Annotated Bibliography on Contrast and Complexity
This document was compiled by members of the project on Contrast and Complexity.
Gengbe and Yoruba have the same vocalic inventories, /i,e,a,o,u/ (but different nasal vowel inventories). In Gengbe, /e/ patterns as if it had no features while in Yoruba, /i/ patterns in this manner. Conclusion: underspecification on a language-particular basis.
Akinlabi presents evidence from Yoruba in favour of both radical and restricted underspecification theory. He shows that in intervocalic /r/-Deletion and vocalic spreading in loan vocabulary, /r/ has just a root node and is unspecified for place features at the time the rules apply. However, he states that while place features are unspecified for liquids, the feature [coronal] is present for obstruent coronals.
The article examines the phenomenon of fixed default segmentism in reduplicated forms and argues that it should be understood as emergence of the unmarked. Constraints on segmental markedness, such as Place markedness hierarchy, are satisfied at the expense of exact copying, even when they are violated freely in the language as a whole. The issues addressed include the nature of markedness constraints, and the character of featural versus segmental faithfulness.
An historical perspective on phonological issues, including underspecification.
This paper argues for Radical Underspecification, in which only unpredictable values for feature are included in the underlying representation; predictable values are inserted by rule during the course of the derivation. This model assumes some Markedness conditions from the UG.
Through a conisderation of [ATR] harmony in Yoruba, the issue of whether certain types of predictable information should be included in UR is addressed, arguing for an approach referred to as "radical underspecification."It us demonstrated that Yoruba requires that all predictable information be excluded from UR: both context-free and context-sensitive values for ATR must be excluded, including their omission from the mid Vs (for which ATR is contrastive). In addition, information concerning autosegmental associations between ATR values and ATR-bearing units must consistently be excluded from UR. The essence of the analysis is that [-ATR] only may be present in UR, a floating specification that associates to the rightmost eligible [ATR]-bearing unit, subsequently spreading leftward where possible. Underlying, conventional and rule-governed assignments of [-ATR] to particular Vs are prohibited from creating a [+high, -ATR] V; it is proposed that this is accounted for by a single co-occurrence constraint ruling out such a configuration. Results obtained include (1) a strong argument for a right-to-left option in the linking of unassociated features, (2) evidence that some undespecification must be determined on a language-specific basis (not solely through universal markedness) and (3) the argument that coocurrence constraints can govern multiple aspects of derivation.
Phonetically similar voicing contrasts can arise from distinct phonological representations. Three possible representations of a voiced obstruent are proposed: 1) it may be marked as laryngeal voice (LV); 2) it may pattern as a sonorant sound and thus have a Sonorant Voicing (SV) node, or, 3) it may be the result of contextual voicing (CV), meaning that the voiced obstruent has neither LV nor SV but surfaces as voiced when surrounded by other voiced sounds. Each system is discussed and illustrated with examples from a variety of languages, including a discussion of Northern Turkic languages supporting the SV hypothesis. In addition, examples from Dutch and Turkish show that more than one system of representation may be found in one language to account for the different behaviour of stops and fricatives.
The authors present arguments in favour of underspecification of the Coronal node. This theory of underspecification is constrained by universal markedness conditions and the Node Activation Condition and is concerned with how segments come to be specified. This raises the issue of how language learners build up representations that are consistent with universal markedness conditions and the Node Activation Condition.
Through an analysis of Old English stress assignment and Early Middle English Trisyllabic shortening, this paper demonstrates two of the advantages of OT as applied to historical linguistics. Firstly, OT is efficient in accounting for phonological patterns exhibiting non-uniformity. Secondly, OT opens up the possibility of devising a restrictive model of language change. It is argued that UG only allows a single constraint to be reranked at any one time. The latter hypothesis is called Reranking maximality hypothesis.
Complex codas in English syllables have an asymmetrical distribution: rimes of more than two positions are limited to word edges. This fact is attributed to a Coda Condition which restricts syllabification to two rime positions, but which no longer holds at the word level. At Level 1, the principle of Structure Preservation (Kiparsky 1985) enforces conformity with the Coda Condition, thus explaining the distribution of complex codas as well as the application of vowel shortening. Apparent exceptions to the Coda Condition result from an independent principle which licenses an additional rime position if the position is half of a partial geminate (Ito1986). After Level 1, Structure preservation is turned off, and as a result, syllable structure is less restrictive, allowing larger codas and making vowel shortening unnecessary.
