Spektator. Jaargang 2

[p. 421]

N.S.H. Smith
The phenomenon of D-deletion in standard Dutch^1.

^*

In this article we will consider the rules governing d-deletion in Dutch and other phenomena connected with them. In addition to the phonological rules involved we will also consider the somewhat complicated morphological rules concerned.

Phonologically speaking there are clearly two sub-processes at work here. On the one hand we have: -

/ī/	verbieden → verbieën^2.	.īdə..→.ījə.	to forbid
/ē/	breder → breeër	..ēdə.→..ējə..	broader
/ā/	kwade → kwaaie	..ādə.→..ājə..	angry
/ō/	rode → rooie	..ōdə..→..ōjə..	red
/ū/	goede → goeie	..ūdə.→..ūjə..	good
/∊j/^3.	rijden → rijen	..∊jdə..→..∊;jjə..	to ride
/aw/	houden → houen	..awdə..→..awwə..	to hold
/äẅ/^4.	kruiden → kruien	..äẅdə..→..äẅjə..	herbs

where we have the loss of /d/ and generally its replacement by /j/. The other sub-process can be illustrated with the following examples: -

/ē/	slede → slee	..ēdə..→..ē..	slegde
/ā/	lade → la	..ādə..→..ā..	drawer
//	Brederode → Bredero	..ōdə..→..ō..	personal name
/ū/	broeder → broer	..ūdə..→..ū..	brother
//	De Neude → De Neu	..də..→....	place name
/∊j/	zeide → zei	..∊jdə..→..∊j..	said
/aw/	Abcoude → Abcou	..awdə..→..aw..	place name
/äẅ/	bruidegom → bruigom	..äẅdə..→..äẅ..	bridegroom

Here not only is the /d/ lost but the following schwa as well.

What are the conditions for d-deletion and what determines whether it is true d-deletion that occurs or də-deletion? It is at once clear that D-deletion (under which label we will refer to both d- and də-deletion) requires the presence of a tense vowel or a lax diphthong before the /d/.

1. I am very grateful for the valuable criticism and advice of Daniel Brink, Simon Dik, Gerard Hubers and Henk van Riemsdijk.

* Wegens redenen van technische aard is de samenvatting van dit artikel geplaatst na de bibliografie (p.437).

2. This example is rejected by most but not all informants.

3. In the case of words like rijden → rijen, the pronunciation appears to vary between [r∊jjə] and [r∊jə], that it is seems to be optional whether the /j/ appears in both syllables or only in the first. A similar problem occurs with houden. In the light of kruiden → kruien [kräẅjə] where we have two different glides, I have assumed that the forms with two glides are basic and that a low-level phonetic rule will take care of the details.

4. For the diphthong represented by the Dutch spelling ui, I have employed the symbols äẅ. This sound can be described as beginning in a centralized front vowel halfway between open and half-open positions and ending in a half-close centralized front rounded semivowel.

[p. 422]

Our first sub-process occurred in all such environments except that following /ȳ/ and //, the second in all except /ī/, and /ȳ/. We will regard these as accidental gaps (the /ȳ/ is certainly subject to the rule in certain dialects. Compare the Brabant place name Uden (locally /ȳjə/. Its nonoccurrence here is due to its extreme rarity in standard Dutch). Thus it is clear that the conditioning for D-deletion is at least partly phonological. As to when we get d-deletion and when də-deletion, this is not so obvious. Clearly it is not solely phonological. Thus compare the following: -

broeder → broeier	..ūdə.→..ūjə..	brooder
broeder → broer	..ūdə..→..ū..	brother
kwade → kwaaie	..ādə..→..ājə..	angry
kade → ka	..ādə..→..ā..	quay
breder → breeër	..ēdə..→..ējə..	broader
veder → veer	..ēdə..→..ē..	feather

Clearly, morphological facts are of some relevance here. Let us examine some of the morphological environments where the two sub-processes occur.

