Associative Syntax and Maximal Repetitions reveal context-dependent complexity in fruit bat communication

TL;DR

This paper presents an unsupervised approach to quantify repertoire and syntax in graded fruit-bat vocalizations, introducing Maximal Repeats (MR) as a measure of combinatorial complexity. It investigates how dimensionality reduction affects unsupervised unit labeling (RQ1) and how syntax and temporal structure encode context (RQ2). Results indicate associative syntax, context-dependent syllable usage, and heavy-tailed MR distributions, with longer MR lengths and small-world transition networks in conflict contexts, implying higher communicative complexity when disagreement arises. The work provides a refined pipeline and MR-based framework applicable to other species, highlighting the role of temporal structure in compressibility and information richness of animal communication.

Abstract

This study presents an unsupervised method to infer discreteness, syntax and temporal structures of fruit-bats vocalizations, as a case study of graded vocal systems, and evaluates the complexity of communication patterns in relation with behavioral context. The method improved the baseline for unsupervised labeling of vocal units (i.e. syllables) through manifold learning, by investigating how dimensionality reduction on mel-spectrograms affects labeling, and comparing it with unsupervised labels based on acoustic similarity. We then encoded vocalizations as syllabic sequences to analyze the type of syntax, and extracted the Maximal Repetitions (MRs) to evaluate syntactical structures. We found evidence for: i) associative syntax, rather than combinatorial (context classification is unaffected by permutation of sequences, F 1 > 0.9); ii) context-dependent use of syllables (Wilcoxon rank-sum tests, p-value < 0.05); iii) heavy-tail distribution of MRs (truncated power-law, exponent α < 2), indicative of mechanism encoding combinatorial complexity. Analysis of MRs and syllabic transition networks revealed that mother-pupil interactions were characterized by repetitions, while communication in conflict-contexts exhibited higher complexity (longer MRs and more interconnected vocal sequences) than non-agonistic contexts. We propose that communicative complexity is higher in scenarios of disagreement, reflecting lower compressibility of information.

Figures (6)

• Figure 1: Improved clustering quality of continuous-type vocalizations. The left panel shows the original benchmark results, which primarily separate isolation calls from adult vocalizations. The right panel demonstrates our improved clustering, which identifies seven distinct syllable types through optimized dimensionality reduction and segmentation techniques applied to the graded vocal system.
• Figure 2: Diagnostic of the local dimensionality of manifold learning. Clustering obtained from Non-parametric UMAP applied on Mel-Spectrograms (6x32) preprocessed by Mel-filterbank (hop size equal to FFT length) and dynamic segmentation. Inputs: 152,578 data-points from all bats (41 individuals). Bluish colors represent the lowest local dimensionality, which corresponds to underdeveloped vocalizations from the Isolation context (i.e., simpler, more uniform spectrograms). Warmer colors (yellow/red) indicate regions of higher local dimensionality and greater acoustic complexity.
• Figure 3: Importance of features used for the Random Forest classifier. Features representing richness of contextual syntax, unpredictability of sequences, commonness of patterns and strength of short transitions (respectively, features: e, f, g, p) account for about 50% of the total feature importance, suggesting a predominant temporal organization of short transitions and repetitive patterns.
• Figure 4: Syllable-type unique to the Isolation context (mother-pupil interactions), isolated through agglomerative clustering. (\ref{['fig:isolation_syllable_single']}) Randomized sampling of the syllable-type, displaying its uniform spectral structure. (\ref{['fig:isolation_syllable_sequence']}) A sequence of this syllable, demonstrating the characteristic repetition patterns consistent with underdeveloped vocalizations. Note -- Unsupervised labeling in these figures used the acoustic segments from the original dataset prat2017annotated, whose boundaries were computed by thresholding the amplitude envelope above a fixed noise floor. When using algorithms that estimate the noise floor dynamically (as in sainburg2019parallels), syllables could be further subdivided into smaller segments; in this example, the three bursts visible in the waveforms were separated into distinct sub-units.
• Figure 5: Distribution of Maximal Repetition (MR) lengths across behavioral contexts (sequences with at least 50 support). Conflict-related contexts (Mating Protest, Fighting, Threat-like) show heavier-tailed distributions with longer MRs, indicating more complex temporal structures and lower compressibility of information. Cooperative contexts (Feeding, Grooming, Kissing) exhibit shorter MR distributions, suggesting higher redundancy and more compressible communication patterns. The Isolation context shows a unique pattern dominated by short, repetitive sequences.