Defence-to-Civilian Language Translation using LLMs

This project explores how defence-specific terminology and phrasing can be translated into civilian, academic, or general English and vice versa. The goal is to build a bidirectional natural language translation layer that bridges domain-specific language used by defence and national security personnel with general-purpose language models and user interfaces. The outcome will support improved interoperability and accessibility of AI systems within defence and national security contexts.

The expected activities will involve collecting and aligning paired examples of defence and civilian terminology, developing an approach to fine-tune or prompt-tune existing large language models, and evaluating translation accuracy and contextual fidelity. This work sits within a broader collaboration between UNSW and Cyndr on AI capability intelligence, funded by Defence Trailblazer. The translation layer will contribute to Cyndr’s ontology transformer concept, which seeks to interpret user intent across different technical and cultural domains (e.g. military, research, industry).

Australian citizens and permanent residents will be given preference due to project context. This project suits candidates with strong, hands-on technical skills in pre-training and post-training language models, along with good foundation in natural language processing.