Distinctive Features

Why phonemes are not atoms

If you make a list of English phonemes — /p, b, t, d, k, g, f, v, s, z, m, n, ŋ, …/ — you have a list of about 24 consonants and 14 vowels. Treated as atoms, they have nothing in common. /p/ is /p/ and /t/ is /t/, end of story. But every English phonological rule disagrees with this picture.

Word-initially in a stressed syllable, /p/ becomes [pʰ]. So does /t/. So does /k/. The same rule, applied to three different segments. If you state it three times — "/p/ → [pʰ]," "/t/ → [tʰ]," "/k/ → [kʰ]" — you have missed a generalization: it is the voiceless stops, as a class, that aspirate. The class is what the rule sees. The phonemes are accidents of which class members happen to exist.

Distinctive features encode the class memberships explicitly. /p/ is a bundle [+consonantal, -voice, -continuant, -nasal, +labial]. /t/ is the same bundle except [+coronal] instead of [+labial]. /k/ swaps in [+dorsal]. The aspiration rule is [-voice, -continuant] → [+spread glottis] / #__ — one rule, applies to all three.

Natural classes

A natural class is a set of phonemes that share a feature specification. Phonological rules target natural classes; the test of a feature system is whether the classes it picks out match the classes rules actually need.

Class	Members in English	Feature specification	Rule that needs this class
Voiceless stops	p, t, k	[-voice, -continuant, -nasal]	Aspiration word-initially
Voiced stops	b, d, g	[+voice, -continuant, -nasal]	Pre-nasal devoicing in some dialects
Sibilants	s, z, ʃ, ʒ	[+strident, +coronal, +continuant]	Plural -es selection (church-es)
Sonorants	m, n, ŋ, l, ɹ, w, j, vowels	[+sonorant]	Always voiced; never trigger devoicing
High vowels	i, u, ɪ, ʊ	[+high, -consonantal]	Trigger palatalization in many languages
Front vowels	i, ɪ, e, ɛ, æ	[-back, -consonantal]	German umlaut; many vowel-harmony systems

Not every set of phonemes is a natural class. /p, ɹ, æ/ has no shared feature specification simpler than enumerating the three; no phonological rule targets exactly this set, anywhere, ever. Feature theory predicts this — and the prediction is borne out empirically.

Jakobson's distinctive feature matrix

The first comprehensive feature system was Jakobson, Fant & Halle's 1952 Preliminaries to Speech Analysis. They proposed twelve acoustic features and represented every phoneme as a column of plus/minus values.

Feature	p	t	k	b	d	g	m	n	ŋ
vocalic	−	−	−	−	−	−	−	−	−
consonantal	+	+	+	+	+	+	+	+	+
nasal	−	−	−	−	−	−	+	+	+
compact	−	−	+	−	−	+	−	−	+
grave	+	−	+	+	−	+	+	−	+
voiced	−	−	−	+	+	+	+	+	+

The matrix exposes the contrasts. /p/ vs /b/ differ in one cell: voicing. /p/ vs /t/ differ in one cell: gravity. /m/ vs /n/ differ in one cell: gravity. The minimal contrast between any two phonemes is exactly the cell they disagree on, which is the contrast a learner has to acquire and a sound change might neutralize.

The SPE feature inventory

Chomsky & Halle's Sound Pattern of English (1968) overhauled Jakobson's acoustic features into an articulator-based system, which remains the basis of pedagogical phonology. The features split into five categories.

Category	Features	Distinguishes
Major class	[±consonantal], [±sonorant], [±syllabic]	Vowels vs consonants vs glides; obstruents vs sonorants
Manner	[±continuant], [±nasal], [±lateral], [±strident], [±delayed release]	Stops vs fricatives, oral vs nasal, etc.
Laryngeal	[±voice], [±spread glottis], [±constricted glottis]	Voicing, aspiration, glottalization
Place	[±anterior], [±coronal], [±high], [±back], [±round], [±distributed]	Where in the vocal tract the obstruction occurs
Vowel-specific	[±high], [±low], [±back], [±round], [±tense], [±ATR]	Tongue height, frontness, lip rounding, root position

Roughly 25 features, drawn from a universal inventory; each language activates a contrastive subset. English uses [voice] but not [spread glottis] contrastively (aspiration is allophonic). Hindi uses both. Korean uses neither but contrasts [tense] / [aspirated] / [lax] in stops. Akan uses [ATR] (advanced tongue root) for a cross-cutting vowel contrast English has no parallel for.

Cross-linguistic feature use

Languages differ in which features are contrastive. The same articulatory parameter can be a phoneme in one language and an allophonic detail in another.

