What contribute to conceptNet55 (21 million edges and over 8 million nodes) Facts acquired from Open Mind Common Sense(oMcS) Singh 2002 )and sister projects in other languages( Anacleto et al. 2006) Information extracted from parsing Wiktionary, in multiple languages, with a custom parser (Wikiparsec Games with a purpose"designed to collect common knowledge von Ahn, Kedia, and blum 2006) (Nakahara and Yamada 2011) Kuo et al. 2009) Open Multilingual WordNet(Bond and Foster 2013), a linked-data representation of WordNet (Miller et al. 1998 and its parallel projects in multiple languages IMDict(Breen 2004), a Japanese-multilingual dictionary OpenCyc, a hierarchy of hypernyms provided by cyc Lenat and Guha 1989 ), a system that represents commonsense knowledge in predicate logic A subset of DBPedia auer et al. 2007 ) a network of facts extracted from Wikipedia infoboxes Most of them are entity-centric knowledge, there are only 74, 989 nodes among 116,097 edges about events Speer, Chin, and Havasi, ConceptNet 5.5: An Open Multilingual Graph of General Knowledge. AAAI 2017
What contribute to ConceptNet5.5 (21 million edges and over 8 million nodes)? • Facts acquired from Open Mind Common Sense (OMCS) (Singh 2002) and sister projects in other languages (Anacleto et al. 2006) • Information extracted from parsing Wiktionary, in multiple languages, with a custom parser (“Wikiparsec”) • “Games with a purpose” designed to collect common knowledge (von Ahn, Kedia, and Blum 2006) (Nakahara and Yamada 2011) (Kuo et al. 2009) • Open Multilingual WordNet (Bond and Foster 2013), a linked-data representation ofWordNet (Miller et al. 1998) and its parallel projects in multiple languages • JMDict (Breen 2004), a Japanese-multilingual dictionary • OpenCyc, a hierarchy of hypernyms provided by Cyc (Lenat and Guha 1989), a system that represents commonsense knowledge in predicate logic • A subset of DBPedia (Auer et al. 2007), a network of facts extracted from Wikipedia infoboxes Speer, Chin, and Havasi, ConceptNet 5.5: An Open Multilingual Graph of General Knowledge. AAAI 2017. Most of them are entity-centric knowledge, there are only 74,989 nodes among 116,097 edges about events 11
Nowadays Many large-scale knowledge graphs about entities and their attributes (property-of)and relations thousands of different predicates have been developed Millions of entities and concepts Billions of relationships FReebase 9 y babeNe NELL WIKIPEDIA The Free Encyclopedia yeGo ProBase The Knowledge Graph Google Knowledge Graph(2012) 570 million entities and 18 billion facts
Nowadays, • Many large-scale knowledge graphs about entities and their attributes (property-of) and relations (thousands of different predicates) have been developed • Millions of entities and concepts • Billions of relationships NELL Google Knowledge Graph (2012) 570 million entities and 18 billion facts 12
However Semantic meaning in our language can be described as ' a finite set of mental primitives and a finite set of principles of mental combination Jackendoff, 1990) The primitive units of semantic meanings include · Thing( (or object, Activity, e State How to collect more · Event e knowledge rather than Place, entities and relations · Property. · Amount, etc Jackendoff, R. Ed. ) .(1990). Semantic Structures. Cambridge, Massachusetts: MIT Press
However, • Semantic meaning in our language can be described as ‘a finite set of mental primitives and a finite set of principles of mental combination (Jackendoff, 1990)’. • The primitive units of semantic meanings include • Thing (or Object), • Activity, • State, • Event, • Place, • Path, • Property, • Amount, • etc. Jackendoff, R. (Ed.). (1990). Semantic Structures. Cambridge, Massachusetts: MIT Press. How to collect more knowledge rather than entities and relations? 13
Outline Motivation: NLP and commonsense knowledge Consideration: selectional preference New proposal: large-scale and higher-order selectional preference e Applications
Outline • Motivation: NLP and commonsense knowledge • Consideration: selectional preference • New proposal: large-scale and higher-order selectional preference • Applications 14
Semantic primitive units Entities or concepts can be nouns or noun phrases Concepts in probase(2012) ° Company · T company, big company big IT company Hierarchy is partially based on headmodifier composition Let's think about verbs and verb phrases How should we define semantic primitive unit for verbs? Wentao Wu, Hongsong Li, Haixun Wang, Kenny Q Zhu Probase: a probabilistic taxonomy for text understanding. SIGMOD, 2012 nowMicrosoftconceptgraphhttpsconceptresearchmicrosoftcom
Semantic Primitive Units • Entities or concepts can be nouns or noun phrases • Concepts in Probase (2012): • Company, • IT company, • big company, • big IT company, • … • Hierarchy is partially based on head+modifier composition • Let’s think about verbs and verb phrases • How should we define semantic primitive unit for verbs? Wentao Wu, Hongsong Li, Haixun Wang, Kenny Q Zhu. Probase: A probabilistic taxonomy for text understanding. SIGMOD, 2012. (now Microsoft concept graph https://concept.research.microsoft.com/) 15