The structure of the German vocabulary: edge marking of categories and functional considerations(1).

WIESE, RICHARD

Abstract

That phonology can signal the boundaries of constituents has been well known since Trubetzkoy identified the delimitative function of phonology. In this paper, it is argued that such edge marking may be much more systematic than observed in earlier work. In particular, the stress domains (foot, word, compound, phrase) and the morphological categories (root, stem, word) of German systematize their edge marking by a repeating alternation in moving from left to right (stress) or by a switch in feature values for [consonantal] (root, etc.). For the latter preference, quantitative evidence is provided. Such alternating patterns of edge marking are argued to be highly functional. Ways of treating these phenomena by means of a metaconstraint in optimality theory are discussed in the final section.

Introduction: the relevance of the "delimitative functions" or "boundary signals"

In the beginning of phonological theorizing, Trubetzkoy expressed the far-reaching hypothesis that the phonological subsystems of languages were by no means restricted to a system of phonemic contrasts between segments. Besides the distinctive function of sounds carried by phonemes, Trubetzkoy (1967) identified two further functions, which he called delimitativ and gipfelbildend `cumulative'. In this paper, I discuss the role of the so-called delimitative function in phonology, with the aim of arguing that it is actually more central than Trubetzkoy could envisage. Second, a treatment of these regularities in terms of optimality-theoretic constraints will lead to the conclusion that constraints enter specific relations with each other, hitherto unnoticed.

The quote in (1) makes it clear that Trubetzkoy, although he clearly saw the importance of phonological delimitation, still claimed that the distinctive function has priority and is the only one for which a language has a logical necessity. In other words, phonology is there to distinguish German Tisch `table' from Fisch `fish'. In addition, it may (but need not) signal where units start and end in a stream of such units.(2)

(1) Trubetzkoy's notion of delimitative functions

"Diese zwei Schallfunktionen, die distinktive und die delimitative, mussen streng unterschieden werden. Die distinktive Funktion ist fur die Sprache als solche unentbehrlich: die einzelnen Schallkomplexe, die den Bedeutungseinheiten entsprechen, mussen unbedingt verschieden sein, damit sie nicht verwechselt werden; (...) Dagegen ist die auBere Abgrenzung der bedeutungsgeladenen Schallkomplexe gar nicht unbedingt notwendig. Diese Komplexe konnen in einem ununterbrochenen RedefluB ohne jede Andeutung ihrer Grenzen aufeinander folgen" (Trubetzkoy 1967: 241-242).(3)

In the following, I will try to combine the insight from Trubetzkoyan phonology with analytical tools provided by optimality theory: linguistic units may, to a large extent, be signalled by marking their left and/or right edges. More specifically, it is postulated that roots, stems, and words in German are much more pervasively signalled by rather concrete phonological properties than is generally assumed, even by Trubetzkoy. Furthermore, when moving from one linguistic level to the next in a hierarchy of linguistic categories, we often find a switch in the value of the feature under discussion, the features being segmental and/or prosodic. This observation, if correct, bears consequences for a theory of constraints, such as optimality theory: constraints are mutually constrained.

I present two types of evidence for these claims. First, it will be argued (as in Wiese 2000 [1996]: 311) that there is indeed a general pattern such that linguistic units are consistently marked on the edges, and that the direction of marking reverses from left to right and back again, when we go through the relevant hierarchy. This is demonstrated by looking, in section 1, at the stress patterns of German. Disregarding cases interpreted as exceptional or marked in the various analyses, the basic generalization is that the main stress switches in its orientation from one linguistic level to the next. In the main part of this article, in section 2, I argue that German roots, stems, and words are surprisingly systematically marked by a contrast of [consonantal] vs. its absence at edges of roots, stems, and words.

1. From right to left in stress patterns

German is a stress-timed language, which, even if this classificatory term has no further function, makes it easy to identify stress patterns at various linguistic levels. There are numerous studies of stress in German (see the thorough survey by Jessen 1999), which disagree in many details. However, it is clear that the units given in (2) are all relevant for the description of stress. In other words, we need to talk about COMPOUND stress, because there is good evidence that it differs systematically from, say, PHRASAL stress and WORD stress.

(2) Stress domains:

Phrase

Compound

Word

Foot

Next, we need to identify the basic stress pattern of the units given in (2). The generalizations are not always uncontroversial, but in most cases, a rather clear picture emerges. As shown in (3), German has left-strong feet, which are sometimes seen as syllabic (Giegerich 1985; Hall 1992; Wiese 2000 [1996]), and sometimes as moraic (Alber 1998; Fery 1998). The major point under debate thus is the precise nature of the foot: the foot is left-strong. Moving up the hierarchy, we find words consisting of several feet, like those in (3b). It turns out that it is the final (rightmost) foot in these cases that bears word stress. There are problematic cases for this analysis, for example in trisyllabic words such as Sansibar, Talisman, etc., but the generalization still seems to be valid. It partly explains that word stress is predominantly on the final or the penultimate syllable, and that word stress is always found on one of the three final syllables. Strong units are italicized in (3).

