51.02.01 · linguistics / phonology-sound-systems

Phonology — phonemes, features, and sound systems

shipped3 tiersLean: none

Anchor (Master): Chomsky & Halle The Sound Pattern of English (Harper & Row 1968); Prince & Smolensky Optimality Theory (Blackwell 2004); Trubetzkoy Grundzüge der Phonologie (1939)

Intuition Beginner

Phonology is the study of how a language organises speech sounds into a system of contrast. Phonetics 51.01.01 described sounds as physical events in the mouth and the air. Phonology asks a different question: which sound differences change meaning, and which are merely automatic variation the speaker never notices?

Two sounds may be physically different yet count as "the same sound" to speakers. English speakers hear the "p" in pin and the "p" in spin as identical. They are not. The first is released with a strong puff of air, written [pʰ]; the second is not, written [p]. The difference is real and measurable on a microphone, but it does not change meaning in English.

A phoneme is one of these meaningful sound categories of a language. A phone is any single physical speech sound. The work of phonology is to find the phonemes of a language, work out how they may combine to form words, and describe the rules that turn abstract phonemes into the actual phones a speaker produces.

The same physical difference can be contrastive in one language and automatic in another. That is why phonology is a property of languages, not of sound itself. The boundary between "two sounds" and "two versions of one sound" is drawn differently in different languages, and the remainder of this unit is the study of how that boundary is found and described.

Visual Beginner

A schematic of the phoneme-to-phone mapping: one abstract phoneme /p/ splits into the allophones [pʰ] and [p] depending on context.

The same idea as a table of English /p/ allophones:

Phoneme	Environment	Allophone	Example
/p/	start of a stressed syllable	[pʰ]	pin, pot
/p/	directly after /s/	[p]	spin, spot
/p/	after another consonant	[p]	happen

One category, three contexts, three phones. Speakers learn the contexts unconsciously; phonology makes the pattern explicit.

Worked example Beginner

English: [pʰ] and [p] are allophones. Say pin while holding your hand in front of your mouth. You feel a puff of air on the release of the "p". Now say spin. The puff is gone. The two sounds are [pʰ] and [p].

Are they separate phonemes in English? Apply the minimal-pair test: search for two words that differ only in this one sound and have different meanings. There is no pair pin versus spin-without-the-s that is distinguished solely by [pʰ] versus [p]. Wherever English has [pʰ] (start of a stressed syllable), it never has [p], and wherever it has [p] (after /s/), it never has [pʰ]. The two phones are in complementary distribution: they never occupy the same slot. English speakers therefore treat them as one phoneme /p/.

Hindi: [pʰ] and [p] are separate phonemes. In Hindi the same two physical sounds do contrast meaning. The word [pal] (with unaspirated [p]) means "moment", while [pʰal] (with aspirated [pʰ]) means "fruit blade" or "knife edge". This is a minimal pair: two words identical except for the one sound in question, with different meanings. Because such a pair exists, [p] and [pʰ] must be assigned to distinct phonemes /p/ and /pʰ/ in Hindi.

What this tells us: the physical sounds are the same in both languages; the phonological analysis is different. Phonology is the layer of structure that decides which physical differences count inside a given language.

Check your understanding Beginner

Formal definition Intermediate+

Phonology studies the sound system of a particular language: the inventory of contrastive categories, the constraints on their combination, and the mapping from abstract representation to surface phone ^[Hayes]. The central objects are the following.

A phone is a single physical speech sound, transcribed in square brackets: [p], [pʰ], [a]. A phoneme is a psychologically real, contrastive sound category of a particular language, transcribed between slashes: /p/, /a/. An allophone of a phoneme is a context-determined surface realisation of it. So English /p/ is realised as the allophones [pʰ] word-initially and [p] after /s/.

Let $Σ$ be the set of phonemes of a language. A phonotactic constraint is a restriction on the strings over $Σ$ permitted as possible words. The set of attested phoneme sequences in syllables is described by a template, e.g. $Onset Nucleus Coda$ with each position filled from a language-particular subset of $Σ$ .