Feature	Contrastive in	Allophonic in	Example contrast
[±voice] (stops)	English, French, Russian	Mandarin, Hawaiian (no /b, d, g/)	English pat /pæt/ vs bat /bæt/
[±spread glottis]	Hindi, Mandarin, Thai, Korean	English, French, Spanish	Hindi pal /pəl/ vs phal /pʰəl/
[±nasal] (vowels)	French, Polish, Portuguese	English (allophonic before nasals)	French beau /bo/ vs bon /bõ/
[±round] (front vowels)	French, German, Turkish	English (no rounded front vowels)	French si /si/ vs su /sy/
[±long] (vowels)	Finnish, Japanese, Czech	English (length is allophonic)	Japanese obasan "aunt" vs obāsan "grandmother"
[±ATR]	Akan, Igbo, many West African	Most non-African languages	Akan kasa [-ATR] "language" vs [+ATR] equivalent
Tone register	Mandarin, Yoruba, Vietnamese	Non-tonal languages	Mandarin 妈 mā vs 麻 má vs 马 mǎ vs 骂 mà

This table is half the argument for distinctive features. The same phonetic parameter (aspiration, nasality, length) is featurally identical across languages; what differs is whether the language has elevated it to a contrastive role. The features are universal, the contrasts are language-specific.

Worked example: English plural -s

The English plural suffix surfaces three ways: [s], [z], or [ɪz]. The choice depends on the final segment of the noun.

1. Data. cat-s [kæts], dog-s [dɒgz], church-es [tʃɝtʃɪz], kiss-es [kɪsɪz], dish-es [dɪʃɪz].
2. Pattern. [ɪz] after sibilants (/s, z, ʃ, ʒ, tʃ, dʒ/). [s] after voiceless non-sibilants (/p, t, k, f, θ/). [z] elsewhere (after voiced sounds).
3. Without features. Three rules listing six sibilants, five voiceless segments, and "everything else." Ugly.
4. With features. Two rules. (a) Insert [ɪ] after [+strident, +coronal] (the sibilants). (b) The remaining /z/ assimilates in voicing: [+strident, +coronal, +voice] → [-voice] / [-voice] __. Each rule references a single natural class.

This is the SPE-style argument feature theory was built to make: the rule is exactly as complicated as the natural classes involved. If you used arbitrary segment names, the rule would look ad hoc. With features, the rule's form mirrors its content.

Competing theories

Distinctive feature theory is not monolithic. Three live alternatives:

Theory	Key idea	Strengths	Weaknesses
SPE binary features (Chomsky & Halle 1968)	Flat unordered bundles of [±F]	Simple, complete, generative-friendly	Cannot represent feature dependencies; binarity is sometimes forced
Feature Geometry (Clements 1985, Sagey 1986)	Hierarchical tree: Place dominates labial/coronal/dorsal nodes	Explains why some features spread together (whole Place node)	Tree topology contested; not all spreading patterns fit
Articulator-based features (Halle 1995, Halle & Stevens)	Features tied to specific active articulators (lips, tongue tip, tongue body)	Articulatorily transparent; matches motor control	Multivalued by nature, complicates symbolic phonology
Articulatory Phonology / Gestural (Browman & Goldstein 1989)	Phonology is overlapping articulatory gestures, not features	Captures coarticulation, timing; falsifiable predictions	Loses some abstract generalizations of feature systems
Element Theory (Government Phonology)	Three to seven monovalent "elements" (|A|, |I|, |U|, |H|, |L|, …)	Privative, fewer primitives	Less coverage; harder for non-experts
Substance-free phonology (Hale & Reiss 2008)	Features are arbitrary symbols; phonetic content irrelevant to phonology	Clean separation of grammar and phonetics	Cannot explain why natural classes recur cross-linguistically

Which theory you adopt has consequences. Feature geometry says assimilation is spreading of a tree node, predicting that "Place" assimilates as a unit. SPE features say it is a set of independent features that happen to spread together. Gestural phonology says it is overlapping articulator activity in time. The data — German place assimilation, Japanese nasal place assimilation — adjudicates the dispute.

Variants and refinements

• Privative features. Single-valued: a segment either has [labial] or does not, with no [-labial] cell. Privative features are common in modern feature geometry.
• Underspecification. Phonemes need not specify every feature in their lexical entry; redundant features are filled in by rule. Saves space and predicts that "default" feature values are easier to alter.
• Contrastive specification (Dresher 2009). Only contrastive features are present in lexical representations. /b/ vs /p/ in English need [voice]; /m/ does not need [voice] (it is redundantly voiced as a sonorant).
• Element theory. Government Phonology uses three to seven monovalent elements (A, I, U, H, L, ʔ, h) instead of features. Vowels are combinations: /e/ = |I, A|, /o/ = |U, A|.
• Substance-free phonology. Features carry no inherent phonetic content; the grammar manipulates symbols, and the symbol-to-articulation mapping is a separate (post-grammatical) module.

Common pitfalls

• Confusing features with phonemes. [+nasal] is not a phoneme — /m/, /n/, /ŋ/ are. Features describe phonemes; they don't replace them.
• Forgetting that the universal inventory is unactivated. Universal does not mean every language uses every feature contrastively — each language activates its own subset.
• Listing segments instead of features. "Rule applies to /p, t, k, f, θ, s/" is a sign you haven't found the right feature class. Likely [-voice].
• Treating SPE as the only feature theory. SPE is the canonical pedagogical baseline, not the current research frontier. Feature geometry, articulator-based features and gestural models all extend it.
• Confusing binary and privative. [-nasal] in SPE means "not nasal." In privative theory, the absence of [nasal] means the same thing — but you can't refer to it as [-nasal]. The notational difference matters for which classes are expressible.
• Assuming features are observable. Features are theoretical constructs. They are validated by the rules and patterns they let you state, not by direct observation.