Turning to compounds, we find that binary compounds ([A B]) predominantly have initial stress, again with exceptions for which no completely satisfactory analysis exists (see, e.g., Giegerich 1985; Benware 1987). Finally, phrases predominantly bear final stress, at least if not under special focus conditions. The classical stress rules for English, the compound rule and the nuclear stress rule of SPE (Chomsky and Halle 1968: chapters 2, 3) apply equally well (or badly) to German.(4)

(3) Stress directionality in German:

a. Feet:
[Hau ser], [Ko nig] (syllables within syllabic feet)
[La ti] [tu di] na [ris] mus (morae within feet)(5)
b. Words:
[Symme trien], [Uni versi taten] (feet within [phonological] words)
c. Compounds:
[Fahr zeug], [Streich holz] (words within compounds)
d. Phrases:
[Ottos Geburtstag], [alte Freunde],
[heute abend], [auf dem Tisch] (words within phrases)

The crucial and, to my knowledge, new observation here is that stress alternates between the right and the left edge from one stress unit to the next higher or lower one, a generalization summarized in (4). This alternation is completely systematic, only perturbed by the various exceptions and provisos mentioned above with respect to the individual stress patterns.

(4) Accentual units and orientations of main stress:

Phrase [right arrow] right

Compound [right arrow] left

Word [right arrow] right

Foot [right arrow] left

Stress was discussed by Trubetzkoy as an instance of the "delimitative function" of phonology. Presupposing stress-relevant structures that are always or at least preferably binary, stress can only highlight either the right or the left edge of the domain. If stress is edge-related in this way, it serves to mark boundaries of units. Consider now a stress system in which stress for all units is uniform with respect to its left/right orientation. Such a system would not be very efficient in terms of its delimitational power: edges would be indicated through stress, but there would be no way of expressing the TYPE of unit whose edge is indicated by stress. The functional logic of a system such as the German one is now obvious: in contrast to the hypothetical system just discussed with uniform stress marking, the type of unit is in fact quite consistently indicated through the position of its main stress. If a hearer encounters a linguistic form of the shape [a B] (with capitalization indicating main stress), she can conclude that the form is either a phrase or a word, but not a compound. (The conclusion may be wrong, but it will constitute a good guess.)

One might speculate whether the pattern can be extended to lower as well as higher categories. The next lower category below the word is the syllable. Remarkably, in an analysis of the syllable using a strong-weak labelling for its constituents, Kiparsky (1979) assigns to the syllable a weak--strong pattern, such as [[[onset.sub.w]--[rhyme.sub.s]].sub.[Sigma]]; see also Giegerich (1985). This is once again a switch in direction from the next higher level. Furthermore, as a reviewer of this paper points out, it is possible to say that the next higher level from the phrase is the clause (in the sense of the main verb and its arguments). But it has been well known since Kiparsky (1966) that the stress pattern of such structures is left-strong, as in ein Buch lesen `read a book', auf den Tisch klopfen `knock on the table'. The issue of stress patterns here is difficult (see Cinque 1993 for a syntax-oriented approach), but arguably, there is yet another switch in stress directionality involved here.

The conclusion from these bird's-eye observations on stress patterns is that the stress rules of German refer to each other and do not seem to be independent of each other, although they are always stated in this way. I am not aware of any analysis claiming that, for example, the phrasal stress rule puts main stress on the final constituent because the compound rule behaves in the opposite way.

The term "stress rule" in the preceding paragraphs is intended to be used in a pretheoretical sense. Stress regularities can be expressed by a number of different means. Within optimality theory, they are expressed by means of standard constraints, in particular those that align the prominent daughters (heads) with the left or right edge of their respective mother node (see Prince and Smolensky 1993; McCarthy and Prince 1993a, 1993b, and numerous subsequent literature). A description of German stress by means of constraints is not the topic of the current paper, but see Alber (1997, 1998) and F6ry (1998) for modelling German word stress in OT. I return to a constraint-based description of edge marking in section 3 below.

One advantage of using a constraint-based framework for the point under discussion, namely the switch in left-right orientation over a hierarchy of categories, may be that this aspect is not mandatory. That is, the observation made for German is not without counterexamples in other languages. For Finnish, for example, it seems that both feet and words are left-strong (see Hanson and Kiparsky 1996; Alber 1997). In other words, the left-to-right alternation noted for German and argued to be a highly functional organization of stress patterns is not universal. In terms of optimality theory, a system of violable constraints may be responsible for the situation found. At the same time, optimality theory makes a prediction here: the Finnish-type, one-sided edge marking over adjacent categories can only arise if enforced by some higher-ranked constraint. I can only speculate at this point that a preference for a marking of initial edges is higher than that of marking final edges. The preference for the marking of left edges (as opposed to right edges) makes sense in terms of speech processing, which is generally left-to-right.

2. From vowel to consonant in lexical categories

Pursuing the idea that such edge marking of units with a pervasive reversal of values within a hierarchy is inherent in the linguistic system (of German), I will, second, argue that other categories are similarly marked at their edges, and that units that are vertically adjacent in a hierarchy are marked with the opposite value of some feature. In particular, I present some results of a study in which roots, stems, and words are analyzed and their edge marking is under scrutiny. The study presupposes an arrangement of these categories as in (5), with recursion allowed for the stem (see, e.g., Selkirk 1982; McCarthy and Prince 1993a, 1993b).

(5) Morphological categories:

Word

Stem

Root

A second theoretical background assumption of the study is the proposal made in direct optimality theory (Golston 1996a) that the representation of any linguistic unit is identical to the set of constraints it violates; see (6). According to direct OT, there is no need to see the (partial) representation of linguistic units as separate from the set of constraints that are violated for the particular linguistic unit. In this conception, a particular constraint may be obeyed by a large number of particular entities, but it can still be violated by a minority of them, and violations of this constraint are part of the phonological representation of each of these entities (simple or complex).

(6) Direct OT (Golston 1996a):

Morphemes are represented by the constraints they violate.

It turns out that the constraints used in standard OT are so varied and manifold that violations of them can be used to specify morphemes: classes of linguistic forms and even individual items can be characterized as specific sets of constraint violations. This makes standard representations redundant since constraint violations, which are routinely used to evaluate forms, can also be drawn upon to represent the very same forms. This is the central claim of direct OT, one I will make use of in the following.

Constraints thus penalize lexical items for their violations of particular edge-marking constraints (just as for any other violation of constraints). For present purposes, it is hypothesized that roots are marked by initial and final consonants. In other words, a root beginning or ending in a vowel violates an alignment constraint that requires the opposite. Therefore, such a root (take See `sea') is represented (partially) by this particular constraint violation; see (7). The display in (7) is nothing but a specific elaboration of the bipartite Saussurean sign: a combination of a meaning (abbreviated as SEE) with a form (the constraint violation).

(7) Partial representation for See:

ALIGN-R (root, [cons.]) SEE

2.1. Roots

With this conception of linguistic representation providing the basic analytical tools, Golston and Wiese (1998) look at the edge marking for roots in German. On the basis of a computer-readable, independently derived root dictionary of German (compiled on the basis of the database of roots by Ortmann [1993] containing 6,512 native roots), they find the results displayed in (8).

(8) Edge marking of roots (from Golston; Wiese 1998: 178-179):

 Consonant Vowel
 no. % no. %
Left edge 6147 94 365 6
Right edge 6268 96 244 4

The result of this quantitative survey is that there is a very strong tendency to mark both edges of a German root by the feature [consonantal]. That is, the two constraints ALIGN-L/R (Root, [cons.]) are obeyed by almost all roots. The minority of roots that violate one of the two constraints are characterized by exactly this constraint violation and are in this respect more marked than the unmarked cases (see [9]). Note that only the first result (left edge marked by consonants) is possibly reducible to the onset requirement of syllables, but the second result (consonant at right edge) is in fact unexpected by syllabic requirements. A typical (and good) root such as Tisch `table' is a bad syllable in that it is a closed syllable.

Expressed in terms of standard OT constraints, vowel-initial roots are represented (inter alia) by a violation of (9b), and vowel-final roots by a violation of (9a). As seen in (8), only a minority of native German roots incur these constraint violations.

(9) OT constraints for roots (Golston and Wiese 1998):

a. ALIGN-R (root, [cons.])

b. ALIGN-L (root, [cons.])

The tendency to mark root boundaries by a consonant is actually not restricted to German but is widespread in typologically diverse languages. It is even found for languages that have a strong preference for open syllables, as seen in the Indo-European or the Bantu languages. (For Indo-European roots, see the dictionary by Watkins [1985].) Bantu roots (radicals) are also predominantly consonant-final (see Meeussen 1967; de Blois 1975: 72f.), though Bantu syllables strongly prefer to be open.

To take up a completely different language, Olawsky (1999: 95) presents relevant statistics from the language Dagbani, a Niger-Congo language spoken in Ghana. Olawsky collected 512 nominal roots of Dagbani, of which 71% end in a consonant, although the preference for open syllables is very strong in Dagbani.

2.2. Stems

If roots are marked by [consonantal], what about stems? Stems dominate roots but are less easily identified on the basis of a corpus. While roots are identifiable as some sort of minimal or core morpheme, and while words are the minimal freely occurring units, no such minimality condition appears to work for stems, which are intermediate constituents (see [5]). For current purposes, I therefore rely on existing linguistic descriptions of the stem in general and the stem in German in particular.(6)

If roots are marked by final consonants, stems should be marked by nonconsonants, according to the functional considerations of this study. As a first piece of evidence that this might be the case, I suggest that theme vowels of many languages are primarily stem markers. Their main function is to sort the roots of the language into a set of classes in order to make them available for further derivational and inflectional morphology. Thus, Latin and some of the Romance languages display precisely the edge marking of stems by vowels predicted by the present account. The main point in the present context is that I am not aware of their unpredicted counterparts, namely "theme consonants." But if most languages have more consonants than vowels, why should these not exist? The hypothesis that roots tend to end in consonants, while stems tend to end in vowels, predicts the nonexistence of theme consonants: there is no functional reason to give stems a consonantal ending. Matters change, of course, once we are dealing with a more specific inflectional morpheme carrying other information. In this case, grammarians seem not to use the term "theme vowel/consonant."