A distinctive feature is a binary attribute $[\pm F]$ used to classify phonemes. Each phoneme is represented as a bundle of feature values, written as a feature matrix. For the consonant inventory the SPE feature set includes at least $[\pm voice]$ , $[\pm nasal]$ , $[\pm continuant]$ , $[\pm anterior]$ , $[\pm coronal]$ , $[\pm strident]$ , and so on ^{[ChomskyHalle]}. The example below shows a partial matrix.

Phoneme	voice	nasal	continuant	anterior	coronal
/p/	−	−	−	+	−
/b/	+	−	−	+	−
/m/	+	+	−	+	−
/t/	−	−	−	+	+
/s/	−	−	+	+	+

A phonological rule has the form

A ⟶ B / X_Y,

read as "segment $A$ becomes $B$ in the environment between $X$ and $Y$ ," where $X$ and $Y$ are feature descriptions (possibly empty, and possibly mentioning word or syllable boundaries). The English aspiration rule is $/ p / \to [p^{h}] / #_V_{[+ stress]}$ .

A prosodic system layers information above the segment: tone (pitch used to contrast words, as in Mandarin), stress (relative prominence of syllables), and intonation (pitch contour over a phrase). Prosody is represented on a separate tier from segmental features.

Counterexamples to common slips

Phones are not phonemes. [tʰ] and [t] are distinct phones; their phonemic status depends on the language. The brackets carry the distinction: square for phone, slash for phoneme.
Complementary distribution alone is not enough. Two phones in complementary distribution must also be phonetically similar to be assigned to one phoneme; otherwise one risks grouping arbitrary sounds ([h] and [ŋ] are in complementary distribution in English but are not one phoneme).
Features are not just labels. SPE features have articulatory and acoustic correlates; later work (Clements & Halle 1991, sage dict) organises them into a feature geometry whose nodes govern phonological processes.

Key concepts Intermediate+

The contrast–distribution diagnostic. Two phones $[x]$ and $[y]$ of a language are assigned to distinct phonemes if and only if there is a minimal pair, equivalently if they occur in the same environment (overlapping or contrastive distribution). If no minimal pair exists and the phones are phonetically similar, they are tested for complementary distribution: if their environments do not overlap, they are allophones of a single phoneme ^[Hayes]. The decision procedure is:

Collect the surface phones from a corpus of the language.
For each pair, search for a minimal pair. If found, assign distinct phonemes.
If none is found, test for complementary distribution. If the environments partition cleanly and the phones are phonetically similar, assign one phoneme with allophones.
If the phones occur in the same environment without contrasting, they are in free variation and again belong to one phoneme.

Distinctive-feature theory. The phoneme inventory is not an unstructured list but a partially ordered set under feature inclusion. A natural class is a set of phonemes that share a value for some feature bundle, so that a single rule can refer to all of them at once. In English the class ${/ p, t, k /}$ is the natural class $[- voice, - continuant, - nasal]$ ; it is exactly the set targeted by the rule that aspirates voiceless stops. Natural-class structure is what makes phonological rules compact and typologically recurrent ^{[ChomskyHalle]}.

Rule-based derivation. A phonological derivation is a sequence of rule applications mapping an underlying representation to a surface representation. Each rule applies whenever its structural description is met; rule ordering (feeding, bleeding, counter-feeding, counter-bleeding) determines the output. This serial architecture was the dominant framework from SPE (1968) through the 1980s and remains the reference point against which later theories are measured.

Bridge. This contrast-and-distribution analysis builds toward 51.04.01, where the same rule-system methodology organises syntax, and appears again in 47.01.01, whose formal-language machinery treats phonological rules as string-rewriting systems over a finite alphabet. The foundational reason phonemes must be abstractions over phones is that meaning lives in contrast, not in raw acoustics; this is exactly the move that lets a finite inventory generate an unbounded set of word forms; the central insight is that discrete categorisation is imposed on a continuous signal. Putting these together, the bridge is that phonology supplies the first layer of discrete structure on top of continuous speech, the layer every higher level of grammar inherits.

Exercises Intermediate+

Exercise 7 (hard, symbolic).

