Crossing Morpheme Boundaries in Dutch

Marc van Oostendorp

1. Introduction

Not all affixes behave in the same way. Prefixes in a given language may behave quite differently from suffixes for instance. Asymmetries of this type are encountered in many, if not all, languages of the world. They have been mentioned sometimes in the literature, but they have never received systematic attention, let alone a thorough linguistic interpretation, as far as I am aware.

Also particular classes of suffixes (or prefixes) may show different types of behaviour. For many Germanic languages, for instance, it has been claimed that we should distinguish between so-called Class I and Class II suffixes (or between morpheme boundaries + and #, or between lexical levels I and II). In many cases, the two classes of suffix have completely different shapes. For instance, class I suffixes are typically vowel-initial and at most monosyllabic, whereas class II suffixes often are consonant-initial and have more material than fits in one syllable. The issue arises wether we should set up morpheme structure constraints to account for these differences, or rather we should make the morphological status from the phonological form.

My goal in this paper is to show that morphological diacritics are mostly unnecessary. Differences in phonological behaviour of different morphemes can be mostly derived from the underlying phonological shape of these morphemes, provided we have a theory of violable constraint interaction such as Optimality Theory (Prince and Smolensky 1993).

One particular difference between prefixes and suffixes is their respective behaviour in syllabification. In this paper I argue that this difference can be reduced to an inherent assymmetry in syllable structure: the fact that onsets appear on the lefthand side of the vowels heading syllables, not on their righthand side. My data will be taken from Dutch. (Cf. Van Oostendorp to appear for discussion of the data in Italian and Kihehe). The difference between Class I and Class II suffixes, on the other hand, follows from the fact that suffixes in the former class all start with a vowel, while those in the second class start with a consonant.

In order to make this work, I will have to make a few assumptions which are not uncommon in the literature. First, I have to assume that every syllable has a head, which is most typically a vowel. Second, I have to accept the idea that there are no language-specific constraints, all conditions are universal, be it that they are ranked on a language-specific basis (this is the basic tenet of Optimality Theory, Prince and Smolensky 1993, McCarthy and Prince 1993a, etc.).

I expect that a lot of theoretical apparatus becomes obsolete, once we take these assumptions seriously. In particular, I argue that it is no longer necessary to stipulate a difference between prefixes and suffixes in terms of their phonological behaviour. It will also not be necessary to distinguish between lexical levels or between `Class I' versus `Class II' or `cohering' versus `non-cohering' suffixes in Dutch. This distinction furthermore is problematic in the sense that there is a class of suffixes which behave as members of Class I according to one criterion and as members of Class II for other criteria. I will argue that also this `paradoxical' behaviour can be derived from the phonological shape of these suffixes.

This paper is structured in the following way. In the sections 2 and 3 I briefly sketch an approach to asymmetries between prefixes and suffixes in Dutch. I give the relevant facts in section 2 and an analysis in section 3. In section 4 I then try to extend my analysis to the difference between `Class I' and `Class II' suffixes, as far as syllabification is concerned. One remaining problematic case will then be the suffix -achtig `-like', which will be discussed in section 5. In sections 6 and 7 stress -- a factor which could also draw the difference between the two types of suffixes -- is discussed. Based on the brief analysis of stress in monomorphemic forms in section 6, I show that the analysis from the earlier sections can also account for the behaviour of the Dutch suffixes in section 7. The last remaining criterion for the morphological status of afixes -- coordinate construction reduction -- is discussed in section 8. Section 9 discusses some so-called paradoxical suffixes which are ambiguous with respect to their phonological word status. It is argued that the behaviour of these suffixes can be derived from the segmental content of the vowels occurring in them: they all have a schwa. Section 10 discusses suffixes with paradoxical behaviour that have a non-schwa vowel in their head position. Section 11 concludes this paper.

Before laying out the theoretical apparatus in full, however, I will first turn to one set of examples illustrating the type of data we are interested in here: syllabification across morpheme boundaries.

2. Asymmetries between prefixes and suffixes

Monomorphemic sequences of consonants and vowels in Dutch are syllabified together (1a), as is indeed quite common in languages of the world. The same happens if the consonant is at the end of a stem and the vowel is initial in the following suffix (1b). However, the picture changes if the consonant belongs to a prefix and the following vowel to another prefix or to a stem. In this case, the syllable boundary will fall between the consonant and the vowel (1c).

(1) a. ode [o.d@] `ode'
b. er+en [e:.r@n] `to honour' (=honour+INF)
c. ont+ehr [Ont.e:r] `dishonour' (=dis+honour)

Two slightly more complicated examples of the same discrepancy between morphological and phonological structure can be found in (2) and (3) below:

(2) a. morphology: [[ ont- [ er ]] -en ]
dis ?honour? inf

b. phonology: Ont.e.r@n

(3)a. morphology: [ on- [[ en ] -ig ]]
un- one -y
`disagreeing'

b.phonology: on.E.nIx

The examples in both (2) and (3) consist of a prefix, a root and a suffix. The syllable boundary always falls on the boundary between prefix and root, whereas syllables cross the boundary between root and suffix if necessary. Together, the two examples show that the order in which the affixes are added is irrelevant. The fact that (2a) is derived by suffixation to the prefixed form ont+eer (dishonour), or that (3) is derived by suffixation to the prefixed form enig 'agreeing', is irrelevant for the strength of the respective syllable boundaries. The only morphological distinction that seems to matter is the one between prefixes and suffixes.

Another morphological property that seems irrelevant is headedness. The German prefix ent-, for instance, counts as a morphological head according to the usual criterion: it determines the categorial status of the derived form (it is a verbalizing sufix). Yet the phonological behaviour of ent- is not significantly different from that of the prefix un-, which does not behave as a morphological head at all. Similarly, derivational suffixes such as -ig are not significantly different from the inflectional suffixes -en, -em, -e, etc. in their phonological behaviour, even though the former, but not the latter, count as heads from a morphological point of view.

It should be noted that the syllable boundaries assigned here do not just correspond to native speakers' judgments. They also have a clear effect on phonological alternations. The most important one of these, in my view, is a schwa-zero alternation found in Dutch, and exemplified in (4) and (5) below:

(4) elite elite+air(Dutch)
[e.li.t ][e.li.tEr]/*[e.li.t@.Er] `elite' `snobbish'

(5)adem be+adem (Dutch)
[a.d@m] [b@.a.d@m]/*[ba.d@m]
`breathe' `breathe upon'

In Dutch monomorphemic forms we never find a schwa immediately preceding another vowel. This restriction can be understood in Optimality Theory as a result of syllable optimization. Schwa as a vowel has minimal feature content. Therefore, we may assume that it can be relatively easily deleted; it does not have a lot of underlying features that surface structure should faithfully reflect. In particular, we may assume that the faithfulness requirements demanding schwa to surface are ranked below the constraint Onset (see Van Oostendorp 1995 for a detailed analysis). This can be observed in the derivation of the following (hypothetical) underlying form /m@an/ for man (id.):

(6)

/m@an/
Onset
Keep-@
m@.an
*!
 
man
 
*

In affixed forms we find again an asymmetry between prefixes and suffixes, as can be learned from (4) and (5). This asymmetry can be reduced to the one we have just seen. The schwa cannot be deleted if it ends a prefix because the resulting surface syllable would cross a prefix-stem boundary. The schwa at the end of a suffix on the other hand, can be deleted under the appropriate circumstances, because syllabification over a stem-suffix boundary is not blocked.

In terms of OT, we could say that there is a constraint which blocks syllabification over a prefix-stem boundary, but not over a stem-suffix boundary. Let us provisionally call this constraint *PrefixSyll. This constraint should be responsible for the effects we have just seen:

(7) a.

/Ont+er/
*PrefixSyll
Onset
Ont.er
 
*
On.ter
*!
 

b.

/er+@n/
*PrefixSyll
Onset
er.@n
 
*!
e.r@n
   

c.

/b@+adem/
*PrefixSyll
Onset
Keep-@
b@.a.d@m
 
*
 
ba.d@m
*!
 
*

d.

/elit@+Er/
*PrefixSyll
Onset
Keep-@
e.li.t@.Er
 
*!
 
e.li.tEr
   
*

The combination of stem plus suffix thus is phonologically indistinguishable from a monomorphemic form. A prefix, on the other hand, seems to behave as a prosodically independent unit. We can find various descriptions for the phenomena involved in the literature: it has been suggested for instance that prefixes are always attached later in the derivation than suffixes, and to be more precise, after syllabification has applied (Booij 1981). Alternatively, it has been suggested that prefixes are specified in the lexicon with an underlying prosodic word (Van der Hulst 1984).

