Abstract Summary
In this study, we combined corpus analyses and memory recall experiments to show that experience with natural language input influences chunking ability. As predicted, our results indeed suggest that our ability to use chunking to process linguistic input online is to a large degree affected by our experience with language.
Abstract :
Our ability to chunk information is often thought to be independent of experience, depending largely on biological constraints, as reflected, for example, by memory measures such as the digit span task. Here we combine corpus analyses and memory recall experiments to show that experience with natural language input influences chunking ability. We used the Corpus of Contemporary American English to extract trigrams of digits (e.g., 911) and consonants (e.g., nsw) that were either of High or Medium frequency. Experimental recall strings were composed of three such trigrams of the same character type (e.g., n s w m p r n c t). Matching control strings consisted of the same characters but pseudo-randomized to remove statistical information (e.g., w s r t p m n c n).
Recall performance was found to be significantly better on experimental recall strings compared to matching pseudo-randomized control strings, both for trigrams of digits and consonants. However, this random vs. experimental string distinction was found to be only relevant in predicting recall performance on High frequency trigrams of digits, whereas for trigrams of consonants it was relevant in predicting recall performance on both Medium and High frequency items. Thus, recall was improved for both High and Medium frequency chunks (i.e., whole trigrams) of consonants, whereas recall was only improved for High frequency chunks of digits. Collapsing ‘chunk’ recall across character type, the data indicates that the random vs. experimental distinction was less relevant in predicting recall performance on the Medium frequency items. Additionally, significantly more digit trigrams were recalled than consonant trigrams.
Overall, our results suggest that the ability to use chunking to process linguistic input online is to a large degree affected by our experience with language. We also observed significant individual differences in chunking ability which may predict variation in language processing skills.
