SEGMENTS WITH INHERENTLY FALLING SONORITY*

Transcription

1 SEGMENTS WITH INHERENTLY FALLING SONORITY* Martin Neef Abstract. This paper deals with an inconsistency in the application of the concept of sonority in such non-linear approaches to phonology that make use of both the segmental tier and the skeletal tier. I will argue that these approaches are forced to take the skeletal tier as the relevant level for the application of sonority. However, since long vowels occupy two positions on this tier, they form a sonority plateau, a configuration that is generally forbidden by basic sonority constraints. To solve this inconsistency, I will propose the concept of segments with inherently falling sonority. In addition, I will consider other candidates for elements with inherently uneven sonority. The intricacies of sonority will be discussed considering German as an example, following the assumption that phonological concepts, despite having a universal core, are characterised by language-specific adjustments. 1. The concept of sonority The concept of sonority has a long history going back to Whitney (1865), Sievers (1876), and Jespersen (1912). While sonority was only of minor importance in structuralist linguistics (with the notable exception of Sigurd 1955), it has become one of the core concepts of generative phonology since Foley (1970). 1 Despite this long history and the diversity of theoretical backgrounds proponents of sonority adhere to, there is a constant core in the conception of sonority in that a syllable has to show a specific sonority contour in order to be wellformed. The most famous formulation of the respective condition is the following Sonority Sequencing Principle (SSP) given by Selkirk (1984:116): (1) Sonority Sequencing Principle (Selkirk 1984:116) In any syllable, there is a segment constituting a sonority peak that is preceded and/or followed by a sequence of segments with progressively decreasing sonority values. In order to be well-formed, a syllable has thus to show a specific sonority contour. The centre of the syllable has to constitute a sonority peak and the sonority has to fall to both edges of the syllable. * I would like to thank Moritz Neugebauer, Cosima Scholz, Jessica Schwamb and several anonymous reviewers for their comments on earlier versions of this paper. 1 Sketches of the role and history of the concept of sonority in linguistics are given by Clements (1990) and Neef (2002). Studia Linguistica 58(3) 2004, pp Published by Blackwell Publishing Ltd., 9600 Garsington Road, Oxford OX4 2DQ, UK, and 350 Main Street, Malden, MA 02148, USA

2 Segments with inherently falling sonority 253 This conception presupposes that segments have differing values of sonority. These values are usually stated in terms of a sonority hierarchy. Phonologists are in disagreement about the adequate formulation of a sonority hierarchy. One of the aims of this paper is to suggest a new kind of such a hierarchy. For expository reasons, I give the following much cited hierarchy from Clements (1990:296) as an example: (2) A sonority hierarchy (Clements 1990:296) obstruents < nasals < liquids < glides < vowels In this hierarchy, vowels are the elements of highest sonority and obstruents are the elements of lowest sonority. This hierarchy is meant to have universal status. The only modifications allowed by specific phonological systems are to either unite adjacent sound classes into one class relevant for sonority or to divide up one class into several adjacent sonority sound classes. The first case is relevant e.g. for languages that only allow syllables not larger than one consonant followed by one vowel like Hawaiian (cf. Kenstowicz 1994:254). The second case concerns e.g. the division of the class of obstruents into two distinct sonority classes of stops and fricatives (cf. e.g. Blevins 1995:211). An open question concerns sounds that do not readily subsume under any of the sound classes mentioned in (2) like clicks. 2. A note on violability ClementsÕ proposal is embedded in a derivational framework that allows rules and constraints to hold at abstract intermediate levels between the underlying lexematic representation and the more or less observable surface representation. Clements explicitly claims ÔÔthat the SSP holds at a more abstract level than surface representationõõ (Clements 1990:287). Thus, in this conception sonority is not a surface-true generalisation but a principle that may be violated by other rules that apply later in the derivation. With respect to violability, such a derivational conception of generative phonology resembles currently prevalent optimality theoretic approaches: Derivational phonology employs violable rules and constraints for several different levels of representation while Optimality Theory is restricted to violable constraints for surface representations only. In such violability frameworks, however, it is impossible to make substantive claims about specific rules or constraints. This is because the soundness of a violable rule or constraint fundamentally depends on its relation to all other parts of the phonological system. Only if constraints are conceived of as unviolable devices, it is possible to assess a constraint in isolation. At least for methodological reasons I therefore subscribe to