A rule $R_{1}$ changes /i/ to [e] before a consonant, and a rule $R_{2}$ deletes a word-final consonant. Show that ordering $R_{2}$ before $R_{1}$ (counter-feeding) versus $R_{1}$ before $R_{2}$ (feeding) yields different surface forms for an underlying /VCP/ shape, and state which ordering the linguist must justify.

Hint

Track the consonant that conditions $R_{1}$ : if $R_{2}$ removes it first, $R_{1}$ 's environment is gone.

Answer

Take an underlying shape /ik#C/ where the final C would condition /i/ → [e] by $R_{1}$ and then be deleted by $R_{2}$ . Under feeding order $R_{1}$ then $R_{2}$ : /ik#C/ $R_{1}$ [ek#C] $R_{2}$ [ek#]. The vowel surfaces as [e]. Under counter-feeding order $R_{2}$ then $R_{1}$ : /ik#C/ $R_{2}$ /ik#/ and now the conditioning consonant is gone, so $R_{1}$ does not apply, giving [ik#]. The two orders yield [ek] versus [ik], so the linguist must justify the order against the actual surface forms of the language. This is the substance of rule-ordering argumentation.

Exercise 8 (hard, symbolic).

In Optimality Theory, suppose the constraints $FAITH_{IO}$ ("do not change the input") and $*_{VOICE}$ ("avoid voiced stops") are both active. For an input /b/, list the two main candidate outputs and explain how the ranking $*_{VOICE} ≫ FAITH_{IO}$ versus $FAITH_{IO} ≫ *_{VOICE}$ selects different winners.

Hint

The candidates are faithful [b] and devoiced [p]. The higher-ranked constraint decides.

Answer

The two candidates are [b] (faithful to the input /b/, but violates $*_{VOICE}$ ) and [p] (satisfies $*_{VOICE}$ , but violates $FAITH_{IO}$ by changing voicing). Under ranking $*_{VOICE} ≫ FAITH_{IO}$ , the constraint that bans voiced stops dominates, so the devoiced [p] wins: a language with final devoicing. Under ranking $FAITH_{IO} ≫ *_{VOICE}$ , faithfulness dominates, so faithful [b] wins: a language that preserves underlying voicing. Identical constraint set, different ranking, different typology.

Lean formalization Intermediate+

lean_status: none. Phonology is an empirical science whose correctness gate is corpus evidence and typological argument, not formal proof. The genuinely formal content (rules as string-rewriting over a feature-decorated alphabet; constraint ranking as a total order evaluated by an $EVAL$ function; the natural-class lattice) belongs to the formal-language theory of 47.01.01 rather than to Mathlib's mathematical core, and Mathlib currently has no phonological-rule or OT-tableau formalism.

Advanced results Master

Feature geometry (Clements 1985; Clements & Halle 1991). The flat SPE feature bundle was reorganised into a hierarchical feature geometry: a tree whose root branches into laryngeal, supralaryngeal (place), and manner nodes, with individual features as leaves. A phonological rule may then target an entire node, e.g. delinking the place node produces placeless consonants attested in many languages. This accounts for why features spread, assimilate, and delete as groups rather than individually.

Autosegmental phonology and prosody (Goldsmith 1976). Tones and other prosodic features live on autosegmental tiers linked by association lines to segments on the skeletal tier. This cleanly describes tone spreading, floating tones, and the stability of tones under vowel deletion. The tier idea extends to stress (metrical phonology, Liberman & Prince 1977), where stress is represented as relative prominence on a metrical tree rather than as a segmental feature.

Optimality Theory (Prince & Smolensky 1993). OT replaced serial rule ordering with parallel constraint evaluation. A generator $GEN$ maps an input to a set of candidate outputs; a constraint set $CON$ contains ranked, violable constraints of two families — markedness constraints penalising structurally dispreferred configurations and faithfulness constraints penalising deviation from the input; an evaluator $EVAL$ selects the candidate that best satisfies the ranking. A toy tableau illustrates the scheme:

/b/ →	$*_{VOICE}$	$FAITH_{IO}$
☞ [b]	$*$
[p]		$*$

With the ranking $*_{VOICE} ≫ FAITH_{IO}$ , [b] wins despite its markedness violation because faithfulness is lower-ranked. The framework's strength is typological: the same universal constraint set, reranked, generates the attested range of languages ^{[PrinceSmolensky]}.

