Postersession 2
Poster #: 68
Topic: Memory and perception
Thursday, Sep 10, 2015
1st floor

Sensitivity to the statistics of rapid, stochastic tone sequences

Sijia Zhao1, Marcus Pearce2, Frederic Dick3, & Maria Chait1

1UCL Ear Institute, University College London, London, United Kingdom
2School of Electronic Engineering and Computer Science, Queen Mary University of London, London, United Kingdom
3Birkbeck-UCL Centre for NeuroImaging, Birkbeck, University of London, London, United Kingdom

Accumulating MMN work suggests that the human brain is remarkably sensitive to patterns in sound. However, much of this research has focused on deterministic patterns. This study investigates what statistics of random sequences are acquired by listeners, the time-scales associated with these processes and the underlying brain mechanisms. The paradigm measures listeners' ability to detect transitions (changes in statistics) within stochastic, 50ms tone-pip sequences. The sequences are too rapid for conscious scanning thus tapping automatic statistical learning processes. Response measures are hit-rates and response-time (how much information, in terms of number of tone-pips, was accumulated by listeners before making a response) and are compared with an ideal-observer model.

Tone frequencies are drawn with replacement from a fixed pool of 20 values between 200-2000Hz. The basic signal is a random sequence of these values (alphabet size = 20; RAND20). In Experiment1 we measure sensitivity to reduction in alphabet size (RAND20 to RAND2, RAND5 or RAND10), demonstrating that performance declines monotonically within this range. In Experiment2 we focus on transitions from RAND20 to RAND10 while varying the statistics of the RAND10 sequence so tones are drawn from the 10 highest frequencies (RAND10_H), 10 lowest values (RAND10_L), 10 middle values (RAND10_M), 10 edge values (5 highest & 5 lowest; RAND10_E), or sampled equally from the entire range (RAND10_A). Performance reveals tuned sensitivity to the mean, variance, and step-size between elements of random sequences. M/EEG data, reflecting brain response dynamics evoked by these sequences will be reported.