GraphKR Semantic Parsing for Portuguese

Quem: 

Valeria de Paiva (Topos Institute)

Onde: 

Via Zoom

Quando: 

1 de Outubro de 2020, às 16h

Semantic parsing to construct graphical meaning representations is an active topic of research at the moment. Representation languages such as AMR, Alexa Meaning Language, Deep Universal Dependencies are some examples and these are getting more accurate as time goes by, and more annotations are accumulated.Traditional, old fashioned Natural Language Understanding uses logic and knowledge representation to build-up meaning representations, that are more useful, as one can reason with them. But these have a reputation for being brittle, hard to build and not easy to improve. We would like hybrid systems that keep the best of both worlds.

Three broad approaches have been attempted to combine distributional and structural/symbolic aspects of meaning representation construction: a) injecting linguistic features into distributional representations, b) injecting distributional features into symbolic representations or c) combining structural and distributional features in the final representation. This work focuses on an example of the third and less studied approach: it extends the Graphical Knowledge Representation (GKR) of Crouch and Kalouli to include distributional features and proposes a division of semantic labour between the distributional and structural/symbolic features. In this talk we discuss the possibilities and difficulties of extending this work from English to Portuguese.Inspired by Kalikow-type decompositions, we introduce a new stochastic model of infinite neuronal networks, for which we establish sharp oracle inequalities for Lasso methods and restricted eigenvalue properties for the associated Gram matrix with high probability. These results hold even if the network is only partially observed. The main argument relies on the fact that concentration inequalities can easily be derived whenever the transition probabilities of the underlying process admit a sparse space-time representation..

Registre-se com antecedência para este seminário:  https://ide-fgv-br.zoom.us/meeting/register/tJclceygrD4vGNQGsydh6omKYBi519NDspWy
Após o registro, você receberá um e-mail de confirmação contendo informações sobre conexão no seminário.

*Texto informado pelo autor. 

Palestrante: 

Valeria de Paiva is a mathematician and computer scientist based in Cupertino, CA. She's working on a 'stealth' start-up in Berkeley, the Topos Institute, while 'visiting' the Department of Informatica, PUC-Rio de Janeiro. She worked as a Principal Scientist at Samsung Research America, and as a senior applied scientist at Nuance Communications, Sunnyvale, CA. Earlier she was at Rearden Commerce and she was a search analyst at Cuil, Inc. in Menlo Park, CA. Before that, she was a research scientist at the Intelligent Systems Laboratory of PARC (Palo Alto Research Center), California (2000-2008). She received her Ph.D. in Mathematics from Cambridge University in 1988 for work on "Dialectica Categories", under Martin Hyland's supervision, and has ever since worked on logical approaches to computation, especially using Category Theory.