Production of constrained associates and the informational uncertainty of the constraint.

When a subject attempts to list, as rapidly as possible, words falling in some specified category (e.g. well-known artists), his rate of listing words begins at a relatively high and stable level and then gradually declines as he progressively exhausts the specified category. Indeed, Bousfield and Sedgewick concluded that the rate of producing words at any given time is proportional to the number of words in the specified category that have not yet been produced.1 At least on the average, this relation may hold across different categories as well as within the same category. Certainly the initial rate of producing words from a large category (like 'famous people') is generally greater than the initial rate for a smaller category (like 'famous artists'). Now, the uncertainty associated with N equiprobable alternatives has been defined as log2N. Usually, however, the category is so large that a determination of the number of words, N, by complete exhaustion of the category is not feasible. To the extent, however, that the initial rate of listing is determined by N, that number, and hence the uncertainty, could be estimated from the initial rate. For example, the degree of statistical constraint exerted upon a given word by its surrounding context could be determined simply by (a) deleting the given word from its context and (b) measuring the initial rate at which Ss produce guesses as to what the missing word might be. Unfortunately, however, the alternative words are not usually equiprobable; some artists are more famous than others, for example, and would therefore tend to be listed first. Accordingly, this study attempts a combined theoretical and experimental examination of how initial rate depends upon uncertainty when the variation that exists in the probabilities of the alternative words is taken into account.