Computing reverse associations
Having seen that word associations with single stimulus words can be computed with a quality similar to that achieved by human subjects, let us now turn to the main question of this paper, namely whether it is also possible to reverse the task, i.e. to compute a stimulus word from its associations.
Let us look at an example: According to the Edinburgh Associative Thesaurus, the top three most frequent responses to clown are circus (produced by 26 out of 93, i.e. 28% of the test persons), funny (9% of the test persons) and nose (8% of the test persons). The question is now: given only the three words circus, funny and nose, is it possible to determine that their common stimulus word is clown? And if it is possible, what would be the quality of the results?
The above is an illustrative example, but, in other cases, it is often more difficult to guess the correct answer. To give a feeling for the difficulty of the task, let us provide a few more examples involving varying numbers of given words, with the solutions provided in Table 4.4:
apple, juice ^ ?
water, tub, clean ^ ?
grass, blue, red, yellow ^ ?
drink, gin, bottle, soda, Scotch ^ ?