d-deletion

i)	before the agentive suffix -er	eg, broeder → broeier	brother
ii)	before the adjectival suffix -e	eg, rode → rooie	red
iii)	before the comparative suffix -er	eg, breder → breeër	broader
iv)	before the infinitive suffix -en	eg, rijden → rijen	to ride
v)	before the past partcpl. suffix -en	eg, gereden → gereeën	ridden
vi)	before the N-plural suffix -en	eg, tijden → tijen	times
vii)	before the V-plural suffix -en	eg, luiden → luien	strike
viii)	before the locative suffix -en	eg, beneden → beneeën	beneath
ix)	before the directional suffix -en	eg, zuiden → zuien	south
x)	before the adjectival suffix -en	eg, gouden → gouwen	golden

də-deletion

i)	before the plural suffixes -er-en	eg. bladeren → blaren	leaves
ii)	cpnd. suffix -er^5.	eg. kledermaker → kleermaker	tailor
iii)	adjectival suffix -(e)lijk	eg. dadelijk → dalijk	presently
iv)	in the past tense morphemes -d-e	eg. zeide → zei	said
v)	morpheme-internal -de	eg. veder → veer	feather

Thus the d-deletion cases all involve the environment Vd + ə(c)#. No such uniformity seems to be present in the də-deletion cases. We have there the following environments: -

Vd+əC+əC#, Vd+əC# #X. Vd+əCəC#, V+d+ə#, VdəX.

5. This suffix is the old plural allomorph -er. In simple plural forms in present-day Dutch the words that retain this suffix add one of the two productive suffixes, -en or -s, in addition. For instance:

kind-er-en	blad-er-en	hoen-der-s
kled-er-en	volk-er-en	spaan-der-s
kalv-er-en	lamm-er-en	ei-er-en

In compound formation, where plural forms of count nouns often appear, eg. bloemenveiling (flower sale), we get just the -er forms, eg. kindervervoer (child transport), kle(d)ermaker (tailor).

[p. 423]

There is also a stylistic difference between the occurrence of d-deletion and the occurrence of də-deletion. In the case of d-deletion we have a distinction between a colloquial form when the /d/ is deleted and a more formal one when it is not. With-də-deletion however we have a quite different situation. The forms with də-deletion are the normal forms of the words concerned whereas the forms without də-deletion are archaistic in effect. This is not the case with the place names but these often display untypical behaviour. Another significant difference is that the first process is quite productive while the second is decidedly unproductive.

We will attempt first to give a purely phonological-morphological analysis of D-deletion and then later consider how we can best take cognizance of the facts concerning style and productivity.

Phonologically we could account for our first sub-process, d-deletion, by simply replacing /d/ by /j/ in the environment given above. This would be unsatisfactory in the light of the following phonological relationships: -

singular	plural	singular	plural
drie	drieën	..ī.	..ījə..	three(s)
wee	weeën	..ē..	..ējə..	birth pain(s)
vlo	vlooien	..ō..	..ōjə..	flea(s)
koe	koeien^6.	..ū..	..ūjə..	cow(s)
paraplu	parapluën^7.	..ȳ..	..ȳjə..	umbrella(s)
reu	reuën	....	..jə..	male dog(s)
lei	leien	.∊j..	..∊jjə..	slate(s)
ui	uien^8.	.äẅ..	..äẅjə..	onion(s)
touw	touwen	..aw..	..awwə..	rope(s)

From these examples we can see that a rule is basically involved here whose prime function is to insert /j/ between a tense vowel or a lax diphthong and a following schwa.^9. Since this rule will be needed in any case it would seem to be reasonable to let it explain also the /j/'s that appear in the case of d-deletion. Our rule of j-insertion can be formulated as follows:

J-INSERTION

This rule will need to be followed by an assimilation rule that will change the sequence /wj/ into the sequence /ww/.

