Skip to content
EarthConvergenceTry the interspecies signal model

Cross-species communication

The animalsand the channels they are actually using

Cross-species communication is not one problem, it is 5very different ones. A honeybee’s message is a geometry performed in the dark. A firefly’s is a latency. A rat’s is an affective state broadcast above the top of human hearing. Each channel fails for a different reason and each one needs a different instrument.

What follows is the ethology, not the marketing. Every species gets two columns of equal weight: what is established, and what is genuinely open.

One system on this list has been decoded to a checkable meaning, and not by us: the honeybee waggle dance, by Karl von Frisch (1946–67), because the thing it refers to — a place — can be walked to and checked. MasiDyn's own honeybee row is a synthetic baseline. Everything MasiDyn has measured on real recordings is structure — timing, temporal structure, contour shape, call detection, species and location association — and calling any of it translation would be false.

Counted from the records below

The shape of the problem

These figures are computed from the species records on this page, not typed in beside them. They describe the scope of this programme and nothing about the animals’ populations, ranges or numbers — claims this site does not make.

5

anchor species, each with a binomial and a published literature

src/lib/content/species.ts

5

distinct signal modalities across those species

computed from the modality field

12

problems recorded as genuinely open, not as teasers

computed from the open field

  • mechanical / vibrational1 of 5
  • bioluminescent / visual1 of 5
  • acoustic (audible)1 of 5
  • acoustic (ultrasonic)1 of 5
  • acoustic (click rhythm)1 of 5

Jump to

The five

Hymenoptera · Apidaemechanical / vibrational

Western honeybee

Apis mellifera

The channel

A dance run on the vertical face of the comb, in the dark. Followers track it by antennal contact and by the vibration the dancer puts into the comb — not by sight.

What a human cannot perceive

The message is geometric and it is performed in darkness on a surface. There is nothing to hear and nothing to see; the whole signal is posture, angle and duration.

Established

  • The angle of the straight waggle run relative to vertical equals the bearing of the resource relative to the sun's azimuth, as seen from the hive. Straight up means toward the sun. This is arithmetic, not analogy.
  • The duration of the waggle run encodes distance, approximately linearly over the middle of the forager's range.
  • The distance calibration is subspecies-specific — the classic 'dance dialects' (von Frisch & Kratky; Boch). A slope fitted on one population does not transfer cleanly to another.
  • This is the only animal communication system decoded to the point of quantitative prediction: you can read a dance and go to the food.

Open

  • The distance term still needs a per-colony calibration before it is exact. Fitting that dialect is itself part of the decode.
  • Waggle-dance work is the exception, not the template. No second natural system has been solved to this level, and the reason it worked here is that the referent — a location — is externally checkable. Most animal signals have no such external ground truth.

Mechanism

Two independent scalars in one gesture. Direction is carried by an angle transposed from the horizontal world onto a vertical surface, with gravity substituted for the sun. Distance is carried by time. Because the sun moves, the same food gives a different dance angle through the day — which means the receiver is doing a coordinate transform too.

What decoding it would actually mean

It already is — by von Frisch, not by us. The honeybee is the existence proof the rest of this programme leans on: a natural, non-human signal that carries a quantitative referent and that a human can read. Everything else on this site is measured against how far short of that bar it falls.

Literature
Karl von Frisch, 1946–67 (Nobel Prize in Physiology or Medicine, 1973); von Frisch & Kratky; Boch, on dialects
In our synthetic harness
Our module implements the exact angle rule and a fittable distance dialect and runs on synthetic dances generated by that same module, not recorded dances. On that synthetic run the nearest-neighbour baseline's bearing error was 26.64° against a permutation null of 90.69° (18 held-out dances, p = 0.005); the distance half is reported separately (493.6 m error) and is much weaker. The synthetic-baseline row (generated data, not recordings) is bearing_err_deg = 26.64 against a permutation null of 90.69 — the full statistics are on the language-studies page.
On real recordings
No real-recording result for this species.

Coleoptera · Lampyridaebioluminescent / visual

Firefly

Photinus pyralis

The channel

Light, in time. A flying male emits a species-specific flash pattern; a perched female answers after a species-specific delay. The code is carried by the intervals, not by the brightness.

What a human cannot perceive

Humans see the flashes perfectly well and read nothing in them. The information is in sub-second timing relationships across a two-party exchange — a channel we can see but cannot parse.

