Monday, August 27, 2007

(more on) whether we have strong linguistic knowledge ingrained at birth..(optional reading)

In the class, I mentioned that all human infants come into this world with what can be thought of as a "universal grammar" that they can "tune" to the local language that they are hearing around them. In otherwords, language is not wholly learned outside--contrary to the common wisdom (the story of how "universal grammar" came about is also a fascinating one--see the mail that I sent to the class last year  )
One question is whether there is really something special about the set of "human languages" as a whole that is different from any other languages.
In particular, if the human baby were to be given away to martians (or other aliens that regularly visit Area 51 and certain phoenix suburbs), would the baby be able to master the martian language? Conversely, if the human babies were to get together--without any intervention from adults, and were to make a brand new language, would it be closer to all the other human languages than it is to any other language?
We can answer the latter very much in affirmative thanks to the fascinating real life story of Nicaraguan Sign Language. A bunch of nicaraguan deaf kids who were ignored by their war-torn society and over a period of time developed a new sign language all their own from scratch. And it is *very* similar to other human languages  (see the URL http:/ / ).
We don't yet have a clear and convincing evidence that babies can't learn martian and other alien languages, but we do know that human kids brought up without human contact are unable to develop language (see )--in otherwords, the underlying universal grammar is able to identify and adapt only to "human" languages!
In our zeal to accentuate differences, we fail to note that in the spectrum of possible languages, all human languages form a really tight cluster--and would be seen so by a martian visiting earth..


Tejaswi said...

Great post!

Language is a way of representing and modeling the world. The reason it's considered to be important in our evolutionary history is that it allowed us to collaborate and not invent the wheel from scratch. We also tend to use other modes of modeling and expression like mathematics and art for the same purpose.

If we agree that language is only one modeling approach we use to understand the world, I think the answer to your question is yes, we can learn the Martian language. We developed quantum physics which is somewhat counterintuitive. It should be, since our ancestors did not deal with quasars and black holes. They therefore did not develop intuitions to understand these objects. We however were forced to develop these models when we started observing them. In a way we had to invent a new language to make sense of our new observations. Statistics comes to mind as well.

We have done so all the time. For example folding a paper 50 times makes its thickness greater than the distance between the earth and sun. This is absolutely counterintuitive at first. But it becomes straightforward when we think in terms of geometric progression. A similar puzzle is the reward a king promised to the inventor of chess.

A somewhat related yet tangential issue is touched upon by Richard Dawkins in this talk .

In all I think that nature is more imaginative than us. So there could be another life form which comes up with languages we are incapable of inventing on our own. However, if and when we meet such life forms we should be able to learn their languages.

This article gives how quantum mechanics was invented only when we were pushed for an explanation. I borrowed "nature being more intelligent than us" line from it!

Subbarao Kambhampati said...

Good comments.

One clarification--I am not saying that "we the humans" cannot ever learn martian language, but that human infants, using just their hard-wired prior knowledge--would most likely not be able to learn anything other than a human language.

(of course, this does bring up the deeper question as to when exactly humanity's intellect surpasses that of the scope of the hardwired prior knowledge that evolution gave us...

but thinking about it makes my head hurt..


"If human brain were to be so simple that we can understand it,
we would be so simple we can't"


Tejaswi said...

I'm with you on the head hurting part. It's such a paradox! I sometimes think that researching the human brain (its capabilities and such) really violates the tradition of science -- the observer and the experiment are not disjoint. But then if you're studying someone else's brain works is this research validated. Neuroscience really is like time machine research in some ways. The result is undoubtedly important, but nobody has a clear idea where to start.

In any case the real take-away from your post was the comment on the accentuation of differences we routinely indulge in. I fully agree with you.

Vidya said...

During our discussion in the class about the in-born intelligence in the human infants, I was quite lost when the ~104 possible phonemes were mapped to intelligence.

I think it is more reasonable to attribute an infant's ability to speak all phonemes to it being untrained and unbiased in speech, than to any understanding of grammar. The fact that grownups are linguistically more inept with a few sounds is just a matter of practice and getting used to making a few sounds and never having tried a few others, and might not be because they grew any dumber.

Jay said...

Do you know of any studies or papers which assert that we "forget" these sounds? This statement confused me too. How does the Nicaraguan Sign Language fit into this? What Vidya says makes more sense to me. Kids seem open to more possibilities and they get used patterns which they hear (or see). Then they mimic them and practice them. I know from watching my sons they try very hard to do what we do (they're still young :) ). I'm not discounting the universal grammar theory, but the "tuning" into a language being a matter of forgetting doesn't seem right to me.