As for German, it is possible to argue that final schwa is a stem marker. This proposal was first suggested by Wurzel (1970), who interprets final schwa in most of its occurrences as a STEM marker. In (10), the final subrule is one instance in a complex rule for noun-stem formation. In other words, noun stems are often marked by the vowel schwa in final position.(7) In a way, final schwa is a theme vowel, assigning a substantial number of German nominal roots to a particular inflectional class. The linguistic analysis of words such as Glaub-e `belief', Lieb-e `love', Ros-e `rose', Lug-e `lie' would thus be that they contain a root (consonant-final!) plus a stem-marker schwa, to the effect that these words constitute a noun class. Other stems (not in this class) are identical to roots and thus violate a constraint requiring a stem to end in a vowel. The first two subrules in (10) are further stem-formation rules, regulating more specific noun classes (with fewer members). Clearly, the schwa-insertion rule in the final line is the default rule with respect to the two preceding ones. While the first subrule (adding /r/) derives specific forms, namely the plural of a particular (small) class of nouns, the second one requires specific morphological features that are not needed for the schwa-insertion rule.

(10) Noun-stem formation rule (Wurzel 1970: 47)

[ILLUSTRATION OMITTED]

The number of such schwa-final nouns (noun stems according to the present analysis) is large in present-day German. A count over the CELEX database of German words (Baayen et al. 1993) reveals the following: of 4,174 monomorphematic German nouns found in this database, 1,034 end in schwa, among them many relatively recent assimilated loan words such as Orange `orange' or Fassade `facade'. On the other hand, only 152 of these nouns end in [en].(8) That is, while this type of schwa still occurs in a minority of cases, the minority is sizable. It must also be taken into account that this ending is productive only with feminine nouns.

I draw the conclusion from this independent description that stems of German tend to end in a vowel, namely schwa in the more general case, and that it is one of the major functions of the German schwa in final position to provide stems with a final vowel.(9)

For further evidence that this interpretation of final schwa might be correct, we may note how nominal inflectional markers for nouns are treated by Wurzel in his account. As shown in (11), there are inflectional markers (Flexive), which are always ordered after the Stammbildungselement (stem-formation element). I interpret this ordering as being related to the categories of stems and words: stem-formation elements are attached to stems (hence the name), inflectional markers are attached outside of those, at the word level. Examples in the final column of (11) contain both stem-formation elements (r, n) and inflectional markers (n, e, plus s from the preceding column).(10)

(11) Difference betweeen stems and words (Wurzel 1970: 49):

ohne Affix Stammbildungs- Flexiv Stammbildungselement
 element und Flexiv

kind ?? ?? - kind ?? + s kind + r + n
das Kind des Kindes den Kindern
bar ?? ?? bar + n ?? - bar + n + e
der Bar des Baren die Baren
- stras + e ?? - stras + n + n
 die StraBe den StraBen
- end + e ?? - end + n + e
 das Ende die Enden
tank ?? ?? - tank ?? + s -
der Tank des/die Tanks

Note now that the inflectional markers are always consonants -- with the exception of e, which requires some discussion. Wurzel postulates a final schwa (again written as e) for some of the nominal paradigmatic classes, but this segment is always deleted afterward. That is, the surface form of assumed bar + n + e is invariably [b?? Ren) (Baren). In other words, to postulate this final schwa as an inflectional marker is in spirit with the theoretical approach of the time (SPE framework) but is in strong contrast to the surface structure and therefore highly dubious on theoretical grounds. A unit is first introduced and then deleted without a trace (similarly for + n + n in the final column). If we therefore neglect this item, we can conclude that the set of word-related inflectional markers postulated in Wurzel (1970) for the German noun system invariably consists of consonants, specifically /s/ and /n/.

2.3. Words

We have now moved upward from the stem to the word and have seen some very tentative evidence that the words might be marked by consonants in inflectional paradigms of nouns. But for words, further independent evidence for the claim is available. I suggest that proper nouns (names) are distinct from common nouns in that they are always words, but not stems or roots. The evidence for this is morphological: names cannot be inflected in the way common nouns can be. In accordance with the analysis of inflectional suffixes above, which separates stem-related suffixes from word-related ones, it is indeed the case that proper names can only be suffixed by the word-related suffixes, namely genitive singular, or plural /s/, see Peter-s (gen. sg. or pl.).

Furthermore, we can ask again how the edges of names are marked by looking at a corpus of such names. As one first result, in (12) the number of vowel-final and consonant-final second names in an electronic telephone directory for the city of Rostock containing more than 50,000 names is tabulated.(11)

(12) Edge marking of words: last names

 Consonants Vowels
 no. % no. %

Right edge 42421 85.2 7352 14.8

The result is that a large majority of last names (85%) end in a consonant. Before discussing the relevance of this finding, a second set of figures can be added: (13) presents a quantitative summary for a list of German first names, this time from a smaller database found on the internet.

(13) Edge marking of words: first names

 Consonants Vowels -- total Vowels -- a
 no. % no. % no. %

Right edge 194 63.8 110 36.2 69 22.7

Here, the result is blurred somewhat by the fact that many first names for females end in -a which can actually be seen as a (borrowed) suffix deriving female names, such as Martin-Martin-a, Josef--Josef-a, and numerous others. Subtracting the female names ending in -a from the total, the result is that there are only 13.5% of vowel-final names. Again, the observation is confirmed: names are predominantly consonant-final.

