The vervet baseline
Seyfarth and Cheney's work in Amboseli National Park, Kenya, characterized three distinct vervet alarm calls: a 'leopard' alarm causing monkeys to run up into trees, an 'eagle' alarm causing them to look up and dive into bushes, and a 'snake' alarm causing them to stand bipedally and scan the ground. Playback experiments confirmed the calls themselves carried the predator-type information — vervets responded appropriately to recorded alarms played without any predator present. The acoustic differences between the three call types were measurable and consistent. The finding is the canonical example of referential animal communication: not just 'something is bad,' but 'something specific is bad.'
Multiple research groups have looked for predator-specific alarm calls in American crows over the past few decades.
What's been tested in crows
Multiple research groups have looked for predator-specific alarm calls in American crows over the past few decades. The evidence is mixed. There's clear acoustic variation in mobbing call structure that correlates with predator type — calls given to aerial threats (accipiters, owls) sound different than calls given to terrestrial threats (cats, humans), with measurable differences in rate, pitch range, and harmonic structure. This is sender-side variation. Whether the variation is treated by receivers as predator-type information — whether a receiving crow responds differently to a hawk-context recording versus a cat-context recording when both are played without context — is less clearly established. Some playback studies have found different response patterns; others haven't replicated cleanly.
Why the answer is hard to nail down
Three things complicate clean tests. First, crow mobbing responses are influenced by the responding bird's own visual context — what predators are actually nearby, what predators it's recently encountered, what the social group is doing. Controlling for visual context in playback experiments is methodologically demanding. Second, individual variation is substantial — different birds respond differently to the same playback. Sample sizes need to be larger than most studies have run. Third, mobbing behavior is multi-modal in practice — vocalizations are paired with visual cues, body posture, head orientation. Disentangling the vocal signal from the multi-modal context is hard. The combined effect: studies that look for crow predator-specific alarm have produced suggestive but inconsistent results.
What's been clearly established
Sender-side acoustic variation: crows produce measurably different alarm vocalizations in different predator contexts. This is not in dispute. The data supports it across multiple studies. What's less clearly established is whether receiver-side decoding treats those acoustic differences as predator-type information that triggers predator-type-appropriate responses. The receiver-side question remains open. CrowLingo's atlas captures the sender-side variation as part of the mobbing cluster's internal structure; we don't claim the cluster encodes referential predator-type information, because that claim isn't yet supported.
Why this matters more broadly
Referential alarm calls — calls that carry information about specific predator types rather than general 'something is dangerous' messaging — are one of the closest analogues in animal communication to lexical meaning in human language. If crows have them, the cognitive distance from American crow communication to language-relevant capacities shrinks meaningfully. If they don't — or if the sender-side variation exists without receiver-side decoding — then the cognitive distance stays larger and the field has to keep being modest about translation claims. The stakes of the question are high, and the empirical situation is not yet clean enough to settle them.
What would settle it
Carefully designed playback experiments with multiple predator-context exemplars, naive receivers, controlled visual contexts, and behavioral measurement of response specificity. The methodology exists; what's lacking is scale. Most studies have run with sample sizes in the dozens of trials; the cleaner answer probably requires hundreds. The ethical floor (no playback within ten meters of active nests, IACUC review for any vertebrate-wildlife playback) constrains how aggressively these studies can scale. Wearable-logger work (Demartsev[3] 2026 for carrion crows) may eventually provide enough synchronized predator-response data to settle the receiver-side question without requiring intensive playback campaigns. Until then, the question stays open.