Detection of anaphoric ambiguity of expert formulations in decision support systems

• Date: 2022
• Main Author: Дубок, М. Ю.
• Format: Article
• Language: Ukrainian
• Published: Інститут проблем реєстрації інформації НАН України 2022
• Subjects: підтримка прийняття рішень; система підтримки прийняття рішень; неоднозначність; зниження неоднозначності; виявлення неоднозначності; анафорична неоднозначність; decision support; decision support system; ambiguity; ambiguity reduction; ambiguity detection; anaphoric ambiguity
• Online Access: http://drsp.ipri.kiev.ua/article/view/262928
• Journal Title: Data Recording, Storage & Processing

The quality of decision support system recommendations directly depends on the unambiguous interpretation of expert formulations. Because formulations are provided by experts in natural language, which is characterized by ambiguity at all language levels, it is critical to reduce the ambiguity of expert formulations. Using ambiguity detection to reduce it, attention is paid to the subtype of syntactic ambiguity — anaphoric ambiguity, which is a significant part of all existing ambiguities and manifests itself in the form of the possibility of referring to several previous words. The article proposes a simple method for searching and classifying anaphors, as well as finding their antecedents for the detection of this subtype of syntactic ambiguity. The developed method is implemented as a standalone computer program and has absolute recall, high accuracy and satisfactory precision. Unlike the method of Yang et al., the proposed method requires only an accurate list of anaphors for a particular language and a part-of-speech tagger that can determine gender and number of nouns, which allows one to use a small amount of information to detect anaphoric ambiguity. Since one anaphor in different genders can coincide in certain cases, each possible gender is attributed to the anaphor. Also, unlike the method of Yang et al., anaphoric ambiguity is detected only if at least two antecedents precede an anaphor within one formulation, because in all available formulations no case of pragmatic anaphoric ambiguity is found, i.e. reference of an anaphor to an antecedent in the previous formulation. Matching the gender and number of anaphors and antecedents has been introduced to increase precision without losing recall. For each formulation, the method also records anaphora, the number of their possible antecedents, the antecedents themselves and preserves the text of the formulation. The method will be used with other developments that detect ambiguity of different types in order to reduce the ambiguity of expert formulations in decision support systems.