Inspired by large language models (LLMs) such as GPT-3 and BLOOM, as well as recent improvements in low-resource NLP capabilities, the Army wants up to two small businesses “to leverage those advances.”
The goal of the technology is twofold: to enable communication with “previously inaccessible populations” while avoiding knowledge losses incurred in “default to-English approaches.”
Commercial-access machine translation (MT) falls short on a few counts. In addition to confidentiality and privacy concerns, most products do not offer bespoke model training.
“Most commercial services perform higher level analysis (stance detection) on already translated media where best practices would necessitate development in the source language,” reads the June 21, 2023 press release.
The Army already has a few leads for potential candidates, among them zero-shot approaches and crowd-sourced manual labeling of training data, and the press release even cites several research papers on MT, by Meta, Microsoft, University of California, Berkeley, and Johns Hopkins University.
The Army Applied Small Business Innovation Research (SBIR) Program, led by the Office of the Assistant Secretary of the Army for Acquisition, will oversee the application and award process, which is described as “extremely competitive.” Small businesses, as defined by SBIR eligibility requirements, can have a maximum of 500 employees.
In Phase I, the SBIR Program will award a maximum of two businesses up to USD 250,000 each, to be used over an anticipated six-month period to develop MT technology, resulting in a proof of concept white paper. The deadline for Phase I applications is July 25, 2023.
One company will receive a Phase II contract worth up to USD 1.5m to last about 18 months, during which time the recipient will continue to research, develop, and eventually prototype the solution. The deployed MT model should be tested and evaluated in line with standard metrics.
Phase III consists of continuous training and development for use cases specific to the US Army Pacific, with the goal of putting the technology into soldiers’ hands, though the timeline for actual implementation is unspecified.