Project summary

Annotated acoustic datasets of animal vocal communication are largely functionally segmented by human perceptions of non-human animal interactions in social and environmental settings. Bioacoustics attempts to identify acoustic features of these signals for small, nuanced features like pitch changes, intensity features, and rhythmicity. Machine learning techniques typically further characterize the nuance of the categories formed by these original observations. While some animal communication research has focused on linguistic categories like semantics and syntax, I have primarily been motivated by Yip (2026), which was an initial proposal for the search for phonology in other species.

In my research, I would like to pursue the concept of the prosodic word (ω) in an attempt to interpret animal communication signals. Intonation, prosody, and the prosodic word are features of manual languages, written language, and spoken languages. My hope here is that these are also features of animal communication, if not a marker of animal dialects in naturally occurring or encultured populations. The prosodic word in phonological theory can be mapped to a few criteria, with no single feature being a sole indicator of a categorization, but they can be independently tested for comparison (Selkirk 1984; Nespor & Vogal 1986).

The framework adapts some concepts from the phonological literature for matching to acoustic features which can be interpreted by computational systems, but requires some oversight by experienced linguists to ensure basic competency, integrity, and alignment with the current state of knowledge within phonological theory. The tests themselves allow for qualitative descriptions of these features, with components that are falsifiable.

The criteria I propose for this specific project include culminativity, minimality, domain-restricted phonotactics, edge marking, and recursive features.

What are this project's goals? How will you achieve them?

The proposed framework for attempting to define the boundaries of a prosodic word includes the following operational definitions:

For cumulatity: Exactly one primary stress/pitch accent per ω (Selkirk 1984) measured by exactly one prominence peak (F0 or amplitude) per candidate unit

For minimality: Binary lower bound: bimoraic syllable or disyllabic foot (McCarthy & Prince 1986) measured by Minimum duration/element-count threshold; sub-threshold tokens rare or show compensatory repair

For domain-restricted phonotactics: Segmental rules apply asymmetrically within vs. across ω (Peperkamp 1999) measured by Collostructional/transition  analysis: within-unit vs. across-boundary sequencing ONSET (ω), DEP(ω-init) asymmetries at the word edge (Ito & Mester)

For edge marking: Pause duration, F0/amplitude reset at hypothesized boundaries ω dominates ω only via adjunction, not free recursion (Peperkamp 1999), measured by nested integrity of sub-units when combined into multi-unit bouts vs. produced in isolation.

Ideally, I would like to be able to define a segmental unit in a pilot study of an annotated dataset with this protocol. Features like prosody, tone, and intonation often give a nuanced meaning to even our own communication. I am hoping that by demonstrating similar patterns in a sample dataset, this might help us to better understand animal communication without having to directly translate the semantics of animal communication or animal languages.

How will you know it worked?

Whether results were significant would be assessed through a permutation test against a null segmentation baseline. Support for ω-like structure would rest on how many criteria are independently satisfied, not on any single result. The following provides a general list of what would be planned for each criterion.

Culminativity: binomial GLMM on whether a unit has exactly one prominence peak.

Minimality: binomial GLMM on sub-threshold occurrence

Domain-restricted phonotactics: Poisson log-linear model of sub-element bigram counts, with the position × bigram interaction evaluated by a likelihood-ratio test.

Edge marking: Gaussian linear mixed model (LMM) on pause duration and F0/amplitude reset at hypothesized boundaries.

Layering: Gaussian LMM comparing isolated versus bout-embedded sub-units.

How will this funding be used?

The funding will primarily be used to support tools for analysis, written materials required for references, and minimal stipends for the researcher and collaborators to allocate time to complete this work.

Who is on your team? What's your track record on similar projects?

Most of this baseline for this research was assembled through independent study in a university environment over the last ten years with facilitation from professors in philosophy, linguistics, and neuroscience. I don't currently have a team, but if I had funding, maybe that is something that people would have time to do. I have previously managed projects, and currently have faculty mentors in animal welfare that could be consulted for assistance with project planning.

What are the most likely causes and outcomes if this project fails?

If the current dataset that I evaluated and picked due to completeness, notoriety, and quality is simply not good enough for a study of this kind, finding someone to consult to repair the protocol to accommodate this is likely the biggest hang-up. Additionally, finding linguists with specializations in phonological theory that are willing to participate and have an ear for interspecies communication might also be somewhat of a barrier. If the project were to fail in this regard, I would keep reading and revising until I figured out a solution, outside of the context of any funding mechanism.

How much money have you raised in the last 12 months, and from where?

This is my first attempt at any kind of funding application for this project.