Phonological rules capture the regular sound alternations that occur in natural languages. When English speakers pronounce "impossible" rather than "inpossible," they are applying a nasal assimilation rule that changes /n/ to [m] before a bilabial consonant. When Turkish speakers alternate between "ev-de" (at the house) and "okul-da" (at the school), they exhibit vowel harmony. These alternations are systematic and predictable, and phonological rules formalize them as context-sensitive transformations from underlying (stored) representations to surface (pronounced) forms.
Rule Formalism
"Segment A becomes B when preceded by C and followed by D"
Examples:
Nasal assimilation: n → m / _ [+labial]
Final devoicing: [+voice, -sonorant] → [-voice] / _ #
Vowel nasalization: V → [+nasal] / _ [+nasal]
Flapping: t → ɾ / V _ V (in American English)
Feature-based: [αF] → [βG] / [γH] _ [δK]
The Sound Pattern of English (SPE) notation, introduced by Chomsky and Halle (1968), represents phonological rules as rewrite operations over feature matrices. Segments are represented as bundles of binary distinctive features ([+voice], [-continuant], etc.), and rules specify changes to these features in phonological contexts. This notation captures natural classes — groups of segments that behave uniformly — through underspecification: a rule targeting [+voice, -sonorant] applies to all voiced obstruents regardless of their place or manner of articulation.
Rule Ordering and Interaction
A central issue in rule-based phonology is how rules interact when multiple rules apply to the same form. In feeding order, one rule creates the context for another: if epenthesis inserts a vowel, then a subsequent stress rule can assign stress to the new syllable. In bleeding order, one rule removes the context for another: if deletion removes a segment, a subsequent assimilation rule can no longer apply. Counterfeeding and counterbleeding orders produce opaque interactions where the surface form does not transparently reflect the rules that applied. Rule ordering was a major topic of debate in generative phonology, and its complex interactions motivated the development of Optimality Theory.
Kaplan and Kay (1994) proved the landmark result that any ordered sequence of phonological rewrite rules can be compiled into a single finite-state transducer, provided each rule replaces strings of bounded length. The compilation proceeds by translating each rule into a replace transducer, then composing all replace transducers in the specified order. The resulting composite transducer maps directly from underlying to surface representations in a single pass, achieving O(n) runtime regardless of the number and complexity of the original rules. This result established the regularity of phonological rule systems and enabled practical implementation in tools like Xerox xfst.
Phonological Rules in Speech Technology
Phonological rules play a central role in speech technology. In text-to-speech systems, rules model the pronunciation changes that occur in connected speech: assimilation across word boundaries, vowel reduction in unstressed syllables, and deletion of segments in fast speech. In speech recognition, inverse phonological rules (or their equivalents in pronunciation lexicons and acoustic models) map surface pronunciations back to underlying word identities. The Festival and MaryTTS speech synthesis systems use explicit rule components for phonological processing.
The formalization of phonological rules has also contributed to phonological theory itself. By making rules computationally explicit, linguists discovered that certain types of rules were never attested in natural languages — for example, rules that count to three or more — suggesting that the computational complexity of phonology is bounded below the full power of context-sensitive rewriting. This observation, formalized in the subregular hierarchy, provides a computational explanation for the restricted typology of phonological processes.