Style levels in conflict resolution

Marc van Oostendorp

Appeared in F. Hinskens, R. van Hout and L. Wetzels, eds. Language Variation and Phonological Theory Benjamins, Amsterdam, 1998.
The HTML of this version is not yet perfect.

1. Registers in Generative Grammar

A language user may accept certain linguistic forms in formal contexts that he would not accept in less formal contexts, or vice versa. A generative grammar however is always set up in such a way that it will either unconditionally accept or otherwise bluntly disallow a given input form. A generative grammar therefore in itself is not able to describe differences between formal and less formal styles of speech.

In order to account for linguistic phenomena related to the formality of the extra-linguistic context, generative phonologists have traditionally taken recourse to the notion style level or register (Selkirk 1972). Every register is a generative grammar in the technical sense just outlined and a language system consists of several registers. The language user selects his register within the system depending on the situation.

To my knowledge there has hardly been made any attempt to restrict the ways in which two styles within one language system can differ from each other. In many rule-based theories of phonology two styles A and B can differ from each other because A has more rules than B, of because the forms of some of the rules in A is more general (contains a smaller number of specified feature values) than the form of those in B, or because A and B have the same rules ranked in different orders. All of these possibilities are attested in the literature. This means that in principle two registers could differ from each other in the same way as any two grammars can differ. If we do not limit these differences, we would expect languages in which for instance one informal style has the phonology of informal Chinese, involving tone, tone sandhi, and a fairly well-developed syllable structure, whereas a more formal style would resemble for instance formal Turkish, involving vowel harmony, stress and a simple syllable structure. It seems improbable that such a language system can actually exist; the possible difference between any two registers within one language system seems much more restricted.

It should be noted that I use a restricted definition of the notion language system. It is of course possible that some language user speaks one language -- e.g. Spanish -- in informal situations and some other language -- e.g. English -- in formal situations but obviously in these cases the differences between those two languages have no limits but those imposed by universal grammar. The same is true for situations in which two distinct regionally determined dialects of the same language are at stake, for instance York English and RP. The cases I am considering here involve situatiosn in which the speaker is considered to use one and the same language or dialect in every situation, be it more and less formal variants of that language. In this article I discuss registers of Standard French, Standard Dutch and Standard Turkish, respectively.

During the past few years many generative phonologists have turned their attention from theories based on ordered rewrite rules to theories in which surface constraints play a key role. Most popular among these constraint-based approaches is probably Optimality Theory (OT, Prince and Smolensky 1993, McCarthy and Prince 1993ab, 1995). Within OT, a specific grammar is an extrinsic ordering of a universallly given set of constraints. Two grammars may differ from each other other only in their ranking of the same set of constraints.

If registers are grammars, this should also be the difference between two registers within one language system. It may seem that OT already solves the problem of how we may restrict the ways in which two styles can differ: it is impossible that constraints are present in one style but not in another, or that constraints are formulated more generally in one style than in the other. The only possible source of difference is the ranking of constraints. Yet this can still not be the complete solution since constraint ranking is a very powerful tool. If the assumptions of OT are correct, also the differences between the phonologies of informal Chinese and formal Turkish can be characterised by differences in constraint ranking only. For this reason it is still useful to try and find a more restrictive interpretation of the notion 'register'. I would like to propose such an interpretation in this paper.

One of the most important results of the work on the universal set of constraints (Con) within OT, is that this set can be divided into two subsets. The first subset consists of well-formedness, such as the constraint against onsetless syllable Onset or the constraint disallowing front rounded vowels *[-back, +round]. The latter constraint would be ranked high in languages, such as English, which disallow these vowels. The second subset consists of so-called faithfulness constraints requiring phonological output forms to be maximally similar to the input. [ It is possible that there are other types of constraints needed for some areas of analysis, such as constraints comparing output forms with other parts of the lexicon (Polgárdi 1994). This would not really affect my argument, although atmittedly it would complicate it.]

If Con did not also include faithfulness constraints, OT would make the incorrect predictions that (i) no language has an onsetless syllable and (ii) no language allows the front rounded vowel /ø/. In the extreme case, all words in all languages would surface for instance as [tata] or [ba] (Chomsky 1995). Faithfulness constraints block exactly this. In order to transform e.g. the English underlying form /pIstOl/ into a surface [tata], too many faithfulness constraints would have to be violated too many times.

