Slashdot Log In
Computer Program Learns Baby Talk in Any Language
Posted by
samzenpus
on Wed Jul 25, 2007 07:13 PM
from the what-was-your-machines-first-word dept.
from the what-was-your-machines-first-word dept.
athloi writes "Researchers have made a computer program that learns to decode sounds from different languages in the same way that a baby does. The program will help to shed new light on how people learn to talk. It has already raised questions as to how much specific information about language is hard-wired into the brain."
This discussion has been archived.
No new comments can be posted.
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
Full
Abbreviated
Hidden
Loading... please wait.
for all you techies let me translate (Score:5, Funny)
Re:for all you techies let me translate (Score:4, Funny)
Parent
Re:for all you techies let me translate (Score:4, Funny)
Ha, that's what she said...
Parent
Baby talk? I swear at my computer! (Score:5, Funny)
Re:Baby talk? I swear at my computer! (Score:5, Insightful)
And what about this "hard wired vs soft wired" stuff? What is this supposed to prove? If I build a virtual machine, does this "prove" that the machine was made of software?
Researchers examined the hardware of a babys brain, mimic it, and argue that it proves the baby learning language is in software.
None of which is to say that I think language is hardwired, but this is such ridiculous logic it makes me feel stupider for having read it.
Parent
Re:Baby talk? I swear at my computer! (Score:5, Interesting)
Parent
Re:Baby talk? I swear at my computer! (Score:5, Insightful)
You don't say if you knew your dad at all growing up, or if you looked at him as a father figure. If either or both of those fit, then even the child behavior of mimicking the mannerisms of adults could explain a lot of those traits.
On the nature side of the argument, how much of your gate and posture is controlled by your muscle structure? Same goes for your voice.
My opinion, you start with the genetic and add the environment later. It is hard for the environment to over come strong traits presented by genetic predisposition, but easy for it to mold how minor traits present.
Parent
Re:Baby talk? I swear at my computer! (Score:5, Funny)
Ewww, that was worse than a goatse link. I feel dirty.
Parent
Genetics IS a form of memory. (Score:4, Insightful)
Genetics IS "memory", your DNA "remembers" what traits your parents passed on. It's in a baby's genes to "discover" their hands and practice moving them until the hands learn how to look after themselves (eg:touch typing).
Same with language, a baby's genes will make them pick up on the phonetic sounds made by it's parents and try to copy them. It is more difficult for an adult to learn a radically different language (eg Asian vs European) because the adult brain refuses to hear the different phonetics, the adult brain long ago rejected those sounds as irrelevant to language and no longer even hears them in speech. This is why you get almost universal mistakes such as "engrish".
Parent
Re:Genetics IS a form of memory. (Score:5, Interesting)
The feedback loop is essential. There is an anecdote Linguists learn on the subject of language acquisition: A couple, both of whom were deaf for non-genetic reasons, had a hearing child. Since the parents could only communicate in sign language they plopped the kid in front of the TV a lot, thinking he could pick up spoken English from the TV. At 3 the child had developed at a completely normal rate in acquiring... sign language; he had not learned one word of spoken English.
As others have pointed out, this is one of the genetic aspects of learning a language. We are "hard-wired", if you will, to socialize, particularly with our parents, and are predisposed to ascribing meaning to the sounds we make to each other. This is of course a vast over-simplification, but I'll leave the detailed explanations to others in this thread; I just wanted to add that anecdote.
Parent
Re:Baby talk? I swear at my computer! (Score:5, Interesting)
It would appear that Chomsky et al have found that there is a "grammar engine" hard wired in the mind which assimilates the local grammar until about the age of seven when the brain reorders itself. He makes interesting case studies of pidgin languages where the several different languages are forced together, the first generation develops a common vocabulary but children born into this culture develop the formal grammar. Worth a read.
Parent
Meh. (Score:3, Funny)
Re:Meh. (Score:4, Funny)
Parent
not all languages (Score:5, Informative)
yes but... (Score:5, Funny)
"ikky wikky gaga googoo hehe hoohoo gaga, Dave"
People are afraid of new things (Score:4, Funny)
How about one that deciphers it for parents ... (Score:3, Funny)
... and integrates it into a baby monitor ...
2 PM:
She: Look, the baby said "mama."
He: No, the baby said "dada."
She: "Mama!"
He: "Dada!"
2 AM:
She: The baby's crying for you - it said "dada."
He: No, the baby said "mama."
She: "Dada!"
He: "Mama!"
how much hard-wired information (Score:5, Interesting)
Silent Little Johnnie (Score:5, Funny)
Finally, one morning at breakfast, Silent Little Johnnie suddenly pounded the table with both teenage fists, spit out a maw full of FruitLoops, and loudly announced, "This cereal tastes like shit!"
