This paper presents a knowledge-based software engineering tool, DASERT (Detection of Anomalies in Software Engineering Requirements Texts), to acquire and validate functional requirements in natural language. The user describes the functional specifications through informal methods, using graphics with comments in natural language. During this elaboration step the system validates the document by processing the comments semantically to detect ambiguities or inconsistencies. To do so it uses natural language processing and knowledge base engineering. DASERT's kernel is a KL-ONE-like semantic network, which helps the semantic parsing of the comments and their semantic representation. This knowledge base is first initialized by the acquisition of the lexical domain knowledge, then progressively enriched with the domain terminology given by the user and with the requirements knowledge extracted from the user's graphics and texts. During initialization and enrichment, the network manager validates the knowledge structurally. This ensures the logical consistency of the base which is then checked for inconsistencies and ambiguities specific to the domain of software requirements. From a software engineering point of view, the originality of DASERT is that it provides a semantic checking of an informal specification by interpreting the natural language comments. From a knowledge acquisition point of view, DASERT allows acquisition from texts to build the kernel of a knowledge base which is then used to guide the semantic parsing of texts during the acquisition of the specification itself. Moreover, the representation formalism provides a unified view of acquisition and validation
