346 63 10MB
English Pages [133] Year 1994
THEORIEDES LEXIKONS Arbeiten des Sonderforschungsbereichs 282
Nr. 56 Recent Developments in Lexical Phonology Richard Wiese, ed.
Dlisseldorf, Februar 1994 Projekt C4
Table of Contents
Richard Mese: Preface
1
I. Morphology-phonology interaction
Geert Booij: Lexical Phonology: a Review Heinz Giegerich: Base-driven stratification: morphological causes and phonological effects of 'Strict Cyclicity'
3
31
Sharon Inkelas/Orhan Orgun: Level economy, derived environment effects and the treatment of exceptions
63
Richard Mese: Phonological vs. morphological rules: German umlaut and ablaut
91
II. Constraints on rules and representations Kees-Jan Backhuys/Mm Zonneveld: Accentual catalexis, non-accentual catalexis and peripherality
115
Tracy Hall:
f::U7,G2
+R Y37' J
q t:."}l,.,.
' .,,: "··
r
\ c)
Extrasyllabicity and resyllabification
151
Sharon Hargus: Historical Change in Athabaskan Rule Domains
167
Ellen Kaisse/Sharon Hargus: When do linked structures evade Structure Preservation?
is5
Andrew Spencer: Syllabification in Chukchee
205
List of contributors
227
Preface
Richard Wiese Heinrich-Heine-Universitiit Di.isseldorf
Ever since the phoneme was defined as that minimal element of sound structure which constitutes a lexically distinctive segment in a language, there is a special relationship between the word level and the theory of phonological structure. Lexical Phonology is a theory which is in line with the recognition that there must be a strong lexical element in any model of phonology. From the earliest research by Pesetsky (1979), Booij (1981), Kiparsky (1982) and Mohanan (1982) to the present (as documented in Kaisse & Hargus (1993) and in this volume), Lexical Phonology has been an attempt to find the principles behind phonological word forms. The discussion documented in this volume on current topics within Lexical Phonology is witness to the fact that debate on the proper way of characterising phonological word forms is alive and well. While there is hardly any statement with which all active researchers in the field would agree, there is considerable focus on the issues addressed. Therefore an attempt has been made to sort the contributions according to the general topics and problems they focus upon. In spite of the considerable overlap in the issues addressed in the contributions, it turns out that the papers can be arranged according to two central and recurrent issues of research within Lexical Phonology, namely that of the morphology-phonology interaction, and the question of how phonological rules should be properly constrained, both in their form and in their interactions. The current collection of papers is the outgrowth of a three-day workshop "Recent developments in Lexical Phonology'', held at the Heinrich-Heine-Universitiit Diisseldorf in August 1993. It was part of the research activities of the Sonderforschungsbereich Theorie des Lexikons, a special research unit financed by the German Research Foundation. Theorie des Lexikons, which combines the efforts of linguists at the universities in Diisseldorf, Wuppertal, and Cologne, has included research on questions of Lexical Phonology and Morphology since its beginnings in 1991.The meeting received some additional support from the funds for the Dutch/North Rhine-Westphalian scientific cooperation.
3
2
Besides the contributions collected in this volume, the following papers were read at the conference, but could not be presented for various reasons in this volume: - Abigail Cohn (Cornell University): Lexical Phonology and the status of phonetic and phonological rules, - Rene Kager (Utrecht University): Blocking effects in lexical stress rules, - Paul Kiparsky (Stanford University): Underspecification and dominance, - Mieke Trommelen (Utrecht University): Dutch evidence for the WORD-level model
Lexical Phonology: a Review Geert Booij (Vrije Universiteit Amsterdam / Holland Institute of generative Linguistics)
1.
Introduction
of Lexical Phonology. Lexical Phonology is the name of a theory concerning the interaction of phonology with morphology. The problem often encountered in discussions and evaluations of this theory is that 'Lexical Phonology' does not stand for one specific hypothesis but rather for a cluster of related, but not necessarily logically interdependent theoretical claims. My aim in this paper is to first discuss what the basic hypothesis of Lexical Phonology is, and to provide arguments for it. Subsequently, I will review a number of related, but logically independent hypotheses concerning the organization of the lexicon such as level ordering. I will argue that the theory of Lexical Phonology derives from some very simple basic assumptions concerning rule application.
2. The basic claim The basic hypothesis of Lexical Phonology (cf. Booij 1981, Kiparsky 1982) is the following: (1)
Morphology and phonology apply in tandem
To put it simply, you take a word, and apply the applicable phonological rules right away; you may then apply a morphological rule to that word, which creates a new domain of application for the phonological rules of the language, the second cycle, which in its turn can be input for another morphological operation that creates a third cycle, and so on. 1 In other words, hypothesis (1) predicts that2 (2)
a. the phonological rules of a language that apply to words apply cyclically, at least in principle; b. the morphological rules of a language may refer to derived (predictable) phonological properties of their input words.
The formulation "you take a word, and apply the applicablephonologicalrules" implies that it is words that form cyclic domains, not morphemes.This is in line with the conclusionsof Brame (1974) and Harris (1983) who argue that cyclic domains must be dominatedby a lexical category node. It is also in harmony with the word-based view of morphology. 1
2
The derived nature of cyclicityis argued for in Booij (1981) and Kiparsky (1982), who mentions Pesetsky (1979) as also having proposed this idea; the point that morphology may refer to derived phonological information is stressed in Anderson (1979: 136). Inkelas (1993a)also argues in favour of the derivednature of cyclicity.
5
4
possibilities for the application of the rule are created by the syntax: Dutch has a number of vowel-initial enclitics (weak forms of pronouns and determiners) that can be incorporated phonologically into the preceding prosodic word. Thus, we get new schwa-vowel sequences at the postlexical level that are subject to the rule of Prevocalic Schwa-deletion:
The cyclic application of phonological rules in its turn predicts that phonological and morphological rules may make use of phonological information that is no longer present in the phonetic form of words.
(3)
Below, I will summarize the evidence for these claims. Principle (1) is in fact a consequence of an even simpler idea, a principle of 'virtual necessity', as Chomsky would call it, namely the following minimal assumption: 3 (4)
Apply a rule when possible
This principle predicts, for instance, that the rule of word stress of a language applies immediately to a given word, before it is subject to morphological operations. Thus, cyclicity of stress assignment follows from principle (4). Of course, principle (4) does not predict that all phonological rules of a language apply in the lexicon. Phonological rules can be specified as to the domain in which they apply. Rules that apply within the domain of the syllable, the foot, or the prosodic word can already apply in the lexicon because these prosodic categories are already available during the construction of words (Booij 1988, Inkelas 1989, 1993a). On the other hand, many rules have domains larger than the word, e.g. the phonological phrase. Such rules are by definition postlexical (post-syntactic) rules since their applicability depends on the availability of domains created on the basis of syntactic structure. 4 . Secondly, principle (4) may also be restricted by specific constraints which force a rule to apply post-cyclically, even though the rule has the word as its domain. In particular, the cyclic application of rules has been argued to be subject to the condition of Strict Cyclicity that says that rules only apply cyclically in derived environments, at least when they apply in a structurechanging fashion (Kiparsky 1982). So we get three stages at which phonological rules can apply: (5)
cyclic level word level postlexical level
Note that this reconstruction of the basic idea of Lexical Phonology implies that a rule can apply at more than one level, as has been argued in Mohanan & Mohanan (1984) and Kiparsky (1985). This can be illustrated by the rule of Prevocalic Schwa-deletion in Dutch. This rule obligatorily deletes a schwa before an adjacent vowel within the domain of the prosodic word. Since the sequence schwa-vowel is a derived environment, the rule can apply cyclically, and there are independent arguments why cyclic application of this rule is necessary (see Booij 1981, Booij & Rubach 1984:8-9). Interestingly, new
'The point that Lexical Phonology is simply a consequence of this minimal assumptionis also made by Kaye (1992: 141). The principle 'Apply a rule when possible' does not exclude the possibility of extrinsic rule ordering. Extrinsic rule ordering may be required in case two qr more rules are applicableto a certain form in virtue of 'Apply when possible'. 4
(6)
I
Ik zette het .ik haalde hem
'I put it' /zeta at/ [zetat] 'I fetched him' /halda am/ [haldam]
A similar example of a rule that applies lexically (in this case at the word level), and reapplies to word+clitic combinations is the rule of Low Vowel Assimilation in Basque (Hualde 1991: 29-31). This view of Lexical Phonology (also advocated in Kiparsky 1985) goes against the older view that the three levels mentioned above imply that the phonological rules of a language are divided into three ordered blocks corresponding to these levels. Moreover, it also implies that there is no need to provide each rule with a label as to the block to which it belongs.' It may be useful to point out that this version of Lexical Phonology is also compatible with syntactic approaches to morphology, in which complex words are derived by means of syntactic transformations such as head-to-head movement. What Lexical Phonology means is that as soon as a complex word is created (either by a morphological or a syntactic operation) the resulting word is susceptible to phonological rules that apply within domains up to (and including) the prosodic word. That is, one might have Lexical Phonology without Lexical Morphology. The view outlined above implies, to begin with, that syllabification of words is a cyclic process, and there is ample empirical evidence for this claim, the hypothesis of continuous syllabification (Kiparsky 1979, Booij & Rubach 1984, Ito 1986, 1989, Booij 1988, Hayes & Abad 1989, Rubach & Booij 1990a, Inkelas 1989, Booij & Lieber 1993, Rubach 1993: 221ft). Note that syllabification rules also nicely illustrate that rules are not necessarily confined to one level, since they reapply at the cyclic level, the word level, and the postlexical level. For instance, after insertion or deletion of a vowel by a rule of phonology at the post-lexical level, syllabic structure has to be redetermined. In other words, syllabification rules are indeed 'anywhere rules', but this needs not be stipulated. The classic example of cyclic rule application is the Main Stress Rule for English (Chomsky & Halle 1968: 39. It concerns pairs of deverbal nouns like compensation(derived from compensate)versus condensation(derived from condense).The second syllable of condensationstill bears some degree of stress since on the previous cycle (condense)it had main stress, and therefore is does not allow for vowel reduction. However, Halle & Vergnaud (1987: 104) argue
'Booij & Rubach (1987) and Rubach (1993) defend the ordered blocks view. Their argumentationruns as follows. The followingreasoning is only possible in an ordered blocks view: if a rule A is extrinsicallyordered before a cyclic rule B, then the cyclicity of rule A is predictable. Inversely, if a rule D is extrinsicalyordered after a word level rule C, it cannot be a cyclic rule. Thus, this argumentationdepends on. whether one must allow for extrinsicorderingof rules within a certain level.
7
6
that this cyclicity effect is a language-specific property of English, to be expressed by a rule of Stress Copy that copies line 1 asterisks assigned on preceding cycles. Normally, the stress of previous cycles is erased, and the stess pattern of the new cycle is determined completely anew (the Stress Erasure Convention of Halle & Vergnaud 1987: 80-85). Yet, this does not mean that cyclicity of stress assignment is no longer motivated empirically since the application of certain word formation rules (for instance, suffixation of the deverbal -a[) is dependent on the stress pattern of the input words. There are quite a number of cases in the literature in which a morphological operation requires information concerning the prosodic structure or the stress patterns of its input words. A clear example of the role of cyclic stress assignment in the operation of phonological rules can be found in Dutch. In complex words created by the addition of a vowel-initial non-native suffix, the last vowel of the input word is lengthened unless it bears the main stress of that word on the first cycle (Booij, in press): (7)
alcoh[::,]I
id.'
profess[::,]r 'id' but: , tromp[e]t 'trumpet' fag[6]t 'bassoon' t[S]n 'id.' bl[:,]k 'block'
alcoh[6]1isch 'alcoholic, A' alcoh[o]Hst 'alcoholic, N' profess[ o]raal 'professorial' tromp[e]ttfst 'trumpet player' fag[::,]ttfst'bassoon player' t[::,]nnage 'id.' bl[::,]kkeer 'to block'
A nice near-minimal pair is formed by: tir[a]n 'tyrant'- tir[a]nniseer 'to tyrannize' versus can[::,]n'id.'- can[o]niseer'to canonize'. It is the difference in stress pattern between the two base words that explains why vowel lengthening only occurs in the second verb in -iseer. Note that on the second cycle, the stress pattern is redetermined, and therefore vowel lengthening must apply on the second cycle on the basis of the stress pattern derived on the first cycle. For instance, in both alcoholist and trompettist main stress is on the final syllable. Yet, vowel lengthening only takes place in alcoholist, reflecting the difference in location of stress between alcohol and trompet. This also implies that on the second cycle the lengthening rule has to precede the reapplication of the Main Stress Rule. 6 Evidence for the cyclic application of segmental rules that do not interact with stress is also available. For instance, in Ilokano, as analyzed by Hayes & Abad (1989), the form /pag+bao+an/ 'place where rats live' from bao 'rat' is subject to two rules: Glottal Epenthesis inserts a glottal stop between two vowels, and Glide Formation turns non-low vowels into glides. The phonetic form to be derived is [pag.ba?wan]. This form cannot be obtained without cyclicity: if we applied Glottal Epenthesis first, we would get [pag.ba?o?an],
and if we applied Glide Formation first, we would derive [pagbawan]. Cyclic derivation. with Glide Formation preceding Glottal Epenthesis gives the right result: 7 1st cycle: syllabification Glottal Ep. 2nd cycle: syllabification Glide Formation
(8)
bao ba.o ba.?o pag-ba. ?o-an pag.ba.?o.an pag. ba?. wan
The facts mentioned in (7) and (8) form a problem for a theory that does not allow for the cyclic interaction of phonology and morphology such as that advocated by Odden (1990, 1993). Similar evidence for the cyclic interaction hypothesis is provided by Kiparsky (1993: 282). Note, moreover, that cyclic interaction does not count as an independent principle since it follows from principle (4), "Apply a rule when possible'. Therefore, it is a simpler theory than a theory that wants to forbid cyclic interaction, and separates phonology from morphology. As stated in (2b), the principle 'Apply when possible' also implies that the application of morphological rules may be preceded by th~ application of phonological rules. A classic case is that of reduplication which may copy part of an input word that clearly has already undergone phonology, as in Bloomfield's (1933) example of internal reduplication in Tagalog pa-mu-mutul, derived from pamutul that in its turn derives from pang-putul, a case of the well known phenomenon of 'overapplication' in reduplication, discussed in Wilbur (1973). 8 Another kind of evidence for cyclic rule application is that morphological operations may impose phonological restrictions on their input words which pertain to derived phonological properties, in particular the number of syllables of the input words and their stress properties. A survey of such phenomena is given in Booij (1992). A simple example is English comparative formation which is only possible for monosyllabic words, and for disyllabic words with a second light syllable such as happy. Furthermore, the affixation of English deverbal -al is subject to the condition that the last syllable of the verbal stern bears stress: try-trial, but organize - *organizal. A number of examples from German are provided in Wiese (to appear: ch. 4). For instance, lhe German prefix ge- that is used in the formation of past participles requires the following stern to begin with a syllable with main stress. The relevance of the prosodic structure of the input of morphological operations is particularly clear in the framework of Prosodic Morphology. The basic claim of this theory is that morphological operations such as reduplication and infixation, which are formalized in terms of templates and prosodic circumscription, are "defined in terms of the authentic units of prosody: rnora,
7
A similar example is provided by Italian, in which stress has to be assigned cyclically, and a rule of UnstressedVowel Deletionoperates on the second cycle before reapplicationof the Main Stress. This means that the final /a/ of algebradeletes before the suffix -ico. Non-cyclic application would lead to the wrong form algebraico 'algebraic' (Vogel 1991). 6
The ordering of Glide Formationbefore Glottal Epenthesisneeds not be stipulated, but follows from the assumption that Glide Formation is part of the Syllabification Algorithm, which applies on each cycle before the segmentalrules (cf. Rubach & Booij 1990a). 8
A similar examplefrom Javaneseis given in Anderson(1979).
9
8
syllable, foot, prosodic word" (McCarthy & Prince, to app.: 1), which clearly implies that these prosodic units must be present in the input words.' In the case of reduplication, one might avoid this kind of interaction by representing reduplication as an abstract prosodic template attached to a stem which is then filled in after the application of phonology to the stem (Odden 1993). In case if infixation, however, this is no feasible solution since the location of the infix depends on the prosodic structure of the stem, which must therefore be available when infixation applies.
'musician';
3. The word level In the original model as proposed by Kiparsky (1982) it was assumed that all lexical (applications of) rules are cyclic. However, it appeared that certain lexical rules should apply after all morphology has been done. The earliest level at which they can apply is the word level, i.e. after all morphology has been performed. For example, the rule of Final Devoicing in Dutch that devoices obstruents in codas should apply after morphology (Booij 1981). Otherwise, wrong phonetic forms would be derived, such as [heltm] instead of the correct [heldm] for he/din 'heroine' with the morphological structure [[held]in], derived from held /held/ 'hero'. An example of a word level rule from English is the rule of 1-velarization: l's are alveolar in onsets, and velarized in codas: (9)
onset: [l]ap, Yo[l]anda, vigi[l]ant coda: pa[I], ca[I], vigi[I]
As shown by the pair vigil - vigilant the rule must be postcyclic, otherwise 10 vigilant would inherit a velarized [I] from its base word vigil. Similarly, Icelandic has a rule of vowel lengthening in open syllables and in word-final syllables ending .in at most one consonant (Booij 1986) which should apply after the morphology given data such as (10) br[o:]t 'breaking' r[i:]ta 'write' v[ei:]kur 'weak'
(ll)a Closed Syllable Adjustment (French) (Booij 1984a): mid vowels are halfclose in open syllables, and half-open in closed syllables (premier [prnmje] 'first, masc.' versus premiere [prnmjer] 'first, fem.'); b High Vowel Laxing (Canadian French): high vowels are [+tense] in open syllables, and [-tense] in closed syllables (Booij & Rubach 1987: 6), e.g. [p:itlt] 'small, fem. sg' versus [p:iti] 'small, masc. sg. '; c R-vocalization: /RI becomes [-cons] in coda position (German, Hall 1992: 57), e.g. Tier 'animal' [ti /I] versus Tiere 'animals' [tiR:i]; d Vowel Shortening: a vowel in an unstressed syllable is short (German, Hall 1992: 32), e.g. Mus{k [muzi:k] 'music' versus Musiker [mu:zik:ir]
br[o]tna 'break' r[i]tna 'Holy Scripture' w[ei]kla 'weaken'
Kiparsky (1985: 88-92) also provides an example of a lexical rule that applies at the word level (English, deletion of morpheme-final /n/ after /mi, as in damning) to which I will return in section 5, and of a rule that applies both at the cyclic level, subject to Strict Cyclicity, and the word level (Icelandic uepenthesis). Other word level rules identified in the literature are listed in (11):
'Cf. also McCarthy & Prince (1993) for an analysis of the relation between morphologicaland prosodic structurein the frameworkof ProsodicMorphology. 101n Durand (1990: 181) it is pointed out that in the English compoundmail order the final /1/ of mail is resyllabified,and is realized as a light [I]. This can be interpreted as a matter of lexicalization:the compoundis syllabifiedas if it is a simplexword, and thus the /1/ is located in an onset.
e Schwa-epenthesis: a schwa is inserted before /r ,1/ followed by a heterosyllabic consonant, and after /r ,1/ followed by a tautosyllabic consonant: [grnk] 'Greek' versus [g:irk:it] 'the Greek' (Bulgarian) (Zee 1988: 117ff). f Vowel raising in Italian: mid vowels become high in unstressed position (Vogel 1991: 189). g Low Vowel Assimilation in Basque (Hualde 1991: 31): vowels become [-low] if the preceding syllable contains a high vowel. h A postvocalic it/ becomes [ts] before ii/ followed by a vowel or a word boundary, as in demokratie /demokrati/ [demokratsi] (Dutch) (Booij, in press); Other languages for which word level rules have been proposed in the literature are Spanish (Harris 1983), Serbo-Croatian (Zee 1988, Inkelas & Zee 1988), Slovak (Rubach 1993), and Finnish (Kiparsky 1993: 284). Rule (llh) will not apply at the cyclic level to a word like demokratie /demokrati/ because in this word the sequence /ti/ is an underived environment. This is as required since the adjective demokratisch 'democratic' derived from the noun demokratie by means of -isch /isl sufffixation and deletion of the stemfinal ii/ does not have [ts] in its phonetic forn1: [demokratis], since the potential trigger ii/ is followed by a consonant whereas the rule requires a vowel or a word boundary after the /ii. In other words, the rule must indeed apply after all morphology has been performed, at the word level, where it also applies to underived words like patio 'id' [patsijo].1' In the case of Basque (rule llg) Strict Cyclicity will block application of the rule to underived forms (derived forms will correctly undergo the rule at the cyclic level). For instance, in aita-a 'the father' with the suffix -a, the rule should not first apply to the stem. The first /a/ is then deleted by a rule of Low Vowel Deletion that deletes /a/'s before an adjacent vowel. Subsequently, the remaining /a/ is turned into /e/ by low Vowel Assimilation resulting in the correct phonetic form [aite]. Nevertheless, stems without suffixes are subject to the rule of Low Vowel Assimilation, and this can take place at the word level, as required, because at that level Strict Cyclicity no longer holds.