Neither of these solutions can be counted very satisfactory, especially in light of the fact that the type of asymmetry attested here is found in many languages of the world (**). On the other hand, as far as I am aware there is no language in which the mirror image effect is attested. I don't know of any language in which suffixes behave as more independent domains with respect to syllabification than prefixes. This fact is difficult to capture in either of the two approaches just sketched. It is unclear why there could be no language in which suffixes are always attached later than prefixes, or in which all suffixes have an underlying prosodic word, whereas prefixes don't.

The same problem arises in the OT analysis just sketched, at least as long as we don't have a satisfactory account of the precise nature of the constraint *PrefixSyll. As it stands now, we cannot account for the fact why natural language does not have a similar constraint *SuffixSyll, or if it does, why its effects are not as clear as those of *PrefixSyll.

The latter point is relevant for the treatment McCarthy and Prince (1993b) give to these facts. They formalize *PrefixSyll as an Alignment constraint (8) and give the tableaux in (7a) and (7b), modified accordingly.

(8) Align-Left: [X0 = (PrWd

This constraint says that the left boundary of every X0 (i.e. every morphological word) should correspond to the left boundary of a phonological word. This constraint is therefore satisified whenever a stem and a preceding prefix are in separate phonological words.

Generalized Alignment theory does not exclude the possibility of the symmetrical counterpart of this constraint, Align-Right. Alignment constraints can refer to the right edges of words just as much as to their left edges. In effect, McCarthy and Prince (1993b) suggest that Align-Right has a role in the grammar as well, be it that its formulation is somewhat weaker than that of Align-Left: it refers to the edge of a syllable rather than to the edge of a Prosodic Word:

(9) Align-Right: ]X0 = )s

The problem with this analysis is that it overgenerates. We now have three constraints (Onset, Align-Left, Align-Right) which could be ranked in four distinguishable ways:

(10) a. Align-Left >> Onset >> Align-Right

b. Align-Left,Align-Right >> Onset

c. Align-Right >> Onset >> Align-Left

d. Onset >> Align-Right,Align-Left

Ranking (10a) gives us languages like German and Dutch: the left edge of stems is stronger than the requirement that syllables have onsets (so that prefixes cannot be built over the prefix-stem boundary) but this latter requirement in turn is stronger than the right boundaries of stems, so that syllables may cross stem-suffix boundaries if necessary. Rankings (10b) and (10d) give us languages in which we can find no difference between prefixes and suffixes at all: in a language which has ranking (10b) both prefixes and suffixes behave as independent from the stem, and in a language with (10d) both types of affix are incorporated into the stem. I will not go into the question of whether these systems are actually attested. In any case, the approach outlined belowmakes the same predictions.

An important problem with this set of constraints however is ranking (10c). A grammar including this ranking would allow prefix-stem boundaries to be crossed by syllables, but not stem-suffix boundaries. This type of language does not exist, as we have just seen, and thus this result is undesirable.

McCarthy and Prince (1993b) acknowledge this problem, and in a footnote they suggest that there is a universal metaconstraint to the efffect that Align-Left is ranked over Align-Right. Of course, this metaconstraint itself is hardly more than a stipulation. McCarthy and Prince refer to psycholinguistics for an explanation: since it is well-known that listeners pay more attention to the beginning of words than to their ends, it is more effectful to try to keep the left edges of words clean than their righthand edges.

Several things can be said about this argument. First, it is not at all clear how this psycholinguistic argument -- assuming for a moment that it is true -- would give us this ranking. Suppose it is clear indeed that left edges of words ought to be respected: this observation could then be translated into the constraint Align-Left. Yet Align-Left can be violated in languages of the world, viz. in those languages in which prefixes are incorporated into the prosodic structure of the stem. From which psycholinguistic fact does it now follow that in these languages the right edges of words should also not be respected?

Furthermore, from the psycholinguistic observation that the left edges of words ought to be respected, it does not follow directly that prefixes should behave differently from suffixes. For instance (derivational) suffixes often behave like morphological heads, which means that they contain categorial and other information that is crucial to the interpretation of the word and the sentence. From this one fact one could draw the conclusion that suffixes should be as clearly distinguished from the stem as possible, whereas for prefixes this is less relevant. Apparently this observation does not help to explain the actual state of affairs in phonology, however.

In the following section I propose an alternative explanation for the types of asymmetries observed here, which does not rely on any presumed property of the human parser. If there is any relation at all, we might as well suppose that the parser works the way it does, because the linguistic structures happen to be the way I will now outline.

3. Phonological integrity

I assume that every syllable has a head. Typically this is a vowel, although under certain circumstances also consonants may serve as a syllable head in some languages. In any case, it is the most sonorous segment of the syllable.

My second assumption is that instead of Alignment constraints, we have two constraints Align and Pr=Lex (Prince and Smolensky 1993) which demand that the edges of morphological constituents should coincide with those of prosodic constituents and vice versa.

A last assumption is that phonological segments have a morphological domain. Typically, this is the smallest morphological word to which they belong. I will demonstrate this on my example from Dutch:

(11) [[Ont [ er]] @n]

The square brackets in this example indicate the boundaries of words: eer can act as an independent word, and so can onteer and onteren. Let us now consider the phonological domains of each of the three vowels. The /e/ in the stem eer has this stem as its morphological domain, since this is the smallest word in which it occurs. The vowel /O/ in the prefix has the derived form onteer as its domain, since the prefix is not a word in its own right. Finally, the schwa in the suffix has the whole word onteren as its domain, since this is the smallest independent word in which it occurs.

Slightly more formally we can now define the notion morphological domain in the following way:

(12) The morphological domain of a segment S equals the smallest word in which S occurs.

Next, we can define the morphological domain of syllables. Since syllables are headed, we can do this in terms of segment domains:

(13) The morphological domain of a syllable equals the morphological domain of the segment heading that syllable.

Thus, in the example above, the domain of the syllable headed by /e/ is eer, the domain of the syllable headed by /O/ is onteer, the domain of the syllable headed by schwa is onteren.

With this theoretical apparatus set up, we can now propose a formalisation for the constraint *PrefixSyll. I will call this constraint Integrity:

(14) Morphological syllable integrity (Integrity): Every segment S dominated by a syllable T: the morphological domain of S =< the morphological domain of T.

This constraint says, roughly, that the segments which are dominated by a syllable should be within the morphological domain of that syllable. In other words, all segments within a syllable should be in the same (smallest) word as the head of that syllable.

To see how this works, let us consider once again our example onteren. The domain of the second vowel /e/ is the root, therefore all the segments in the syllable headed by this vowel should be in the root eer. The t of the prefix is outside of this domain, therefore the syllabification *on-teren is not allowed. The domain of the schwa vowel in the suffix, on the other hand, is the whole word. The r at the end of the root obviously is within this domain and therefore the syllabification enteh-ren is allowed by Integrity. In constraint tableaux:

(15)

Candidates
Integrity
Onset
[on.[e.n]ig]
*
[o.[ne.nig]]
*!
*

(16)

Candidates
Integrity
Onset
[[Ont.er].@n]
**!
[[Ont.e.r]@n]
*

 

Alignment is irrelevant in the cases at hand, and we may therefore assume that Align and Pr=Lex are ranked below the Onset constraint. In all, we have three constraints, which may be ranked in the following three distinguishable ways:

(17) a. Integrity >> Onset >> Align
b. Align >> Onset (ranking of Integrity irrelevant)
c. Onset >> Align,Integrity

The grammar in (17a) gives us the Dutch and German facts; the grammar of (17b) gives a language in which both prefixes and suffixes are separated from the stem; (17c), finally is a language in which both prefixes and suffixes are incorporated into the syllable structure of the stem. A language in which suffixes behave as more separate from the stem than prefixes, cannot be generated, as expected.