3 254 Martin Neef a theoretical framework that disallows violability. 2 In particular, I follow assumptions of Declarative Phonology (cf. Scobbie, Coleman & Bird 1996, Walther 1999). I thus conceive of sonority as a set of unviolable constraints for determining the well-formedness of phonological objects. This decision is supported by the fact that theories allowing violability as opposed to declarative approaches are not falsifiable. A theory that allows unviolable constraints only is thus a stronger theory (cf. Ellison (2000) and Ploch (2003) for very astute and convincing criticisms of Optimality Theory that can be understood as a plea for a declarative conception of grammar). Moreover, I emphasise the autonomy of phonology from phonetics. In Autonomous Declarative Phonology, there is no need to search for a uniform phonetic correlate of phonological entities (cf. Neef 2004). Sonority is a purely phonological means to explain the distributional properties of phonological segments (cf. also Sigurd 1955). To develop a workable theory relying on unviolable constraints only, these constraints have to be both language-specific and domain-specific in the first place. Therefore, I restrict my considerations to one language only, namely German. This approach does not deny that the same constraints may be relevant for many different languages or potentially even for all human languages. This potential universality, however, is not part of the basic axioms of the theory as in Optimality Theory but can only result from a detailed study of many different languages. Moreover, since many of the arguments to be developed are logic in nature, they cannot be bolstered by a broader empirical base. Thus, German may serve as an example to set the basic conditions for the application of sonority constraints in a declarative approach to phonology. The theoretical observations that I will present in the following may have other consequences in a violability framework, but the observations in itself are equally relevant. 3. In which domain does sonority apply? The SSP in (1) makes two assumptions that have been disputed in the relevant literature because there are specific data in many languages that render them problematic: First, the SSP takes the syllable as a whole as the domain of the application of sonority. Second, the SSP demands the decline in sonority from the peak to the edges to be strictly monotonic. The following example from German challenges both these assumptions. Here, the dotted line above the skeletal tier indicates the sonority contour, based on the hierarchy given in (2). The distinction between the skeletal positions labelled as V, C, and X will be discussed later. 2 Of course, violability plays a role in declarative theories, too, namely when dealing with irregular linguistic items. The crucial point is whether regular forms are allowed to violate a grammatical condition or not.

4 Segments with inherently falling sonority 255 (3) Syllable structure of Strolchs rogue (genitive) σ Onset Rhyme X C C V C C X [ t R c l ç s ] Since both the two final segments are voiceless fricatives, there is a plateau in sonority that violates the SSP. The first two elements of this syllable form a plateau, too, if obstruents are taken to belong to one and the same class of sonority as in the hierarchy in (2). If fricatives are, instead, regarded as being more sonorous than stops (a position taken e.g. by Blevins (1995:211) as a cross-linguistic phonological property and by Vater (1998:143) as a property of German phonology), the problem becomes even more serious because under this view the syllable onset of the example given would show a fall in sonority, a structure which is generally not licensed. One way to solve this problem is to simply allow sonority plateaus at the edges of syllables; cf. e.g. Blevins (1995:210, 236). In German (as well as in other languages), however, most pairs of segments that form a sonority plateau are not allowed at the edges of a syllable. For example, syllables must neither end in two nasals nor begin with a labial fricative followed by another obstruent. Therefore, the concept of a sonority plateau is in need of further constraints that restrict the combinations allowed by the sonority contour to the amount that is observable in the language under consideration. A theory that allows strict monotonicity only is, thus, obviously more restrictive than a theory that allows plateaus. The peripheral elements in syllables that suggest the assumption of sonority plateaus have a marked status anyway. They are not only problematic for sonority but they may also exceed the domain of the maximal syllable. In German, a well-formed syllable has at most five positions on the skeletal tier (cf. Hall 1992:59f., Neef 1996:88 96). Only specific segments may be added to these positions at the edges

5 256 Martin Neef of syllables which are usually termed extra-syllabic. 3 German allows one syllable-initial extra-syllabic position and three syllable-final ones (cf. Neef 1996:95), indicated with ÔXÕ in the diagram in (3). The only elements licensed as extra-syllabic are voiceless coronal fricatives in the initial position and voiceless coronal obstruents in the final positions. Thus, the notion of extra-syllabicity indirectly defines a sub-part of the syllable as a specific domain which may be called the core syllable. 4 Given extra-syllabicity, a theory that restricts sonority constraints to the core syllable does hence not increase in complexity. The idea to restrict the application of sonority constraints to a subpart of the syllable goes back at least to Fujimura & Lovins (1978:112) who restrict their Ôvowel affinity principleõ to the Ôsyllable coreõ (which, however, slightly differs from my concept of a core syllable). Clements (1990:290) cites more authors that follow this line of analysis. In this way, it becomes possible to maintain the requirement of strict monotonicity. In the following, I give a reformulation of the SSP in declarative terms, employing the mode of surface-true constraints and being based on the derived notion of a core syllable: (4) Constraints for the sonority contour of syllables (cf. Neef 1996:65f., 105) a. In the onset of a core syllable, sonority has to rise strictly monotonic. b. In the rhyme of a core syllable, sonority has to fall strictly monotonic. c. The first element of the rhyme has to be of higher sonority than the final element of the onset of the same syllable. 5 These syllable contour constraints may well be universals of natural languages. Crucial differences between languages derive from languagespecific conceptions of the core syllable, covering e.g. different constraints on the minimality and the maximality of the syllable. 4. On which representational tier does sonority apply? If the phonological syllable is construed as a flat structure with segment positions directly connected with the syllable node, it is 3 The established term for this phenomenon is extrasyllabic, to be precise, meaning that there are elements that are located outside a syllable. This manner of speaking stems from derivational phonology (cf. Clements & Keyser 1983:39). I prefer a representational reading of this term as extra-syllabic, meaning that there are elements that are additional to a core syllable but nevertheless located inside a syllable. 4 Kenstowicz (1994:255) defines the term core syllable in quite a similar way, based on a derivational syllabification algorithm. 5 An alternative way of characterising the connection between the rising and the falling part of the sonority contour is the concept of a ÔdemisyllableÕ as proposed by Clements (1990:303).

