1 Historical background
Autosegmental phonology emerged in response to limitations in earlier generative models of sound structure. Traditional linear approaches treated speech as a string of segments arranged one after another, but many languages show patterns in which a single phonological property extends over multiple segments or behaves independently of the segmental chain. Tone, harmony, and certain assimilatory patterns were especially difficult to represent in a purely sequential system.
1.1 Origins in generative phonology
The theory developed within the broader framework of generative phonology in the late 1960s and 1970s. Early generative work focused on rule-based descriptions of how underlying representations are converted into surface forms. As researchers encountered languages with nonconcatenative patterns, they found that segment-by-segment rules often required cumbersome or ad hoc formulations. These difficulties encouraged the search for a representation that could capture long-distance dependencies more directly.
1.2 Development of the autosegmental model
Autosegmental phonology proposed that phonological information is distributed across multiple tiers rather than stored on a single line. Features such as tone could be represented separately from the segmental string and connected by association lines. This made it possible to model spreading, deletion, and floating elements without forcing every process into a linear rule format. The model was initially used to analyze tone systems, but it soon expanded to other phenomena involving shared features and layered structure.
1.3 Major contributors
Several linguists played important roles in shaping the theory. John Goldsmith is commonly associated with its formal articulation, especially in work on tone and nonconcatenative representation. Other scholars contributed to related ideas about feature spreading, skeletal organization, and the treatment of harmony and prosody. The model developed through interaction with research on African languages, Semitic morphology, and general phonological theory.
1.4 Influence on later phonological theory
Autosegmental ideas influenced many later approaches to phonology. They helped establish the view that sound patterns may require hierarchical and multidimensional representation. Concepts such as feature geometry, prosodic structure, and metrical organization drew on autosegmental insights. Even where later theories modified or replaced specific mechanisms, the central idea that phonological information can be distributed across multiple linked units remained highly influential.
2 Core theoretical principles
Autosegmental phonology rests on the claim that phonological representations are not strictly linear. Instead, different kinds of information are placed on separate tiers, allowing each to have its own internal organization. Association lines connect these tiers and define which elements belong together in the surface form.
2.1 Multi-tier representation
In a multi-tier system, segments, tones, and other features are represented on distinct levels. A single segment may be linked to several feature bundles, and one feature may extend across more than one segment. This structure is designed to reflect the fact that some phonological properties behave independently of the segmental sequence. It also makes it easier to represent partial association and gaps in the phonological string.
2.2 Autosegments and association lines
An autosegment is an element on a tier that can function independently of the segmental line. It may represent a tone, a nasal feature, a vowel quality specification, or another feature-like unit. Association lines connect autosegments to the units they affect. These lines are central to the theory because they encode the pattern of alignment between independent tiers.
2.3 Nonlinear phonological structure
The model rejects the idea that all phonological processes operate only by replacing one segment with another in sequence. Instead, it allows processes such as spreading, delinking, docking, and reassociation. Nonlinear structure is especially useful when a property persists after its original segment is removed or when a single property influences multiple positions. The result is a more flexible representation of phonological patterning.
2.4 Well-formedness conditions
Autosegmental representations are constrained by formal principles that regulate how tiers may be connected. These constraints are intended to prevent ill-formed configurations and to capture common patterns of phonological organization. They help define which association structures are permitted in a language.
2.4.1 One-to-one and one-to-many associations
Some autosegmental relations involve a single autosegment linked to a single position, while others allow one element to associate with multiple positions. One-to-many mapping is important for explaining spreading phenomena, in which a tone or feature covers a sequence of segments. One-to-one mapping, by contrast, is common when each position has its own distinct specification.
2.4.2 No-crossing constraint
The no-crossing constraint prohibits association lines from intersecting one another in an ordered representation. This condition helps maintain a coherent mapping between tiers and prevents ambiguous structural configurations. It is often used to explain why certain reordering patterns are unattested or why derivations must proceed through specific intermediate states.
2.4.3 Obligatory contour principle
The obligatory contour principle states that adjacent identical elements on a tier may be disfavored or prohibited in certain contexts. It is often applied to tonal sequences, where repeated identical tones may be interpreted as a single spread tone rather than separate adjacent units. The principle has also been extended in broader ways to describe restrictions on repeated features in phonological systems.