4. An extension of integrity: Lexical levels in Dutch

Following the original proposals of Booij (1977), most phonologists have assumed that there are two types of Dutch suffixes: `Class I' suffixes and `Class II' suffixes in the original terminology. The following lists are copied from Booij (1977):

(18) Class I
-aal /al/, -aan /an/, -age /aZe/, -air /E:r/, -ast /Ast/, -eel /el/, -eer /e:r/, -ees /es/, -ees /es/, -egge /EG@/, -ein /Ein/, -erig /@rIx/, erij /@rEi/, -es /Es/, -esk /Esk/, -eur /ø:r/, -eus /øs/, -iaan /ijan/, -ide /id@/, -ief /if/, -iek /ik/, -iet /it/, -ieus /ijøs/, -in /In/, -iseer /ise:r/, -isme /Ism@/, -ist /Ist/, -iteit /itEit/, -ei /Ei/

(19) Class II
-achtig /AxtIx/, -loos /los/, -ling /lIN/, -baar /ba:r/, -dom /dOm/, -heid /hEit/, -nis /nIs/, -schap /sxAp/

At least three phonological differences were supposed to be related to the distinction between class I and class II suffixation. One of these had to do with stress; we will return to this shortly. The other two criteria were syllabification and schwa deletion, which apply across the boundary between a stem and a Class I suffix, but not across a 'Class II boundary':

(20)Syllabification
a. /mohAmEd/ + /an/ -> [mo.hAm.m@.dan] (*[mo.hAm.mEt.an])
b. /hAlv/+/lIN/ -> *[hAl.vlIN] ([hAlf.lIN]) 'hobbit' (litt. half-PERSON)

(21) Schwa deletion
a./sinod@/ + /al/ -> [sinodal] (*[sinod@al]) 'synodal'
b. /ward@/ + /los/ -> *[wartlos] ([ward@los]) 'worthless'

We can see that no resyllabification has applied in (20a) because the stem-final /v/ is devoiced: final devoicing does not normally apply to obstruents in the onset of a syllable.

Since prefixes trigger neither resyllabification nor schwa deletion, as we have just seen, Booij assumed that all Dutch prefixes belong to Class II inherently. We have seen above that this somewhat arbitrary stipulation is no longer necessary if we adopt Integrity.

Similarly, it seems that the distinction is superfluous for almost all of the suffixes as well. Since all Class I suffixes start with a vowel we get the desired result from the constraint ranking immediately, as we have in fact seen above:

(22)

Candidates
Integrity
Onset
[[mo.hAm.mE.d]an]
[[mo.hAm.mEt.]an]
?*

For most consonant-initial suffixes, the miniature grammar developed until now also gives the correct output, but vacuously so, since both Integrity and Onset are irrelevant:

(23)

Candidates

/v/\yl/+/nIs/ 'garbage' ('dirt'+NOM)

Integrity
Onset
[[v/\yl]nIs]

 

If a suffix starts with a liquid (-loos, -ling), and the preceding stem ends in an obstruent, a potentially ambiguous situation arises. The word werkloos `idle' could be syllabified either as werk.loos (respecting the boundary between stem and suffix) or as wer.kloos (satisfying the maximal onset condition).

The former option is actually chosen, and Booij (1977) took this as evidence for his claim that consonant-initial suffixes belong to Class II: the syllabification rules do not apply -- just as they don't apply in the case of roodachtig.

In the framework presented here, however, this line of reasoning does not hold. There is no single 'process' or 'rule' which syllabifies consonants into onset positions. Rather, there are several independent wellformedness constraints on the syllabified output structure. One such constraint is Onset, but this is irrelevant in this particular case, because it does not select between the two competing forms:

(24)

/werk/+/loos/
Integrity
Onset
->[[wer.k]los]    
->[[werk].los]    

Yet there is another syllabification constraint which is relevant. I will provisionally call this constraint Syllable Contact (Vennemann 1988) and formulate it as follows:

(25) Syllable Contact (SC)
*Ci.Cj, where Ci is less sonorous than Cj.

This constraint is normally undominated in Dutch, giving syllabifications suchs as [ta.blo] rather than [tAp.lo] for tableau. Yet it cannot be undominated in this case, because this would give us the incorrect results:

(26)

/werk/+/los/
SC
->[[wer.k]los]  
[[werk].los]
*!

We therefore need to find a constraint which can dominate SC. We have already seen this constraint above: Align (McCarthy and Prince 1983):

(27) Align: Morpheme boundaries should coincide the boundaries of prosodic constituents (i.e. a phonological word).

Obviously, this constraint has a lot in common with the alignment constraint in (9). There is a difference, however, in that (27) does not refer to left or right boundaries of either syllables or morphemes. It does not need to, because its effects are already as desired, if it is ranked appropriately:

(28)

/werk/+/los/
Align
SC
[[wer.k]los]
*!
 
[[werk].los]  
*

We now should establish the relative order between the two subrankings Integrity >> Onset and Align >> SC. The order between Onset and Lex=Syll is readily established. Once we have another look at the vowel-initial suffixes we see that Onset should dominate Align:

(29)

/er/+/@n/
Onset
Align
->[[e.r]@n]  
*
[[er].@n]
*!
 

If Align would dominate Onset, we would get *[er.@n] as the output. We have therefore established the following constraint ranking for Dutch:

(30) Integrity >> Onset >> Align >>SC
>> ?Keep-@

5. Achtig

There is now one suffix left which is problematic for the account presented here. This is -achtig '-like'. The problem with this is that it seems to be the only vowel-initial suffix which does not belong to level I. It does not trigger resyllabification (31a) or scha deletion (31b).

(31) a. roodachtig `reddish' [rot.Ax.tIx, *ro.dAx.tIx]
b. oranjeachtig 'orange-like' [o.rAn.j@.Ax.tIx, *o.rAn.jAx.tIx]

There are several ways to solve this problem. A first approach is to assume that -achtig is lexically prespecified as a foot or a prosodic word. A special type of faithfulness constraint (32) might then be held responsible for the lack of resyllabification:

(32) FaithfulFoot: If a surface foot Fi corresponds to an underlying foot Fj, all segments in Fi should be present in Fj.

(33)

/rod/+/AxtIx/
FaithfulFoot
Onset
(rot)(AxtIx)  
*
(ro)(tAxtIx)
*!
 

Yet there is a theory-internal technical problem with this. There are arguments to assume that we need to specify underlying metrical structure for certain `Class I' suffixes as well.

Take for instance the feminine suffix -in. This suffix bears exceptional stress. A word like bazin 'female boss' would be stressed on the prefinal syllable [bázin] if it followed the default stress rules of Dutch, but in actual fact stress always falls on this suffix. It can be argued that the best way to describe this is to assume that -In gets assigned a lexical stress, i.e. an underlying foot (cf. the discussion of stress in sections 6 and 7).

The problem now is that -in is clearly a `Class I' suffix: bazin is syllabified as [ba.zIn]. This means that FaithfulFoot cannot be operative here. As far as I can see, there is no non-arbitrary way to make this work.

An alternative approach assumes that -achtig has an underlying consonant like all Class II suffixes. An obvious candidate for this would be the glottal stop which is also present on the phonetic surface. Furthermore, Dutch phonology gives us more reasons to posit such a segment in the phonology, as we will see in section 6.

The advantage of this assumption is that the facts about -achtig now follow without any stipulation, because this suffix has the same phonological shape as -loos in all relevant respects:

(34)

Candidates
Integrity
Onset
Align
a. [[ro.d]?AxtIx]
*!
b. [[rot.]?AxtIx]

Both candidates fare equally well with respect to these two constraints. Therefore, there are other constraints which decide between the two. Align is one such constraint, but onset clusters of an obstruent followed by a glottal stop are also never found in Dutch. The constraint responsible for this, will naturally select candidate (b) in the table above.

A similar explanation holds for the derivation of oranjeachtig:

(35)

Candidates
Integrity
Onset
Align
a. [[ro.d]?Ax.tIx]
b.[[rot.]?Ax.tIx]
*

Again, the constraints Integrity and Onset cannot distinguish between the two candidates, and Align (or alternatively the constraint ruling out [j?] clusters) will start being operative and select [o.rAn.j@.?Ax.tIx].

A small technical problem arises here. We noted before that the sequence of a full vowel followed by a schwa is not allowed in Dutch monomorphemic forms. There are no monomorphemic words such as *[xA@s] or *[x@As]. But, as a matter of fact, there are also no words such as *[x@?As]. In other words, the sequence [@?A] is disallowed in monomorphemic words. One possible constraint that can rule this out is the following variant of the OCP (Cohen et al. 1958), since both the schwa and the glottal stops are segments without a specified place of articulation:

(36) *00: It is not allowed to have two placeless segments next to one another.