W-ASSIMILATION

j → w / w -

6. This is an irregular plural since forms in -oe will normally take -s. However it is of relevance since it demonstrates what happens when the -en does occur. Vlo is a parallel case.

7. This, and a number of other words in -u, has for some speakers no /j/ glide. this is probably to be connected with their status as foreign words.

8. Some speaker have here [äẅẅə] instead of [äwjə]. These will have a more general version of W-ASSIMILATION.

9. This will be a different rule from the lowlevel rule inserting glides between high vowels and following unstressed vowels as in Ria (girl's name), [rija], Mantua (Mantua), [mantyẅa].

[p. 424]

Note here that the configurations resulting from these rules do not match the tense vowel - semivowel - boundary sequences present before their application.

INPUT SEQUENCES

īw# ȳw# ūj# ēw# ōj# āj#^10.

J-INSERTION OUTPUT SEQUENCES

īj+ ȳj+ ūj+ ēj+ ōj+ ōj+ āj+

Thus the older sequences all involve either a front vowel followed by a back semivowel or a back vowel followed by a front semivowel.

Now let us consider the rules necessary to account for d-deletion and də-deletion. If we chose to make these separate rules, d-deletion would apply in the environment.

whereas də-deletion would apply in the environment

(Neither of these rules ever applies after an unstressed vowel). In fact, however, these two environments are very similar. They both involve the sequence

This leads one to think that one rule only might be involved here. The reason d-deletion would require a more complicated environment is that it occurs in a more restricted set of cases than də-deletion. Note however that these are precisely the cases where J-INSERTION will apply. And this rule will follow d-deletion in any case. Therefore if we state only one rule of D-DELETION replacing both our putative rules of d-deletion and də-deletion, the differing behaviour of our two sub-processes will still be

10. We will treat tense and lax /a/ as being underlying back vowels. According to Chomsky and Halle (1968) the back low unrounded vowel is the unmarked vowel and it must therefore be present in all underlying vowel systems. In this case the tense /ā/ is clearly behaving like a back vowel.

[p. 425]

stated in our rules. Following J-INSERTION there will be necessary a general rule of SCHWA-DELETION to take care of those cases where J-INSERTION has not operated. These three rules will now be formulated as: -

D-DELETION



J-INSRTION



SCHWA-DELETION

This gives us three rather notural rules. D-DELETION can be regarded as a rule of weakening of a voiced stop in intervocalic position, J-INSERTION as a hiatus-filling rule, and SCHWA-DELETION as a rule simplifying complex vocalic sequences.

There is still a problem with these rules, however. It concerns the string which appears in the environment of all three rules. The statement of the two alternative possibilities within curly brackets suggests that it is chance that these two strings function equivalently in the context of this rule. It also defines one of two entirely irrelevant explicit conjunctive orderings on the two sub-rules. It would be possible to employ angled brackets here, taking advantage of the fact that no forms involving sequences of a lax vowel plus /d/ plus /ə/, ie /də/, will be marked as being subject to D-DELETION, and that no sequences of lax vowel plus schwa will ever appear in the inputs to J-INSERTION or SCHWA-DELETION. Thus we would have




which defines the two sub-environments

[p. 426]

in that order. While this relieves us of the need to make a choice between possible orders of sub-rules, it obscures the real facts here that these rules do not operate after all vowels but only after tense ones. To use angled brackets as we have here, we have merely employed an artefact of the rules and failed to express what is actually going on.

Thus, though the use of angled brackets here is therefore observationally adequate, it is in fact unsatisfactory. It leaves the possibility open that forms with lax vowels could undergo these rules. It is equivalent to stating that it is merely chance that no lax vowel forms do in fact do this. This is clearly untrue.

