Syllable Weight

Phonetics, Phonology, Typology

Matthew Gordon

• Publisher: Routledge
• Year: 2007
• ISBN:
  9781135922269
• Edition: 1
• Topic: Languages & Linguistics
• Subtopic: Languages
• Index: Languages & Linguistics

The book is the first systematic exploration of a series of phonological phenomena previously thought to be unified under the rubric of syllable weight. Drawing on a typological survey of 400 languages, it is shown that the traditional conception that languages are internally consistent in their weight criteria across weight-based processes is not corroborated by the cross-linguistic survey. Rather than being consistent across phenomena within individual languages, weight turns out to be sensitive to the particular processes involved such that different phenomena display different distributions in weight criteria. The book goes on to explore the motivations behind the process-specific nature of weight, showing that phonetic factors explain much of the variation in weight criteria between phenomena and also the variation in criteria between languages for a single process. The book is unlike other studies in combining an extensive typological survey with detailed phonetic analysis of many languages. The finding that the widely studied phenomenon of syllable weight is not a unified phenomenon, contrary to the established view, is a significant result for the field of theoretical phonology. The book is also an important contribution to the field of phonetically-driven phonology, since it establishes a close link between the phonology of weight and various quantitative phonetic parameters.

Chapter One

Introduction

1.0 BACKGROUND

Linguists have long observed that certain phonological phenomena in many languages distinguish between “heavy” and “light” syllables (e.g. Jakobson 1931, Trubetzkoy 1939, Allen 1973, Newman 1972, Hyman 1977, McCarthy 1979a, b, etc.). For example, Latin preferentially stressed closed syllables and syllables containing long vowels over open syllables containing a short vowel (Allen 1973). Closed syllables and syllables containing long vowels were thus heavier in weight than open syllables containing a short vowel in Latin.

While the exact definition of syllable weight is elusive, it may be defined very broadly as that property which differentiates syllables with respect to their prosodic behavior. The difficulty in explicitly defining syllable weight lies in determining which prosodic aspects of language fall under the rubric of weight. As indicated by the Latin stress example above, stress figures prominently among those phenomena considered to involve syllable weight. The domain of weight, however, is not limited to stress. Other phenomena that are potentially sensitive to the weight of syllables include poetic metrics, compensatory lengthening, tone assignment, quantitative aspects of syllable structure, and reduplication. We consider here how these weight sensitive processes instantiate weight. In many languages, only certain types of syllables, the heavier ones, may carry contour tones (Hyman 1985, Duanmu 1994a, b). Weight distinctions are also relevant in many poetic traditions, in which the placement of syllables within the meter is governed by their weight (Hayes 1988). Similarly, many languages have constraints on the minimal size of many classes of words, typically content words. In such languages, words that are subminimal, or not heavy enough, are either disallowed or strongly restricted in their distribution (McCarthy and Prince 1986, 1990, 1995a). Many processes that lengthen or shorten syllables or segments also have been argued to fall under the rubric of weight-based phenomena. For example, long vowels do not occur in closed syllables in many languages, a restriction that has been argued to result from constraints on the maximum weight of the syllable (Steriade 1991, Hayes 1995). Reduplication has also been argued to be a weight sensitive process, because the reduplicant in many languages assumes a certain prosodic shape that appears to conform to some weight standard (McCarthy and Prince 1986, 1990, 1995a). All of these processes superficially have in common that they are sensitive to the phonological weight of syllables.

1.1. FORMAL REPRESENTATIONS OF WEIGHT

As the number of prosodic phenomena argued to instantiate syllable weight has grown, the notion of weight has played an increasingly larger role in phonological theory. In response to the burgeoning role of syllable weight in linguistic theory, phonologists have developed simple yet compelling representations of weight grounded in fundamental concepts such as phonemic length, segment count and sonority.

Of these theories of weight, the two that have gained widest acceptance are skeletal slot models, including CV and X slot models (McCarthy 1979a, b, Steriade 1982, Clements and Keyser 1983, Levin 1985), and moraic models (Hyman 1985, Hayes 1989). The appeal of both of these models is that they assume representations that are projected from independently contrastive properties such as segmental and length distinctions. Units of weight, either skeletal slots (in CV and X slot models) or moras, are assigned to segments. Syllables with a greater number of segments logically receive a greater number of weight units. Similarly, contrasts in segmental length are represented by assigning long segments two weight units, while short segments are associated with one unit of weight. Weight distinctions are thus reducible to differences in the number of units of weight in the syllable. Syllables with a greater number of weight units are “heavier” than syllables with fewer weight units.

The link between syllable weight and the representations designed to model it becomes clearer if we consider the case of Latin, using both moraic and skeletal slot models of weight. Recall from above that Latin preferentially stresses closed syllables and syllables containing long vowels over open syllables containing short vowels. The Latin primary stress rule that demonstrates this weight distinction is as follows: primary stress falls on a heavy penultimate syllable, equivalent to a closed syllable (1a) or one containing a long vowel or diphthong (1b). If the penult is not heavy, stress retracts onto the antepenult (1c) (Allen 1973).