In our zeal to accentuate differences, we fail to note that in the spectrum of possible languages, all human languages form a really tight cluster--and would be seen so by a martian visiting earth.

For some perspective on this: prairie dog languages. To us prairie dogs all sound the same, but it seems there are a few dialects. Perhaps to a Martian our human languages would be likewise similar.

Yin said...

My opinion is that the new born babies do not have any linguistic knowledge but they have the ability to learn our human languages. The question if the baby is able to learn the alien language then is changed to the question if the complexity of that language exceeds our abilities. Obviously, if we have 64 letters rather than just a to z, English becomes harder. How about 10000 letters? How could we expect to master a language when our brains have not enough memory to remember that language’s most basic letters? Distinguishing the ability from the knowledge also explains why isolated children can create a new brand language and why wolf children can communicate with their beast parents. It is simply because our abilities allow us to do that.

I do not know what a neural network can do if its nodes are more than the neurons of human brains. However, I believe the machines may be smarter than humans as long as the machines are complex enough.

Subbarao Kambhampati said...

[Response to Vidya, J. and Yin]


the comment on phoneme forgetting is not meant to be the "complete explanation" of universal grammar but rather illustration of a single consequence of it. Universal grammar also entails, in essense, a disjunctive set of grammars, for all possible human languages. The babies then specialize this to the particular language mileau they are growing up in.

(The limited number of phonemes available in the beginning does make the important point that the ingrained linguistic knowledge is not all that vaccuous. It only captures those languages that can be spoken with the set of these available phonemes--even though our vocal chords can make a larger range of sounds..]

Anyways, please read the posting referenced in my original article--

for more on universal grammar etc


shrutigaur said...
This comment has been removed by the author.
shrutigaur said...

When babies come into this world, they possess universal language (not universal grammar, I belive)i.e.their vocal cords can make all possible sounds which any human possibly can. This, in some sense is like A={any phoneme}*

However as the baby grows up, it realizes (as primitive humans must have realised when they felt the need for a finite "human"language)that such a language with so many different words would be unmanageable and have limited practical purpose. We could be better off with lesser words i.e. a more restrictive and more structured language. Every word would have to be mapped to a concept/idea so that everyone exactly knows that a 'pencil' means a 'pencil' and nothing else.Also, it is better if sentences can only be constructed in some particular ways.This is because even if universal language is richer in variety, it doesnt serve the purpose of efficient communication.

So, the baby starts restricting its language/unlearning the unused words/sounds in order to communicate with others. Another thing I feel is that the baby is actually not actually memorizing words or sentences but adapting itself to the structure of the language.It actually learns the grammar of the language and this is the reason why it can make sentences on its own even if it hasnt heard them before.For the first few years, we can assume that the baby learns on the training data-the language supplied by its parents/surroundings and later develops rules based on its understanding.

So, given that the baby's vocal cords can produce the same sounds as the Martians, it would be able to learn their language too. Essentially it is trying to learn the underlying structure of the language.

About the next question, the language children develop in isolation would emerge from what they see/need to communicate in their environment. If the environment is similar to human environment, the language would be similar to humans as well.

Louis Casillas said...

I was looking at a website of the top 20 strange experiments and came upon this one:

"#13: The Ape and the Child

History contains numerous accounts of children raised by animals. The children in such cases often continue to act more animal than human, even when returned to human society. The psychologist Winthrop Kellogg wondered what would happen if the situation were reversed. What if an animal were raised by humans — as a human. Would it eventually act like a human?

To answer this question, in 1931 Kellogg brought a seven-month-old female chimpanzee named Gua into his home. He and his wife then proceeded to raise her as if she were human, treating her exactly the same as they treated their ten-month-old son Donald.

Donald and Gua played together. They were fed together. And the Kelloggs subjected them both to regular tests to track their development. One such test was the suspended cookie test, in which the Kelloggs timed how long it took their children to reach a cookie suspended by a string in the middle of the room.

Gua regularly performed better on such tests than Donald, but in terms of language acquisition she was a disappointment. Despite the Kelloggs's repeated efforts, the ability to speak eluded her. Disturbingly, it also seemed to be eluding Donald. Nine months into the experiment, his language skills weren't much better than Gua's. When he one day indicated he was hungry by imitating Gua's "food bark," the Kelloggs decided the experiment had gone far enough. Donald evidently needed some playmates of his own species. So on March 28, 1932 they shipped Gua back to the primate center. She was never heard from again."