Real-World Natural Language Processing

Hagiwara's guide is for people who want to build working software that handles human language, without first mastering linguistic theory or the calculus behind deep learning. It concentrates on the methods needed to assemble practical tools such as chatbots, language identification systems, and text classifiers. Readers learn to design, build, and release these programs, and they work through named entity tagging, machine translation, and text generation along the way.

The book also tackles more advanced material, including attention mechanisms and transfer learning. Its examples are written in Python and rely on open source libraries, namely AllenNLP and Hugging Face Transformers. Projects include a sentiment analyzer and a conversational bot. The author points out that the field changes quickly, with new tools appearing constantly, and that the same techniques power translation services, voice assistants, and automatic writing.

The content falls into three parts. The first covers foundations: an overview of the field, a first working application, word and document embeddings, sentence classification, and sequence labeling with language modeling. The second introduces advanced architectures: sequence-to-sequence models, convolutional neural networks, attention and the Transformer, and transfer learning with pretrained language models. The final part addresses building robust applications, covering development best practices and the steps for deploying and serving an NLP system.

The intended reader is a Python programmer. No earlier knowledge of machine learning is assumed. The author earned a computer science PhD from Nagoya University in 2009, has interned at Google and Microsoft Research, and served as a Senior Machine Learning Engineer at Duolingo. He now leads his own research and consulting business.