2026-08-28·by Sijie Wang#node#idea#math

chomsky-hierarchy

The Chomsky hierarchy — four tiers of automata

Parent: logic

Chomsky (1956), the linguist, classified formal grammars into four nested types; each matches a class of automaton (by how much memory it has) and a class of language. The Turing machine is the outermost tier.

TypeGrammarAutomatonMemoryLanguagese.g.
3regularfinite automaton (DFA/NFA)noneregulara*b*
2context-freepushdown automaton (one stack)a stackcontext-freeaⁿbⁿ, balanced parens, most syntax
1context-sensitivelinear-bounded automaton (tape ≤ input)bounded tapecontext-sensitiveaⁿbⁿcⁿ
0unrestrictedTuring machineunbounded taperecursively enumerableanything computable

Strict inclusions: regular ⊊ context-free ⊊ context-sensitive ⊊ recursively enumerable.

The ladder is really about memory

  • none → a finite automaton can only be in one of finitely many states; it can't count, so it fails aⁿbⁿ.
  • one stack → nesting/recursion (balanced brackets), but only last-in-first-out; fails aⁿbⁿcⁿ (needs two counts).
  • bounded tape → context-sensitive; still decidable (finite configurations).
  • unbounded tape → full computation. Turing (Type 0) is the top because unbounded memory = universal computation.

Where undecidability enters

The jump from Type 1 (decidable) to Type 0 (only semi-decidable) is exactly the phase transition: membership in a context-sensitive language is decidable, but membership in an r.e. language is only recognizable — deciding it is the halting problem (history). Unbounded memory buys universality and undecidability in one step.

Linguistics motivation

Chomsky's point was about natural language: finite-state (Type 3) grammars can't capture nested clause structure, so language needs at least context-free; and phenomena like cross-serial dependencies push it to mildly context-sensitive. The hierarchy was born to say "human syntax is not regular."

Full story of why a linguist built this → history.

Each tier (children)