A Marker Passing Approach to Winograd Schemas

Abstract

his paper approaches a solution of Winograd Schemas with a marker passing algorithm which operates on an automatically generated semantic graph. The semantic graph contains common sense facts from data sources form the semantic web like domain ontologies e.g. from Linked Open Data (LOD), WordNet, Wikidata, and ConceptNet. Out of those facts, a semantic decomposition algorithm selects relevant facts for the concepts used in the Winograd Schema and adds them to the semantic graph. Markers are propagated through the graph and used to identify an answer to the Winograd Schema. Depending on the encoded knowledge in the graph (connectionist view of world knowledge) and the information encoded on the marker (for symbolic reasoning) our approach selects the answers. With this selection, the marker passing approach is able to beat the state-of-the-art approach by about 12%.

@InProceedings{10.1007/978-3-030-04284-4_12,
author="F{"a}hndrich, Johannes
and Weber, Sabine
and Kanthak, Hannes",
editor="Ichise, Ryutaro
and Lecue, Freddy
and Kawamura, Takahiro
and Zhao, Dongyan
and Muggleton, Stephen
and Kozaki, Kouji",
title="A Marker Passing Approach to Winograd Schemas",
booktitle="Semantic Technology",
year="2018",
publisher="Springer International Publishing",
address="Cham",
pages="165--181",
abstract="This paper approaches a solution of Winograd Schemas with a marker passing algorithm which operates on an automatically generated semantic graph. The semantic graph contains common sense facts from data sources form the semantic web like domain ontologies e.g. from Linked Open Data (LOD), WordNet, Wikidata, and ConceptNet. Out of those facts, a semantic decomposition algorithm selects relevant facts for the concepts used in the Winograd Schema and adds them to the semantic graph. Markers are propagated through the graph and used to identify an answer to the Winograd Schema. Depending on the encoded knowledge in the graph (connectionist view of world knowledge) and the information encoded on the marker (for symbolic reasoning) our approach selects the answers. With this selection, the marker passing approach is able to beat the state-of-the-art approach by about 12{%}.",
isbn="978-3-030-04284-4"
}
Authors:
Johannes Fähndrich, Sabine Weber, Hannes Kanthak
Category:
Conference Paper
Year:
2018
Location:
Joint International Semantic Technology Conference
Link: