This project will explore how artificial intelligence (AI) can support the translation of emergency warnings for culturally and linguistically diverse (CALD) communities. It will examine where AI adds value, its limitations and how those can be addressed through human oversight or complementary processes. The aim is to understand how translation quality affects community comprehension and response during emergencies.
This project is currently open for Expressions of Interest.
Project teams responding to this call for Expressions of Interest (see PDF in top-right corner) are required to submit their response using the Centre’s current EOI submission form.
EOI proposals are due by 5:00 pm AEDT on 19 December 2025 to research@naturalhazards.com.au
An online webinar, scheduled for 1:00 pm AEDT on 27 November 2025, provided a more detailed briefing of the project and the opportunity for interested parties to pose specific questions. Watch the recording below.
Project details
This project concept was submitted by the NSW Department of Customer Service.
This project aims to understand both the opportunities and limitations of using AI to translate emergency warnings. It will look at:
- the areas AI provides real value in supporting emergency warning translation
- the limitations, risks or challenges this presents, and how any limitations can be mitigated
- any additional tools or processes that can work with AI to supplement any shortcomings
- the impact on the CALD communities receiving the warnings, particularly in terms of usability, comprehension and trust
- the extent communities are willing or able to accept mistakes in AI-generated translations
Critically, this project will also investigate whether any limitations in AI translation impact the way CALD communities understand and respond to emergency warnings.
Frequently asked questions
Q) Are the word limits in the submission form a guideline?
A) Each question in the submission form has a required word limit. Submissions with statements exceeding the word limit will be deemed non-compliant.
Q) What do you mean by “Total cumulative FTE contribution over the life of the project"?
A) This means the total FTE per person over the life of the project. Maximum total FTE for each person is 1.0. Cumulative is to add up all the personnel.
For example:
- If someone is contributing 1 FTE per year for three years, then their FTE is 1.000 FTE
- If someone is contributing 0.1 FTE per year for three years, then their FTE is 0.100 FTE
- If someone is contributing 0.05 FTE for two years of a three-year project, then their FTE 0.033 FTE
- If someone is contributing 0.5 for two years and 0.1 for 1 years of a three-year project, then their FTE is 0.367 FTE
Q) Can the Centre connect interested parties with other agencies or organisations involved in developing this project, to discuss it further and assist in developing an EOI submission?
A) While an EOI is open for submissions, the Centre procurement processes don't allow Centre staff to connect interested parties, as this can cause conflicts of interest. However, interested parties are able to contact whomever they wish, excluding those organisations listed on the EOI, to discuss the project and explore opportunities for collaboration or potential consortium submissions. If an online project briefing has been organised for an EOI, we encourage interested parties to attend the briefing to ask questions and meet other interested parties.
Q) Can interested parties contact the Centre with questions about the project design before submitting?
A) While an EOI is open for submissions, Centre procurement processes don't allow Centre staff to share knowledge of the project design directly with interested parties. However, if you have questions, you can email research@naturalhazards.com.au and you will receive a timely response if appropriate, according to our procurement processes.
Q) Is there a preference for a project team to be from a single research organisation, or from across multiple organisations?
A) The Centre has no preference for either a single organisation or a multi-organisation project team. EOIs will be accepted from either and will be evaluated against the evaluation criteria in the same way.
Q) Can the proposed project team include researchers from government agencies or research consultancies, as well as university-based researchers?
A) EOIs will be accepted from multi-organisation project teams, and project teams can include researchers at government agencies and research consultancies. The proposed project team's capacity to undertake the project will be evaluated in the same way whether researchers are university-, consultancy- or agency-based.
Q) Would academic salaries (excluding administrative overheads) be eligible to budget?
A) Yes, academic salaries can be included within the project budget, but only when they are direct project costs and their salary is not already covered elsewhere. E.g. the salary costs of a contract researcher who is actively working on the project can be included up to the FTE component they are contributing. However, the salary costs of a tenured academic who is already paid through the university/research organisation would be included as an in-kind contribution up to the amount of their FTE contribution to the project.
Q) What do you mean by "peer review"?
A) This is a quality control process. Where requested, the final report must be reviewed by someone who is an expert in the field and independent of the project.
Q) At what rate can salary on-costs be charged?
A) Administrative overheads are limited to direct salary-related on-costs. Indirect cost recoveries are not to be included. Other administrative overheads and indirect costs can be included as in-kind contributions. Where the salary on-cost rate exceeds 28% this will need to be justified.
Q) Can equipment costs be included in the budget?
A) Ordinarily, project funds are not to be used to purchase equipment. Where funds for equipment are requested, they will need to be justified.
Q) Can international research teams apply? Alternatively, can international research teams be part of a consortium bid?
A) International research teams can be part of research projects when they are part of a consortium submission that is led by an Australian research organisation. The international team would need to be subcontracted by the lead Australian based research organisation. Please note that all budget submissions must be in Australian dollars and the lead organisation must bear the cost of funds transfer and responsibility for due diligence as required under Australian Foreign Interference regulations.
Q) Is there an option of registering my interest as an independent researcher that could assist a larger team or are you only able to consider proposals that address the entire project?
A) Yes, you can submit an EOI as an independent researcher addressing part of the project requirements. In the EOI submission form, please outline which aspects of the project you can address. Once the EOI closes, we can consider linking researchers together, although whether or not we can do that will depend on a range of factors, so we can’t guarantee that outcome. Of course, the other option is to reach out to potential collaborators now about putting in a joint EOI submission that addresses the entire project.
Q) Are you interested in comparing specific large language models in the evaluation?
A) Many languages do not have large language models. The original proposal focused on languages most used in New South Wales. But in places like Western Australia, Indigenous communities need access to warnings, and models for their languages don’t exist.
It would be useful to check which models are available. For languages with strong models, we can look at their impact. For others, like sub-Saharan African languages spoken in Victoria, there are no models, even though large communities use them.
The evaluation should consider:
• The impact where models exist.
• The impact where they don’t.
• Possible ways to fill the gaps.
Q) Do you prefer focussing on open-source models?
A) Since government organisations would be using these systems, open-source models could be a good starting point. However, we would not restrict the evaluation only to open-source models.
Q) How many languages we need to AI translate? Do we need to consider a very large number of languages and communities, or can the research be exploratory with a smaller sample of selected communities and languages?
A) It does not need to cover a very large number of languages. What matters is having a range:
• Some languages that are already well supported by AI (for example, certain Chinese languages).
• Some that are less developed in terms of AI capability.
• And ideally, some in between.
The actual number is less important than ensuring diversity across these categories. The focus should remain on languages within the Australian context, since evaluation requires engagement with communities. Research should therefore select languages where connections with communities can be made to assess translation quality and impact.
Q) Do we have to be a native speaker of that language?
A) No, being a native speaker is not required. However, assistance will be needed to evaluate the quality of translations. Different translation tools produce different results, so part of the work involves comparing these tools.
Access to speakers of some languages is essential, because without them there is no reliable way to assess whether the translations are accurate.
Q) Can the research be focused on specific geographic areas or does it have to be one on a large or national geographic scale? Especially on the point of demographic data.
A) It could be focused on specific areas if you are able to extrapolate to a broader scale. As long as geographic areas have good mix of demographic languages. It should be also considered the hazard exposure (e.g. flood-prone or fire-prone areas), to ensure messages are relevant to participants’ real-life experiences. Research can be focused on specific geographic areas, provided the findings can be extrapolated more broadly. The key requirement is that selected areas include a good mix of language types—both those with established large language models and those without. This approach allows the research to capture insights across different contexts and then extend them to a wider, potentially national scale.
Q) Do you expect any industry contribution to the project?
A) Industry contributions are welcome, and we often work with research consultants. The best way to outline contributions is through the budget template provided on the project’s website. Contributions can be in-kind or included as part of the budget for services, depending on project needs. There is no restriction on industry involvement. The only consideration is when industry organisations are already part of Natural Hazards Research Australia’s participant base, in which case some additional questions may be asked. Otherwise, industry participation is fully supported.
Q) Would it be required to include a member of the research team who identifies as First Nations to ensure appropriate cultural understanding and engagement in the project?
A) It is not a requirement. Projects are offered to the research community to enable applicants to propose the approach they consider most effective. Accordingly, the composition of the research team is at the discretion of the applicant and should reflect the expertise needed to address the project’s challenges. While such expertise may be appropriate in some cases, it is not formally mandated. From the Centre’s perspective, this relates to overall research capability. The key consideration is not individual skills in isolation, but whether the team’s collective expertise is relevant to deliver the project and meet the expected outputs.
Q) Could part of the project outputs include an evaluation pipeline or technical tools that can be reused?
A) Yes, certainly. The work will involve not only research but also defining the processes required to achieve reliable outcomes. It will not be a simple matter of inputting data into an AI tool and receiving results. For emergency service organisations, there will likely need to be changes in workflows and additional steps along the way. Developing tools or pipelines that can be reused would therefore be valuable.
Q) Has any past research been done on which languages AI has already been explored for disaster warnings?
A) No, not specifically. New South Wales State Emergency Service (SES) has identified seven languages of particular interest, based on significant local communities who speak them and may be at risk from natural hazards.
This does not reflect prior research into AI translation for those languages. Rather, it highlights known communities and priority languages within New South Wales. Similarly, each state would have its own set of languages tied to local demographics, but there has not been research conducted in this area yet.