11
Shaw (1985) argued that in Dakota Strict Cyclicity is valid for all lexical levels (she distinguishesthree levels in the lexicon). Note, however, that the word level we are discussinghere is not to .be identifiedwith the last lexical level, but rather with a level after all morphologyhas been done, so this is no problem for the idea that there is a lexical level where Strict Cyclicityis no longer in force. The same remark applies to the three cyclic levels (root, stem, word) proposed in Clark (1990).
11
10
Dutch has a rule of Homorgauic Glide Insertion that inserts glides between adjacent vowels of which the first should be non-low. This rule applies both in derived and underived words. At the cyclic level it will insert a Ul in, for instance, knie-en 'knees' /kni-;m/ [knij;m], a derived environment. At this level, the rule can not yet apply to underived words. At the word level, it will also apply to underived words like patio [patsijo]. Thus, we get the right result that the rule holds for both derived and underived words. Note again that rules must be allowed to apply to more than one level. A parallel case is the rule of Mid Vowel Harmony in Southern Bantu (Harris 1989). The regularity involved is that mid vowels are half-close (i.e. /e, o/) when the next syllable contains a high vowel, and half open otherwise, and the rule must apply to derived and underived words alike. In the case of derived words, it can already apply at the cyclic level, but underived words will be subjected to the rule at the word level. This will give the right results. Nevertheless, in this case there is another constraint that predicts it to apply at the word level, the constraint that at the cyclic level rules must be structurepreserving (cf. section 4.2.). In a number of cases, the word level status of the rules does not follow from Strict Cyclicity (rules lla-f). That is, these rules have to marked as word level rules. A shared property of these rules is that they are conditioned by prosodic structure. However, we cannot assume that prosodic structure does not count as derived environment because cyclic rules may also be triggered by prosodic information, for instance, the rule of Ilokana discussed above, and the rules of 12 Lateral and Nasal Depalatalization in Spanish discussed by Harris (1983: 50). The word level is not to be seen as a particular part of the lexical component of the grammar. The rules involved just apply when possible. Since their domains are the lower category prosodic domains syllable, foot and prosodic word, which are available before syntax, the rules can apply right away, i.e. before syntax. The only constraint they are subject to and that has to be encoded on the relevant rules is that they cannot apply until after the end of the morphological operations. The main effect of applying rules at the word level instead of the cyclic level is that the phonological generalizations involved will be true at the surface level, and hence they form transparent generalizations. In the case of the rule of Final Devoicing in German and Dutch, moreover, cyclic application of this rule would make the underlying distinction between voiced and voiceless obstruents in morpheme-final position completely useless, since these obstruents always occur in coda position on the first cycle, and hence the voiced obstruents would never get the chance to surface. A concomitant correct prediction is that word level rules are not bled by resyllabification effects at the postlexical ( = post-syntactic) level. For instance, the Dutch enclitic -ie Iii induces resyllabification of the final consonant of its host, thus shifting it to an onset position, but in case that consonant is an obstruent, it is realized as a voiceless consonant, although it stands in onset position, as is correctly predicted by the theory. To give an example, the phonetic form of vond-ie 'found he' is [v:mti], and not *[v:mdi], although the syllabification pattern is (v:m).(ti). (Booij 1985a). Similar observations are presented by Wiese (to appear) for German: he observes that Final Devoicing
Clements · & Keyser (1983) also argued that syllable structure counts as. derived environment. 12
has to apply before postlexical cliticization, except in (lexicalized) expressions a~ hab ich, 'have I', hab es 'have it' which have two possible pronunciations: [hap/bi~], [hap/b:is]. The same applies to the resyllabification effects of Enchainement in French that do not affect the vowel quality as regulated by Closed Syllable Adjustment (Booij 1984a). Another illustration is formed by 1velarization in English. For instance in call Andy, with resyllabification, we get (ko).(lan).(dy). with a velarized I in the onset. This follows from applying 1velarization at the word level, before syntax. Similarly, Zee (1988) pointed out with respect to the Bulgarian rule of Schwa-epenthesis that enclitics do form part of the preceding prosodic word at the postlexical level, but that this does not influence the location of the epenthesis schwa. For instance, we get / grnk = c/ 'he is Greek' with the syllabification pattern (grn).(kc).. Yet the schwa occurs after the /r/ because at the lexical level we have the prosodic structure (grk)., causing the schwa to be inserted after the Ir/ since it is followed by a tautosyllabic consonant at that level. Similarly, Kiparsky (1993: 284) pointed out that Finnish Consonant Gradation applies to phonological words at the word level, but not to word +clitic combinations that form phonological words at the postlexical level. As far as German is concerned, Hall (1992: 109) also points out that the rule of R-vocalization is not bled by resyllabification of a word final C to the onset of the next vowel-initial word, as follows from the word level application of that rule. One may wonder how the application of rules that should only apply cyclically are blocked from applying at the word level. A possible solution is the idea that rules apply as soon as possible, but can be turned off at a later level. For instance, the Polish rule of Palatalization discussed below, at least if it is a rule, should be blocked from applying morpheme-internally at the word level. Otherwise, we would also palatalize the first Isl of servis +e 'service, Joe. sg. ', after all, and the desirable effect of Strict Cyclicity would be undone. In the case of Polish Palatalization, this problem does not arise at all if, as proposed by Kiparsky (1993), Strict Cyclicity is seen as the effect of structurebuilding rules applying to underspecified representations. For instance, the first is/ of servis will be fully specified as [ +anterior], whereas the second Isl will be unspecified for [anteriorJ. When this underspecified segment occurs before a [-ant] segment, it will be [-ant], and otherwise, by default, [ +ant]. However, in other cases such as Finnish Consonant Gradation it remains necessary to restrict application of the rule to phonological words available at the lexical level, hence this rule must be turned off at the postlexical level, unlike Prevocalic Schwadeletion in Dutch. Another general constraint on the lexical application of rules is that they should be obligatory. This reflects the classical distinction in European structuralist phonology between word phonology and sentence phonology. Take the case of the Dutch rule of inf-deletion that is optional and deletes /n/'s in rhymes consisting of a schwa plus in/ at the end of a morpheme:
(12) wapen 'weapon' wapen-en 'weapons' molen 'mill' molen-aar 'miller'
[wap:in] or [wap:iJ [wap:m:in] or [wap:in:i] [mol:in] or [mol:i] [molana:r]
13
12
Apart from its optionality it could have been a word level rule. However, it is clear that it should apply post-lexically since it must be bled by cliticization and the ensuing resyllabification: (13) zetten het 'put it' /zct;in ;it/ phonetic form (zc)(t;i)(nat) If we allowed for deletion of the /n/ at the word level, we would expect the
phonetic form [zct;it] to be possible through /n/-deletion followed by Prevocalic Schwa-deletion. However, this phonetic form is completely out as realization of zetten het, although it is a normal realization of zette het, with the verb in the past sg. form.
4. Properties of lexical rules In the course of the development of the theory of Lexical Phonology, a number of specific distinguishing properties have been proposed for lexical rules, summarized in Kaisse & Shaw (1985): (14) Lexical rules a. apply within words only b. are cyclic c. are subject to Strict Cyclicity d. are structure-preserving e. may have exceptions A related issue is whether the inverse of these properties hold for postlexical rules. For instance, is it indeed true that postlexical rules do not have lexical exceptions? The answer to this question is far from clear (cf. Hayes 1990 for discussion). We may also add the property of obligatoriness to this list, but clearly the inverse does not hold for postlexical rules, which might be obligatory as well. We should also note that these defining properties should not be taken as defining properties of the rules themselves, but rather of the way in which they apply, since the same rule may apply both lexically and postlexically. Moreover, the fact that lexical rules apply within words only is simply a matter of the domain specification that each rule requires. Furthermore, as we saw above, cyclicity is not an inherent property of rules, but follows from principle (4): "Apply a rule when possible'. Therefore, the question is rather whether the following statements are true (15) a. The application of phonological rules at the cyclic level is subject to Strict Cyclicity b. Phonological rules that apply lexically are structure-preserving These two hypotheses are discussed in the next subsections. 4.1. Strict Cyclicity The formulation of (15) makes it clear that cyclicity and strict cyclicity must be distinguished: · one can also envisage a theory in which rules apply cyclically
'
without Strict Cyclicity being in force. Classic cases in favour of Strict Cyclicity are the English Trisyllabic Laxing Rule that accounts for alternations like sanesanity and the Polish rule of Palatalization of consonants before front vowels and glides discussed above. The analysis of Trisyllabic Laxing and Palatalization in terms of Strict Cyclicity is not without problems, as Szpyra (1989: 69-83) noted. Quite a number of exceptions to these rules remain unexplained. For instance, the following complex English words are exceptions to Trisyllabic Laxing: (16) hyphenate, momentary, dangerous, vaporous, alienate, motorist In the case of Polish Palatalization it appears that this rule does not apply before the following suffixes (Szpyra 1989: 79): 13 (17) -e 'fem. pl', -ego 'gen. sg', -emu 'dat. sg.', -ej 'dat. sg', -em 'instr. sg', -e 'nom. pl' However, these criticisms do not imply that Strict Cyclicity is incorrect, but only that this concept may not suffice to account for the phenomena under discussion here. For instance, the rules may have lexical exceptions, or may be conditioned by specific morphemes. 14 Moreover, there are other cases in the literature where Strict Cyclicity appears to make the right predictions (cf. Clark 1990). Evidence for the necessity of Strict Cyclicity can also be found in the realm of syllabification. As argued by Guerssel (1986), Kenstowicz & Rubach (1987) and others the distinction between (high) vowels and glides is not a difference in feature specification, but a difference in prosodic positions: vocoids are vowels when in nuclear position, and glides otherwise. It is usually predictable whether a vocoid will surface as a vowel or as a glide. However, in some languages the vowel-glide distinction may be distinctive in certain positions. In those cases vocoids that unpredictably surface as nuclei will be specified as such in the lexicon, for instance by assigning them the node Nucleus lexically. Note now that this lexical specification has to block the Gliding rule, whicli is part of the syllabification algorithm, and assigns vocoids in that position to a non-nuclear position. Otherwise, the lexical specification would not even make sense. That is, Strict Cyclicity will block such structure-changing applications of the syllabification algorithm, as required. This makes Strict Cyclicity, just as the principle 'Apply a rule when possible' a principle of 'virtual necessity'. Note that the Strict Cyclicity effect is obtained automatically if syllabification rules are seen as rules that only build structure. Thus, they will respect existing lexically given prosodic structure. As argued in Rubach & Booij (1990a), the only structure changing operation involved in syllabification is the Coda Erasure rule which removes the coda node of the last syllable of a base word when a
13
Rubach ( 1984) adduces independent evidence that these suffixes begin with an abstract mid back unroundedvowel that surfaces as its front counterpart, which accounts for the fact they do not have a palatalizingeffect. "Similarly, Amason (1992) argued that the applicationof U-umlaut in Icelandic is governed by specific suffixes, and cannot be dealt with in terms of Strict Cyclicity and level ordering.
14
suffix is attached, ·after which the structure building rules reapply. The same remark applies to languages that require lexical (i.e. unpredictable) marking of stress in some words. Such lexical specifications make only sense if they will not be undone by the stress rule(s) of the language. An example of such a language is Greek (Malikouti-Drachman & Drachman 1992: 203). The Strict Cyclicity effect follows from the fact that stress rules are metrical structure building rules that respect metrical structure already provided in lexical representations. 4.2. Structure Preservation So far, we have seen that the way in which phonological rules apply almost
completely follows from principle (4) and the domain specification of each rule, which is part of its structural description. The only additional hypotheses are: (i) optional rules are always postlexical, (ii) Strict Cyclicity - which can be seen as a derived notion - and (iii) Structure Preservation. Structure Preservation is the hypothesis that lexical rules do not introduce types of segment that do not occur in the underlying segment inventory of the language. For instance, the distribution of the aspirated stops in English is completely predictable, and hence there are no underlying aspirated stops in English. Consequently, it is predicted that the rule of aspiration applies postlexically. Kiparsky (1985: 92) explicitly hypothesizes that all lexical rules are structure-preserving, including the word level rules. However, certain evidence suggests that this claim cannot be completely upheld: certain lexical rules are not structure-preserving. To begin with, Dutch has a rule of Pre-r-lengthening which lengthens vowels before /r/ in the same prosodic word. This rule introduces non-distinctive properties, it creates segments that do not occur at the underlying level: (18) biet /bit/ [bit] 'beet' bier /bir/ [bi:r] 'beer' eet /et/ [et] 'eat' eer /er/ [e:r] 'honour' hoop /hop/ [hop] 'hope' hoor /hor/ [ho:r] 'hear' Yet, the rule appears to be a lexical rule since it has exceptions: the past tense forms of a number of ablauting verbs do not participate in the lengthening, e.g. wierp 'throwed', hielp 'helped', bedierf 'spoiled' (Booij, in press, Gussenhoven, to app.). Suppose now that we assume that it is only the cyclic level that is subject to the constraint of Structure Preservation. Kiparsky (1985: 135, note 3) suggests the possibility that the marking conditions that express generalizations concerning underlying segments may be turned off at a later level in the lexicon). This solves the problem, at least for Dutch, since the rule is clearly a rule that should not and cannot apply cyclically: the environment in which it applies is never a derived environment. The word level status of the rule 1s confirmed by the following observation: the agentive suffix -eur surfaces as -eus before the feminine suffix -e. Cyclic application of the lengthening rule would thus derive the phonetic form [m::mto:sa]for monteuse 'fem. technician' derived
15
from monteur, whereas [m:mtosa], without lengthened vowel is the correct phonetic form. Other examples of word level rules that are not structure-preserving are the Dutch rules of Homorganic Glide Insertion (that introduces, besides /j/ and /w/ that also occur as underlying segments, a front round glide after /y/ that does not occur underlyingly), the Dutch rule that turns /t/ into [ts] (there are no underlying affricates in Dutch), and the English rule of 1-velarization (there are no underlying velarized Ill's in English). One might of course claim that such rules are postlexical rather than postcyclic. However, it requires an extra stipulation to make them postlexical: if nothing is said they will apply at the word level. A similar observation was made in Booij & Rubach (1987) with respect to the rule of Canadian French mentioned above that predicts the tenseness of high vowels ([ +tense] in open syllables, [-tense] in closed syllables): it is a word level rule that introduces a distinction that does not occur underlyingly for these vowels. The same problem occurs for German, as pointed out by Hall (1989). The dorsal fricative consonant in German is a velar [x] after a back vowel ("Achlaut"), and a palatal fricative [9] ("Ich-laut") elsewhere. In other words, the distribution of [x] versus [9] is predictable by rule. The rule that specifies the velarity of the fricative after back vowels is therefore not structure-preserving since there is no underlying opposition [x] versus [9] (the other allophone is introduced by a default rule). Yet, the rule must be lexical since it only applies within morphemes; the palatal [9] does occur after a back vowel as the first consonant of the diminutive suffix -chen, hence minimal pairs like (19) Kuchen 'cakes' [kuxan] versus Kuh+chen 'little cow' [kupn] This issue is also discussed by Macfarland & Pierrehumbert (1991) and Iverson & Salmons (1992). The latter argue that the rule is a word level rule, and that therefore Structure Preservation should be seen as a constraint that only applies at the cyclic level. 15 German R-vocalization also introduces a segment that does not belong to the underlying segment inventory (Hall 1993). Yet, it must apply at the word level, because it is not affected by resyllabification. This supports the idea that Structure Preservation is not a condition on the application of rules at the word level. A potential problem for the hypothesis that it is only the word level or a later level16 at which Structure Preservation is turned off, is the rule of Glottal Stop insertion in Ilokano. This rule is clearly cyclic, as illustrated above. If there are no underlying glottal stops, as Hayes & Abad suggest, there is a cyclic rule that
15
According to Wiese (to app.) the suffix -chen forms a prosodic word of its own, which explains why the rule does not apply there, its domain being the prosodicword. We then have to allow for prosodic words without a full vowel for German. On the other hand, as Wiese pointed out, the suffix -chen does behave as an independent phonologicalword with respect to reduction as in Mutter- und Viiterchen 'dear mother and father'. 16
Myers (1991) adduces cases in which the Structure Preservationconstraint also holds for a part of the post-lexicalphonology.
17
76
introduces new segments. However, as Hayes & Abad (1989: 341) point out, in order to predict all instances of glottal stop insertion, they have to allow for contrastive syllabification in the lexicon. Alternatively, they suggest, one may allow for glottal stop phonemes but only in the environment C--V. In the latter analysis glottal stops do belong to the underlying inventory, and hence there is no problem for the Structure Preservation hypothesis.17
5. Level ordering A typical feature of many analyses in the framework of Lexical Phonology is that they make use of levels and level ordering, to such an extent that Lexical Phonology is sometimes equated with a theory of level ordering. 18 The idea of level ordering is that the lexicon consist of two or more ordered levels or strata. Morphological rules are assigned to specific strata, and phonological rules can also be assigned to one or more of these strata. The classical hypothesis of level ordering for English is there is a level 1 at which stress-shifting affixes are attached to stems, and at which the Main Stress Rule of English applies, and a level 2 for the attachment of stress-neutral affixes, where the Main Stress Rule no longer applies. The morphological prediction then is that stress-neutral affixes are peripheral to stress-shifting affixes. For English different numbers of lexical levels have been proposed: two (Siegel 1974), three (Kiparsky 1985), and four (Halle & Mohanan 1985). Other languages for which levels have been claimed to be relevant are: (20) Dutch (3 levels, Paulissen & Zonneveld 1988, 2 levels, Kager 1989) Malayalam (4 levels, Mohanan & Mohanan 1984) Dakota (3 levels, Shaw 1985) Mandarin (4 levels, Packard 1990) Icelandic (2 levels, Kiparsky 1985) German (3 levels, Wiese 1988) Igbo (3 levels, Clark 1990) Italian (3 levels, Scalise 1984) Basque (2 levels, Hualde 1991) Turkish (4 levels, Inkelas 1993b) The basic idea behind level ordering is that the phonological behaviour of affixes correlates with their morphological behaviour with respect to affix ordering. As will be clear from the exposition on Lexical Phonology given above, the
The problem of Structure Preservation pertains to all analyses in which there are more kinds of segment at the end of the lexical derivation than at the underlying level. For instance, Mohanan & Mohanan (1984) argue that Malayalam has three underlying nasals, but seven lexical nasals. It remains to be seen whether the relevant rules can be interpreted as word level rules. In Mohanan (1988) it is suggested that structurepreservingness is a tendency rather than a principle.
hypothesis of level ordering does not belong to the theoretical core of Lexical Phonology. In other words, rejection of level ordering does not imply rejection of Lexical Phonology. From the beginning it was clear that there is not always a perfect correlation between the phonological and the morphological behaviour of word formation processes. For instance, Mohanan & Mohanan (1984) had to allow for loops to earlier strata in order to derive the correct phonetic forms of certain types of complex word, which undermines the basic idea. A number of types of objections to level ordering can be found in the literature: it makes incorrect predictions (i.e. there is counterevidence), it is a superfluous mechanism because the facts already follow from other principles, and it creates conceptual problems. The classic example of empirical counterevidence for level ordering in English is the word ungrammaticality,in which first the stress-neutral prefix unhas been added to the adjective grammatical, and subsequently the stressshifting suffix -ity. Therefore, ungrammaticalitywas called a bracketing paradox since in this case morphology requires a structure different from phonology. Similar examples are extrametricality and governmental, and words in -ation derived from words ending in the stress-neutral suffix :ize such as legalizeation, divinize-ation,palatalize-ation(Szpyra 1989). A similar example from Dutch is the word molenares 'fem. miller', in which the stress-shifting suffix -es follows the stress-neutral suffix -aar (an allomorph of -er) (cf. Schultink 1980 for more examples). The second type of objection is that the required affix ordering already follow from independently established principles (Booij 1982, 1989; Fabb 1988). For instance, in languages such as English, Dutch and German, [-native] suffixes can only be added to [-native] (underived or derived) stems. We do not find *yellowity besides yellowness, whereas both absurdity and absurdness are possible words of English. This makes the correct prediction that [ +native] suffixes are always peripheral to [-native] suffixes (and [ +native] prefixes peripheral to [-native] prefixes). Since [-native] suffixes are often stressshifting, and [ +native] suffixes stress-neutral, the observation that stress-neutral suffixes are peripheral to stress-shifting suffixes follows from the constraint w.r.t the feature [native]. Interestingly, there are crucial cases that show what the correct generalization is: Dutch has three [ +native], but stress-shifting suffixes. The generalization in terms of the feature [native] rightly predicts that such stress-shifting suffixes can be external to stress-neutral suffixes: molen-aar: es 'fem. miller', ler-aar-es 'fem. teacher', etc .. Another example of the use of level ordering for morphological purposes is the following. It has been observed both for English and Dutch that the irregular formation of the past tense by means of Ablaut does not occur in verbs with ablauting roots that are derived from compounds: they take the regular past tense suffix.
