Status: ConceptWeights availability: Not releasedLicence: Licence not yet confirmed
Auslan Pose Encoder (concept only)
A concept for a multimodal pose encoder for Auslan research. No model has been trained. No weights exist. It is documented to be honest about scope, not to claim capability.
Intended use
- Research exploration only, subject to community governance.
Out-of-scope use
- Interpreter replacement.
- Production translation of Auslan.
- Any claim of language understanding from pose accuracy alone.
Metrics
No verified metrics are published for this model.
Model type
Concept: multimodal encoder over pose/landmark sequences.
Architecture
Not built. Any architecture would need to model hands, body, face and timing jointly.
Training data
None. A responsible dataset does not yet exist (see the Auslan Community Corpus concept).
Performance note
No performance exists to report. Any figure would be fabricated.
Ethical considerations
Building this without community consent and governance would be unethical. Documenting it as a concept keeps the roadmap honest.
Limitations
- This is a concept, not a model. No weights, no metrics, no deployment.
- Gesture-level accuracy would not imply language equivalence.
Known failure modes
- Not applicable — no model exists.