Faithfulness constraints therefore have an important role to play in the overall architecture of Optimality Theory. I would like to suggest that they play an important role in the characterisation of the notion of register as well. I propose that the difference between two registers within one language system can be made in the following way: [ Cf. Itô and Mester (1995) for a proposal that a similar distinction can play a role in the organization of the lexicon. According to these authors, 'native words' in a given language show a relatively low ranking of faithfulness constraints, whereas 'non-native words' show a somewhat higher ranking.]

(1) The more formal the register, the higher ranked the faithfulness constraints.

Every language system has a fixed ranking of the faithfulness constraints and a fixed ranking of the well-formedness constraints. In addition, the hypothesis in (1) allows us to take any two registers in a language system and predict which of the two is the more formal. These are the reasons why I think the hypothesis in (1) is worth to be explored.

In this paper I will explain how three well-known style-related phenomena can be described using this hypothesis: French liaison, Dutch vowel reduction and Turkish vowel epenthesis. I will argue that an optimality theoretic account of these phenomena is superior to previous, rule-based accounts because it is more restrictive and therefore, offers a higher level of explanatory adequacy.

The remainder of this paper is organized in the following way. In section 2 I discuss French liaison, in section 3 Dutch vowel reduction and in section 4 Turkish vowel epenthesis. Section 5 will be concerned with two important questions connected to the hypothesis in (1): first, what is the status of this statement in the overall linguistic theory -- is it a universal principle or a mere artifact of some independent factor common to the languages or cultures of the French, Dutch and Turks; and second, given the fact that faithfulness constraints according to some authors play an important role also in language acquisition, what is the relation between the findings presented here and the ones of those scholars.

I will not discuss the exact formulation of the faithfulness constraints. I will assume that in general they have the following definition, following essentially McCarthy and Prince (1995):

(2) Input/output faithfulness (Correspondence, Parse/Fill, etc.):

Another formulation of essentially the same constraints is given in McCarthy and Prince (1993); the differences between the two views of faithfulness seem immaterial to the present discussion. The only reason why I use the definition in (2) is that it seems the most generally accepted among present-day OT phonologists.

2. French Liaison

The style related phenomena which are probably best studied among generative phonologists, are French enchaînement and liaison (Selkirk 1972, Encrevé 1988 and references cited there). Enchaînement is the process by which a word-final consonant is prononounced as the onset of the first syllable of the next word, if that word starts with a vowel. Liaison can be seen descriptively a special case of enchaînement, involving consonants which do not surface if they cannot occur in an onset position. In this section we only discuss liaison. [ In Wetzels (1987), it has been argued that one of the differences between liaison and encha înement and liaison involves sensitivity to style register. Whereas liaison does show up such a sensitivity, the level enchaînement is approximately the same for all registers. Although a thorough discussion of these enchaînement falls beyond the scope of this article, it should be noted that this is what the theory presented here predicts: liaison involves consonants which do or do not surface, hence faithfulness constraints. This type of process is therefore sensitive to the principle in (1). Enchaînement, on the other hand, only involves syllabification. Faithfulness is therefore irrelevant.]

The conditions under which liaison can occur are rather complex. They are not only dependent on register but also on parameters such as syntactic structure, syllable structure, sonority of the segments involved, frequency of a certain word combinations (see Encrevé 1988, De Jong 1989, etc. In this paper I will concentrate on two factors: syntactic structure and register. These seem also to be the relevant factors involved in the more general case of enchaînement. The others seem more specific for liaison.

It has been a tradition in the generative literature since Selkirk (1972) to follow Fouché (1958) and distinguish between three registers in French speech: an informal style I (Conversation Familière), a rather formal style II (Conversation Soignée) and a very formal style III (Discours and Lecture). We find some liaison in every style of speech, but the more formal the register, the more liaison we find.

This means that we find the minimal number of liaisons in the least formal register. Here we find liaison between a determiner and a following noun or adjective (3a), between an adverb and an adjective (3b), a (monosyllabic) preposition and its complement (3c) [ The situation is complicated by the fact that polysyllabic prepositions such as pendant 'during' only allow liaison in more formal styles of speech. I assume the reason for this is that polysyllabic forms behave more like lexical than like functional elements. This is probably due to prosodic restrictions on the division between functional elements and affixes on the one hand and lexical stems on the other; see Dixon (1977) for discussion of polysyllabic suffixes in Yindiny , Van Oostendorp (1994) on the Dutch suffix -achtig , Monachesi (1996) on the Italian pronoun loro.] and object clitics and the following word in the verb phrase (3d). In these examples an underscore indicates a liaison context:

(3) Style I Conversation Familière

We find these types of liaison also in other registers. In the more formal style of conversation soignée we also find liaison between an adverb and a following element in the verb phrase (4):

(4) Style II Conversation Soignée

In the most formal style of speech we find all of the liaisons illustrated in (3) and (4) and also between a noun and a postnominal PP (5a) or AP (5b) and between a verb and its object (5c):

(5) Style III Discours et Lecture

We can summarize the literature about the syntax-phonology interface with respect to these facts as follows, using the ideas about the syntactic extended projection of Grimshaw (1992): roughly, an extended projection is the complete projection of a given word, i.e. involving all syntactically realised complements, modifiers and functional heads. The only intervening bracket between the two segments therefore is the bracket of an incomplete extended projection. In style I we only find liaison if the first element is a functional element and the following word belongs to the same macroprojection. This is illustrated for the examples in (3) in (6a). In (6b) we give the tree structure of one of these examples for the sake of clarity:

(6) a. [DP s õ n [l a m ã

[AP t r è s [l i n k

[PP e n [l é t é

[VP v o u s [l e n

b. DPi

son NPi


The symbol [l denotes the bracket between a functional element in a macroprojection and a following head in the same projection. This is essentially the weakest type of bracket.

In style II we also find liaison between an element which is part of an adjunct in a extended projection and a following head within the same extended projection. [ In this particular example, the adverbial phrase consists of only one word -- be it a word which is probably lexical, not functional. The same facts are found if the adverbial is a phrase ( très naivement ).] In this case we have an extra bracket between the two words: the righthand bracket of the adjunct, indicated by the symbol ']f'.


naivement exprimé

In style III finally liaison can also cross the lefthand bracket ('[f') between one functional projection and another:

(8) a. v i n s [f [l a c u i r
g e n s [f [l a g e
>d o n n a i t [f [l u n c o u r s
b. DPi
les NPi
vins PPj
à cuire

We can summarize the differences between the three styles as follows:

(9) liaison in:
Style I : [f X [l Y ... ]]
Style II: [f X [l Y ... ]] [[ ... X ]f Y ]f
Style III: [f X [l Y ... ]] [[ ... X ]f Y ]f [f X [f Y ... ]]

In style I we are only allowed to let a syllable cross what is essentially the weakest syntactic boundary, [l. In style II we can also cross the somewhat stronger ]f, and in style III also the strongest boundary [f. In OT we can formalize these observations by introducing the three constraints of (10) and ranking them in the order of (11). [ One boundary, ] l , is obviously missing from the discussion. The reason for this is that French extended projections are quite uniformly head-initial. As a consequence two words are never only seperated by only a left lexical bracket, as the left boundary of a lexical projection is always at the same time also the left boundary of a functional projection. The relative ranking of * Cross -] l therefore cannot be determined. For this reason I will ignore this constraint.]

(10) *Cross-[l: A syllable should not cross the left boundary of a lexical projection.

*Cross-]f: A syllable should not cross the right boundary of a functional projection.

*Cross-[f: A syllable should not cross the left boundary of a functional projection.

(11) *Cross-[f _ *Cross-]f _ *Cross-[l

The relative strength of the boundaries is reflected in the hierarchy in (11). I do not know whether this ranking is specific for French or universal, nor whether it should be stipulated as is done here or can be derived from universal principles.

The Cross constraints are pure wellformedness constraints: they express conditions on the output form of the linguistic derivation alone. In liaison, another constraint is at stake, a constraint which lets the consonant surface whenever liaison is allowed. I call this constraint SurfaceC, an instance of the general constraints scheme in (2):

(12) SurfaceC: A consonant in the underlying form should correspond to a consonant in the surface form.

SurfaceC is a faithfulness constraint: it puts a requirement on the relation between input and output and it therefore is a special case of the input/output constraint in (2).

According to SurfaceC, if /s o n/ is the underlying form of a word, [son] is preferred as the ouput form over [so] or [sõ].

