Two intriguing properties of reduplicative affixes are that they are invariant in shape or weight and that their segmental makeup is dependent upon the base to which they are attached. Previous autosegmental explanations of shape- invariance and base- dependent melody adopt two mechanisms: i) a skeletal template to account for the consistency in weight and ii) a copying mechanism that transfers base segmental content into a reduplicative template (Marantz 1982, Broselow and McCarthy 1984, and Levin 1985). These accounts have since come under attack for a number of reasons. One major criticism concerns the unconstrained nature of templates; in, principle., any string of skeletal slots can be licensed as a template under these earlier accounts (Clements 1985, McCarthy and Prince 1986, 1987, and Steriade 1988). As a means of coping with this problem, McCarthy and Prince (1986, 1987) propose that reduplicative affixes be characterised by prosodically defined templates: i) core syllables (ad, ii) light syllables (σ(μ)), iii) heavy syllables (σ(μμ)), iv) syllables (σ), v) bimoraic feet (F(μμ)), vi) iambic feet (F(μμμ)), vii) disyllabic feet (F(σσ)), and viii) prosodic words. An important contribution of prosodic templates is that they establish prosodic constituency as a criterion for constraining templates, for a unit that does not correspond to a prosodic constituent can never be licenced as a reduplicative template. In addition, McCarthy and Prince (1986, 1987) exploit a copying mechanism. Copying can take two forms: i) an entire sequence of base segmental melody can be targeted for copying; or ii) a portion of it can be targeted for copying as long as that portion is prosodically defined or "circumscribed" (McCarthy and Prince 1990). In either case, copying is selective in that it excludes suprasegmental structures. Hence, I will refer to this model as Selective Copy. In contrast with Selective Copy, Steriade (1988) puts forward a different prosodic model of reduplication. Within her model, shape- invariance results from an interplay of weight and syllable markedness parameters. Among the weight parameters are: i) light syllables, ii) monosyllabic feet, iii) bimoraic feet, iv) disyllabic feet. These weight parameters differ from prosodic templates posited by Selective Copy. They are not templates that possess independent prosodic structures even though they are stated in terms of prosodic units that specify which prosodic constituent is targeted as the reduplicant. This conception of weight parameters forces a different copying apparatus. Copying must be exhaustive: it must target both the segmental melody and the prosodic structure of the base. This is critical; without the prosodic constituents of the base, weight parameters cannot select which unit of the copied base to retain as the reduplicant. In the following, I will refer to this model as Exhaustive Copy. This brief contrast demonstrates two radical differences between Selective Copy and Exhaustive Copy: i) templates vs. parameters and ii) selective vs. exhaustive copying. These differences impose a further contrast in expressing insertion of base - independent melody. Whereas insertion can make reference to the prosodic units of a template with Selective Copy, it cannot rely on parameters with Exhaustive Copy. Insertion can and must be defined by an existing prosodic structure of the base. This distinction becomes significant in examining Swati diminutive reduplication, which shows a base-independent vowel a in its reduplicant. What is important about this melody is that it is present only when a base cannot supply sufficient segmental content for a foot-sized reduplicant lingi-lingis 'resemble a little' vs. goba-gob 'bend a little'. The question addressed here is whether or not these two proposals can express a insertion. I demonstrate that a insertion cannot be formally stated by Exhaustive Copy. Crucial to the argument is that an insufficient base is lacking not only in segmental melody but also in prosodic constituency essential for defining the locus of insertion. Motivation for an insertion treatment takes two steps. First, internal evidence from passive formations suggests that i functions as the default vowel -not a. Second, this a behaves in Kikuyu like an architypical instance of "prespecification" in that it overrides any vowel in the base. As mentioned in footnote 4, Kikuyu and Swati are both members of the Bantu family. Moreover, reduplication as a diminutive marker is an important and widespread property of Bantu morphology (Meinhof 1932). This suggests that Swati a should be treated on a par with Kikuyu a since it cannot be filled in by default. The layout of this paper is as follows. Section 1 contrasts the two models of reduplication through examples from Tagalog. Section 2 applies Selective Copy to an account of Kikuyu and Swati reduplication. In particular, I demonstrate that Swati a requires an insertion treatment in consideration of both internal and external evidence. Section 3 spells out the argument against Exhaustive Copy from Swati a. In Section 4, I discuss some further implications of the templatic vs. parametric opposition for an overall theory of morphology. The critical point emerging from this discussion is that morphological processes such as Semitic roots and Japanese hypocoristics require access to templates; the parametric approach falls short on this score.