Apparently this constraint only holds within certain morphological (or possibly prosodic) boundaries. For instance, it certainly does not apply across morphological word boundaries: oranje asbak 'orange ashtray' [orAnj@AsbAk] is perfectly wellformed in Dutch. We could now say for instance that *00 only holds within the phonological word. As we will see in section x /?AxtIx/ automatically projects a phonological word of its own. Therefore, [(o.rAn.j@)(?Ax.tIx)] will conform to *00. On the other hand, [x@?As] is still not allowed, presuming that the constraint which requires the relation between morphemes and phonological words to be one-to-one (Pr=Lex), is sufficiently high-rankin, so that monomorphemic */x@As/ cannot project two phonological words.

6. Dutch word stress

Three properties are usually considered indicative of the morphological level to which an affix belongs. The first two of these are syllabification and schwa deletion. These have been discussed in some detail in the previous two sections. The third property is stress: level I suffixes are 'stress-shifting' (i.e. they require stress to be placed in a syllable in their immediate neighbourhood) or 'stress-bearing' (they get word stress themselves). Level II suffixes on the other hand are 'stress neutral': the word stress on the stem is exactly the same as when this stem occurs as an independent word (and usually the suffix gets some secondary stress).

In this section and the next I will go into this distinction in some more detail. It is my aim to show that also it can be understood in terms of the phonological shape of these morphemes.

The general idea is that since vowel-initial suffixes are forced to blur the morpheme boundary because of the Onset constraint anyway (and thereby violate the constraint Align), they will merge with the prosodic word of the stem also metrically. However, for consonant-initial suffixes (and -achtig), Onset is irrelevant, as we have just seen. For this reason, Align gets a chance: every morpheme should get its own prosodic word. This difference between vowel-initial and consonant-initial suffixes -- the former have a reason to integrate, the latter do not have such a reason -- will then account for a difference in stress.

I will make this idea more precise in the section 7. In this section I first give a general overview of the stress pattern of Dutch monomorphemic words, following the general idea of Nouveau (1994) and Van Oostendorp (1997).

Dutch stress is not too different from its English counterpart. Variation is possible, but only within certain limits. For instance, in words consisting of open syllables only, main stress can fall on any of the last three syllables of the word, but not on any other syllable:

(37)a.Panama [pánama] (id.)
b.pyjama [pijáma] 'pyjamas'
c.chocola [Sokolá] 'chocolate'
d. *[mákaroni]

It is quite generally agreed upon that some of the stress patterns in (37a-c) are more marked than others. Here, I will assume that penultimate stress (37b) is the most unmarked, that antepenultimate stress is slightly more marked, and that final stress is the most marked stress pattern.

If the final syllable is light and the penultimate syllable is heavy (i.e. contains a diphthong or is closed), the latter is always stressed.

(38) agenda [axÉnda] `diary' *[áxEnda], *[axEndá]

??Gibraltar [xibrÁltAr] (id.) *[xíbrAltAr], *[xibrAltÁr]

According to Nouveau (1994), the basic facts of unmarked Dutch stress can be described by the constraints in (39), ranked in the order given in (40):

(39)NonFinal:Stress should not be on the final syllable of the word.
Align-Right:Stress should be as much to the right as possible.
*Clash:Don't put two heads of feet in adjacent positions.
FtBin: Feet are binary at some level of representation.
Trochee: Feet are left-headed.
Superheavy: The segments of a 'superheavy syllable' should be together in a foot.
WSP: A heavy syllable should be in the head position of a foot.

(40)?*Clash, FtBin, Trochee, Superheavy >> NonFinal >> WSP >> Align-Right

Most of these constraints are known from the literature (e.g. Prince and Smolensky 1993, McCarthy and Prince 1993ab, 1995, Van de Vijver 1997).

A special word needs to be said about 'superheavy syllables'. These units consist of a tense vowel followed by one consonant, or a lax vowel followed by two consonants. Superheavies only occur at the end of the word. Within the word, tense vowels only occur in open syllables and lax vowels in syllables closed by one consonant. They are analysed as bisyllabic by Nouveau (1994) and Zonneveld (1993) among others and this is the analysis I will follow here as well; I assume that the 'extra' consonant at the end of the word forms a degenerate syllable. In order to make sure that stress always falls on the non-degenerate syllable of a superheavy, I postulate the constraint Superheavy.

Their working is briefly illustrated in the tableaux in ()-(), in which the constraints *Clash, FtBin, Trochee have been put together under the heading Top:

(41)

/piama/
Top
NonFinal
WSP
Align-R
(píja)ma      
**!
->pi(jáma)      
*
pija(má)
FtBin!
*
   
(pijá)ma
Trochee!
     

(42)

/xibrAltAr/
Top
NonFinal
WSP
Align-R
(xíbrAl)tAr    
**!
**!
->xi(brÁltAr)    
*
*
xibrAl(tÁr)  
*!
*
 
(xibrÁl)tAr
Clash!
*
   

(43)

/agEnda/
Top
NonFinal
WSP
Align-R
(ágEn)da      
**!
->a(gÉnda)      
*
agen(dá)
FtBin!
*
*
 
(agÉn)da
Trochee!
   
*

As one can see, the main stress is always on the prefinal syllable in the default case. There is one notable exception: word-final superheavy syllables always get stressed. Superhevay syllables in Dutch either consist of a tense vowel followed by a consonant, or of a lax vowel followed by two consonants:

(44)

/myzik/ 'music'
Top
NonFinal
WSP
Align-R
->mu(zík)  
*
   
(myzík)
Trochee!
*
   
(my'zik)
Superheavy!
 
*
*

The problem we now have to solve is: how can we account for the fact that exceptional stress patterns such as [SLS] and [SLS] are possible, whereas [SHS], [SHS] and [SSu] are not?

In Van Oostendorp (1997) it is proposed that the lexical marking we need to posit is underlying prosodic structure. In accordance with the spirit of Optimality Theory, it is proposed that feet can occur underlyingly anywhere in the word. The interaction of a faithfulness constraint on underlying feet with the constraint hierarchy in (40) will give us the desired results.

I formulate the faithfulness constraint as follows:

(45)Max-Foot: An underlying foot needs to have a correspondent in the output.

If we rank this constraint just below the WSP, we can explain why pánama is a possible stress contour, whereas ágenda and múziek are not:

(46)

/(pána)ma/
Top
NonFinal
WSP
Max-Foot
Align-R
(pána)ma        
**
pa(náma)      
*!
*
pana(má)
FtBin!
*
 
*
 
(paná)ma
Trochee!
   
*
 

(47)

/(ágEn)da/
Top
NonFinal
WSP
Max-Foot
Align-R
(ágEn)da    
*!
 
**
->a(gÉnda)      
*
*
(á)(gÉn)da
Clash!
   
*
*
(agen)(dá)
Bin!
*
     

(48)

/(m'yzik)/ 'music'
Top
NonFinal
WSP
Max-Foot
->my(zík)  
*
 
*
(myzík)
Trochee!
*
   
(m'yzik)
Superheavy!
 
*
 

(49)

/(máka)roni/
Top
NonFinal
WSP
Max-Foot
Align-R
(máka)roni        
***!
(maka)(róni)        
*
(mákaro)ni
Bin!
       
makaro(ní)  
*!
     

The constraints given until now account for most of the relevant facts of Dutch primary word stress. However, they still cannot explain why chocolá is a possible word of Dutch:

(50)

/choco(lá)/
Top
NonFinal
WSP
Max-Foot
Align-R
(chóco)la      
*
**!
->cho(cóla)      
*
*
choco(lá)
FtBin!
*
     
cho(colá)
Trochee!
*
 
*
 

An underlying foot on the final syllable would always violate two high ranking constraints, FtBin and NonFinal, and therefore it would not be allowed to surface. Words with such a marking would therefore surface with default stress.

Now, although it is true that chocolá is more marked than Pánama as far as its stress pattern is concerned (Nouveau 1994), it still is a wellformed word of Dutch and it should be generated by the grammar. We therefore need an extra type of lexical marking. One possibility we might entertain in this connection is catalexis (Kiparsky 1991, Kager 1995): there is an empty mora or an empty syllable at the end of the word chocolá. This word thus really ends in a trochee, be it that its final syllable is phonetically null. The possibility of a catalectic analysis for Dutch is also considered (but later rejected) in Nouveau (1994).