According to the hypothesis in (1) we predict that the difference in register among the various dialects is mainly a difference in the relative ranking of (12): the more formal the style of speech, the more highly ranked the constraint SurfaceC. It appears to be the case that we can describe the facts above in exactly this way. We give the three rankings for the three different styles:

(13) Style I : *Cross-[f _ *Cross-]f _ SurfaceC _ *Cross-[l

Style II: *Cross-[f >> SurfaceC >> *Cross-]f >> *Cross-[l

Style III: SurfaceC >> *Cross-[f >> *Cross-]f >> *Cross-[l

To make this clear I will first give the tableau for the most informal register. The relevant consonant-vowel sequences in the forms son amant (3), naivement exprimé (4) and vins à cuire (5). In these examples, n, t and s are used to indicate that the segment at issue is not phonetically realised.

(14) Style I

*Cross-[f *Cross-]f SurfaceC *Cross-[l
F n [l a (3)

n [l a

t ]f E (4)
F t ]f E

s [f a (5) *! *
F s [f a


In son amant only a left lexical bracket is crossed. Since the relevant constraint is ranked below SurfaceC, the latter constraint takes effect and lets the consonant surface. In the other cases, the relevant Cross constraints are ranked higher than SurfaceC, blocking liaison.

In style II, the faithfulness constraint is ranked somewhat higher. The effects of this are shown in the tableau in (15):

(15) Style II

*Cross-[f SurfaceC *Cross-]f *Cross-[l
F n [l a (3)

n [l a

F t ]f E (4)

* *
t ]f E

s [f a (5) *!
* *
F s [f a

In this case, SurfaceC also outranks *Cross-]f. The righthand boundary of functional projections may therefore now also be crossed by a syllable. Only in cases such as vins à cuire liaison is still not allowed. In order to make this happen, the constraint SurfaceC should be ranked even higher, outranking all relevant Cross constraints. This happens in the most formal register -- the one Fouché labeled Discours et Lecture:

(16) Style III

SurfaceC *Cross-[f *Cross-]f *Cross-[l
F n [l a (3)

n [l a *!

F t ]f E (4)

* *
t ]f E *!

F s [f a (5)
* * *
s [f a *!

The faithfulness constraint now outranks all the relevant wellformedness constraints. Liaison therefore becomes possible (and obligatory) in this register. We thus divide the language system in a discrete -- but possibly large -- number of registers. In every register liaison is allowed in certain syntactic contexts only, and it is obligatory wherever it is possible. This is not to say, of course, that a language user might not switch registers within a given utterance.

The hypothesis in (1) makes the correct prediction for French liaison. Two points have to be added, however. In the first place, liaison is never unrestricted, not even in the most formal style of speech. This means that there are more relevant constraints than have been introduced here. Some of them should outrank SurfaceC even in the most formal registers. The most trivial instance is probably a constraint *Cross-[S, prohibiting liaison across sentence boundaries. This constraint -- if indeed it is one -- would be ranked over SurfaceC also in Style III.

This brings us to our second point. If the hypothesis in (1) were unconditionally true, we would expect there to be an extremely formal register in which liaison would be possible between any two words which occurred adjacent to one another, even if they would occur in different sentences. Inversely, we would also expect there to be some very informal register in which no liaison was allowed whatsoever. Yet these predictions are not borne out. It is therefore necessary to put some upper and some lower limit on every faithfulness constraint within the language system. In the case of French, we would have to say that SurfaceC always dominates at least the constraint *Cross-[l and it should always be dominated by *Cross-[S. Between these two extremes, the choice is relatively free, i.e. determined by considerations of style alone.

3. Dutch Vowel Reduction

We now turn to another phenomenon which has a clear relation to differences in style of speech: Dutch vowel reduction. Like many (Germanic) languages, Dutch has an active process reducing full vowels in unstressed position to schwa. Unlike in English, however, the reduction rule is not absolutely obligatory. Whether or not a vowel reduces is dependent on many factors, including register, vowel quality, level of stress, position in the word, frequency of use of the word, etc. I refer to Martin (1968), Kager (1989), Van Oostendorp (1995) and references cited there for discussion. Here, I will concentrate on the first three factors and their interaction.

