NLP Seminar: Natural Language Inference with Multilingual Supervision – University of Copenhagen

NLP Seminar: Natural Language Inference with Multilingual Supervision

Jörg Tiedemann, Full professor at the University of Helsinki, Finlandwill give a talk at University of Copenhagen as part of the NLP Seminar Series, funded by CCC. The talk will start at 17:00, there will be refreshments from 16:30.

Title

Natural Language Inference with Multilingual Supervision

Abstract

Natural Language Inference (NLI) is the task of identifying inferential relationships between a premise p and a given hypothesis h. Both, p and h are expressed in natural language, typically as a pair of sentences with a specific relationship between them coming from a limited set of inferential relations (e.g. entailment, contradiction, neutral). NLI requires semantic analyses and, therefore, can be seen as a test of text understanding capabilities of a system. Modern NLI models are based on deep neural nets and either cross-sentential encoding or independent sentence embeddings. In this talk, I will present our work on sentence representation learning and its appliction to common NLI benchmarks. I will start with the introduction of a state-of-the-art supervised NLI model with hierarchical bi-LSTM architectures and, after that, discuss our research in multilingual supervision for representation learning. The latter is motivated by the use of translations as semantic mirrors and the idea of applying highly multilingual data sets in neural machine translation to learn language-independent meaning representations.

Bio

Jörg Tiedemann is a professor of language technology at the Department of Modern Languages at the University of Helsinki. His main research interest is in cross-lingual NLP and machine translation.