The results of a forced choice picture selection task given to first language learners showed that while children acquire the ability to contrast phonemic oppositions in a consistent order, that is, they distinguished phonemes with less structure before those with more structure, the order did not parallel the order in which phonemic inventories expand cross-linguistically. These findings contradict the geometry proposed by Rice and Avery (1991) on the basis of phonemic inventories that occur cross languages. An alternate geometry is proposed.
The author proposes a markedness-based explanation for universals concerning stem-initial consonant clusters. It is argued that there is a universal set of redundancy rules (in the form of neutralization rules) which place constaints on distribution of marked features. At least some of these universals are claimed to be phonetic, arising from limitations of the speech production and perception systems.
It is proposed in this paper that Universal Grammar (UG) includes hierarchically organized lists of constraints on feature co-occurrence. The constraints are of two kinds: prohibitions and marking statements. The former absolutely exclude the cooccurrence of two features; the latter define the use of a given feature value in a given segment as being marked or complex. Segments characterized by a feature configuration mentioned in a prohibition are absolutely impossible and will never occur in a phonological system. Segments characterized by a feature configuration mentioned in a marking statement may occur in a phonological system only if the relevant marking statement is deactivated. The deactivation of a marking statement, however, is costly, since in this way, a complex feature configuration is introduced into the phonological system. Establishing the structure of a phonological system, then, simply means establishing which marking statements are active and which are deactivated in it. The more marking statements that are deactivated in a system, the more complex it will be. (p. 2)
On the basis of roundness harmony in Nawuri, it is argued that blocking effects show that labial consonants and round vowels specified for the [labial] articulator node must exist on the same plane. This view supports the feature geometry model presented by Sagey (1986) but contradicts more recent geometry models in which the place features of consonants and vowels are arrayed on different planes (Clements 1991, Odden 1991, Hume 1992, Clements and Hume 1992).
Drawing on place assimilation facts, this paper argues in favour of underspecified input representations in Optimality Theory. It is argued that allowing underspecified representations into the input and using general faithfulness constraints explains why unmarked elements are targets of assimilation and why these elements fail to trigger assimilation processes.
The article proposes a single set of phonological features to characterize place of articulation in both consonants and vowels. The set includes the oral cavity features labial, coronal, dorsal, and a pharyngeal cavity feature radical or constricted pharynx under the pharyngeal node. Thus the features back and round are eliminated form feature theory. The article also proposes that place features for vowels and glides be partially segregated form those of consonants in the sense that they are assigned at different planes in phonological representation.
This paper argues for contrastive specification, in which all and only contrastive features are specified in the underlying representation.
This paper argues for the feature representations as multi-tiered, hierarchical structures and for the theory of assimilation processes based on these representations. Place features are subdivided into two sets: the ones universally present in consonants and vowels and ones underspecified. The prediction is made that all cases of assimilation (including vowel harmony and long-distance assimilation phenomena) can be handled within the proposed theory of representations.
The author contends that the feature [+/-Nasal] is contrastive for French vowels in the phonological output, and that vowels with the feature [+Nasal] displays significant nasal air flow. By comparison, the feature [+/-Nasal] is not contrastive for English vowels in the phonological output, so no vowel is specified for this feature at this level. The gradient nasality in vowels preceding a nasal is the result of phonetic (nasality) interpolation.
This paper uses Morpheme Structure Constraints as criteria for determining underspecification. It is shown that MSCs in English holding between non-adjacent consonants treat coronal stop /t/ differently than non-coronal stops. Based on these facts, the author argues for a view of underspecification in which the presence of the Place Node for coronals is a parameterized option.
The paper argues that the prosodic development of children's early words can be accounted for in terms of prosodic constraints on output form. It is assumed that children's segmental representations are full from the beginning, but the early emergence of unmarked prosodic structures combined with the low initial ranking of Parse conspire to yield early word structures that differ from adult target forms. A given constraint ranking accounts for variability at a given point in time whereas constraint reranking accounts for development over time.
