet, as I discovered when I typed in, word-by-word, Catullus’s Poem 101, I initially received a predictably awkward rendition of the poem’s lines. Halfway through, however, those lines disappeared and were replaced by a crafted translation of the lines.  I haven’t yet found their source—a 100% match did not appear when I googled the translated lines—but they are not the same as the one provided by the initial algorithm. Instead, they reach into the internet, filching how others (most likely humans) have translated those lines.  What I was given felt closer to the results that would have been provided by the person hidden inside Benjamin’s deceptive chess-playing machine.

The quest for reliable machine translation seems partially propelled by fear of the fallible, “wrongly” biased translator.  As Slater discusses, however, even the most mechanized translation devises embed the biases of their designers.  And the latest translation software that relies on human translators cannot escape the preferences of certain powerful groups of translators.

Of course, these “failures” are not really failures, for all they do highlight what makes language such a fascinating and endless source of study: its essential aleatory nature that allows it to convey information both compactly and expansively, efficiently and gloriously inefficiently.

For these reasons, I’m enthralled by the parlour game of translation tag that retranslates translations through google translator.  Here, I begin with a Modern English translation of the first four lines and then retranslate it into Afrikaans, Albanian, Amharic, Arabic, and Armenian. For Afrikaans, I provide only the translator’s rendition into Afrikaans and then back translated into ModEnglish. For the other four, I first provide the translation from ModEnglish into the receiving language and then the back translation into (just as I did with Afrikaans).  Beneath that is the translation from the language just above it and its back translation into ModEnglish.  Thus, the third entry under Albanian is the translation from Afrikaans (rather than from ModEnglish), and the fourth entry is the back translation into ModEnglish.

ModEnglish

When April’s gentle rains have pierced the drought of March right to the root, and bathed each sprout through every vein with liquid of such power.

Afrikaans

Wanneer sagte reën April se die droogte van Maart reg deurboor om die wortel en gebaai elke spruit deur elke aar met vloeibare van so ‘n mag.

When pierced soft rain, April’s drought from March right around the root and bathed every stream through each ear with liquid of such power.

Albanian

Kur shirat e butë të prillit kanë shpuar thatësirë e marsit të drejte në rrënjë, dhe bathed çdo dal nëpër çdo mënyrë me të lëngshme e fuqi të tillë.

When soft April rains have drilled the March drought right at the root, and bathed every pack in any way with such force liquid.

Kur shpuar shi të butë, thatësira prillit nga mars djathtë rreth rrënjës dhe i larë çdo rrjedhë nëpërmjet çdo vesh me lëng e fuqi të tillë.

When perforated soft rain, drought from March to April, right around the roots and wash each flow through each ear with such force juice.

Amharic

ሚያዝያ የአምላክ የዋሆች ዝናብ ሥር ወደ መጋቢት መብት ድርቅ የወጉትም ያዩታል, እና ከመታጠብ ጊዜ እያንዳንዱ እንዲህ ያለ ኃይል ፈሳሽ ጋር ሁሉ የደም ሥር በኩል ያበቅላል.

April’s rain under mild drought pierced right into March, and bathed each such power to grow through every vein with fluid.

ቀኝ ሥሮች ዙሪያ, ሚያዝያ ከማርች እስከ ለስለስ ያለ ዝናብ, ድርቅ ቀዶ እንዲህ ኃይል ጭማቂ ጋር በእያንዳንዱ ጆሮ በኩል እያንዳንዱ ፍሰት ታጠብ.

Right around the roots, from March to April, soft rain, drought surgery to power wash with juice in each ear through each flow.

Arabic

عندما الأمطار لطيف في ابريل نيسان قد اخترقت الجفاف من حق مارس إلى الجذر، واستحم كل برعم من خلال كل عرق مع السائل من هذه السلطة

When the rain really nice in April, has penetrated the drought of March to the right of the root, and bathed each bud through every race with the liquid from this authority

الحق حول جذور، في الفترة من مارس إلى أبريل المطر الناعمة، جراحة الجفاف لغسل السلطة مع عصير في كل أذن من خلال كل التدفق.

 Right around the roots, in the period from March to April of soft rain, drought surgery to wash power with juice in each ear through the whole flow.

Armenian

Երբ ապրիլյան Նուրբ անձրեւները պիրսինգով երաշտի մարտի իրավունքի արմատի, եւ bathed յուրաքանչյուրը բողբոջել միջոցով ամեն երակային հեղուկ նման իշխանության.

When drought root March April Gentle rains pierced right through every vein and bathed each sprout like fluid power.

Ճիշտ է շուրջ արմատներին, ի մարտից մինչեւ ապրիլ փափուկ անձրեւի, երաշտը վիրահատություն է լվանում իշխանությունը հյութի յուրաքանչյուր ականջին ամբողջ հոսքը:

Right around the roots, from March to April in the soft rain, drought surgery to power wash the entire flow of juice in each ear.

I’m not certain why Amharic, the ancient Ethiopian language, would be the one to introduce such modern notions as surgery and power washing, but am certain that the insights gained through such a game are limited.  Obtaining the insights revealed by translation requires a deeper knowledge of language than currently available via machine translation.

 

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