Speech and Language Processing, 3rd Edition

What this book is

Speech and Language Processing by Dan Jurafsky and James H. Martin is the textbook most NLP courses assign. The third edition is written in public: the authors post updated draft chapters on the book's site, free to read, and the latest draft on the site at the time of writing is dated 19 August 2026. The subtitle now reads An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition with Language Models.

What it covers

The draft is organised in three volumes. The first covers words and tokens, n-gram language models, logistic regression and text classification, embeddings, neural networks, transformers and pretraining, and post-training. The second covers masked language models, interpretability, information retrieval and RAG, machine translation, RNNs and LSTMs, and speech: phonetics, speech recognition and text-to-speech. The third covers sequence labelling, parsing, information extraction, semantic role labelling, sentiment, coreference, discourse and conversation.

Is it worth your time?

Yes, as the deeper reference behind practical NLP work. It explains each idea clearly with worked examples, then connects it to the modern approach. When you need to know what TF-IDF really computes, how named entity recognition is framed, why tokenisation matters or how retrieval systems are evaluated, the relevant chapter answers it properly.

How to use it

Do not read it front to back. Read the chapters on words and tokens, text classification, embeddings and transformers early, then pick up sequence labelling, information extraction or retrieval when your project needs them. Each chapter ends with historical notes that point to the key papers Note which draft date you read, because chapters are revised between releases.

Prerequisites

Programming experience and some probability and linear algebra. The early chapters are accessible to undergraduates.

Free and paid

The draft chapters are free on the authors' site. The authors are clear that the book is still in progress.

Common questions

Is a draft reliable? Yes. It is written by two of the field's best-known teachers and revised regularly, which also keeps it more current than most printed textbooks.

Best for: learners who want a clear, thorough reference for classic and modern NLP.