Phonological features are the sub-segmental properties (such as [+voice], [+nasal], [+continuant]) that decompose speech sounds into their component articulatory and acoustic dimensions. Rather than treating phonemes as unanalyzable atoms, feature theory reveals that sounds sharing features form natural classes — groups that behave uniformly in phonological rules. For example, all nasals ([+nasal]: /m, n, ng/) pattern together in assimilation processes across unrelated languages because they share a feature specification. Feature geometry extends this by organizing features into a hierarchical tree structure, where nodes represent articulatory regions (laryngeal, place, manner), predicting which features can spread, delink, or assimilate independently and which must move as a unit.
Classify all English consonants by their feature specifications and then identify which natural classes are needed to state common phonological rules (e.g., aspiration of voiceless stops, nasalization of vowels before nasals). Work through assimilation processes in a language like Turkish or Sanskrit to see how feature spreading operates over natural classes. Draw feature geometry trees for a set of segments and practice delinking and spreading nodes to model observed alternations.
When you studied phonological systems, you learned that languages select a subset of all possible sounds — phonemes — and use contrasts between them to distinguish words. But what makes two sounds contrast? What makes /p/ and /b/ different in a way that matters, while many other sound differences are ignored by the phonological system? Feature theory answers this by decomposing phonemes into smaller, independently meaningful components called distinctive features.
The intuition is that phonemes are not atoms — they are bundles of properties. The phonemes /p/ and /b/ are nearly identical: both are bilabial stops, formed by closing both lips and releasing a burst of air. The single difference is voicing: /b/ is produced with the vocal folds vibrating ([+voice]), and /p/ is not ([-voice]). Feature theory represents this as the only difference in their feature specifications. This is more than notational convenience — it makes a testable prediction: any phonological rule that affects /b/ for voicing-related reasons should also affect /d/ and /g/, since they share the feature [+voice]. Cross-linguistic evidence overwhelmingly confirms these predictions.
The concept of natural classes follows directly. A natural class is a set of sounds that share at least one feature value and that behave as a unit in phonological rules. The nasals /m/, /n/, and /ŋ/ all carry the feature [+nasal]. In language after language — English, Japanese, Swahili — vowels nasalize before nasals, nasals assimilate to adjacent places of articulation, and nasals pattern together in ways that non-nasal consonants do not. Feature theory predicts exactly which classes should appear as natural classes, and which groupings (say, /p/, /r/, and /ŋ/) should not pattern together — because they share no defining feature.
Feature geometry extends the theory by organizing features into a hierarchical tree rather than a flat list. Not all features behave independently: place features (where in the mouth the articulation happens — labial, coronal, dorsal) can spread as a unit in assimilation without affecting laryngeal features (voicing). Feature geometry captures this by grouping place features under a single dominating node; spreading that node moves all place features together. This predicts which assimilation patterns are possible and which would be typologically unnatural.
A critical distinction to carry forward is that features operate at the phonological level, not the phonetic level. [+/-voice] is a categorical contrast used to distinguish phonemes; in actual speech, voicing is a continuous acoustic property that varies with context, speaker, and rate. Feature theory is a model of the grammar, not a direct description of acoustics. This is the same distinction as the phoneme/allophone distinction you encountered in phonological systems, now extended downward to the sub-segmental level.