6 Segments with inherently falling sonority 257 evident that the sonority constraints have to apply on the segmental level. Ever since Clements & Keyser (1983), however, it is standard to distinguish at least two distinct tiers of syllabic representation, namely a segmental tier and a skeletal tier (or CV-tier respectively). Given a conception of this type, it is necessary to decide on which of these tiers the sonority constraints apply. 6 Wiese (2000:259) explicitly deals with this question and decides in favour of the skeletal tier. Other authors like Clements (1990:299), Kenstowicz (1994:255), Walther (1999:72) and Ewen & van der Hulst (2001:120) connect sonority values immediately with elements on the segmental tier. ClementsÕ Core Syllabification Principle (1990:299, 317) explicitly links sonority with the level of segments. In the following, I will give arguments supporting Wiese s view (which may represent the implicit opinion of the majority of researchers). The discussion of extra-syllabicity already suggests that the skeletal tier is the domain for sonority because on this tier only the notion of the core syllable is present. A further argument for this assumption derives from the behaviour of the affricate [pf]. 7 Since affricates are combinations of a stop and a fricative, they constitute two different segments with incompatible phonological features (cf. Broselow 1995:176). Their phonotactic behaviour, on the other hand, is similar to that of single segments. The introduction of the skeletal tier facilitates to treat affricates as monosegmental and bisegmental at the same time: Affricates have two positions on the segmental tier but only one position on the skeletal tier (cf. e.g. Broselow 1995:176, Wiese 2000:40 42). Given this non-isomorphic relation of the representation of affricates with respect to the two tiers, affricates may give a hint as to whether the segmental tier or the skeletal tier is the relevant level for the application of sonority. The following diagram shows that it is sensible to treat affricates as units with respect to sonority and therefore to regard the skeletal tier as the relevant tier for sonority: 8 6 An alternative theory that replaces the skeletal tier in favour of a moraic representation (but retains the segmental tier) does not suffice to explain the sonority contour because onset segments are usually regarded as having no moraic value. Therefore, the sonority contour of onset elements cannot be checked on the moraic tier (cf. Broselow 1995: ). 7 The other affricates of German are less telling because they consist of coronal obstruents which are licensed as extra-syllabic. 8 The distinction between C-positions and V-positions on the skeletal tier is not really necessary if sonority is taken into consideration as a primary feature: The V-position is the one position on the skeletal tier that has the highest sonority value. Therefore, it would be enough to assign both these types of positions the same label as proposed by Levin (1985) (cf. Broselow 1995:184). In any case, these positions inside the core syllable have to be distinguished from extra-syllabic positions.

7 258 Martin Neef (5) Syllable structure of Pfropf stopper σ Onset Rhyme C C V C [ p f R p f ] On the segmental tier, affricates are two distinct units. If the sonority constraints would hold on this tier, the two segments would receive two values for sonority. Under the assumption that stops are less sonorous than fricatives, the sonority contour of the affricate in the onset would rise as demanded by the sonority constraints, but it would also rise in the rhyme, violating the relevant constraint. Under the assumption that stops and fricatives have the same value for sonority (assumed e.g. by Clements (1990:292) as a cross-linguistic property and by Wiese (2000:260) with respect to German), both affricates in example (5) would violate the sonority constraints by forming sonority plateaus. Moreover, neither [p] nor [f] can be analysed as extra-syllabic elements. The typical property of extra-syllabic segments is that they combine freely with other segments, in particular with other obstruents. The labial obstruents under discussion do not combine with most other obstruents in a peripheral position in the syllable. A German syllable must neither begin with, e.g., [pt] or [pv] nor end with [xf] or [tf]. Thus, the elements of the affricate [pf] clearly differ from extra-syllabic segments. A general prediction of the sonority concept is that combinations of segments that can be found in the onset of syllables have the reverse order in the syllable rhyme. Affricates differ from other segmental combinations in not showing this mirror-image. From these considerations I conclude that affricates are units with respect to sonority (as indicated by the dotted sonority line in (5)). Since affricates constitute units on the skeletal tier but two elements on the segmental tier, sonority applies on the skeletal tier only. 9 c 9 This argumentation is not relevant for approaches that substitute the segmental tier by a root node tier since affricates are monopositional Ôcontour segmentsõ on this tier (cf. Clements & Hume 1995:254, 257). Extra-syllabicity, however, is not present on this level of representation. Therefore, it is not straightforward in these types of phonological models to simply declare the root node tier as the tier relevant for sonority.