An overview of the idea of contrast in the phonological literature. Compares Radical Underspecification, Contrastive Underspecification, and Modified Contrastive Specification.
The first half of this thesis argues for Modified Contrastive Specification, specifically for the claim that [high] is present in underlying inventories only when there is a contrast between mid and high vowels within the front/back region, based on the patterning of metaphony (high harmony) in Spanish and Italian dialects. The second half of this thesis outlines a mechanismÑenhancementÑfor more fully specifying underspecified phonological representations in the phonetic component.
Discusses the problem of ternary power, and several related issues, including: Steriade's (1987) complement and default rules, and opacity.
This dissertation presentas evidence that phonological patterning is determined by the auditory/acoustic properties of speech sounds as well as by articulation. A primary source of constraints on auditory representations is the theory of phonological contrast. Specifically, it is proposed that one of the fundamental requirements on contrasts is that they should be maximally auditorily distinct. This is one of the three conflicting goals that shape the selection of contrasts: (i) Maximize the number of contrasts; (ii) Maximize the auditory distinctiveness of contrasts; (iii) Minimize articulatorily effort.
The paper proposes that the difference between child and adult phonology lies in relative rankings of constraints against markedness and constraints demandinf faithfulness to the input. In the child's initial stage the markedness constraints outrank faithfulness constraints. As acquisition progresses the appropriate faithfulness constraints are promoted. Th epaper examines the treatment of syllable onsets in an intermediate stage phonology.
Consonant weakening processes involve gradient changes from long closure to complete deletion. Such gradient changes are difficult to express within feature geometry since no unified set of features is able to account for these types of gradient changes. In this paper, Aperture Theory is used to explain consonant weakening as the deletion of an aperture node while consonant strengthening involves the insertion of an aperture node. To account for the cross-linguistic variation in targets of consonant weakening and the surface realizations of weakened segments, the author assumes Rice's (1993) version of underspecification.
The thesis gives a descriptive and a theoretical analysis of distributional co-occurrence constraints on consonants of Australian Aboriginal languages. Phonotactic positions of clusters are shown to be positions of neutralization of feature contrasts which are present intervocalically. The patterns of permissible clusters are argued to take the form of implicational relationships between segments and clusters, showing that all Aboriginal languages share a core of cluster types while they vary in elaborating additional marked clusters. It is argued further that phonotactic patterns are phonetically grounded; unmarked clusters correspond to structures which are gesturally and/or perceptually simple while marked structures are gesturally and/or perceptually complex. A phonetically-based theory of markedness is proposes to account for neutralization patterns.
Using data from coronal phonotactics in Australian languages that have four contrastive coronal , the author aruges that markedness cannot be all derived from the structural complexity of a segment and that there are inherent markedness relations encoded into the segment structure.
This article provides an analysis of partial assimilation in Toba Batak to argue for fully autosegmental account, which is shown superior to alternative segmental treatments. The analysis supports for constraints on the Inalterability of linked features, and argues for the claim that the Obligatory Contour Principle should be formulated at underlying representation
Incomplete representations in Dependency Phonology include: head |i| without dependent |i| (central vowels); head |u| without dependent |u| (unrounded back vowels); head |a| without dependent |a| (the empty vowel) (p. 269). The redundancy rule is: f Ñ> fÑf (i.e. head |f| is redundantly specified for dependent |f| (p. 262)
Coronal consonants and front vowels in Maltese pattern together in default assignment, in vowel-to-coronal consonant assimilation, and in the more general process of coronal assimilation. As a result, it is argued that coronal consonants and front vowels form a natural class of coronal sounds and the unmarked status of [Coronal] place of articulation can be extended from consonants to vowels.
The paper deals with vowel heights. There are two problems inherent in the SPE: how do we account for sytems with four vowel heights; and for rules that raise /lower vowel heights by one step each. By examining vowel height facts in Esimbi, which is spoken outside the Bantu region of Cameroon, the paper argues for a binary approach to distinctive features within the framework of underspecification.
Argues that a phonological analysis of Japanese requires reference to the stratification of the Japanese lexicon into morpheme of different classes. While most of segmental and sequentional constraints hold at lexical core, they gradually "turn off" toward the lexical periphery. Shows the necessity of the development of a constraint-based model of lexical organization.
