Research programme
I study how animals initiate vocal exchanges, organize calls into sequences, acquire group-specific structure, and encode social information. Vampire-bat communication is the principal system for my doctoral research; earlier and ongoing work with marmosets, budgerigars, other parrots, and free-ranging dogs provides comparative evidence and methodological foundations.
The six themes below describe linked levels of analysis. Together they connect an acoustic event to the identity of its producer, its position within a sequence, the social and behavioural context in which it occurs, and experiments that can test its function.
01 Exchange
02 Combination
03 Representation
04 Laws
05 Behaviour
06 Identity
01
I test whether acoustic similarity between consecutive callers predicts response probability, whether the association is concentrated at the onset of calling bouts, and whether playback can distinguish directed matching from general response biases.
02
I examine non-random call transitions, repetition, and higher-order sequence structure; test whether combinations predict behavioural context; and investigate whether sequence conventions converge within social groups.
03
I compare acoustic features, temporal distance measures, learned representations, dimensionality reduction, and clustering procedures. The central question is whether biological conclusions are robust to how acoustic similarity is represented and validated.
04
I evaluate proposed compression and efficiency laws as statistical hypotheses. Systematic deviations may identify vocal elements for which redundancy, identity information, or interactional function outweighs signal economy.
05
Acoustic localization and synchronized video permit calls to be assigned to individuals and aligned with recipient behaviour, spatial configuration, and ongoing social interactions.
06
Vocalizations may reveal identity as a by-product of anatomical variation, or individuals may learn and maintain vocal structures that function as identity signals. I am developing criteria that distinguish these alternatives.
Method development · simulation and empirical validation
This project is analytically distinct from constructing an acoustic space. It tests whether distributional models such as word2vec can recover relationships among discrete call types from their sequence contexts. Simulations establish the conditions under which the embedding is identifiable; the empirical test will compare model-derived relationships with independently annotated social and behavioural contexts.
The project pages describe the datasets, experimental designs, publications, and ongoing analyses that support these themes. Individual projects often contribute to several themes: the marmoset study links sequence learning, audience effects, acoustic representation, and identity, while the current vampire-bat work connects vocal matching, call combinations, behavioural context, and representation choice.
I am always happy to hear from people working with longitudinal bat recordings, behavioural annotation, playback experiments, acoustic representation, social learning, or computational models of vocal sequences. Send me a short email if you want to exchange ideas or think through something together.
Status labels distinguish peer-reviewed articles, public preprints, completed theses, manuscripts in preparation, and active analyses. Preliminary results are identified explicitly and should be interpreted in light of the stated sample, model, and validation limitations.