Individually distinctive acoustic features in sperm whale codas

Ricardo Antunes, Tyler M. Schulz, Shane Gero, Hal Whitehead, Jonathan Gordon, Luke RendellAnimal 2011 · DOI

TL;DR

Sperm whales, known for their complex social structures, use "codas" – series of clicks – for communication. This study investigates whether specific variations within these codas can act as individual "signatures," allowing whales to recognize each other. The research found that while most coda types don't reliably distinguish individuals, one specific type, "5Reg," shows promising individual-specific variations, suggesting different coda types may serve distinct communicative functions.

What Problem Does This Paper Address?

For highly social animals living in stable groups, the ability to recognize and discriminate between individuals is crucial for maintaining social bonds, coordinating activities, and potentially managing social hierarchies. Sperm whales, particularly females and juveniles, form long-term social units, and their interactions suggest a need for such an individual recognition system. Sperm whales produce "codas," which are stereotyped sequences of clicks, and these have been hypothesized to function as individual identifiers. However, previous research has shown that many individuals share common coda types, challenging the idea of simple individual signatures. This paper addresses the question of whether more subtle variations within specific coda types, rather than the repertoire of types themselves, could carry information about individual identity.

Methods

The researchers analyzed recordings of codas from a single, well-studied sperm whale social unit, the "Group of Seven," over 41 days. They employed a multi-step analytical approach:

  • Coda Classification: Codas were first classified into types using k-means clustering based on the timing patterns of clicks within the coda (interclick intervals or ICIs).
  • Feature Extraction: For each coda, they extracted features including:
  • Absolute ICIs: The raw time intervals between consecutive clicks.
  • Relative ICIs: ICIs standardized by the total duration of the coda.
  • Amplitude: The relative envelope amplitude of each click within the coda.
  • Linear Discriminant Analysis (LDA): This statistical technique was used to determine if codas from different individuals could be statistically separated based on the extracted features for specific coda types. The classification error rate (how often a coda was misassigned to an individual) was a key metric.
  • Jackknife Procedure: To assess the robustness of the LDA results, a jackknife procedure was used, where one coda at a time was removed and the remaining data used for classification.
  • Random Assignment Comparison: The calculated error rates were compared against error rates generated by randomly assigning codas to individuals to determine if the observed discrimination was better than chance.
  • Partial Mantel Test: This statistical test was used to evaluate the influence of individual identity and recording day on coda similarity, helping to disentangle individual-specific variation from day-to-day variations in vocalizations.

Datasets

The study utilized a dataset of sperm whale coda recordings collected from a single social unit ("Group of Seven") off the coast of Dominica. This unit consisted of five adult females, one juvenile male, and one male calf. The data were collected over 30 days at sea between January and March 2005. The paper does not introduce a new benchmark dataset but rather uses existing recordings from a specific, well-documented social unit.

Key Insights and Ideas

  • Coda Type Specificity: The most striking insight is that individual-specific information is not uniformly distributed across all coda types. Some types appear to be more "individualized" than others.
  • Beyond Repertoire: This study moves beyond the idea that individual recognition relies solely on the range of coda types an individual produces. It highlights that variations within a single coda type can carry identity information.
  • Hierarchical Coding: The findings suggest a hierarchical coding scheme in sperm whale communication, where a general coda type (e.g., 5Reg) serves as a broad signal, and subtle variations in timing around its stereotyped rhythm provide individual-level information. This is analogous to how birdsong can convey species identity and individual identity.
  • Functional Differentiation: The differential individual specificity across coda types strongly suggests that these types have distinct functions. This implies a more complex communication system than previously understood.

Key Results

  • Coda Type 1+3: Discriminant analysis for this type showed classification error rates that were not significantly better than random assignment, indicating poor individual discrimination.
  • Coda Type 1+1+3: Similar to 1+3, this coda type also failed to provide reliable individual discrimination, with classification error rates falling within the range of random assignment.
  • Coda Type 5Reg: This type showed the most promising results.
  • Absolute ICIs: Using Absolute ICIs as features, the data showed an "obvious separation of individuals using linear discriminants."
    • Partial Mantel Test: This test revealed a significant positive correlation between coda similarity and being made by the same individual for 5Reg codas (correlation coefficient of 0.55, P < 0.001), indicating a stronger effect of individual identity compared to other coda types.
  • Classification Error Rates: For 5Reg, classification error rates for Absolute ICIs were outside the range of random assignment, suggesting a degree of successful individual identification.

Limitations and Future Work

  • Single Social Unit: The study is based on data from only one social unit. The authors acknowledge that this limits the generalizability of findings regarding the 5Reg coda type and its individual specificity. Further research on multiple social units is needed.
  • Limited Individuals: The analysis for some coda types was restricted to a subset of individuals within the group due to data availability.
  • Behavioral Context: Recordings were made in a restricted range of behavioral contexts. The authors suggest that future work should investigate the contextual use of codas, as different functionalities might be reflected in how and when they are used.
  • Generalizability of 5Reg: While 5Reg showed individual specificity, its ubiquity across different populations suggests it might play a role in group or clan recognition as well, in addition to individual identity. Further research is needed to disentangle these potential roles.
  • Discarding Information: The authors propose that standard analytical practices of standardizing intervals might be discarding valuable individual-specific information, particularly for types like 5Reg.

Summary updated

BibTeX
@inproceedings{antunes2011individually,
  author = {Ricardo Antunes and Tyler M. Schulz and Shane Gero and Hal Whitehead and Jonathan Gordon and Luke Rendell},
  title = {{Individually distinctive acoustic features in sperm whale codas}},
  booktitle = {Animal},
  year = {2011},
  doi = {10.1016/j.anbehav.2010.12.019},
}
Figure