Learning about word learning in ambiguous situations

2014-07-14How learners map words to meanings is a central question in language acquisition. Yu & Smith (2007) presented evidence that humans are able to use cross‐situational statistics in ambiguous referential situations. According to this their account, so long as the word and its referent co‐occur, this sensory co‐occurrence of the word and its referent can induce robust word learning. Indeed, the authors claim there is “simple associative learning” from “the sheer amount of statistics”. To what extent is learning words simple and robust? ❧ To test the limits of associative learning, I designed an experiment that mirrored Yu & Smith’s, but added an additional manipulation. During a training phase, I presented college students with a series of trials in which two objects from a set are paired with a nonsense word, similar to Yu & Smith. I also present trials with two objects from another set of objects paired with a beep (interleaved with the word trials). One group of subjects was instructed that they were to figure out the meaning of the words, and I replicate Yu & Smith’s finding of robust word learning. However, another group of subjects was told that they needed to categorize images as ones that occurred with beeps or with words. When tested on word‐object associations, those subjects showed no evidence of making the word‐to‐object associations, despite identical exposure as the first group. ❧ I conclude that simple associative learning would be unable to produce robust learning unless children are engaged in active hypothesis testing (Trueswell et al., 2013). Without hypothesis testing, word learning from associative learning is weak at best