Established

  • Photinus courtship is a timed dialogue: male flash pattern, female response delay, both species-specific. That pattern is how the species are told apart in the field.
  • The signal is a species-recognition system first. It is what keeps sympatric species from wasting a season on each other.
  • Predatory Photuris females mimic the response delays of Photinus females and eat the males that answer (Lloyd). The code is exploitable, which is the strongest possible evidence that it is a code.

Open

  • Species-level timing parameters are documented in the taxonomic literature, but they are not one clean equation across the genus — they are a table of species, and getting them right is field work.
  • How much beyond species identity the exchange carries — male quality, female receptivity — is an active question, not a settled one.

Mechanism

A two-sided temporal handshake. Neither party's signal means anything alone: identity is established by the interval between them. That makes it one of the few natural signals whose 'grammar' is a latency.

What decoding it would actually mean

A per-species timing model good enough to run in the field would let a passive optical sensor census fireflies by species from flash timing alone. That is a survey instrument, not a translator.

Literature
Stanger-Hall & Lloyd (2015); Lloyd, on aggressive mimicry in Photuris
In our synthetic harness
The weakest module in our set, and we say so on the results page. The interval→delay line it learns is OUR working calibration, not a published regression, and the synthetic data is generated from that same line — so the learner is recovering our rule, not nature's. The synthetic-baseline row (generated data, not recordings) is delay_err_s = 0.125 against a permutation null of 0.481 — the full statistics are on the language-studies page.
On real recordings
No real-recording result for this species.

Primates · Cercopithecidaeacoustic (audible)

Vervet monkey

Chlorocebus pygerythrus

The channel

Airborne calls in the human hearing range. Acoustically distinct alarm classes, given to different predators.

What a human cannot perceive

Almost none, physically — you can hear a vervet alarm call. The gap is interpretive: telling the classes apart reliably, and knowing what the distinction actually is to the animal.

Established

  • Vervets give acoustically distinct alarm calls for leopards, eagles and snakes. Playback of a recorded call, with no predator present, elicits the escape appropriate to that predator: into the trees, down and looking up, up on the hind legs scanning the ground (Seyfarth, Cheney & Marler, 1980).
  • That playback result is one of the foundational experiments in animal communication, because it separated the signal from the situation.
  • Infants produce the calls broadly and refine them with experience — the mapping is partly learned.

Open

  • Whether 'referential' means the call denotes a leopard, or means it triggers an escape routine that happens to suit leopards, is a debate the field has not closed. Functionally referential is the careful term, and it is careful for a reason.
  • The alarm repertoire is a handful of call types. Reading it is a classification problem with very few classes — it is not evidence of an open-ended vocabulary, and it has never been shown to combine.

Mechanism

A small, discrete set of call types, each reliably paired with a class of danger and each eliciting a different, ecologically correct response. That is functional reference: the call does the work the predator would have done.

What decoding it would actually mean

Reliable automatic classification of alarm classes from field audio would give an acoustic early-warning readout of what a troop is reacting to. It would not give you a vervet dictionary, because outside the alarm system there is not much to look words up in.

Literature
Seyfarth, Cheney & Marler (1980)
In our synthetic harness
Our recorded row is a three-class problem on synthetic acoustics. The pitch, duration and harmonic values are invented by our own generator and must not be read as vervet measurements — only the task structure is real. The synthetic-baseline row (generated data, not recordings) is accuracy = 0.643 against a permutation null of 0.464 — the full statistics are on the language-studies page.
On real recordings
No real-recording result for this species.

Rodentia · Muridaeacoustic (ultrasonic)

Brown rat (laboratory rat)

Rattus norvegicus

The channel

Ultrasonic vocalisations, emitted above the human hearing limit. Two adult call classes with opposite affective valence.

What a human cannot perceive

Total, without instruments. Human hearing tops out around 20 kHz; both adult call classes sit above it. A room of rats calling constantly is a silent room to a person — this is the clearest case on the site of a conversation happening in the open that we are simply deaf to.

