Field of Science

We know toddlers can't count, but are they good at statistics?

Toddlers can't count. And, to be honest, statistics is the one field of math that has never really clicked for me. But there is mounting evidence that children and adults are very sensitive to the statistical nature of the world.

This has been shown in vision studies, some of which I've worked on, but which I won't talk about more here (but check out Brian Scholl's lab, among others). 

Statistical learning has also been found for language-like material. Certainly, statistics could help with learning language. For instance, a given words (e.g., the) is more likely to be followed by some words (e.g., book or table) than other words (e.g., contemplate or earn). Moreover, rare sound combinations typically mark boundaries between words (no word contains the sound combination thst, but it can occur -- rarely -- between words, such as with steam). Babies could, at least in theory, use that information to break the sound stream that they hear into words (contrary to popular imagination, people rarely speak in single words, even to infants, so determining that winsome is one word but withsome is two is non-trivial).

So, statistical information that could be valuable for language learning is available in what babies hear. A number of studies have shown that babies are in fact sensitive to such information (check out Gomez & Gerken, 2000, for a review). Whether or not statistical learning is actually used in real live language learning as opposed to the laboratory experiments just mentioned is an open question, but it certainly could be.

The issue that is bothering me lately is that I'm not aware of any evidence that infants are better at statistical learning than adults. Given that children tend to be much better language-learners than are adults, this raises an important question: what aspect of language learning are adults bad at? (Some people think that adults aren't impaired at learning languages; I don't think that's true and may eventually write a post about it.) The evidence lately seems to be that adults are bad at syntax, so this makes me wonder just how likely it is that statistical learning is used to help learn syntax, as some people have claimed.



Gomez, R.L., Gerken, L. (2000). Infant artificial language learning and language acquisition. Trends in Cognitive Sciences, 4(5), 178186.

6 comments:

ajowen said...

Consider the way that accumulating statistics might interfere with your ability to form new patterns. That is, it's not that children are better at statistics than adults. It's that children have less accumulated experiences to interfere with forming the statitical relationships.

See Kuhl, P. K. (2004). Early language acquisition: Cracking the speech code. Nature Reviews Neuroscience, 5, 831-843.

Amanda Owen (U of Iowa)

coglanglab said...

Seems reasonable, but as I said, why isn't there evidence of this? Adults seem to be very good at statistical learning. Where is the evidence of interference (or children's lack of interference)?

I'm not sure anybody has been pushing this angle, because I think the people working on statistical learning are for the most part interested in other things. But *I* am interested in what makes kids different from adults, and it would be nice to have data that speak directly to that point.

Which is also not to say that there isn't any. I just don't know of it. If somebody does, please let me know!

Anonymous said...
This comment has been removed by a blog administrator.
Coppe said...

There are basically two important research results in this respect, to my mind.

One comes from the series of Saffran et al. (1996; 1999) studies that show that infants have an amazing ability to use statistical mechanisms to segment words.

The other comes from a study by Johnson & Jusczyk (2001) which shows that when infants are presented with an abundance of statistical cues AND prosodic cues, they prefer to use prosody to segment words.

In other words, children have the ability to use powerful statistical learning mechanisms (in which they could differ from adults, but not necessarily) and they have extensive innate knowledge of prosodic structure (in which they definitely differ from adults, because they proceed from a different hypothesis space).

Yang (2004) is a must-read on this. He also provides some compelling empirical arguments why innate knowledge is necessary.

E.K. Johnson and P.W. Jusczyk, Word segmentation by 8-month-olds: when speech cues count more than statistics, J. Mem. Lang. 44 (2001), pp. 1–20.
J.R. Saffran et al., Statistical learning by 8-month old infants, Science 274 (1996), pp. 1926–1928.
J.R. Saffran et al., Statistical learning of tone sequences by human infants and adults, Cognition 70 (1999), pp. 27–52.
C. D. Yang, Universal Grammar, Statistics or Both?, Trends in Cognitive Sciences 8 (2004), pp. 451-456.

Coppe said...

Actually, here is an interesting article by Newport and Aslin (2000) in which they report finding that adults and children perform roughly the same in word segmentation using only statistical mechanisms. In fact, the adults perform a little better.

E. L. Newport and R. N. Aslin, Innately Constrained Learning: Blending Old and New Approaches to Language Acquisition, in S. C. Howell, S. A. Fish, and T. Keith-Lucas (eds.), Proceedings of the 24th Annual Boston University Conference on Language Development (2000). pp. 1-21. Somerville, MA: Cascadilla Press.

Chris said...

This is really a god question to ask, My son is just three, he's good at counting, My hubby is a teacher so he teaches our son, in fact he started teaching our son one year back, so our kid is quite good, can't really answer whether my son is good at statistics as well or not.

Medela shoulder Bag