Nell’ambito dei Seminari del CoLingLab (Laboratorio di Linguistica Computazionale http://colinglab.fileli.unipi.it/), mercoledì 20 settembre alle ore 11 presso la Sala Riunioni di Pal. Venera (Via S. Maria 36, secondo piano), la Prof. Vera Demberg (Università della Saarland) terrà un seminario dal titolo:

Thematic fit modelling with neural networks

ABSTRACT: Being able to generate thematic fit estimates on a large scale can help tasks like inferring missing or dislocated arguments, predicting upcoming words or inferring common sense knowledge. We have in recent years been working on a vector space model building on the Distributional Memory (Baroni & Lenci 2010) as well as a neural network model addressing that same task. I will describe the original neural network model (Tilk, Demberg, Sayeed, Klakow and Thater, 2016), as well as newer extensions to it which attempt to overcome problems with role swapping (e.g., apple eats boy) or the lack of data for the most common sense objects (e.g., eating with knife and fork).

VERA DEMBERG is professor for Computer Science and Computational Linguistics at Saarland University, Saarbrücken, Germany. From 2010 till 2016, she held a position as an independent research group leader at the Cluster of Excellence “Multimodal Computing and Interaction”, Saarland University. She received her PhD in 2010 from the University of Edinburgh, on the topic of building a broad-coverage model of syntactic processing in humans. Her research interests include psycholinguistic experimental research and computational modelling on human sentence processing at the levels of syntax, thematic role assignment, event cognition and coherence relations in discourse. In particular, her research has focussed on incremental processing and prediction.