17
18
Sometimes, the term 'Lexical Morphology' is nsed to refer to the level ordering hypothesis.
(21) English: stand-stood/ grandstand-grandstanded Dutch: fluit-floot 'to whistle, to play the flute' /blokfluit-blokfluitte 'to play the recorder' Kiparsky (1982) and Paulissen & Zonneveld (1988) accounted for this difference by locating the Ablaut rule at level 1, and conversion of nominal compounds to verbs at a later level, at which also regular past tense formation takes place.
-- ----------------
18
19
However, level ordering is superfluous here if we assume a feature [+ Ablaut] that triggers the irregular past tense formation is not inherited by the complex verbs since they are conversion of nominal compounds, and features can only percolate across a proper path of nodes of the same syntactic category. In other words, the verbs grandstand and blokfluit do not get the feature [ + Ablaut], and hence they are subject to regular past tense formation (cf. Booij 1989).19 Level ordering is also used for phonological purposes. For instance, it has been proposed that in Italian derivation is level 1, and compounding is level 2 because certain phonological rules apply to derived words, but not to compounds. However, once we realize that compounds consist of more than one prosodic word, this difference in application follows straightforwardly from the domain assignment of the relevant rules: they apply within the domain of the prosodic word (Vogel 1991). The same point can be made with respect to Basque: levels are made superfluous once the domain of application of the rules is specified properly. For instance, the Basque has a rule of Low Vowel Assimilation (Hualde 1991: 23): (22) V
•
[-low] / [+high] C0
--
In the Baztan dialect this rule applies morpheme-internally, in some derived words (Derivation I), in inflected words, and in host+clitic combinations, but not in compounds and certain types of derived words (Derivation II). Therefore, Hualde (1991: 38-39) concludes to the following organization of the grammar of Baztan Basque: (23) Stratum I: inflection and derivation I Stratum II: derivation II and composition Postlexical: cliticization (back to Stratum I) The rule of Low Vowel Assimilation is then assigned to Stratum I. Hualde argues that this solution is better than making use of a distinction between two types of boundary, + versus #, as was done in standard linear generative phonology (Chomsky & Halle 1968), because the level approach eliminates the possibility of unrestricted or arbitrary use of boundary symbols. Note, however, that the ordering proposed implies that inflection is ordered before compounding and derivation II, and that a loop is necessary from a postlexical stage back to Stratum I. The correct application of Low Vowel Assimilation can be achieved without level ordering, by assigning this rule the prosodic word (w) as its domain of application. Inflectional suffixes, and derivational suffixes of class I can be interpreted as cohering suffixes, i.e. they form one prosodic word with the preceding material. Compounds, on the other hand, consist of more than one prosodic word. The suffixes of class II will be qualified as non-cohering suffixes which form a prosodic word of their own (cf. Booij 1984b, Booij & Rubach 1984, Booij 1985b for cross-linguistic evidence for this approach). That the attachment of clitics induces application of Low Vowel Assimilation follows from the assumption that cliticization means that a function word is prosodically
incorporated into the preceding prosodic word, its host. Thus, at the post-lexical level, cliticization creates new domains of application for rules that apply with the prosodic word. Independent evidence for this assumption in the case of Basque is that clitics undergo other rules that are usually confined to words. For instance, in Baztan Basque the initial /b/ of the determiner bat deletes intervocalically, i.e. after a host word ending in a vowel, whereas this rule does not apply across word boundaries (Hualde 1991: 29). Note, again, that this implies that a rule can apply both lexically and postlexically, as was already argued above. In sum, an analysis in which rules are properly specified for their domain of application in terms of prosodic categories, can make level ordering for phonological purposes 20 , and the concomitant complication of a loop, superfluous. Moreover, it avoids the awkward morphological consequence that inflection is ordered before certain types of word formation (derivation II and compounding). , The phonological use of level ordering has also been invoked when languages have lexicons with different historical layers with their own phonological generalizations, as is the case for English, Dutch, and Malayalam. In Germanic languages such as English and Dutch we find a Greek-LatinRomance stratum, and in Malayalam there is a Sanskrit stratum and a Dravidian stratum. Phonological rules that apply to only one layer of the lexicon can be seen as rules that are conditioned by a diacritic feature that refers to the relevant stratum. For instance, in Dutch and English the Latinate prefixes in- and conexhibit a more radical form of place assimilation than the negative prefix onlun-: before liquids there is total assimilation in the case of the Latinate prefixes: il-, ir-, car-, col-. The relevant generalization can be made by means of reference to the feature [-Native], a features that is needed anyway for the morphology of English and Dutch. Clearly, this type of assimilation is a process borrowed from Latin. There is also a conceptual problem with level ordering in word-based morphology: word formation rules may take existing words listed in the lexicon as their inputs since clearly idiosyncratic properties of words recur in complex words derived from them. This implies a loop in the organization of the lexicon, and thus the morphological component may add an affix of level n to an already existing word listed in the lexicon with the structure 'base-affix of level n+l', thus making level ordering empty (Booij 1982, Aronoff 1988). This conceptual problem can be solved by interpreting a difference in levels as a difference in constituency. For instance, in line with Inkelas (1989, 1993a) we might interpret the English levels 1 and 2 as corresponding to stem constituents (s) and phonological word constituents (w) respectively, with the wnode dominating the s-node, and with rules applying in the s-domain or in the 21 w-domain. This implies that we also classify each affix as an s-affix or a waffix. An s-affix can only be affixed to a stem, thus creating a larger stem, a waffix is attached to a w-constituent, thus forming a larger w-constituent. A stem
wThe same point is made with respect to Malayalamby Sproat (1986), and with respect to Korean by Kang (1992). 21
19Similar reanalyses of the morphologicalfacts of Mandarin without level ordering are presentedin Sproat & Shih (1993), a reply to Packard(1990).
An interpretationof levels in terms of hierarchicallyordered constituentshas also been proposed by Selkirk (1982). The differencewith Inkelas' proposal is that Selkirk's constituents are morphological constituents, whereas Inkelas' consiituents are phonologicalconstituents,which exist in additionto the morphologicalconstituents.
21
20
can form a w, and a w can consist of a stem only. So the following configurations are well-formed and ill-formed respectively: (24)
w
w
I s
*s
w
Since the s-node cannot dominate the w-node that is higher in the hierarchy, it will be impossible to add an s-affix to an existing word that is dominated by a w-node. Thus, we maintain the generalization that w-affixes (native affixes) are peripheral with respect to s-affixes (non-native affixes). The structure of the words ungrammaticaland ungrammaticalitywill be as follows: (25) w
w
s un
grammatical
w
w
I I
I I
s
s
un
suffixes. The formal reinterpretation of level distinctions as a distinction between types of constituent is also proposed by Goldsmith (1990: 241), who proposed three hierarchically ordered constituents for Bantu languages: root, stem, word, which also parallels Clark's (1990) distinction of three levels for Igbo. Note that the notion 'stem' is used here in a phonological sense. For instance, in English and Dutch inflectional suffixes, which are attached to stems in the morphological sense ( = words minus inflectional endings) as far as morphological structure is concerned, are attached to phonological words at the level of prosodic structure, since they are w-suffixes. This is the formal expression of Bloomfield's (1933) observation that these languages have a word-based rather than a stem-based morphology. 22 Halle & Vergnaud (1987: 78) concluded on the basis of 'bracketing paradoxes' such as ungrammaticality that although there may be levels for phonology, morphological rules are not assigned to particular strata. In the case of ungrammaticality,un- is qualified as a non-cyclic prefix, and -ity as a cyclic suffix. Each affix must therefore be specified for the feature [cyclic], but this is often predictable from its specification for the feature [native]:
I~
grammatical 1ty
That is the suffix -ity has been added to ungrammaticalat the s-level, which is possibl; since there is no intervening w-suffix. Thus, there is no or~ering paradox, as in the level ordering approach. The prefix un- counts as a w itself, although it is a bounded morpheme, as argued in Booij & Rubach ( 1984). The Latinate prefix in- can now be considered as an s-prefix, with the stem functioning as the domain of the rule of assimilation discussed above. Hence, this assimilation process will not affect the prefix un-. The notion w-constituent can also be used to express the observation that inflection is usually peripheral with respect to derivation. This follows from the following constraint:
(28) [-native] • [+cyclic] Cyclic affixes induce erasure of the existing stress pattern, and reapplication of the Main Stress Rule. Thus, ungrammaticality is derived as follows: (29) 1st cycle MSR 2nd cycle 3rd cycle Stress Erasure MSR
grammatical
a
un[grammatical] [un[grammatical]Jity a
a
(26) Inflection must occur at the edge of w
In Dutch, there is a number of suffixes that, although they are [ +native], nevertheless shift the stress of their input words:
This constraint implies that level ordering as a means of ordering inflection with respect to derivation superfluous. Moreover, it correctly predicts that inflection can occur word-internally if it occurs at the edge of a w. For instance, in Dutch we find plural forms of nouns as the first, non-head constituents of compounds (where each constituent form a w of its own), and before suffixes that form prosodic words of their own such as -achtig '-like, fond of' (which stands in contrast to the 'cohering' suffix -ig I-J"flwith the same meaning as -achtig:
(30) -in /mi ,es /cs/ -ij lei/ -lijk /1-Jk/ -isch /is/ -ig fa'Y/
(27) dames-toilet 'ladies' room' dames-achtig 'lit. ladies-like, lady-like' *dames-ig 'lit. ladies-like' steden-raad 'cities' council' steden-achtig 'fond of cities' *steden-ig 'fond of cities' In sum, the prosodic constituency approach appears to be superior tq !eve! ordering as a means of expression the 'closing morpheme' nature of infleclional
held 'hero' leraar 'teacher' voogd 'guardian' afschuw 'horror' afgod 'idol' afkeer 'aversion'
heldfn 'heroine' lerares 'fem. teacher' voogdfj 'guardianship' afschi\welijk 'horrible' afg6disch 'idolatrous' afkerig ,averse'
This underscores the point that there is no one-to-one correlation between the morphological and phonological properties of the different sets of affixes. However, Halle & Vergnaud draw a more radical conclusion: (31)
22
"In our theory then, as in SPE, morphology is distinct and separate from phonology. Morphology interacts with phonology in that it creates the objects on which the rules of phonology operate." (Halle & Vergnaud
Wiese (to app.) discusses the possibility and advantages of interpreting levels as constituents for German.
23
22
1987: 78)
This claim, all morphology before all phonology, does not follow from their objections to level ordering as a morphological mechanism. Of course, in the model in which all morphology precedes all phonology, it is still possible to have cyclic application of phonological rules (Odden 1990, 1993), by keeping the internal morphological structure of the complex word intact, but this now has to be stipulated again. Note, moreover, that one could as well describe a language in which first all rules must apply to the largest domain, and then to the next, smaller domain, until the smallest domain has been reached. That is, it is a much Jess restrictive theory. As we saw above, the application of morphological rules may be dependent on derived phonological information concerning the sterns to which they apply. In the SPE-rnodel of the interaction of morphology and phonology to which Halle & Vergnaud return, phonological conditions on morphological attachment will have to be implemented as a checking procedure that cyclically checks for each affix on a given cycle, whether the string on the previous cycle meets the phonological requirements of that affix. Again, this can be done, but nothing in the theory excludes the possibility that an affix exist that imposes requirements on the phonological string of the next cycle rather than the previous cycle, a situation that one would like to exclude in a principled way. Although Halle & Vergnaud are right in disconnecting certain morphological and phonological generalizations, this does not lead to the conclusion that the insights of Lexical Phonology with respect to the interaction of phonology and morphology have to be given up. Rather, we should conclude that levels might not be an adequate mechanism for expressing the fact that certain phonological rules have a restricted domain of application within words. In sum, when the use of level ordering cannot be made superfluous by assigning prosodic domains to rules, as was the case for Basque and Italian, or by independently motivated morphological generalizations, as was suggested, for instance, by Sproat & Shih (1993) for Mandarin Chinese, the generalizations expressed by level ordering should be expressed in terms of types of wordinternal hierarchically ordered constituency. Languages with a rich morphology for which a ternplatic or a position class approach has been advocated (for example, Slave in Rice 1993) may thus be seen as languages with a hierarchy of word-internal domains. Note also that Clark (1990) who distinguished three ordered (and cyclic) levels for Igbo calls them root, stern, and word respectively, with a hierarchical ordering of the corresponding constituents implied. 5 .1. The word level as lexical stratum
Let us now have a closer look at another case in which level ordering has been invoked for English. The process involved is the deletion of /n/ after !ml in words like damning and damned, whereas the In/ remains in words like damnation and damnable. Kiparsky's formulation of the rule is as follows: (32) n-,. 0 / rn-]
The rule cannot apply to a level 1 word like damnation because of the SSC (level 1 is the level of underived words and complex words with non-native
affixes). At level 2 we create words like damning. If at this level the SCC does not hold, as proposed by Kiparsky, we can delete the morpheme-final /n/, as required. Deletion does not apply to damnation since the word-internal morpheme bracket after damn has been erased in damnation before entering level 2 (Bracket Erasure at the end of level 1). An alternative analysis is proposed in Borowsky (1993) who also distinguishes two levels. In her proposal, at the non-cyclic level 2 all phonology precedes morphology. Hence, the coda-final In/ of damn is deleted at that level before the suffix -ing is attached. On the other hand, in damnation, the /n/ha; shifted to onset position at level 1, and hence it is immune to deletion. Another example of a rule of English that should apply before level 2 morphology is the rule of schwa-insertion that inserts a schwa in morphemes ending is a consonant-r cluster, for instance filter with the underlying form /filtr/. At level 1, schwa-insertion does not take place because the attachment of one of the vowel-initial suffixes of tliat level has the effect of shifting the Cr cluster to onset position, as in filtration. On the other hand, before the level 2 suffix -ing the schwa must nevertheless be inserted, as infiltering. A nice aspect of Borowsky's proposal is that it enables us to make a significant generalization, because the rules involved all have the effect of saving consonants which would be extrasyllabic at the end of level 1 in case no affixation takes place at level 1. So the rules can be formulated as follows: (33) After level 1 m'-,,,.0 0 -,. a before r'
where the ' indicates extrasyllabicity. What is at stake here is that non-native morphemes must be pronounceable as words before entering the native morphology of English. A serious objection to Borowsky's proposal is that she does not allow for lexical phonological rules to apply after Je,el 2 morphology. Note that the obligatory resyllabification that is typically induced by the attachment of vowelinitial suffixes, also occurs before a level 2 suffix such as -ing. For instance, in walking the /kl forms the onset of the second syllable. Therefore, we would like to let syllabification reapply lexically after the level 2 morphology. In Dutch, we find similar alternations, for instance with respect to the occurrence of the schwa before /r/: · (34)
center 'id.' /sentr/ [sentar], centr-eer 'to center' [sentre:r] versus centeren 'to center' [sentarnn] filter 'id.' [flltar], filtr-eer 'to filter' [ftltre:r], filtr-aat 'filtrate' [flltrat] versus filter-en 'to filter' [ftltaran], filter-ing 'id.' [ftltanu] integer 'honest' [mteyar], integr-iteit 'integrity' [mteyriteit], integr-eer 'to . integrate' [mtayre:r] versus integer-e 'honest, inflected form' [mteyara]
Interestingly, in Dutch there are also lexical phonological rules that have to apply after level 2 morphology. For instance, the diminutive suffix is a level 2 suffix, and is attached to non-native morphemes in their pronounceable, i.e. w form, but nevertheless exhibits a regular allornorphy conditioned by the phonological shape of its base words. Moreover, some of the level 2 suffixes
25
24
are not stress-neutral, and induce reapplication of the Main Stress Rule of Dutch (Booij, in press), for instance the level 2 suffix adjectival suffix -ig. Therefore, an alternative to Borowsky's analysis is called for. Such an alternative is available once we make use of the difference between sconstituents and w-constituents proposed and discussed above. A lexical morpheme like filter with the underlying form /flltr/ should be specified in the lexicon as an s-constituent. We now assume the rule that all s-constituents can be turned into w-constituents without affixation. This is necessary when such a morpheme is to be used as a word, or as the input for w-affixation. The rules discussed above will be assigned the domain w. Hence, as soon as the wconstituent is created, a schwa wil be inserted before the extrasyllabic Ir! of /flltr/, in accordance with the principle 'Apply a rule when possible', and subsequently, the word can be used as a free form or as a base for wsuffixation. An independent argument for this approach is provided by the following observation. Quite a number of non-native words of Dutch have two allomorphs, one to be used in s-suffixation (non-native suffixation), and another one in w-suffixation (native suffixation), as shown by the following examples, where the first line represents s-suffixation, and the second line w-suffixation: (35) trauma 'id.'
traumat-isch 'traumatic', traumat-oloog 'traumatologist' trauma's (pl.) ratio 'id.' ration-eel 'rational', ration-ale 'reason' ratio's (pl.) Plato 'id.' platon-isch 'platonic', platon-ist 'id.' Plato's (genitive) cursus 'course' curs-ist 'student', curs-orisch 'by means of a course' cursuss-en (pl.) apostel 'apostle' apostol-isch 'apostolic', apostol-aat 'apostolate' apostel-en (p.) orkest 'orchestra' orkestr-atie 'orchestration' orkest-en (pl.)
be dominated by a w-constituent' in virtue of the Elsewhere Condition. If a lexical entry is not represented as s-constituent, it can only function as w. This is the case with native morphemes and words. Hence, the difference between non-native and native morphemes can be encoded by assigning sconstituency to non-native morphemes only, whereas the native morphemes remain unspecified as to constituency at the underlying level, and are given the w-status by rule. The following generalizations fall out from this approach: (36) a. Non-native affixes do not attach to native morphemes b. Native affixes are peripheral with respect to non-native affixes In short, Borowsky's word level rules should be reinterpreted as rules that apply in the domain of the phonological word w, as soon as that domain of application has been created. Thus, we can keep the notion 'word level rule' to refer to those rules of lexical phonology that apply after all morphology has been . performed. Moreover, we can now avoid a very language-specific organization of the lexical phonology of English. 24
6. Conclusions In this paper I have tried to identify the theoretical core of Lexical Phonology as the following minimal assumption: 'Apply a rule when possible'. The application of rules is further determined by its domain of application, and the principles of Strict Cyclicity and Structure Preservation. With these minimal assumptions it is possible to correctly predict the way in which phonology and morphology appear to interact. It is not necessary to assign a rule to a specific level of the grammar (although it may be the case that certain rules have to be turned off at a later level of derivation). This implies that, in principle, rules can apply at more than one level, for which we saw empirical evidence.
