To create effective and accurate two-way, free form, spoken language translation devices, they must be supplied with appropriate training data. The goal of the DARPA TRANSTAC (Spoken Language Communication and Translation System for Tactical Use) program is to demonstrate capabilities to rapidly develop and field this technology so speakers of different languages can communicate with one another in real-world tactical situations. A critical component is to generate data sets to both train and evaluate the technologies. A novel approach was developed to collecting these data through the creation of innovative data collection and evaluation scenarios. This paper describes the scenario generation methodology used to support the TRANSTAC data collections and evaluations.
Proceedings Title: Proceedings of the 2009 International Test and Evaluation Association Symposium
Conference Dates: September 28-October 1, 2009
Conference Location: Baltimore, MD
Pub Type: Conferences
Performance evaluation, TRANSTAC, speech-to-speech translation systems, military technology