What individual American crow calls 'mean' to the receiver
We know American crow calls cluster acoustically into around nine emergent categories. We know individual identity is acoustically recoverable from a single caw. We have statistical associations between cluster identity and behavioral context. We do not know whether a receiving crow, hearing a specific call type, makes specific inferences about caller identity, behavioral context, or intent in a way that demonstrably changes its behavior. The Marzluff[1] face-recognition work establishes that crows can track individual identity socially. It does not establish that crows use vocal individual-signature information to do so in real time. The gap is the receiver-side decoding problem, and it remains the largest open question in the field.
Statistical models hint at non-random sequence structure in caw-rattle combinations and other multi-call exchanges.
Whether crow vocal sequences have compositional structure
Statistical models hint at non-random sequence structure in caw-rattle combinations and other multi-call exchanges. The 'hint' is a real finding. Whether the sequence structure carries meaning beyond the sum of the parts — whether caw-then-rattle means something different from rattle-then-caw to the bird hearing it — is undertested. Behavioral evidence for crow compositional structure (playback experiments with sequence-permuted versions of natural exchanges) hasn't been collected at the scale required. This is the second-largest open question, and probably the most popular-coverage-relevant one.
How dialect actually functions
Family groups carry distinct acoustic centroids. Geographically separated populations differ. Cultural transmission is the leading hypothesis for the differences. Whether the differences function — whether a crow from group A actually responds differently to a call from group B than to a call from group A — is not established. Functional dialect requires playback experiments with cross-group exemplars, which is ethically expensive and methodologically demanding. Until those experiments run, the dialect claim sits at the descriptive level. Calling crow vocal differences 'dialects' in the functional human-linguistic sense is at present overreach.
What quiet grunts encode
American crow quiet grunts — low-amplitude vocalizations exchanged at close range between paired adults and between parents and offspring — are vastly under-recorded. Most public corpora capture what humans can hear from twenty meters; quiet grunts disappear at that distance. The wearable-logger studies (Demartsev[3] 2026 for carrion crows) are starting to recover them at scale for one corvid species. For American crows specifically, the quiet-grunt repertoire remains mostly unmapped. CrowLingo's atlas has zero quiet-grunt exemplars in the v1 corpus; the cluster sits as a placeholder for what we know exists but haven't sourced.
How crows respond to non-corvid acoustic information
Crows mob predators in response to alarm signals from their own species. Whether they extract information from other species' alarm calls — heterospecific eavesdropping — is well-documented in adjacent songbirds but less rigorously characterized in American crows. Do they know which neighborhood predator is around because they heard a robin's alarm sequence? Probably yes; the evidence is suggestive. Rigorous yes-or-no requires controlled playback with naturalistic stimuli and synchronized behavioral observation. The work is doable, hasn't been done at scale.
Whether AI methods miss what hand-engineered methods see
AI pipelines recover what hand-engineered features recovered, often at finer resolution. But the AI methods optimize for what their training distribution rewards — which is mostly species discrimination and broad acoustic similarity. Whether the embedding spaces are missing dimensions of variation that classical bioacoustic methods captured (specific acoustic features that don't correlate with the major principal components of natural variation) is an open question. Most of the field assumes the answer is 'unlikely to be significant.' The assumption hasn't been rigorously stress-tested.
What honest gap-naming does for the field
Naming the gaps explicitly prevents the field from drifting toward over-claiming. If a popular article says AI is translating crow language, the gap inventory provides the receipts for what would actually be required and what hasn't been established. If a funding proposal claims a methodological breakthrough is imminent, the gap inventory clarifies which gaps the proposal might close and which it won't. CrowLingo's journal exists partly as a public version of this discipline: we name what's known, we name what isn't, and we resist conflating the two. The unknown quadrants are where the next decade of meaningful work will happen.