Autosegmental or non-linear phonology is the analysis of speech sounds in several autonomous levels or dimensions, called tiers, rather than as a single string of phonemes. One tier holds articulatory features, another holds syllable size and shape, and another (in tone languages) holds tone. These may be further separated out: if a language has vowel harmony it is useful to look at its vowel features as forming one tier and its consonants another.

The tiers are linked to each other with association lines. This allows many important phonological processes to be explained in terms of linking, dissociation, or spread of associations. The phoneme theory did not explain why some phonetic processes were more natural than others.

Older versions of feature theory, such as that put forward in Chomsky and Halle's profoundly influential The Sound Pattern of English (1968), regarded phonemes as composed of many binary features, such as [+voice], [-nasal], [+coronal], and so on, the whole set building up to specify a phoneme such as [d]. These features could be used in phonological rules that operated on, say, all [+coronal] sounds.

In autosegmental theory a much more minimal feature specification is used, reflecting universally unmarked values: vowels are not nasal, stops are voiceless, nasal and lateral consonants are voiced, front vowels are unrounded, and so on. Only when a language contains a marked contrast with these does it have to be specified as a feature attached to the feature tier. For example, languages often have assimilation of voiced or voiceless obstruents, but this virtually never affects or is affected by so-called voiced nasals. Therefore it is better to regard nasals as not even specified for voice.

Also, many of the features are now regarded as unary or privative: they are either present or not. Nasal stops and vowels have a feature [Nasal] but oral stops and vowels just lack it: they don't possess an opposite feature or value. This again captures the reality better, in that languages always have phonological processes that operate on natural classes of sounds, marked out from the rest of their sound systems.

The feature tier is also known as the melody tier, because the autosegmental approach was originally used to describe tones. The other main tier is the timing or prosodic tier. This shows a word's division into syllables or morae. A mora is a unit of timing inside a syllable, and a syllable can be light or heavy depending on how many morae it has. Phonological processes often target syllable weight: one common example is stress, which in many languages is on heavy syllables, or makes stressed syllables heavy. Also, the possible word shapes in a language are strongly constrained by what kinds of syllable it allows.

Here as an illustration is a depiction of the English word income. At the top the number of x's marks the degree of stress on each syllable. Above the SAMPA phonemic transcription ["InkVm] are the X-slots of the timing tier, assigned to their place in the syllable (q.v. for details of structure). Below it are the (main) features that compose the phonemes.

          x
          x               x

          σ               σ
          |             / |
          R            /  R
          | \         /   | \
          |  \       /    |  \
          |   \     /     |   \
          N    Co   O     N    Co
          |    |    |     |    |
          X    X    X     X    X
          |    |    |     |    |
          I    n    k     V    m
          |    |    |     |    |
        +hi   Cor   Dor  -hi   Lab
        -bk   Nas        +bk   Nas
        -ATR             -ATR
                         -rd
Many people pronounce this word as in-come, with the coronal nasal of in, but others give it the dorsal nasal of ink. To show this process of assimilation, the association line between [n] and its original [Coronal] feature is delinked and a new association with the neighbouring [Dorsal] feature is created:
               X    X
               |    |
               n    k
              /|.   |
             / = .  |
            /  |  . |
          Nas Cor  Dor

Log in or register to write something here or to contact authors.