Recent developments in metrical theory have led to the situation in which there are now at least four different approaches to stress assignment. One approach uses only a grid to represent the relative prominence of syllables in a word (cf. Prince 1983); aside from representational conventions, the grid -only approach differs from the other three in that it does not posit any metrical constituency. Second, the constituentized grid approach also represents stress with a grid, but by enhancing the representations with parentheses, metrical constituency is also indicated (cf. Halle and Vergnaud 1987). Hayes (1987) has recently developed an approach employing representations like those in the constituentized grid approach; I will refer to this as the templatic approach. This approach is different insofar as the constituents which are available in the theory are not derived from parameters, but rather it is the constituent templates themselves which are the primitives of the theory. The fourth approach is one in which relative prominence is indicated with arboreal structures, rather than with grids (cf. Hayes 1981, Hammond 1984). In this paper I will present an analysis of the stress pattern of Pacific Yup'ik which follows Rice (1988), and I will claim that this analysis has important implications for each of the approaches mentioned above. Pacific Yup'ik is a particularly interesting testing ground for metrical theories; for our purposes here, the interesting aspect is that an adequate analysis of the stress pattern has broad implications for various approaches to stress assignment.

In Nootka, there is a strategy for forming hypocoristic names, or terms of endearment, from the normal form of the name by a combination of truncation, vowel mutation and affixation. The nature of this formation is highly suggestive of the type of morphology described by many linguists as subtractive. In this paper, however, we will show that what actually occurs is a pattern of template -filling based on the prosodic structure of the language. We will argue that the building of hypocoristic forms is, in fact, highly reminiscent of reduplicative strategies employed in this language as argued for in Stonham 1987 for the closely related Nitinaht language, the difference being that reduplication subsequently concatenates with the structure it has drawn from, while Nootka hypocoristic formation, henceforth H.F., abandons the remainder of the original structure, retaining only the copied portion required for the template. Before investigating the nature of H.F., we will first present certain aspects of Nootka structure which will be important for a clear exposition of the problem.

No description available.

In loanword phonology we seek to uncover the processes by which speakers possessing one phonological system perceive, apply native representational constraints on, and ultimately produce forms which have been generated by a different phonological system. In other words, loanwords do not come equipped with their own phonological representation. For any phonetic string, it is only native speakers for whom a fully articulated phonological structure is present. As host language speakers perceive foreign forms solely in accordance with their own phonological system, they instantiate native representations on the acoustic signal, fitting the superficial input into their own phonological system as closely as possible. Given these assumptions, it should not be surprising that despite the identity of a given acoustic signal when impinging upon the inner ear of speakers of different languages, this input may be represented, and ultimately produced in a distinct manner in each language it enters. The loanword phonology under investigation here, that of Cantonese, will be shown to possess two distinct levels. The first level of loanword phonology consists of a parsing of the input signal into unprosodized segment-sized chunks, for which native feature matrices are provided. As this level of loanword phonology is solely concerned with perceiving the input, and providing a preliminary linguistic representation, we may refer it as the Perceptual Level. It is only when full prosodic structure is supplied for the incoming form that the raw segmental material may undergo phonological processes, so that it may be realized in conformity with native prosodic constraints on syllable structure. As this stage of the loanword phonology admits the possibility of true phonological processes acting on segments, it may be regarded as the Operative Level of the loanword phonology. The processes which apply at the Operative Level of the Cantonese loanword phonology do not exist in native phonological derivations. As these operations were not acquired during the initial acquisition period, they exist in a separate domain from native phonological operations, presumably supplied by Universal Grammar. Their only property common with native phonological processes is that the same constraints exert an influence on the output of both systems. I will provide evidence for the Perceptual Level and the Operative Level of the loanword phonology by showing that loanwords undergo two distinct, ordered scansions during the course of the derivation. Scansion One will be shown to correspond to the Perceptual Level of the loanword phonology, providing raw segmental representation to incoming forms. Scansion Two will be shown to correspond to the Operative Level of the loanword phonology, providing prosodic representation which will be shown to trigger various phonological operations on the perceived segments.

In this dissertation I examine partial vowel assimilations, where more than one but less than all vowel features pattern together in a phonological rule. The result of this dissertation research is the 'Height Node Geometry'. The particular innovation this geometry makes is to group the height features ( (high) and (low)) under a separate Height Node, as opposed to having the height features report to the Dorsal Node or the Place Node. Motivation for the Height Node Geometry comes from analyses of several phonological processes. First, removing the height features from under the Dorsal Node and the Place Node facilitates a more natural explanation for reduplication in the Petit Diboum dialect of Fe?fe?-Bamileke. Second, placing the height features above the Place Node but still directly or indirectly under the Supralaryngeal Node provides an account for Tunica partial translaryngeal harmony. Finally, vowel harmony in Ewe involves spreading of (+high) and (+low) in the same environment, arguing for a simpler rule of node spread; hence, I propose that the height features stem from a separate Height Node, instead of directly to the Supralaryngeal Node.