A third possible solution is to be found in the work of Kenstowicz (1970), Pyle (1970) and Fidelholtz (1971). These articles all involve languages where it appears that for some rules of the phonology it is necessary that long vowels be analysed as unitary long vowels and for other rules as sequences of two short vowels. There are some problems involved in the utilisation of this solution here. Thus not all the tense vowels are long in Dutch. The high tense vowels are short except before /r/, where they are long. The point is that if we could represent our tense vowels as sequences of two short vowels, then it would be easy to abbreviate them together with the sequences of lax vowel and semi-vowel. Thus before we can utilise this solution it will be necessary to assume that all the tense vowels are long at the stage at which these rules apply, and that the shortening of the high tense vowels takes place later. To account for the fact that our tense vowels are long, we may be able to make use of a marking convention. Thus it may well be the case that the unmarked value of the feature [long] will be [+long.] for tense vowels. It does seem to be case that for many languages the tense vowels are also the long vowels, whether or not a quality distinction is present. This is the case with Standard English, for example. We will assume therefore the existence of the following marking convention: -

Thus we can now define a high tense vowel shortening rule that will apply after the rules that we have already formulated.

HIGH TENSE VOWEL SHORTENING

[p. 427]

Here the long vowels have to be treated as units, demonstrating that in Dutch too these segments have a variable status. Thus for D-DELETION, J-INSERTION and SCHWA-DELETION they have to be regarded as sequences of two short vowels while for the rule above they have to be treated as unitary long vowels.

Our formulation of D-DELETION, J-INSERTION and SCHWA-DELETION is now: -

D-DELETION^11.



J-INSERTION^12.



SCHWA-DELETION



We have now the problem of how to specify the words that can undergo D-DELETION, and represent all the various sub-regularities, redundancies and exceptions present. Thus as we pointed out earlier there was a significant difference, both in productivity and in stylistic effect, of our two sub-processes, d-deletion and də-deletion. Thus d-deletion is very productive (although not equally so after all vowels) in a colloquial style. We could represent this fact by a redundancy rule of the form:

Note that this rule has a very similar environment to J-INSERTION. This is because it is precisely the words that show the sub-process of d-deletion that undergo J-INSERTION. The feature [ - RR2] appears here since it is necessary to refer to another redundancy rule that states: -

RR2) X → [-D-DELETION]

which expresses the greater regularity that in general most of the words that could do not undergo D-DELETION. The fact that D-DELETION does not normally take place after /ī/ can be expressed by a third

11. [-syll] is required here to prevent D-DELETION occurring after long diphthongs, which would be analysed [-cons] and therefore provide an environment for this rule unless so blocked.

12. Here no more than two [-cons] 's are allowed since J-INSERTION does not occur after long diphthong's which would be [-cons] [-cons] [-cons].

[p. 428]

redundancy rule, preceding RR 1 as follows: -

RR Z) X → [-RR1] √Yīd + ə(C)#

The fact that verbieden can, despite its structure /vərbīd + ən/, undergo D-DELETION can be expressed by making it an exception to RR Z, ie marking it [ - RR Z ] in the lexicon. Thus we represent the truly exceptional facts in the lexicon, and sub-regularities in the redundancy rules which must come after the morphological rules proper, at the end of the morphological component, since we need to know the shapes of morphological affixes.

The forms with də-deletion show much less systematicity. Thus although the words that can undergo D-DELETION here, but don't, all have perhaps an archaistic flavour (except for the place names), it is impossible to draw up any kind of rules. Thus in compounds involving these words as first elements, we usually get one form or the other, with little scope for variation. For example wedergeboorte (re-birth) not* weergeboorte from we(d)er (again), Nederland (the Netherlands) not *Neerland but neerlandicus (scholar of Dutch language and literature) not* nederlandicus. Among the different words too, there seems to be considerable variation in usage. Thus lade (drawer) will be less normal than kade (quay). Generally with words of this group most people will just use one from themselves though they will of course understand both. Since each word behaves in its own particular way, its behaviour is a matter for the lexicon. However in as much as two forms exist for these words, related in a systemic way, this has to be accounted for. We will do this by saying that, when these words, for which the speaker has alternate forms, with and without /d/, are used without the /d/, the conditions determining this varying with the word, the feature [ - RR 2 ] will be assigned, presumably dependent on conditions stated within a particular lexical entry.

