Hi all,

Reminder - this is happening today!

Dima

On Tue, 6 Jun 2017 at 08:42 Dzmitry Bahdanau <dimabgv@gmail.com> wrote:

Hi all,

Our next tea-talk will be given by Jackie C. K. Cheung, who is an assistant professor at McGill university. It will take place at Z-209 (Pavillon Claire-McNicoll), on June 13, at 14:00. Hope you to see you in great numbers!

Please note the unusual day and time of the talk!

Title: Leveraging External Knowledge for Machine Comprehension of Rare Entities

Abstract: Thanks to the "long tail" of natural language, entities that only appear a handful of times in a training corpus account for the vast majority of entities overall. Standard techniques for training word embeddings such as word2vec and GloVe do not scale well to modelling these entities, because they depend on having many entity or word occurrences for reliable word co-occurrence statistics. We propose to use external knowledge in the form of short entity definitions as a source of information to train entity representations. We describe experiments in two settings. In the first, we present a simple scheme for initializing entity representations for knowledge base completion. In the second, we introduce a new machine comprehension dataset, the Wikilinks Rare Entities data set, and an accompanying entity cloze task which requires predicting rare named entities in context. We describe a hierarchical double encoder model which reads a text passage and an entity definition in order to predict which entity fits in context. Our model substantially outperforms a number of strong baselines which do not utilize entity definitions, demonstrating the impact of external knowledge in making inferences about rare entities.

Bio: Jackie CK Cheung is an assistant professor in the School of Computer Science at McGill University. He received his Ph.D. at the University of Toronto, and was awarded a Facebook Fellowship for his doctoral research. He conducts research in computational semantics and natural language generation, with a focus on topics such as inducing event structure from distributional semantics, and automatic summarization. His work is funded by the Natural Sciences and Engineering Research Council of Canada (NSERC), and the Fonds de recherche du Québec - Nature et technologies (FRQNT).

Dima