Markedness and typology. The set of constraints and their universal rankings predict which inventories and processes are attested. Voiceless stops are more common than voiced ones (markedness); languages with final devoicing outnumber languages with final voicing. OT makes these gradients first-class objects of the theory rather than incidental observations.

Synthesis. The progression from structuralist phonemes through SPE features to Optimality Theory builds toward a unified, constraint-based view of grammar, and appears again in 51.04.01 where ranked constraints model syntactic well-formedness. The foundational reason OT displaced rule ordering is that typology is shaped by conflicting universal pressures; this is exactly the tension between markedness and faithfulness; the central insight is that every surface form is a compromise among ranked, violable constraints. Putting these together, the bridge is that phonology becomes the cleanest laboratory for studying how competing universal pressures resolve into language-particular systems, a pattern that generalises to morphology, syntax, and first-language acquisition.

Full proof set Master

Proposition (Complementary distribution entails no minimal pairs). Let $[x]$ and $[y]$ be two phones of a language $L$ , and let $E ([x])$ and $E ([y])$ denote the sets of phonetic environments in which they occur. If $[x]$ and $[y]$ are in complementary distribution, that is $E ([x]) \cap E ([y]) = \emptyset$ , then no minimal pair distinguishes them, and consequently no evidence exists in the corpus to assign them to distinct phonemes.

Proof. Suppose, toward a contradiction, that a minimal pair exists for $[x]$ and $[y]$ . By definition a minimal pair consists of two words $w_{1}$ and $w_{2}$ that are identical in every segment except that one has $[x]$ and the other has $[y]$ in the same position, and that differ in meaning. The shared position fixes a single phonetic environment $ε$ (the identical left and right contexts). Therefore $ε \in E ([x])$ (because $[x]$ occurs there in $w_{1}$ ) and $ε \in E ([y])$ (because $[y]$ occurs there in $w_{2}$ ). Hence $ε \in E ([x]) \cap E ([y])$ , contradicting $E ([x]) \cap E ([y]) = \emptyset$ . The contradiction defeats the supposition, so no minimal pair exists. $□$

Corollary (Allophone assignment is consistent). Given complementary distribution and phonetic similarity, the two phones may be assigned to a single underlying phoneme without loss of contrastive generalisation, because the proposition guarantees that no contrast is being collapsed. The converse fails: absence of minimal pairs alone does not establish complementary distribution, since the phones might be in free variation instead.

Connections Master

Phonetics 51.01.01. Phonology operates on the continuous phonetic signal described by phonetics: every phoneme category is an abstraction over measurable acoustic and articulatory events, and the allophone rules of this unit are stated in terms of the places and manners defined there.
Syntax 51.04.01. The rule-system and constraint-ranking methodology developed for phonology was the template for generative syntax: SPE-style derivations and OT-style ranked evaluation both transfer to constituency, movement, and well-formedness in syntax.
Formal languages and automata 47.01.01. Phonotactic templates and phonological rules are string-rewriting systems over a finite alphabet decorated with features; their generative power and their position in the Chomsky hierarchy are studied as formal-language theory.
Speech perception 29.03.01. The phoneme categories posited here are the categories the listener recovers from the acoustic signal, so perceptual evidence (categorical perception, compensation) constrains which phonological analyses are psychologically real.

Historical & philosophical context Master

Phonology as an autonomous discipline was founded by Nikolai Trubetzkoy in Grundzüge der Phonologie (1939), which fixed the phoneme as the unit of sound that participates in oppositions and classified those oppositions as privative, gradual, or equipollent ^{[Trubetzkoy 1939]}. Trubetzkoy's Prague School colleague Roman Jakobson, with Gunnar Fant and Morris Halle, recast the phoneme as a bundle of binary distinctive features grounded in acoustics (Preliminaries to Speech Analysis, 1952), giving phonology its first principled algebraic vocabulary.

