Speech and Language Processing (3rd Edition Draft)
by Dan Jurafsky and James H. Martin
The standard NLP textbook, free as a draft and rebuilt around large language models.
Overview
Jurafsky and Martin's Speech and Language Processing is the standard NLP textbook, and the third edition has been rewritten around large language models. The current free draft is dated January 6, 2026 and runs to 25 chapters across two volumes plus eleven web-only appendices. Volume I, Large Language Models, moves from Introduction, Words and Tokens, N-gram Language Models, Logistic Regression and Text Classification, Embeddings and Neural Networks into Large Language Models, Transformers, and Post-training: Instruction Tuning, Alignment, and Test-Time Compute, then Masked Language Models, Information Retrieval and Retrieval-Augmented Generation, Machine Translation, RNNs and LSTMs, and four speech chapters covering Phonetics and Speech Feature Extraction, Automatic Speech Recognition and Text-to-Speech. Volume II, Annotating Linguistic Structure, covers sequence labeling for parts of speech and named entities, constituency and dependency parsing, information extraction, semantic role labeling, sentiment and affect lexicons, coreference resolution and entity linking, discourse coherence, and conversation structure. Appendices A through K hold the material displaced by the LLM rewrite — hidden Markov models, naive Bayes, Kneser-Ney smoothing, CCG, logical semantics, WordNet and more. Recent revisions added the ASR and TTS chapters, restructured the LLM material, and introduced Direct Preference Optimization. Per-chapter PDFs and both PowerPoint and PDF slide decks are published alongside a combined PDF, and the authors explicitly invite classroom use.
At a Glance
- Topic
- ML
- Level
- All Levels
- Format
- Book
- Cost
- Free
- Duration
- 25 chapters plus 11 appendices — a semester-length read
- Provider
- Dan Jurafsky and James H. Martin
- Hands-on
- No
- Certificate
- None
What You’ll Learn
- ✓The full arc from n-gram models and embeddings to transformer language models
- ✓How transformers work at the level of detail needed to implement one
- ✓Post-training methods: instruction tuning, alignment, DPO and test-time compute
- ✓Information retrieval and retrieval-augmented generation explained from first principles
- ✓Speech pipeline fundamentals including feature extraction, ASR and text-to-speech
- ✓Classical structure tasks such as parsing, coreference and semantic role labeling
- ✓The statistical foundations preserved in the appendices, from HMMs to Kneser-Ney smoothing
Highlights
- •The reference textbook for the field, free as a draft and updated as of January 6, 2026
- •Rebuilt around LLMs rather than bolted onto a pre-2020 structure
- •Ships PowerPoint and PDF slide decks per chapter, so it doubles as a ready course kit
- •Legacy statistical material is preserved in eleven appendices instead of being deleted
- •Repeatedly front-paged on Hacker News across editions, most recently in December 2025
Who It’s For
Best For
- ✓Engineers who want the theory underneath the libraries they already use
- ✓Self-taught practitioners filling gaps in NLP and speech fundamentals
- ✓Instructors assembling an NLP or LLM course from open materials
Prerequisites
- •Undergraduate linear algebra, probability and calculus
- •Basic programming ability, ideally Python, to follow the algorithms
FAQ
What is Speech and Language Processing (3rd Edition Draft)?
Jurafsky and Martin's field-standard textbook, whose third-edition draft has been restructured around large language models. Twenty-five chapters across two volumes plus eleven appendices take you from tokenization and n-grams through transformers, post-training and RAG to speech recognition and linguistic structure. Free, with per-chapter PDFs and slide decks, and updated as recently as January 2026.
Is Speech and Language Processing (3rd Edition Draft) free?
Speech and Language Processing (3rd Edition Draft) is free to access.
What level is Speech and Language Processing (3rd Edition Draft) for?
Speech and Language Processing (3rd Edition Draft) is aimed at a all levels audience. Recommended background: Undergraduate linear algebra, probability and calculus, Basic programming ability, ideally Python, to follow the algorithms.
How long does Speech and Language Processing (3rd Edition Draft) take?
Expect roughly 25 chapters plus 11 appendices — a semester-length read. Most learners work through it at their own pace.
What will I learn from Speech and Language Processing (3rd Edition Draft)?
You'll learn: The full arc from n-gram models and embeddings to transformer language models; How transformers work at the level of detail needed to implement one; Post-training methods: instruction tuning, alignment, DPO and test-time compute; Information retrieval and retrieval-augmented generation explained from first principles; Speech pipeline fundamentals including feature extraction, ASR and text-to-speech; Classical structure tasks such as parsing, coreference and semantic role labeling; The statistical foundations preserved in the appendices, from HMMs to Kneser-Ney smoothing.
Topics
Sources
This page was written from 2 sources, 1 on domains other than web.stanford.edu.