An Investigation into Fuzzy Negation in Semantic Similarity Measures

Machine computation of semantic similarity between short texts aims to approximate human measurements of similarity, often influenced by context, domain knowledge, and life experiences. Logical negation in natural language plays an important role as it can change the polarity of meaning within a sentence, yet it is a complex problem for semantic similarity measures to identify and measure. This paper investigates the impact of logical negation on determining fuzzy semantic similarity between short texts containing fuzzy words. A methodology is proposed to interpret the implications of a negation word on a fuzzy word within the context of a user utterance. Three known fuzzy logical not operators proposed by Zadeh, Yager and Sugeno are incorporated into a fuzzy semantic similarity measure called FUSE. Experiments are conducted on a sample dataset of short text inputs captured through human engagement with a dialogue system. Results show that Yager's weighted operator is the most suitable for achieving a matching threshold of 90.47% accuracy. This finding has significant implications for the field of semantic similarity measures. It provides a more accurate way to measure the similarity of short texts that contain fuzzy words combined with logical negation. Whilst validation of the approach on more substantial datasets is required, this study contributes to a better understanding of how to account for logical negation in fuzzy semantic similarity measures and provides a valuable methodology for future research in this area.