The generative turn came with Noam Chomsky and Morris Halle's The Sound Pattern of English (1968), which unified the feature set with a rule-based derivation architecture and made phonology a chapter of universal grammar ^{[ChomskyHalle 1968]}. SPE's exhaustive feature system and its ordered-rule derivations set the research agenda for two decades, but accumulating problems with rule ordering and opacity motivated Alan Prince and Paul Smolensky's Optimality Theory (circulated 1993, published 2004), which replaced serial derivation with parallel evaluation of ranked, violable constraints ^{[PrinceSmolensky 1993]}.

The philosophical stakes are real. Phonology claims that the sounds of a language fall into discrete, psychologically real categories whose existence is independent of any particular utterance — a strong form of mentalism about linguistic structure. Competing empiricist accounts (usage-based phonetics, exemplar theory) deny that abstract phonemes are needed at all and locate structure in the statistical distribution of remembered tokens. The longevity of the phoneme concept, despite these challenges, is itself evidence that some level of discrete categorisation is doing genuine explanatory work.

Bibliography Master

@book{Trubetzkoy1939,
  author = {Trubetzkoy, Nikolai S.},
  title = {Grundzüge der Phonologie},
  publisher = {Travaux du Cercle Linguistique de Prague},
  year = {1939},
}

@book{ChomskyHalle1968,
  author = {Chomsky, Noam and Halle, Morris},
  title = {The Sound Pattern of English},
  publisher = {Harper \& Row},
  year = {1968},
}

@book{PrinceSmolensky2004,
  author = {Prince, Alan and Smolensky, Paul},
  title = {Optimality Theory: Constraint Interaction in Generative Grammar},
  publisher = {Blackwell},
  year = {2004},
}

@book{Hayes2009,
  author = {Hayes, Bruce},
  title = {Introductory Phonology},
  publisher = {Wiley-Blackwell},
  year = {2009},
}

@book{OGrady2010,
  author = {O'Grady, William and Archibald, John and Aronoff, Mark and Rees-Miller, Janie},
  title = {Contemporary Linguistics: An Introduction},
  edition = {7},
  publisher = {Bedford/St. Martin's},
  year = {2010},
}

@article{JakobsonFantHalle1952,
  author = {Jakobson, Roman and Fant, Gunnar and Halle, Morris},
  title = {Preliminaries to Speech Analysis},
  journal = {MIT Acoustics Laboratory Technical Report},
  number = {13},
  year = {1952},
}

Prerequisites

51.01.01

Tier anchors

beginner: O'Grady, Archibald, Aronoff & Rees-Miller Contemporary Linguistics Ch. 3; Hayes Introductory Phonology Ch. 1 (which sound differences change meaning)
intermediate: Hayes Introductory Phonology (Blackwell 2009) Ch. 2–5 (phonemes, allophony, features, rules); O'Grady et al. Ch. 3
master: Chomsky & Halle The Sound Pattern of English (Harper & Row 1968); Prince & Smolensky Optimality Theory (Blackwell 2004); Trubetzkoy Grundzüge der Phonologie (1939)

References

Hayes — Introductory Phonology · Ch. 1–5: phonemes, allophones, distinctive features, rule writing; the contrast-and-distribution diagnostics
Chomsky & Halle — The Sound Pattern of English · Ch. 4–5: distinctive feature system (SPE), phonological rules and representations
Prince & Smolensky — Optimality Theory · Ch. 1–4: GEN, CON, EVAL; ranked markedness and faithfulness constraints
Trubetzkoy — Grundzüge der Phonologie · the Prague School oppositions: privative, gradual, equipollent; the phoneme as a bundle of distinctive features

Estimated time

beginner: 15m
intermediate: 40m
master: 90m

Phoneme	voice	nasal	continuant	anterior	coronal
/p/	−	−	−	+	−
/b/	+	−	−	+	−
/m/	+	+	−	+	−
/t/	−	−	−	+	+
/s/	−	−	+	+	+

Phoneme	voice	nasal	continuant	anterior	coronal
/p/	−	−	−	+	−
/b/	+	−	−	+	−
/m/	+	+	−	+	−
/t/	−	−	−	+	+
/s/	−	−	+	+	+

Phoneme	voice	nasal	continuant	anterior	coronal
/p/	−	−	−	+	−
/b/	+	−	−	+	−
/m/	+	+	−	+	−
/t/	−	−	−	+	+
/s/	−	−	+	+	+