(1) Latin stress

a. kar'pentum ‘carriage’

b. a'mi:ku:s ‘friend’

c. 'simile ‘similar’ nom., acc.sg. neuter

1.1.1. Skeletal slot models of weight

First let us consider the representation of Latin weight in a skeletal slot model (McCarthy 1979a, b, Clements and Keyser 1983, Levin 1985); the one presented here is that of Levin (1985). In skeletal slot models, the syllable is divided into constituents. Syllables consist of a nucleus, typically a vowel, which may be preceded by one or more consonants (the syllable onset) and also (in many languages) may be followed by one or more consonants (the coda). Together the nucleus and the coda form a constituent termed the rime (or rhyme). Short segments each project a timing position while long segments project two. In Latin, as in virtually all languages, the onset is ignored for purposes of calculating weight (but see Chapter Four). Only segments (and their associated skeletal slots) that belong to the rime contribute to the weight of a syllable.

The Latin weight distinction has a fairly straightforward representation in this model, as shown in Figure 1.1. (A syllable with both a long vowel and a coda consonant is also heavy, of course, since it contains three timing positions in the rime.)

Figure 1.1. Skeletal slot representations of three syllable types

1.1.2. Moraic models of weight

Now let us consider the representation of Latin weight in moraic theory (Hyman 1985, Hayes, 1989). The units of weight in moraic theory are moras. The weightless nature of onsets is directly captured by assuming that onsets are non-moraic. Contrasts in duration between short and long segments are represented as differences in mora count, parallel to the representation of duration contrasts as differences in the number of timing positions in skeletal slot models. Short segments receive one mora and long segments receive two, as shown in Figure 1.2. In Latin, consonants following a tautosyllabic vowel, i.e. those corresponding to coda consonants in skeletal slot models, are also moraic. The heavy vs. light distinction is thus captured succinctly in terms of mora count; syllables with at least two moras are heavy.

Figure 1.2. Moraic representations of three syllable types

1.1.3. Representations of Weight and Cross-Linguistic Variation in Weight Criteria

Clearly both skeletal slot and moraic models are well equipped to handle weight distinctions of the Latin type according to which closed syllables and syllables containing long vowels are heavy. The Latin weight distinction, however, is not the only weight criterion observed cross-linguistically. Another quite common weight distinction is one that treats only syllables containing long vowels as heavy. For example, in Khalkha Mongolian (Bosson 1964, Walker 1995, 1996), syllables with long vowels, including diphthongs, are heavy, while those containing short vowels are light, whether they are open or not.

The Khalkha weight distinction requires a slight expansion of the principles underlying the representations presented in Figures 1.1 and 1.2. Skeletal slot models must assume that the domain over which weight is calculated may differ between languages. Weight may be calculated over either the entire rime, as in Latin, or over just the nucleus, as in Khalkha. The moraic model must assume that the weight of coda consonants is subject to language specific parameterization, Hayes’ (1989) Weight by Position parameter. In languages like Latin, coda consonants are moraic, whereas in languages like Khalkha, they are not.

Interestingly, the Khalkha and Latin type weight distinctions do not exhaust the range of cross-linguistic variation in weight systems. As the database on weight sensitive phenomena available to theoretical phonologists has expanded to include information on a larger cross-section of languages, a diverse array of weight systems has been unearthed, necessitating expansions of the formal apparatus available to theories of weight.

Several languages, e.g. Komi Jaz’va (Itkonen 1955), Chukchi (Skorik 1961, Kenstowicz 1997), Kobon (Davies 1980, Kenstowicz 1997), Yimas (Foley 1991), which base their weight distinctions on neither segment count nor phonemic length contrasts, but rather on vowel quality, have attracted attention in the literature (see Chapter Two). Representing weight contrasts based on vowel quality in terms of differences in the number of weight units is problematic, since moras and skeletal slots are assumed to be projected from contrasts in segment length, not contrasts in segment quality.

Phonologists have also relatively recently noted the existence of languages with greater than binary weight distinctions (see Chapter Two). For example, stress systems in several languages, e.g. Klamath (Barker 1964), Chickasaw (Munro and Willmond 1994), Mam (England 1983, 1986), draw a ternary weight distinction with long vowels and diphthongs (CVV) at the top of the weight hierarchy, closed syllables containing a short vowel (CVC) in the middle, and open syllables containing a short vowel (CV) at the bottom. The representation of a ternary weight distinction of this type as a contrast in numbers of weight units requires that the heaviest syllable in the hierarchy, CVV, receive three moras. This practical necessity, however, violates the principle that representations of weight are projected from contrasts in length. This principle dictates that long vowels should receive two and not three moras. Recent work has even documented the existence of languages with greater than three levels of weight for stress assignment (see Chapter Two), e.g. Kobon (Davies 1980, Kenstowicz 1997), Kara (de Lacy 1997).

