Natural Language Processing for Detecting Undefined Values in Specifications
Daniel Davis, Chia-Chu Chiang · 2022
In this project, we present an algorithm with the use of the parse trees and parts of speech tagging in detecting missing quantified values in user requirement specifications and how the natural language processing results impact the quality of the detections. A comparative adjective found in quantifier phrases is tagged by the parts of speech tagger as a comparative adjective but is phrased as a noun in the parse tree. This could happen because dependencies for phrases of the parse trees may favor nearby words causing the sentence to be phrased incorrectly. We conclude that the overall accuracy of the Artificial Intelligence/Machine Learning parse trees are inadequate to detect quantitative statements as quantifier phrases even when it is very clear of their meaning. When this occurs in the parse tree, we search for patterns in the parts of speech tags to reveal the missing quantitative values. We found this solution effective because the parts of speech tagger has over 56% accuracy tagging sentences with a much higher rate for single words.