8 Segments with inherently falling sonority Long vowels as segments with inherently falling sonority Given these premises, it follows logically to assume segments with inherently falling sonority (at least in a theory that banishes violability). Usually, long vowels are analysed as bipositional on the skeletal tier in order to explain why long vowels combine with one element less in a syllable than short vowels. 10 However, this assumption leads to a sonority plateau in long vowels as noted in Neef (1996:93): (6) Inconsistent syllable structure analysis of Mond moon σ Onset Rhyme C V C C X [ m o: n t ] In most analyses that make use of sonority this problem is not mentioned. 11 A solution to this dilemma could be to relate sonority to the skeletal tier in general and to the segmental tier in the case of long vowels only. Because of its theoretical inconsistency, this solution is not truly worth discussing. A more promising way is to simply allow sonority plateaus within (core) syllables. This solution is adopted by Maas (1999:123) and by Walther (1999:66). However, if sonority plateaus are not only allowed at the syllable edges affecting extra-syllabic elements but also within the core syllable, the concept of sonority loses much of its explanational force. This is because most of the potential sonority plateaus in fact do not occur. In German, there is never a tautosyllabic adjacent occurrence of two noncoronal stops, or of two non-coronal fricatives, or of two nasals, or of two non-high vowels. Even the restriction of sonority plateaus to the nucleus domain would not help much because in German, sonorants may be syllabic but they are never allowed to form sonority plateaus. If sonority plateaus were allowed in general, a number of constraints would be 10 These combinatorial properties have already been discussed by Menzerath (1954) with respect to German. 11 Even Blevins (1995:210, 236) who assumes sonority plateaus for tautosyllabic geminates does not relate the concept of sonority plateaus to the problem of long vowels.

9 260 Martin Neef necessary to rule out these non-occurring combinations. If sonority plateaus are forbidden in principle, these regularities are explained. A natural and straightforward solution to the long vowel problem is to attribute the property of an inherently falling sonority contour to long vowels. This means that long vowels which are associated with two positions on the skeletal tier have a higher value of sonority on the first of these positions than on the second one. The exact sonority value of this second position has to be determined by a detailed look at other phonological segments, a task I will perform in the following. (7) Inherently falling sonority contour of long vowels V C long vowels As a consequence of this consistent definition, long vowels do not form sonority plateaus, and the sonority contour constraints given in (4) can be kept as unviolable constraints. In this notation, by the way, there is an explicit connection between the segmental tier, the skeletal tier, and sonority. In Neef (1996:92 et passim), I argued (based on a reflection of the phonological status of the feature length by Vennemann (1982:273ff.)) for the association of tense vowels to two positions on the skeletal tier (not of long vowels). In German, long vowels form a true subset of tense vowels: Tense vowels are long under stress but short elsewhere (cf. Mo bel ÔfurnitureÕ as [Umø:.b l] and the derived form mo blieren Ôto furnishõ as [mø.ubli:.r n]). Lax vowels, on the other hand, are always short. Since the combinatorial properties of vowels do not depend on stress, it makes sense to extend the bipositional analysis to tense vowels. In particular, I assume that vowel length is not phonological at all in German because it does not have the potential of distinguishing meaning (cf. Neef 2004). This assumption allows to explain, for example, the strict regularity in German that full syllables may end in unstressed tense vowels but never in lax vowels by constraining the rhyme of full syllables to a minimum of two skeletal positions. 12 e 12 A prerequisite for this analysis is that both the first vowel in Ma dchen ÔgirlÕ and the final vowel in Oma ÔgrandmotherÕ are tense vowels, a position that has a good tradition in German phonology (cf. e.g. Eisenberg 1998) and that is argued for by Neef (1996:89ff.; 2004). e