In these cases it is impossible to derive one allomorph from the other by means of a regular phonological rule. Hence, we have to list both allomorphs in the lexicon23 , one as the s-form, and one as the w-form. In a few cases, the wallomorph is derivable from the s-form, for instance in the case of the morphemes ending in /Cr/, as shown above. In other cases this does not make sense. For instance, the Dutch word orkest has the following s-form: /:irkcstr/. Unlike what happens in the case of filter, the morpheme-final extrasyllabic Ir! is not saved by schwa-insertion, but is deleted instead, witness the singular noun orkest and the plural form orkesten. A similar example from English is the word bomb which has a stem-allomorph with !bl, as in to bombard, and an wallomorph without !bl as in to bomb. it does not make sense to postulate a separate rule of /bl-deletion for such isolated cases. In those cases where the lexical entry contains both an s-form and a w-form because the latter is not regularly derivable from the former, the existence of the w-form will block the application of the general rule 'an s-constituent may A similar analysis is possible for the word level rules of Dutch that Trommelen (1993) adduced in favour of Borowsky's model. Trommelen's proposal is also at odds with the observationgiven above that in Dutch certain lexicalphonologicalrules have to apply after level 2 morphology. 24
The listing of both allomorphsin the lexicon was also proposed by Lieber (1981) in relationto Umlaut-phenomena. 23
25
24
are not stress-neutral, and induce reapplication of the Main Stress Rule of Dutch (Booij, in press), for instance the level 2 suffix adjectival suffix -ig. Therefore, an alternative to Borowsky's analysis is called for. Such an alternative is available once we make use of the difference between sconstituents and w-constituents proposed and discussed above. A lexical morpheme like filter with the underlying form /ftltr/ should be specified in the lexicon as an s-constituent. We now assume the rule that all s-constituents can be turned into w-constituents without affixation. This is necessary when such a morpheme is to be used as a word, or as the input for w-affixation. The rules discussed above will be assigned the domain w. Hence, as soon as the wconstituent is created, a schwa wil be inserted before the extrasyllabic Ir! of /ftltr/, in accordance with the principle 'Apply a rule when possible', and subsequently, the word can be used as a free form or as a base for wsuffixation. An independent argument for this approach is provided by the following observation. Quite a number of non-native words of Dutch have two allomorphs, one to be used in s-suffixation (non-native suffixation), and another one in w-suffixation (native suffixation), as shown by the following examples, where the first line represents s-suffixation, and the second line w-suffixation: (35) trauma 'id.'
traumat-isch 'traumatic', traumat-oloog 'traumatologist' trauma's (pl.) ratio 'id.' ration-eel 'rational', ration-ale 'reason' ratio's (pl.) Plato 'id.' platon-isch 'platonic', platon-ist 'id.' Plato's (genitive) cursus 'course' curs-ist 'student', curs-orisch 'by means of a course' cursuss-en (pl.) apostel 'apostle' apostol-isch 'apostolic', apostol-aat 'apostolate' apostel-en (p.) orkest 'orchestra' orkestr-atie 'orchestration' orkest-en (pl.)
be dominated by a w-constituent' in virtue of the Elsewhere Condition. If a lexical entry is not represented as s-constituent, it can only function as w. This is the case with native morphemes and words. Hence, the difference between non-native and native morphemes can be encoded by assigning sconstituency to non-native morphemes only, whereas the native morphemes remain unspecified as to constituency at the underlying level, and are given the w-status by rule. The following generalizations fall out from this approach: (36) a. Non-native affixes do not attach to native morphemes b. Native affixes are peripheral with respect to non-native affixes In short, Borowsky's word level rules should be reinterpreted as rules that apply in the domain of the phonological word w, as soon as that domain of application has been created. Thus, we can keep the notion 'word level rule' to refer to those rules of lexical phonology that apply after all morphology has been performed. Moreover, we can now avoid a very language-specific organization of the lexical phonology of English.24
6. Conclusions In this paper I have tried to identify the theoretical core of Lexical Phonology as the following minimal assumption: 'Apply a rule when possible'. The application of rules is further determined by its domain of application, and the principles of Strict Cyclicity and Structure Preservation. With these minimal assumptions it is possible to correctly predict the way in which phonology and morphology appear to interact. It is not necessary to assign a rule to a specific level of the grammar (although it may be the case that certain rules have to be turned off at a later level of derivation). This implies that, in principle, rules can apply at more than one level, for which we saw empirical evidence.
In these cases it is impossible to derive one allomorph from the other by means of a regular phonological rule. Hence, we have to list both allomorphs in the lexicon23 , one as the s-form, and one as the w-form. In a few cases, the wallomorph is derivable from the s-form, for instance in the case of the morphemes ending in /Cr/, as shown above. In other cases this does not make sense. For instance, the Dutch word orkest has the following s-form: /:,rkl:str/. Unlike what happens in the case of filter, the morpheme-final extrasyllabic /r/ is not saved by schwa-insertion, but is deleted instead, witness the singular noun orkest and the plural form orkesten. A similar example from English is the word bomb which has a stem-allomorph with /b/, as in to bombard, and an wallomorph without /b/ as in to bomb. It does not make sense to postulate a separate rule of /bl-deletion for such isolated cases. In those cases where the lexical entry contains both an s-form and a w-form because the latter is not regularly derivable from the former, the existence of the w-form will block the application of the general rule 'an s-constituent may 24
23The
listing of both allomorphs in the lexicon was also proposed by Lieber(I981) in relation to Umlaut-phenomena.
A similar analysis is possible for the word level rules of Dutch that Trommelen (1993) adduced in favour of Borowsky's model. Trommelen's proposal is also at odds with the observation given above that in Dutch certain lexical phonological rules have to apply after level 2 morphology.
26
References
Anderson, S.R. 1979. 'Notes on the development of phonological theory'. In E.Fischer-J0rgensen et al. (eds.) Proceedings of the 9th Int. Congress of Phonetic Sciences, Copenhagen: Inst. of General Linguistics, 133-142 Amason, K. 1992. 'Problems in the Lexical Phonology of Icelandic'. In: Dressler et al. (eds.), 5-14 Aronoff, M. 1988. Review of A.-M. di Sciullo & E. Williams, On the Definition of Word, Language 64, 766-770 Bloomfield, L. 1933. Language. New York: Holt, Rinehart, Winston Booij, G.E. 1981. 'Rule ordering, rule application and the organization of grammars'. In: W.U. Dressler et al. (eds.) Phonologica 1980. Innsbruck: Institut fiir Sprachwissenschaft, 45-56 Booij, G.E. 1982. 'Lexicale Fonologie en de organisatie van de morfologische component'. Spektator 12, 169-188 [English version in E.Gussmann (ed.) Rules and the Lexicon. Lublin: Catholic Univ. of Lublin, 1987, 43-65] Booij, G.E. 1984a. 'French C/0 alternations, extrasyllabicity and Lexical Phonology'. The Linguistic Review 3, 181-207 Booij, G.E. 1984b. 'Neutral vowels and the autosegmental analysis of Hungarian vowel harmony'. Linguistics 22, 629-641 Booij, G.E. 1985. 'Lexical phonology, Final Devoicing, and subject pronouns in Dutch'. In H.Bennis & F. Beukema (eds.) Linguistics in the Netherlands 1985. Dordrecht: Foris, 21-26 Booij, G.E. 1985b. 'Coordination reduction in complex words: a case for prosodic phonology'. In: H. van der Hulst & N. Smith (eds.) Advances in Non-linear Phonology. Dordrecht: Foris, 143-160 Booij, G.E. 1986. 'Icelandic vowel lengthening and prosodic phonology'. In: F.Beukema & A. Hulk (eds.) Linguistics in the Netherlands 1986. Dordrecht: Foris, 9-18 Booij, G.E. 1988. 'On the relation between Prosodic Phonology and Lexical Phonology'. In: P.M Bertinetto & M. Loporcaro (eds.) Certamen Phonologicum. Torino: Rosenberg & Sellier, 63-76 Booij, G.E. 1989. 'Complex verbs and the theory of level ordering'. In: G.E. Booij & J. van Marie (eds.) Yearbook of Morphology 1989. Dordrecht: Foris, 21-30. Booij, G.E. 1992. 'Lexical Phonology and Prosodic Phonology'. In W.U. Dressler et al. (eds.), 49-62 Booij, G.E. in press. The Phonology of Dutch. Oxford: Oxford University Press Booij, G.E. & J.Rubach 1984 'Morphological and prosodic domains in Lexical Phonology'. Phonology Yearbook 1, 1-27 Booij, G.E. & J.Rubach 1987. 'Postcyclic versus postlexical rules in Lexical Phonology'. Linguistic Inquiry 18, 1-44 Booij, G.E. & J. Rubach 1991. 'Lexical Phonology'. In W.Bright (ed.) The Oxford Encyclopedia of Linguistics, Vol 2. Oxford: OUP Booij, G.E. & R. Lieber 1993. 'On the simultaneity of morphological and prosodic structure'. In: Hargus & Kaisse (eds.), 23-44 Brame, M. 1974. 'The cycle in phonology: stress in Palestinian, Maltese and Spanish'. Linguistic Inquiry 5, 39-60 Borowsky, T. 1986. Topics in the Lexical Phonology of English. Doct. diss. UMass at Amherst Borowsky, T. 1993. 'On the word level'. In: Hargus & Kaisse (eds.), 199-234
27
Chomsky, N. & M. Halle 1968. The Sound Pattern of English. New York: Harper & Row Clark, M. 1990. The Tonal System of Igbo. Dordrecht: Foris Clements, G. N. & S. J. Keyser 1983. CV Phonology. Cambridge Mass.: MIT Press Durand, J. 1990. Generative and Non-linear Phonology. London & New York: Longman Dressler, W.U. et al. (eds.) 1992. Phonologica 1988. Cambridge: Cambridge University Press Fabb, N. 1988. 'English suffixation is constrained only by selectional restrictions'. Natural Language and Linguistic Theory 6, 527-539. Goldsmith, J. 1990. Autosegmental and Metrical Phonology. London: Basil Blackwell Guerssel, M. 1986. 'Glides in Berber and syllabicity'. Linguistic Inquiry 17, 112 Gussenhoven, C. to app. 'The Dutch foot and the chanted call.' Journal of Linguistics Hall, T. A. 1989. 'Lexical Phonology and the distribution of German [~] and [x]'. Phonology 6, 1-17 Hall, T. A. 1992. Syllable Structure and Syllable-related Processes in German. Tiibingen: Niemeyer Hall, T. A. 1993. 'The phonology of German /RI'. Phonology 10, 83-106 Halle, M. & K.P. Mohanan 1985. 'Segmental phonology of modern English'. Linguistic Inquiry 16, 57-116 Halle, M. & J.-R. Vergnaud 1987. An Essay on Stress. Cambridge Mass.: MIT Press , Hargus, S. 1993. 'Modeling the phonology-morphology interface'. In: Hargus & Kaisse (eds.), 45-74 Hargus, S. & E. Kaisse (eds) 1993. Studies in Lexical Phonology. San Diego: Academic Press (Phonetics & Phonology, 4) Harris, J. 1983. Syllable Structure and Stress in Spanish. Cambridge Mass.: MIT Press Harris, J. 1987. 'Non-structure preserving rules in Lexical Phonology. SouthEastern Bantu harmony'. Lingua 72, 255-292 Harris, J. 1989. 'Towards a lexical analysis of sound change' in progress'. Journal of Linguistics 25, 35-56 Hayes, B. 1990. 'Precompiled phrasal phonology'. In: S.Inkelas & D. Zee (eds.) The Phonology-Syntax Connection. Chicago; Chicago University Press /Stanford: CSLI. Hayes, B. & M. Abad 1989. 'Reduplication and syllabification in Ilokano'. Lingua 77, 331-374 Hualde, J.I. 1991. Basque Phonology. London & New York: Routledge Inkelas, S. 1989. Prosodic Constituency in the Lexicon. Doct. diss. Stanford University Inkelas, S. 1993a. 'Deriving cyclicity', In: Hargus & Kaisse (eds.), 75-110 Inkelas, S. 1993b 'Level telescoping and Lexical Phonology'. Paper given at the Diisseldorf Workshop on Recent Developments in Lexical Phonology', Aug. 26, 1993 Inkelas, S. & D. Zee 1988. 'Serbo-Croatioan pitch accent'. Language 64, 227248 Ito, J. 1986. Syllable Theory in Prosodic Phonology. Doct. diss. Univ. of Mass.
28
Amherst Ito, J. 1989. 'A prosodic theory of epenthesis'. Natural Language and Linguistic Theory 7, 217-259 Iverson, G. K. & J. Salmons 1992. 'The place of Structure Preservation in German diminutive formation'. Phonology 9, 137-144 Kager, R. W. J. 1989. A Metrical Theory of Stress and Destressing in Dutch. Dordrecht: Faris Kaisse, E.M. & P. Shaw 1985. 'On 'the theory of Lexical Phonology'. Phonology Yearbook 2, 1-30 Kang, D. 1992. 'Word-internal prosodic words in Korean'. Proceedings of NELS 22. 243-257 Kaye, J. 1992. 'On the interaction of theories of Lexical Phonology and theories of phonological phenomena'. In W.U.Dressler et al. (eds.), 141-155 Kenstowicz, M. & J.Rubach 1987. 'The phonology of syllabic nuclei in Slovak'. Language 63, 463-497 Kiparsky, P. 1979. 'Metrical structure assignment is cyclic'. Linguistic Inquiry 10, 421-442 Kiparsky, P. 1982 'From Cyclic Phonology to Lexical Phonology'. In. H. van der Hulst & N. Smith (eds.) The Structure of Phonological Representations, Part 1. Dordrecht: Faris, 131-175 Kiparsky, P. 1984. 'On the Lexical Phonology of Icelandic. In C.C. Elert et al. (eds.) Nordic Prosody Ill. Stockholm: Almqvist & Wiksell, 135-164. Kiparsky, P. 1985. 'Some consequences of Lexical Phonology'. Phonology Yearbook 2, 85-138. Kiparsky, P. 1993. 'Blocking in non-derived environments'. In: Hargus & Kaisse (eds.), 277-314 Lieber, R. 1981. On the Organization of the Lexicon. Bloomington: IULC [Reprinted by Garland Press, New York] Macfarland, T. & J. Pierrehumbert 1991. 'On ich-laut, ach-laut, and Structure Preservation'. Phonology 8, 171-180 Malikouti-Drachman, A. & G. Drachman 1992. 'Greek clitics and Lexical Phonology'. In: W. U. Dressler et al. (eds.), 197-206 McCarthy, J.J. & A. Prince 1990. 'Foot and word in Prosodic Morphology'. Natural Language and Linguistic Theory 8, 209-283 McCarthy, J. J. & A. Prince to appear. Prosodic Morphology I. Constraint Interaction and Satisfaction. Ms. UMass Amherst/ Rutgers University McCarthy, J.J. & A. Prince 1993. 'Generalized alignment'. In: G.E.Booij & J. van Marie (eds.) Yearbook of Morphology 1993. Dordrecht: Kluwer, 79-153 Mohanan, K.P. 1985. 'Syllable structure and lexical strata in English'. Phonology Yearbook 2, 139-157 Mohanan, K.P. 1988. 'Universals in phonological alternations'. Ms. Mohanan, T. 1989. 'Syllable structure in Malayalam', Linguistic Inquiry 20, 589-625 Mohanan, K.P. & Mohanan, T. 1984. 'Lexical Phonology and the consonant system in Malayalam'. Linguistic Inquiry 15, 575-602 Myers, S. 1991. 'Structure Preservation and the Strong Domain Hypothesis', Linguistic Inquiry 22, 379-385 Odden, D. 1990. 'Phonology and its interaction with syntax and morphology'. Studies in the Linguistic Sciences 20, 69-88 Odden, D. 1993. 'Interaction between modules in Lexical Phonology'. In: Hargus & Kaisse (eds.) , 111-144
29
Paulissen, D. & W.Zonneveld 1988. 'Compound verbs and the adequacy of Lexical Morphology'. In M.Everaert et al. (ed.) Morphology and Modularity. Dordrecht: Faris, 281-301 Packard, J. 1990. 'A lexical morphology approach to word formation in Mandarin'. In: G. E. Booij & J. van Marie (eds.) Yearbook of Morphology 1990. Dordrecht: Faris, 21-37 Pesetsky, D. 1979. 'Russian morphology and lexical theory'. Ms., MIT Rice, K. 1993. 'The structure of the Slave (Northern Athabaskan) verb'. In: Hargus & Kaise (eds.), 145-172 Rubach, J. 1984. Cyclic and Lexical Phonology. Dordrecht: Faris Rubach, J. 1993. The Lexical Phonology of Slovak. Oxford: Clarendon Press Rubach, J. & G.E. Booij 1990a. 'Syllable structure assignment in Polish'. Phonology 7, 121-157 Rubach, J. & G. E. Booij 1990b. 'Edge of constituent effects in Polish'. Natural Language and Linguistic Theory 8, 427-464 Scalise, S. 1984. Generative Morphology. Dordrecht: Faris Schultink, H. 1980. 'Boundaries, word classes, and the accentuation of derived words in Dutch'. In: W. Zonneveld et al. (eds.) Studies in Dutch Phonology. 's Gravenhage: Martinus Nijhoff, 205-222 Selkirk, E. 0. 1982. The Syntax of Words. Cambridge Mass.: MIT Press Shaw, P. A. 1985. 'Modularisation and substantive constraints in Dakota Lexical Phonology'. Phonology Yearbook 2, 173-202 Siegel, D. 1974. Topics in English Morphology. Repr. by Garland, New York Sproat, R. 1986. 'Malayalam compounding, a non-stratum ordered account'. Proceedings of the West Coast Conference on Formal Linguistics 5, 268-288 Sproat, R. & C. Shih 1993. 'Why Mandarin morphology is not stratumordered'. In: G.E.Booij & J. van Marie (eds.) Yearbook of Morphology /993. Dordrecht: Kluwer, 185-217 Szpyra, J. 1989. The Phonology-Morphology Interface. Cycles, Levels and Words. London: Routledge Trommelen, M. 1993. 'Lexical word-processes in Dutch'. The Linguistic Review 10, 161-184 Vogel, I. 1991. 'Level ordering in Italian Lexical Phonology?'. In: P.M.Bertinetto et al. (eds.) Certamen Phonologicum II. Torino: Rosenberg & Sellier, 81-101 Wiese, R. 1988. Silbische und lexikalische Phonologie. Studien zum Chinesischen und Deutschen. Tiibingen: Niemeyer Wiese, R. to appear. The Phonology of German. Oxford: Oxford University Press Wilbur, R. 1983. The Phonology of Reduplication. Bloomington Ind.: IULC Zee, D. 1988. Sonority Constraints on Prosodic Structure. Diss. Stanford University
37
Ba.:se-tl:riv-er1. morphological
causes
:st:ra.tifica.tior1.: and phonological
effects
of 'Strict
Cyclicity'
Heinz J. Giegerich ( University
1.
