Neural Models for Implicit Discourse Relation Recognition
Andre Cianflone (cianflone <at> encs (point) concordia (point) ca)
Wednesday 7 December 2016 at 11:30 AM
Salle 3195, Pavillon André-Aisenstadt
In order to understand a coherent text, humans infer semantic or logical relations between textual units that may or may not be explicitly stated. For example, in "I am hungry. I did not have lunch today." the reader infers a "causality" relation, even if it is not explicitly stated via a term such as "because". This linguistic mechanism is called "implicit discourse relations". The current state of the art demonstrates that automatically identifying implicit relations is much more difficult than for explicit relations. In this talk, we will present our work on the use of various traditional machine learning models and Convolutional Neural Networks with pretrained word embeddings to improve discourse relation recognition (DRR). Results at the recent CoNLL-2016 shared task show early promise with deep neural networks, but no added gain with pretrained embeddings. As recent research has also shown how correlation mechanism in neural networks can increase DRR, we will then introduce our recent focus on Encoder-Decoder type networks with attention mechanism for DRR.
See all the weekly talks for the year:1991 1992 1993 1994 1995 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021