Established

  • Adult rats emit two acoustically distinct ultrasonic classes: long, monotonous, narrowband calls near 22 kHz, and short, frequency-modulated calls near 50 kHz. Our decoder module states these bands and cites Brudzynski, Panksepp & Burgdorf, and Wöhr & Schwarting.
  • The two classes map to opposite affect. 22 kHz calls accompany social defeat, predator exposure and withdrawal; playback drives avoidance and threat-circuit activation. 50 kHz calls accompany play, mating and reward anticipation; playback drives approach and nucleus accumbens dopamine release.
  • This is the best-validated affect-to-call mapping in any rodent — validated by playback and by neural response, not by correlation alone.

Open

  • It is not laughter. 50 kHz is a positive-affect and social-appetitive marker; the popular framing outruns the evidence and we do not repeat it.
  • The real methodological problem is not the classifier, it is the null. On real recordings the held-out split must be by SUBJECT — many calls from one animal are not independent samples, and pseudoreplication is the standard way this literature goes wrong.
  • Real repertoires are messier than the two clean bands: there are short aversive calls and flat appetitive ones. Any decoder that works on the textbook split and not on the edge of the repertoire has learned the textbook, not the animal.

Mechanism

An affective state readout rather than a message about the world. The call does not point at a thing; it reports the internal state of the animal producing it, and other rats act on that report.

What decoding it would actually mean

A validated affect readout is a welfare instrument. It would let a laboratory measure the affective state of an animal continuously and non-invasively, which is a real and immediate use — and it is a use that does not require the word 'language' anywhere.

Literature
Brudzynski; Panksepp & Burgdorf; Wöhr & Schwarting
In our synthetic harness
Our row runs on synthetic feature data, not recorded calls. The generator deliberately injects atypical, edge-of-repertoire calls so that no single-feature threshold separates the classes. The recorded row is nonetheless one shuffle away from failing its own significance gate, and the results page says so. The synthetic-baseline row (generated data, not recordings) is accuracy = 0.929 against a permutation null of 0.701 — the full statistics are on the language-studies page.
On real recordings
No real-recording result for this species.

Cetacea (Artiodactyla) · Physeteridaeacoustic (click rhythm)

Sperm whale

Physeter macrocephalus

The channel

Stereotyped bursts of echolocation clicks, called codas. The type of a coda is defined by its rhythm — the pattern of intervals between clicks.

What a human cannot perceive

The clicks are audible on a hydrophone, but the structure is in millisecond-scale interval patterns inside a burst, in an ocean, from an animal that spends most of its life below the depth anyone can watch it.

Established

  • A coda is a stereotyped click burst, and its TYPE is defined by the pattern of inter-click intervals. Types are conventionally written as rhythms — 5R is five regular clicks; 4+1 is four regular then a gap then one.
  • Several coda types share a click count and differ only in where the gap falls, so typing a coda is a rhythm problem and not a click tally.
  • Sperm whales form vocal clans that differ in their coda repertoires — the repertoire is cultural, transmitted, and clan-specific (Rendell & Whitehead).
  • Published machine-learning work on codas (e.g. Sharma et al. 2024) operates at this level: acoustic structure, evaluated structurally, not semantically.

Open

  • There is no verified coda-to-meaning map. None. Structure can be measured; meaning has not been decoded, by us or by anyone, and our own decoder module carries that as a hard guardrail in its source.
  • The hard part is not the model, it is the referent. Nobody has a reliable external ground truth to align codas against — no equivalent of the honeybee's food source, which is the only reason the bee was solvable.
  • Whether coda structure carries anything beyond identity is exactly the open question. Assuming it does, and calling structural clustering 'translation', is the standard failure mode in this field.

Mechanism

Rhythm as identity. The best-supported function of coda repertoire is social — marking clan and group membership — which makes it closer to an accent than to a vocabulary.

What decoding it would actually mean

Reliable coda typing at scale would give clan-level identification from passive acoustics — who is present, from which clan, where. That is population science and it is genuinely valuable. It is not talking to whales, and anyone who says it is has skipped a step nobody has taken.

