Appeared in S. Barbiers, J. Rooryck and J. van de Weijer, eds. Small words in the big picture; Squibs for Hans Bennis. Leiden: HIL.
Optimality Theory (OT) is primarily a theory on language typology and language variation, its main point being that the (phonological) differences between languages can be described in terms of a language-particular ranking of universal constraints. It is therefore quite surprising that hardly any work has been done within this framework on the microvariation we find between dialects; most of the studies within OT concentrate on the larger-scale variation between languages. This squib intends to show that even one single fact of one single dialect of one single language can cause certain problems for the theory, while at the same time also affirming some of its fine-grained predictions. The fact in question is that the diminutive of the word bank (id.) [bANk] in Bergen Dutch is bangeske [bAN@sk@].
In many ways, Bergen Dutch -- spoken in the town Bergen op Zoom, in the Southwestern part of the Netherlands -- patterns with many non-standard dialects of Dutch in that the basic, unmarked, allomorph of the suffix is -ke1: boerderij - boerderijke, [burd@rEi- burd@rEik@], `farm'. After a velar consonant, the allomorph -ske is selected, because the hypothetical form *boekke would feature two adjacent velar obstruents and this is not allowed (by a principle called the Obligatory Contour Principle): boek - boekske, [buk - buksk@], `book'. If the last syllable of the stem has a short lax vowel and a sonorant consonant, the selected allomorph is -eke: man - manneke, [mAn - mAn@k@], `man'. Most of the discussion on the diminutive in Standard Dutch (cf. De Haas and Trommelen 1993 for an overview) has concentrated on the question of why the bisyllabic variant is chosen in exactly these circumstances.
The facts just mentioned are not overly exciting from a dialectological point of view: Bergen Dutch shares them with many other dialects. If we look at the diminutive of words ending in the cluster /Nk/ however, something very surprising happens: bank - bangeske, [bANk - bAN@sk@], `bank'.2
What is going on here? Under a rule-based approach, it looks as if the processes involved are opaque in various ways, as if the stem-final /k/ has been deleted for one reason or another. The /s/ should have been inserted before this, because we do not find it in forms such as tangeke (*tangeske) from underlying [tAN]. But insertion of schwa can only have occurred after k-deletion, because we have seen that schwa is only inserted after sonorants. We would thus have evidence for a rule ordering of depth 3:
Ordering these rules in any other way would give us unattested outputs such as *bangske, *bangke or *bangeke. As far as I know, these facts are quite unique for Bergen Dutch. They cannot be found in neighbouring towns such as Breda, and indeed Heestermans reports that they are also missing in the speech of younger Bergen speakers, who say benkske, perhaps under the influence of the neighbouring dialects. In rule-based theory, it is quite easy to account for this fact, viz. by assuming that the k deletion rule gets lost.
On the other hand, these facts pose a problem for Optimality Theory, which is purely output based. If we look at the output alone, it is not clear for instance why the diminutive of [bANk] cannot be *[bAN@k@], if /s/ insertion is triggered by two adjacent instances of /k/. The two output forms are the same, so it is unclear why one is allowed but the other is not. In the recent OT literature, several solutions have been proposed for this type of (`counterbleeding') effect. The problem with these solutions is that they invariably involve complicated machinery that is not necessary in the neighbouring dialects where one just says bankske. The difference between Bergen Dutch and the other dialects is then rather large. It seems preferable to be able to say that small dialect differences involve small differences among grammars.
It might not be necessary to invoke all of this machinery, if we could assume that -eske is simply an allomorph of -(s)ke: in that case the younger generation would simply be replacing one allomorph by another. Unfortunately, it probably is very hard to find good evidence pro or contra this assumption of allomorphy. I adopt it here only for reasons of space, there is another interesting phenomenon connected to this fact: the apparent deletion of the underlying /k/ of /plANk/ in [plAN@sk@]. The question that concerns me here is: why is the final segment deleted in this particular environment and not in e.g. [buksk@ - *busk@] (book) or [lEmp - lEm@sk@] `lamp'?
Let us first consider the contrast of *[lEm@sk@] vs. ok[plAn@sk@]. Why is the plosive deleted in the latter case, but not in the former? The sketchy rule-based analysis just outlined does not provide us with an answer to this: it just stipulates that the relevant rule is `k deletion', and not for instance `stop deletion'. As far as I can see, it is not clear what could be the motivation behind this rule. On the other hand, within the framework of Optimality Theory, we might get at least a first approximation to what is going on. One very important set of constraints consists of the so-called faithfulness conditions, regulating the relation between input and output, and thus taking care of the fact that not all inputs form surface as [tata] or [bi] or whatever the optimal phonological form in the language may be.
The most popular formalism to implement faithfulness nowadays is Correspondence Theory (McCarthy and Prince 1994). Within this theory, we can see the input and the output of the derivation as two parallel representations, as illustrated with the following fairly trivial example:
Correspondence constraints say (among other things) that all segments which are in the input representation should have a correspondent in the output representation (thus forbidding deletion of segments), and that all segments in the output should have a correspondent in the input (thus preventing unnecessary epenthesis of segments). In the example in (5), mi corresponds to m1, A j to A2, nk to n3, and vice versa. All the correspondence constraints are thus satisfied in a straightforward way.
