Dear all,
You are invited to attend this semester's first Cognitive Area Seminar talk on *Friday Jan 15, 3:30 - 5 PM (STBIO, Room S3/4*).
The talk will be given by *Professor Jackie C.K. Cheung (McGill Computer Science, Reasoning and Learning Lab)*, and is titled "Towards Large-Scale Natural Language Inference with Distributional Semantics". For a full abstract, see below dashed line.
If you are unable to attend this week's talk, there will be more opportunities. Please see our full talk schedule @: http://www.mcgill.ca/psychology/events-colloquia-0/brownbag-series. We have a line-up of speakers from McGill and beyond, whose work covers a wide range of topics in psychology and cognitive neuroscience, from conversational speech timing to neural correlates of human rhythm perception.
Looking forward to an exciting semester!
Best, Anna
*Apologies for cross-posting*
---------------------------------------
*Towards Large-Scale Natural Language Inference with Distributional Semantics*
*Jackie Chi Kit Cheung*, Assistant Professor
*Reasoning and Learning Laboratory*
*Department of Computer Science*
*McGill University*
Language understanding and semantic inference are crucial components of complex natural language applications, from intelligent personal assistants to automatic summarization systems. Current systems often require hand-coded information about a domain of interest (e.g., medical records, current events, celebrity gossip), as well as the semantic properties of linguistic expressions within that domain. In this talk, I demonstrate that the distributional behavior of linguistic expressions within a corpus can characterize many aspects of their meaning. I present a method that integrates distributional-semantic information into a probabilistic content model, in order to derive a structured representation of the important events and participants in a domain. The model achieves state-of-the-art performance on a structure learning task and on multi-document summarization for systems that do not rely on hand-coded domain knowledge. I also show that distributional information can be used to discover a class of semantic operators known as downward-entailing operators, of which the most prominent member is the standard negation marker, "not". Knowledge of these operators is crucial for correctly reasoning about entailment relations in text. These results elucidate both the utility of distributional semantics for current natural language applications, and their potential to form the basis of a fully fledged theory of semantics.