Chapter 6: Operational Engines 6.2 Implementation of Operational Engines 6.2.1. Bridging the Gap between Research Advances and Operational Users 6.2.1.1. Introduction Handbook of Natural Language Processing and Machine Translation 6.2.1.2. Pre-gale Efforts: Early Applications of Hlt in the Intellige

The goal of the GALE Program is to empower the warfighter through the use of language technologies. In order to accomplish this goal, the technologies must be integrated into operational engines that meet the needs of both warfighters and those who support them. That integration imperative leads directly to the two key parts of this chapter: how to perform that integration, and assessing how well we have met operational needs. The chapter begins with a look at the deployment history over the past decade and a half for speech and language technologies in operational environments. As Kathleen Egan and Allen Sears (Section 6.2.1) explain, this has been an evolutionary process driven by an increasingly rich understanding of warfighter needs. The second contribution provides details on the design of an operational engine that has recently been deployed in operational environments around the world, including in direct support of operational forces in the field. As Daniel Kiecza and his colleagues (Section 6.2.2) explain, meeting the need for responsive and accurate performance by integrating systems that each have a different development heritage can be a substantial undertaking, sometimes requiring development of new technologies. As an example of that process, Amit Srivastava and Daniel Kiecza (Section 6.2.3) describe the development of a new technique for establishing sentence boundaries during speech recognition that substantially improves the latency for subsequent translation of the recognized speech. The integration challenge in GALE extends well beyond pairing components, however. John Pitrelli and his colleagues (Section 6.2.4) introduce the full scope of the challenge by describing the GALE Interoperability Demonstration (IOD) system. Coordinating real-time processing by eleven different types of language technologies would be a substantial undertaking under any circumstances; doing so in a distributed manner using the facilities of seven research groups spread over thousands of miles in three countries raises the challenge to an entirely new level. But, it is exactly that level at which we must learn to work if we are to learn how best to integrate the diverse language technologies that we are now able to create. The core technology behind the GALE IOD system is the Unstructured Information Management Architecture (UIMA), a flexible integration platform designed to meet the unique demands of distributed language processing. Eric Nyberg (Section 6.2.5) and his colleagues describe how UIMA-enabled components can be shared using GALE’s UIMA Component Repository (UCR), and how individual researchers can flexibly configure entire UIMA processing environments using GALE’s UIMA Component Container (UCC).