3 Phonological representations
Autosegmental analysis divides phonological structure into several interacting components. The segmental string remains important, but it is no longer the only organizing principle. Features, tones, and timing units each contribute to the full representation.
3.1 Segmental tier
The segmental tier contains the consonants and vowels that make up the surface sequence. In many analyses, this tier is linked to other levels through skeletal positions or timing units. It provides the basic temporal order of the utterance, while other tiers specify properties that may extend beyond individual segments.
3.2 Suprasegmental tiers
Suprasegmental tiers contain features that are not confined to a single segment, such as tone, stress, or harmony-related specifications. These tiers often have wider scope than the segmental line and may attach to multiple positions. Their independence is a key reason autosegmental models were developed.
3.2.1 Tone tiers
Tone tiers represent pitch categories such as high, low, or contour tones. A tone may belong to a vowel, syllable, or other tone-bearing unit, but it can also remain unattached temporarily. This separation makes it possible to represent tonal floating, spreading, and reassociation.
3.2.2 Stress tiers
Stress tiers encode prominence relations among syllables or moras. Although stress is often analyzed through metrical theory, autosegmental treatment can represent it as a distinct layer associated with timing structure. This is useful for capturing nonlocal prominence patterns and secondary stress systems.
3.2.3 Vowel harmony tiers
Vowel harmony tiers represent features such as frontness, roundness, or [ATR] that spread across a word or phrase. Rather than assigning the same feature repeatedly to each vowel through separate rules, the model allows one specification to govern a sequence of vowels. This highlights the shared behavior of harmony domains.
3.3 Feature geometry in autosegmental analysis
Feature geometry extends autosegmental ideas by organizing features into hierarchical bundles. Instead of treating all features as flat and independent, it groups related properties under intermediate nodes. This structure can explain why some assimilations affect only part of a feature set, while others target a broader class of articulatory properties. Feature geometry preserves the nonlinear spirit of autosegmental representation while adding internal structure to feature systems.
3.4 Representation of moras and timing units
Moras and timing units provide a way to represent weight and duration. A mora is often associated with syllable weight, while timing units offer a skeletal framework for segmental association. These devices make it possible to model length contrasts, heavy syllables, and timing-based processes such as gemination. They also connect autosegmental theory with broader prosodic structure.
4 Tone and tonal processes
Tone is one of the classic domains for autosegmental analysis. Tonal patterns often involve properties that are not tied rigidly to individual segments, and they can survive segment deletion or move across a word in systematic ways. Autosegmental notation captures these behaviors especially well.
4.1 Tone-bearing units
Tone-bearing units are the positions that host tonal specifications, often syllables or moras. Not every segment carries tone in every language, and some units may host more than one tonal element. Identifying the correct tone-bearing unit is essential for describing how tones align with the speech stream.
4.2 Tone spreading
Tone spreading occurs when a tone extends from one tone-bearing unit to adjacent units. Instead of assigning the same tone through repeated independent rules, autosegmental analysis represents the tone as linked to several positions. This captures the sense that a single tonal property covers a larger domain.
4.3 Tone shifting
Tone shifting refers to a tonal feature moving from one position to another, often leaving behind an untoned unit. The process is useful for analyzing systems in which tones appear displaced from their expected location. Autosegmental derivations can represent this by delinking a tone from one unit and reassociating it with another.
4.4 Tone docking and floating tones
A floating tone is a tone that is present in the representation but not linked to a tone-bearing unit at a given stage. It may later dock onto an available unit, affecting pitch patterns or tone sandhi. Floating tones are a major advantage of the autosegmental model because they explain surface effects that seem to come from an absent but still active tonal element.
4.5 Downstep and upstep
Downstep is a lowering of a tone relative to a preceding tone, often triggered by an intervening floating tone or tonal configuration. Upstep is the corresponding raising in some systems, though it is less common. Autosegmental theory provides a way to treat these effects as consequences of tonal structure rather than purely phonetic fluctuation.
4.6 Tonal stability and deletion
Tonal stability refers to the tendency of a tone to persist even when the segment it was originally associated with is removed. Deletion of a tone-bearing segment may leave the tone floating, after which it can attach elsewhere. This behavior illustrates one of the central insights of autosegmental phonology: features and segments can have different lifetimes in a derivation.