There are independent arguments to assume that these words actually do not end in an empty syllable or an empty mora, but rather in an empty consonant. The evidence for this is these words sometimes do indeed behave as word-consonantal. One indication in this connection is the behaviour of adjectival inflection. This inflection shows up after consonant-initial words, but not after vowel-initial words:

(51) a.het aardig+@ meisje
the nice girl

b.de grot+@ tafel
the big table

c.het sexy(*@) meisje
the sexy girl

d.de mica(*@) tafel
the micaceous table

If the final vowel is stressed, inflection does show up, just as if the word ended in a consonant:

(52) a. een wee(j)+@ geur
a sickly smell

b.een continu(w)+@ stroom
a continuous flow

For this reason I assume that words like these do indeed end in a consonant, and that this consonant is empty. For the sake of concreteness we might also assume that the empty consonant phonologically [h] or [?], since neither of these two segments is 'phonemic' word-finally in Dutch. Furthermore, we have seen that there are independent reasons to assume an underlying [?] in the analysis of the suffix -achtig.

Assuming an underlying empty consonant in chocola explains the word-final stress in this word, which now ends in a superheavy syllable (consisting of tense [a] plus the empty consonant):

(53)

/chocolaC/ 'chocolate'
Top
NonFinal
WSP
Align-R
->choco(láC)  
*
   
(chóco)laC
Superheavy!
*
*
**
cho(cólaC)
Superheavy!
 
*
*

Because these words now end in a superheavy syllable, they get word-final stress.

Unfortunately, the picture just sketched is somewhat too simplistic. In the first place, the simple account just sketched does not address the fact that there is evidence that the stress pattern of chocolá is more marked than that of pánama. Nouveau (1994) shows, for instance, that both children acquiring the language and adult native speakers tend to 'simplify' words like chocolá as chócola, whereas pánama is never changed into panamá. The account just sketched cannot deal with this. Chocolá has an empty segment and pánama an lexical foot, and there is no reason for any speaker of Dutch to 'simplify' one form to the other.

The second problem is that the distribution of empty consonants seems very restricted; they occur in a very limited number of positions only.

In order to properly describe the latter phenomenon we need to refer to two constraints. The first disallows empty segments anywhere in the word:

(54)*Empty: Empty syllables (e.g. degenerate syllables with an empty consonant) are not allowed.

The second constraint prohibits deletion of empty material:

(55) Max-C: Don't delete underlying consonants.

Clearly, if we rank *Empty over Max-C, we get the result that empty consonants do not surface, except where they are absolutely necessary for other reasons.

The interaction between these two constraints now solves our first problem: why is chocolá more marked than pánama, not just marked differently?

Assuming that (54) >> (55) are ranked below all other constraints we have just seen, there is no reason any more why chocola would not surface as chocóla if it only has an underlying empty consonant:

(56) *Clash, FtBin, Trochee, Superheavy >> NonFinal >> WSP >> Max-Foot >> Align-Right >> *Empty >> Max-C

(57)

/chocolaC/ 'chocolate'
Top
NonFinal
WSP
Align-R
*Empty
Max-C
choco(láC)      
*
*!
 
(chóco)laC
Superheavy!
 
*
**
   
cho(cólaC)
Superheavy!
 
*
*
   
(chóco)la      
**
   
->cho(cóla)      
*
   

The only way to let chocolá surface with final accent is by providing it with both an underlying empty consonant and a lexical foot:

(58)

/(choco)laC/ 'c`hocolate'
Top
NonFinal
WSP
Max-Ft
Align-R
*Empty
Max-C
->

(choco)(lá.C)

       
*
*
 
(chóco)la.C
Superheavy!
 
*
 
***
*
 
cho(cólaC)
Superheavy!
 
*
*
**
*
 
(chóco)la        
**!
 
*
cho(cóla)      
*!
*
 
*

We can now understand why chocolá can be regularized to chócola: the form may loose its lexical marking for an underlying empty consonant, only retaing the lexical foot, so that the underlying form of the word becomes the same as that of pánama. This lexical foot itself may actually also be lost, in which case the default pattern arises: chocóla, panáma.

7. Underlying stress and morphological form

After this summary of the basic facts of Dutch word stress within an OT frame, we can now start to analyze the behaviour of suffixes with respect to stress. `Class II' suffixes are always stress-neutral. The stress pattern on their base is exactly the same as it would have been if the suffix were not attached. Furthermore, primary stress stays on the stem, even though the suffix may get a secondary stress:

(59) televisie `television' [tèl@vízi]
televisie-achtig
`television-like' [tèl@vízi-?ÀxtIx]

`Class II' suffixes on the other hand are either `stress-attracting' or `stress-bearing'. In the former case, stress falls on the stem, but on some other position than where it would be if the suffix were not attached (60). This is always a position closer to the suffix, hence the name `stress-attracting'. In the case of `stress-bearing' suffixes, stress falls on the suffix (61).

(60) a. eenvoud `simplicity' [énvAut]
b. eenvoud+ig `simple' [envÁud@x]

(61) a.respect `respect' [rEspÉkt]
b. respectabel `respectable' [rEspEktáb@l]

It is widely agreed upon (see e.g. the recent textbook of Booij and Van Santen 1995, and the recent handbook of De Haas en Trommelen 1995) that the distinction between `stress-bearing' and `stress-attracting' suffixes can be derived from their respective phonological shapes. -abel `-able' is stress-attracting because it is bisyllabic (and furthermore, its final syllable contains a schwa; as a rule syllables immediately preceding schwa always attract main stress). -ig `-y' on the other hand is monosyllabic and therefore is more likely to be stress attracting: as we have seen in the preceding section, final syllables only get stress in exceptional cases also in monomorphemic words.

The conclusion we can draw from this is also drawn by most authors: that the stress in `Class I' suffixed words is not in any essential way different from that in underived words. As far as stress is concerned, the boundaries between Class I suffixes and stems are invisible.

This observation was quite easily captured in the derivational framework of Lexical Phonology. In this framework, we can assume that the stress rules applied at the end of Class I, i.e. after Class I suffixation, but before Class II suffixation. If we assume furthermore that metrical structure is respected after it is built, we get the proper characterisation of the facts: words derived at Class II receive a stress pattern much like a compound.

(62) respect+abel televisie-achtig
Level I: stress ass.??[rEspEktáb@l]?[tel@vízi]
-----------
Level II: compound stress -- [tel@vízi-ÀxtIx]

Yet these same facts can be made to follow just as easily without the stipulation of lexical levels. The reason for this has already been sketched in section 6: Class I suffixes have an independent reason to cross morpheme boundaries. Onset forces them to do this.

Particularly relevant is also the constraint Align, the constraint which requires every morpheme boundary to correspond to a phonological word boundary. This constraint plays a decisive role in the derivation of a word like televisie-achtig. In forms like this, both the root and the affix get their own phonological word, because of Align. The stress on this form therefore is compoundlike.

(63)

/televisie/+/achtig/
Align
(televísie)(?àchtig)  
(televisie?áchtig)
*!

The stress within each of the phonological words is determined by the constraints outlined in the previous section. As far as I can see, Align does not interact with the constraint ranking in (56). However, we have seen in section 4 that this constraint does interact with (is dominated by) Onset. This was the core of our analysis why level I suffixes are integrated into the syllabic structure of the stem. It can now also provide us with an explanation why they are integrated with the metrical structure of the stem:

(64)

respectabel
Onset
Align
(rEspEktáb@l)  
*
(rEspÉkt)(àbel)
*!
 

The difference in stress between Class I and Class II suffixes is therefore already explained: it is parasitic on their difference in segmental shape.

An explanation in terms of underlying prosodic structure, on the other hand, cannot work. Such an analysis would provide -achtig with an underlying foot, and would leave Class I suffixes without any underlying structure. Yet, in order to explain the stress in bazín (`female boss'), we need to assign it a lexical foot (and an empty consonant):

(65)

/baz/+/inC/
Top
NonFinal
WSP
Align-R
*Empty
Max-C
ba(zín.C)      
*
*!
 
(bá)zin.C
Superheavy!
 
*
**
   
(bá.zin)C
Superheavy!
 
*
*
   
->(bá.zin)      
*
 
*

(66)

/baz/+/(In.C)/
Top
NonFinal
WSP
Max-Ft
Align-R
*Empty
Max-C
->ba(zÍn.C)        
*
*
 
(bá.zIn).C
Superheavy!
 
*
*
**
*
 
ba(zÍn).C
Superheavy!
   
*
*
*
 
ba(zÍn)  
*!
 
*
   
*
(bázIn)    
*!
*
*
 
*

Assigning an underlying foot to -achtig therefore would not be sufficient; it would not explain why this suffix behaves differently from -in.

