Back

Does Zipf's law of abbreviation shape birdsong?

Gilman, R. T.; Durrant, C.; Malpas, L.; Lewis, R.

2023-12-17 animal behavior and cognition
10.1101/2023.12.06.569773 bioRxiv
Show abstract

Zipfs law of abbreviation predicts that in human languages, words that are used more frequently will be shorter than words that are used less frequently. This has been attributed to the principle of least effort - communication is more efficient when words that are used more frequently are easier to produce. Zipfs law of abbreviation appears to hold for all human languages, and recently attention has turned to whether it also holds for animal communication. In birdsong, which has been used as a model for human language learning and development, researchers have focused on whether more frequently used notes or phrases are shorter than those that are less frequently used. Because birdsong can be highly stereotyped, have high interindividual variation, and have phrase repertoires that are small relative to human language lexicons, studying Zipfs law of abbreviation in birdsong presents challenges that do not arise when studying human languages. In this paper, we describe a new method for assessing evidence for Zipfs law of abbreviation in birdsong, and we introduce the R package ZLAvian to implement this analysis. We used ZLAvian to study Zipfs law of abbreviation in the songs of 11 bird populations archived in the open-access repository Bird-DB. We did not find strong evidence for Zipfs law of abbreviation in any population when studied alone, but we found weak trends consistent with Zipfs law of abbreviation in 10 of the 11 populations. Across all populations, the negative correlation between phrase length and frequency of use was several times weaker than the negative correlation between word length and frequency of use in human languages. This suggests that the mechanisms that underlie this correlation may be different in birdsong and human language.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.