5 Harmony and spreading phenomena
Autosegmental representations are also useful for describing harmony systems and other long-distance assimilations. In these patterns, a feature often extends across multiple segments within a defined domain. The model helps explain why such processes can be both regular and nonlocal.
5.1 Vowel harmony
Vowel harmony involves agreement among vowels in a word or phrase with respect to one or more features. Autosegmental analysis treats the harmonic feature as spreading across the relevant domain. This allows a language to be described without repeating the same feature on every vowel through separate local rules.
5.2 Nasal harmony
Nasal harmony spreads nasal specification across segments, typically from a nasal segment or nasal morpheme to neighboring sounds. The effect may extend over vowels, consonants, or both, depending on the language. Autosegmental theory is well suited to this phenomenon because nasality can be represented as a separate feature tier that associates with multiple segments.
5.3 Consonant harmony
Consonant harmony refers to the agreement of consonants in features such as place of articulation or secondary articulation. Some systems show long-distance similarity between consonants separated by other material. By representing the relevant feature independently, autosegmental analysis can model these dependencies without treating them as simple local assimilations.
5.4 Feature spreading across segments
Feature spreading is a general process in which a phonological property extends from one segment to another. The spread may be partial, affecting only part of a feature bundle, or broad, affecting a wider domain. This mechanism is one of the central explanatory tools of autosegmental phonology.
5.5 Transparent and opaque segments
A transparent segment does not block the spread of a feature, even though it may itself not show the feature. An opaque segment, by contrast, interrupts or alters the spread. Autosegmental theory accounts for these differences by examining whether the segment is linked to the relevant tier and how the association structure is configured. These notions are especially important in harmony systems with exceptions or directional effects.
6 Syllable and timing structure
Autosegmental approaches often rely on timing structure to connect segments with higher-level patterns. This framework separates the question of sequence from the question of feature content. As a result, it can represent length, alignment, and rhythmic organization in a compact way.
6.1 Timing slots and skeletal representations
Timing slots, sometimes called skeletal positions, provide abstract anchors for segments. A segment is linked to one or more slots, and a slot may remain empty or receive later association. Skeletal representations help explain how a language can preserve timing while changing segmental content.
6.2 Association of segments to timing units
Segments are mapped to timing units in ways that can vary by language and process. A single consonant may occupy one slot, while a long segment may occupy two. The association pattern is central to understanding how phonological length and temporal organization are encoded.
6.3 Gemination
Gemination is the doubling or lengthening of a consonant or, in some languages, another segment type. Autosegmental analysis often treats geminates as linked to multiple timing positions rather than as two separate segments. This captures their phonological behavior, including their participation in syllable structure and timing contrasts.
6.4 Elision and epenthesis
Elision deletes a segmental position from the surface form, whereas epenthesis inserts a new one. In autosegmental terms, these processes may create empty timing slots or additional associations. The model provides a framework for describing how a phonological system maintains well-formed structure when segments are lost or added.
7 Rule interactions and derivations
Autosegmental phonology often describes change through a sequence of structural operations. Rather than applying only linear rewrite rules, it uses linking, deletion, and reassociation to modify tier relations. This makes derivations easier to visualize and more faithful to the patterns found in natural language.
7.1 Linking and delinking
Linking connects an autosegment to a host position, while delinking removes that connection. These operations are fundamental to many analyses because they allow features to move, spread, or become floating. They also explain how a feature can remain in the representation after its original association has been disrupted.
7.2 Feature insertion
Feature insertion adds a new autosegment or feature specification to the representation. It may be used to account for epenthetic segments, default tonal patterns, or other phonological material not present in the underlying form. In autosegmental terms, insertion often involves creating a new unit that can later associate with the relevant tier.
7.3 Feature deletion
Feature deletion removes an autosegment, a link, or an entire feature bundle. Depending on the analysis, deletion may leave behind a floating element or an empty position. This distinction is important because the remaining structure can still influence later stages of the derivation.
7.4 Reassociation processes
Reassociation occurs when an autosegment changes its attachment from one position to another. This can happen after delinking, feature movement, or the creation of an empty site. Reassociation is especially useful for describing tonal phenomena and harmonic adjustments that cannot be expressed as simple local substitution.