As far as register is concerned we can observe that there is more reduction in informal styles than in formal ones. As for vowel quality, we can observe the following: /e/ reduces always in unstressed position; /a/ reduces less often than /e/; /o/ and /ø/ less often than /a/; and /y/, /u/ and /i/ reduce hardly at all. [ Next to the set of tense vowel mentioned in the text, Dutch also has the lax vowels /A, E, oe, O, I/. These hardly ever reduce but it is unclear whether this has to do with their vowel quality or with the fact that these vowels always get at least secondary stress. I refer to Van Oostendorp (1995) for discussion.] With respect to the level of stress, next to two levels of stress (primary and secondary), we can distinguish between two types of unstressed position: one immediately following another unstressed position (called a 'semi-weak position') and the other one in other positions in the word (called a 'weak position'). As an example, let us consider the Dutch word fonologie /fonoloxi/ 'phonology'. This word has primary stress on the final syllable gie and secondary stress on the first syllable fo. The syllables no and lo are unstressed -- no is weak and fo is semi-weak -- and in the most informal registers both can be reduced to schwa: fon@l@gie. Yet in somewhat less informal styles we can see a difference: in those styles only the vowel in the weak syllable no can be reduced: fon@logie. The form fonol@gie is not found in any style of speech: the semi-weak syllable lo can only reduce if the weak syllable does as well.

The three factors register, vowel quality and stress level interact with one another in a rather complicated way. The following table is copied in a slightly adapted form from Kager (1989):

(17) Weak Positions Semi-weak positions

/e/ formal formal
/a/ formal semi-formal
/o/, /ø/ semi-formal informal
/y/, /u/, /i/ informal excluded

In the Dutch literature (Booij 1981, 1995, Kager 1989) one usually distinguishes between three registers, informal, semi-formal and formal.

Some examples are given in (18) below. In (18a) and (18b) it is shown that unstressed /e/ should always be reduced, even in weak positions and the more formal styles of speech. [ The given judgments represent those of my own idiolect. In some other variants of Dutch, forms like [ teját@r] are considered quite good. Importantly, however, reduction of /e/ is considered more normal in these dialects than reduction of any other vowel.] The word in (18c) shows that the /a/ in semi-weak position can stay unreduced in more formal registers. The example in (18d) repeats the facts of the example fonologie. Example (18e) shows that the high vowel /i/ cannot be reduced in any register:

(18) a. theater `theatre' [t@ját@r/?*teját@r]
b. plezier `fun' [pl@zír/?*plezír]
c. democratie `democracy' [dèmokr@tí/dèmokratí]
d. fonologie `phonology' [fòn@l@xí/fòn@loxí/fònoloxí]
e. kabinet `cabinet' [*kàb@nÉt/kàbinÉt]

We will formalize the notion of vowel quality by reference to standard features. We propose the following features play a role in the specification of each of the vowels. Schwa is supposed to be either unspecified or otherwise to have the default specification [-front, -low, -round, -high]. [ I will remain neutral on the question of whether the feature values [-front], [-low] etc. are absent (or 'unspecified') at any stage of the derivation, or whether they are only unmarked feature values. For the purposes of the present exposition, it is irrelevant whether e.g. the vowel /e/ is specified only as [+front], or as [+front, -low, -round, -high].]

(19) /e/ [+front]
/a/ [+low]
/o/ [+round] /ø/ [+round, +front]
/u/ [+high, +round] /y/ [+high,+round,+front] /i/ [+high, +front]

Following standard assumptions (e.g. Mascaró 1987) we analyse reduction as loss of marked feature values. In the case of /e/ this means loss of [+front], in the case of /a/ loss of [+low], etc. I propose the following constraints are responsible for the blocking of vowel reduction:

(20) Parse-[+high]: [+high] in the underlying form should be present in the output form.
Parse-[+round]: [+round] in the underlying form should be present in the output form.
Parse-[+low]: [+low] in the underlying form should be present in the output form.
Parse-[+front]: [+front] in the underlying form should be present in the output form.

These constraints clearly are faithfulness constraints, requiring the output to be maximally similar to the input. Reduction of /e/ to schwa gives a violation of Parse-[+front] only, whilst reduction of /y/ gives violations of Parse-[+front], Parse-[+round] and Parse-[+high].

In addition to these faithfulness constraints blocking reduction, we also need the following two wellformedness constraints which force unstressed vowels to reduce to schwa:

(21) Reduce-1: Weak and semi-weak positions should be schwa.
Reduce-2: Weak positions should be schwa.

We can now establish the relative ranking of the several wellformedness constraints and the one of the several faithfulness constraints. The relative ranking of the faithfulness constraints should of course observe the differences between the vowel qualities observed above: Parse-high should be ranked very high because high vowels never reduce, while Parse-[+front] is ranked very low, since it can always be violated when /e/ reduces to schwa.

