Capability 3.1.2.2

Cultural & Linguistic Alignment

Definition

Refining models so they accurately reflect diverse local languages and First Nations communities, while guarding against exploitation and respecting cultural norms and Indigenous data rights.

Examples

Te Hiku Media Māori language models
NLLB-200 (Meta).

Indicators

Number of culturally aligned models released domestically
Inclusion of Australian English and First Nations languages data with community consent

National AI Plan

PLAN COMMITMENT

Limited commitment

Acknowledges importance of inclusive AI development and culturally appropriate datasets – and signals intent to explore unlocking local data (Action 2) – but doesn't have specific commitments on model adaption and alignment. Limited commitment by government is a missed opportunity to increase domestic maturity in a globally scarce capability with high public interest value. It is particularly important to increase agency in this capability if investments are not made in General Purpose & Frontier Models (3.1.1.10).

Stocktake

Building Speech Recognition Systems for Language Documentation: The CoEDL Endangered Language Pipeline and Inference System (ELPIS). Foley et al. 2018. The 6th Intl. Workshop on Spoken Language Technologies for Under-Resourced Languages. 200-204. https://www.researchgate.net/publication/328067867_Building_Speech_Recognition_Systems_for_Language_Documentation_The_CoEDL_Endangered_Language_Pipeline_and_Inference_System_ELPIS
Example: SAIGE developed by Sapia.ai

Is the Stocktake missing evidence?

Share additional information for potential future Assessments