Presents a prosodic analysis of Rendaku in Japanese supporting underspecification theory, phonological cyclicity, and nonsyllabified underlying representations. Motivation is given for a theoretical distinction between floating autosegments and autosegments linked to an unspecified x-slot.
Within the framework of OT, discusses features which are "...unspecified under all theories of underspecification, even under the 'contrastive' view (as well as of course the 'radial' view), e.g. [voice] in sonorants. In this case, the problem lies in the presence of redundant values in some contexts but not in others." (p.1) See also Ito and Mester 1993 (CJL article).
This paper outlines the history of the ideas that came together in Lexical Phonology and provides an overview of the model. It discusses the characteristics of postlexical rules, interactions between morphological and phonological strata, the nature of lexical representations, constraints on lexical rules, and structure preservation (among other questions). The authors emphasize the role the theory ascribes to underspecification and markedness theory.
This paper argues for a modified feature geometry in which the feature [consonantal] is a daughter of the root node by showing that it is phonologically active (i.e. the feature [consonantal] does spread or dissimilate). Empirical evidence for assimilation comes from Cypriot Greek, Rato-Romash and Uyghur. Evidence for dissimilation comes from Ahtna and Halland
The paper argues for the view that underspecification is not restricted to phonology but may persist into phonetic representation. The presented phonetic data (the occurrence of coarticulatory patterns in speech) can be interpreted as reflecting the presence or absence of feature values in surface forms. It is shown that the surface specification may depend on segmental contrasts.
Contains a chapter entitled "Feature Geometry, Underspecification, and Constraints."
Outlines the implications of Structure Preservation for underspecification: "...[SP] determines point-blank that any rule which introduces marked specifications of lexically non-distinctive features must be postlexical." (p. 93) E.g. a rule introducing values of [±voice] for [+sonorant] segments is blocked by SP because voicing is non-distinctive for sonorants in the language in question. Underlying assumption: Lexical Minimality, i.e. underlying representations contain minimal redundancy.
The author employs contrastive specification and binary features in her analysis of the Mandarin vowel system /i, ü, u, «, a/. Examination of the diminutive r-suffixation process in Mandarin.
The author argues that glottal is less marked than coronal place, and that we find coronal epenthesis only where circumstances do not permit the glottal stop. The author also argues in favour of ranked markedness constraints rather than underspecification.
This article examines the case for [voice] as a privative feature. While Dahl's Law in Bantu presents a potential problem for the theory of privative voicing, the problem is only apparent. Lombardi argues that the feature [-voice] is not required to explain this dissimilation process and that an analysis of this phenomenon that is consistent with privative voicing can be maintained .
The author reanalizes the Palestinian Arabic material (Davis 1995) to show that the constraint interaction in OT is superior to the parametric rule-based theory, providing a more explanatory account of the material. He discusses the the nature of process specificity under OT and the question of relevance of representations in OT.
Drawing on both phonetic and phonological evidence from Semitic and other Afro-Asiatic languages, McCarthy argues for a place feature [pharyngeal] in addition to the oral place features [coronal], [dorsal] and [labial]. This feature characterizes the class of gutturals, which includes pharyngeals, laryngeals, and uvular continuants in these languages. The phonological evidence includes avoidance of gutturals in codas, guttural degemination, root cooccurrence constraints, and vowel lowering.
This paper argues for a Restricted Theory of Underspecification and the use of privative features. Using the mimetic palatalization in Japanese as an example, the authors show that all coronal consonants (except /r/) must be specified for Coronal underlyingly, as these segments attract the mimetic morpheme. /r/, however, is the only liquid in the language and does not contrast with any consonants in place of articulation. /r/, therefore, is not specified for Coronal, and it does not attract the mimetic morpheme. The authors also reanalyse Rendaku and Lyman's Law, using Voice as a privative feature.
In this paper, Odden proposes a hierarchical, feature geometry representation in which vowel place features are dominated by a single Vowel Place node. This representation differs from an earlier one proposed by Steriade (1987) in that Odden's representation does not segregate the feature [round] from the other vowel features as is the case in Steriade's representation. Evidence from vowel harmony is brought in to support the revised representation.