Deriving
the Strict
Cyclicity
There is widespread agreement side of lexical derivations should of ( 1) below. ( 1)
Strict
Cvclicity
Effect
Structure-changing
of Edinburgh)
cyclic
(where a 'derived environment' morphological or a phonological
Effect' in the literature that the phonological be subject to a constraint to the effect
(SCE)
rules
apply
in derived
environments
is an environment created rule on the same cycle).
by
only·
either
a
Less widespread is the agreement as to the status of (1) in the grammar. Proposed in similar form as the Strict Cycle Condition ('SCC': Mascaro ( 1976), also Kean ( 1974) ), ( 1) represented a clear improvement over its predecessor (the Alternation Condition, see Kiparsky (1982, 1993) and § 5 below for discussion): but it retained a number of problems connected with its essentially stipulative nature. The confinement of sec to structure-changing rules remained arbitrary, as did its confinement to cyclic rules. Moreover, the notion of 'derived environment' is in itself heterogeneous in that it refers to the products of both morphological and phonological derivation. Kiparsky's (1982) argument whereby SCE is an automatic consequence of the independently-motivated Elsewhere Condition ( (2) below) removed sec from the inventory of conditions on the grammar, and with it went some of the arbitrariness arising from its stipulative character. (2)
Elsewhere
Condition
(EC)
Rules A, B apply disjunctively to a form$ iff: ( i) SDA is a proper subset of SDB; and (ii) SCA is distinct from SCB. In that case rule A, applying first, blocks rule Crucial to Kiparsky's argument is the assumption that morphologically simple or complex, re-writes itself as Both [sa,: n] (sane) and [sa,n1tI] (sanity) are identity first and second cycles respectively, stating on their what counts there as an 'underived environment' in
B. every lexical item, an 'identity rule'. rules sited on the respective cycles the sense of SCE.
32
33
Given its one-off nature. such an identity rule will constitute 'rule A', as in (2) above. in relation to any phonological rule whose structural description it matches: and given the identity of its input and output, its output will be distinct from the output of the phonological rule provided that rule is structure-changing. Hence, underived environments block structure-changing rules. Structure-building rules, in contrast, do not produce outputs that are 'distinct' in the technical sense and are therefore not subject to blocking. While motivating the confinement of the blocking effect to structurechanging rules and to derived environments. Kiparsky's proposal effectively severed the link between blocking and cyclic rule application. While this link was stipulative in the first place. and may well have been accidental (Giegerich ( 1988) and § 5 below), its removal nevertheless had the adverse consequence of also removing from SCE the well-founded prediction (Kiparsky 1982, 1985, Halle & Mohanan 1985 and others) that the 'word level'. the (presumably) noncyclic final stratum, is not subject to SCE. Poorly motivated as it is in the first place (Mohahan & Mohahan 1984, Giegerich 1988), the notion of identity rules is further undermined by the unavoidable stipulation that the word level contains no such rules: only then can SCE be avoided on that level (assuming EC to hold throughout the derivation). I return to this issue in § 5. Finally. through 'rule A' Kiparsky. present
Giegerich ( 1988) endorsed Kiparsky's proposal of achieving SCE the operation of EC but argued that the rules that figure as in (2) above are not identity rules in the sense defined by To set the stage for the contribution to the debate made in the paper, I summarise the argument briefly.
Giegerich's ( 1988) proposal draws on Selkirk (1982). who argued that the bases for morphological processes in English fall into the two categories Root and Word, and that the division of the lexical derivation into strata 1 and 2 is an automatic consequence of this categorisation: the inputs to all morphological operations on stratum 1 are roots, as are their outputs (required by the fact that stratum-! affixation is recursive). Root is thus a recursive category which may in itself be morphologically simple or complex. Stratum 2 is word-based: the inputs and outputs of all morphological processes on that stratum are members of the category Word. Members of that category may consist of morphologically simple items ('free roots'): or they may contain forms that are stratum-! complex: or they may recursively contain members of the category Word that have undergone morphological processes on stratum 2 (the word level). In English, Selkirk's proposal receives strong support from the frequent occurrence of 'bound' roots in stratum-1 derivations (matern+ity, vis+ual, re+fer! - roots that cannot enter into the word level without prior affixation - while word-level affixation does not operate on bound bases. (Apparent counterexamples such as gormless. wistful etc .. where bound forms appear to have stratum-2 affixes, differ from those found on stratum 1 in that they are always non-recurrent (*gormful, *wistless !: we shall see below that such formations are amenable to alternative analysis involving stratum-! derivation. if they are morphologicaliy complex at all.)
subscribing to Selkirk's analysis, I observed in Giegerich ( 1988) that for a given form to transit from stratum 1 to stratum 2, it must undergo a rule that converts the root outputs of stratum-[ affixation into the word inputs required by stratum-2 morphological processes (see Selkirk 1982: 45). This root-to-word rule must itself be sited on stratum 1: given that it takes roots as inputs, it cannot be a stratum-2 rule. It is this rule that produces SCE in the way Kiparsky's ( 1932) identity rules did. While not losing any of the improvements that Kiparsky's ( 1982) proposal made on previous accounts of SCE, this analysis has the further advantages, firstly, of not depending on poorly-motivated identity rules (unlike those. the root-to-word rule employed here has strong motivation in the grammar), and secondly, of making the automatic prediction that the final lexical stratum displays no SCE: members of the category Word exit from the lexicon without further (word-to-Xl conversion. Below is a sample derivation demonstrating (in italics l the operation of the root-to-word conversion: for reasons of exposition· I treat nation as monomorphemic (but see § 3.2 below) and simplify the bracketing.
Stratum
( 3)
1
Cycle 1
[nationlr
{
•
[[nationlr
•
[{nationl,.lN
Stratum 2 Cycle :?
allr
• • { •
[[[nationlr f[[nationlr [{{nation],.
al Jr ity]r .. . allr iselr .. . allrlA ------ • •
(nationally l (nationhood)
On the first cycle, the root nation either undergoes suffixation yielding the complex root national: or it converts into the noun nation under the root-to-word rule. Ineligible for further stratum-1 (root-based) affixation, the latter form exits from the stratum to undergo word-level affixation ( e.g. nationhood) or to emerge as nationN from the lexicon. The complex root national is on the second cycle subject to further affixation /nationalit.v, nationalise), or it will convert fato the word nationalA• in which case it will again exit from the stratum to form further words on stratum 2 ( e.g. nationally) or to exit from the lexicon. On the third cycle (not demonstrated in (3)). the root nationality is subject only to the root-to-word conversion that turns it into nationalitvN. while the n,ot nationalise is a candidate for either the further root nationalisation or the conversion into the word nationalisev, prompting the form's exit from the stratum. And so forth. On the phonological side of the derivation, the conversion nationr • nationN blocks structure-changing rules from applying to the simplex form on the first cycle under EC, while permitting structure-building operations ( syllabification, stress etc.). The derived form nationalr, on the other hand, is eligible for structure-building as well as structurechanging rules: in this particular case, Vowel Shortening will apply (Kiparsky 1982, Myers 1987). The shortened (as well as syllabified and stressed) form nationalr enters the second cycle, where the stress pattern is modified appropriately for nationality. nationalise; but the conversion nationalr • nationalA prevents the application of further
34
35
structure-changing relevant to the counterfeeding ( Further details
rules to that form. In other cases (although not present example). this conversion will prevent next-cycle (Kiparsky 1982, Giegerich 1988). And, again. so forth. in § 5 below.)
The analvsis presented in (3) above departs from that given in Giegerich ( 1988) in one important respect: I had assumed there, following Selkirk ( 1982). that roots bear lexical category (Noun/Verb/Adjective) labels, and that the root-to-verb conversion that crucially fi;,ures in mv account of SCE is essentiallv a read iustment rule in that it does nothing but turn into a word a ro'ot that i~ already (underlyinglv) categorised for N. V or A. I shall argue in § 3 below that this is not the case. It must be a characteristic-of the morphological category Root that it is not specified for the lexical categories NVA: those are assigned to roots as part of the conversion into words. in operations that are idiosyncratic to individual roots not onlv in that their outcome (N. V or Al is essentially unpredictable bi.it also in that such conversions ( and the associated NVA specification) may unpredictably fail to apply. in the cases of bound roots /matern+it_v etc.). I give the appropriate rule in (4 l below. As will be shown in § 4.2. the German equivalent to (4i is a 'Root-to-Stem Conversion'. with a further 'Stem-to-Word Conversion' of essentially the same format acting on the second stratum (of threei. (4)
Root-to-Word [
Jr
•
Conversion [ [
IL=
N. V. Al
(4) is short for three separate rules. converting roots into words of the categories Noun. Verb or Adjective respectively. As we shall see in § J below. it is not predictable for a given root whether or not it is subject to (4). nor which subrule of (4) will apply.' This crucial idiosyncrasy in application. 14) shares with Kiparsky's (1982) original identity rules. It is quite
clear that. as long as EC as well as rule (4) form part of the grammar. that rule must cause SCE in the fashion described above. Moreover. (4) must form part of the grammar as long as Selkirk's I 1982) proposal regarding lexical stratification js adopted (although the additional claim that roots are unspecified for lexical categories still has to be substantiated). The case for deriving SCE from EC is therefore only as strong as is Selkirk's case for stratification on the grounds of base category distinctions rather than distinctions among affixes. I shall argue in this paper that Selkirk's case - a case for base-driven stratification and against affix-driven stratification -- is indeed stronger than has been hitherto assumed. The argument will be structured as follows. In 'i 2 . I shall review a number of arguments. in relation to English, against stratification models that rely crucially on affix ordering generalisations rather than base category distinctions. In fj 3 J argue that (in English and German) the morphologies of stratum l and stratum 2 differ from each other in crucinl respects. an cl that these differences are due to the absence of lexical category specifications in roots. This argument will confirm the crucial status of the root-word distinction in tex'ical stratification. a distinction that will be further explored. and
tentatively amended by the intermediate category Stem for German, in 4. In ,; 5. finally. I explore the phonological implications of the particular version of SCE predicted by base-driven stratification. 2.
Problems
2.1
Weaknesses
of affix-driven
§
stratification
of affix ordering
generalisations
Affix-driven stratification models depend crucially on affix ordering generalisations of the kind first proposed for English by Siegel (1974). It is unfortunate for such models that, as is widely documented in the literature (Aronoff 1976, Aronoff & Sridhar 1985, Sproat 1985, Fabb 1988, Katamba 1993). the existence of such generalisations is neither a necessary nor a sufficient condition for strata! splits. The exceptionless generalisation whereby English inflexion does not occur inside the products of the derivational morphology (*legsless. *eventsful) had prompted earlier researchers to posit a final lexical stratum containing regular inflexion as the sole set of morphological processes (Kiparsky 1982. Halle & Mohahan 1985): but this stratum has been abandoned in more recent work (e.g. Sproat 1985. Booij & Rubach 1987, McMahon 1990\. On the other hand. the distinction between nonneutral and neutral affixation continues to be invoked in support of the split between stratum 1 and stratum 2 despite the weakness of that particular ordering generalisation: a given affix may be stress-shifting in some and stress-neutral in other formations ( protestantism vs. catholicism etc.; see Aronoff & Sridhar (1985)); other affixes (e.g. -ment Aronoff (1976) and § 4.1 below) may on morphological grounds be members of both stratum 1 and stratum 2 - an indeterminacy that seriously challenges the principle of affix-driven stratification. Even the basic claims of this ordering generalisations have been called into question (Fabb 1988); but all such weaknesses. although well-known, have been overridden by arguments in favour of the strata! split on the phonological side. It is clear that the case for the division of stratum 1 and stratum 2, already strong in phonological terms. would benefit from morphological arguments not reliant on the affix ordering generalisation. 2.2
Nonnative
and native
morphology
Both in English (Kiparsky 1982 and others l and in German (Giegerich 1985. Wiese forthcoming), the distinction between stratum-1 and stratum-2 affixes correlates strongly (if by no means one-to-one) with the distinction between (essentially Romance) affixes of the learned vocabulary and those. mainly Germanic, suffixes that characterise the derivational morphology of the everyday language. Various researchers have used the feature [± native] to express this distinction (e.g. Aronoff 1976. Booij 1977. Plank 1981 ). If this feature merely served to classify affixes it would be redundant in an affix-driven stratification model: *fearlessit_v is ruled out by the affix ordering generalisation without necessary recourse to the etymology of the offending affix. But the
37 36
.. • rtv of the feature [± native! in morpholodcal processes b1fu_rcatm g prope . . t· • 1 ut·t·i·xes not on lv foil to att_ach to than . .that· s · · hut· ,1\so · to I+ native I roo t s goes j .ur ther· . . . 1- na ..ne •] ··uffixe•, bases already contammg I.+ natnc t f [+ native I · *s/JorritY *bookici. while conversclv the ,ltLlc.nmc~ h ( ·f·t·· ·t· r native! roots is possible (soiemn/v. contmne1l Just a:" ,:sue: su 1xes o _. ~ . r· t' . ·1 -ut·fives ( n ·,tiona/lv. scemcness1. ·t·· f lv occur outside - na 1ve ,, " • • .. . . . su f 1xes rce. . . · · rd·J"•· t·10 that rule•, out *fearless1t_v 1s m Clearly, the affix ordering gene i.,a .11. . ' th 3 t rontrols the possible l' d · t'·ve of a lar er genera 11:,at1on ,._ -·. · , rea ity er1va i -. d ~·- .··ve ,material in rnorpholo_g:ica1 processes. combinations of nanve an nonnan · b. of researchers [" t· ha· been noted bv a num er This genera isa 1°1 /' ld l'P0 Saciuk ·1969 , Booij 1977. Plank 19811. 0
.
•
•
•.••
•
0,
,
·
0
;}6:he ~:::
~:rrC:~~1e_ ~~exical 'st~atification' whas. _oct_c~sliorE1_ aalmh·e, wuos1ek~ ~~c°ri~ · · · the present t core 1c,1 r, . · it acquired its formal sen_se rn_th· this framework. due probably to the has receivt~d swca1_tnht aaii~;ttoe:a:~ou'; rather than base behaviour that has preoccupa.
2.J
~
10n
~haractcrised
research.
Productivitv
differences
Beyond the expression of the blocking effect, affix-driven stratification offers no facilities for dealing with less-than-full productivity in a principled way. The framework is only able to make relative productivity predictions among competing affixes: it does not rule out the siting of a fully productive morphological rule on stratum 1. Such a rule may be subject to blocking by a rule of lesser productivity disjunctively ordered earlier on the stratum (see above); but it will have no competitor on a later stratum: given its complete productivity. it would block such a competitor completely. For English (and German), the prediction that such rules mav exist is clearly false: all stratum-1 morphological processes have severely restricted productivity. Partly this is due to the [± native] distinction: while the nonnative affixation processes were clearly productive in their donor languages, they arc unproductive in English: it seems, for example, that the set of attested -itv derivatives was borrowed wholesale (over a bUbstantial time-span, of course) and has not increased since. ( See Marchand ( 1969), Bauer ( 1983) for discussion). Typically. such affixations have large numbers of unpredictable gaps (fabb 1988). Native stratum-! affixes (for example -th, noted above) are typically fossilised, attaching to a mere handful of bases ( wa.rm, wide, strong, long) which have no generaliseable properties beyond, in this case, being adjectives. I shall argue in the next section that the assumption of a standard format of affixation rules throughout the lexical strata is at fault in the model's failure to express incomplete productivity. 2.4
the bl k" o- effect within a given stratum. competing To nchie".e . ,o_c mr ct·,. tivelv ordered within that stratum morphological . procesf:'es ".re l 1s,1luvntchecase -th I wannrh. width) is even 'K' arskv 198~· 7) I as is ccar ,. . , ... t \ 1p· " . . . ~· . ·_. : . d if this differential in productzv1tv is o less prod~-c~v~h~~~~h lt·10~7zing effect, then -th attachment has rn be be expres_,,e ~ f _- ,·t1, -ottachznent within stratum l. Note thaE . · ctivelv ordered be me " , .. d lSJUTI ·. ' ' ·,c, it>. f natiVt·' ,1.ffix (Lttact1i.fl'½ to nauvc the -th :.lff1xatJon is an t',>;..ll11pf,._· i..) tic\ bl,1ck1·-no-- -itrr is therefore noc a t 1· the case or - 1 l bases on stra_ um .h th .d.18·tr·bution of r ± native l alone accounts for 1 strong one (given t at e · . ·. f )etino- affixes to
It{e
c,_•s•
C
_c
.•
'
:~~1;~~
t~e ~~ret;,en;{~~~~~t~;;~).s~:ste::~(c a~ai~~ree ~ss~:ilon_ ,;f cydc:ic_ity s ac 1 1 . al rules probablv means that d1sJunctn e or enng among P 1ono ogic · · ·d G·ven thcre'ore within strata cannot be supportedd tb_yto.thed11:ff::~;;~cefr/.~ o;derin,/. o1: ' ·ther predicts a pro uc n i v · t h a t EC , · 81 . ' · . f · d ths.t other evidence for t 11e ordering from a pr~ductivity d1f erence_,n are st;atum is unavailable. the ordering: of competing: 1affixes k_on ~o s~r!ctictiom; regarding productivity ~i~~fnwC:~i:::e:tl-~~~~t
ial~~i~g1n~he
device
of expre;sing
such
observed
is available).
The format
of affixation
processes
It is at least consistent with an affix-driven approach to lexical stratification that affixation rules should themselves be affix-driven in the sense that a given affix should attach to a range of bases with generaliseable properties (expressed in a subcategorisation frame). Hence it is generally assumed that all affixation rules have the form (5) below, regardless of their strata! siting and productivity (Lieber 1980, Kiparsky 1982):
(5)
Insert
A in the environment
[Y --~
(where Y and Z express the subcategorisation the lexical category specification of the base).
Z]L frame
of the affix A, and
L
If a given affix is subject to local productivity restrictions then these have to be stated, as part of the rule, in the subcategorisation frame; in the case of suffixes such as -it_v. -ive etc. ( abusive vs. -tfindive: Lieber 1980: 34 ff.) the feature [- native] forms part of the subcategorisation frame. But the following problems. in turn connected in part with the model's failure to deal with incomplete productivity. cast doubt on the appropriateness of (5) as the general affixation format.
38
39
First, productivity restrictions ty picalJ,, can not be ,:,Lated in a generaliseable form: the (adjectival) bases to which -th attacties, for example, have to be listed. This listing obviate,; the su bco.tegorisation frame: and it typically occurs with stratum-! affi::ations. Second, such a general format fails to express the fact that unproduct,vc• morphology typicaliy gives rise to semantically idiosvncratic (noncomponential) outputs. (See here Aronoff's (1976: .15 ff,) discussion of -ity vs. -ness; also Anderson ( l 992: 195 ff.)) The fact that certain aspects of the rule's output have to be listed even when the nonsemantic mechanics of the rule are regular (albeit perhaps including the feature [± native]) clearly affects the rule's status quite profoundly; but the framework treats such rules as identical in nature to the ones whose outputs do not require any listing. Again, the listing of the outputs of (51 with reference to their semantics is typically associated with stratum-1 affixations. Third, recall
stratum-!
typically nble to attach to bound roots mentioned in s 1 above, as well as and many more. In such cases, the lexical category of the base ('L' in (5) above) cannot be determined in a nonarbitrary way: assuming the validity of (5) on stratum I, probably every such rule would have to be amended by a separate list of uncategorised bound :roots that are also candidates for affixation by the rule .
matern+ity, Jev+ity, deterg+ent
affixes
are
vis+ual,
re+fer
Fourth, as we have seen, nonnative affixes typicallv attach to nonnative bases. and again this happens predominantly on stratum L While an affix-driven model can express the stratum-! siting of such processes indirectly (if not on principled grounds) by simply placing the relevant affixation rules appropriately. it can express the nonnativeness requirement of the base onlv bv writing [- native] into the subcategorisation frame of every rule that is so re.stT]cted. so long· as such rules have the format (5). This misses a generalisation that i, (at least) as strong as the original affix ordering generalisation. and again one that pertains to stratum 1. 0 ;
_,. '
Rase-driven
J.1
The nature
affixation
of the
on st rntum
categorv
Root
I observed above that Selkirk's ( 1982) proposal whereby the distinction between strata 1 and 2 is a consequence of the category distinction between Root and Word is motivated by the occurrence of 'bound roots' in stratum-1 affixation processes (nrntern+ity. vis+ual), where a bound root is a form that cannot enter into the word level without prior affixation. Such bound roots have the additional property of hearing no non-arbitrary lexical category specifications: vis- could only be aso;ismed the label 'N' on the strength of the arn.ument that in other instances, -ual attaches to nouns (textual, sensua!J; the assignment of the label 'Adj' to matern- (motivated only by the generallsation that -ity ,:,Jsc,where attaches to adjectival bases) is even more arbitrar)' in that the same root
would
have
to
paren ta1. proced
be ll
specified
as
'N'
in
maternal
(-al attaches
to
nouns:
ra]).
