Research

The two main current research projects of our group are the study of vocal learning in songbirds in a naturalistic social setting of a small colony and the adaptation of computational methods for the analysis of both birdsong and natural language.

Our group studies vocal learning in diverse species, but mainly in the songbird. Similar to speech learning in humans, songbirds learn their songs by adjusting their own vocalizations in reference to a memorized tutor song. They learn their songs in a first sensory phase during which they memorize a template of tutor song and in a second sensorimotor phase, in which they refine their vocalizations to gradually approximate the tutor song. They achieve this remarkable feat using a highly specialized set of brain areas, termed the song system. Our research is directed at elucidating the various neural mechanisms of song production and song learning by performing experiments in freely behaving and singing birds. Using our recently developed high-tech observatory, we are currently working towards resolving every single vocalization performed in a small colony of birds. Our goal is to understand the diverse social interactions that promote and constrain cultural learning of a song in juvenile birds.

Song Learning Projects

Algorithms:
Machine learning approaches for sound separation and vocal segmentation
Multimodal action recognition from video and audio 
Statistical inference of social learning strategies
Reinforcement learning of vocal repertoires

Neuroscience:
Sensory substitution
Neural population recording and analysis
Zebra finch welfare
Wireless brain imaging
Connectomics
Development of viral tools

There are remarkable parallels between birdsong learning and NLP. For example, birds use the same strategy for learning their song vocabulary that computational linguists use to retrieve similar documents to a given text. Such similarity might provide clues about the evolutionary origins of language. Our aim is to explore the parallels between birdsong and natural language, by doing so we hope to discover computational primitives ideally suited for NLP. As part of our outreach effort, we are working on a web application to help researchers in assimilating the scientific literature and in assessing their own findings and hypotheses within the context of published facts.

Projects in Natural-Language-Processing (NLP)

Multi-document summarization
Local citation recommendation
Scientific argument mining

MSc and BSc students: if you want to contribute to our ongoing projects or propose your own projects, look at our student projects or just send us an email.