Program

OR27-23-3

Probabilistic syntax model of the birdsong using a simple artificial neural network with feedback of the previous sequences.

[Speaker] Koumura, Takuya:1
[Co-author] Okanoya, Kazuo:1
1:University of Tokyo (Japan)

Learned sequential vocalizations, such as language and birdsong, requires highly complex cognitive skills. Especially, songs in Bengalese finches are excellent models of complex sequential vocalizations because song elements are sequenced according to certain probabilistic rules, referred to as song syntax.
Conventionally, song syntax is modeled by conditional probabilities of element types given the previous element sequences. However, such symbol-based models are not very much informative about the neural mechanisms of song syntax. In this study, to bridge the gap between observed symbol sequences and the recorded neural activities, we designed a probabilistic model of the song syntax using a simple artificial neural network with feedback of the previous sequences. We found neural activities in anesthetized birds that were correlated with the mode inputs. These results suggested that the complex sequencing pattern could be controlled by the feedback of the previous vocalizations. (Supported by Kakenhi #26240019 & #15J09948)

Advanced Search