Non-invasive · public data · claim-bounded dictionary
Respectful research dictionary

One dictionary for call units, examples, and evidence.

This is not a translation dictionary. It is the project map: real FEROP catalogue samples, larger DCLDE model-label inventory, synthetic listening sketches, and the field-test language needed to ask whether any unit has receiver-interpretable content.

Current word count: 0 validated orca words. The dictionary lists acoustic units and hypotheses. Human shorthand like "food is here" is an analogy for field-test design, not a claim that the call has been translated.
Browse call examples Open Talk Orca Back to main site
60real FEROP catalogue clips shipped on this page
22FEROP K-call types with local audio examples
8,552labelled call-type detections in the model manifest
199distinct call-type labels in the larger dataset

DCLDE/model label inventory

The model-side dictionary is much larger than the public FEROP sample player. It covers 8,552 labelled detections, 199 distinct call-type strings, and 43 call types with at least 30 examples. These are classification labels and counts, not translated words.

43 usable labels199 distinct labels7,748 usable detectionspublic data

Top usable call types by labelled detections

Synthetic sketches

These are generated listening sketches for interface and protocol design. They are not validated orca speech and should not be used as biological evidence. Real FEROP clips remain the evidence anchors.

Call-type recovery

14 SRKW call-type classes recover around 0.709 and 18 NRKW classes around 0.968 in site-controlled heads. This supports stable acoustic labels, not words.

classificationsite controlled

Production context

DTAG calls predict movement-defined context: 0.770 for foraging vs non-foraging and 0.577 for three-way context, with individuals held out.

production sidenot receiver meaning

Playback response

Published FEROP playback re-analysis shows same-pod replies and different-pod silence: 6/6 vs 0/6. The response tracks dialect membership, not content.

receiver responsedialect

Sequence structure

SRKW sequences show information beyond first-order transitions. This is a prerequisite for richer coding, not evidence of semantic phrases.

structurefrontier

What the dictionary can say

It can say this is a named call type, this sample is a real catalogue recording, this synthetic file is a contour sketch, and this evidence supports a field-test design.

safe claim

What it cannot say yet

It cannot say K1 means hello, K7 means come here, or any other one-to-one word mapping. That is what a content-controlled playback study must test.

0 wordsno translation

Glossary

WordA receiver-interpretable unit with content. Current count in this project: zero validated words.
Call typeA stereotyped acoustic category from a catalogue or model head. FEROP K-types and DCLDE labels are call types.
VariantA catalogue example or subform within a call type, such as K5i, K5ii, or K7X.
DialectGroup- or pod-linked call structure. The playback response evidence here is dialect-sensitive response.
Production contextWhat the vocalizing whale was doing when the call was recorded. This is not the same as what a listener understands.
Receiver responseWhat a whale does after hearing a broadcast. This is the evidence needed to move from call structure toward meaning.
Synthetic contourA generated audio sketch for listening and UI demonstration. It is not a validated orca vocalization.
Field kitThe protocol, run sheet, scoring sheet, and stimulus manifest used to test the next receiver-response question.