Transition network grammars for natural language analysis
Communications of the ACM
| Hi-index | 0.00 |
The requirements specifications phase of the software life cycle plays a vital role within the overall development process. Existing methodologies provide a formalized specification document for use by the design phase. However, such methods generally rely on heuristic or ad-hoc methods for their initial conversion from the English requirements document to the formalized system specification.A methodology is under development by which the English requirements document is converted into a reliable set of knowledge for use within the specification process. Such a method would be utilized to help improve the overall quality and reliability of the resulting specification document. This method provides a three-phase conversion from the initial English document into a knowledge-based tool for use by the system specification group.The initial phase in the methodology uses as input a standard English requirements document and produces a structured parse of this text. This step utilizes the concept of Augmented Transition Networks [1] for the conversion. The grammar used in constructing the ATN is based on Chomsky's transformational English grammar [2]. Inherent ambiguities within the document are resolved interactively with the user/analyst.The second phase converts this structured parse into a set of Prolog-based facts and rules that encapsulate the knowledge within the initial document. These facts can be divided into three major classifications: hierarchies, properties and events. Hierarchy facts detail the relationships that exist between the entities themselves, as described within the requirements document. Property facts specify the specific characteristics (physical and logical) of an individual entity. Finally, event facts provide information regarding the actual actions that take place within the proposed system. The specific action being performed, the entity invoking this action, and all other entities in this event are detailed.In addition to these facts, other relevant information is stored within the knowledge base. Rules exist for the definition of the specific actions and entities referenced within the requirements specification document. The quantification of all abstract modifiers used within the document is performed interactively with the user/analyst. In addition, the method links the origin of all knowledge-base information back to the original requirements specification document itself. A final set of rules permits restrictions to be placed on specific entity characteristics (properties) or upon events themselves.The final phase checks the accumulated knowledge for reliability, understandability and traceability [3]. Reliability checking includes examining the aspects of correctness, completeness, consistency and necessity. Testability relates to the overall quantification, structure, conciseness and self-description within the document. Traceable information augments the user's ability to track the correspondence of the entities and actions within the generated knowledge base back to the initial requirements document.