10 6. Diphthongs as units If tense vowels have an inherently falling sonority contour, it is natural to ask whether there are other segments with an inherently uneven sonority contour. This should be the case for all segments that are associated with more than one position on the skeletal tier within one syllable. A self-suggesting candidate for this property are diphthongs. Standard generative analyses, however, take diphthongs as combinations of two discrete phonological segments (e.g. Wiese 2000:14). In the following, I will discuss problems of such analyses with regard to sonority and propose an analysis that treats diphthongs as phonological constructions. In particular, I will look at the following three kinds of analyses: (8) Possible analyses of diphthongs a. Diphthongs as combinations of a non-high lax vowel plus a high lax vowel b. Diphthongs as combinations of a lax vowel plus a glide c. Diphthongs as phonological units with a biskeletal representation and with an inherently falling sonority contour The analysis in (8a) is the most simple one with respect to the phonological inventory because it regards the notion of diphthong as a derived one: Diphthongs are combinations of segments that also occur in other constellations. The properties of these segments should therefore be the same as parts of a diphthong compared to any other phonological context. Specifically, diphthongs are tautosyllabic sequences of a non-high lax vowel followed by a high lax vowel. However, since German has three diphthongs [ai ^], [au^], and [ ^] i only13 but three high lax vowels ([i], [y], [U]) and four non-high lax vowels ([E], [{], [ ], [a]), this approach has to explain the restricted combinatorial possibilities of lax vowels. Wiese (2000: ) shows how this task can be carried out in a derivational conception of phonology, leaving open the question how this argumentation can be translated into an output-oriented model of phonology. I will not try to evaluate this analysis in general but merely show that it is not tenable if the concept of sonority is considered. A necessary assumption for an analysis of diphthongs like in (8a) is that high vowels are of lower sonority than other (full) vowels. 14 However, this assumption leaves unresolved the properties of combinations of diphthongs and rhotics: c Segments with inherently falling sonority 261 c 13 A fourth diphthong [Ui ^] is marginal in German and appears in the interjections ui, hui, and pfui only. 14 The marginal diphthong [Ui ^] cannot be explained in this way because it consists of two high vowels.

11 262 Martin Neef (9) Combinations of vowels and diphthongs with rhotics non-high vowel high vowel a. lax vowel plus rhotic starr, Herr, schnorr Ôstiff, master, cadge wirr, Gesurr, du rr Ôconfused, humming, b. tense vowel plus rhotic c. diphthong plus rhotic (imperative)õ Haar, leer, Tor, Sto r Ôhair, empty, gate, sturgeonõ a dryõ Bier, pur, Tu r Ôbeer, pure, doorõ teuer, Feier, Mauer Ôexpensive, party, wallõ The gap in this table is a necessary one because diphthongs end in a high vowel. Hence, a sequence of a diphthong plus a rhotic must not have a non-high vowel preceding the rhotic. Since rhotics may tautosyllabically combine with all lax vowels (cf. (9a)) and with all tense vowels (cf. (9b)), these rhotics should be of lower sonority than all full vowels, either non-high or high. If a rhotic follows a diphthong (cf. (9c)), however, both elements must not be in the same syllable (cf. *[.mau^r.] or *[.mau^.]). This fact does not follow from maximality constraints for the skeletal tier because diphthongs may well combine with other sonorants in the same syllable (cf. fein ÔfineÕ, Maul ÔmouthÕ). This distribution suggests that the segment preceding the rhotic must not be of higher sonority than the rhotic itself. Since free vowels are of higher sonority than rhotics (evidenced by the fact that they are capable of appearing tautosyllabically followed by a rhotic), the second element of a diphthong differs from a vowel. The analysis of diphthongs as combinations of a lax vowel plus a glide in (8b) (cf. e.g. Wurzel 1981:921) solves this sonority problem if glides are taken to be of lower sonority than full vowels but of the same sonority as the vocalic rhotic. The price of this analysis, however, is that it adds a new sound class to the phonological inventory, namely the class of glides which consists of the two distinct phonemes [i] and [U]. This is less of a problem if this sound class can be shown to appear outside of diphthongs as well and may be thus independently motivated. The main problem for this analysis, however, is to explain the restricted combinatorial properties of glides. The persuasiveness of this approach very much depends on the ease with which this latter task may be fulfilled. A straightforward way to solve the problem of combinatorial restrictions is to take diphthongs as fixed patterns belonging to the phonological inventory. This assumption underlies the traditional view of diphthongs and motivates the existence of a specific term for this phenomenon. The analysis in (8c) takes this position. The general idea is that diphthongs are phonological constructions. A construction is a combination of smaller units (in this case phonological segments) with properties that cannot be entirely derived from the properties of the

