High-capacity auditory memory for vocal communication in a social songbird

Abstract

Effective vocal communication often requires the listener to recognize the identity of a vocalizer, and this recognition is dependent on the listener's ability to form auditory memories. We tested the memory capacity of a social songbird, the zebra finch, for vocalizer identities using conditioning experiments and found that male and female zebra finches can remember a large number of vocalizers (mean, 42) based solely on the individual signatures found in their songs and distance calls. These memories were formed within a few trials, were generalized to previously unheard renditions, and were maintained for up to a month. A fast and high-capacity auditory memory for vocalizer identity has not been demonstrated previously in any nonhuman animals and is an important component of vocal communication in social species.

Fig. 1. Learning ladder for assessing auditory memory capacity.

(A) The structure of a single trial. Subjects initiate a trial by pecking a key. A randomly chosen 6-s stimulus file is then played (20% of trials are rewarded, and 80% of trials are nonrewarded). If the stimulus is interrupted by another peck on the same key before the 6-s playback is completed, then a new trial is immediately initiated. If the stimulus is not interrupted and the stimulus is in the rewarded group, then the subject receives 12 s of seed access from a mechanical food hopper. (B) The learning ladder procedure gradually introduces new rewarded and nonrewarded vocalizers to the stimulus set each day. Ten stimuli are used for each vocalizer and vocalization type. Each stimulus is, in turn, composed of random sequences of renditions of DCs or songs sampled from our repertoire library for that vocalizer (see also fig. S1 for full-size exemplar spectrograms). (C) The lines show the probability of stimulus interruption of individual vocalizers by a single subject in 20 trial bins (blue, rewarded; red, nonrewarded). Tick marks above the plot indicate interrupted trials, and those below the plot indicate noninterrupted trials. (D) Average odds ratio (OR) for song and DC assessed after training, on days 4 and 5, for all subjects (n = 19). Birds perform better on songs (OR, 15.5; 95% CI, 9.9 to 24.4) than on DC (OR, 8.4; 95% CI, 5.6 to 12.9) (P = 0.004, log-transformed paired t test). Error bars show 2 SEM.

Fig. 2. Memory capacity for vocalizer identity for all subjects.

Discrimination performance per vocalizer and subject (n = 19) for songs (A), DCs (B), and both songs and DCs (n = 4) (C). The mixed condition (C) was performed by four subjects who were additionally tested with a total of 56 vocalizers: 24 vocalizers of DCs and 32 vocalizers of songs. For each subject (white/gray plot background), the dots indicate the OR of interrupting a given vocalizer. Red dots correspond to nonrewarded vocalizers (NoRe) and blue dots to rewarded vocalizers (Re). The number of vocalizers that are discriminated significantly above chance (P < 0.05, controlling for false discovery rate using Benjamin-Hochberg procedure) are indicated above each subject’s plot (maximum number of vocalizers are 12 for DCs, 16 for songs, and 56 for the mixed condition). Note that the order of the dots on the x axis is random and that the rewarded and nonrewarded vocalizers are not paired. Error bars correspond to the one-sided 95% CI (Fisher’s exact test). OR of 1 corresponds to chance. Error bars for nonrewarded stimuli are generally smaller because they are played more frequently. The same data are shown in terms of probabilities in fig. S3.

Fig. 3. Speed of memory acquisition.

(A and B) Learning rates are analyzed by plotting the behavioral response (probability of interruption) as a function of informative trials (see Results section) for rewarded (blue) and nonrewarded vocalizers (red). (C and D) The separation between the red and blue curves in A and B quantifies the learning and is shown in C and D as an OR of odds for nonrewarded divided by the odds of rewarded as in Fig. 1D [(C), song; and (D), DC]. Shaded regions show 2 SEM. Asterisks indicate region where OR was significantly greater than 0 (n = 19, P < 0.05, false discovery correction).

Fig. 4. Generalization and long-term memory.

(A and B) The plots show the average probability of interruption across all subjects (n = 19) for each of the 10 renditions the first time they are heard by the subject; the renditions are ordered on the x axis according to the presentation order. Error bars are 2 SEM. (C and D) Interruption rates for nonrewarded and rewarded vocalizers in two subjects (S1 and S2 of Fig. 2) during three epochs for songs (left) and DCs (right). The three epochs shown are Naïve (initial exposure to the stimuli), Learned (last two sessions of initial learning ladders), and Month later (1 month after Learned without any reinforcement). The interruption rates to a particular vocalizer are restricted to trials before the second informative trial of that vocalizer during the relevant epoch. Asterisks indicate epochs during which nonrewarded stimuli were interrupted at a significantly higher rate than rewarded stimuli (P < 0.05, one-sided t test). Error bars indicate 2 SEM. n.s., not significant.