The paper argues for the claim that coronals lack a Place node in their internal structure. Evidence comes from phonological processes in Fula, Guerre and Mau, in which vowels might spread features or be fused across an intervening coronal.
This introduction to the volume on coronals sumarizes three approaches to segmental representation: Radical Underspecification, Contrastive Specification, and Modified Contrastive Specification.
The whole volume is devoted to Coronal segments. Many contributors derive the representations of coronals from a theory of underspecification.
This paper evaluates Radical Underspecification theories which assume that underlying representations cannot contain predictable information. This information is supplied by structure building rules. The author shows that the mechanism of structure building rules is unmotivated; radical underspecification cannot express the asymmetry of feature values and leads to many false predictions
A range of asymmetric behaviours of the vowel [i] in Yoruba is accounted for by analyzing [i] as completely underspecified. The feature hierachy is adopted to account for the cases where symmetric behaviour of [i] is attested.
The paper examines the effect of [ATR] vowel harmony on low vowels in Okpe. The fact that an underlying low vowel surfaces as [-low], [-ATR] in certain [+ATR] environment is accounted for by claiming that low vowels are underlyingly unspecified for vocalic features. This account shows there is no need to posit special, ad hoc feature-changing rules
The author argues that the representation of vowels in verbal roots in Tiv require a type of radical underspecification: i. no features are assigned to vowels underlyingly; ii. surface forms result from the interaction of morpheme level specifications with rules of spreading and redundancy. The behaviour of assimilation rules is predicted by the proposed theory of featrue geometry.
Variability in the early acquisition of segments is examined. Individual and cross-individual variation in both child and adult language are investigated, and it is argued that the difference between child and adult variability is a quantitative rather than a qualitative one in that children show more variability because their immature phonological system has fewer contrasts. As the child's system becomes more marked with language-specific features, it becomes more like the adult's. A model that accounts for universal uniformity and individual variability is proposed whereby a segment is formed from feature hierarchies that establish constituency and markedness. A theory of specification accounts for the elaboration of structure during developments; variability results from underspecified segments. As contrasts are established during the course of phonological development, variation is eliminated.
In this paper, two cases used to support the claim that redundant features can be phonologically active are re-examined. Adopting a representational approach, Rice argues that surface exceptions to Rendaku in Japanese and Labial Attraction in Turkish indicate that the segments in question have underlyingly contrastive features, putting into question the claims of a lexical stratification approach in which redundant features can be phonologically active.
The author argues for the unmarked status of Coronal place of articulation for vowels, as proposed by Hume 1992 for Maltese. She argues that when a contrast exists between front unrounded vowels and central unrounded vowels, the front vowels must be specified for Coronal. In systems like this, front vowels do not show any exceptional behavior in phonological processes. When no contrast exists, front unrounded vowels are unspecified for Coronal. In systems like this, front unrounded vowels show neutrality in vowel harmony processes, undergo assimilation, and appear as epenthetic segments.
This paper addresses the dual nature of laryngeals, which may be placeless or may act as members of the class of gutturals, along with pharyngeals and uvular continuants. "I argue that, in fact, the representation of laryngeals is intrinsically tied to the range of guttural consonants in a given language. Laryngeals are specified as Pharyngeal only when pharyngeals or uvular continuants are also present in the inventory of the language; otherwise, they are Placeless."(73) Rose invokes two organizing nodes: Pharyngeal, a Place node that covers a region of articulation that includes the larynx, and [RTR], a tongue feature dominated by Phar that characterizes all gutturals except laryngeals.
Shaw argues that those coronals that do not require distinctive specification in terms of a subcoronal feature will be unspecified for a Coronal articulator node.
Distinguishes between two types of predictable values, and argues that only one type is missing from phonological representations. The missing ones are R-values, or those values which are predictable from feature coocurrence restrictions (e.g. [+voice] for [+sonorant] segments. The ones that can be present in phonological representations are D-values, or values which are contrastive within a given class of segments (e.g. [±voice] for obstruents).
A critical review of underspecification theory, aimed at arguing against Lexical Minimality and underspecification, and in favour of a declarative, licensing approach to the phenomena of positional neutralization. "Although the question of derivational vs. declarative phonology cannot be considered directly here, we do have to ask whether the phenomena attributed to underspecification must be analyzed derivationally, by distinguishing an earlier, less-specified stage of the derivation from a more fully specified, later stage. This will turn out to be the key issue."