12 constituting elements. The construction-specific properties of diphthongs relate to sonority, as I will argue for in the following. An advantage of this analysis is that it allows us to assume that the components of a diphthong are segmentally identical to ordinary lax vowels. In general, lax vowels must not appear adjacently because they have the same value of sonority which prevents them from appearing inside one syllable. At the same time, they are associated with one position on the skeletal tier only, which prevents a lax vowel from being the final element in a syllable. Only as diphthongs may two lax vowels be adjacent in a syllable. This constellation presupposes that the second lax vowel of a diphthong has a lower value of sonority than usually. Since this property cannot be derived from the property of lax vowels in free position, diphthongs have to be analysed as phonological constructions. This approach allows us to deal with the quite similar behaviour of diphthongs and long (tense) vowels with respect to both synchronic and diachronic variation. Tense vowels and diphthongs are phonologically identical in that they both occupy two positions on the skeletal tier and in that they both have a falling sonority contour. This similarity may foster the diachronic tendency of tense vowels to become diphthongs, which can be found in many Germanic languages (cf. e.g. Schane 1995). If a diphthong is taken as a phonological unit associated with two positions on the skeletal tier, this second skeletal position may be said to have the same sonority value as a (vocalic) rhotic, thereby explaining the distributional data. This analysis is captured in the following extract from a sonority hierarchy: (10) Sonority hierarchy of vocalic segments Segments with inherently falling sonority 263 lax vowels tense vowels diphthongs vocalic rhotic, schwa Since lax vowels must not combine with diphthongs in the same syllable, the first skeletal position of a diphthong must have the same sonority value as a lax vowel. This analysis allows a straightforward definition of the notions full syllable and reduced syllable: A full syllable is a syllable that has an element of the absolutely highest sonority value as its syllable peak, while reduced syllables have an element of a lower sonority value as their peak. The reason for grouping schwa together with the vocalic rhotic is that both elements must never appear adjacent within a syllable

13 264 Martin Neef (cf. Neef 1996:80). The fact that neither tense vowels nor diphthongs may be divided into two syllables follows from the assumptions made so far: Since the first skeletal position of both elements is of absolutely highest sonority, this position must be the peak of a full syllable; since full syllables have minimally a VC-rhyme in German, the second skeletal position must belong to the same syllable as the first skeletal position. 7. The velar nasal as a biskeletal element A third candidate for a biskeletal analysis is the velar nasal. In sharp contrast to the two other German nasals [n] and [m], this sound must neither follow a tense vowel nor a diphthong (cf. Wiese 2000:229). Derivational analyses attribute this property to a complex underlying representation /Ng/ of the velar nasal, invoking rules of assimilation and of /g/-deletion. In this paragraph, I want to reanalyse the velar nasal in declarative phonology. A straightforward way to deal with the limited distribution of the velar nasal is to assign it two positions on the skeletal tier (either VC or CC). As the maximal full syllable rhyme in German is VCC, the velar nasal does not fit into a syllable together with either a long vowel or a diphthong which both have the skeletal value VC. Since the velar nasal may appear in the rhyme following a lax vowel as in lang ÔlongÕ, pronounced as [lan], it has to have a falling sonority contour associated with its two skeletal positions. As a consequence, velar nasals must not appear in onsets where a rise in sonority is obligatory. In fact, this is another well-known property of the velar nasal. Only in ambisyllabic position may the velar nasal appear in an onset. A consonant is ambisyllabic if it immediately follows a syllabic lax vowel and if there is no other segment preceding the next syllable peak. This is compatible with the velar nasal s sonority contour if the two skeletal positions are split up on the two relevant syllables. The following diagrams illustrate the properties of the analysis proposed: (11) Syllable structure of Rings ring (genitive) and Ringe ring (plural) a. σ b. σ σ Onset Rhyme Onset Rhyme Onset Rhyme C V C C X C V C C V [ R I s ] [ R I ] e