Literature
Watkins & Schevill; Rendell & Whitehead, on vocal clans; Project CETI / Sharma et al. (2024)
In our synthetic harness
Our leaderboard module classifies coda TYPE on synthetic interval templates (the baseline typed 7 of 14 against a permutation null of 0.157) and says nothing about meaning. The module was filed as orca.py — a misnomer we are not going to hide; it handles sperm whale codas and has been renamed sperm_whale_coda.py. The leaderboard's species field was always the correct one. The synthetic-baseline row (generated data, not recordings) is accuracy = 0.5 against a permutation null of 0.157 — the full statistics are on the language-studies page.
On real recordings
Using 1,501 real coda recordings from the Dominica Sperm Whale Project (Project CETI, CC BY 4.0), MasiDyn's research pipeline detected clicks and measured their timing: median inter-click interval 0.175 s over 1,429 codas. This is consistent with published ranges but is not a measured error rate, since the dataset has no click annotations, and the value tracks the detector's 0.10 s minimum click spacing (0.115–0.196 s at settings of 60–120 ms). Coda types in neighbouring files (by file number, inside 9 same-rig blocks) were more alike than in a shuffle of the whole corpus (conditional entropy 3.62 vs 4.04 bits, z = -20.0, 1,419 file-to-file pairs). The files are isolated snippets with no timestamps, the shuffle mixes coda types across rigs, and part of the effect is the same coda type recurring, so this shows coda types clustering in the dataset's file order, not a whale sequence or grammar. On 1,313 Watkins sperm whale clips (WHOI), 82% of whose codas were recorded off Dominica (an audit count from a committed file), the same pipeline gave a median inter-click interval of 0.1727 s: a consistency check on largely the same population, with the same detector floor, not an independent replication. No coda was mapped to a meaning and behavioural context could not be tested. Every number with its null, sample size, receipt and caveats.

A separate track · real recordings

Species with real-recording structure tests

Real recordings were analysed for 35 marine mammal species, for structure only. Five have dedicated entries, summarised here; the other 30 have contour-structure results from the Watkins archive only. None of this decodes meaning. Every number, with its null, sample size, receipt and caveats, is on the results page.

Real recordings · structure only

Sperm whale

Physeter macrocephalus

Also an anchor species: its real-recording summary is in the sperm whale panel above, under “On real recordings”, kept separate from its synthetic-harness row.

4 committed results · evidence: Receipt reproduced; Committed receipt, not re-run; Reproduced from a committed cache · no meaning decoded. Every number, null and caveat

Data source and licence

  • Dominica Sperm Whale Project / Project CETI, CC BY 4.0 (Paradise et al., NeurIPS 2025, arXiv:2512.02206). Licence: CC BY 4.0 (attribution required; commercial use permitted).
  • Watkins Marine Mammal Sound Database, Woods Hole Oceanographic Institution (WHOI). Licence: Research and educational use only (WHOI terms).

Real recordings · structure only

Killer whale (orca)

Orcinus orca

On a public 30-minute Orcasound recording of Southern Resident killer whales (Haro Strait, 2019; CC BY-NC-SA), MasiDyn's call detector matched human-marked calls with F1 0.69, against 0.35 for randomly placed detections (398 annotated calls). Contour shapes of calls that share a Ford catalogue type were closer together than chance (z = -2.9 on 38 calls, a small sample). On Watkins orca clips (WHOI), using wild clips only and one sample rate at a time, contour type was associated with recording location: mutual information 0.2272 bits, z = 11.2 against a shuffle null (627 clips, Andenes vs Kikvika, Norway), and 0.43 bits, z = 14.0 (245 clips, Dabob Bay vs Vancouver); a third subset (37 clips, 3 sites) showed no association (z = 0.7). These used a saturated 64-label contour alphabet, were computed before an estimator fix and were never re-run, and location cannot be separated from recording session in that archive, so this is not evidence of dialect. Orca and false killer whale contour shapes were told apart at 0.58 against chance 0.5 (z = 2.77, 150 clips per species), but clips from the same recording session were not kept apart, so part of this may be session rather than species. Nothing here decodes meaning.

  • 1,452 of the 2,646 Watkins orca clips are captive (Oregon Coast Aquarium); the other 1,194 are wild (Norway, British Columbia, Washington, Massachusetts, Newfoundland).

10 committed results · evidence: Receipt reproduced; Committed receipt, not re-run · no meaning decoded. Every number, null and caveat

Data source and licence

  • Orcasound (Orcasound Lab hydrophone, Haro Strait, 2019-07-05; call annotations by Scott Veirs). Licence: CC BY-NC-SA (non-commercial, share-alike).
  • Watkins Marine Mammal Sound Database, Woods Hole Oceanographic Institution (WHOI). Licence: Research and educational use only (WHOI terms).

