Lexical semantics for software requirements engineering – a corpus-based approach
In: Corpus Linguistics 25 Years onSearch for other papers by Kerstin Lindmark in
Current site
Google Scholar
PubMed
Search for other papers by Johan Natt och Dag in
Current site
Google Scholar
PubMed
Search for other papers by Caroline Willners in
Current site
Google Scholar
PubMed
Purchase instant access (PDF download and unlimited online access):
Abstract
In companies that constantly develop new software releases for large markets, there continually arrive new requirements, written in natural language that may affect the development work. Before any decision is made about the requirements, these must be analysed and understood, and related to the current set of implemented and queued requirements. This task is time-consuming owing to the high inflow of requirements, and decision-making would be facilitated by any support that would reduce the requirements analyst’s workload. One of the main tasks is finding requirement duplicates and requirements with similar content and different NLP methods have been tried for this. Simple word matching is one of the methods used for linkage between requirements. If links could be set up not only between words, but also between concepts at different semantic levels, the chances of finding content-corresponding requirements would be greater. One goal of this project is to establish a terminology for requirements as well as to establish (Wordnet-type) semantic relations between terms, in order to enable multilevel linkage. For this purpose, we use a corpus consisting of 1,932 authentic software requirements, written in English of varying grammatical and stylistic quality. First, term candidates were extracted using the WordSmith Keyword function, with BNC Sampler as reference corpus. To find out whether there is any terminology specific to the ‘requirements’ sub-domain of the ‘software’ domain, the documentation associated with the software to which the requirements relate was also used as a reference (separately). Then, lexico-semantic patterns according to Hearst (1992) were used to find hyponymy–hyperonymy relations, and to confirm manually established relations. These analyses were performed on the text both ‘as is’ and, reducing noise somewhat, after POS-tagging by means of the Brill tagger (Brown Corpus tag-set). The results so far suggest that corpusbased methods are of importance to the management or requirements analyses.