Purpose
These notes cover the basic pipeline for turning text into something a model can work with. Start with tokenization, then move to classification and multinomial logistic regression. Neural networks shows what changes once linear decision boundaries stop being enough.
The section also holds practical notes that sit closer to modern language models than to textbook NLP. Prompting is the bridge between the older modeling material and the newer model-facing workflow, and decoding strategies covers how a trained model is turned into text at inference time.
Suggested path
- Represent and retrieve text: tokenization, PPMI, word embeddings, information retrieval and the PPMI experiment
- Build classifiers: classification, multinomial logistic regression
- Add nonlinear models: neural networks
- Use current models: prompting, decoding strategies