Real recordings · structure only

Humpback whale

Megaptera novaeangliae

One humpback song clip (40 s analysed; its source is documented only in a code comment) beat 200 frame-shuffled versions of itself on two statistics (z = 33.1 and 32.1), and a stationary-noise control failed. The strongest self-similarity is at the shortest lag tested (0.4 s), so this reflects short-range spectral continuity, not repeated phrases or themes; almost any non-stationary natural sound would pass this test. It is a single-clip structure test, not a decode of themes or meaning. Across 488 Watkins humpback call contours (WHOI), contour shapes beat a within-contour shuffle (z = -38.364), a weak test that mostly reflects smooth pitch tracks.

3 committed results · evidence: Receipt reproduced; Committed receipt, not re-run · no meaning decoded. Every number, null and caveat

Data source and licence

  • Watkins Marine Mammal Sound Database, Woods Hole Oceanographic Institution (WHOI). Licence: Research and educational use only (WHOI terms).
  • Source given only in a code comment: supporting audio of PLOS ONE article pone.0051214 (Antarctic feeding ground). The file-level match is unverified. Licence: Not recorded; to be confirmed.

Real recordings · structure only

Common minke whale

Balaenoptera acutorostrata

Species label unverified.

A single 2.9-second clip labelled as a minke whale passive-acoustic recording (source not documented) contains a regular pulse train about 0.34 s apart that beats a frame-shuffle null (z = 22.2 and 19.0). This is a structure test on one clip, not a decode. The 24 Watkins minke clips were recorded at too low a sample rate to measure.

  • Species label unverified: the clip's label was entered by hand and its source is undocumented.

1 committed result · evidence: Values reproduced with audit code · no meaning decoded. Every number, null and caveat

Data source and licence

  • Source undocumented: no dataset, URL or recordist is recorded, and the species label was entered by hand. Licence: Unknown.

Real recordings · structure only

Common bottlenose dolphin

Tursiops truncatus

In 167 Watkins contours from common bottlenose dolphins (WHOI), contour shapes beat a within-contour shuffle (z = -14.71), a weak structure test. A within-tape test of whether individual dolphins can be told apart by their signature whistles was run on 33 Watkins clips, but its result file was never committed, so MasiDyn does not report it as a result.

  • Its only committed structure result is the single August Watkins receipt (non-default branch, never reproduced).

1 committed result · evidence: Committed receipt, not re-run · no meaning decoded. Every number, null and caveat

Data source and licence

  • Watkins Marine Mammal Sound Database, Woods Hole Oceanographic Institution (WHOI). Licence: Research and educational use only (WHOI terms).

The 30 Watkins-only species — pilot whales, right whales, dolphins, walrus and seals among them — are in the Watkins contour table, including the 3species whose contours did not beat their null (Ross, ringed and leopard seal; the leopard seal set’s species identity is uncertain).

How this page is written

The rules we hold ourselves to here

Writing about animal communication is unusually easy to do dishonestly, because the exciting claim and the defensible one use the same words. These are the constraints applied to every line above.

No invented numbers
There is no population count, range size or lifespan on this page. The only quantities stated are physical properties of a channel, frequency bands stated in a decoder module and attributed to the literature it cites, synthetic leaderboard figures labelled as synthetic, real-recording results transcribed from committed receipts with their nulls and sample sizes, and a few caveat figures computed by the audit from committed files, labelled as such. Where a number is a parameter of our own synthetic generator, it is labelled as ours.
Established means published
Every claim in an Established column names the work behind it. Where the field is divided — as it is on whether vervet alarm calls are referential in a semantic sense — the page says the field is divided rather than picking the more quotable side.
Open means open
The Open column lists problems nobody has solved, MasiDyn included: our real-recording work in the project’s research and workspace repositories is structure only and closes none of them. That includes the one that matters most: for four of these five species there is no external ground truth to align a signal against, which is the actual reason the honeybee is the only solved case.
Translation is never claimed
Nothing on this site translates anything. What is done is structure testing (on real recordings and on synthetic data), classification (on synthetic data, and species discrimination on Watkins contours) and affect-label classification on synthetic data only, and each is named as what it is on both this page and the results page.
Only real animals
Every species on this page, in the Journey and in the signal model is a real organism with a binomial name. Where the signal model covers an extinct or ancestral lineage, its template is labelled lore.