In establishing the properties of the category Root, we first modify the definition of 'bound': a bound form is one which cannot enter the next stratum without undergoing a morphological operation. From what was said in 'i I, where the root-to-word rule (4) was po,sited as part of the morphological operations of stratum 1. it follows that all inputs to stratum-I affixation, as well as all outputs of such affixation. are bound: neither underived roots (nation, divine) nor complex forms produced on that stratum !national, divinity) can enter into the word level without under.;wing rule (4) - recall the sample derivation (3). 'Bound' roots under the former, narrower definition (e.g. matern-J differ from other root.", oiliy in that they cannot enter the word level without prior affic:;ition (with subse::iuent application of (4)); such roots (which I shall informally call 'defective') are not in themselves subject to (4). This is dernc,nstratecl fo (6):
Stratlllll .1
( 6)
Stratllm Cycle
C~vcfe 1 •
[[maternlr
allr
• { •
{
[maternlr
-/ [[maternJi.h
1
-,
(f(maternJr
alJrlA
---
•
:
(n/a)
Next. as also demonstrated in (3) and (6), I take the view that all roots (and not just defective roots such as matern-l lack lexical category specifications. an cl that such lack of specification is the defining property of the morphological category Root, Moreover, affixation processes on stratum 1 fail to assign lexical categories. This is a necessary consequence of the assumption (Selkirk 1982) that Root is a recursive category: in the present framework it means that the root-to-word rule (4) has the property of both producing words and assigmng lexical categories (whose presence is in turn one of the defining properties of the morphosyntactic category Word), For the moment 1 assume furthermore that all lexical items owe their lexical categorv specifications to rule (4) (but see § 4.JL The decision not to associate lexical category assignment automatically with affixation processes ( -ity formations. for example, do not in such a mode.I automatically become nouns qua affixation ), is justified by the fact that all stratum-! affixations are attested as attaching to defective roots, which cannot be mernber::: uf lexical categories. -_ity attaches to maternctc .. and of course to a number of roots that become adjectives through rule (4). -ise attaches to the defective root bapt-, as well as to terror (becoming a noun through (4)), to the affixed form national (which becomes an adjective through (4). and so forth. Tt follows that stratum-1 affixation is not crucially sensitive to any lexical category labels that its
40
41
bases may carry. In the present account, buses carry no such labels; lexical categories are assigned (by (4)) onlv to forms that do not constitute inputs to stratum-1 affixation. I deal with this question further in the next section. For the moment, we note firstly that the recursive morphological category Root is defined as (a) bound and (b) not specified for lexical categories, secondly that rule (4) performs the root-to-word conversion in that it (a) supplies the lexical catcg01·y to any given form, (bl removes that form from the possible inputs te straturn-1 affixation and (c) facilitates that form's entry into the word level: unlike roots. words are members of lexical categories. Thirdly, note that rule (4) is a necessary consequence of the definition of Root as a recursive unlabelled category (a definition strongly supported bv the facts of stratum-1 affixationl. Given the fact that (4) is the cause of SCE (6 1 .1bovel on stratum 1. it is now clear that the categorial distinction between Root and Word is SC:E's ultimate cause. 3
°
The format
of root affixation
It was argued in ,; 1.4 that the affixation format (5) ('Insert A in the environment [Y____ZJr,') is inappropriate to stratum-] affixation on various counts: (a) productivity TC'strictions, wbieh nre 2\:trcrnely common. are not amenable to the type of generalisation demanc!cd by /5). (b) oc1tputs have to be listed with regard to noncomponentia1 sernanLics. (cl lists of defective roots have t~ be added to each rule's structu rcli description. and (cl) [ - native l has to be entered into all su bc::ttegorisation frames for [- native] affixes. I return to (cl) in § 4.3 below. for the moment merely noting that the definition of Root given in " J.1 b particularly pertinent to items that are of nonnative etymologies and that such items are therefore more likely than others to be handled nn stratum I. (a)-(c) all call for the listing of the outputs of stratum-] affixation. In addition. it is now clear that the format (5 ), if it were to be maintained lln stratum L could not contain the lexical category specification 'L' in its structural description. Without this essential part of its •;tructural description, (5) becomes untenable as the rule format for root affixation.
I propose that all lexica.l items (morphologically simple roots as well as affixes) are listed in the lexicon. and that moreover each such lexical entry contains a list of ::tll morphological operations that the particular item is potentially subject to: all available affixes as well as, where appropriate, rule (41 with the lexiciil category that it specifics. are t.hus listed with any given simple root as well as with any given affix. Some examples are given below. ( 7)
serene
-ise
{ {
-al
• • •
-i ty -ade Adj
•
-at ion
-ade
•
N ( 4)
•
V ( 4)
-ity
•
N
nation ( 4)
(
•
-al •
t~ (4 ~
matern (4)
(
• • •
-i ty -ise Adj
fl
•
-i ty
•
-al
( 4)
lt p,oeE; 1.v·irhout with :-:d1 TfTixes
sayinµ that not every spenLe:_· •Nill h /X+ta/ Rule B:
Disjunctivity Stipulation: A ::i -B (-- or, equivalently: A and B form a "block" of rules that are exclusively disjunctive with each other) On the basis of these mechanisms, the following derivations for sahe and machte (both intended as past subjunctive forms) are possible. (LPD is Janda's equivalent to the Elsewhere Condition. The general past rule (18 A) does not apply to sah as this entry is more specific than the rule.) Crucially, the Disjunctivity Stipulation (STD) prevents umlauting (by rule B) of /a/ in machte. That is, in the regular form the more
general rule, and not the more specific one, must apply. With the Elsewhere Condition in force, this would be impossible. 13 (19) (= Janda (25)); (see also Anderson 1982: § 4.2.2) UR /sah/[ + p ast l A --LPD--
vs.
a
e PR [sahe] B
/mach-/ te ----StD---[machte]
However, an alternative to this analysis is readily available. Suppose that, in contrast to Janda's treatment of the subjunctive, we separate the Umlaut rule from that of forming the past subjunctive. (Note that Janda's past subjunctive rule (18 B) is just one instance of umlauting, except that it adds schwa is there isn't any.) The rules needed are then t!J_osein (20). (20) Rules for regular past tense: a. X +V ] • X + It/ (level 2/3) [ + past b.
X
[~fast]
•
X + /a/
(level 2/3)
This treatment presupposes that the irregular past is formed at level 1 by means of lexical entries, as discussed below, and that the regular past is formed at level 2 (or 3) of the lexicon, by the two rules given in (20). 14 Note that Janda's rule (18A) is decomposed into two rules here. Rule (20b) provides second and third person singular forms with schwa, both for indicative and subjunctive past tense forms. This rule, however, is not the only source of a final schwa: one other source is a general constraint for all subjunctive forms of verbs. It is an exceptionless generalization that all subjunctive forms end in a schwa syllable. A final schwa (as in kame, sahe) appears simply to fulfill this prosodic requirement. Finally, Umlaut is taken to be a phonological rule not disjunctive with any of these rules, but one which never applies at the level of regular inflection (level 2/3). This then accounts for the "disjunction" between affixation of past marker /t/ and umlaut: they are in complementary distribu13: The use of orthographic forms as phonemic representations is Janda's. 14: The exact number of lexical levels for German is not relevant here. The need for
12: I have changed Janda's example haben to sehen. The former verb belongs to the class of mixed verbs, with additional complications I do not want to discuss here.
or at least usefulness of a level distinction has been argued for by Giegerich (1985), Wiese (1988), Hall (1992), Wiese (1993), Giegerich (1994) and others.
107
106 tion over the lexical levels. The fact that irregular verbs umlaut, and that regular verbs are suffixed by past tense /ti, motivates this proposal. The application of these rules under the assumptions made is sketched in (21).
15
As a
result of the assumptions on rule application, irregular past forms (strong verbs) are
grounds of parsimony. More particularly, approaches such as Optimality Theory (McCarthy & Prince 1993) demonstrate how independently motivated static constraints can derive words with alternations which, on first sight, seem to require a process morphology approach. Second, it is worth repeating that Umlaut in German is
not a morphological marker such as the affixes are. If it is a marker, it is only a very
not subject to the rules in (20).
abstract one, marking nothing but the presence of a derived category.
(21) derivation of past subjunctive kiime and klagte:
It is also worth noting here that Umlaut is equally present in connection with deriva-
level 1 derivation: a.
/k a: m/ V, [+ past]
entries
/kla:g/v
b. /k a: m/ V, [+past] + subJ.
tional and with inflectional morphology. This creates a problem for a theory (such as that of Anderson 1992) which claims, first, that these two kinds of morphology are
adding subjunctive feature
fundamentally different, and, second, that Umlaut is a morphological rule. For Anderson, derivational •morphology is a lexical process, while inflectional morphology is
/k
E:
m/ V, [+past] + subJ.
Umlaut for past subj.
part of syntax. But if Umlaut is a morphological rule, it must then straddle the lexi-
d. /k
E:
m Q/ V, [+ a
prosodic schwa
cal and the syntactic component of German. In such a theory, Umlaut might as well be called a syntactic rule.
C.
t] + ~UgJ.
The overall evidence for Umlaut as a morphological rule is thus weak. While in phon-
level 2/3 derivation:
e.
/k 1 a: g t/ V, [ + past]
reg. past formation, (20a)
f.
/k 1 a: g t Q/V, [ + past]
2/3.ps. affix (20b)
g.
/k 1 a: g t :l/v,
[+ +
a
t]
adding subjunctive feature
~UgJ.
umlaut for past subj.
h.
ological terms, Umlaut is a homogeneous and local process, and fits the constraints known for phonological rules (it is a simple linking rule), it is not clearly related to any one morphological process. Furthermore, assuming a morphological Umlaut rule leads (in the case of the past subjunctive) to an analysis which violates an otherwise reasonably well motivated general principle (the Elsewhere Condition).
The point of the present re-analysis is that Umlaut can count as a potential counterexample to the Elsewhere Condition only if it is taken to be a morphological rule. Only then is a stipulation reversing the priority in application (Disjunctivity Stipula-
4. Ablaut
tion) required. Under the alternative analysis, no such violation of the EC is found.
contrasting it with the other major vowel alternation
Theories which accept Umlaut as a morphological rule are almost by necessity theories of process morphology (item-and-process
models), unless one accepts the mas-
sive amount of allomorphy, as Lieber (1987) in effect does. While process morphology is currently one widely accepted paradigm of morphology, it is arguably the case that a more static model of morphology (item-and-arrangement)
is more desirable on the
The properties of the Umlaut alternation
can be brought to light more clearly by in Modern German, that of
Ablaut, as briefly introduced in (2). Anderson (1985: 167) summarizes one common view of the relationship between the two phenomena as follows: " .•. from the point of view of the synchronic morphological system, it [Umlaut] is simply another alternation of vowel quality like Ablaut, correlated with certain grammatical categories". It is this view, commonly held in the present literature, which I will now challenge. It can be attacked on two grounds: firstly, the internal systematics of Ablaut is funda-
15: The verb klagen is a more adequate example for demonstrating that the subjunct-
mentally different from that of Umlaut; secondly, arguments exist that the place of
ive forms do not umlaut. It is a fact about Janda' s verb, machen, that none of its
Ablaut in the grammar of German is different from that of Umlaut.
forms ever umlaut; see Mach-er, etc. In contrast, for klagen we find Kliig-er and
kliig-lich. The theory proposed above expresses this contrast by assigning the floating [+
front] to klag, but not to mach.
The patterns of Ablaut are represented in (23) by one verb each. In addition to the changes in vowel quality illustrated
here, there are differences
in length and/or
108 tenseness of the vowels, which are ignored here. (For this reason, the vowel patterns are stated in the table by orthographic symbols only.) (23) Ablaut patterns for strong verbs:
vowel pattern aia au a ea e ea o eoo eia i au iao iae i
O 0
oio oao 000
DUO
uiu uaa ii O 0 ai ii au i au au o o
Infinitive simple past fiel fallen trug tragen las lesen stahl stehlen hob heben hing hangen fand finden schwimmen schwamm bitten bat bot bieten stieJ3 stoilen kam kommen erlosch erloschen schwur schworen rief rufen tat tun log liigen glitt gleiten lief laufen soff saufen
past participle gefallen getragen gelesen gestohlen gehoben gehangen gefunden geschwommen gebeten geboten gestoJ3en gekommen erloschen geschworen gerufen getan gelogen geglitten gelaufen gesoffen
'fall' 'carry'
'read' 'steal' 'lift' 'hang' 'find'
109 16
subject to the alternation. But such observations still have a status different from those to be made about Umlaut. As Bybee notes on the irregular verbs of English, but as can be said with equal justification for the strong verbs of German, possible generalizations are those "formed over the shape of the derived past tense form, rather than as a relation between the basic, present form and the past form" (Bybee 1993: 2). There are other ways of demonstrating that Umlaut and Ablaut are not integrated into the derivation of words in the same way. I will discuss here the relation of these two phenomena to conversion (i.e., zero-derivation) of word class. There are several types of conversion in German, which are distinguishable in terms of their productivity and their semantic effects, and not just by their input and output categories.
'swim'
'ask' 'offer' 'push' ,,come'
'go out' 'swear'
'call' 'do'
'lie'
'slide' ,,run'
'booze'
"Umlaut" is, as I demonstrated in §§ 2 to 4, the name of exactly one pattern of vowel alternation. In contrast, 'l'\.blaut" is a cover term for at least 20 different such patterns, all found within the narrow domain of verb conjugation. Furthermore, all types of bidirectional relations between vowels exist; the relations are therefore not statable as simple linking or delinking rules. See, e.g., the change from i to o in bieten bot, vs. the change from o to i in stoBen - stieB. In other words, while Umlaut can be stated straightforwardly as a phonological rule such as (11), the same is impossible to
One such type of conversion, V-to-N conversion, is illustrated in (24). Here, the derivation of nouns from verbs has the following properties: it is non-productive, it derives nouns with an idiosyncratic meaning, and the gender of the noun is non-predictable. All three genders of German are found (although masculine gender is dominant). From these properties, we may conclude that this type of conversion is a rule belonging to level 1 of the lexicon. (24) V-to-N conversion
kochy > (der) Koch, schauy > (die) Schau, graby > (das) Grab The reverse type of conversion, from nouns to verbs, exists as well, but has different properties. It is quite productive, it is possible with compounds and affix-derived words, but not with inflected words. A few examples are given in (25) for simplex, compound and suffixed words. An assignment of this conversion to level 2 of the lexicon would be one way of accounting for these properties. 17
achieve for Ablaut. At best, a moderately large number of rules would be needed. This is not to say that there are no generalizations to be made about the ablauting · forms. For example, it is almost always the case that Ablaut affects only the rhymes of monosyllabic verbs. Neither onset consonants nor the forms of disyllabic verbs are
16: Wunderlich & Fabri (1993) observe additional generalizations for the ablauting verbs.
17: For similar types of conversion in English, this differential level assignment was already proposed by Kiparsky (1982).
109
108 tenseness of the vowels, which are ignored here. (For this reason, the vowel patterns are stated in the table by orthographic symbols only.) (23) Ablaut patterns for strong verbs: vowel pattern Infinitive simple past fiel fallen aia tragen trug au a lesen las eae stehlen stahl eao heben hob eoo hangen hing eia finden fand iau schwimmen schwamm iao bitten ia e bat i O0 bieten bot oio stoBen stieB kommen kam oao erloschen erlosch 000 schworen schwur DUO rief uiu rufen tun tat uaa log li.igen i.ioo gleiten glitt ai ii laufen lief au i au saufen soff au o o
past participle gefallen getragen gelesen gestohlen gehoben gehangen gefunden geschwommen gebeten geboten gestoBen gekommen erloschen geschworen gerufen getan gelogen geglitten gelaufen gesoffen
'fall' ,.carry" 'read'
'steal' 'lift' 'hang' 'find'
subject to the alternation. 16 But such observations still have a status different from those to be made about Umlaut. As Bybee notes on the irregular verbs of English, but as can be said with equal justification for the strong verbs of German, possible generalizations are those "formed over the shape of the derived past tense form, rather than as a relation between the basic, present form and the past form" (Bybee 1993: 2). There are other ways of demonstrating that Umlaut and Ablaut are not integrated into the derivation of words in the same way. I will discuss here the relation of these two phenomena to conversion (i.e., zero-derivation) of word class. There are several types of conversion in German, which are distinguishable in terms of their productivity and their semantic effects, and not just by their input and output categories.
'swim' 'ask'
'offer' 'push' 'come,. 'go out' , swear' 'call' 'do'
..lie"
'slide' , , run 'booze'
"Umlaut" is, as I demonstrated in §§ 2 to 4, the name of exactly one pattern of vowel alternation. In contrast, 'l ts / _
(18)
1:
{ V ,#
b.
predikat-fef atlet-fek parachut-fst grot-esk kwart-et moment-eel profet-es accurat -es se
senat-or potent-aat
'senator' 'potentate'
(20)
homeo-paat psycho-paat neuro-paat
'homeopath' 'psychopath' 'neuropath'
10
let us now consider the
homeopaat-fe psychopaat-fe neuropaat-fe
At first blush, these pairs seem to belong to (14a), cf. acrobaat/acrobaa[ts]-(e.(20) are phonologically odd, however, in that the failure of Affrication comes as a complete surprise. Is -paat- an uninteresting exception, requiring no further statement than that, or is there more to it? We think the latter is the case, and that the 'invisible person' plays a crucial part in this. Towards this, consider data such as those immediately below. (21)
apologfe energfe
apologet-isch energet-isch
theorfe patos
theoret-isch patet-isch
}
To the left-hand side of t, there is always a 'sequence' of sonorants: either a long vowel or a vowel-sonorant combination. The rule does not apply in numerous words in which /-i:-/ is followed by a consonant in the same suffix (cf.19a); nor is there a change before numerous other suffixes, even if they have front vowels (cf.19b); let alone back ones (cf.19c). (19) a.
'frontal' 'debutant'
Given this introduction to the Affrication phenomenon, following data, involving the stem -paat-:
There are two relatively marginal '/-i:/#'-suffixes the rule does not accept as a context: plural -i, and the nominalizing person-suffix -ie, cf. (17). (17)
front-aal debut-ant
C.
'predicative' 'athletics' 'parachutist' 'grand' 'quartet' 'currently'
despot-fame bismut-fne fanat-icus experiment-eel statut-air advocaa:t-uur
'prophetess' 'accuracy'
compliment-ens preparat-eur
'despotism' 'bismuthine' 'fanaticist' 'experimental' 'statutory' 'the Bar' 'complimentary' 'taxidermist'
These examples show that there are pairs of (i) nouns ending in stressed -(e and (ii) adjectives suffixed with -isch, which have the following coinciding properties: (a) the nouns are underived (at least there is no reason to assume otherwise, stems without -ie do not exist for them), and (b) in the adjective -et- 'replaces' -(e. What is strikingly missing, however, is a *pat(e noun: we find pdtos instead. The other way around, (21) lists the three (apparently idiosyncratic) morphological options available to nouns in -os: (22)
a.
10
k6smos thermos(-fles)
- k6sm-isch - therm-isch
ethos
- eth-isch
Note that the Dutch version of Velar Softening (k->s) applies 'before non-round front vowels': - impli[s]-fet fabrfek - fabri[s]-eer implik-atie - introdu[s]-e publfek - publi[s]-fst introduk-tie - medi[s]-fjn katolfek - katoli[s]-isme medik-us opaak fannak-ologfe - fanna[s]-fe - opa[s]-iteit 'implication' 'implicit' 'factory' 'to produce' 'introduction' 'guest' 'public' 'publicist' 'medicine' 'catholic' 'catholicism' 'doctor' 'phannacy' 'opaque' 'opacity' 'phannacology'
130
1 31
b. C.
