Linguistic law-like compression strategies emerge to maximize coding efficiency in marmoset vocal communication

Human language follows statistical regularities or linguistic laws. For instance, Zipf’s law of brevity states that the more frequently a word is used, the shorter it tends to be. All human languages adhere to this word structure. However, it is unclear whether Zipf’s law emerged de novo in humans or whether it also exists in the non-linguistic vocal systems of our primate ancestors. Using a vocal conditioning paradigm, we examined the capacity of marmoset monkeys to efficiently encode vocalizations. We observed that marmosets adopted vocal compression strategies at three levels: (i) increasing call rate, (ii) decreasing call duration, and (iii) increasing the proportion of short calls. Our results demonstrate that marmosets, when able to freely choose what to vocalize, exhibit vocal statistical regularities consistent with Zipf's law of brevity that go beyond their context-specific natural vocal behavior. This suggests that linguistic laws emerged in non-linguistic vocal systems in the..., , , each row indicates a single vocalization
\"monkey\": ID of the experimental animal;
\"stage\": the different training stages;
\"call label\": type of call (see legend below);
\"duration\": call duration;
\"interval\": interval between call onset and preceding call onset;
\"call\_length\": call classification;

Legend for call types:
1\_1 first phee syllable;
1\_2 second phee syllable;
2 trill;
3 tsick;
4 ekk;
5 chirp;
6 twitter;
7 trill\_phee;
8 peep

nan \"not any number\", resp. first call of call sequence