The relative ranking of the two wellformedness constraints is determined by the Paninian theorem of Prince and Smolensky (1993): the more specific constraint Reduce-2 should outrank the more general Reduce-1, otherwise it would not have made sense to introduce Reduce-2 into the analysis in the first place. We thus have the following rankings as part of the Dutch language system: [ It might well be that this ranking reflects some universal difference between these features, rather than a language-specific ranking for Dutch; cf. Van Oostendorp 1995.]

(22) Parse-[+high] >> Parse-[+round] >> Parse-[+low] >> Parse-[+front]
Reduce-2 >> Reduce-1

Again we assume that the relative ranking of the faithfulness constraints and wellformedness constraints with respect to each other is free in the language system

and that that the Parse constraints should have a higher relative ranking in the formal registers. It appears that we can formalize the three traditional styles from the Dutch literature in exactly this way:

(23) formal: Parse-[+high] >> Parse-[+round] >> Parse-[+low] >> Reduce-2 >> Reduce-1 >> Parse-[+front]

semi-formal: Parse-[+high] >> Reduce-2 >> Parse-[+round] >> Reduce-1 >> Parse-[+low] >> Parse-[+front]

informal: Parse-[+high] >> Reduce-2 >> Reduce-1 >> Parse-[+round] >> Parse-[+low] >> Parse-[+front]

In the most formal register, all faithfulness constraints are unviolated, except Parse-[+front]. This causes all vowel reduction to be blocked, except for that of /e/:

(24) formal

P-[high] P-[round] P-[low] Rd-2 Rd-1 P-[front]

*! *
F pl@zir

F fatal

* *


F fonologi

* **



F bizAr

* *
b@zAr *!

In the semi-formal registers also /a/ and /o/ in weak positions can be reduced. This can be achieved by ranking both Parse constraints somewhat lower relative to the wellformedness constraints.

(25) semi-formal

P-[high] Rd-2 P-[round] Rd-1 P-[low] P-[front]

F pl@zir


F f@tal


F fon@logi

* *



F bizAr

b@zAr *!

In the most informal style of speech, all faithfulness constraints get ranked below the wellformedness constraints except Parse-[high].

(26) informal

P-[high] Rd-2 Rd-1 P-[round] P-[low] P-[front]
*! *

F pl@zir

*! *

F f@tal

*! **


*! *

F fon@l@gi


F bizAr
* *

b@zAr *!

Again the hypothesis in (1) seems to make the correct prediction, but again we should impose some limits on the variation: Parse-[high] always dominates the wellformedness constraints, which in turn always dominate the constraint Parse-[front].

The formalisation of (23) brings to light the rather arbitrary character of the division into three registers, usual in the earlier generative literature. There seems to be little reason to make exactly this distinction and not to distinguish e.g. two registers or four. Given the limits just outlined, we now actually have the opportunity to distinguish more differences in style. For instance we also can have a ranking like the following:

(27) Parse-[+high] >> Parse-[+round] >> Reduce-2 >> Reduce-1 >> Parse-[+low] >> Parse-[front]

In this style only low and front unstressed vowels (i.e. /e/ and /a/) could be reduced, regardless of their precise position. Given the rather arbitrary nature of the division into styles in (23), there seems to be no reason why we shouldn't consider this to be a valid style of speech in Dutch as well as any of the others.

4. Turkish Vowel Epenthesis

The next phenomenon to consider is vowel epenthesis in Turkish. Informal Turkish doesn't allow complex onsets. If words with such onsets are borrowed into the language, the situation is repaired by epenthesizing a vowel between the two consonants.

We can distinguish between two styles of informal Turkish. In the most informal style, the epenthetic vowel shows harmony on the features roundness and backness with the following vowel. In a somewhat less informal style, the epenthetic vowel is the default (high central) vowel I. Some examples are given in (28) (from Clements and Sezer 1982):

(28) careful less careful colloquial

form form form

`prince' prens pIrens pirens

`test' prova pIrova purova

`protest' protesto pIrotesto purotesto

`France' fransa fIransa fIransa

`transit' transit tIransit tIransit

`direct' drek dIrek direk

`announcer' spiker sIpiker sipiker

`Dracula' drakula dIrakula dIrakula

`crèche' kreû kIreû kireû

`group' grup gIrup gurup

In order to analyse these facts we need four constraints. Two wellformedness constraints (29a) and (29b), requiring vowel harmony and vowel epenthesis respectively, and two faithfulness constraints blocking insertion of non-underlying material: (29c) blocks insertion of vowel roots and (29d) blocks insertion of association lines.

