• Ph.D. program (Mathematics) Massachusetts Institute of Technology, 1965-1970 (advanced to candidacy 1968).
• B.S. (Mathematics) Massachusetts Institute of Technology, 1965.

Honors and Professional Societies

Sigma Xi (national scientific honorary society).

Association for Computing Machinery (special interest groups on AI, Symbolic Algebraic Manipulation, Automatic and Computation Theory, Computer Uses in Education).

American Association for Artificial Intelligence.

Association for Computational Linguistics.

Experience

Summary

Wide experience for over a decade in artificial intelligence research and development, with strength in natural language processing, knowledge representation, expert systems, rule-based systems, constraint techniques and perception. Interest in applying these tools to real-world problems.

Highlights

Mr. Bobrow has wide experience for over two decades in artificial intelligence research and development, with strength in natural language processing, knowledge representation, expert systems, rule-based systems, constraint techniques and perception. He has a strong interest in applying these tools to real-world problems.

Since May 1996 Mr. Bobrow has been technical director of the BBN component of the Logistics Anchor Desk project of the Joint Logistics ACTD. This project involves the operation of a distributed, collaborative decision support system pulling data from a wide variety of legacy military database systems deployed on more than thirty Sun workstations deployed from Bosnia-Herzegovinia through various major military command centers in Europe and throughout the Continental United States. Mr. Bobrow is leading the development of the next generation of LogAD which will be a web-based system capable of operating on networks of PCs and Sun workstations.

Mr. Bobrow has been the technical lead for several generations of spoken language systems.

1995

Technical manager of project to develop VALAD, the Voice Activated Log Anchor Desk. This project involved building a new spoken language system from scratch and integrating it with a large GUI based database and mapping project used for Logistics analysis. Brought system from concept to prototype in four months, successfully demonstrated at Army's Prairie Warror ’95 to general officers and high-ranking DoD personnel over the course of two week exercise. Overall design of VALAD. Designed and oversaw development of totally new interpreter and database query generator.

Designed and built new RTN parsing. Developed wholly new technology for integrating NL and speech grammars and lexicons, did initial implementation and oversaw final development. Designed new way to build high-quality semantic RTNs for natural language understanding.

1994

Technical manager of natural language components of BBN spoken language effort (ATIS). Major technical contributor to new statistical model of natural language understanding (HUM). Designed and helped implement new socket-based interprocess communicati on mechanism which is being used for next generation of BBN text-understanding projects. Played a key role on the DARPA Semantic Evaluation committee -- helped design semantic annotations for evaluation of all DARPA spoken-language systems, and directed effort to produce an annotation tool. Technical contributor to HOOKAH.

1993

Developed meaning representation for natural language understanding that would support both a priori meaning model and an efficient search procedure. Developed methodology for annotating large amounts of data for HUM by use of Delphi. Sped up context-free parser for HUM by a factor of 5-10. Ability to “find concrete, feasible solutions to hard technical problems”.

1992

Technical manager of SLS Natural Language group. Helped design and oversaw development of the Semantic linker and Frame Combiner which made it possible to deal with the ill-formed input characteristic of speech.

1991

Developed grammar for air-traffic control interactions used in Gisting project and which substantially reduced speech error rate. Supervised effort on discourse analysis of pilot-controller interactions which was used in dialogue clustering algorithms for Gisting. Extended tools for finite state speech grammars which are used by Gisting, CADRE, SLS and Tipster projects. Developed natural language interface to a GeographicInformation System. Development of probabilistic parsing strategies which led to another order of magnitude speedup in Delphi system and improved parse accuracy. Research on multiple underlying systems. Won SDP publication award for paper on multiple-underlying systems.

1989

Parlance. Work on SLS project. Sped up Delphi system by more than an order of magnitude. Parlance and Delphi were the top performing systems in the first SLS systems evaluation.

1988

Technical manager of the Parlance product. Design and implementation of Learner. Porting Parlance to a Navy domain.

1987

Technical manager of the Parlance product, released Version 1 of the Parlance system. Developed the concept for and led the implementation of the Parlance LEARNER™ which made it possible for customers to build new Parlance applications with minimal support from BBN personnel.

Mr. Bobrow was appointed Division Scientist in 1986 in recognition of his strong and continued technical leadership role within the Information and Computer Science Division of BBN Laboratories. He played a key role within BBN to establish an intrapreneurial commercial natural language interface project, funded by an R&D limited partnership. The purpose of this project is to produce and market a set of high quality commercial language products based on the IRUS transportable natural language research system. This intrapreneurial unit was set up in October 1984 with a three-year R&D partnership funded by a subsidiary of Prudential-Bache.