This dissertation proposes a formal model of phonological performance, Double Lookup, that also has empirical consequences for theories of phonological competence. The most significant of these is the Productivity Hypothesis, the claim that the ordering of rules derives from their relative productivity. According to Double Lookup, the use of phonological knowledge during speech production occurs in two steps. First, forms are retrieved from memory; second, phonological rules are retrieved from memory and applied, if appropriate, to the retrieved forms. Phonological patterns may be applied during speech in this way or be prepatterned (stored as patterns across lexical items in memory). The productivity of a rule is defined to be the likelihood of its being retrieved and applied during speech production. In general, less productive rules are more likely to be prepatterned than more productive rules. The Productivity Hypothesis then follows: Because prepatterned forms are retrieved before rules are retrieved and applied, less productive rules will be ordered before more productive rules. Double Lookup and the Productivity Hypothesis are tested in several ways. First it is shown that the ordering of partially productive rules in English, as determined using standard linguistic methods, corresponds with their ranking in productivity, as determined through experiments described in the literature and through original surveys of speech errors. The application of fully productive rules in English is also shown to be consistent with the Productivity Hypothesis; fully productive rules do not apply in a linear sequence, but rather interact in accordance with universal principles. All apparent counterexamples actually involve less than fully productive rules. Next it is shown that the phenomenon referred to in the literature as cyclicity is correctly predicted to arise under certain well-defined circumstances, as when a rule is both prepatterned and very productive. In addition, it is shown that there are large categories of examples that cannot be handled by the notion of cyclicity at all, but find a simple account within Double Lookup. Finally, evidence for the model is summarized by comparing it with other models of rule ordering which face conceptual and empirical problems Double Lookup avoids.

Assimilation and coarticulation both involve extending the duration of some property or feature. The similarities between these phenomena can be seen by comparing Basque vowel raising with vowel -to -vowel coarticulation in a language like English. In Basque the low vowel /a/ is raised to [el following a high vowel. This gives rise to alternations in the form of the definite suffix, /-a/ (de Rijk 1970): (1) sagar –a; 'apple (def.)'; mutil-e 'boy (def.)'. In an English sequence containing a low vowel preceded by a high vowel, like [-ilæ-] in 'relapse', the high vowel also conditions raising of the low vowel. But in spite of the parallels between these cases, standard analyses regard Basque vowel raising as phonological whereas the English vowel raising is regarded as non-phonological, being attributed to a phonetic process of coarticulation. In this paper, we will argue that this distinction is untenable. We will see that coarticulation can affect the distribution of contrasts, and therefore must be specified in the phonology. This opens up the possibility of giving a unified analysis of assimilation and coarticulation. Analyzing coarticulation as phonological implies that phonological representations contain far more phonetic detail than is usually assumed to be the case. Vowel-to-vowel coarticulation involves fine degrees of partial assimilation in that vowels assimilate only partially in quality, and the effects may extend through only part of the duration of a segment (e.g. Ohman 1966). This conclusion thus flies in the face of the standard assumption that the richness of phonological representations should be severely restricted in order to avoid over-predicting the range of possible phonological contrasts. So before we turn to evidence that coarticulation is phonological, we will lay the groundwork by examining the arguments for limiting the detail in phonological representations and show that they are based on very questionable assumptions.

This paper presents an analysis of vowel harmony in Yawelmani and its interaction with vowel epenthesis and lowering, within the non -derivational, constraint -based model of Optimal Domains Theory (ODT). Kisseberth's (1969) analysis of the Yawelmani system, formulated within classical generative phonology, demonstrates an opaque rule interaction among the rules governing vocalic phonology, and was taken as an important piece of evidence for the notion of rule ordering in generative theory. The challenge in providing a non -derivational analysis of Yawelmani lies in accounting for conditions on vowel harmony which factor in phonological structure that is "inserted" in surface form, as well as structure that is "deleted" from underlying form. This paper presents a restricted means of bringing together information from underlying and surface representations in a theoretical framework that eschews intermediate representations, through the use of abstract (ie., unrealized) feature-domain structure. We discuss problems that arise under an alternative approach in which individual constraints are able to freely inspect structure at both underlying and surface levels of representation.

The neutralization of the laryngeal features of a consonant that is not directly followed by a vowel is a common process cross -linguistically. Laryngeal neutralization in this position has a clear phonetic cause: laryngeal features are not salient unless they are immediately followed by a vowel. Since laryngeal neutralization has a phonetic cause, it seems reasonable to characterize it directly in phonetic terms, without positing any additional layer of phonological abstraction. However, a phonetic explanation is not sufficient to account for all cases of laryngeal neutralization. For example, in Korean, laryngeal neutralization occurs in a nonneutralizing phonetic environment; in Nisgha, laryngeal neutralization occurs only in the reduplicant, although the phonetic environment for neutralization is found in both the reduplicant and the base. Although phonetics is the major factor leading to the development of these types of restrictions on laryngeal features, I argue that a phonetic account is not adequate for all such restrictions. Abstract phonological constraints and representations are necessary. Hence, two types of neutralization are possible: (i) phonetic neutralization, which results directly from the lack of saliency of cues and occurs in every instance of the neutralizing environment; and (ii) abstract phonological neutralization, which may occur where the neutralizing environment is absent (as will be demonstrated for Korean), and may fail to occur in every instance of the neutralizing environment (as will be demonstrated for Nisgha).