1.2. INCONSISTENCY OF WEIGHT CRITERIA

Another standard notion of weight which recent research has shown to be problematic is the view that weight is consistent across phenomena within the same language (Hyman 1985, McCarthy and Prince 1986, 1995b, Zec 1988, Hayes 1989). According to this hypothesis, which I will term the “moraic uniformity hypothesis,” all weight sensitive phenomena within a single language observe the same weight criterion and thus employ the same weight representations. Standard representations of weight have captured the assumption that weight is a property of languages by parameterizing weight criteria. For example, Hayes’ (1989) moraic theory assumes that coda weight is parameterized; some languages assign a mora to syllable-final (coda) consonants by the Weight by Position parameter, while others do not. Similarly, in skeletal slot models (e.g. Levin 1985), the syllabic affiliation of sonorant consonants is parameterized on a language specific basis: some languages syllabify postvocalic sonorant consonants in the nucleus, while others syllabify them as codas.

Several exceptions to the moraic uniformity hypothesis have surfaced in recent literature, e.g. Steriade (1991), Crowhurst (1991), Hyman (1992), Hayes (1995). For example, Steriade (1991) shows that the stress system, the system of poetic metrics, and the minimal root requirement of Early and Classical Greek are sensitive to different weight criteria from the pitch accent system. At both historical stages of Greek, the stress and metrical systems as well as the minimal root requirement treat both CVV and CVC as heavy. Pitch accent weight criteria are more stringent, however, at both stages. In Early Greek only CVV and syllables closed by a sonorant consonant (CVR) are heavy, while in Ancient Greek only CVV is heavy for purposes of pitch accent placement. A process of vowel shortening in syllables closed by a sonorant also points to the greater weight of CVR relative to syllables closed by an obstruent (CVO) in Early Greek. Crowhurst (1991), Hyman (1992), and Hayes (1995) present additional cases of non-uniformity of weight criteria within a single language.

Cases of conflicted weight criteria are problematic for two reasons. First, they necessarily require reference to at least three levels of weight in a single language. To see this, consider the case of Classical Greek (Steriade 1991). In Classical Greek, CVV is heavy for pitch accent assignment, minimal root requirements, and poetic metrics. CVC is heavy only in the metrical system and for the minimal word requirement but not for pitch accent placement. CV is light for all phenomena. Thus, collapsing all phenomena, CVV is heaviest, followed by CVC, followed by CV. For reasons discussed above, representing this ternary weight distinction is problematic in theories like moraic theory that encode weight distinctions as differences in number of timing positions. A ternary weight distinction requires that the heaviest syllable types, those containing long vowels, carry three moras, but long vowels should only be bimoraic. In fact, the potential for complex weight hierarchies involving more than three levels of weight grows as the number of weight-sensitive phenomena considered increases.

A second challenge presented by cases of conflicted weight criteria concerns the fundamental conception of weight as a language-driven rather than a process-driven phenomenon. Given the increasing number of cases of conflicted weight criteria reported in the literature, it seems worthwhile to explore systematically the alternative and equally plausible hypothesis that weight is more a function of process rather than language. Under this view, variation in weight criteria would be attributed principally to differences between weight-based phenomena in the weight distinctions they characteristically employ, rather than to differences between languages. For example, it could turn out that weight-sensitive tone tends to observe different weight criteria than weight-sensitive stress and that this process specificity accounts for many cases of conflicted weight criteria. If this scenario turned out to be true, the focus of the theory of weight should shift from explaining how and why languages differ in terms of their weight criteria to addressing how and why weight criteria differ between weight-sensitive phenomena. Exploring weight as not only a language-driven but also a process-driven property also has the potential to provide insight into cases of weight uniformity. To see how examination of the process specific nature of weight is potentially useful, consider the following hypothetical scenario. Let us suppose that coda consonants did not count in determining minimal word requirements in the majority of languages. Similarly, suppose that coda consonants also did not count in determining weight for tone in most languages of the world. This would raise two questions. First, we might ask why codas are characteristically weightless for computing minimal word requirements. Second, we would also want to know why codas are also weightless for purposes of tone in most languages. Crucially, in this hypothetical scenario in which codas are characteristically weightless for both tone and minimal word requirements, even if we were to find a language (in fact, even if we found many such languages) in which coda consonants were weightless for both tone and minimal word requirements, this would not provide support for the view that weight is uniform as a function of language. Rather, assuming that other weight-sensitive phenomena did not display the same cross-linguistic distribution of weight criteria as tone and minimal word requirements, the convergence of weight criteria for tone and minimal words within the same language would be an artifact of the process specifi...

Table of contents

• Cover
• Halftitle
• Title
• Copyright
• Contents
• Preface
• Acknowledgments
• Chapter One: Introduction
• Chapter Two: The Typology of Weight
• Chapter Three: Weight-Sensitive Tone
• Chapter Four: Weight-Sensitive Stress
• Chapter Five: Other Weight-Sensitive Phenomena
• Chapter Six: Conclusions
• Appendix One
• Appendix Two
• Appendix Three
• Appendix Four
• Appendix Five
• Appendix Six
• Notes
• References
• Index
• Language Index