(29) a. Spread F If a feature F is linked to one segment in a word, it should be linked to all segments in that word.; the relevant instances for this constraint scheme in Turkish are Spread [front] and Spread [round].

b. NoCluster: *C1C2 in the onset

c. NoEpenthesis: A vowel in the ouput form should be present in the underlying form.

d. NoSpreading: An autosegmental association between a feature and a segment in the output form should be present in the underlying form.

All of these constraints can be found in the literature, although one could argue about their actual formulation as give here. In any case it seems reasonable to distinguish between two types of process, feature spreading and vowel epenthesis, since the two processes seem separate: Turkish has harmony also in cases where there is no epenthesis, for instance in stem-affix combinations, and Turkish also has epenthesis in cases where there is no harmony, viz. in the examples of 'less careful forms' in (28) above.

The relative ranking in the language system of the wellformedness constraints Spread and NoCluster in this case is as hard to establish as the relative ranking of the faithfulness constraints with respect to each other. This is not crucial however, since we can still see how the hypothesis in (1) gives the correct registers (plus one more, to be discussed below):

(30) a. careful register: NoEpenthesis >> NoCluster,

NoSpreading >> Spread

b. less careful register: NoCluster >> NoEpenthesis,

NoSpreading >> Spread

c. colloquial register: NoCluster >> NoEpenthesis,

Spread >> NoSpreading

Again, there should be put some upper and some lower limits on the variation: every style has vowel harmony in the case of stem-affix combinations, meaning that we should probably refine the definitions of the constraints Spread and NoSpreading in the appropriate way, perhaps adding some more constraints to be able to account for the full range of complexity of the harmony processes in the language.

In the most careful register, both faithfulness constraints dominate the wellformednesss constraints. Therefore, epenthesis and harmony are disallowed:

(31) careful register

NoEpenthesis NoCluster NoSpreading Spread
F prens

pIrens *!

pirens *!

In a somewhat less careful register, the NoSpreading constraint still dominates its wellformedness counterpart Spread but NoCluster and NoEpenthesis have traded places:

(32) less careful register

NoCluster NoEpenthesis NoSpreading Spread
prens *!

F pIrens
* *!

In the colloquial register, finally, both well-formedness constraints outrank their faithfulness competitors and vowel epenthesis and harmony freely occur:

(33) colloquial register

NoCluster NoEpenthesis Spread NoSpreading
prens *!

* *!
F pirens

One other possible ordering of the constraints would be the following:

(34) NoEpenthesis >> NoCluster,

Spread >> NoSpreading

This would have approximately the same level of informality as the 'less careful' register: one faithfulness constraint dominates a wellformedness constraint and one faithfulness constraint is crucially dominated by a wellformedness constraint. Yet the result of NoEpenthesis >> NoCluster is that no epenthetic vowels occur. This makes the actual ordering of Spread and NoSpreading irrelevant and the register in (34) would have the same outputs as the one in (30a).

5. Discussion: Universal Principles and Language Acquisition

We have now seen that style related phenomena in French, Dutch and Turkish seem to pattern with the predictions made by the principle in (1). In this section we will briefly speculate on some implications of this finding.

In the first place, it becomes very attractive to the generative linguist to now start considering (1) as a universal principle of the language system. The learnability problem which is one of the foundations of generative grammar (e.g. Chomsky 1986) obviously also arises in the case of registers. It seems unlikely that a child will learn three (or more) separate grammars like she would have to if there were no formal relation between the registers of one language system. If (1) were to be a universal principle, the child would only have to learn (i) the relative rankings among the faithfulness constraints, (ii) the relative rankings among the wellformedness constraints and (iii) the upper and lower limits for the (upper and lower) faithfulness constraints with respect to the hierarchy of wellformedness constraints. This is still a considerable task but in any case it is easier to fulfill than acquiring three completely different systems.

The question now arises whether we have any right to postulate (1) as a universal principle. For instance, William Labov (p.c.) has pointed out to me that French, Dutch and Turkish all have a fairly well-established spelling tradition. It might be that in such language (1) is actually a corollary of (35):

(35) The more formal the style of speech, the closer the pronunciation to the spelling.

