Assessment of Winograd schemes and the use of commonsense knowledge for the resolution of ambiguities

• Grant number: 18/09681-4
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: May 01, 2019
• End date: December 19, 2021
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: Fabio Gagliardi Cozman
• Grantee: Hugo Neri Munhoz
• Host Institution: Escola Politécnica (EP). Universidade de São Paulo (USP). São Paulo , SP, Brazil

Abstract

The primary objective of this project is to make an in-depth evaluation of how much the tests based on the Winograd schemes embrace the phenomenon of common sense in dealing with the semantic problem of ambiguity. Although common sense is a phenomenon constantly present in human life, it is a phenomenon of difficult characterization, whether in Philosophy, Social Sciences or Artificial Intelligence (AI). One reason for this is that the very purpose of using commonsense knowledge is not clear. However, we argue in this project that although ambiguity is always present in situations of human communication, it can be solved relatively easily by humans by using common sense knowledge for disambiguation. Therefore, we will also evaluate how common-sense ontologies implemented in knowledge bases can help solve the problem of ambiguity. To do so, we propose a) to classify the characterization of the common sense phenomenon in the literature of AI in order to know if everyone is dealing with the same problem or different problems; b) to classify the various schemes of Winograd trying to follow the previous classification; c) replicate two attempts to solve the problem that presented the best performance up to the present moment to try to explain the reasons for success - and failure; and d) to improve the tests based on our central hypothesis, the need to consider an interlocutor for the application of common-sense knowledge.

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

NERI, HUGO; COZMAN, FABIO. The role of experts in the public perception of risk of artificial intelligence. AI & SOCIETY, NOV 2019. (19/07665-4, 18/09681-4)

COZMAN, FABIO GAGLIARDI; MUNHOZ, HUGO NERI. Some thoughts on knowledge-enhanced machine learning. INTERNATIONAL JOURNAL OF APPROXIMATE REASONING, v. 136, p. 308-324, SEP 2021. (16/18841-0, 19/07665-4, 18/09681-4)

NERI, HUGO; COZMAN, FABIO. The role of experts in the public perception of risk of artificial intelligence. AI & SOCIETY, v. 35, n. 3, p. 11-pg., 2019-11-15. (18/09681-4, 19/07665-4)