Another problem arises in those verbs where D-DELETION has analogically spread to forms where the phonological environment for D-DELETION is not present. Thus in rijden (/r∊jdə/) and houden (/hawdə/), the infinitive or present tense plural forms of the verbs to ride and to hold, we expect D-DELETION to apply. However it also applies in the first person singular present tense form, giving us rij (/r∊j/) and hou (/haw/) respectively. Cases like these can be characterized as positive exceptions tot D-DELETION. This seems to be possible with all verbs in /-awd-/ and / -∊jd / which are able to undergo D-DELETION. We could express this fact by a redundancy rule as follows:

[p. 429]

Note that this redundancy rule has quite a different effect from RR 1). And herein lies the reason why we specified the də-deletion words as [ - RRα]. In the case of words lexically marked as [ - RR 2 ] or words where this is a feature introduced by a redundancy rule, like RR 1, we have to do with a set of forms that are subject to what basically is a minor rule. But they are only affected by this rule when the right context is present. Thus the adjective oud (old) will be marked by RR 1 as being unaffected by RR 2 and therefore in effect subject to D-DELETION. However D-DELETION will only actually take place in the attributive form oude (/awdə/) which will become ouwe (/ awwə/). The form oud (/awd/) will never be affected since the right context is not present. With rijden we have quite a different state of affairs. D-DELETION not only takes place where the right context is present, as in rijden (/r∊jdə/), but also in cases where it is not, as in rijd (/r∊jd/). Thus we have in rijden a case of a positive exception to D-DELETION, as we have said. Now the traditional treatment of negative exceptions, that is cases where we expect a rule to apply and it does noet, has involved their being marked with a rule feature like [ - Rule X ]. Now it is quite clear that with positive exceptions we have precisely the opposite state of affairs. Thus with these we have a situation where we expect a rule not to apply and it does. So if we are to treat exceptions consistently we should assign positive exceptions the feature [ + Rule X ] which is what we did above with RR 3. For this reason we have to use some other method of marking forms that can undergo minor rules. Thus we can define a redundancy rule, like RR 1 above, that does this indirectly. What we have done is to say that forms that do undergo a particular minor rule can be regarded as negative exceptions to the redundancy rule stating the more general truth that forms normally do not undergo this rule. Thus whenever a form has the feature [ - RR2], whether this was present in the lexicon or introduced by a redundancy rule, this will precent that form from undergoing RR2 and so prevent it from being made a negative exception to D-DELETION.

D-DELETION will then apply to it and effect a change only if the right context is present.

A similar state of affairs reveals itself with the verbs laden (to load) and raden (to guess). Here we get not only D-DELETION where we do not expect it to apply but also J-INSERTION. Thus we have forms like:

raden /rādə/∼	raaien /rājə/ guess (pres. plur.)
laden / lādə/∼	laaien / lājə/ load (pres. plur.)

[p. 430]

raad / rād/→/rāt/∼^13.	raai /rāj/	guess (1st sing.)
laad / lād/→/lāt/∼	laai /lāj/

However if we were to say that these verbs were positive exceptions to J-INSERTION, what would this imply? It would imply that )Φ./j/ whatever the environment. But how do we know which of the three zeroes in /Φ r Φ ā Φ/, which is what we would have after the application of D-DELETION, to replace by /j/? We get a similar problem in the paradigm of doen(to do):

deden /dēdə/∼	deeën /dējə/ did (past plur.)
deed /dēd/→/dēt/∼	dee /dē/ did (past sing.)

