Skip to content
CrowLingo

PILLAR V — THE FRONTIER

The honest map of where the field is.

What we can do today. What we cannot do yet. What we should not do. Ethics, open data, contribution pathways.

NOW vs AIMED

Demonstrated vs aspirational.

What works today

Automatic detection

Segmenting crow calls from hours of field audio

Unsupervised discovery

Finding call categories from geometry alone

Caller identification

Recognizing known individuals from voice

Zero-shot captioning

NatureLM-audio generates plausible descriptions

What's still ahead

?

Compositional decoding

Understanding calls as combined meaningful units

?

Bidirectional dialogue

Real-time two-way communication between human and crow

?

Crow dictionary

Human-readable glosses for every call type

?

Cross-species transfer

Which structures generalize across corvids

ETHICS

Six rules for listening back.

01

We do not deploy playback features

02

We cite primary research, always

03

We distinguish demonstrated from aspirational

04

We do not overclaim translation

05

We credit recording contributors

06

We open the dataset and the code

Frequently asked

What people ask about this.

What is on the CrowLingo v1-v5 roadmap?
v0 (live): interactive atlas with 16 real recordings, AI narration, cluster pages. v1: Perch 2.0 embeddings + similarity search. v2: upload-your-own-crow with cluster placement. v3: extend to other corvid species. v4: Hugging Face dataset release. v5: real-time bidirectional synthesis — speculative end-state, no promise.
Can I contribute crow recordings?
Yes. Field recordings under CC-BY-SA license, with city-coarsened location metadata and behavioral notes when available, can be submitted via the contribute page. We don't accept recordings under restrictive Macaulay or commercial licenses, or recordings with identifiable human voices in background.