Classification of hybrid crows in quail using artificial neural networks

In galliforms, calls are strongly determined genetically and no influence of learning has ever been demonstrated. Hybridization is a useful tool for investigating patterns of heritability. The vocal repertoire of the European quail (Coturnix c. coturnix) and of the Japanese quail (C. c. japonica) are similar except for their crows. The European quail possesses two forms of crows and the Japanese quail only one form. We produced hybrids from the following crosses; F(1), F(2) and backcrosses. Visual analysis of spectrograms showed that hybrid crows presented all intermediaries between the three forms of crows produced by the two subspecies. According to the level of analysis of a crow, visual classifications of spectrograms probably include some subjectivity. Artificial Neural Networks (ANN) are now widely used as a powerful classification technique in behavioural sciences. We trained an ANN to recognize the three crows of the two subspecies. Then we analysed its classification of hybrid crows. The ANN revealed important inter-individual variability between the crows of the F(1) and the F(2) crosses. Birds issued from backcrosses produced crows similar to those of the European quail to which they were backcrossed. This study confirms that ANN is a useful tool to classify spectrograms rapidly.