Minos eros
- Min6-isch - er6t-isch
chaos
- cha6t-isch
Nouns in -os lose this ending before -isch, lose only final -s, or undergo an 's->t' rule (see fn.9). 11 Thus, *pdt-isch, *pat6-isch or *pat6t-isch should have been found, where patet-isch is. These observations have the following two implications. First, let as assume that the missing stem *pat(e is blocked by the idiosyncratic existence of pdtos, but is available in word-formation: what so far looked like homeo-paat-ie actually is homeo-patie. This immediately allows us to explain the failure of Affrication, since no triggering suffix is present. Second, the fact that kosm-isch 'cosmic' and therm-isch 'thermic' in (22a) do not have t in the derived form, shows that the m->t-rule must be limited to the m of the suffix -isme. The Dutch version now is (recall (11) with regard to the triggering suffixes): (23)
ism -> ist /
{ -isch, -icus, -N }
A similar comment with respect to the suffixal focus of the rule holds for English, where cosmic and thermic occur, too, as does cataclysm(-ic). 12 Aronoff (1976:121) apparently also intended his original rule (3) to apply to suffixal -ism. This can be indirectly deduced from the fact that he cites as a counterexample English embol-ismic, not bothering to discuss the much more obvious cosmic. As cases in which the rule "might also be at work", he cites sarcasm/sarcastic, and a handful of similar others. Interpreting the Dutch picture in this area is not helped by the fact that these are almost all dictionary-words. The following is an overview: (24) a.
chiasme chiliasme
'chiasmus' 'chiliasm'
chiast-isch chiliast
11
There is a considerable group of nouns able of losing its final -Vs rhyme when suffixed; examples are f(scus!fisc-aa/ ('(the) exchecquer'/'fiscal'), Arist6teles/Aristoti!l-isch, and Lazarus/Lazar-fst. This process will play an important role in section 3.8. Some further · examples of simple loss of -s are the following: Jezus - Jezu-fet commissaris - commissari-aat salaris - salari-eer secretaris - secretari-aat 'Jesus' 'Jesuit' 'commissioner' 'commissionership' 'salary' 'to pay' 'secretary' 'secretariat'. 12
The Dutch equivalent is cataclysme, but a derived adjective of this form is somehow hard to imagine.
b.
enthousiasme iconoclasme miasme orgasme orgiasme pleonasme sarcasme spasme (cinema) (fantasie) (gymnast- iek) (gymnasi-um) (pederast-ie) (scholast-iek)
'enthusiasm' 'iconoclasm'
enthousiastN iconoclast
!miasma'
'orgasm' 'frenzy' 'pleonasm' 'sarcasm' 'spasm' 'cinema' 'phantasy' 'gymnastics' 'gymnasium' 'pederasty' 'scholastics'
orgiast sarcast cineast fantast gymnast gymnasiast pederast scholast
enthousiastA iconoclast-isch miasm-atisch orgast-isch orgiast-isch pleonast-isch sarcast-isch spast-isch fantast-isch gymnast-isch (gymnasi-aal) pederast-isch scholast-isch
Although more irregular (see enthousiast,. for instance) and synchronically completely unproductive, this picture strongly resembles that found for -isme/-ist/istisch. It may well be therefore that, after Aronoff, rule (23) applies as well to -asme, producing forms such as e.g. chiliast by derivations involving the 'invisible person' suffix. So we can explain the unexpected lack of Affrication in (20) by an analysis that seems reasonably motivated. But what is our analysis of the former base-words homeopdat, psycho-paat and neuro-paat if the stem ends in -ie? This is where we have to introduce one more independently motivated rule of Dutch phonology: Prevocalic Vowel Deletion - recall naz(i)-isme in (lla), and also consider data such as those in (25): (25) Prevocalic Vowel Deletion 13 propagand(a) -ist b6eddh(a) -fst parod(ie) -fst chfn(a) -ees agend(a) -eer ch6ler(a) -isch chfl(i) -een
'propagandist' 'buddhist' 'parodist' 'chinese' 'to memorate' 'choleric' 'chilean'
Tr6tsk(y) -fst s61(o) -fst Esperant(o) -fst p6r( ie) -eus rabb(i) -fjn anim(o) -eer mar6kk(o) -aan
'Trotskyite' 'soloist' 'Esperantist' 'porous' 'rabbi' 'to stimulate' 'moroccan'
The multiple stresses indicate the main stress of the stem (leftmost) and of the derived word (rightmost). The only twist occurs when main stress is (always) immediately before the adjectival suffix -isch. Vowel deletion is indicated by parentheses. 13
132
133
din(e) -eer pur(e) -eer therap(fe) -eut
'to dine' 'to mash potatoes' 'therapist'
euf6r(fe) -isch chem(fe) -isch fricand(6) -el
'euphoric' 'chemical' 'fricandel'
Notice that there is little reason to assume that the deletee is necessarily a suffix. Over and above this, stress does not seem to play a crucial role in the sense that both stressed and unstressed vowels disappear. The latter situation is only expected, if Trommelen & Zonneveld (1991) and Zonneveld (1992) are right in claiming that, immediately upon (cyclic, Romance) affixation, Halle & Vergnaud's (1987) universal Stress Erasure Convention erases the stress pattern of the base. If this is so, the original stress-pattern of the base cannot be relevant to vowel-deletion. 14 Given this well-motivated phonological rule, our -N suffix immediately pays of. See
requires final dental obstruents in the stem, whereas the stressed suffix strongly prefers non-dentals. It is not our intention here to back up these observations with a full-fledged analysis of these suffixes' behaviour. Instead, let us consider the following sub-part of this last observation: with respect to stems ending in the dental obstruent -t, those to which the stressed suffix exceptionally attaches all denote persons: acrobaat, aristocraat, democraat, etc. in (14a), and forms with -paat- in (20). Clearly, this cries out for an explanation, and one that would preferably remove these stems from the stressed -fe input. This is precisely what we have done for the -paat- words, and we will now
continue to do for the others. 3.2. Aristocraat (vs. aristocra[ts]ie)
the derivation in (26): homeo -
(26) person-suffix -N stress erasure prevoc. vowel de!. stress assignment
patfe
(psycho-, neuro-) -N
ie
0 paat
In the entire derivation, there is no representation to which Affrication can apply, either because there is no suffixal -ie, or the appropriate suffix (-N) is not a trigger. Of course, psycho-pdat and all other forms in the same stem are derived in precisely the same way. Notice how the ostensible surface situation is the reverse of the underlying one: homeopdat is morphologically more complex than homeopatfe. As Aronoff stated: word formation is a much more abstract matter than one might at first be led to believe. One final observation provides both an interesting piece of independent evidence for our analysis, and a link to the next section. In (14a/b) above we listed, as an introduction to our discussion, examples of the two -ie suffixes of Dutch, one stressed the other unstressed. These two suffixes appear to have a number of properties with respect to their input, among which is at least the following: the unstressed suffix
Reconsider the human-based ...t-fe forms of (14), which we have found exceptional in that this suffix strongly prefers not to be added to a dental obstruent; and notice the situation we are in: we find ourselves in the bind that we do not want -fe to attach to a t-final base, but at the same time we must trigger Affrication, so there must be a stage where -t is immediately before the suffix. This can be accomplished rather easily in our 'invisible person' framework, at no extra cost. Recall from (14a) that -fe attaches to person-nouns such as tirdn(-fe), and so on. The assumption we will make is that the tfinal forms actually have the -N suffix at that stage of the derivation where -fe is added. E.g, aristocraat /aristocraat-N/ will be assumed to have a 'root' never surfacing in isolation followed by the 'invisible -N'. Such roots can also be found in examples such as partn-er, Bosn-ie 'Bosnia', stagn-eer 'to stagnate', frustr-eer 'to frustrate', and a host of other examples. The derivation of the abstract noun is then as follows: (27)
aristocra[ts]fe:
[aristocraat-N] -ie
ieN-suffix
prevoc. vowel deletion affrication stress assignment
0 ts -fe
And similarly for the remaining t-forms of (14a). Prevocalic Schwa deletion (2) and Prevocalic Vowel Deletion (17) might well be called the same rule, if not for the fact that they differ in their 'range': schwa is simply always deleted, whereas full vowels can be retained: individu -eel Mao-fsme algebra -isch h6bby -fsme trochee -isch ego -fst hfndoe -fsme discussie -eer 'individual' 'Maoism' 'algebraic' 'dilettantism' 'trochaic' 'egotist' 'hinduism' 'to discuss' 14
3.3. Psychoot (vs. psychose) At two points in the above discussion we have referred to a rule performing the operation of 's-to-t' in some environment. The examples encountered so far are the ones repeated below:
135
134
(28)
Laos chaos eros
Laot-iaan (cf. (16)) chaot-isch (cf. (22c)) erot-isch (id.)
A brief look at the relevant data shows that words of the following types, combined with the suffixes mentioned, are subject to this rule: (29)
a.
b.
extase [-z@] catechese synthese hypothese parenthese enclise analyse
extat-isch catechet-iek synthet-isch hypothet-isch parenthet-isch enclit-isch analyt-isch analyt-icus
propedeuse neurose narcose afasie [-zf:] poezie
skepsis [-sis]
skept-isch skept-icus deict-isch ellipt-isch syntact-isch syntact-icus
epilepsie [-sf:]
deixis ellipsis syntaxis
'ecstatic' 'catechetics' 'synthetic' 'hypothetic' 'parenthetic' 'enclitic' 'analytic' 'analyst' 'sceptic' 'sceptic (N)' 'deictic' 'elliptic' 'syntactic' 'syn tactician'
krisis [-zis]
eros [-s] apocalyps romance [-s@] Florence
propedeut-isch neurot-isch narcot-ica afat-isch poet-isch poet-ica krit-isch krit-icus epilept-isch epilept-icus erot-iek apocalypt-isch romant-isch Florent-ijn 'introductory' 'neurotic' 'narcotics' 'aphatic' 'poetic' 'poetics' 'critical'
These data show that the rule applies to z (29a) and s (29b) alike. With respect to words in -is, recall from (22) that -Vs can be lost upon suffixation.15 Clearly, there is a striking resemblance with the ism->ist rule in (23) in two ways: the structural change is the same, and there is a considerable overlap between the triggering suffixes, although (29) contains some additional ones. Although we have no positive evidence that these suffixes can be readily added to -isme, it seems appropriate to propose that the s-to-t rule is not a completely new one, but rather that (23) be expanded to cover the data above. This '(expanded) rule (23)' differs considerably from Affrication. In fact, the suffixes they take as triggers are mostly different ones: the two 'bare' -fe suffixes are not among the [t]-triggers, as shown by, e.g., precfs-ie in (14b); the forms orgasme 'orgasm' / anorgasm-(e 'frigidity' show the same, for stressed -(e. Conversely, there are only a small handful of forms to which both rules apply: (30)
Laos stimulans Laurens tendens
-
(cf. (28))
'Laotian' 'stimulants' 'Laurentia' 'tendentious'
Lao[ts]-iaan stimulan[ts]-ia Lauren[ts]-ia tenden[ts]-ieus
The suffixes in question will also appear as the triggering context of rule (23). The rule now is, informally: (31)
{ism -> z/s ->
~st} I
-isch -icus -N
(23)
-iek -ica -ijn (29)
-iaan -ia -ieus (30)
'critic'
But now observe the following: if -N is part of the environment of the m->t-rule, what about the newly added z/s->t branch of the rule? If we take our proposals seriously, we predict the existence of human nouns such as those in (3 la), which bear a hitherto completely opaque relation to their nominal and adjectival kin. The proposed derivations are those in (32b ):
'epileptic' 'epileptic (N)'
(32)
a.
'erotics'
'apocalyptic' 'romantic' 'Florentine'
neur6ot psych6ot asceet exegeet
'neurotic (N)' 'psychopath' 'ascete' 'exegete'
neur6sei, psych6seN ascesei, exegeseN
neur6t-ischA psych6t-ischA asceet-ischA exeget-ischA
There is a brief and inconclusive discussion of the English version of this process in Chomsky & Halle (1968:232). 15
136
137
3.4. Chaoot (vs. chaos) b. person suffix stress erasure prevoc. schwa deletion s -> t stress ass.
neur6se psych6se -N ose
ascese exegese
An implicit part of some of the cases of suffixation we have been discussing so far, is 'presuffixal vowel lengthening'. Examples from the text above are those in (35a) below, and (35b) adds some other cases:
-N ese
0
0
oot 6ot
eet eet
(35)
Based on the abstract noun (ending in -Vz@), we derive the person noun by independently motivated means, given our analysis. The adjective can then be derived from either noun (or both). Instead of prevocalic schwa deletion, derivations of this type may also involve Prevocalic Vowel Deletion (as earlier encountered in the derivation of homeo-paat in (26)). In this way, we derive person-nouns such as those below: (33)
poe[z]fe person suffix stress erasure prevoc. vowel deletion s -> t stress ass.
hypocrisfe -N
ie
0
0
novies noviet
'novice' 'freshman'
La[o:]t-iaan cha[6:]t-isch er[6:]t-isch er[o:]t-fek
b.
satan demon alcohol babylon eter alfabet israel
sat[a:]n-isch dem[6:]n-isch alcoh[6:]l-isch babyl[6:]n-isch et[e:]r-isch alfab[e:]t-isch isra[e:]1-isch
The details of this process, including a precise listing of suffixes that trigger it, does not concern us here. The more important question is, whether it is involved in -N derivations. The answer is yes; consider (36).
- novi[s]-iaat 'noviciate' - novi[ts]-iaat 'freshmanship'
chaos
(36) person suffix stress erasure presuffixal lengthening s -> t stress assignment
iet fet
This gives the nouns poeet 'poet', and hypocr(et 'hypocrite'. Again, there is no extra charge in the derivation of these nouns. Finally, consider the small but interesting group in (34). (34)
Laos chaos eros
~N
ie eet eet
a.
-N OS OOS
oot 6ot
This is the noun cha6ot, ie. 'chaotic person'. Again, the hitherto opaque derivation makes maximal use of the devices already available in Dutch morphophonology: each step is independently necessary in the grammar.
3.5. Analfabeet (vs. alfabet) By simply adding a suffix (as indicated), we may derive one noun from another in the top row pair. 'Freshman' is derived from 'novice' by adding -N and applying the s->t rule. Finally, another noun is derived from that output by suffixation, of the type automatically triggering Affrication. We know of no earlier analysis that so mechanically and satisfyingly lays bare the relation between these forms.
An independent theoretical point is illustrated by the following data: (37)
portier
(i) 'door' (ii) 'gate-keeper'
(def.art.: het portier) (def.art.: de portier)
As argued in Trommelen & Zonneveld (1986), Dutch is, after Allen (1979) and Williams (1981), a Righthand Head Rule language. This is so not only for major lexical category, but also for its gender system: suffixes are associated with one specific
138
139
gender, whereas prefixes are not. If we take as our example the choice between the iw• definite articles de ('common') and het ('neutral'), determined by gender, the followi · is typical of what we find: (38) a.
raar]A normeer]v puur]A probeerlv de magneet]N de anemoon]N b. de missie]N de aktie]N de harmonie]N het conto]N het conformisme]N het kolonialisme ]N
-
de raar-iteit]N de normeer-ing]N het puur-isme ]N het probeer-sel]N het magneet-isme]N het anemoon-tje]N de trans-missie]N de re-aktie ]N de dis-harmonie]N het dis-con to]N het de-conformisme lN het neo-kolonialisme ]N
'strangeness' 'standardization' 'purism' 'attempt' 'magnetism' 'little anemone' 'transmission'
anoniem]A de senaat]N het schip]N het dorp]N het advies JN b. speciaal]A het kapitaal]N het loket]N C. de man]N de vrouw]N de koning]N de sopraan ]N de acrobaat]N
-
de anoniem-us]N de senaat-or]N de schipp-er]N de dorp-eling]N de advies-eur lN de speciaal-ist]N de kapitaal-ist]N de loket-ist]N het mann-etje]/' het vrouw-tje]N het konink-je]N het sopraan-tje]N het acrobaat-jelN
(het) alfabet
(40)
'alphabet'
'reaction'
'disharmony' 'discount' 'deconformism' 'neocolonialism'
Relevant in this context is the behaviour of 'person' suffixes. As (39) below they take the gender associated with de. But the generalization is not a general sema one on 'human words', because the diminutive suffix takes het, even if the input output) is human: (39) a.
The examples in (39c) can be multiplied by adding the diminutives of the human words in the righthand columns of (39a-b). Given this, observe the following interesting observation: in its non-human sense, portier is a neuter het-word, whereas in its human sense it is a common de-word. In our framework, there exists an automatic explanation: the -N suffix is part of the morphological make-up of the (ii)-option in (37). 17 This same situation can be complicated with one additional step. Consider the data of (40).
'anonymous (person)' 'senator'
'skipper' 'villager' 'advisor' 'specialist' 'capitalist' 'ticket clerk' 'little man' 'little woman' 'little king' 'little soprano' 'little acrobat'
(1983).
For a treatment of the interesting allomorphy of the diminutive suffix, -cf. Tromlfi
alfab[e:]t-isch (de) an-alfab[ e:Jt 'alphabetical' 'illiterate (person)'
The relevant phenomenon here is vowel lengthening. All instances of it can be attributed to the presuffixal position of the relevant vowels, except one: an-alfabeet. But notice again, that this is the only 'person' -denoting word in the entire group. The suffix -N, again, triggering lengthening, explains why the situation is as observed. But this suffix also performs a second task. Het alfabet shows that this noun has neutral gender. But in de analfabeet 'the illiterate', gender is common. This cannot be due to the negation prefix an-, which does not determine gender (cf. (38b)).1' To say that it is due to vowel lengthening would be very peculiar indeed, and unparallelled in Dutch morphophonology. The source for the article de must therefore be a suffix, and that suffix, we propose, is -N.
17
There is independent evidence that the correct view of this phenomenon is morphological rather than semantic. In some cases similar in all respects to (36), neutral gender is maintained (which suggests that the word simply has 'two meanings'): model (i) 'model' (het) (ii) '(artist's) model' (het) talent (i) 'talent' (het) (ii) 'talented person' (het) We would also expect there to be common de-words, which simply cannot be classified in either group; some cases in point are: bas (i) '(double) bass (instrument)' (de) (ii) 'bass' (singer)' (de) hulp (i) 'help, assistance' (de) (ii) 'help, assistant' (de) 18
16
(het) alfab[e:]t-isme (het) an-alfab[e:]t-fsme (cf. (37a)) 'literacy' 'illiteracy'
Alfabeet, in the sense of 'literate person', does not exist; instead, ge-letter-d-e is used. If one of the Dutch genders is 'unmarked', it is no doubt the 'common' one; hence all nominal input to -N derivation discussed so far, was 'common' to begin with. If pressed to give one's intuition, however, it would no doubt be de alfabeet.
141
140
vowels precede other final voiced obstruents (d, v, z, and so on). As is well known, these are abundantly available (cf. Zonneveld 1978, op.cit.). Third and strikingly, both exceptional forms denote persons, an observation that obtains significance in the present context. Thus, Zwaab can easily be releaved of its offending consonant by adding to it -N, the 'invisible person'; and the derivation of homof6ob, and other similar forms in this paradigm, has precisely the steps of (23), with fob(e replacing pat(e. Hence, although the raison d'etre of the phonotactic constraint remains as yet unclear, a group of otherwise awkward forms receives an explanation.
3.6. Homofoob (vs. homofobie) Many aspects of the -pat(e paradigm of (18a/b) reoccur in that of fobfe Notice that fobfe exists, where *patfe did not (20/21). (41) a. b.
fobfe f6b-isch
C.
homo-fobfe, agora-fobfe, claustro-fobfe, xeno-fobfe homof6b-isch, etc. homof6ob, etc.
Zonneveld (1978:49) discussed the (41c)-form homof6ob, observing that it violates. otherwise valid phonotactic constraint that in Dutch final -b cannot be preceded h Jong vowel (nor, similarly, by a diphthong or a VC-sequence). Consider data sue those in (42): heb (42) - krab schrob rob kabbel - gabber knibbel kibbel kabel - fabel kaliber wiebel bijbel - heibel september november - *kraab, *kreib, *kramb 'to have' 'to scrub' 'to ripple' 'mate' 'to haggle' 'to bargain' 'cable' 'fable' 'to wiggle' 'calibre' 'bible' 'racket' 'September' 'November' 'crab' 'seal'
web snob sabbel dobbel sabel nobel spijbel december
krib tob lebber tobbe syllabe troebel amber Humber
rib schub bibber rubber meubel puber somber Elbe
bob
'web'
'crib'
'rib'
'to bob'
'snob' 'to suck' 'to dice' 'sabre' 'noble' 'to play truant' 'December'
'to worry' 'to sip' 'tub' 'syllable' 'turbid' 'ambre' 'Humber'
'plate' 'to shiver' 'rubber' 'furniture' 'puber' 'sombre' 'tomb' '(river) Elbe'
dribbel
. 3.7. PakisUllim (vs. Pakistan) In Dutch, as in many other languages, the morphology of geographical affixation is a '•>complicated issue. To give an example from this area: sometimes the country or region appears to be derived from the inhabitant (43a), at other times the relation is the reverse of this (43b); in a third case, two morphemes seem to replace one another (43c); in a fourth, there is no country to the 'inhabitant' (43d); and in a fifth, no inhabitant to the country (43e ).
amoebe
Roemeen Pool Fries
tombe
b.