However, these results from given names and family names must be held against some consideration of what would be expected by chance. Suppose, in a first approximation, that consonants and vowels are distributed randomly over words, constrained only by hard phonotactic constraints, such as the effect of the German rule of final devoicing, that is, without any preferences for particular sounds or classes of sounds. If this distribution of consonants vs. vowels should be indistinguishable from those found in (12) and (13) above, we would have no reason to argue that words preferably end in consonants. Finding such a chance distribution is not trivial, but one possible approach is presented here. One possible count (presented in [14]) then reveals that 14 possible consonantal phonemes can appear word-finally, while 13 possible vowel phonemes can appear in the same context. Two affricates, /pf/ and /ts/, were counted separately here, as well as three diphthongs. The two major classes of phonemes excluded are the voiced obstruents and the short, lax vowels, neither of which occur word-finally according to most descriptions. Alternative classifications of phonemes are always possible, of course, but would not dramatically change the picture. In particular, if we add both classes of nonoccurring sounds, namely voiced obstruents and short vowels, the situation is basically the same again.

(14) Chance distribution (phonemes in word-final position):

Consonants Vowels
no. % no. %

14 51.9 13 48.1

A possible objection to this way of calculating a chance distribution is that the types and qualities of consonants and vowels (and thus their respective numbers) are in fact irrelevant. This would lead to the simpler null hypothesis that words could end in either an open syllable (i.e. final vowel) or a closed syllable (i.e. final consonant). With this binary split (a word can only end in a closed or in an open syllable), final consonants and vowels would have a completely even probability of occurring. For German, this distribution is in fact very close to the one found in (14), and again, it does not match the factual situation.

The empirical results reported above for first and last names are expected not by the chance distribution based on the numbers of consonants and vowels, nor by the chance distribution of open and closed syllables, nor by the requirements of syllabic well-formedness, which would give vowel-final words a clear majority. The general picture found in the data is again surprisingly uniform and can be sketched as in (15): for the right edge, there is a switch from consonant to vowel and back to consonant when we move from roots to stems to words in German.

(15) Edge marking of categories:

Word [right arrow] Consonant

Stem [right arrow] Vowel

Root [right arrow] Consonant

As in the case of roots and stems, we can easily state optimality-theoretic constraints to express the findings of words; see (16) for one possible attempt. Of these, the left-alignment constraint in (16b) was not justified above. However, as there is no mechanism at all, in either the phonology or the morphology of German, that shows a tendency to form initial vowels, we may actually assume for the time being that it holds.

(16) OT constraints for words:

a. ALIGN-R (word, [cons])

b. ALIGN-L (word, [cons])

The alignment schema for constraints provides an obvious mechanism for the description of all the facts enumerated so far. More important, however, is the functional view of phonological structures made possible by these results: phonological features mark the boundaries of linguistic units and thereby fulfill the demarcative function first proposed by Trubetzkoy (see [1]). It must be added, though, that the proposal does not imply that words are identical to roots. This is obviously not the case: if roots are generally monosyllabic, words tend to be polysyllabic. Also, the word-final consonants are often a subset of those found for roots (see Golston 1996b on Sanskrit and Greek).

The description of rightmost edge marking for roots and stems given above allows for a straightforward prediction: suppose it is true that roots are generally marked with a consonant at their right edge, as the evidence strongly suggests. Constraint (9a) then holds. If, in addition, the switch from consonant to vowel in moving up the morphological hierarchy to the stem is the preferred move, then we have all the grounds needed for a prediction: we should be able to find evidence for a constraint (17a), but there could be no constraint (17b). In fact, constraint (17a) is something like a derived constraint, one that does not need to be stated independently, while (17b) cannot be thus derived.

(17) Predictable and unpredictable constraints

a. ALIGN-R (stem, [-cons])

b. ALIGN-R (stem, [cons])

There is the additional aspect that the constraint in (17a) is not even statable if a feature model with a unary (not binary) feature [consonantal] is adopted. Suppose that only [consonantal] exists, but [-consonantal] does not. Introducing a katathetic vowel for stems could not count as a violation of feature insertion since no such feature is available. Is there evidence that the predictions just made ([17a] is either a derived constraint or unstatable, while [17b] is disallowed) are true? As we have seen in the preceding section, theme consonants do not seem to exist, while theme vowels are widespread. (In the perspective of the present paper, noun-final schwa in German may well be called a theme vowel.) Furthermore, a notion developed in the minimalist theory of Chomsky (1995) may be used to ensure that not all the structural categories available in morphology are actually present. Consider a simple, and consonant-final, root of German and its structure, as in (18).

(18) Minimal structures for simplex roots

Word

Stem

Root

[ti]

While the structural categories presented here for this item (root, stem, word) are those admitted by the system introduced in (5), it is possible that not all of them are actually needed. In particular, an intermediate node such as stem that does not branch may well be pruned from the structure (see Ito and Mester 1998: section 5 for a similar argument with respect to German compounds). In other words, for the simplex word Tisch only the root and the word are visible, and constraints referring to the stem are irrelevant, as this category is invisible or virtually nonexistent. If such a view can be defended, the need for linguistic items to obey all constraints (an impossible achievement anyhow) is greatly reduced.

