Eureka moments of insight, rather than repeated exposure, appears to be how children learn new words, according to new research from the University of Pennsylvania published recently in the Proceedings of the National Academy of Sciences.
The researchers speculate that these, and future findings from their work, may overturn current theories of “associative learning”, and that spending time with children in their natural environments rich in stimuli is better than using simple object by object picture books for learning new words.
They write that current theories maintain that children learn their first words by being aware of the associations and various contexts in which they become exposed to them and then over time, through a series of statistical “cross-tabulated” comparisons, they narrow down to the common element.
But as a result of their findings the researchers suggest that what really happens is that children have a moment of insight, a “eureka” moment, about the meaning of a word, even if it is not absolutely clear what that is, and they cling onto this until subsequent exposures disprove it; they appear to use a “one-trial ‘fast-mapping’ procedure, even under conditions of referential uncertainty”, write the researchers.
First author and postdoctoral fellow at Penn, Dr Tamara Nicol Medina, told the press that the current theory is “appealing as a simple, brute force approach”.
“I’ve even seen it make its way into in parenting books describing how kids learn their first words,” she added.
Co-author and psychology professor at Penn, Dr Lila Gleitman, said:
“This sounds very plausible until you see what the real world is like.”
“It turns out it’s probably impossible,” she explained.
Experiments that back the associative word learning model usually expose the subject to a series of two, three or four pictures of various objects, shown against a neutral background. But the researchers argue that in the “real world”, the contexts are more complex and less certain than this: there are an infinite number of possible “referents” for a word, and they can change in type and appearance from exposure to exposure, and not all are present each time.
The authors have long maintained, with other psychologists and linguists, that the vast number of statistical comparisons that would have to be made in order to learn words in this association model, would be beyond the capability of human memory.
If you were to set some of the most sophisticated computers a task like this, they would have to use shortcuts and would not guarantee optimal learning, they say.
Co-author and Penn psychology professor Dr John Trueswell said that this doesn’t mean humans can’t use the statistical approach at all in learning, “only that we do this kind of tracking in situations where there are a limited number of elements that we are associating with each other”.
“The moment we have to map the words we hear onto the essentially infinite ways we conceive of things in the world, brute-force statistical tracking becomes infeasible. The probability distribution is just too large,” he explained.
For their study, the researchers carried out three experiments. In all three experiments, the subjects (adults and pre-school children), watched short, 40-second, videos of parents interacting with their children.
The subjects watched the videos with the sound turned off. The only time they heard a sound was when the parent spoke the target word: in the first experiment the sound was a “beep” noise, and in the second and third experiment it was a nonsense word.
The purpose of the first experiment was to find out how informative the short video scenes were at giving a sense of the meaning of the target word from its context.
If more than half the subjects were able successfully to guess the target word, the short video scene was classed as HI: High Informative.
If fewer than one third of the subjects were able successfully to guess the target word, then the scene was classed as LI: Low Informative.
The first experiment showed that out of 288 short video scenes, only 7% were HI and 90% were LI.
In the second experiment the order of HI and LI video scenes was carefully arranged so that sometimes an HI scene for a word was viewed early in a sequence, and sometimes it was shown later.
Trueswell said that previous studies tended to use artificial stimuli (pictures of objects on a neutral background), with a much smaller number of options for each word. And they also only only measured the final outcome, that is whether the subject was able successfully to guess the word or not.
But in this study they measured the success after each observation, so that they could get a sense of whether the subject got the meaning of the word as a result of a cumulative process or from a moment of insight: a “eureka” moment.
As Trueswell explained they could look at the “trajectory of word learning throughout the experiment, using natural contexts that contain essentially an infinite number of meaning options”.
The results strongly suggest that the “eureka” moment was the most likely way that subjects learned the meanings of the target words. Repeated exposure did not lead to improved accuracy, pointing away from association hypotheses as the most plausible explanations.
In support of this conclusion was the fact that when an HI video scene occurred early in the sequence for a given target word, the final guesses improved. The researchers said this was because early HI scenes gave the subjects the best opportunity to learn the correct word, and in fact most of them did guess correctly in those instances. Thus starting on the right track, and viewing subsequent scenes, helped them to “lock in” the correct meaning.
Gleitman said it was as though you “make something like an insightful conjecture” when you know the evidence is good.
When subjects were first presented with LI scenes (and the HI scenes were much later in the sequence for a target word), they tended to give incorrect guesses, and their final guess was also more likely to be wrong, even though they changed them as the experiment went on.
The researchers said this showed that the subjects did not remember plausible alternative meanings, including the correct one, that they may have seen earlier in the sequence.
The third experiment was done a few days later. The subjects were shown again the video scenes of the words they had guessed wrongly in the previous experiment. They were not able to recall their incorrect guesses and the researchers concluded this showed that “forgetting” incorrect meanings is important to being able to learn correct meanings of new words.
“All of those memories go away,” said Gleitman “And that’s great!”
Memory failure, that is not remembering the incorrect meanings, is “rescuing you from remaining wrong for the rest of your life”, she explained.
As a next step the Penn team wants to explore what makes some interactions informative while others are less so, and what order people process visual information.
They think their work will show that having rich interactions with children, with lots of patience, in their everyday environment, is a more fruitful way to help them learn words than with abstract picture books and “drills”.
“How words can and cannot be learned by observation.”
Tamara Nicol Medina, Jesse Snedeker, John C. Trueswell, and Lila R. Gleitman.
PNAS published ahead of print 16 May 2011.
DOI:10.1073/pnas.1105040108
Additional source: University of Pennsylvania.
Written by: Catharine Paddock, PhD