The Stanford Dependency Relations for Commonsense Knowledge Representation of Winograd Schema Challenge (WSC)

Abstract

An alternative to the Turing Test that could offer a more accurate measurement of artificial intelligence is the Winograd Schema Challenge (WSC). It presents a number of coreference resolution issues that cannot be resolved without the use of human behavior reasoning. A certain type of Commonsense Knowledge (CSK) is necessary for Winograd schema. In order to handle the representation of Winograd appropriately, this research used a Deep-learning Stanford dependency parser as a natural language processing tool created by the Stanford NLP Group. The purpose of this tool is to use dependency grammar to represent sentences based on their grammatical analysis which helps understand the connections between words in a sentence such as which words rely on other words for meaning or grammar which is the task of dependency parsing. In addition, Extracting these dependency relations reflects commonsense knowledge representation for WSC. Then, we integrate common sense knowledge with the Syntactic ontology graphical representation by substituting synonyms for the main events in each sentence. To assess the entire system, we employed Precision and Recall as natural language performance evaluation metrics. Precision and recall measures for Root and advc1 dependency types are 0.94 and 0.92 respectively. Precision and recall measures for the nsubj dependency type are 0.96 and 0.94 respectively. Precision and recall measures for dobj, idobj, and pobj dependency types are 0.92 and 0.83.