Using Large Language Models in Political Science

Hands-on sandbox exercises for learning Large Language Models (LLMs), organized by the chapters of Jurafsky & Martin's Speech and Language Processing (3rd ed.) — the draft is free and openly published online. Each exercise pairs a short theory recap with a runnable coding task, oriented toward political-science text (speeches, parliamentary debates, reviews, news, social media).

This repo is the practice companion to the LLM concept notes in Notion: every concept page there has a matching exercise here.

How this repo is organized

Folders are numbered to match the J&M chapters (3rd ed., latest draft). Each chapter folder contains one subfolder per exercise, and each exercise has:

• README.md — the task spec (objective, dataset, tasks, success criteria, reference)
• *_sandbox.py — a starter script with TODOs to fill in

Chapter map

Folder	J&M Chapter	Exercises
02_tokenization/	Ch. 2	Tokens · Training Data & Corpora
03_ngram_language_models/	Ch. 3	N-gram Language Models · Perplexity & LM Evaluation
04_naive_bayes_sentiment/	Ch. 4	VADER vs. Star Ratings · Classification Metrics
05_logistic_regression/	Ch. 5	Logistic Regression & Text Classification · Bag of Words & TF-IDF
06_vector_semantics/	Ch. 6	Embeddings · How Embeddings Get Trained
07_neural_networks/	Ch. 7	Neural Networks
08_rnns_lstms/	Ch. 8	RNNs & LSTMs
09_transformers/	Ch. 9	Transformers · Attention
10_large_language_models/	Ch. 10	Pretraining vs. Fine-tuning · Decoding & Sampling · Context Window · Scaling Laws
11_masked_language_models/	Ch. 11	Masked Language Models (BERT)
12_alignment_prompting/	Ch. 12	Prompting & In-Context Learning · RLHF & Alignment · Bias & Fairness · Agents & Tool Use
13_machine_translation/	Ch. 13	Machine Translation & BLEU
14_ir_qa_rag/	Ch. 14	Information Retrieval · RAG Pipeline · Hallucination & Validation
capstone_text_annotation/	Capstone	LLMs for Text Annotation & Classification · Inter-rater Reliability

Note on numbering: this repo follows the latest J&M 3rd-ed draft, where Ch. 8 = RNNs & LSTMs and Ch. 9 = Transformers. Some Notion course reading pages still use an older numbering (e.g. Transformers as Ch. 8); those should be reconciled to match this map. See issue #1.

Setup

Requires Python 3.11 (see .python-version). Dependencies are pinned in requirements.txt for reproducibility.

python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt

Datasets go in data/ (git-ignored except its README). A tiny runnable sample lives in data/sample/ so the early exercises work out-of-the-box before you supply real data.

Suggested workflow

1. Read the concept note in Notion.
2. Open the exercise README.md here.
3. Fill in the TODOs in the *_sandbox.py stub.
4. Commit your solution next to the stub (e.g. tokens_solution.py).

Reference

• Daniel Jurafsky & James H. Martin, Speech and Language Processing (3rd ed. draft) — free online: https://web.stanford.edu/~jurafsky/slp3/

License

MIT — see LICENSE.