Distributional Modeling of Entities: The Case for Precision

Entities (“Mozart”) are ontologically distinct from concepts (“composer”). Distributional methods are very good for capturing the fuzzy, graded meaning of concepts (Italy is more similar to Spain than to Germany), but comparatively little attention has been paid to entities, which presumably call for a more precise representation.
Sebastian Pado will report on two studies where we use distributional models to represent (named) entities, with a focus on improving our understanding of their strengths and problems. In the first study, they target fine-grained attributes of the kind typically found in structured knowledge bases (Italy has 60 million inhabitants). In the second study, they contrast the hypernymy relation between two concepts (“physicist — scientist”) with the instantiation relation between an entity and a concept (“Mozart — composer”). They conclude that entities are in principle amenable to the same distributional techniques applied to concepts, but do not always behave in parallel.

Sebastian Pado is professor of computational linguistics at Stuttgart University, Germany. He studied in Saarbrücken and Edinburgh, receiving his MSc in 2002 (cognitive science) and his PhD in 2007 (computational linguistics). After a postdoctoral position at Stanford, he become professor of computational linguistics at Heidelberg University in 2010 and in Stuttgart in 2013.
His core research concerns methods to learn, represent, and process aspects of natural language meaning from and in text. Examples include distributional models of linguistic concepts (e.g.
compositionality, polysemy), multilingual processing, discourse and dialogue structure (politeness, narrative text vs. reported speech),and inference over textual data.