Natural Language Processing

Every sentence of human language is ambiguous, context-dependent, and full of knowledge the words never state — and we want machines to read it, translate it, answer from it, and write it back.

The field begins with the basics: regular expressions to match patterns, tokenization to split text into units, and edit distance to measure how far apart two strings are. Then the first real model of language — the n-gram, which predicts the next word from the last few.

The modern turn is representation: instead of treating words as atomic symbols, embed them as vectors so that meaning becomes geometry — king minus man plus woman lands near queen — and let a neural network compose those vectors into the meaning of a whole sentence.

The transformer lets every word attend to every other, and scaled up on enough text it became the large language model — one architecture that translates, summarizes, answers, and converses.