Machine Translation and Interlingua: are we there yet? / Fine-Grained Information Extraction from User' Comments with Semantic Context Paths
Vassilina Nikoulina & Caroline Brun (Naver LABS)
Machine Translation and Interlingua: are we there yet? - Vassilina Nikoulina (Naver LABS)
Abstract
In this talk we address the problem of Machine translation and how it has been approached through years of research. We will then talk about recent trends in Machine translation and in particular Multilingual Machine translation. We present the analysis of these models that have been done (including ours) trying to understand to what extent we have reached the “interlingua” representations.
Bio
Dr. Vassilina Nikoulina is a Researcher in Naver LABS Europe. She has been working on the problem of Machine Translation since 2003. More recently she started to be interesting in broader problems related to multilinguality, pretrained Language models, and analysis of what kind of knowledge is encoded in these Language models.
Fine-Grained Information Extraction from User' Comments with Semantic Context Paths - Caroline Brun (Naver Labs)
Abstract
Supervised methods provide the best performances for Information Extraction (IE), but require labeled data that is costly to create.
We introduce a new IE task, Semantic Context Path (SCP) tagging, that extracts richer information than simple named entity recognition, while allowing for easier creation of training data than for relation extraction.
In this presentation, we (1) define the SCP tagging task, (2) describe an efficient method to create training data for the task,
and (3) design and evaluate baselines extraction models for SCP tagging on English and Korean.
These models are integrated into a dedicated map interface allowing semantic navigation through user' comments.
Bio
Dr. Caroline Brun is a Senior Scientist at Naver Labs Europe. She has been working on a wide range of NLP themes, including word sense disambiguation, robust parsing and named entity recognition. Her current research interests include
aspect-based sentiment analysis and fine-grained information extraction.