3. Constraints on constraints

There is a more general conclusion to be derived from the analyses presented in this paper: constraints as conceptualized in optimality theory are not necessarily independent of each other.(12) Rather, the existence of one constraint can impose a heavy restriction on the existence of some other constraint. This interaction of constraints within the grammar will be discussed in this final section.

Note first that the findings summarized in (4) for stress and (15) for consonants and vowels at morphological edges can be captured somewhat abstractly by a principle of contrast. I propose a formulation of such a principle in (19), assuming that such a principle is indeed more general than our two examples have shown. As contrast is surely one of the most deeply rooted notions in phonology, such a principle makes a lot of functional sense.(13)

(19) Principle of contrast (CONTRAST-P):

For any two categories A and B that are syntagmatically or paradigmatically adjacent:

If category A bears a phonological property P, B should NOT bear P.

Crucially, the formulation of CONTRAST-P generalizes over structures that are either syntagmatically or paradigmatical (hierarchically) arranged. Therefore, we note that this principle is akin to a notion well known in recent phonology, namely the principle of obligatory contour (OCP; see Leben 1973; McCarthy 1986; Odden 1986, and others). The OCP demands contrast, or, rather, disallows the absence of contrast; see the schematic example for a situation disallowed by the OCP in (20). This constraint says that two units A, B should not bear the feature [Labial] twice, at least not if the two units are adjacent on their tier and are not parts of different linguistic units.

(20) OCP as an instantiation of CONTRAST-P:

*A B

[Lab] [Lab]

The close similarity between the two German cases of paradigmatic contrast discussed above and the OCP is only blurred by the fact that OCP cases typically involve syntagmatic relations (like the syntagmatic adjacency between A and B in [20]), whereas CONTRAST-P, see (19), generalizes to all types of adjacency, syntagmatic or paradigmatic.

But what is the status of the principle formulated in (19)? Several options are available: CONTRAST-P could be a metaconstraint, a principle regulating the relationship between two or more constraints, or it could be a standard constraint, of the OCP type, or it could be something completely different. What I would like to suggest here is that the current evidence suggests that a metaconstraint is indeed operative here.(14) Consider again the possible OCP constraint disallowing two adjacent labial segments (or other structures). What the formulation in (20) hides is the fact that each of the two labial elements is already a constraint violation. In optimality theory, each occurrence of a feature in a representation counts as the violation of the constraint militating against its occurrence, in the most general version the constraint *STRUC. That is, an OCP constraint such as (20) already contains reference to two independently needed constraints. More generally, CONTRAST-P, (19), refers to two independent constraints, which of course may entertain different rankings in the language-specific hierarchy. (21) is an attempt to reformulate CONTRAST-P in this fashion.

(21) Principle of contrast as a metaconstraint

CONTRAST-P (CAT-A, P; CAT-B, NON-P)

Again, given that the marking of category A by some phonological property P is expressed by means of a constraint (schema), the constraint militating against the marking of category B by the same property is a metaconstraint regulating the relationship between two constraints. Furthermore, the formulation in (21) adopts the view that metaconstraints must be included into the set of constraints that are operative in a grammar.

Ito and Mester (1998) propose to interpret the OCP by means of more elementary, independently needed, mechanisms. They develop a formalization of the OCP in terms of constraint conjunction in the sense of Smolensky (1995). For example, the ban against two labials as in (20) is expressed, according to their proposal, as the local conjunction of the constraint against the feature labial with itself, crucially within a given domain (morpheme, word, etc.). The cases discussed in the present paper are different as they deal with dissimilarity BETWEEN domains. I do not see at present how these cases can be expressed via local constraint conjunction. Rather, a treatment in terms of a metaconstraint seems necessary.

The right-to-left switch observed for German stress patterns can equally well be captured by a principle of contrast. In current OT, edge-related stress is captured by an alignment of strong elements (heads) to a left or right edge of their mother constituent; see (22) for a constraint schema of this sort, with PRC as a variable over prosodic constituents (cf. EDGEMOST; Prince and Smolensky 1993).

(22) L/R alignment

ALIGN (PRC, L/R, HEAD (PRC), L/R)

The CONTRAST-P, (21), would require that the value for the edge parameter in this schema is not identical for any two vertically adjacent constituents. Left and right will alternate, as seen above for German stress.

4. Conclusions

The main point of this paper is that left-right asymmetries (see stress) and vertical asymmetries (see the switch in feature values between morphological levels) are pervasive. The presupposition is that the lexical categories as outlined in (2) for stress domains and in (5) for morphological domains exist to give structure to the (German) vocabulary. The more structure these categories give to the lexical items, the less the vocabulary is just a long list of arbitrary items. One empirical conclusion from the studies presented above is that lexical categories in German are rather strongly signalled by boundary signals, in particular by consonants and vowels. This fact can easily be overlooked in linguistic frameworks that have no means of integrating quantitative evidence of the type adduced above. But once a more direct connection between markedness and representation is made, as in direct OT, patterns emerge. It should be added here that the patterns found for German do not seem at all restricted to this language. That roots of Indo-European languages generally behave in this way ([cons] initially and finally) is rather well known; that it is also true for typologically or genetically distant languages such as Bantu languages is less obvious, but equally true.

