How to Learn a Language Quickly
Children learn the meanings of about ten words per day, but it isn’t clear which techniques they use to achieve this fast rate. A research team simulated word learning and showed that a specific strategy, where the learner assumes there are no exact synonyms, is so effective that it can reduce the total learning time to the shortest time possible, which is just as soon as every word has been heard at least once. The results may give insight into the development of language in human ancestors.
A typical child learns approximately words by the time she is . Children use many strategies to identify word meanings, including techniques to deal with ambiguous situations. For example, a child hears the word “cup” and at the same time sees a cup, a ball, and a book. She might remember this experience the next time she hears “cup” in conjunction with a cup and a different set of objects (the “confounders”). If the cup was the only object present in both situations, the child learns that “cup” means cup.
If the child further assumes that there is only one name for each object (meanings are mutually exclusive), then she can learn words faster. For example, if she hears “cup” and already knows the meanings of “ball” and “book,” the two other objects present, then she learns immediately that “cup” refers to the only non-assigned object. “It’s a boot-strapping technique, where you use information from previous learning [of words] to eliminate certain meanings,” explains Richard Blythe of the University of Edinburgh in the UK. Small-scale lab tests have shown that children and adults use mutual exclusivity to determine word meaning . But researchers don’t know how effective this strategy is compared with others when dealing with hundreds or thousands of words.
To address this question, Blythe and his colleagues used a physics analogy that others have exploited in the past: word learning resembles some problems in nonequilibrium statistical physics, where a large number of entities (such as molecules) interact, and the probability distributions for certain states evolve over time. In language learning, a word like “cup” will start off with many confounders, and so the probability of “cup” meaning cup will be low. But over time this probability—and that of other word-meaning pairs—will grow to one, analogous to the system approaching equilibrium.
The researchers at first assumed a language, or “lexicon,” with or words, which appear with a range of different frequencies. In their computer simulations, the “learner” is repeatedly presented with a single word and a set of “objects,” one of which is the target meaning and the rest of which are confounders. The learner gradually learns the words by comparing many of these events. The team mathematically derived the total learning time for the entire lexicon, and it was strongly dependent on the number of confounders presented in each event.
Blythe and his colleagues compared two cases. The first assumed simple elimination without mutual exclusivity, in which the learner still considers already-named objects as potentially correct meanings for a new word. Learning words with this strategy would take more than a lifetime, they found, unless the number of confounders was less than a few. When the team included mutual exclusivity in the model, they found that the learning time dropped dramatically. For a modest number of confounders (around ten), the entire lexicon was learned in the minimum time it takes to hear every word at least once. Words were learned nearly as quickly as they were encountered, suggesting that the mutual exclusivity assumption is extremely effective. The authors speculate that acquiring this word learning strategy may have been an important step for early humans as they developed their language ability.
Linda Smith, a cognitive scientist from Indiana University in Bloomington, says that mutual exclusivity is a common theme in brain studies. “Competition is how the brain works—in all domains, at all levels,” she says. If the brain forms an association between a word and an object, this will inhibit other words from forming a similar association with the same object. She expects some psychologists will take issue with the idea that learners retain a set of confounders for each word from one utterance to the next , but she says that similar kinds of ambiguity are included in theories of the brain’s memory retrieval.
Michael Schirber is a Corresponding Editor for Physics based in Lyon, France.
- E. M. Markman and G. F. Wachtel, “Children’s Use of Mutual Exclusivity to Constrain the Meanings of Words,” Cognitive Psychol. 20, 121 (1988)
- T. N. Medina, J. Snedeker, J. C. Trueswell, and L. R. Gleitman, “How Words Can and Cannot Be Learned by Observation,” Proc. Natl. Acad. Sci. U.S.A. 108, 9014 (2011)
Learning meaning: Why language learning seems hard (book chapter by Michael Gasser of Indiana University)