Research and Development in Natural Language Understanding 
Ralph Weischedel 
BBN Systems and Technologies Corporation 
Duration: December 21, 1984 - September 30, 1989 
Brief Summary of Objectives: There are three objectives of the contract: to perform research and 
development in parallel parsing, semantic representation, ill-formed input, discourse, and tools for 
linguistic knowledge acquisition, and to integrate software components from BBN and elsewhere to 
produce Janus, DARPA's New Generation Natural Language Interface, and to demonstrate state-of-the- 
art natural language technology in DARPA applications. The following software has been distributed: 
natural language system; IRACQ, knowledge acquisition system; System components and knowledge 
bases of Janus; KL-TWO knowledge representation and inference system integrated with Janus; various 
components for DARPA's Spoken Language Systems Project at BBN. 
Summary of accomplishments: 
1. The first 20 months of the effort were devoted to technology transfer of IRUS, the 
understanding component of Janus to the Government. To this end, BBN delivered IRUS, 
its knowledge acquisition tools (IRACQ and KREME), and knowledge bases for lexical 
semantics, lexical syntax, a domain model, and transformation rules to data base structure, 
to Texas Instruments for integration in DARPA's Fleet Command Center Battle 
Management Program (FCCBMP). Working with the Naval Ocean Systems Center, BBN 
officially demonstrated IRUS-86 in summer 1986 as part of the FCCBMP to representatives 
of CINCPACFLT, DARPA, SPAWAR, and NOSC. As of August, 1986, all of the 
components were transferred to NOSC, and BBN began focus on the component research 
goals and on the system integration goals of Janus. 
2. In conjunction with the USC/Information Sciences Institute, USC/ISI's Penman language 
generation component was integrated with IRUS, to provide paraphrases and answers in 
English. This was demonstrated in May, 1987. 
3. In the second version of Janus, the Penman generation component was replaced by the 
Mumble grammar for generation and BBN's Spokesman text planner. This was delivered to 
Lockheed for integration with DARPA's AirLand Battle Management Program in late 1987. 
We estimate that this second version of Janus had greater functionality (e.g., not just 
paraphrase and answer generation, but also multi-paragpraph, multi-page output 
generation), and that the generator was 2-10 times faster than the first version. 
4. Our component research made several direct contributions of technology to BBN's Spoken 
Language Systems effort, including the initial grammar (a unification-based grammar), the 
semantic representation language, a parallel parsing algorithm, and components for 
mapping from the semantic representation (an intensional logic) to code for one or more 
application systems. In addition, we are in process of publishing our research results on 
dealing with errorful, novel, or unclear language; clarification dialogue; centering algorithms 
for reference resolution; knowledge acquisition; and a hybrid semantic representation based 
on intensional logic and a terminological knowledge representation. 
5. Our design for seamless interfaces has been adopted in the human-machine interface for 
DARPA's CASES expert system. 
6. Software resulting from our research and development effort has now been distributed (for 
R&D purposes) to the FCCBMP at the Pacific Fleet Command Center, NOSC, the 
University of Pennsylvania, the University of Massachusetts, Harvard University, USC/ISI, 
Texas Instruments, Lockheed Austin Division, RADC, and NSA. 
441 