Yet there are also other ways of satisfying the relevant faithfulness constraints. I propose that one of these is at work in the case of bangeske. The input-output pair could here be something like:
|(2)||input:?||bi||A j||N k||kl||+||@ m||sn||ko||@ p|
|output:||b1||A 2||N 3||+||@ 4||s5||k6||@ 7|
In this example, both kl and ko could correspond to k6. This satisfies the faithfulness constraints just mentioned, although it fails to meet with another type of correspondence constraint, viz. that the relations between input and output segments should be one-to-one. This latter constraint apparently is less strong in Bergen op Zoom than it is in other variants. It is not possible to construct an input-output pair like (2) for the input /lEmp+@k@/. In this case, the stem-final /p/ would have to correspond to the output [k] in the suffix. This would violate identity constraints on corresponding segments; apparently, this is not allowed.3
We may thus get an initial understanding as to why it is only the /k/ that can get deleted in this environment; in correspondence relation this can be related to the fact that the suffix also contains a /k/. It is still not clear however why the stem-final /k/ has to get deleted. As we just discussed, the output [bAN@sk@] still violates the one-to-one correspondence constraint; this has have been enforced by another constraint. What can this other constraint be?
It might be useful to understand why an alternative analysis for this same candidate is not grammatical: [plANk@sk@]. In the latter form, we have the `long' (bisyllabic) form of the suffix -@ sk@ after a stop. But this is excluded in Bergen op Zoom, just as it is excluded in all dialects of Dutch: there is a contrast between the word kam /kAm/ `comb' with diminutive [kAm@k@] and hap /hAp/ `bite' with diminutive [hApk@].
4. Unmarked Allomorphs
A sketch of an analysis of this in terms of ranked constraints may run along the following lines. In the first place we may assume that the long form of the diminutive is the unmarked form, for instance because this links the synchronic shape of the suffix to its historical form (-îkîn). An constraint UnmarkedAllomorph may then select the output candidate [kAm@k@] rather than *[kAmk@], but we still need an outranking constraint which prevents *[hApk@] from surfacing, and similarly for all other nouns ending in an obstruent. For this I will use the constraint *ObstruentSchwa, that forbids schwa-headed syllables with an obstruent in the onset. We thus have established a constraint ranking *ObstruentSchwa >> UnmarkedAllomorph for the cases without a cluster.
We can now evaluate the relevant candidate outputs for /plANk+@k@/:
The form [plAN@sk@] offers a way to satisfy both constraints, be it at the cost of low-ranking one-to-one correspondence. Let us now consider the contrast *[bu:sk@] (from /bu:k/) vs. ok[plAN@sk@]. In the former form there is no reason why the /k/ should be deleted, because we still see the short form of the diminutive affix (in other words, *[bu:sk@] does just as well as the actual output [bu:ksk@] vis à vis the constraints *ObstruentSchwa and UnmarkedAllomorph, so that one-to-one correspondence will select the latter). But why isn't the long form chosen in a case like this? I think that the crucial factor here is that in Bergen Dutch, like in other dialects, it is not allowed to have a sequence of a full vowel followed by a schwa. This prevents the form *[bu:sk@] from surfacing; there thus is no serious candidate output which could select the long form of the diminutive in this case.
One of the most hotly debated issues in present-day phonology is whether phonological theory is based on ordered rules or on ordered constraints on outputs and input-output relations. The fact that the Bergen op Zoom diminutive of bank is bangeske seems relevant for this discussion. On the one hand, it shows a type of opacity that is hard to capture in Optimality Theory without introducing rather complicated machinery. On the other hand the deletion of the stem-final /k/ in exactly this environment is something that remains a mystery in a rule-based approach, while (a correspondence-theoretic approach to) Optimality Theory seems able to shed some light on it. Further work within both frameworks is clearly needed in order to see whether any of them can successfully resolve this paradox.
2. According to Heestermans (1998), the same allomorph is chosen for stems ending in a short vowel plus a velar fricative, e.g. weg-weggeske [wEx-wEG@sk@]. This fact could not be confirmed by my informant, but it is at least as intriguing as the facts discussed in the body of the text.
3. This leaves open the question why the diminutive of kerk [kEr(@)k] is not [kEr@k@], but rather [kEr(@)ksk@]. It is possible that featural alignment is at stake: in the case of [bANk] the form ends in a velar even after deletion of the stop, but this is not true for [kErk].
Haas, W. de & M. Trommelen (1993) Morfologisch Handboek van het Nederlands; Een Studie over Verschillende Vormen van Woordformatie. SDU, Den Haag. Aan het Woord 7.
Heestermans, H. (1998) West-Brabants Zogezeed. Dagblad De Stem. 18 mei 1998.
McCarthy, J. and A. Prince (1995) Faithfulness and Reduplicative Identity. Ms., University of Massachusetts, Amherst & Rutgers University.
Oostendorp, M. van (1995) Vowel Quality and Phonological Projection. Doctoral Dissertation, Tilburg University.