SLJ's parents were shocked. His Mother somewhat regained her composure and asked, "Johnnie...what happened? We thought you couldn't speak!"
"I can speak just fine", responded the no longer silent little Johnnie. "But why haven't you said anything before now?" his Father asked.
"Because", NLSLJ replied, "...up to now, everything s'been OK..."
And so it begins... (Score:4, Funny)
I like the theory... (Score:4, Insightful)
Re:Skeptical (Score:5, Informative)
From TFA:
Expanding on some existing ideas, he and a team of international researchers developed a computer model that resembles the brain processes a baby uses when learning about speech.
This sentence means nothing. How do they know their computer model resembles the brain processes? Because they got the same outcome? Is that enough to verify what goes on in the mind of a child?
How about this: as soon as their program can distinguish allophones, I will be impressed. Allophones are different sounds in a language that native speakers do not distinguish, but which nevertheless occur in certain environments. For instance, in English we do not distinguish the voiced th sound and the voiceless th sound, but we do distinguish f and v, even though the only difference in both pairs is voicing. The difference is that exchanging f and v can change the meaning of a word, but changing voiced th and voiceless th only makes the word sound funny.
Parent
IAAL, too (Score:5, Interesting)
IAAL, and although not a child language specialist, I will say one thing: children make plenty of meaningless sound before the start making sense, and more interestingly, they become able to tell their future native language apart from other languages quicker than they become able to understand it. (And I'll even be as daring to suggest that it simply has to be this way; you need to be able to tell signal from noise before you can decode a signal.)
I also think that by calling this a "technology," you're fundamentally misunderstanding it. It's a computer program being used as a test of a model of phonological learning.
I think you've got it exactly backwards here. The whole point this is demonstrate a model that loses the ability to tell allophones apart. I.e., that makes the jump from perceiving a speech stream as a continuous sequence of sounds laid out on a continuous acoustic space, to perceiving it as a sequence of discretely distinct segments.
Of course, a major disclaimer: I haven't seen the actual research, so I don't know to what extent they've met these goals.
Parent
Re:IAAL, too (Score:4, Informative)
Rumelhart and McLelland worked on the groundbreaking "can a neural network learn how to pronounce words based on their spelling?" paper, which used back-propagation to train a neural net to do just that. That was in the 1980s. (Sejnowski at the Salk Institute followed up with a lot of neural net training studies too.)
Their little cheat was that there was no temporal component to the data. Words were represented as sets of triplet-letters: catalog is represented as "-ca", "cat", "ata", "tal", "alo", and "log". (Actually, I don't remember if they used special sequences to represent start and stop, so --c -ca og- and g-- may not have been part of the sets.}
And of course the neural net didn't really have audio output, though of course the rejoinder is that this would be trivial.
My key question is how they deal with the issue of time in this study, and if there is any actual audio output which would act as feed-back for the training system or whether the output is representational only, as an output set of phonemes.
Having real audio output and real audio input would let it correlate its output with real language examples. Having representational blobs would only mean that: given inputs of the hash that represents "hard TH" vs the hash that represents "soft TH" the system could yield a result of different outputs.
And you're saying that the key result would be if the system learned to conflate or ignore the two sounds of "TH", hard or soft, in trying to interpret words. Remember that the initial Rumelhart-McLelland model was "content/meaning free", and I suspect that this one is too. Learning to conflate "x" and "y" in a neural net would be trivially implemented and trainable: the links for "x" and "y" into the model would have similar weights in the right contexts (the context being the set of predecessor and successor phonemes).
It sounds like an agglomerator: given a large dataset of valid words in a given language, this system learns the rule for "predecessor" and "successor" probabilities of a particular phoneme vs another phoneme and then produces random output with the same Bayesian probability, producing gibberish nonsensical sounds which follow the probability distribution of the input training language.
or that's my guess at least trying to be the typical slashdotter commenting without reading the article.
I'll try to get at the article from the Uni with journal access tomorrow.
Kris
Parent
Chomsky Is Most Probably Wrong (Score:4, Interesting)
Chomsky's argument is that there are specific areas of the brain (Broca's and Wernicke's areas) that are dedicated to language and are prewired for grammar. Truth is, people who are born unable to speak, use other areas of their cortices to learn to communicate in sign language. I see no fundamental difference between learning motor skills (such as walking, running, reaching and grasping) and learning how to speak. Every type of motor learning has to do with generating precisely timed sequences of motor commands. It is all in the timing. It just so happens that Broca's area is genetically prewired to control the mouth, tongue, throat and lung muscles. It's still motor learning. No special wiring is needed other than what is avalaible for other types of motor behavior. One man's opinion.
Parent