CourseNatural Language Processing: The Complete Course · Module 2: Tokenisation and Word Normalisation · part 15 of 73
Part 15 · Module 2: Tokenisation and Word Normalisation

Lesson 2: Subwords, how models read words they never saw

55 min read·14 Oct 2026
Module 2 · Lesson 2 · About 40 minutes · Needs: basic Python, and what a word token is (Lesson 1 of this module)

The situation

Monday 14 September 2026, project week 2. Dev forwards a test-split message from 25 August: "I got your email about battery batch VB-0426. My pack has that code and the battery runs flat very quiickly. What happens now?" A battery-batch fault: the kind he needs first.

A word list built from the training messages holds every word here except the typo "quiickly", so a word-level model sees an unknown. This lesson replaces words with subwords, pieces smaller than a word, so nothing is ever unknown. This message runs through every section.

The rest of this part opens with an account

Signing up is free and takes one email. Nothing to pay, and nothing renews.