Stimulus Sets and Song Recording:
The 6 best-performing males (those that learned the task in the fewest number of trials) were designated Solvers, while the 6 worst-performing males (those that did not solve the task) were designated Non-Solvers. 6 Stimulus sets were created by randomly pairing 1 Solver song with 1 Non-Solver song. All Solvers used as stimulus males solved the task in under 30 trials (mean=22.67 trials); all Non-Solvers took at least 60 trials before they were pulled from testing (mean=90.67 trials). The 6 fastest Solvers were chosen to maximize disparity between the abilities of Solvers and Non-Solvers. Five out of the six Non-Solvers dropped out during stage 5, in which they could no longer see the food reward. The remaining Non-Solver dropped out during stage 1, in which food was placed in wells. Because of concern that this Non-Solver was categorically different in ability than the 5 Non-Solvers who failed at the same stage, statistics were run on female preference including and not including this stimulus set. Stimulus Solver and Non-Solver males were recorded in sound attenuation chambers (Industrial Acoustics) using Shure SM57 directional microphones and Sound Analysis Pro (see Tchernichovski et al. 2000). Males were placed in divided cages with a female (not a study subject and not housed with study subjects) in the other half of the cage in order to elicit directed song (ten Cate 1985). Males and females were thus in visual and auditory contact but not physical contact. To acclimate, pairs were placed in sound attenuation chambers for 24 hours prior to recording. Song output was recorded during the subsequent 24 hours. The majority of males produced hundreds of directed songs in this time period. Males that produced fewer songs were re-recorded in separate sessions with different females until enough songs for analysis were produced. Female zebra finches have been shown to prefer longer songs (Neubauer 1999), so we selected the longest song from each male’s repertoire.
Zebra finch recordings were analyzed after the experiment using Raven Pro Interactive Sound Analysis Software (The Cornell Lab of Ornithology, Ithaca, NY, version 1.5; Bioacoustics Research Program 2014). Typically, song complexity is analyzed by selecting random motifs from across a male’s repertoire and averaging their complexity values (e.g. Boogert et al. 2008). However we were not interested in whether problem-solving ability correlated with general song complexity in males, but in whether song complexity explained the female preference we observed. We thus analyzed only the songs used as stimuli. We measured several potential measures of complexity in this species: song length, number of phrases, average phrase length, average elements per phrase, and total unique elements per song. Data were collected by visual inspection of spectrograms (256 pt. transform, frequency resolution = 86.1 Hz) and elements were categorized following Airey and DeVoogd (2000). To categorize elements as same or different, we used characteristics of the number and distribution of harmonics, frequency modulation, and element duration. Introductory elements were counted as phrases but excluded from phrase length and elements per phrase analyses.