« An Unusual Exclusionary Rule Case
Debating Green Jobs »

A suboptimal newspaper headline, reported by the Language Log. The post closes with a simple but important bit of applied linguistics: An “extra second or two ... is a long time in sentence processing.”

Categories: Uncategorized    

    12 Comments

    1. PatHMV says:

      In all seriousness, though, Google is making some incredible progress in this area. Their translation service, for most languages, is good enough that I can gain a basic understanding of what is being said, which is all I’ve ever hoped for from machine translation.

      I’ve recently been using Google Voice. All calls to my home phone number are forwarded to it. I’ve chosen to have them go all to Google’s messaging service, which records the message like normal voice mail, but also transcribes it automatically and e-mails (or texts, at my option) the transcription to me. A neat feature of the transcription is that the words it is sure about are in black text, while the words it’s not sure about are in grey text. I’ve yet to spot a black-colored word as being incorrectly transcribed. It’s far from perfect, but once again, the transcription is good enough to give me the gist of the message, so I can decide whether I need to listen to it or not.

      Who knows... maybe the first sentient computer will arise in a Google server complex somewhere!

      Quote

      March 9, 2010, 1:23 pm

    2. arch1 says:

      Geoffrey Pullum:

      And this one (New York Times, March 8; thanks to Helenmary Sheridan) is a really good one, which they might better have rephrased:

      Google’s Computer Might Betters Translation Tool 

      Took me a full extra second or two (and that is a long time in sentence processing) to find the main clause verb.

      It took me an extra second or two to realize that Pullum was providing the original phrasing, not a rephrasing. And more than 10 seconds to properly interpret that phrasing. I next skimmed the comments and was initially surprised, and a bit intrigued, to see PatHMV admitting to having used Google Vice.

      I plan to avoid major decisions today.

      Quote

      March 9, 2010, 3:09 pm

    3. geokstr says:

      PatHMV says:decide whether I need to listen to it or not.
      Who knows... maybe the first sentient computer will arise in a Google server complex somewhere! 

      Where’s Sarah Conner when you need her?

      Quote

      March 9, 2010, 3:53 pm

    4. PatHMV says:

      Incidentally, the actual article bearing the difficultly-worded headline is available here. It describes how Google truly is providing cutting-edge general translation services by harnessing its immense computing power.

      [EV says: I should have included the link in the first place, and your comment prompted me to do that. Thanks!]

      Quote

      March 9, 2010, 4:17 pm

    5. PatHMV says:

      On a linguistic note, the headline suffers from two particular weaknesses. “Might” of course has two very different meanings in the English language. In the sense of “maybe,” it is paired with a verb, to indicate the possible course of action. “He might play.” “We might eat at 6pm.” “Google might improve its computers.” Thus, when you see “might” paired with a verb, you expect that “might” is being used in its “maybe” sense, not its “power” sense.

      The second issue is “betters” which, while a proper use of the word, is an archaic one. The most common form is “better,” used strictly in a comparative sense. While I have no statistics to back me up, I suspect that the current second-most-common usage of “betters” is to refer to one’s superiors (usually referring to social class), though that is more, I think an archaic and a British usage. “Don’t contradict your betters, young man.” In the sense of “improve,” “betters” is, I think, fairly uncommon. It just doesn’t initially strike us as a verb at first glance.

      Thus, we see in the headline “might betters.” The “might” following a concrete noun (computer) leads us to believe that the next word will be a verb. Then we see the next word and it does not appear at first glance to be a verb. Thus, we must retreat to a manual parsing of the sentence rather than an automatic, autonomous one.

      Incidentally, the New York Times has rewritten the headline. It now reads: Google’s Computing Power Refines Translation Tool.

      One almost wonders whether the original headline was some sort of inside joke, written in another language and translated to English automatically via Google’s service.

      Quote

      March 9, 2010, 4:28 pm

    6. jellis58 says:

      I thought the “s” in “betters” was a typo for a while until I finally figured it out maybe 8 seconds in. I thought it was saying that a computer developed by google could make some translation tool better. 

      My very intitial reaction, though, was that the headline was missing a “be” (as well as having a superfluos s) and that it was trying to say that a new translation computer made by google could be better than some other unidentified translation tool.

      Quote

      March 9, 2010, 9:14 pm

    7. John Moore says:

      Language translation and AI were hampered for a long time by folks using formal linguistic approaches. It’s good that the statistical approaches are winning out. It’s especially good for Google, which has all the raw data for the approaches.

      It would be interesting to find out what their models are like.

      Decades ago, as an undergrad project, I once fed a whole bunch of text from a couple of books into my program that ran a simple Markov model (I was a very fast keypuncher). The results were, well, humorous at times and amazing at others.

      Quote

      March 9, 2010, 9:16 pm

    8. kimsch says:

      I need to translate to Spanish from English fairly regularly and I don’t speak, read or write Spanish (but I am getting a bit better, day by day). I end up starting with Google translate for English to Spanish, then I double check using Babelfish Spanish to English to see how it comes back...

      Quote

      March 9, 2010, 9:28 pm

    9. salinabad says:

      Hello,
      It’s really nice post
      Language translation and AI were hampered for a long time by folks using formal linguistic approaches. It’s good that the statistical approaches are winning out. It’s especially good for Google, which has all the raw data for the approaches.
      Thanks for this post
      Salina

      Quote

      March 10, 2010, 1:04 am

    10. How do you revise iTunes / iPhone / iPod Transfer? (For Windows users) | Weight Loss Tips for Your Diets says:

      [...] The Volokh Conspiracy » Blog Archive » “Google’s Computer Might Betters Translation ... [...]

      March 10, 2010, 10:00 am

    11. mascali says:

      Io posso assicurarvi che la traduzione dall’inglese all’italiano è molto meglio che dall’italiano all’inglese: provate a tradurre ciò che ho scritto...

      Quote

      March 13, 2010, 11:59 am

    12. çeviri says:

      I study French online through a foreign language website, mostly by chatting with French speakers. I help them with English and they help me with French. I also know people in other countries. Most people I know this way have at least a basic understanding of English. I use Google’s translator. It is helpful, but now I know enough French to know when to ignore it.

      Quote

      May 7, 2010, 6:15 pm

    Leave a Reply