8. Prosodic word deletion

In the preceding sections I have shown that the prosodic word status of Dutch prefixes and suffixes can be derived from their phonological shape. All the major phonological tests for prosodic wordhood but one have now been discussed The one exception is due to Booij (1977, 1981, 1995), who notes that Dutch has a rule of deletion that is sensitive to prosodic word status. In a conjunction of two partially similar elements, the first of these two can be deleted, provided that it has phonological word status:

(67) a.roodachtig en groenachtig
redlike and greenlike

b.eetbaar en drinkbaar
eatable and drinkable

c.*groenig and rodig
greenish and reddish

I will not go into the details of the deletion process here. We might surmise, however, that the constraints involved have the following effect:

(68) AllOrNothing: Do not delete only part of a phonological word.

Another constraint might roughly have the following effect:

(69) *XandX (special case of Telegraph, Pesetsky 1997): Delete redundant material in conjunctions.

Together these two constraints will give the appropriate results:

(70)

a.

/eet/ + /baar/ + /en/ + /drink/ + /baar/
AllOrNothing
*XandX
(eet)(baar)(en)(drink)(baar)  
*!
->(eet)(en)(drink)(baar)    

b.

/groen/ + /ig/ + /en/ + /rood/ + /ig/
AllOrNothing
*XandX
->(groenig)(en)(rodig)  
*
(groen)(en)(rodig)
*!
 

9. Paradoxical suffixes

Several suffixes are problematic for approaches trying to provide a unitary explanation for the syllabification and the stress behaviour of suffixes. A list of these is given in (), which is copied from Booij (1995):

(71) -e /@/ (several functions)

-el /@l/ (denominal verbalizing suff.)

-en /@n/ (pl.suff.)

-er /@r/ (several functions, plus allomorph /ar/)

-erd /@rd/ (creates deadject. pej. names)

-erig /@r@G/ `-ish'

-ing /IN/ `-ing'

-nis /nIs/ `-ness'

-s /s/ substantivizing suff, pl. suff., gen. suff.)

-sel /s@l/ (creates de-verbal nominalizing suff.)

-st /st/ (super., de-verbal nominalizing suff.)

-ster /st@r/ (feminizing suff.)

-t /t/ (de-verbal nominalizing suff.)

-te /t@/ `-ness'

-tje /tj@/ diminutive (plus allomorphs)

The problem with these elements is that they behave as Class I suffixes as far is syllable structure is concerned, i.e., they integrate with the syllable structure of the stem they are attached to (Booij 1977, 1995, Langeweg 1988, and references cited there).

A classic example is the following:

(72) a. Judas [j'y.dAs] `Judas'

b.judasserig [j'y.dAs.s@.rIg] 'Judas-like' (i.e. treacherous)

Booij (1995) has shown that even the stress behaviour of these suffixes can sometimes provide us with arguments that they are incorporated in the phonological word. For instance the plural suffix for nouns has two allomorphs: -s and -en. The choice among these forms is dependent on many factors, but at least one of these seems to be prosodic well-formedness. Words with a stressed syllable usually adopt the form -en (/@n/ or /@/, in most variants of Dutch the final n never surfaces), whereas words with penultimate stress select -s:

(73) familie 'family' [famíli] families [famílis]

encyclopedie [Ensiklopedí] encyclopedieën [Ensiklopedi(j)@]

It is easy to see that this follows from the general tendency in Dutch to build trochaic feet and to have stress on the antepenultimate syllable of the word. Assuming that allomorph selection is free, the constraints Nonfinal and Align-R together will select between families and familiën and between encyclopedies en encyclopedieën. They will prefer the forms presented in (73).

This line of reasoning is based on two assumptions. First, the stress in both encyclopedíe and famílie should be fixed independent of the plural suffix. This could be due to an underlying marking in encyclopedíe, as we have seen in section 6, but it is not so clear that there is any lexical marking in famílie. However if stress in this form is not established independent of plural suffixation, we cannot understand why we could not get the form familí(j)en, where stress is also on the penultimate syllable?

The second assumption we have to make is that the plural suffix is indeed incorporated into the prosodic word structure of the base, wherever this is possible. If the choice of the plural suffix is indeed determined by considerations of prosodic wellformedness, we have no other option than making this assumption. On the other hand, these suffixes are still 'stress neutral'. If families were an underived word, it would get stress on the final syllable, because that syllable is superheavy.

These suffixes thus behave in a paradoxical way: they behave as if they are part of the phonological word as far as syllable structure is concerned and also with respect to the rhythnmic organisation of the resulting word, but on the other hand they behave as invisible for stress assignment itself.

Up unto this point I have been rather unspecific about the exact way in which the constraint Align in (27), repeated here for convenience, is satisfied.

(27) Align: Morpheme boundaries should coincide the boundaries of prosodic constituents (i.e. a phonological word).

As far as Align is concerned, we have two options. We can either assign a separate prosodic word to every morpheme, as is illustrated in (74a), or we can build a recursive structure such as (74b). Both of these structures would contrast with (74c) in which Align is violated:

(74) a. PrWd

PrWd PrWd

[[r o o t] a f f i x]

b. PrWd

PrWd

[[r o o t] a f f i x]

c. PrWd

[[r o o t] a f f i x]

Align is trivially satisfied in (74b): since there is a one-to-one mapping from morphemes to prosodic words, every morpheme boundary as a matter of course corresponds to a prosodic word boundary. In (74b), on the other hand, the prosodic structure is homomorphous to morphological structure in all relevant respects. Also in this case every boundary of every morpheme corresponds to the boundary of some prosodic word. The only difference between (74a) and (74b), as far as Align is concerned, is that the left boundary of the suffix and the right boundary of the stem correspond to the same prosodic word in (74b), while they correspond to different prosodic words in (74a).

There are ways to empirically distinguish between (74a) and (74b). All segments in (74b) are still part of the same phonological word. We may therefore assume that they can be syllabified together, while this cannot be done in (74a): segments in two separate phonological words cannot be part of the same syllable, due to a (possibly universally) high-ranking constraint on Prosodic Word Integrity:

(75) Prosodic syllable integrity (PSI): Every segment S dominated by a syllable T: the prosodic domain of S =< the prosodic domain of T.

But this means that the structure in (74b) offers a way to satisfy Onset and Align at the same time.

If the 'paradoxical' suffixes in (71) could be assigned such a structure, their behaviour can be accounted for rather straightforwardly, because the recursive structure can easily accomodate this paradoxical behaviour: we may assume that main stress is assigned within the most deeply embedded Prosodic Word, even though syllable structure can be embedded within the larger domain. The stress pattern could be described for instance by reference to a high-ranking constraint HdWd (Kager 1996):

(76) HdWd: The head of PrWd is the innermost PrWd.

Also the syllabification facts can be made to follow quite straightforwardly (the relative ranking of PSI cannot be determined at this point):

(77)

Candidates
Onset
Align
PSI
a.[[ro.d]@] (+rec. PW)
b.[[ro.d]@] ()
*!
c.[[rot.]@]
*!
d.[[ro.d]@]
*!

What determines allomorph selection in (73)? General rhythmic principles might be involved here, such as the constraint NonFinal, blocking stress on a final syllable, and a constraint NoLapse, which forbids sequences of unstressed syllables.

The problem now resides with the `real' Class I and Class II suffixes. Why would not these suffixes end up with a similar structure to the one in (77a), in other words, why do not all suffixes behave as paradoxical?

If we compare the list of suffixes in (71) to the ones in (18) and (19), we may observe that (almost) all elements of the last two classes contain at least one full (i.e. non-schwa) vowel. The suffixes in (71), on the other hand, have a schwa as their only vowel or otherwise consist of consonants exclusively. There is a handfull of apparent exceptions to this generalisation: a few suffixes in (71) contain a full vowel. I will return to these in section 10 below. For now, I will stick to the generalisation in (78):

(78) A suffix behaves as `paradoxical' iff it does not contain a full vowel.

Since we have established that paradoxical behaviour equals the structure in (74b), a structure which seems very desirable in the constraint system established hitherto, we can conclude from (78) that (79) should hold as well:

(79) Full vowels are not allowed in a position that is adjoined to the prosodic word.

We may assume that there is a relation between this constraint and the constraint HdWd in (76). Schwa in Dutch, as in many other languages, is the typical unstressed vowel. Full vowels tend to reduce to schwa in unstressed positions, and underlying schwa itself never surfaces as stressed. The domain of stress is the innermost Prosodic Word in an adjunction structure, and apparently, this is the domain of the full vowels as well. We may therefore stipulate principle (80):

(80) License: Full vowels are only allowed in the head word.

