Why Tokenization: Computers Need Numbers, Not Text

~11 min read

Neural networks only understand numbers, so text has to be chopped into pieces and mapped to numeric IDs before a model can touch it. That chopping is tokenization, and the piece size you choose creates a fundamental trade-off.

Why Tokenization: Computers Need Numbers, Not Text is a Pro topic

Sign in, then upgrade to Pro or Power to unlock this topic and the full AI Engineering curriculum.

Key points

  • Neural networks only compute on numbers, so text must be converted to a sequence of integer token IDs before any model can process it
  • Word-level tokenization needs a huge vocabulary (millions of words) and still fails on any word it's never seen (out-of-vocabulary/'unknown' tokens)
  • Character-level tokenization keeps the vocabulary tiny and has no unknowns, but produces very long sequences and harder-to-learn meaning
  • Subword tokenization is the practical middle ground: common words stay whole, rare words split into meaningful pieces — manageable vocabulary, reasonable length, handles unseen words
  • The choice of tokenization scheme directly trades off vocabulary size, sequence length, and how gracefully the model handles unfamiliar words