SBIR-STTR Award

This Small Business Innovation Research Phase I project will explore a potential solution to one of the key problems hindering the development of practical computer systems that understand natural language-English, German, etc. In particular, although several different approaches have been taken to codifying linguistic rules, all have suffered from a common scaling problem: a the number of rules is increased to approach coverage of a usable subset of the language, the interactions become so complex and unmanageable that the systems tend to collapse. adding new rules to cover some previously neglected aspect of the language tends to `break` old rules, introducing bugs so that sentences that were once properly interpreted no longer are. This situation is very similar to what happens in the development of large software systems: the process of adding more and more features to existing bodies of software tends to break down as conflicts rapidly accumulate. The recent advent of object-oriented programming (OOP) has had a major beneficial impact on this general software problem. The opportunity exists to adapt OOP techniques to the codification of linguistic rules, allowing the creation of systems that break through the complexity barrier that has previously prevented the rapid development of natural language systems for a wide range of applications. The research objective is to demonstrate a concrete means of structuring linguistic rules as independent `objects` with inheritance, following the OOP paradigm. Goodbyte, Inc. anticipates that this technique will make it far easier to extend natural language grammars without disrupting previous coverage.

This Small Business Innovation Research Phase II project conducted by Goodbyte, Inc. seeks to adapt Object-Oriented Programming (OOP) technology to the construction of computer systems that understand statements and questions posed in natural human language, such as English or Japanese. Previous Natural Language Processing (NLP) systems have suffered from a common scaling problem: As the number of linguistic rules is increased to cover the language more fully, interactions among rules become so complex that the systems eventually collapse. This situation is similar to adding feature upon feature to large software systems. The Phase I studies indicated that OOP concepts, which have been of great aid in organizing conventional software, can be adapted to linguistic rules as well, allowing them to be structured as independent objects with inheritance, while the encapsulation of information within an object holds the promise of extending rules without disrupting previous coverage. With the basic methodology for Object-Oriented Grammars (OOG) now in place, the chief objective for Phase II is to put OOG to the acid test by developing a prototype system with substantive coverage. In the process, the equivalent of an OOP class library will be developed, upon which new NLP applications can be based. It is anticipated that such a class library will allow natural language interfaces to be developed rapidly for a variety of applications, just as OOP and associated visual programming tools have made it possible to rapidly develop graphical user interfaces. Potential applications of a computer system that understands statements and questions posed in natural language include conversational access to information in databases, catalogs, almanacs, encyclopedias, and documents distributed across networks such as the Internet; automatic translation of documents from one language to another; document classification based on content; and automated response to e-mail mes sages, including electronic commerce. In combination with speech technology, the possibilities for machines to converse with people are enormous.