This constraint might be formalized in several ways. For instance we may assume that schwa is an `empty vowel', formally represented by a bare [-cons] root node, whereas full vowels would have at least some specification for height and/or place of articulation. We could then say that it is these latter features that have to occur in the head of the word. Generalizing this observation might help us to understand why the suffixes in (71) all seem to have coronal consonants only (with the exception of -ing, on which see section 10). I will not attempt to give a full formal account of this intuition here. License will do the job, as far as necessary.

We need to rank License topmost in the hierarchy, to get the required result. Since it only mentions full vowels, the constraint will be irrelevant for suffixes that do not have a vowel, or that only have a schwa. Therefore, these suffixes will end up in an adjoined structure. Such a structure is not attainable however for true Class I suffixes (or true Class II suffixes, for that matter):

(81)

Candidates
License
Onset
Align
a.[[mo.hAm.me.d]an]
*!
b.[[mo.hAm.mE.d]an]
*!
c.[[mo.hAm.met.]an]
*!

The structure of the paradoxical forms with a schwa suffix can now be accounted for. The last question to answer is what happens to the three paradoxical suffixes that seem to have a full vowel on the surface: -ing, -aar and -aard.

10. Paradoxical suffixes without schwa

It is the goal of this article to diminish the number of morphological diacritics in the lexicon by deriving the phonological behaviour of sets of affixes from their underlying form. The only three suffixes for which this cannot be done, are -ing, -aar and -aard. These are the topic of this section. I will discuss the latter two suffixes together in section 10.2. The former will be discussed in section 10.2.

10.1 -ing

The Dutch suffix -ing is mentioned by most scholars as an example of a paradoxical affix. It derives nouns from verbs, just like its counterpart in English. It is however much less productive than the corresponding English form (cf. Van Haeringen 1971, De Haas and Trommelen 1993). In essence, productive -ing formation seems to be restricted to words derived by the suffixes -eer (or -iseer, 82a) or by prefixes (82b), although there are a few words which seemed to be formed of a simple verb plus the suffix -ing (83):

(82) a. standaardisering `standardisation', accentuering `accentuation', democratisering `democratisation'

b.onderbreking `interruption' (*breking `breaking'), aanraking `touch' (*raking `touch'), afschrijving `copy' (*schrijving `writing')

c.wrijving `irritation' (from wrijf `rub'), speling `leeway' (from speel `play')

The reputation of -ing as a stress-neutral suffix probably derives from forms such as those in (82b). The word afschrijving for instance has the stress pattern [ÁfsxrEivIN], which is not the stress pattern of an underived word, because in such a word stress cannot occur to the left of a diphthong. In other words, the form of afschrijving is different from that of an underived form. Presumably, this is the reason why this form has been called `stress-neutral'.

It should be observed, however, that the stress of schrijving is not unusual for a simple phonological word at all. In other words, there is nothing in the facts in (82) that argues against prosodic analyses such as those in (83), in which -ing is always incorporated in the phonological word of the base:

 (83) a. democratis eer ing
b. af schrijving

The problem is that -ing does not seem to be sufficiently productive to provide us with those cases which would really be a good testing bed to see whether it is stress attracting or not. All the stems I have found to which -ing could be attached either ended in a stressed syllable (wrijving) or in a stressed syllable followed by schwa (gijzeling [gÉiz@lIN] `kidnapping' from gijzel [gÉiz@l]). In both cases, it is hard to see where stress could have actually been other than in the position than where it is.

We can actually observe that the rhyme sequence [IN] never gets stressed in Dutch, not even in underived forms: although there are quite a few words such as koning `king', paling `eel', honing `honey', each of them with stress in the first syllable, there is no word in Dutch which ends in stressed -ing (with the exception of those words in which -ing is the only available rhyme, such as zing `sing' and ring `ring'.) As a matter of fact, there is a series of place names in Dutch, such as Scheveningen, Wateringen, etc., which are exceptions to the three-syllable window requirement, because they have stress on the first (the preantepenultimate) syllable of the word. Typically, the final three syllables in these words contain either schwa or -ing. This provides us with extra indications that the rhyme sequence -ing is not likely to be stressed, even if it is not a suffix. Its behaviour in derived forms therefore can probably be derived from its phonological shape.

10.2 -aar and -aard

A more problematic pair of suffixes is formed by -aar ([a:r]) and -aard [a:rt]. These behave as really paradoxical. On the one hand, they are vowel-initial and consequently they attract a preceding onset into their syllable. In the following, I will concentrate on -aar, but I presume that similar things can be said about -aard. We also find a few differences. These will be discussed at the end of this section.

On the other hand, however, they seem to form superheavy syllables. If they would really be 'Class I', they would consequently attract main stress. Yet they are perfectly stress neutral (cf. (b) with (c), where the latter has the regular class I suffix -ier).

(84) a.wand@lwAn.d@.la.r
walk walker

b. wÁn.d@.la.r

c.wink@l wIn.k@.lí.r
shop shopkeeper

-aar in (84) is one of the very few superheavy syllables that does not attract main stress. This could be explained if the suffix were in an adjoined position, but [a] clearly is at least as much a full vowel as the [i] of winkelier in (84). An important observation here, however is that -aar has an allomorph -er, with a schwa (and -aard similarly has an allomorph -erd). Indeed, it has been argued that -er is the unmarked allomorph of the two: -aar is chosen after a stem ending in a schwa-headed syllable, -er is selected elsewhere:

(85)wand[@]laar `walker' danser `dancer'

bewond[@]raar `admirer' schrijver `writer'

tek[@]naar `illustrator' voorzitter `chairperson'

In order to account for the fact that -aar is the more marked suffix of the two, we have to introduce a constraint prohibiting it. For the sake of concreteness, we might call this constraint *aar. *aar is dominated by a constraint prohibiting two schwa syllables next to one another: *@@ (see Van Oostendorp 1995 for full formal discussion). This admittedly rather crude analysis at least has the virtue of giving the desired result:

(86)

/wAnd@l/+{/ar/,/@r/}
*@@
*aar
[wand@lar]
 
*
[wAnd@l@r]
*!
 

(87)

/sprek/+{/ar/,/@r/}
*@@
*aar
[sprekar]
 
*!
[sprek@r]
   

There should be some connection between the fact that -aar has an allomorph with schwa, and the fact that it behaves phonologically as if it has a schwa itself. In Smith (1976), it is argued that the agentive suffix is /@r/ underlyingly and that /@/ is `strengthened' to /a/ under certain circumstances, but only after the relevant stress rules have applied. In an abstract way, this seems to be the appropriate analysis. It is the one I will adopt here as well, modulo certain technical details.

The odd behaviour of -aar is a consequence of opacity, from the point of view of Smith (1976): a superheavy syllable is created after stress has applied. One way to deal with opacity phenomena within Optimality Theory has been proposed by McCarthy (1997): a specific relation called 'Sympathy' among candidate outputs in the set evaluated by the function Eval.

In McCarthy's view, a candidate output can be declared sympathetic if it obeys some faithfulness constraint and if it is more harmonic than all other candidates that obey the same faithfulness constraint. A sympathy constraint can subsequently require all candidates in the `real evaluation' to be as close to the sympathetic candidate as possible.

A simple example (from McCarthy 1997) might show how this works. In Tiberian Hebrew we find a process of vowel epenthesis in consonant clusters, as illustrated in (88a). Next to this, we also have a process of glottal stop deletion, as illustrated in (88b). The processes of vowel epenthesis and glottal stop are in an opaque relation. A vowel can be epenthesized before an underlying glottal stop, even if this glottal stop is deleted on the surface (88c). This latter fact can not be understood by simple constraint ranking in a theory of input-output comparison alone: no matter how the constraint responsible for epenthesis (*Complex) and the one responsible for deletion (*?]) are ranked, we never get the right result (89).

(88)a./melk/ -> melex `king'

b./qara?/ -> qara `he called'

c./deS?/ -> deSe `tender grass'

(89)

/deS?/
Complex
*?]
a. deS?
*
*
b. deSe?  
*
c. deSe    
d. deS    

The ranking between these constraints seems irrelevant, because two candidates (89c) and (89d) satisfy both of them. Since (89c) does violate a faithfulness constraint (Dep-V) unnecessarily, the optimal output in most probability will be (89d).