14 Segments with inherently falling sonority 265 If the velar nasal occupies two positions on the skeletal tier, only extrasyllabic elements may follow this phone in the same full syllable, a prediction that is borne out by the facts. However, there is one exception to the generalisations stated so far which poses a severe problem for the analysis given: The distributional restrictions of the velar nasal in the syllable rhyme do not hold if the velar nasal is followed by a velar stop. In this context, the velar nasal may well follow both a tense vowel and a diphthong, and a subsequent velar stop (which is not licensed as extra-syllabic) fits into the rhyme if the velar nasal follows a lax vowel: (12) Monoskeletal velar nasal a. den Kreis [{de:n}{krai ^s}] (colloquial) Ôthe circle (acc.)õ b. eingreifen [{ai ^N}{grai ^.fn }] (colloquial) Ôto meshõ U c. Bank [bank] ÔbankÕ While the velar pronunciation of the nasal is obligatory for all stylistic levels for constellation (12c), the nasal s pronunciation is coronal in explicit pronunciation in (12a and b) because on this stylistic level, assimilation is restricted within the domain of the phonological word (indicated by curved brackets). However, a velar pronunciation of the nasal is not ruled out in these cases in general as witnessed by the colloquial data. The examples in (12) represent cases of surface assimilation, indicated by the velar nasal being followed by a velar stop (either voiced or voiceless). Thus, the velar nasal behaves as a simple element in this context. Only if the velar nasal is not subject to surface assimilation, is a bipositional analysis adequate. The close connection of the monopositional and the bipositional version of the velar nasal is especially obvious in Northern-German dialects where a word-final velar nasal is in free variation with a sequence of a velar nasal plus a velar stop (e.g. Ring ÔringÕ as either [rin] or [rink]). Consequently, in the sonority hierarchy in (13) I distinguish two velar nasals, depending on the surface phonological context. While as a default the velar nasal is biskeletal (and shows an inherently falling sonority contour), there is also a marked case, namely if the velar nasal is followed by a velar stop. In this constellation only the velar nasal is monoskeletal. 8. Summary and prospect As a result of the preceding argumentation, I propose a sonority hierarchy for German that differs from all preceding proposals 15 by assuming segments with inherently falling sonority: 15 A survey of 13 different sonority hierarchies explicitly proposed for German is given by Neef (2002:39).

15 266 Martin Neef (13) Sonority hierarchy of German decline in sonority lax vowels tense vowels diphthongs vocalic rhotic, schwa [R], [1] [m], [n], [ ] obstruents, [ ]/_ [G] 16 affricates Concerning the categorisation of the other sounds, this hierarchy does not differ crucially from standard assumptions. The only striking feature is that rhotics are given two different sonority values, depending on whether the rhotic is vocalic or consonantal (cf. Neef 1996:74). This step is motivated by the fact that rhotics are allowed to appear as adjacent pairs within a syllable (cf. the second syllable in Lehrer ÔteacherÕ as [le:.r ]). 17 While the sonority hierarchy in (13) suffices to explain many distributional properties of German, some phenomena remain in need of further investigation. Problematic are the consonant clusters [kv], [ps], and [ks] that are allowed to appear as syllable onsets (cf. quer ÔacrossÕ [kve ], Psalm ÔpsalmÕ [psalm] and Xaver (proper name) [ksa.v ]) although they form a sonority plateau according to the hierarchy in (13). A solution to this problem could be the distinction of stops and fricatives with respect to sonority. However, only eight of 48 conceivable onset pairs of a stop and a fricative exist in German. In the syllable rhyme, only the pair [sk] exists, compared to ten constructible combinations (cf. Neef 1996:84). Thus, under this assumption a number of further constraints would have to explain the large number of non-existing stop-fricativecombinations. Furthermore, words like Ferien ÔholidaysÕ and Linguist ÔlinguistÕ both have an unproblematic trisyllabic pronunciation as [Ufe.ri. n] and [lin.gu.uist] respectively, but also a bisyllabic one that raises questions about the status of the phonological segment which corresponds to the second vowel letter in orthographic representation in both examples. If [Ui] in the bisyllabic pronunciation of Linguist were analysed as a rising ^ a a 16 In this formula, the [g] is meant to represent a velar stop underspecified for voiceness, thus either [g] or [k]. 17 The idea that rhotics have different values of sonority is first discussed by Jespersen (1912:191) and later considered (but not implemented in the respective models) by Heike (1992:12f.) and Vater (1992:109). a e

