Learning from Disagreements
Massimo Poesio (joint work with Alexandra Uma, Tommaso Fornaciari, Dirk Hovy, Silviu Paun, and Barbara Plank)
There is plenty of evidence from NLP and Computer Vision (CV) that humans disagree on many cognitive interpretation tasks, from the simplest tasks such as Part of Speech tagging to more subjective tasks such as classifying an image or deciding whether a proposition follows from certain premises. Although most AI work still relies on the assumption that a single interpretation exists, there has been extensive research in recent years on developing methods for learning from data that do not rely on such assumption. In this talk, I will review the evidence about disagreements and the main datasets that contain such information, discuss the best known approaches to build models from data containing them, and compare them on some of the best datasets to assess how their performance is affected by a dataset’s characteristics.
Massimo Poesio is a Professor in Computational Linguistics in the School of Electronic Engineering and Computer Science, Queen Mary University of London. He is a cognitive scientist with a particular focus on Computational Linguistics / Natural Language Processing. His research interests include computational models of anaphora resolution (coreference); disagreement in language interpretation; the creation
of large corpora of semantically annotated data (an area in which he pioneered the use of games-with-a-purpose with the development of Phrase Detectives, http://www.phrasedetectives.org); semantic interpretation of verbal and non-verbal communication in interaction; the study of conceptual knowledge using a combination of methods from CL and neuroscience; and the application of text analytics methods to real life problems, such as deception detection, or the identification of reports of human rights violations from social media.
For papers and info, see http://www.massimopoesio.org or http://www.dali-ambiguity.org
Events at the University
Browse upcoming public lectures, exhibitions, family events, concerts, shows and festivals across the University.