Yet, a sympathetic form is selected by the constraint Max-C, the constraint against deletion of consonants which is itself ranked in a position that is sufficiently low (because the glottal stop is in fact deleted).

(90)

/deS?/
Complex
*?]
Max-C
a. deS?
*
*
 
b. deSe?  
*
&
c. deSe    
*
d. deS    
*

 

(90b) is the sympathetic candidate (indicated by &), because it satisfies the constraint Max-C, and it is the most harmonic candidate that does this (e.g. it is more harmonic than (90a) because the latter candidate violates *Complex).

Now we can rank a faithfulness constraint in the appropriate position in the hierarchy (below Complex and *?], but above Dep-V), to get the desired result: the optimal output should be as close to the sympathetic candidate as possible, although as a matter of course also sympathy is violable (otherwise, the sympathetic candidate would always also be the actual winner). The faithfulness constraint is notated here as Max-V(&): all vowels in the the sympathetic candidate should have a correspondent in the actual output:

(91)

/deS?/
Complex
*?]
Max-V(&)
Dep-V
Max-C
a. deS?
*!
*
*
   
b. deSe?  
*!
 
*
&
c. -> deSe      
*
*
d. deS    
*!
 
*

Epenthesis in the actual winner is not required because it would solve any problem with *Complex, but because it makes this form resemble the sympathetic candidate. The latter candidate itself crucially cannot surface because it violates the constraint against word-final glottal stops.

An analysis of the -aar/-er alternation could run along the same lines. This time, the sympathetic candidate would be selected by the constraint *aar: it would be the most harmonic form that satisfies this constraint. The optimal candidate would be required to resemble the sympathetic form as much as possible. This comparison would involve faithfulness constraints on foot and word structure, abbreviated in the following tableau as Faith.

(92)

/wAnd@l/+{/ar/,/@r/}
*@@
Sympathy
*aar
License
Align
[wÁnd@l@r]
*!
 
$
   
[wAnd@l'@r]
*!
     
*
[wÁnd@lar]
   
*
*
 
[wAnd@lár]
 
*!
*
 
*

[wÁnd@l@r] is the most sympathetic candidate: it is more harmonic than the other candidate that satisfies the constraint *aar ([wAnd@l'@r], which violates Align). The form [wÁnd@l@r] itself is not considered, because it violates *@@. Yet it influences the choice of the actual output, which is the form that resembles it most from a prosodic point of view: [wÁnd@lar].

The notion of sympathy thus offers a way to capture Smith's (1976) insight in optimality theoretic terms. The importance of Smith's idea for the approach developed here is that it allows us to maintain the original hypothesis: that the morphological status of Dutch affixes is determined by their phonological form.

11. Conclusion

I have argued in this paper that it is not necessary to stipulate an underlying lexical marking for `Class I' versus `Class II' suffixes. Their phonological behaviour follows from their phonological shape. It is also not necessary to stipulate a morphological difference between prefixes and suffixes beyond the obvious fact that prefixes are attached on the lefthand side of the stem and suffixes on the righthand side.

Underlying morphological diacritics seem no longer necessary once we assume a theory of violable and rankable output constraints. The main theme of this paper is that isomorphy between phonological and morphological structure is enforced by two constraints: Integrity and Align. Yet these constraints interact with constraints on phonotactic wellformedness, such as Onset. The latter constraint may force a less-than-perfect mapping from the morphology onto the phonology.

In due course the following constraint rankings have been established for Dutch:

(93)

Integrity>>Onset >>Align>>SC

>> Keep-@

Parse-@ >> Max-x,y >> *Clash, FtBin, Trochee, Superheavy >> NonFinal >> WSP >> Max-Foot >> Align-Right >> *Empty >> Max-C

AllOrNothing >> *XandX

*@@>>Sympathy>>*aar>>License>>Align

This miniature grammar can account for the basic facts of syllabification and primary stress assignment in monomorphemic forms in Dutch, as well as the interaction of these processes with the affixation. It should be noted that these constraints do not refer to `left' or `right' edges (except, irrelevantly, for the constraints responsible for stress) and that there is only one constraint which could be used to distinguish between two types of affixes: Max-x,y. Yet this constraint is not used to distinguish between the traditional Class I and Class II suffixes, or between prefixes and suffixes, but to draw a distinction that is traditionally completely orthogonal to these distinctions.

Acknowledgement

The author is grateful Geert Booij, Harry van der Hulst, Jan Kooij, Michael Redford and Anthi Revithiadou for comments and discussion. The research that we report on here is financed by the Dutch organisation for scientific research, NWO, grant xx.

References

Booij, G. (1977) Dutch Morphology; A Study of Word Formation in Generative Grammar. Foris, Dordrecht.

Booij, G. (1981) Generatieve Fonologie van het Nederlands, Het Spectrum, Utrecht/Antwerpen.

Booij, G. (1995) The Phonology of Dutch, Clarendon Press, Oxford.

Booij, G. and A. van Santen (1995) Morfologie; De Woordstructuur van het Nederlands. Amsterdam University Press, Amsterdam.

Cohen, A., C.L. Ebeling, K. Fokkema and A.F.G. van Holk (1959) Fonologie van het Nederlands en het Fries. Martinus Nijhoff, Den Haag.

Haas, W. de and M. Trommelen (1993) Morfologisch Handboek van het Nederlands; Een Overzicht van de Woordvorming, SDU Uitgeverij, 's-Gravenhage.

Haeringen, C.B van (1971) `Het Achtervoegsel -Ing: Mogelijkheden en Beperkingen', Nieuwe Taalgids 64:449-468.

Harris, J. (1993) `Integrity of Prosodic Constituents and the Domain of Syllabification in Spanish and Catalan', in: K. Hale and S.-J. Keyser, eds., The View from Building 20, MIT Press, Cambridge.

Hulst, H. van der (1984) Syllable Structure and Stress in Dutch, PhD Dissertation, Leiden University.

Kager, R. (1995) `Consequences of Catalexis', in: H. van der Hulst and J. van de Weijer, eds., Leiden in Last; HIL Phonology Papers I, Holland Academic Graphics, The Hague, pp. 269-298.

Kager, R. (1996) *****

Kiparsky, P. (1991) `Catalexis', Manuscript, Stanford University.

Langeweg, S. (1988) The Stress System of Dutch, PhD Dissertation, Brandeis.

Oostendorp, M. van (1995) Vowel Quality and Phonological Projection, PhD Dissertation, Tilburg University.

Oostendorp, M. van (1997) `Lexicale Variatie in Optimaliteitstheorie', Nederlandse Taalkunde 2:133-154.

Oostendorp, M. van (in press) `Italian s-Voicing and the Structure of the Prosodic Word', to appear in M. Davenport and S.J. Hannahs, eds., Phonological Structure, Foris.

McCarthy, J. (1997) `Sympathy and Phonological Opacity', Handout HIL Lectures, Sept. 1997.

McCarthy, J. and A. Prince (1993a) `Prosodic Morphology I: Constraint Interaction and Satisfaction', Manuscript, University of Massachusetts, Amherst, and Rutgers University, New Brunswick.

McCarthy, J. and A. Prince (1993b) `Generalized Alignment', Yearbook of Morphology 1993; 79-154.

McCarthy, J. and A. Prince (1995) `Faithfulness and Reduplicative Identity', Papers in Optimality Theory; University of Massachusetts Occasional Papers. 18:249-384.

Nouveau, D. (1994) Language Acquisition, Metrical Theory, and Optimlaity: A Case Study of Dutch Word Stress. Led, Utrecht.

Pesetsky, D. (1997) `Optimality Theory and Syntax; Movement and Pronunciation', in: D. Archangeli and T. Langendoen, eds., Optimality Theory; An Overview. Blackwell, Oxford. pp. 134-170.

Prince, A. and P. Smolensky (1993) `Optimality Theory; Constraint Interaction in Generative Grammar', RUCCS Technical Report 2, Rutgers University Center for Cognitive Science, Piscataway, N.J.

Smith, N. (1976) `-Aar', Leuvensche Bijdragen 65:485-496.

Vennemann, T. (1988) Preference Laws for Syllable Structure, De Gruyter, Berlin.

Vijver, R. van de (1998) The Iambic issue; Iambs as a Result of Constraint Interaction, PhD Dissertation, Free University Amsterdam.

Zonneveld, W. (1993) `Schwa, Superheavies, Stress and Syllables in Dutch', The Linguistics Review 10:59-110.