Since that time, Mr. Bobrow has been the research and development manager for a group of seven scientists and systems programmers. He has been a major direct technical contributor to the project, including the development of a novel algorithm that efficiently combines spelling correction and morphological analysis for a large dictionary, the design of a compiler for semantic interpretation rules, the design and implementation of a new grammar compiler and parser control structure, and new techniques for anaphora resolution, ellipsis and conjunction. His work has resulted in an order of magnitude speed improvement over the original research base, as well a substantial improvement in linguistic capability. He has also played an important role in the overall project management and marketing effort, with the goal of producing a system that is not only technically advanced but meets market needs.

Prior to his appointment as Division Scientist, he was responsible for the design, implementation and continued development of the RUS natural language system, which has formed the basis for all BBN natural language research since 1976. This system has also been used by the USC Information Sciences Institute, the University of Delaware, the University of Illinois, the National Library of Medicine, General Motors Research Laboratory, and Univac. The highly efficient system contains one of the three largest computational grammars of English, and makes use of semantic processing during parsing to guide the syntactic analysis process. The close interaction between semantics and syntax is mediated by a highly modular interface which has made it possible to use the identical parser with four different semantic processors.

The system operates on the TOPS-20 and TENEX systems, the VAX-11, the Xerox 1100 series processors and the Jericho, a BBN developed LISP machine. He led a group of four people who produced IRUS, a transportable natural language database interface system based on RUS, which currently operates in conjunction with the SYSTEM 1022 dbms, and was also interfaced with the Britten-Lee IDM 500 database machine. Mr. Bobrow also led a multi-site (BBN and ISI) design effort for NIKL, a new implementation of the KL-ONE knowledge representation language used within the DARPA community.

Participated in the of implementation group for NIKL. He played a major role in the design and implementation of KL-ONE, the primary knowledge representation system used in the AI department at BBN.

Mr. Bobrow was the principal investigator on several Navy contracts to determine the applicability of artificial intelligence techniques to tactical situation assessment afloat. Designed an expert system to assist tactical action officers in threat assessment, situation projection and detection of deception techniques used by opposing forces. Led group of four senior scientists in examining the use of rule-based systems, knowledge representation, constraint propagation and other inference techniques in identifying composition of hostile forces, determining possible threats and identifying deceptive maneuvers.

He designed and implemented the parser and interpreter that provided an English interface for an instructional system for teaching causal reasoning. As part of the ARPA intelligent terminal project, Mr. Bobrow designed and implemented a rule-based system in INTERLISP capable of running all code written for the RAND RITA rule-based system.

Led a group of three people in developing a natural language interface to the HERMES message system, including the embedded data base manager for messages. As part of the effort in implementing INTERLISP on a specially microcoded PDP-11/45, developed a system for determining the detailed memory use of the PDP-10 INTERLISP system. This involved extensive modifications to a large assembly language program that interpreted PDP-10 assembly language instructions and was able to run the entire INTERLISP system and collect data on the memory usage, paging performance and critical code segments of the system. These results were also used to double the speed of the PDP-10 version of INTERLISP.

Developed BCPL programs to transmit memory usage data on an operating INTERLISP system, over an internal network to be displayed on the IMLAC display computer. He designed and developed a flexible control framework for AI-based computer aided instruction used in NLS Scholar, a system that engaged in an interactive teaching dialogue with a student learning to use the NLS document preparation and editing system. This system provided a tutorial on text editing, let the student the NLS system for exercises, diagnosed the results and provided appropriate remedial lessons. Also, he worked on the problems of using simulation models to aid inferencing techniques in the SOPHIE system for teaching electronic troubleshooting.

Designed memory structures for a simulated Martian robot exploration vehicle. Worked to set up the initial JPL robotics project and helped design techniques for representing spatial information for robot manipulation and navigation. Developed techniques for running programs distributed on several computers on the ARPANET.

While at the University of California, Mr. Bobrow developed graduate and undergraduate programs in artificial intelligence, data structures, and combinatorial analysis and logic for computer science majors. Supervised the development of UCI-LISP (which is now in use at many universities as a major language in teaching and research in artificial intelligence) as well as the development of versions of the L6 data structure language of the DEC PDP-10 and XDS Sigma-7 computers. Continued theoretical research on artificial intelligence and perception.

While working part time at BBN, he developed several large BBN-LISP programs including an interactive database management system EDMS with flexible facilities for data security, and the initial list structure editor for LISP.

At Stanford Research Institute, he developed and tested the initial algorithms for the visual system of the SRI robot project (SHAKEY). Helped design hardware to perform edge extraction.

At IBM, Mr. Bobrow, produced the automatic algebraic expression simplification program for FORMAC, one of the first systems for symbolic algebraic manipulation. He went to IBM Hursley laboratories to help design the interface between FORMAC and the first implementation of PL/1.

As a graduate student in mathematics at Massachusetts Institute of Technology, he worked with Professors J. Y. Lettvin and M. Minsky on developing a theory interrelating neural functioning and cognitive and perceptual processes.