Requirement Mining in Technical Documents

In this paper, we first develop the linguistic characteristics of requirements which are specific forms of arguments. The discourse structures that refine or elaborate requirements are also analyzed. These specific discourse relations are conceptually characterized, with the functions they play. An implementation is carried out in Dislog on the platform. Dislog allows high level specifications in logic for a fast and easy prototyping at a high level of linguistic adequacy. 1 The Structure of Requirement Compounds Arguments and in partticular requirements in written texts or dialogues seldom come in isolation, as independent statements. They are often embedded into a context that indicates e.g. circumstances, elaborations or purposes. Relations between a requirement and its context may be conceptually complex. They often appear in small and closely related groups or clusters that often share similar aims, where the first one is complemented, supported, reformulated, contrasted or elaborated by the subsequent ones and by additional statements. The typical configuration of a requirement compound can be summarized as follows: CIRCUMSTANCE(S)/CONDITION(S),PURPOSE(S)--> [REQUIREMENT CONCLUSION + SUPPORT(S)]* <-PURPOSE(S), , ELABORATION(S) CONCESSION(S) / CONTRAST(S) In terms of language realization, clusters of requirements and their related context may be all included into a single sentence via coordination or subordination or may appear as separate sentences. In both cases, the relations between the different elements of a cluster are realized by means of conjunctions, connectors, various forms of references and punctuation. We call such a cluster an requirement compound. The idea behind this term is that the elements in a compound form a single, possibly complex, unit, which must be considered as a whole from a conceptual and argumentative point of view. Such a compound consists of a small number of sentences, so that its contents can be easily assimilated. 2 Linguistic Analysis 2.1 Corpus characteristics Our corpus of requirements comes from 3 organizations and 6 companies. Our corpus contains 1,138 pages of text extracted from 22 documents. The main features considered to validate our corpus are the following: specifications come form various industrial areas; documents are produced by various actors; requirement documents follow various authoring guidelines; requirements correspond to different conceptual levels. A typical simple example is the following: Inventory of qualifications refers to norm YY. < /definition> Periodically, an inventory of supplier’s qualifications shall be produced. < /mainReq> In addition, the supplier’s quality department shall periodically conduct a monitoring audit program.< /secondaryReq> At any time, the supplier should be able to provide evidences that EC qualification is maintained. < /ReqCompound>