Event Date: Thursday, 1 March, 2018, 11 a.m.
Location: Via Santa Maria, 36, Pisa, PI, Italia [2nd floor seminar room]
Speaker: Prof. Raffaella Bernardi (University of Trento)
Title: Grounded Textual Entailment
Abstract: A long standing challenge for computational semanticists has been to build models able to capture the semantic relations holding between sentences. This task has been at the core of logical models which are based on the logical entailment relation between propositions. Sentence A logically entails sentence B iff in all “worlds” where A is true also B is true — there cannot be a world in which B is false and A is true; whereas A contradicts B iff there are no worlds in which both A and B are true. A “world” is an interpretation, or in other words, a situation. In the era of statistical computational models, entailment has been defined in more probabilistic terms: a sentence A entails B if, typically, a human
reading A would infer that B is most likely true. After the pioneering shared tasks on Recognizing Textual Entailment (RTE), recently, rather large datasets, SICK and SNLI, have been released. Both datasets have been built taking Flickr, a dataset of images and caption descriptions, as starting point, but the sentences have been de-coupled from the images. We are interested in bringing back the idea of “world” proper of the logical approach, so to have models able to recognize whether two sentences are in an entailment or contradictory relation given a specific world. We take the world to be an image. By doing so we introduce the task on Grounded Textual Entailment which we believe can bring to the light strenght and limitation of current Language and Vision models, and may result into a more human friendly task than both logical and textual entailment. The work is carried out as a collaboration between the University of Trento and the University of Malta.
Raffaella Bernardi is Assistant Professor at DISI (Department of Information Engineering and Computer Science) and CIMeC (Center for Mind/Brain Science), University of Trento. From 2002 till 2010, she has been assistant professor with a temporary contract at Faculty of Computer Science, Free University of Bozen-Bolzano where she taught Computational Linguistics and acted as local coordinator of the Erasmus Mundus European Masters Programme in LCT. She studied at the Universities of Utrecht and Amsterdam specializing in Logic and Language, in 1999 she joined the international PhD Programme at the University of Utrecht and wrote a dissertation on categorial type logic (defended in June 2002). In 2011, she has started working on Distributional Semantics investigating its compositional properties and its integration with Computer Vision models. She has supervised PhD projects on controlled natural language to access ontology and databases and on Interactive Question Answering Systems on the Library Domain, and Multimodal Models. She is the local coordinator of the Erasmus Mundus European Masters Programme in LCT and of the Language and Multimodal Interaction track of the MSc in Cognitive Science offered by the University of Trento. She is the author of more than 50 publications in proceedings of international workshops, conferences and journals. She has been the PI within the EU Project “CACAO” (CP 2006 DILI 510035 CACAO Program: eContentplus), she has been part of the team which won the ERC 2011 Starting Independent Research Grant COMPOSES (project nr. 283554), and she is member of the Management Board of the Cost Action The European Network on Integrating Vision and Language.