16 Segments with inherently falling sonority 267 diphthong, this would be an example of a segment with inherently rising sonority. References Blevins, J The syllable in phonological theory. The handbook of phonology, ed. J. Goldsmith, Cambridge, MA/Oxford: Blackwell. Broselow, E Skeletal positions and moras. The handbook of phonology, ed. J. Goldsmith, Cambridge, MA/Oxford: Blackwell. Clements, G. N The role of the sonority cycle in core syllabification. Papers in laboratory phonology I, eds. J. Kingston & M. Beckman, Cambridge: Cambridge University Press. Clements, G. N.& Hume, E. V The internal organization of speech sounds. The handbook of phonology, ed. J. Goldsmith, Cambridge, MA/Oxford: Blackwell. Clements, G.N.&Keyser, S. G CV phonology. A generative theory of the syllable. Cambridge/London: MIT Press. Eisenberg, P Grundriss der deutschen Grammatik. Das Wort. Stuttgart/ Weimar: Metzler. Eisenberg, P., Ramers, K. H.& Vater, H. eds., Silbenphonologie des Deutschen. Tu bingen: Narr (¼Studien zur deutschen Grammatik 42). Ellison, M. T The universal constraint set: convention, not fact. Optimality theory: phonology, syntax, and acquisition, eds. J. Dekkers, F. van de Leeuw & J. van de Weijer, Oxford/New York: Oxford University Press. Ewen, C. J.& van der Hulst, H The phonological structure of words: An introduction. Cambridge: Cambridge University Press. Foley, J Phonological distinctive features. Folia Linguistica 4, Fujimura, O.& Lovins, J. B Syllables as concatenative phonetic units. Syllables and Segments, eds. A. Bell & J. B. Hooper, Amsterdam/New York/Oxford: North-Holland. Goldsmith, J. A The handbook of phonological theory. Cambridge, MA/ Oxford: Blackwell. Hall, T. A Syllable final clusters and schwa epenthesis in German. Silbenphonologie des Deutschen, eds. P. Eisenberg, K. H. Ramers & H. Vater, , Tu bingen: Narr. Heike, G Zur Phonetik der Silbe. Silbenphonologie des Deutschen, eds. P. Eisenberg, K. H. Ramers & H. Vater, Tu bingen: Narr. Jespersen, O Lehrbuch der Phonetik. Leipzig/Berlin: Teubner. Kenstowicz, M Phonology in generative grammar. Cambridge, MA/ Oxford: Basil Blackwell. Levin, J A metrical theory of syllabicity. PhD dissertation, MIT. Maas, U Phonologie. Eine Einfu hrung in die funktionale Phonetik des Deutschen. Opladen/Wiesbaden: Westdeutscher Verlag. Menzerath, P Die Architektonik des deutschen Wortschatzes. Bonn: Du mmler. Neef, M Wortdesign. Eine deklarative Analyse der deutschen Verbflexion. Tu bingen: Stauffenburg (¼Studien zur deutschen Grammatik 52). Neef, M Das Maß aller Dinge: Sonorita t. Sprache als Form. Festschrift fu r Utz Maas zum 60. Geburtstag, eds. M. Bommes, C. Noack & D. Tophinke, Wiesbaden: Westdeutscher Verlag. Neef, M Die phonologischen Einheiten des Deutschen. Ms. University of Cologne.

17 268 Martin Neef Ploch, S Metatheoretical problems in phonology with Occam s Razor and non-ad-hoc-ness. Living on the edge. Papers in honour of Jonathan Kaye Berlin/New York: Mouton de Gruyter (¼Studies in Generative Linguistics 60). Schane, S. A Diphthongization in particle phonology. The handbook of phonological theory, ed. J. Goldsmith, Cambridge, MA: Blackwell. Scobbie, J. M., Coleman, J. S.& Bird, S Key aspects of declarative phonology. Current trends in phonology: models and methods, Vol. II, eds. J. Durand & B. Laks, Salford: European Studies Research Institute. (¼Proceedings of the Royaumont meeting 1995). Selkirk, E On the major class features and syllable theory. Language sound structure, eds. M. Aronoff & R. T. Oehrle, Cambridge, MA/ London: MIT. Sievers, E Grundzu ge der Lautphysiologie zur Einfu hrung in das Studium der Lautlehre der indogermanischen Sprachen. Reprint Hildesheim/New York: Olms. Sigurd, B Rank order of consonants established by distributional criteria. Studia Linguistica 9, Vater, H Zum Silbennukleus im Deutschen. Silbenphonologie des Deutschen, eds. P. Eisenberg, K. H. Ramers & H. Vater, Tu bingen: Narr. Vater, H Zur Silbenstruktur im Deutschen. Festschrift fu r Georg Heike, eds. B. J. Kro ger, C. Rieck & G. Sachse, Frankfurt: Hector (¼Forum Phoneticum 66). Vennemann, T Zur Silbenstruktur der deutschen Standardsprache. Silben, Segmente, Akzente, ed. T. Vennemann, Tu bingen: Niemeyer (¼Linguistische Arbeiten 126). Walther, M Deklarative prosodische Morphologie: constraint-basierte Analysen und Computermodelle zum Finnischen und Tigrinya. Tu bingen: Niemeyer (¼Linguistische Arbeiten 399). Whitney, W. D On the relation of vowels and consonants. Journal of the American Oriental Society 8, Reprinted 1971: Whitney on language. Selected writings, ed. M. Silverstein, Cambridge, MA: MIT Press. Wiese, R The phonology of German. Oxford: Clarendon. Wurzel, W. U Phonologie: Segmentale Struktur. Grundzu ge einer deutschen Grammatik, eds. K. E. Heidolph et al., Berlin: Akademie- Verlag. Received August 3, 2003 Accepted April 27, 2004 Martin Neef University of Cologne German Department D Ko ln Germany neef@uni-koeln.de