7.5 Autosegmental derivations
An autosegmental derivation shows the step-by-step transformation from underlying to surface representation. It typically includes separate changes on different tiers, along with the resulting association pattern. Such derivations emphasize the independence of tiers while also showing how they combine to produce a complete pronunciation.
8 Formal devices and notation
The theory depends on a set of visual conventions for representing layered structure. Diagrams are used to show how tones, segments, and features align across tiers. These notations are central to the analytic clarity of the model.
8.1 Tree diagrams and tier diagrams
Tree diagrams and tier diagrams display the hierarchical organization of phonological elements. A tier diagram usually presents the segmental line alongside one or more autonomous tiers, with lines indicating connection. The format makes it easier to observe spreading, floating, and alignment relations at a glance.
8.2 Feature matrices
Feature matrices list the phonological features associated with a segment or autosegment. They are often used alongside tier diagrams to specify the internal makeup of segments. When integrated with autosegmental notation, matrices provide a compact record of feature content and contrast.
8.3 Association conventions
Association conventions determine how elements are linked in a representation and how ambiguous cases are interpreted. Some conventions specify default attachment, while others regulate directionality or scope. These standards help keep analyses consistent and make derivations easier to compare.
8.4 Comparison with linear phonology
Compared with linear phonology, autosegmental theory offers a more flexible account of patterns that cannot be reduced to strict sequence. Linear models are efficient for many segmental alternations, but they struggle with spread, floating material, and independent tiers. The autosegmental framework addresses these difficulties by separating representation into multiple interacting layers.
9 Applications in language analysis
The theory has been widely used in the description of diverse languages and sound systems. Its main strength lies in showing how a small set of structural principles can account for phenomena that otherwise seem irregular or complex. It has therefore become an important descriptive tool in phonological analysis.
9.1 Tonal language description
Autosegmental phonology is particularly well suited to tonal languages, where pitch contrasts may function independently of the segmental inventory. It can represent tone melodies, tone sandhi, and tonal alternations with precision. This made it a major framework for the analysis of many African and Asian languages.
9.2 Semitic root-and-pattern morphology
Semitic languages have been central to discussions of nonconcatenative morphology because roots and patterns interact in ways that do not resemble simple prefixing or suffixing. Autosegmental approaches can represent consonantal roots and vocalic templates on different tiers. This allows the analysis of interleaving forms as structured combinations rather than as exceptions to linear morphology.
9.3 African language phonology
Many African languages display tone systems, harmony processes, and assimilatory patterns that became foundational case studies for autosegmental analysis. Researchers used the model to describe tonal melody, downstep, and spreading effects with greater elegance than was possible in earlier frameworks. These languages helped establish the explanatory reach of the theory.
9.4 Case studies in vowel harmony systems
Vowel harmony systems in languages from several language families have been analyzed autosegmentally. Such studies often focus on how one feature spreads across a domain, how transparent vowels behave, and how harmony interacts with morphology. The framework has been especially useful where harmony appears to ignore ordinary segmental adjacency.
10 Criticisms and later developments
Although influential, autosegmental phonology has not remained unchanged. Later theories refined some of its assumptions, and critics pointed to areas where the model could be underspecified or overly descriptive. Even so, many of its ideas continue to shape phonological analysis.
10.1 Limits of the model
One criticism is that autosegmental representations can become complex when many tiers and associations are required. Some analyses risk using the framework to restate the phenomenon rather than explain it. Others note that not all phonological alternations fit neatly into spreading or delinking accounts, especially when fine-grained timing or gradient effects are involved.
10.2 Comparison with metrical phonology
Metrical phonology developed partly alongside autosegmental theory and shares its concern with hierarchical structure. It focuses more directly on stress and rhythmic prominence, while autosegmental work is broader in scope. The two approaches are often complementary, with metrical theory handling prominence and autosegmental theory handling feature association and spreading.
10.3 Integration with feature geometry
Feature geometry incorporated autosegmental insights into a more articulated model of feature organization. It preserved the principle that features can operate on separate levels, but it added internal structure to the feature system itself. This integration made it possible to capture more subtle patterns of assimilation and feature dependency.
10.4 Role in contemporary phonological theory
Contemporary phonology continues to use autosegmental concepts, even in frameworks that differ from the original model. Ideas such as tiered representation, association lines, and floating elements remain analytically useful. The theory’s legacy lies in its demonstration that phonological structure is often multidimensional rather than strictly linear.