If spelling reflects the underlying form of words, this would explain why (1) works in the languages involved, but it would not allow us to make (1) into a universal principle. Now there is an obvious way to test these two competing hypotheses, viz. by examining register related phenomena in languages which do not have an established spelling system. It is rather hard to find the relevant data and I therefore prefer to leave this issue open for further research.

Many issues therefore remain to be explored. But I believe the three case studies discussed in this paper show an interesting correlation worth to be explored.

6. Conclusion

In this paper we have seen that the differences between style levels within a certain language can be insightfully described if we adopt one of the central assumptions of Optimality Theory: that grammars consist of a ranking of universal constraints. Dividing the set of constraints into two subsets -- the set of wellformedness constraints and the set of faithfulness constraints -- makes the description of style levels in a given language much easier and furthermore helps us solve some of the problems involving the learnability of style levels.


Acknowledgments The author wants to thank Ben Hermans, Frans Hinskens, Harry van der Hulst, William Labov and Leo Wetzels for comments. All usual disclaimers apply.


Alderete, J. (1995) 'Faithfulness to Prosodic Heads.' Manuscript, University of Massachusetts.

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

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

Chomsky, N. (1986) Knowledge of Language. Praeger, New York.

Chomsky, N. (1995) The Minimalist Program. MIT Press, Cambridge, Mass.

Clements, N. and E. Sezer (1982) 'Vowel and Consonant Disharmony in Turkish.' In: H. van der Hulst and N. Smith, eds., The Structure of Phonological Representations. Foris, Dordrecht. pp. 213-255.

Dixon, R. (1977) 'Some phonological rules in Yidiny.' Linguistic Inquiry 8:1-34.

Encrevé, P. (1988) La liaison avec et sans enchaînement; Phonologie tridimesionelle et usages du français. Éditions du Sueil, Paris

Fouché, P. (1958) Traité de prononciation française. Paris, Klincksieck.

Gnanadesikan, A. (1995) 'Markedness and Faithfulness Constraints in Child Phonology.' Manuscript, University of Massachusetts at Amherst.

Grimshaw, J. (1992) 'Extended Functional Projections.' Talk presented at Utrecht Lexicon Workshop.

Kager, R. (1989) A Metrical Theory of Stress and Destressing in English and Dutch. Foris, Dordrecht.

Itô, J. and A. Mester (1995) 'The Core-Periphery Structure of the Lexicon and Constraints on Reranking.' In: J. Beckmann, Laura Walsh and Suzanne Urbanczyk, eds., UMOP 18; Papers in Optimality Theory. GLSA, Amherst. pp. 181-209.

Jong, D. de (1989) The Sociolinguistic Aspects of French Liaison. Doctoral Dissertation, Free University of Amsterdam.

Kager, R. (1989) A Metrical Theory of Stress and Destressing in Dutch and English, Foris, Dordrecht.

Martin, A. (1969) 'Klinkerreductie: Een casus.' Manuscript Utrecht University.

Mascaró, J. (1987) 'Vowel Reduction as Deletion.' Handout Going Romance Conference, Utrecht.

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

McCarthy, J. and A. Prince (1993b) 'Generalized Alignment.' In:Geert Booij and Jaap van MArle, eds., Yearbook of Morphology 1993. Dordrecht, Kluwer. pp. 79-153.

McCarthy, J. and A. Prince (1995) 'Faithfulness and Reduplicative Identity'. In: J. Beckmann, Laura Walsh and Suzanne Urbanczyk, eds., UMOP 18; Papers in Optimality Theory. GLSA, Amherst. pp. 249-384.

Monachesi, P. (1996) A Grammar of Italian Clitics. PhD Dissertation, Tilburg University.

Oostendorp, M. van (1994) 'Affixation and Integrity of Syllable Structure in Dutch.' In: R. Bok-Bennema and C. Cremers, eds., Linguistics in the Netherlands 1994. John Benjamins, Amsterdam. pp. 46-57.

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

Polgárdi, K. (1994) 'Derived Environment Effects in Optimality Theory.' Manuscript, University of Leiden.

Prince, A. and P. Smolensky (1993) Optimality Theory; Constraint Interaction in Generative Grammar. Manuscript, Rutgers University, and University of Colorado.

Selkirk, L. (1972) The Phrase Phonology of English and French. PhD Dissertation, MIT.

Wetzels, L. (1987) 'The Timing of Latent Consonants in Modern French,' in Carol Neidle and Rafaël Nuñez Cadeño, eds., Studies in Romance Languages. Dordrecht, Foris. Pp. 283-317.