Measuring context dependency in birdsong using artificial neural networks

Area of Science:

• Animal Communication
• Bioacoustics
• Computational Linguistics

Background:

• Context dependency is crucial for understanding sequential structures in human language and animal signals.
• Birdsong serves as a model system for studying context dependency in non-human animals, but previous research faced methodological limitations.
• Assessing the duration of past context's influence is key to understanding complex sequential behaviors.

Purpose of the Study:

• To estimate context dependency in birdsong using a scalable, modern neural-network-based language model with a long accessible context length.
• To compare the detected context dependency with traditional Markovian models and previous experimental findings.
• To investigate the relationship between birdsong vocabulary size (syllable classification granularity) and context dependency.

Main Methods:

• Utilized a modern neural-network-based language model capable of handling long sequences to analyze birdsong.
• Estimated the context dependency within birdsong sequences.
• Examined the correlation between assumed birdsong vocabulary size and detected context dependency.

Main Results:

• The study detected context dependency in birdsong that extends beyond the scope of traditional Markovian models.
• Findings align with previous experimental investigations into birdsong context dependency.
• A larger assumed vocabulary size (finer-grained syllable classification) was associated with shorter detected context dependency.

Conclusions:

• Modern language models offer a scalable approach to studying context dependency in animal vocalizations like birdsong.
• Birdsong exhibits significant context dependency, challenging simpler models.
• The granularity of syllable classification influences the observed context dependency, suggesting a trade-off between vocabulary size and memory.