Computer imitates a baby

Talk to me, baby: Infants begin learning speech from their first month.  Photo by Torbein on flickr.com
Talk to me, baby: Infants begin learning speech from their first month. Photo by Torbein on flickr.com

No, not by crying and pooping, but by recognizing speech. Researchers in Chicago have written a computer program that learns language sounds the same way a baby does. Exposed to tape recorded speech in English and Japanese, the computer learned to recognize all the basic vowel sounds in the language at the same pace as a baby.

This supports the theory of categorical perception. The human brain, faced with an infinite variety of sensory information, reduces that complexity by grouping similar phenomena into a manageable number of categories. Research has shown that babies can distinguish subtle variations in spoken sounds but, by their first birthday, have figured out what sounds occur in their native language. Any other sounds are lumped together with whatever native sound is closest, thus reducing the aural universe to a few manageable chunks.

(Once established, these categories can persist throughout life. My girlfriend, born and raised in Indonesia, say “dee” instead of “the” – not because she can’t make the “th” sound (if asked, she can), but because there is no “th” in her native language. The closest they have is “d,” and so every English “th” is lumped into that category. My few pathetic attempts to speak Indonesian have generated similar issues in reverse, as I substitute the English sounds I know into foreign words that are actually pronounced slightly differently.)

The computer research indicates that the human brain can do something very complicated, like learn a language, from just a few simple rules. Specific instructions do not have to be hard-wired in. This has important ramifications for understanding human intelligence, as well as for creating artificial intelligence.

It does not, however, explain why teenage girls talk so much. (Shameless plug.)

Your Comments, Thoughts, Questions, Ideas

Liza's picture
Liza says:

I know your post was about learning sounds, not grammar, but it got me thinking about the debate over the Junie B. Jones books. Some parents are upset because the books are written in the voice of a child, and they use some incorrect grammar, spelling, etc, to reflect that (i.e. "funnest", adverbs missing -ly, etc.). They argue that children reading the books are being mis-educated.

But I agree with Jill Ratzan, of Rutgers University. According to the article,

"Jill Ratzan, a doctoral student in library and information science at Rutgers University, said Junie B.’s English is actually more complex and interesting than most realize — and possibly more “correct.”

“I believe ‘perfect grammar’ is any grammar that works,” said Ms. Ratzan, whose paper on the series, “You Are Not the Boss of My Words,” was published in the journal Children and Libraries in 2005. “Junie B is actually following the precise rules of English. What she’s not following are the exceptions.”

For example, she said, “As adult English speakers, we know that the word ‘run’ has an exception in the past tense and is therefore ‘ran.’ But other verbs, you’d just add ‘ed,’ and she’s following that rule to the letter, even though she’s at an age where she has not yet been taught formal grammatical rules. She just knows them.”

Ms. Ratzan also notes that the trend of language’s evolution is toward this kind of regularization, which means Junie B. might be teaching children the English of the future. But, she said, “Just because they read ‘funnest’ doesn’t mean they’ll learn to say that. I’ve never heard a kid speak in a Yorkshire accent because they read ‘The Secret Garden’ or say ‘Have you any wool?’ ”"

posted on Fri, 07/27/2007 - 9:34am
Liza's picture
Liza says:

I'm also curious about what kind of feedback the researchers gave the computer.

When you're chatting with a child, the wee one says something, and, especially if it's mispronounced, you repeat it back so you're sure you understand what they meant. For example, if a child tells you, "I'm put on my thocks," you might reply, "Oh, you're putting on your socks? Good for you. Need any help?" It's a natural thing; it happens reflexively, but it also serves as a gentle correction that helps improve grammar, pronunciation, vocabulary, etc., over time.

It's hard to have that kind of interaction with a computer, though!

posted on Fri, 07/27/2007 - 9:44am

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Allowed HTML tags: <a> <h3> <h4> <em> <i> <strong> <b> <span> <ul> <ol> <li> <blockquote> <object> <embed> <param> <sub> <sup>
  • Lines and paragraphs break automatically.
  • You may embed videos from the following providers vimeo, youtube. Just add the video URL to your textarea in the place where you would like the video to appear, i.e. http://www.youtube.com/watch?v=pw0jmvdh.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Images can be added to this post.

More information about formatting options