Formalizing accessibility using linguistic productivity
Kyle Mahowald, J. Tenenbaum, T. O’Donnell
MIT Press
Abstract
Tversky and Kahneman famously claimed that people use the availability heuristic to reason about the probability of events, relying on the ease with which information about the events can be retrieved from memory—a property known as accessibility (or sometimes availability)—to estimate probability, rather than using more veridical mechanisms. As an example of this phenomenon, Tversky and Kahneman showed that people guess that word patterns that include an English suffix (e.g., _ _ _ _ i n g) occur with a greater number of English words than patterns consisting of non-linguistic sub-sequences, (e.g., _ _ _ _ n _ ), despite the fact that any word matching the former must necessarily match the latter. T&K’s explanation is that this discrepancy results because -ing is a linguistic unit—a suffix—and, therefore, more accessible in memory than the non-linguistic sequence _ n.. In this paper, we propose that the accessibility of linguistic units should be related to their productivity–how readily the units can be combined to form novel expressions (e.g., pinescented/pine-scentedness). Since highly productive units, such as the suffix –ing, are frequently needed as independent units during the production and comprehension of novel words (e.g., pine-scentedness) they should be more accessible in memory than less productive suffixes, which are typically only accessed as parts of words (e.g., -th in warmth). We tested this idea in a large-scale behavioral experiment using T&K’s paradigm with a variety of English suffixes of differing levels of productivity. We found that people systematically overestimated the frequency of word patterns that contained English suffixes and that the rate of overestimation was related to the suffix’s productivity, such that highly generalizable suffixes (like -ness) led to more overestimation than less productive suffixes (like –th; warmth). This result provides support for the idea that (a) morphological productivity can be measured in behavioral paradigms and that (b) morphological productivity is connected to traditional notions of accessibility.