The authors distinguish a set of primary features that are perceptually the most salient. The strength with which a primary feature is manifested in a given sound is influenced by the secondary features that co-occur with it. They argue that their analysis provides support for the view that the frequently occurring feature combinations in the languages of the world come about because those combinations maximize perceptual distinctiveness through the mechanism of feature enhancement.
Trubetzkoy uses both contrast and substantive properties to characterize segmental systems. Trubetzkoy's discussion of German liquids vs. Czech and Gilyak liquids is illustrative of his views on contrast.
McCarthy's analysis of Pasiego high harmony has been widely cited as an example which requires that both values of a feature be present underlyingly. Vago argues against McCarthy's analysis, developing a non-feature-changing analysis which maintains that there is only one underlying feature value (the other being inserted by default rule).
The author argues for a universal procedure of contrastive specification in which contrasts are established on a step-wise basis following the feature hierarchy of Height>Labial>Coronal (>prior to). The evidence for ranking Labial above Coronal, i.e., rounding contrasts specified before front vs. non-front contrast comes from vowel harmony in Turkish, Khalka and Buriat. Vowels in different inventory shapes will pattern in predictably different ways.
Using a theory of modified contrastive specification and privative features, the author shows that the four Mandarin underlying vowls /i, u, «, a/ display a two height system, with /a/ marked for Low and other vowels unmarked for a height feature. In terms of place marking, /i/ is marked for Coronal, due to the presence of a central vowel /«/, and /u/ is specified for Labial. This specification provides a satisfactory account of the assimilation processes and co-occurrence restrictions in Beijing Mandarin. The author also examines the r-suffixation processes in Beijing Mandarin, proposing that the diminutive suffix -r is a featural morpheme rather than a segmental one.
The paper argues for two claims: (i) coronals occur freely due to their lack of Place features, (ii) restrictions on Place specifications may hold for specific syllable positions. A number of languages is examined under two conditions: Modified Coda condition (codas may not have Place features) and Cluster condition (Adjacent consonants are limited to at most one Place specification).
This paper relies on the notion of "distinctness" in the discussion of the segments. Two segments distinguished only by the presence and absence of some feature, say [distributed], are non-distinct, and are considered "identical" for the computation of the OCP. The use of the Coronal feature [+/-distributed] follows the Restrictive Specification theory, whereas the use of the feature [anterior] follows the theory of Radical specification.
The paper argues that the OCP can act as a rule blocker and a rule trigger. It is also shown that the OCP is able to disambiguate an otherwise ambiguous rule, and a constraint on the form of possible rules. It is argued that this view captures a generalization missed in the traditional statement of the rule: the condition that two of the segments involved share feature values. Evidence mainly comes from English, Italian, Seri, Berber, Chumash, Cantonese, Axininca Campa and Kasem.
This is a synchronic and diachronic study on the vowel phonology of five Manchu-Tungus languages of China. The author proposes that the vowel system of written Manchu (old form of spoken Manchu, Xibe, and Hezhen) is a tongue root system, drawing evidence from vowel harmony processes. The author also proposes a feature hierachy of Low>Coronal>Labial>ATR/RTR (> prior to) for contrastive specification in vowel inventories in the Manchu-Tungus languages.
The idiosyncratic behaviour of ghost segments distinguishes them from similar cases of epenthesis or syncope and shows that they must be represented in underlying form. Examined are the nature and conditions for the surfacing of structure, the implications of which are shown to be borne out in data from French, Yokuts, Polish and Afar. The conditions for a ghost segment to materialize are linked to epenthsis: the ghost segment surfaces when epenthesis is motivated by prosodic or cluster constraints (eg. in OT), and takes the form of a C or V as determined by well-formedness conditions on epenthesis. Examples of this are shown in French and Polish. Because ghost segments result from the interaction of epenthesis and floating features, they are probably more prevalent than presently thought. An apparent case of syncope in Afar is shown to possess all the traits of a ghost, and a dialectal exception to syncope is handled as a ghost missing in underlying form, obviating the need for special marking.