In the last few years, there have been several breakthroughs concerning the methodologies used in Natural Language Processing (NLP). These breakthroughs originate from both new modeling frameworks as well as from improvements in the availability of computational and lexical resources.

In this seminar booklet, we are reviewing these frameworks starting with a methodology that can be seen as the beginning of modern NLP: Word Embeddings.

We will further discuss the integration of embeddings into end-to-end trainable approaches, namely convolutional and recurrent neural networks.

The second chapter of this booklet is going to cover the impact of Attention-based models, since they are the foundation of most of the recent state-of-the-art architectures. Consequently, we will also spend a large part of this chapter on the use of transfer learning approaches in modern NLP.

To cap it all of, the last chapter will be abour pre-training resources and benchmark tasks/data sets for evaluating state-of-the-art models followed by an illustrative use case on Natural Language Generation.

This book is the outcome of the seminar “Modern Approaches in Natural Language Processing” which took place in the summer term 2020 at the Department of Statistics, LMU Munich.

Creative Commons License

Creative Commons License

This book is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.