A total of 26 male zebra finches (Taeniopygia guttata) from our breeding facility were used, 5 were lesioned, 5 were used as controls, 4 were used as tutors, and the other 12 were unfamiliar adult birds whose previously recorded songs were compared with those of the juveniles.
Birds: Birds were normally raised with both parents until 35-40 days post hatch (dph), giving them enough exposure to song for successful memorizationBöhner 1990 . At 41 ± 1 dph (range 40-42 dph, N = 5), birds received large bilateral excitotoxic lesions targeting NCM; lesions were made as previously describedCanopoli 2014 . Thereafter, control and lesioned animals were housed in sound attenuation chambers (IAC GmbH) on a 14/10 hours day/night schedule with ad libitum access to food and water.
Surgeries: Briefly, birds were anesthetized with isoflurane and head fixed in a stereotaxic apparatus. In each hemisphere they received 500 nl ibotenic acid solution (7 mg/ml in ddH2O) delivered either in a single or in two injections between 1,500 and 2,500 μm ventrally from the brain surface (two injections to extend the lesioned area in the ventral direction). Injections were made at about 750 μm anterior to the bifurcation of the mid-sagittal sinus (751±48 μm, N = 5), and about 450 μm lateral from the midline (438±76 μm, N = 5). This same injection protocol was previously used in adultsCanopoli 2014  and proved to be effective in lesioning a volume of NCM comparable in size to the injected amount of neurotoxic solution. The acid volume we injected per hemisphere (500 nl) is equally large as volumes injected inLondon 2008  (500 nl) and inCanopoli 2014  (500 nl), and much larger than volumes injected inGobes 2007  (110 nl). All birds resumed normal behavior within a few hours after surgery. After observing that lesions did not cause any major learning impairment, we decided that sham injections as controls would be unnecessary.
Histology: All birds with lesions were perfused at either 76 dph (N = 2) or 80 dph (N = 3). Birds were given an overdose of sodium pentobarbital, followed by perfusion via the left ventricle first with ringer solution, followed by a 4% paraformaldehyde solution. To visualize cell bodies, sagittal sections were cut from fixed brains and Nissl stained with cresyl violet. In Figure 1a the volume of the injected acid is shown to indicate the initial lesion size. Based on the large volume and medial-posterior location of our injections we estimate that our lesions encompassed the regions targeted in previous NCM studiesChew 1995 Bolhuis 2000 Gobes 2007 London 2008 .
Song motif selection and analysis: From all birds, we randomly extracted 20 song motifs recorded typically during the first few morning hours. All juveniles' song motifs were recorded at 76.6±2.5 dph (range 72-82, N = 27). For sufficiently developed songs (all adults and most of the juveniles), a stereotyped sequence of syllables (a song motif) was identified and only renditions of that sequence were extracted for the subsequent analysis. In poor learners and in birds in an early stage of song development, a stereotyped song motif could not be clearly identified; in these birds we extracted 'song motifs' as sequences (of comparable length) of syllables which represented most of the bird's vocal range (examples of such subsong "motifs" are shown in Figure 1b). Song motifs from juveniles and adults were compared to each other using Sound Analysis Pro (SAP) 2010bTchernichovski 2000  and the results were further analyzed with custom Matlab scripts (Mathworks Inc). For each pair of birds, we assessed song similarity by averaging the 400 similarity scores (each ranging 0-100) resulting from pairwise comparison between each bird’s 20 song motifs. The consistency of our choice of song motifs, as well as the stereotypy of such motifs (song crystallization), was tested by calculating for each bird the 380 similarity scores among its own 20 motifs, excluding comparisons of identical motifs. One of the 12 additional adult birds used to evaluate similarity to unfamiliar songs (that is, not a tutor) was discarded from the analysis because of low self-similarity (mean similarity score 54.8±8.0). For the remaining adults, average self-similarity scores ranged from 70.9 to 88.5 (mean 80.7±5.3, N = 15), which is well above all tutor-pupil similarities observed in our experiments and also above reports in the literatureFunabiki 2003 Phan 2006 London 2008 Roberts 2012 Dittrich 2013 ; standard deviations of self-similarity scores ranged from 1.3 to 9.5 (mean 4.4±2.5, N = 15). To visualize songs we computed log-power sound spectrograms, which are time-frequency representations of sound intensity.
Note on Sound Analysis Pro (SAP): SAP compares two songs using normalized sound features such as Wiener entropy, spectral continuity, pitch, and frequency modulation. Based on these features, SAP identifies segments in the pupil song that best match segments in the tutor song. The similarity score computed in SAP combines three different measures: percent similarity, sequential match and accuracy. Percent similarity corresponds to the fraction of segments in the tutor song for which a matching segment in the pupil song was found, or in other words it reflects the fraction of the song that has been copied by the pupil. The accuracy measures the local similarity by comparing the song features frame by frame, and sequential match measures the temporal order of matching song segments. The similarity score is the sum of the partial similarities of the selected sections. The procedure is explained in detail inTchernichovski 2000 .