Home » Uncategorized

The problem with machine translation

Written By: Sylvia on December 5, 2010 4 Comments

There’s a joke quoted in the TA Professional Translation Services Guide about a machine translation of the biblical phrase “The spirit is willing, but the flesh is weak.”  Supposedly, someone used a free online translator to translate this phrase from English to Russian, then back to English again, and what came out was, “The vodka is great, but the steak is bloody awful.”

Machine translation technology is constantly developing and being applied in new ways.  Unlike human translators, who can generally translate a maximum of about 3,000 words per day, a machine translator can process a huge amount of text almost instantly.  The superior speed of machine translation can offer enormous advantages in terms of both timeline and cost.  However, machines are not likely to replace human translators anytime soon.

Here’s what a machine translator can do:

  • Like a bilingual dictionary, it can match a word in one language with a word in another language.  However, the same word may have different meanings.  For example, “spirits” can be either souls or alcoholic drinks.
  • When it has to choose between different possible translations, a machine translator can make statistical “guesses” at the context.  For example, in a sentence that talks about both meat and spirit, the machine might guess that the word “spirit” refers to alcohol.

However, the problem with machine translation is that a machine is only a machine.  It matches components and follows rules.  It doesn’t actually know what it’s talking about.

(Here, I’m addressing only the  functional aspects of language, not aesthetics.  To be sure, a machine cannot assess whether a sentence sounds good or bad.  A machine is also incapable of managing nuance, subtext, symbolism or wordplay; it cannot control mood or tone; a machine cannot write poetry.)

For a real-life example of machine translation, I took the Spanish Wikipedia entry about Woody Allen and used Babel Fish, Yahoo’s free online translator, to translate it to English.  Here’s an short example of the  resulting translation, a phrase describing the reason for Woody Allen’s childhood nickname:

“In the school they called Network to him by its hair…”

Here’s the original Spanish phrase:

En la escuela le llamaban Red por su pelo…”  (At school, they called him Red because of his hair…)

What happened with Yahoo Babel Fish?

  1. The Spanish preposition “por” can translate as either “by” or “because of.”  Babel Fish made the wrong choice.
  2. The Spanish possessive pronoun “su” can mean, “his,” “her,” “its,” or “their.”  Babel Fish made the wrong choice.
  3. The Spanish entry leaves the nickname “Red” in English.  However, there happens to be a Spanish word with identical spelling — “una red,” which means “net” or “network.”  The machine translator incorrectly assumed that all of the words in the sentence were in Spanish and translated them.

(Arguably, “Network” is a cooler nickname than “Red.”  But we have to stick with the facts…)

Machine translators definitely have their uses:

  1. As tools for reading/ general comprehension (as opposed to writing/ communication)
  2. As dictionaries, to look up individual words.
  3. As ways to get a quick “gist” translations.  And, increasingly, companies are integrating machine translation into the workflow for human translation; i.e., to save time, machines create initial rough translations which are then revised by human translators.  (There are certainly some risks involved in this process, but that’s a topic for a separate discussion.)

But in the end, machines are only machines.  Like parrots, they can echo human language without understanding its meaning or consequences.   For translations where the choice of words matters, where there is something at stake — when we are translating our employment contract or our medical records, our love letters or the foreign editions of our novels — there is no substitute for a human brain.

N. Strauss is a writer, editor, entrepreneur and the director of a Madrid-based translation agency.  A Cincinnati native, N. Strauss has an M.F.A. in Creative Writing from the University of Michigan, where she also taught creative and expository writing.  She currently lives in Alicante, Spain.

Tags:

Digg this!Add to del.icio.us!Stumble this!Add to Techorati!Share on Facebook!Seed Newsvine!Reddit!

4 Responses to “The problem with machine translation”

  1. Monica Oliveira on: 5 December 2010 at 9:33 am

    This is indeed one view of MT. Another view of MT is it is a translation tool, just like TM. I’m talking about professional MT, not Babel Fish. Translators who use MT can do whatever they want with the output, depending on the quality level they need to deliver to their customers. Some LSP use MT and edit it to sound like human translation, and it is still worth using MT if the volume of words is large. Or edit just to be accurate, correct and understandable. One common argument against MT is that it doesn’t provide good quality to medical documents, contracts, like you mentioned above. But this is not a valid argument because MT doesn’t claim to be adequate for all contents. And again, translators can edit the output. MT has helped the World Health Organization to translate most of its content in many languages. They edit all the translations but as they “train” their machine, the editing time is getting shorter. If translators don’t try, they will never see nor benefit from the technology.

    N. Strauss Reply:

    Those are excellent points, Monica. It’s certainly worthwhile for translators to explore the ways that MT can facilitate their work.

  2. Trip Kirkpatrick on: 13 December 2010 at 2:08 pm

    For a bit of amusement, you might be interested in Translation Party (if you don’t know about it already). Input an English phrase, and TP uses Google Translate to port it to Japanese and then back to English, cycling until the translation stabilizes. Taking your anecdotal starting point from above, “The spirit is willing…”, the trail looks like http://translationparty.com/#8417049.

    N.Strauss Reply:

    Trip, thanks for the link to Translation Party — it’s a lot of fun!

Leave a Reply:

You must be logged in to post a comment.

  Copyright ©2009 Found in Translation, All rights reserved.| Powered by WordPress| WPElegance2Col theme by Techblissonline.com