Software Quality Journal Software
Detecting ambiguous sentences in requirements documents using ambiguous word dictionary
Abstract
In this paper, to automate the detection of ambiguities included in requirements specification documents (simply, documents), we focus on ambiguity at the lexical level and develop a technique to construct a dictionary of ambiguous words. A word becomes ambiguous in some cases and not ambiguous in the other cases, so we cannot decide the ambiguity of words uniquely or definitely. To address this problem, we provide a value of “likelihood” of ambiguity for each candidate of an ambiguous word in the dictionary. Our “likelihood” of ambiguity of a word is based on how many people think it ambiguous. First, we ask several subjects to extract from various documents the words that they think ambiguous. For each extracted word, their judgment may vary and we calculate its “likelihood” of ambiguity based on the majority and the deviation of their judgment. Next, to expand the vocabulary of ambiguous words, we analyze the other documents to collect words which are semantically similar to the words that our subjects extracted. Semantic similarity degree is calculated by Word2Vec and cosine similarity. Furthermore, we develop a tool to automate ambiguous word detection from documents and experimentally evaluate its usefulness. Our tool has the function of expanding the vocabulary of ambiguous words using semantic similarity.