Facebook is used by billions of people around the world in over one hundred languages. The wonderful linguistic diversity of people using Facebook presents challenges for Natural Language Processing (NLP) systems: annotating a separate data set for each language is a resource-intense process that does not scale over all problems we want to solve in all languages. We develop methods for Cross-Lingual Understanding (XLU) that allow us to learn NLP components in one language and apply them in other languages without the need of extra training data.
In this talk I will give an overview of the problems that we are addressing and the common methods for XLU. I will cover the Cross-Lingual Natural Language Inference (XNLI) benchmark that we created recently. I will then talk about recent progress that has pushed the state-of-the-art by using cross-lingual pretrained representations. I will finish by briefly discussing other ongoing work.
Ves Stoyanov is a Research Scientist Manager at Facebook AI focusing on Applied Research in Natural Language Processing (NLP). Previously he was a Research Scientist on the Search and the Applied Machine Learning teams at Facebook. Prior to Facebook, Ves was a postdoctoral researcher at Johns Hopkins University’s Center for Language and Speech Processing (CLSP), where he worked on Machine Learning for Structured Prediction supported by a Computing Innovation Fellowship from the CRA. Ves obtained his PhD from Cornell University under the supervision of Claire Cardie with thesis titled “Opinion Summarization: Automatically Creating Useful Representations Of The Opinions Expressed In Text.” During his PhD, Ves was supported by an NSF Graduate Research Fellowship.