2. Computational ethology of female choice in a natural social context:

Previous accomplishments: Courtship behavior is complex. Song and copulatory solicitation displays therefore only represent part of a larger suite of behaviors (Perkes et al. 2017). Like many other species of songbirds, cowbirds will establish strong pair bonds during the mating season. A robust behavioral read-out of pair bonding is that females are sung to almost exclusively by their consort male. The decision to sing, or be sung to, is bidirectional and females therefore play a profound role in these behavioral interactions. To test the role of the song system on courtship behavior, we performed targeted HVC lesions to pair-bonded females, re-introduced them into large outdoor aviaries and quantified song interactions between the cohort of birds. Compared to sham lesioned birds, introduction of females with HVC lesions into the aviary caused a dramatic change in male behavior.  Pair-bonded males that normally only sang to their pair-bonded mate, would direct their song to the lesioned females. This change in behavior was also accompanied by an overall reorganization of the group’s social network (Maguire et al. 2013). Experimental observations performed by human observers were unable to identify any significant behavioral patterns that might have led males to ignore their pair-bonded mate and sing to HVC-lesioned females.  An overarching hypothesis that has emerged from this work is that the “song system” provides context specificity to a suite of courtship behaviors and that its removal causes females to produce contextually inappropriate signals that males find attractive. We have evidence that this circuit might even control reproductive behavior because lesions cause an increase in reproductive output (more eggs) (White et al., 2017).

Current studies: Quantifying complex social behavior in a cohort of interacting birds is extremely challenging.  In our previous work, males were clearly responding to subtle behaviors that experienced ethologist were unable to identify. To identify and quantify the complex behavioral interactions in social groups, we are building a large outdoor “smart” aviary (figure 3) equipped with an array of cameras and microphones to quantify all possible behavioral interactions in small (12 – 20 birds) social groups. This computational ethological approach to studying complex social behavior is part of a large funded collaborative effort between multiple biology and engineering research groups at Penn (NSF MRI: Development of an observatory for quantitative analysis of collective behavior in animals). We are developing sophisticated computational approaches (in collaboration with the Daniilidis lab, Penn engineering) to identify behavioral interactions (both gestural and acoustic) between individuals in the group and use this information (in collaboration with the Bassett lab, Penn engineering) to quantify dynamic changes in the social network over time. In addition to the computational approach, we are continuing our collaboration with David White, who has several large outdoor aviaries. In one study, we are quantifying, in collaboration with Alex Baugh (Swarthmore College) [LINK], the impact of HVC lesions on the neuroendocrine system given the effect of these lesions on reproductive output. In other experiments, we are evaluating how targeted lesions of different areas of the song system (e.g. RA, LMAN) impact CSD behavior and social behavior. These experiments are critical for an evaluation of the basal ganglia in courtship behavior, a question of great interest given how much is known about the role of this specialized “anterior forebrain” circuit (Area X à DLM à LMAN) in the context of song learning and maintenance in males.


Future directions: Our current approach to behavioral quantification of social behavior remains in the early developmental phase. Our long-term goal is to use deep learning approaches coupled with computer vision methods to classify all relevant behavioral epochs during the breeding season. Using this information, we plan to develop network visualization tools to quantify all relevant network interactions. At the behavioral level, these approaches will allow us (in collaboration with Erol Akcay, Penn Biology [LINK] and Dani Basset, Penn Engineering [LINK]) to build mating network models that will inform how individuals respond to signals and cues of mate availability and quality. At the neurobiological level, full automation of this system coupled with the ability to record from birds in a wireless fashion will allow us to record brain processing during complex social interactions.