OVERVIEW OF LEXIKOS

To ancient Greeks, our corporate name would mean "of or for words." To modern high technology firms, it means high-performance English text-analysis software, which we can efficiently integrate under license into a wide range of computers, specialized peripherals, and advanced business applications.

We are a consulting and software development company specializing in the several interlinked technologies of intelligent text processing. Our multi-disciplinary staff has considerable expertise and commercial experience in software engineering, computational linguistics and lexicography, knowledge representation, parsing systems, symbolic programming, and all facets of software system design. We work as a unified team to produce and customize portable Common Lisp software which can analyze and semantically interpret English text.

Business Orientation

Natural Language Processing (NLP) applications always embody two fundamentally different kinds of software - the general-purpose kernel, devoted to analyzing English inputs via application-independent logic; and the application-specific remainder, which transforms the linguistic analysis done by the kernel logic into useful external work, customized data bases, or other practical results.

We believe it is foolish and counterproductive for each high-tech company to create its own kernel just to develop an NLP application. The kernel is the most critical element in the technical viability of the application, and the most expensive part by far. It includes a lexicon, morphological utilities, a parser, and semantic and discourse aids - modules which often make up 95% of the final application - plus many specialized tools needed to build, debug, install, and use these components.

All of these elements are intrinsically reusable, application-independent software. By relying on Lexikos for them, several client firms focusing on diverse NLP markets and applications can divide their costs of kernel software creation, maintenance, and enhancement without getting in each other's way. Each client who retains us gets a system kernel built by linguistic specialists, not by application coders (who create the rest of the NLP product). Overall, we believe this is a sensible division of technical labor, from which all companies involved benefit.

We always welcome new clients for our linguistic kernel software, the coverage and extent of which we constantly improve. For further information, or to arrange technical presentations or demonstrations, please write or give us a call.

Technical Orientation

Our focus is on English text, under a proprietary knowledge-based architecture that emphasizes the value of detailed, bulk vocabulary data. All parts of our linguistic software use this data heavily at run-time, as our semi-deterministic syntax analyzer calls utilities to handle context-dependent lexical disambiguation and case-frame modeling. The analysis system which results is fast, robust, accurate, and flexible.

As each input phrase is analyzed, interpretive logic in our system transcribes it into application-specific commands which reflect its contextual meaning. When these commands are processed, the net effect is very much as if the original text had been "interpreted" like a high-level program. It will "do" whatever our client has previously specified would be useful within his particular application.

Architecturally, this flexible kernel can be used on text files, dialogue inputs, or voice transcripts. Its parses are based on syntactic clues, common sense naive semantic models, and, optionally, operator feedback. Unlike most current NLP systems, ours is thus not limited to a child-like grammar, a specific semantic domain, perfect input text, or even a predefined vocabulary (though each of these can boost its speed/accuracy tradeoff). Most applications get good accuracy and coverage right away, with non-interactive system throughput approaching human reading rates even on modestly priced, high-end PC hardware.

Ultimately, our system's power derives from the bulk data on English held in our standard English lexicon, which our staff of computational linguists is constantly testing, refining and extending. No lexicon is ever complete, so we let ours be readily tailored and/or expanded as specific clients or applications dictate, both before installation of our kernel and at need later during a job, through well-integrated vocabulary-expansion utilities. These utilities ask simple and direct questions about contextual word meanings and features, and they can be easily used by the on-site clerical staff members who operate the system.

The technical partnership we build with each client is vitally important to our own success. Any client we accept gets a responsive team of expert advisors on all aspects of our NLP software or that of others, plus most-privileged-customer status on any kind of consulting or prototyping efforts we undertake. We will be wise and willing partners in most new efforts our clients propose.