NLP — Overview

LLMs & GenAI starts from Transformers and tokenization already in hand. This section is the layer underneath: how text was understood before (and alongside) large language models, and the task taxonomy — translation, summarization, QA, sentiment, NER — that Transformers now dominate but didn't invent.

Understanding this layer matters for a concrete reason, not just history: a lot of "why does the LLM tokenizer split words this way," "why do NER/parsing still show up as production pipeline stages even in an LLM-heavy stack," and "what does BLEU actually measure" only make sense with the classical grounding this section provides.

What's in this section

• Classical NLP: Tokenization, POS Tagging, NER & Parsing — text normalization, part-of-speech tagging, named entity recognition, and syntactic parsing (dependency and constituency).
• Word Embeddings — word2vec (CBOW and skip-gram), GloVe, fastText, and why contextual embeddings (BERT-style, see Attention & Transformers) superseded these static ones.
• NLP Task Taxonomy — sequence labeling, text classification, sentiment analysis, machine translation, summarization, and question answering, with the classical approach to each contrasted against the modern Transformer-based one.

See the roadmap for the full ordered path.