(1) Recent exposure to a word tunes auditory spectro-temporal receptive fields (STRF) to respond maximally to that word. This effect

(2) is reflected in the EEG as a better entrainment of the EEG to the envelope of this word, and

(3) facilitates subsequent recognition of the same word.



Experimental design

4 subjects participated in the psychophysical experiment. Since one of them used a different set of SNR values for target presentation, his results are not included in the group result plots.

Stimuli were taken from a recorded database of 100 2-syllable words with similar endings., pronounced by a female voice. Each word item was used several times (between 2 and 7) within the same experiment , and was used indifferently as prime, as target, or as both in a given trial. Words occurred on a continuous background pink noise. Primes were always played with a +5 dB signal-to-noise ratio (SNR); the SNR of the targets could take one of five values (-21 dB, -18 dB, -15 dB, -12 dB, and -9dB for the 3 subjects included in group analyses) with equal probabilities.

A central, white fixation cross was continuously presented on the screen faced by the participant. The time course of an individual trial was as follows: the prime was presented 5000 ms after the beginning of the trial; 1250 ms after prime onset, the cross turned blue, indicating target interval; stimulus-onset asynchrony (SOA) between prime onset and target onset was uniformly distributed between 1750 ms and 2750 ms; 1500 ms after target onset, the fixation cross turned white again and 3 word alternatives were displayed as response options. Participants indicated their choice by pressing one of the 3 keys corresponding to the 3 options. After 3000 ms, the response interval timed out and the next trial started.

There were distinct conditions regarding to the matches between prime, target, and response options. In all conditions, the target was always proposed as one of the three options. In condition 1, prime and targets were made of the same word (valid priming), and the 2 remaining options were drawn randomly from the list of remaining 99 words. In condition 2, prime and target were different words (invalid priming), prime was proposed as one of the options, and the remaining option was drawn randomly from the list of remaining 98 words. Conditions were randomized across trials, and, options order was randomized within each trial. Probabilities of conditions 1 and 2 were respectively 1/3 and 2/3. This way, the presence of the prime among the options was uninformative about the identity of the target.

In order to control for a response bias consisting in choosing the prime, we introduced a third condition for one of the subject (the last one to perform the task): in condition 3, prime and target were different words (invalid priming), but the prime was not proposed as an option; instead, the 2 remaining options were drawn randomly from the list of remaining 99 words. For the concerned participant, probabilities of conditions 1, 2 and 3 were respectively 1/5, 2/5 and 2/5. Behavior analysis and results

For each SNR we investigated whether choice accuracy was higher on trials with valid priming than on trials with invalid priming. When this was the case, two effects were susceptible to account for this effect of priming on performance. One was a facilitation effect on the sensory processing of the target. The other one was a response bias that consists in choosing the prime in trials where it is proposed as an option, especially if the target was not recognized.

In order to correct for response bias, we computed d’ estimates for the subject who encountered all 3 conditions. We counted, within condition 2, the number of trials where the subject selected the (always invalid) prime. The ratio of this number over the total number of condition 2 trials served as an estimate of the false alarm rate in condition 1 and 2. Finally, the hit rate for valid priming was estimated as the proportion trials with correct responses in condition 1, and the d’ for valid priming was computed as: 〖d'〗_valid=z(〖hit rate〗_(cond.1) )- z(〖false alaram rate〗_(cond.2)), where z is the z-score. Since no false alarms were possible in condition 3, the d’ for invalid priming was computed as: 〖d'〗_invalid=z(〖hit rate〗_(cond.3) )

In terms of d’, priming drastically improves performance at -15 db and – 18 dB SNRs. Importantly, priming has no more effect on discrimination performance when SNR is as low as -21 dB.

EEG (2 participants)

During EEG recordings, we used the same stimuli and same presentation design as in the psychophysical experiment, except that only one SNR of -15 dB was used for target presentation. Besides, participants were instructed to minimize motor activity and to fixate the central cross during the whole duration of the task apart from response intervals, and to take advantage of response intervals to blink and/or move.