Invitation to seminar - Victor Raskin & Julia Taylor Rayz, Workshop on Ontological and Lexical Acquisition

Venue: BARK/ABC-iRob, Einstein terem (1032 Budapest., Kiscelli utca 82.)

 

Short bio of Victor Raskin

Victor Raskin (Ph.D., Moscow State University, 1970) is Distinguished Professor of English and Linguistics at Purdue University, with courtesy appointments in Computer Science and in Computer and Information Technology. He has published widely in semantic theory and applications, from machine translation to information security, computational humor, biomedical domains, and robotic intelligence. He has taught at his alma mater and then at The Hebrew University of Jerusalem, Tel Aviv University (part-time), University of Michigan (sabbatical), and, since 1978, at Purdue. He has delivered numerous keynote addresses and invited lectures at international conferences and organized several of those. He has authored several books, including Semantic Mechanisms of Humor (D. Reidel, 1985) and Ontological Semantics (MIT Press, 2004, with S. Nirenburg).

 

Short bio of Julia M. Taylor Rayz

Julia Taylor Rayz (Ph.D., University of Cincinnati, 2008) is an Associate Professor of Computer and Information Technology at Purdue University. She has published intensely in computational humor, ontological semantic technology, fuzzy logic, and a wide range of semantic applications. Her experience in the industry includes developing an ontological semantic system for text analytics. She has delivered a number of keynote addresses, plenary sessions, invited lectures, tutorials and workshops at international conferences and hosted a few workshops, sessions, and schools.

 

Workshop on Ontological and Lexical Acquisition

The workshop will focus on the technology for hybrid human-computer acquisition of comprehensive semantic ontology and lexicon for natural language whose ultimate goal is to reflect adequately human understanding of natural language. The acquisition resource, http://engineering.purdue.edu/~ost, has been developed within Purdue’s Ontological Semantic Technology (OST) project for a family of computational processing applications, ranging from information security to health domains to robotics. The semantic ontology represents the human knowledge of the world. It is a vast lattice of concepts, representing events and objects, that are connected to each other with properties. Each concept will have a number of properties connecting it to other concepts. Thus, each even will have the properties of agent, theme, instrument, beneficiary, etc. One property is-a denotes subsumption, the only property that most published ontologies have. The OST ontology is language-independent but marked with English labels for the benefit of human acquirers. Thus, the event go is subsumed by motion-event; it has animate as its agent, and is also connected, as most events, to time and location. The OST lexicons are language-specific, and every sense of every word or phrasal in a natural language is anchored in an ontological concept with the properties restricted to correctly reflect its meaning. Thus, the English drive will be anchored in the concept go with the agent that is adult and human and the instrument automobile. Ontologies may vary in grain size depending on convenience and application: thus, drive can be introduced as a concept subsumed by go or represented by it. The ONT Acquisition Resource provides the format for both ontological and lexical acquisition, with all the ranges for properties explicitly expressed, and the minimally trained human acquirer, often an undergraduate, provides only specific intuitive answers. With some experience, acquirers produce 6-8 concepts or 8-10 lexical senses per hour, and we consider 10,000 concepts and 50,000-100,000 lexical senses to be sufficient for any application. The ontology and lexicons are, of course, usable within any approach and easily adjustable for any application. We believe that their approximation to human knowledge of the world and understanding of their natural languages is much closer than most other, nonrepresentational, ontologies and lexicons.

Kulcsszavak: