DIKU Bits: What do multilingual language models learn about language?

On 13 September the Natural Language and Processing (NLP) Section at the Department of Computer Science, University of Copenhagen, will give a DIKU Bits lecture.

Click here to view the slideshow from Karolina's DIKU Bits presentation.

Speaker 

Karolina Ewa Stanczak, PhD Student in the NLP Section at Department of Computer Science, University of Copenhagen.

Title

What do multilingual language models learn about language?

Abstract

Despite the success of multilingual language models on a number of downstream tasks such as machine translation and language generation, it remains unclear what multilingual language models learn about language. In this talk, I will discuss current research addressing this question including my own results indicating that multilingual pretrained models can learn language-universal representations of grammar. 

Which courses do you teach?: 

Natural Language Processing (NLP), Fair and Transparent Machine Learning Methods (FTML)

Which technology/research/projects/startup are you excited to see the evolution of? 

Multilingual NLP, bias and fairness in NLP