In the present bird's-eye studies of German stress and of edge marking for morphological or lexical categories, we have seen that phonology is functional and delimitative, to a larger extent than the functional view of Trubetzkoy and Jakobson could foresee. A wider perspective for these considerations would consider whether phonemic representations, which are implicitly or explicitly conceived as strings of segments, can be completely reformulated by means of notions of delimitation. In direct OT, such a move is quite natural. As Golston (1996a) demonstrates, a representational account relying solely on constraint violation is possible and does away with most, if not all, of the linear order of phonemic representations as strings of segments. In any case, a representational theory replacing arbitrary strings of segments by edge marking can express the functional significance of the asymmetries noted in this paper more directly.

Notes

(1.) The ideas expressed in this paper were first presented at a colloquium celebrating the 60th birthday of Dieter Wunderlich in Dusseldorf. I also thank audiences at colloquia in Marburg and Santa Cruz, California, for their input, and Birgit Alber and Susanne Niedeggen-Bartke, Marburg, and Chris Golston, Fresno, for their contributions. Finally, two anonymous reviewers helped to improve this paper. Correspondence address: Institut fur Germanistische Sprachwissenschaft, Philipps-Universitat Marburg, D-35032 Marburg, Germany. E-mail: wiese@mailer.uni-marburg.de.

(2.) Trubetzkoy's phonology can be characterized as a functional approach largely because it identifies the three above-mentioned functions of sound structure and views sound structure in relation to these functions. The cumulative function is completely ignored in the following.

(3.) "These two functions of sound, the distinctive and the delimitative, must be strictly distinguished. The distinctive function is a necessary ingredient of language: the individual complexes of sound that correspond to units of meaning must of necessity be kept distinct from each other so that they will not be confused. (...) On the other hand, the external boundaries of meaningful complexes of sounds are not in any way necessary. It is possible for these sound complexes to follow each other in an uninterrupted flow of speech without any hint at their boundaries" (translated by the author).

(4.) I assume that the existence of the categories in (3) is established prior to (and partly independent of) the stress rules. Otherwise, as a reviewer suggests, there would be the danger of circularity: units and stress rules motivate each other. While one of the main points to be made here is that stress alternates because of its function of marking linguistic units of a particular type, stress is not absolutely necessary to differentiate between, for example, compounds and phrases. This is witnessed by the fact that exceptions exist in both directions: German has right-strong compounds and left-strong phrases. The distinction between such compounds and phrases can still be made, for example on the grounds of separability on the syntactic level. In general, there is a (hopefully converging) set of syntactic, morphological and phonological regularities for the existence of the units under discussion.

(5.) Example taken from Alber (1998: 117).

(6.) Very little discussion can be found on the stem category in German. Neef (1996: section 2.2) uses the stem in a sense different from the one used here, namely as the formal representation of any lexical item.

(7.) The vowel is given as /e/, but there is a further rule in this account reducing this vowel to schwa. In general, Wurzel's account is very much within the spirit of SPE (Chomsky and Halle 1968). My argument is based on the assumption that a number of insights given in this work hold independently of the formal framework in which it is couched.

(8.) The search was performed over the so-called stem lexicon in CELEX. The resulting lists of forms reveal some errors (for example with respect to segmentation in morphemes). As these errors probably dc, not introduce a bias, they were not corrected.

(9.) There is a long-standing uncertainty whether to treat final schwa in German nouns as a morpheme or not. Wurzel's proposal basically is that it is a minimal morpheme marking nothing but the category of its base as a noun stem.

(10.) It is not clear to me why the stem marker r (as in kinder, `child, p1.') is not listed in the second column of (11).

(11.) Data were taken from the computer-readable telephone directory D-Info 1996. Given that the names are in orthographic form, there are some uncertainties involved, for example for foreign names. However, the quantitative impact of these problematic cases is negligible. Counting was in terms of tokens and not types here, in contrast to the immediately following data.

(12.) Of course, in OT, constraints also interact with each other in that they enter constraint ranking. In this other sense, they are never independent of each other.

(13.) There is another principle of contrast, as used by Clark (1987, 1993) and Carstairs-McCarthy (1994), which postulates that any two forms should differ in meaning. The present principle, instead, requires contrast WITHIN a single form. This is not to say that there may not be a deeper connection between the two observations.

(14.) McCarthy and Prince (1995: 364-365) introduce another example. They argue for a metaconstraint in which the faithfulness of roots is higher ranked than the faithfulness of affixes. The task of this metaconstraint is to fix the, otherwise variable, ranking between two constraints. In the type of metaconstraint discussed here, the function of the metaconstraint is different: it fixes feature values in an otherwise open constraint schema.

References

Alber, Birgit (1997). Quantity sensitivity as the result of constraint interaction. In Phonology in Progress -- Progress' in Phonology, Geert Booij and Jeroen van de Weijer (eds.), 1-45. HIL Phonology Papers III. The Hague: Holland Academic Graphics.

--(1998). Stress preservation in German loan words. In Phonology and Morphology of the Germanic Languages, Wolfgang Kehrein and Richard Wiese (eds.), 113-141. Tubingen: Niemeyer.

