Selective Entrainment as a Neural Mechanism for Attention

Attending selectively to one stream of input when there is competing irrelevant stimuli is challenging, as exemplified in the well known “cocktail party” problem. Many natural stimuli one might encounter in “cocktail party” situation, such as speech and music, have an underlying rhythmic structure which could facilitate attentional selection. Here we examined the contribution of rhythmicity and temporal predictability to the ability to perceptually select one of 2 competing streams and the neuronal mechanisms underlying this ability. We simulated a “cocktail party of tones” by simultaneously playing two streams of simple tones, with different pitches. Participants were instructed to attend to either high or low pitched tones and detect target sounds. We varied the degree of temporal predictability within each stream from strictly rhythmic to completely random, allowing us to determine the contribution of rhythmicity for selective attention. We recorded intracranial ECoG from 7 epileptic patients, implanted clinically for pre-surgical evaluation. The behavioral results demonstrate facilitation of target detection in a rhythmic vs. non rhythmic context, supporting the assumption that temporal predictability assists stimulus processing and selective attention. Analysis of the ECoG responses indicates several differences between processing rhythmic and non-rhythmic streams. First, the auditory evoked responses were more pronounces in the rhythmic condition. Second, in the rhythmic condition we found selective enhancement of rhythmic neuronal activity at the rate of the attended stream. Third, this selective rhythmic activity was observed in a very broad array of brain regions, greatly exceeding the area where auditory evoked responses are observed. These findings support an “entrainment selection hypothesis” suggested previously by Schroeder and Lakatos (2008), which posits that when there is clear rhythmic structure to an attended input stream, the neuronal oscillations in the relevant frequency band entrain to the stimulus so that events within the stream fall on optimal excitability phases of the entrained oscillations. In contrast, when there is no temporal predictability, events in the stream will fall randomly on optimal and non-optimal excitability phases of ambient oscillatory activity, resulting on average, in a decreased sensory response and decreased performance. Our finding of selective entrainment in a wide spread array of regions further suggests that this phenomenon engages higher order areas in addition to sensory areas, and serves to dissociate the effect of selective entrainment from auditory evoked responses.