In the case of the singular form, deed we have clearly an irregular application of D-DELETION. But how do we know which /d/ is to be deleted. If a form is a positive exception to a rule this means that the rule will operate whether of not the context is met. Thus here D-DELETION would apply, one would suppose, as if it were
d → Φ
that is, as if it were context free. But this would mean that it could apply twice, with the result /ē/, which is patently absurd. It would also be impossible for us to say that only the second /d/ was a positive exception to D-DELETION since it has always been assumed in generative phonology that it is not segments that can be exceptional but forms. Segments behave exceptionally by virtue of their forming part of a larger whole that is itself exceptional. Thus we will seemingly have to alter our conception of how positive exceptions to rules are affected by these rules.

Now, in the case of deed and the other positive exceptions to D-DELETION, this rule appears to operate as if the part of the context to the left of the environment bar were still effective, ie. as if the rule were operating as follows:



In the case of raad J-INSERTION also appears to operate as if the part of the context to the left of the environment bar were still effective, ie. as if it were operating thus:

Thus it is clear that these positive exceptions still have to conform to part of a rule-environment (For an attempt to specify the conditions governing this more precisely, see Smith (in preparation)).

13. This by the rule of final devoicing of stops and fricatives, which we will not discuss here.

[p. 431]

If we examine again the verbs with irregular application of D-DELETION and J-INSERTION, some interesing patterns can be observed

Unaltered form	Altered form	Exceptional application of
/Ik rād/ (I guess)	/Ik rāj/	D-DELETION J-INSERTION
/Ik dēd/ (I did)	/Ik dē/	D-DELETION
/Ik r∊jd/ (I ride)	/Ik r∊j/	D-DELETION
/Ik hawd/ (I hold)	/Ik haw/	D-DELETION

Now, these irregular applications of D-DELETION and J-INSERTION tend to result in less variable paradigms. Thus /rāj/ in the singular now has the normal relationship to the plural /rājə/ in that it is the same, lacking only the plural suffix /- ə/. Historically we can express the extension of these rules to the singular as:

Earlier Situation		Later Situation
Unaltered form Altered form		Unaltered form Altered form
sing rād	-	rād	rāj
plur rādən	rājən	rādə(n)	rājə(n)

If we look at the form /dē/ (did), we can see that it is not exactly parallel to the plural form /dējə/. The reason is that no application of J.-INSERTION has taken place. If it had we would have had /dēj/. But it is not possible for /ēj/ to occur finally. Similarly /r∊jj/ and /haww/ are also not possible. In sub-dialects of Standard Dutch where high-mid vowels are diphthongized however we have a less clear situation. My impression is however that these diphthongal pronunciations of /ē/, /ō/, // are equivalent in length to the lax diphthongs /∊j/, /aw/, /äẅ/ and not to the long diphthongs /ēw/, /ōj/ etc. If this is so we can represent them as /ēj/, /ōw/ and //, so that the /ē j/ deriving from /ē/ will still [-long] [-long] [-long] [- long]
be distinct from the /ēj/ we would get if J-INSERTION applied to /d ē/. We need a rule here then. HIGH MID TENSE VOWEL DIPHTHONGIZATION

[p. 432]

This rule will analyse long vowels as two shorts and express the change as eg. /ēē/ → /ēj/. Thus we assume that J-INSERTION will only apply exceptionally to singular forms whose corresponding plurals normally undergo it when the resulting sequence is one that matches those strings possible in underlying forms, eg. /āj/. An alternative explanation would be that J-INSERTION does in fact analogically apply and that the resulting impossible sequences will be corrected by later rules. In the first place we would have a derivational constraint, in the second a conspiracy.

