Reasoning about Conditional Beliefs for the Winograd Schema Challenge

Denis Golovin, Jens Claßen, Christoph Schwering
2017 International Symposium on Commonsense Reasoning  
The Winograd Schema Challenge has been proposed as an alternative to the Turing Test for measuring a machine's intelligence by letting it solve difficult pronoun resolution problems that cannot be tackled by statistical analysis alone. While many solutions so far are based on machine learning and natural language processing, we believe that a knowledgebased approach is better suited. In particular, we propose to employ a logic for conditional beliefs that is capable of dealing with incomplete
more » ... even inconsistent information (which commonsense knowledge often is). It does so by formalising the observation that humans often reason by picturing different contingencies of what the world could be like, and then choose to believe what is thought to be most plausible. We discuss and evaluate an implementation where relevant commonsense background information is obtained from the ConceptNet semantic network, translated into our formalism, and processed by a reasoner for our logic.
dblp:conf/commonsense/GolovinCS17 fatcat:wja7wctgrvdxjlrulbblxijtfa