I’m in the midst of revising a grant proposal on Ganong effects with my colleagues Adrian Staub and Andrew Cohen, of PBS UMass, and Amanda Rysling, for this week only still Linguistics UMass but as of 1 July 2017 Linguistics UC Santa Cruz. Congratulations Amanda! The crux of the proposal is that there is no Ganong effect, at least in the sense that a listener’s knowledge of words always and necessarily influences their performance in a phoneme categorization task in which one category makes a word with the rest of the string and the other does not. The reason for denying that the effect exists in this sense is that it comes and goes, and when it comes it varies in size, and is often absolutely tiny even when it comes. There is some evidence that the Ganong effect bigger when the categorization task is made harder, either by degrading the stimuli or by imposing a cognitive load on the listener via a simultaneous task. Those findings suggest that otherwise its size is small, perhaps even vanishingly small because the signal suffices in laboratory listening conditions, and the listener therefore needs no help from their linguistic knowledge. But even the effects of degrading the signal or imposing a cognitive load aren’t consistent.
But this presents us with at least a rhetorical problem: what general theory captures such an evanescent effect and accounts for the variation in its size? Our current strategy, which I’m convinced is the right one, because we don’t have the results we need yet, is to say that until we have the results of the proposed experiments, we cannot propose or even outline any general theory.
Nonetheless, here are some ideas. (1) A listeners’ linguistic knowledge is likely to create a response bias for the word-making category when the target sound is final in the string, because the listener has already heard all the sounds but one that are necessary to activate a word, (2) their linguistic knowledge is instead likely to influence evidence accumulation when the target is initial, as they have to wait until all the following sounds have been heard to activate a word beginning with one category and not the other. This predicts that linguistic knowledge should interact with response times differently for initial and final targets. (3) Degrading the signal with noise should increase reliance on lexical knowledge for initial and final targets similarly because it hinders recognition of the sounds that would activate a word. (4) How a cognitive load influences the size of the effect may depend on whether the load consists of a linguistic task.