Another problem arises with the plurals of certain nouns. Thus if we consider the following forms:

	singular		plural
a)	lade	/lādə/	laden	/lādə(n)/	drawer(s)
	la	/lā/	la's	/lās/
b)	slede	/slēdə/	sleden	/slēdə(n)/	sledge(s)
	slee	/slē/	sleeën	/slējə(n)/
c)	weide	/w∊jdə/	weiden	/w∊jdə(n)/	meadow(s)
	wei	/w∊j/	weien	/w∊jə(n)/

since lade has a different plural affix-en, in the non d-deleted form from that in the d-deleted form, -s, and since the intuitions of speakers are that sleeën is the plural corresponding to slee, and similarly that weien is the plural corresponding to wei, we must conclude that the d-less plurals are to be derived from the (to be) d-less stems and not from the plural forms with d's. This is only of significance with words in -ade (or perhaps more generally words in -de with back vowels). La for instance has the normally s-plural that occurs with words ending in a stressed back vowel. However at the time when the morpheme (plural- is rewritten as /s/ or /ən/, La still has the form lade, albert marked with the feature [ - RR 2 ] that will ensure its undergoing D-DELETION. There are two ways of solving this problem. We could say that before we assign the phonological shape of the plural morpheme we look ahead in the derivation to see what is going to happen to the stem. This would be a derivational constraint of the sort known as ‘peeking’ (see Hill, 1970; Cornulier, 1972). An alternative solution would be to list among the various situations where the /s/ plural allomorph appears, the additional environment

This would be unnatural in the sense that it fails to explain that the reason this word gets an /s/ plural is that it comes to end in a back vowel. Thus if we do not wish to make use of derivational constraints we are forced to adopt a solution of a rather ad hoc nature (compare also the alternative

[p. 433]

analyses of the positive exception operation of J-INSERTION above). There are problems also with the plurals of the d-forms lade slede, etc. These forms may only take the plural allomorph -en. In this they contrast with words like methode, palisade (which are never subject to D-DELETION) which behave like most other words in /-ə/ which can take either affix -s or -en with the stylistic difference that -en is more formal. Since our treatment of these first words has involved their being marked only when they will be subject to D-DELETION, the problem of how to distinguish them from words of the palisade-class is a serious one. If we wanted to utilise the notion of trans-derivational constraint, we could say that if another derivation of a word results in D-DELETION, then in any derivation that does not involve D-DELETION, the plural morpheme (is present) must be rewritten /ən/. This however would be an unfortunate step since the idea of trans-derivational constraint is so powerful (see Lakoff 1970 for more details on this).

Now we have to confess that there are a number of forms that are just plain exceptional! These fall into a number of types. First we have the forms moei (/mūj/) and kaai (/kāj/), which are irregularly formed from moede and kade respectively. The more normal d-deleted forms ka and moe exist beside these. The irregular forms can be derived by the irregular application of J-INSERTION and SCHWA-DELETION. This gives us the derivations

/kādə/	/mūdə/
kāə	mūə	D-DELETION
kājə	mūjə	J-INSERTION
kāj	mūj	SCHWA-DELETION

The second type of exception is typified by neer (/nēr/), an alternative form of neder (down), which latter must be analysed as /nēd+ər/ in the light of beneden (underneath), which would be /bə+nēd+ən/. However we do not get /nējər/ which we would expect (cf. /bənējə/). Instead we have the irregular non-application of J-INSERTION. Therefore this form must be marked in the lexicon as [-J-INSERTION]. A similar case is mee (/mē/), a variant of mede (with (particle)). This must be /mēd+ə/ since we also have met (with (preposition)), which we can derive from underlying /m∊d/^14. Here we have another negative exception to J-INSERTION. Possibly a parallel case is kou (cold (noun)). Since we have koud (cold(adjective)) it is possible that we will have to say that here we have *koude, underlying /kawd + ə/, with obligatory D-DELETION. This form will then be marked [- RR2] [- J-INSERTION] in the lexicon.

Thirdly we have some compounds without the application of J-INSERTION we would expect. We have bruidegom-bruigom (bridegroom), underlying [#[#bräẅd+ə#][#gom#]#] and a number of compounds

14. The vowel alternation found here will not be discussed. Suffice to say that the alternation in /m∊d/∼/mēdə/ is parallel to that found in certain nouns (also verbs and adjectives). Eg.