Baayen, Harald R.; Piepenbrock, Richard; and van Rijn, H. (1993). The CELEX Lexical Database [CD-ROM]. Philadelphia: Linguistic Data Consortium, University of Pennsylvania.

Benware, Wilbur A. (1987). Accent variation in German nominal compounds of the type (A(BC)). Linguistisehe Berichte 108, 102-127.

Carstairs-McCarthy, Andrew (1994). Inflection classes, gender, and the principle of contrast. Language 70, 737-788.

Chomsky, Noam (1995). The Minimalist Program. Cambridge, MA: MIT Press.

--; and Halle, Morris (1968). The Sound Pattern of English. New York: Harper and Row.

Cinque, Guglielmo (1993). A null theory of phrase and compound stress. Linguistic Inquiry 24(2), 239-297.

Clark, Eve (1987). The principle of contrast: a constraint on language acquisition. In Mechanisms of Language Acquisition, Brian Mac Whinney (ed.), 1-33. Hillsdale, NJ: Erlbaum.

--(1993). The Lexicon in Acquisition. Cambridge: Cambridge University Press.

de Blois, Kornelis F. (1975). Bukusu Generative Phonology and Aspects of Bantu Structure. Tervuren: Musee Royal de L'Afrique Centrale.

Fery, Caroline (1998). German word stress in optimality theory. Journal of Comparative Germanic Linguistics 2, 101-142.

Giegerich, Heinz J. (1985). Metrical Phonology and Phonological Structure. German and English. Cambridge: Cambridge University Press.

Golston, Chris (1996a). Direct optimality theory: representation as pure markedness. Language 72(4), 713-748.

(1996b). Prosodic constraints on roots, stems and words. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 172-193. Berlin: Akademie-Verlag.

--; and Wiese, Richard (1998). The structure of the German root. In Phonology and Morphology of the Germanic Languages, Wolfgang Kehrein and Richard Wiese (eds.), 165-185. Tubingen: Niemeyer.

Hall, Tracy Alan (1992). Syllable Structure and Syllable-Related Processes in German. Tubingen: Niemeyer.

Hanson, Kristin; and Kiparsky, Paul (1996). A parametric theory of poetic meter. Language 72(2), 287-335.

Ito, Junko; and Mester, Armin (1998). Markedness and word structure: OCP effects in Japanese. Unpublished manuscript, Santa Cruz.

Jessen, Michael (1999). Word stress in Germanic languages: German. In Word Prosodic Systems in the Languages of Europe, Harry van der Hulst (ed.), 515-545. Berlin and New York: Mouton de Gruyter.

Kiparsky, Paul (1966). Uber den deutschen Akzent. Studia Grammatica 7, 69-98.

--(1979). Metrical structure assignment is cyclic. Linguistic Inquiry 10, 421-441.

Leben, William R. (1973). Suprasegmental phonology. Unpublished Ph.D. dissertation, MIT.

McCarthy, John J. (1986). OCP effects: gemination and antigemination. Linguistic Inquiry 17, 207-263.

--; and Prince, Alan S. (1993a). Prosodic Morphology I. Constraint Interaction and Satisfaction. New Brunswick, NJ: Rutgers University Center for Cognitive Science.

--; and Prince, Alan S. (1993b). Generalized alignment. Yearbook of Morphology 1993, 79-153.

--; and Prince, Alan S. (1995). Faithfulness and reduplicative identity. In University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory, Jill Beckman, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.), 249-384. Amherst, MA: Graduate Linguistics Student Association.

Meeussen, Alfred E. (1967). Bantu grammatical reconstructions. Africana Linguistica III, Ann. Ling. 61, 79-121.

Neef, Martin (1996). Wortdesign. Eine deklarative Analyse der deutschen Verbflexion. Tubingen: Stauffenburg.

Odden, David (1986). On the role of the obligatory contour principle in phonological theory. Language 62, 353-383.

Olawsky, Knut (1999). Aspects of Dagbani Grammar. With Special Emphasis on Phonology and Morphology. Munich: Lincom Europa.

Ortmann, Wolf Dieter (1993). Kernmorpheme im Deutschen. 4561 allgemeinsprachliche deutsche Morphemgestalten im Vergleich zwischen zehn Quellen. Munich: Goethe-Institut.

Prince, Alan S.; and Smolensky, Paul (1993). Optimality Theory. Constraint Interaction in Generative Grammar. New Brunswick, NJ: Rutgers University. Boulder: University of Colorado.

Selkirk, Elisabeth O. (1982). The Syntax of Words. Cambridge, MA: MIT Press.

Smolensky, Paul (1995). On the internal structure of the constraint component Con of UG. Unpublished manuscript.

Trubetzkoy, Nikolaj S. (1967). Grundzuge der Phonologie, 2nd ed. G6ttingen: Vandenhoeck and Ruprecht.

Watkins, Calvert (1985). The American Heritage Dictionary of Indo-European Roots. Boston: Houghton Mifflin.

Wiese, Richard (2000 [1996]). The Phonology of German. Oxford: Oxford University Press.

Wurzel, Wolfgang Ullrich (1970). Studien zur deutschen Lautstruktur. Studia grammatica 8. Berlin: Akademie-Verlag.

RICHARD WIESE Philipps University, Marburg

Received 11 July 1999 Revised version received 13 September 2000