December 15, 2025
Old school vs. bot school
Speech and Language Processing (3rd ed. draft)
Classic AI textbook gets a 2025 glow‑up—nostalgia vs. “NLP is dead”
TLDR: The landmark NLP textbook released a 2025 draft updated for the LLM/chatbot era, moving old methods to appendices and putting transformers up front. Comments clash between grounded nostalgia and “NLP is dead” doom, with jokes about chapters aging faster than AI’s relentless pace.
Jurafsky & Martin just dropped the Aug 24, 2025 draft of their classic language tech textbook, and the comments instantly turned into a reunion party meets flame war. The book leans into the chatbot era: large language models (LLMs) arrive early (ch. 7), transformers get the spotlight (ch. 8), old-school stuff like Naive Bayes and PPMI gets pushed to the appendix, and speech/audio gets a makeover with Whisper for speech-to-text and VALL‑E/EnCodec for text-to-speech. There’s even DPO (a way to align AI with human preferences) in ch. 9, and the chatbots chapter is split across LLMs and a new Conversation chapter. The vibe? A mix of “finally modern” and “RIP the old guard.”
Nostalgics like brandonb remember when everything was hidden Markov models and state machines—now “the whole stack” is neural. MarkusQ pushes back on the “antiquated” label, calling the book grounded, a necessary map of the roads that led to today’s AI. mfalcon drops the meme of the day: chapters get written, then LLMs sprint ahead before they’re published. Meanwhile, ivape detonates a doom-bomb claiming NLP is dead and jobs are evaporating. Fans chime in with wholesome throwbacks—Jurafsky meetups, the legendary “Dice Book”—and everyone giggles at the authors’ warning: “When will the book be finished? Don’t ask.” Grab the draft here and join the comment cage match.
Key Points
- •An August 24, 2025 online draft of the 3rd edition of Speech and Language Processing is released with a full PDF and updated slides.
- •Chapter 9 adds LLM posttraining preference alignment using Direct Preference Optimization (DPO).
- •New ASR (Whisper) content in Chapter 15 and TTS (EnCodec, VALL-E) in Chapter 16 are introduced.
- •Classification is now taught via Logistic Regression, with Naive Bayes and PPMI moved to the appendix; tf-idf is consolidated in Chapter 11.
- •LLMs are introduced in Chapter 7, Transformer internals in Chapter 8, RNN/LSTM moved to Chapter 13; dialogue content is redistributed, including a new Chapter 25 on conversation.