Singular	plural
/sxlp/	/sxēpə/	ship(s)
/w∊x/	/wēxə/	road(s)
/bad/ → /bat/	/bādə/	bath(s)
/xƆd/ → /xƆt/	/xōdə/	god(s)

For the final devoicing here compare note 13.

[p. 434]

involving the first element kleer (clothes) eg. kledermaker - kleermaker (tailor), underlying [#[#klēd+ər#][=māk+ər#]#]. In this latter case we might assume that /klēd+ər/, (actually an old plural) had been reanalysed as /klēdər/ without internal structure. This would be difficult to defend, however, in the face of aankleden (to dress) (/ān## klēd+ə/ which becomes aankleeën ([ānklējə]).

A last problem is caused by the word poeder (powder), which cannot have any internal structure, but still undergoes J-INSERTION after D-DELETION, ie [pūdər]∼[pūjər]. This is a positive exception to J-INSERTION.

Thus what might seem to be a rather straight forward phonological process or processes has turned out to involve several interesting theoretical problems, to wit the treatment of exceptions, minor rules and redundancy, analogy, derivational constraints^15., including peeking and transderivational constraints. This kind of situation appears to be typical in Dutch phonology. Simple rules hardly exist!

University of Amsterdam

15. Our alternative analyses in the two examples we give of derivational constraints can be examined in the light of the recent exchange involving Baker & Brame (1972) and Lakoff (1972).

[p. 435]

Appendix the rules

Redundancy Rules

RR Z) X → [- RR 1]√Y īd+ə(C)#

RR 1) X → [- RR 2]√Y[-syll][-cons]d+ə(C)#

(colloquial)

RR 2) X → [-D-DELETION]

RR 3) X → [+D-DELETION]√Y

(colloquial)

[p. 436]

Phonological Rules

D-DELETION d → Φ /[-syll][-cons]-ə

J-INSERTION Φ → j / [-cons]²₂ - + ə(C)#

SCHWA-DELETION ə → Φ / [-cons]²₂ + -

HIGH TENSE V SHORTENING



HIGH-MID TENSE VOWEL DIPHTHONGIZATION

W-ASSIMILATION j → w/w -

Bibliography

Baker, C.L., and M.K. Brame 1972 ‘Global Rules: a rejoinder’ In: Language 48, 51-75.

Chomsky, N. and M. Halle 1968 The Sound Pattern og English, Harper and Row.

Cornulier, B. de 1972. ‘A Peeking Rule in French’ In: Linguistic Inquiry 3, 226-227.

Fidelholtz, J.L. 1972. ‘On the Indeterminacy of the Representation of Vowel Length’. In: Papers in Linguistics 4, 577-594.

1970. ‘A Peeking Rule in Cupeno’ In: Linguistic Inquiry 1, 534-539

Kenstowicz, M. 1970. ‘On the Notation of Vowel Length in Uthuanian’. In: Papers in Linguistics 3, 73-113.

Lakoff, G. 1970. Some thoughts on Transderivational Constraints. Unpublished (OSCULD)

Lakoff, G. 1972. ‘The Arbitrary Basis of Transformational Grammar’ In: Language 48, 76-87

Pyle, C. 1970, ‘West Greenlandic Eskimo and the Representation of Vowel Length’ In: Papers in Linguistics 3, 115-146.

[p. 437]

Smith, N.S.H., in prep. Minor Rules, Exceptions and Redundancy.

[Sommaire]

Samenvatting

Dit artikeltje gaat over d-deletie in het Nederlands. Er wordt hier een poging gedaan een verklaring te geven voor verschijnselen zoals die aan twee soorten alternaties kunnen worden geillustreerd b.v. goede ∼ goeie, lade ∼ la. In het algemeen is het eerste proces produktief en het tweede niet. Desondanks kunnen beide processen door de d-deletie-regel worden verantwoord. De andere regels die direkt relevant zijn voor deze verschijnselen worden ook behandeld. Ook worden problemen besproken in verband met uitzonderlijke analoge vormen.