Techorama Belgium 2019

Language models are a crucial part of many Natural Language Processing pipelines. They are used in Speech-to-Text, Part-of-Speech tagging, Machine Translation, Optical Character Recognition, Intent Classification in chatbots etc. I will explain the overall idea behind language models, how they work and what the differences between them are. A special focus will lie on word embeddings. In the last few years, several novel approaches to word embeddings have been developed: word2vec, character embeddings such as FastText, GloVe embeddings, BERT and ELMO embeddings. Although the core idea behind these types of embeddings is the same, they each have their strengths and weaknesses, making them more or less appropriate for a certain use case. Through illustration with concrete examples and linguistic fun facts, you will gain insight in the (im)possibilities that come with language models, and learn how to pick the right one for your NLP problem.