As a similarly unexpected case, Zonneveld (1978) gives Zwaab 'Swabian person'. one reaction to this picture would be to say that these two forms are appal'., precisely those that fill the gap in (42). That this is the correct view, howev unlikely for various reasons. First, the existence of just two forms as against the more numerous ones of (42) would be quite unimpressive. Second, not only do we· to compare final -b after long vowels with (42), but also with forms· in which;·
C.
d.
e.
Roemen-ie Pol-en Fries-land Vietnam Ital-ie Roi-land Duits-land Frank-rijk Arab-ie
Kelt Eskimo Apache Sovjet Unie Aboe Dhabi
'Rumania' 'Poland' 'Frisia' 'Vietnam' 'Italy' 'Holland' 'Germany' 'France' 'Arabia' 'Kelt' 'Eskimo' 'Apache' 'Soviet Union' 'Abou Dhabi'
Vietnam-ees Ital-i-aan Ho!-land-er Duits-er Frans-man Arab-ier
The default option always seems to be to add the suffix -er to the country, but even that is hard to execute in the bottom two cases; often, native speakers fall back, e.g., on 'Sovjet Rus' in the first case. Of current special interest, however, are those examples involving countries or regions ending in -((i)st)an. The picture is that below:
142
143
(44) a.
b.
Afghaan Koerd Oesbeek Turkmeen (Hindoe)
Although few things come as a surprise in this area, to say that haphazard vowel lengthening is also among the ways of inhibitant formation stretches the imagination and the power of Dutch grammar beyond the limit. But there is a way out, in our framework, precisely because we are dealing with 'persons': we know from previous sections that -N has the capacity of being a vowel-lengthener. This precisely is what is needed here, and again a set of recalcitrant forms receives an explanation. A small handful of further examples may be discussed here: (45)
Twent Drent Toscaan
Twent-e Drent-e Toscan-e
'Tubantia' 'Drente' 'Tuscany'
3.8. Catechismus (vs. catechist) In (20) and footnote 11 we discussed an allomorphic rule deleting the ending -Vs when suffixes are added. Some further examples of this rule are those below: basis Atlantis cftrus ffscus anus
bas-aal Atlant-isch citr-6en fisc-aal an-aal
dactylus Arist6teles Euclides Hercules Texas
dactyl-isch Aristotel-isch Euclfd-isch Hercul-isch Tex-aan
Christ-en Homer-isch abort-eer
Honduras Hellas messfas
'basis' 'Atlantic' 'lemon' 'fiscal' 'anal' 'Christian' 'Homeric' 'to abort'
Hondur-ees Hell-een messi-aans 'dactylic' 'Aristotelean' 'Euclidean' 'Herculean' 'Texan' 'Hondurese' 'Hellene' 'messianic'
In our framework, we expect there to be examples in which the person suffix triggers the allomorphy. The predicted phonetic effect is simply loss of the ending. Such cases are the pairs in (47); notice that our final example involves both -Vs deletion and the m->t rule, operating in the context of the -N suffix.
(47)
These three are an exhaustive list, as far as we are aware. Superficially, the 'region' seems to be morphologically more complex than the inhabitant, but in our analysis they receive the same morphological treatment as those of (44b): Twente-N simply becomes Twent by Prevocalic Schwa Deletion. An earlier example of such a derivation is neuroot from neurose in (32b). As a bonus, in fact, the unproductive topographical suffix -e can be eliminated from the grammar. This is advantageous in the sense that otherwise morphologically complex Toscan-e, as a foreign word, would violate the condition that 'foreign' suffixation involves full vowels (rather than schwa, cf. section 2.3.).
(46)
Christus Homerus ab6rtus
Afghan-istan K6erd-istan Oesbek-istan Turkmen-istan Hfndoe-stan Hindoe-staan Pak-istan Pak-istaan Teheran Teheraan
feutus Christus catechfsmus
'foetus' 'Christ' 'catechism'
- feut - anti-Christ - catechfst
'freshman' 'anti-Christ' 'catechist'
4. Word-stress and the V-suffix As pointed out above, accentual phenomena are lacking from the empirical evidence serving as the basis of our zero suffix analysis. On the other hand, since our account does have potential consequences for stress assignment (by assuming additional righthand prosodic material in words), we will now show that the -N suffix is perfectly compatible with the (word-)stress system of Dutch. Although accounts are readily available (see, e.g., Trommelen & Zonneveld 1989, Kager 1989, Zonneveld 1993), we do not need a full-fledged analysis of Dutch word-stress to make our point. Rather, let us consider the cases in which we assumed -N, and see what their stress properties are, both without and with the additional suffix. Diagram (48) contains an overview.
144
145
(48)
without suffix (trad. analysis)
with suffix (underlyingly)
after phonology in-/output of stress
(5)
conform-fat homeopaat aristocraat neur6ot poeet noviet cha6ot poruer analfabeet homof6ob pakistaan toscaan catech{smus
conform-isme-N homeop[a:]tie-N aristocr[a:]t-N neur[o:]se-N po[e:]sie-N nov[i:]s-N chaos-N port[i:]r-N analfabet-N homo[fo:]bie-N pakistan-N tosc[a:]ne-N catechismus-N
conform-ist-N
(26)
(27) (32)
(33) (34) (36) (37) (40)
(41) (44)
(45)
(47)
homeop[a:]t-N aristocr[a:]t-N neur[6:]t-N po[e:]t-N nov[{:]t-N cha[6:]t-N port[{:]r-N analfab[e:]t-N homof[6:]b-N pakist[a:]n-N tosc[a:]n-N catechist-N
In the leftmost column are the surface output forms of the person nouns discussed in this contribution; they are also the underlying forms assumed in traditional analyses (if discussed there at all). The middlemost column contains the underlying forms assumed here; in many cases we use underlyingly as a 'stem' an abstract noun that is traditionally usually considered derived; to such a stem we add -N as indicated. The rightmost column contains the intermediate representations derived after various forms of phonological and allomorphic adjustment: schwa deletion, prevocalic vowel deletion, s->t, and so on. It is important to reemphasize that all of these adjustments are independently motivated, i.e., they are not specifically required for the proposal in hand. Assuming that these adjustments precede stress assignment, we derive an input to stress that is essentially the required surface output, followed by -N. The output contains the main stresses indicated in this column. In order to demonstrate that the empty suffix is compatible with Dutch word-stress assignment, it is sufficient to point out the following two major generalizations: (49)
(i) if the final syllable is open, main word-stress is penultimate; (ii) if the final syllable is 'superheavy', i.e. contains a '-VXC' rhyme, it receives main stress.
All words in the leftmost column go by rule (ii), which is why by and large traditional analyses can take such forms as underlying ones. But now consider the rightmost column, i.e., the list of intermediate representations in our analysis. Notice that the
dashes do not indicate syllable structure, but rather morphological make-up. If we project syllable structure onto them, which is a mental task quite easily executed, we see that they fall within the domain of rule (i). This is precisely the way we want things to be: penultimate stress in the presence of the empty suffix.
5. The prosodic notion of Catalexis Up until this point, our exposition has consisted of an introduction, an English point of departure, the elaborate development of the idea that Dutch has a suffix suggestively called 'invisible person', and a demonstration that assuming such a suffix does not interfere with Dutch word-stress assignment. Part of our account is the idea that this suffix has no 'phonetico-phonological' content, but is a Nucleus, since it behaves as such in a considerable number of phonological processes, such as Prevocalic Vowel Deletion and others. We have not yet made the connection with the notion of Catalexis, and that is what will take place in this section. Consider the following. Catalexis is the prosodic notion introduced by Kiparsky (1991) in order to account for the situation in a phonological system that can be described intuitively as the reverse of hitherto more well-known extrametricality. If extrametricality means disregarding material (a consonant, a syllable, a suffix) that is actually present in the string at the relevant stage of a phonological derivation, catalexis is the addition of seemingly absent - material. In Kiparsky's words: (50)
The parallellism between extrametricality and catalexis is brought out by the following representation. Assume that extrametricality is the erasure of a metrical constituent, together with any constituents that dominate it and nothing else, at the edge of a prosodic domain. A melodic element which is extrametrical will then be invisible to all rules and constraints that refer to metrical structure; it will still be visible to rules and constraints defined on a purely melodic domain. Catalexis, conversely, is the addition of a metrical constituent at the edge of a prosodic domain, where it is adjoined to the superordinate metrical structure if permitted by the language's well-formedness constraints. (p.3) [emphasis by P.K.]
Kiparsky (1991) and Kager (1993) show that a reasonably articulated theory of catalexis makes a number of testable empirical predictions that are firmly supported by the facts
146
147
of an impressive range of languages, and has a number of clear theoretical advantages; Kiparsky again: (51)
In sum: enriching the theory with the parameter of catalexis allows degenerate feet to be eliminated cleanly, tightens the typology of stress-systems, and obviates the need for several unprincipled devices. (p.18).
In order to briefly see what is meant, consider a language such as Tiibatulabal, analysed in Hayes (1981:60) as an example of a stress-system employing a right-to-left iamb: (52)
/\ /\ I\
wswsws
I\
I\
w s w s
pi:ti'pi:tiYdinat
witanhatal
'he is turning it over (repeatedly)'
'the Tejon Indians'
ponihwi'n 'of his own skunk'
In this account, we have degenerate feet at the left-hand word-edge, resulting from the left-wardly iterative stress-rule. Suppose, however, one were to embark on an interesting program limiting the use of degenerate feet (as in Kager 1989, Hayes 1991). One way of doing this would be to say that Tiibatulabal has a non-iterative right-hand monosyllabic stress rule, and an iterative trochee; and by assuming that universal theory disallows degenerate feet in weak positions (in this case: left-hand word-edge ones). The next step then is (Kiparsky 1991:12-3) to say that degenerate feet are banned completely, and that Tiibatulabal has both a right-to-left trochee and right-hand catalexis. All evidence we have seen for catalexis so far involves accounts of stress-systems, which appear to benefit considerably from its introduction. An exception is a brief passage in Kiparsky (1991:4), who notes that in languages such as Malayalam, Pali, Yupik and others, statements such as "syllables must have onsets except in word-initial position" may be instantiations of catalexis (of a word-initial C) at the level of syllablestructure. Kager (1993:10) ventures into the realm of the accentual phonology/morphology interface, when he shows that in Toba-Batak stress-shift in certain word-classes can be · explained by assuming that they have a 'catalectic suffix'. The language has a twosyllable window of main stress at the righthand word-edge. This in itself forces no choice between an iamb with lexically marked extrametricality, or a trochee with lexically marked catalexis. One of the arguments selecting the latter option is the . relative frequency of penultimate stress as compared to final stress, reducing the number of lexical markings. Then, consider the following situation (Kager, op.cit.):
(53) The interest of Toba-Batak resides in certain morphemes which lack segmental realization but are associated with stress-shift. The passive may be marked by a stress-shift to the final syllable (cf. ldpu 'to smear', lapu 'be smeared'). In nominals, catalexis functions to form the vocative (cf. (nan 'mother', indn 'mother!'), and the honorific possessive (cf. tandmmu 'your hand (informal), tanammu 'your hand (honorific)'). In the theory of catalexis, these morphemes can be elegantly analysed as actually consisting of a catalectic syllable. So individual morphemes can take the shape of a catalectic suffix in order to capture prosodic phenomena such as stress shift in Toba-Batak. In our Dutch case, we are not dealing with stress shift, in fact we are not crucially dealing with stress at all. Still, it will be clear what our actual proposal is: if catalexis is a theoretical possibility, the facts of Dutch can be explained by assuming that our -N is a catalectic suffix. An important aspect of this proposal is that invoking catalexis as the theoretical background of our analysis, also and immediately brings along the much-wanted constraining factors. As pointed out, stress accounts could easily approach vacuity if zero-vowels could be added at will to phonological strings. Relevant in this respect is the so-called Peripherality Condition, that is assumed to constrain both the circumstances under which prosodic categories can be extrametrical (Hayes 1981), or catalectic (Kiparsky 1991): extrametrical or catalectic material may be assumed only at the edge of the pertinent stress-domain, in languages like Dutch and English the (phonological) word. Since we are dealing with a proposed suffix here, this condition is automatically adhered to. There is one specific situation left that is a source of potential worry: a stress-cycle where the (extrametrical or catalectic) material is internal rather than peripheral. The classic example of this occurs in English, which has syllable extrametricality (Hayes 1981), and derivations such as the following: pdr -> parent. In other words: the originally extrametrical (invisible) material becomes visible in non-peripheral position, and is there able to carry stress. Similarly, it seems reasonable to assume that catalectic material is automatically invisible when nonperipheral. A case in point is our derivation of aristocraat-fe from the suffixed person noun aristocraat-N in (27), repeated here for convenience. (54)
aristocra[ts]fe: ieN-suffix prevoc. vowel deletion affrication stress assignment
[aristocraat-N] -ie
0 ts -le
,, 148
In actual fact, it will not be the language-specific rule of Prevocalic Vowel Deletion, but the universal Peripherality Condition that omits -N from the string. Precisely
because of the existence of the rule, however, this is not a critical case: it is an illustration of 'peripherality', but does not provide crucial evidence for it. Although in the abstract the required situation of the latter sort is clear, we have not been able to find crucial examples in our account.
149
processes in a semi-jocular fashion, creating forms such as afaat 'aphasiac person', extaat 'ecstactic person' and spast 'spastic person', demonstrating their productivity. Rather than posit a new device, however, we submit that the phenomenon should be approached along the strategically successful lines, that is, catalexis contributes to the notion that backformation should be implied among the "unprincipled devices" of (48). We leave it for future research to establish whether this line of thinking is worth pursuing.
6. Conclusions and Morphological Perspective
In this article, we have dealt with a number of morphological data from Dutch. Initially, these centered around the suffixes -isch, -ist, and -isme. Taldng as our point of departure Aronoff's analysis of the relation between English -ism and -ist, we proposed that the distributional facts concerning the equivalents of these suffixes in Dutch can be explained only if -ist is derived from -isme. We argued that there is no extra phonological and morphological cost attached to this move, if a 'person' suffix is assumed to trigger the phonological change from m to t. This suffix was presumed to have the shape of an empty nucleus, hence its phonetic invisibility. We then provided a considerable amount of independent empirical evidence from a variety of phonological and morphological areas, to support the existence of such a phonetically invisible person. We proposed that the suffix in question is in fact catalectic, and showed that its presence is constrained by the Peripherality Condition, and is compatible with Dutch stress-assignment. We conclude that the catalectic person suffix -N is part of Dutch morphology. The wider perspective we speculatively presume our discussion may have is the following. There is a clear and successful strategy in current morphology to reduce all sorts of seemingly exceptional morphology to affixation. The most transparent examples of this strategy can be found in non-concatenative morphology such as that of the Semitic languages, and in accounts of reduplication. One need only compare early generative references such as, respectively, Levy & Fidelholtz (1971) and Wilbur (1973) with, e.g., Marantz (1982) and McCarthy & Prince (1990) to see what this entails. Don (1993) may be cited as a recent attempt to analyze morphological conversion among precisely the same lines. It seems clear that one way of looking at our account of Dutch is as a (synachronic) formulation of the phenomenon of back formation. In all cases of (48), the derived form ends up shorter (in some cases considerably shorter) than the underlying one, although a suffix is added. This is largely the result of 'allomorphic' deletion processes, but the principal of these, Prevocalic Schwa Deletion and Prevocalic Vowel Deletion, are natural rules that will recur in many languages. In Dutch one is able to, apply these
BIBLIOGRAPHY
Allen, M. 1979. Morphological Investigations. Ph.D. diss., The University of Connecticut. Aronoff, M. 1976. Word Formation in Generative Grammar. Cambridge, Mass.: MIT Press. Backhuys, K.-J. 1986. De morfologie van romaanse woordvorming in het Nederlands, M.A. Thesis, Vakgroep Nederlands, R.U.U. Backhuys, K.-J., M. Trommelen en W. Zonneveld. 1988. Vlamers en Belgiers. De status van de adjectivische suffixen -en, -s en -isch. Spektator 17:252-267. Chomsky, N., and M. Halle. 1968. The Sound Pattern of English. New York: Harper and Row. Don, J. 1993. Morphological Conversion. Ph.D. Thesis, Research Department for Language and Speech, Utrecht University. Halle, M., and J.-R. Vergnaud. 1987. An Essay on Stress. Cambridge, Mass.: MIT Press. Mass. Hayes, B. 1981. A Metrical Theory of Stress Rules. MIT. Ph.D. diss., distributed by Indiana University Linguistics Club. Kager, R. 1989. A Metrical Theory of Stress and Destressing in English and Dutch. Dordrecht: Foris. Kager, R. 1993. Consequences of Catalexis. Paper read at the HILP Phonology Meeting, Leyden, January 1993. Kiparsky, P. 1982. From Cyclic Phonology to Lexical Phonology. In H. v.d. Hulst and N. Smith (eds.), The Structure of Phonological Representations I, Dordrecht: Foris, 131176. Kiparsky, P. 1991. Catalexis. Unpubl. ms., Stanford University and Wissenschaftskolleg zu Berlin. Kooij, J.G. 1979. Morfologie en het Romaanse vokabularium in het Nederlands. In T. Hoekstra en H. v.d. Hulst (red.), Morfologie in Nederland, Glot Special, 23-41.
'
151
150
Kooij, J.G. 1980. Morphophonology and the Dutch Romance Vocabulary. In W. Zonneveld and F. Weerman (eds.), Linguistics in the Netherlands 1977-1979, Dordrecht:
Final Extraprosodicity and Resyllabification'
Foris, 350-358. Lessen Kloeke, W.U.S. van. 1981. Deutsche Phonologie und Morphologie, Merkmale
Tracy Alan Hall, F. A. S. Berlin
und Markiertheit. Ph.D. Thesis, Leyden University. Levy, M., and J.L. Fidelholtz. 1971 Arabic Broken Plurals, Rule Features, and Lexical
A number
Features. Glossa 5 :57-70. Marantz, A. 1982. Re Reduplication. Linguistic Inquiry 13:435-482. McCarthy, J.J., and A. Prince. 1990. Foot and Word in Prosodic Morphology. Natural
of linguists
have argued
that resyllabification,
defined here as the
reassociation of a consonant from the coda into the following onset, can be eliminated from the grammar by invoking Final Extraprosodicity (e.g. It [1>u], and Irish [fa] --> [E:] /_#. 1.2 Structure Preservation as a special case of the Strong Domain Hypothesis In contrast to Borowsky's position, Myers (1991) suggests that the domain over which Structure Preservation holds is not predictable, but is specified by language-particular grammars just like the Strong Domain Hypothesis (4), also put forward by Kiparsky (1984). (The version below is from Borowsky 1986.)
(4) The Strong Domain Hypothesis a. All rules are available at the earliest level of the phonology. b. Rules may cease to apply, but may not begin to apply, at a later level by stipulation. In fact, Myers suggests that Structure Preservation is simply a special case of the Strong Domain Hypothesis. According to this view, all marking conditions on segment inventories are available at the earliest lexical levels, and languages may only select the point in the grammar where such constraints tum off, just as the Strong Domain Hypothesis predicts that all phonological rules are in principle available to apply at the earliest lexical level, and grammars only mark where such rules cease to apply. Myers suggests that on the one hand, there are languages that allow lexical violations of Structure Preservation, such as Malayalam (Mohanan and Mohanan 1984), German (Hall 1989), and assorted Bantu languages (Harris 1987), and on the other hand, there are language in which Structure Preservation regulates postlexical rule application (e.g. Shona, Myers 1989, Hyman 1993). The difference between these languages is accounted for by Myers as follows: in languages like German, which we review below, structure-preserving constraints tum off at some lexical level, whereas in languages like Shona, analogous constraints persist postlexically. Thus in Myers' view, contrary to Borowsky, the domain of Structure Preservation is not predictable but can vary according to individual languages.
189
188
1.3 Linked structures may violate Structure Preservation
MacFarland and Pierrehumbert (1991) have put forward yet a third hypothesis concerning Structure Preservation violations. They suggest that rules which violate Structure Preservation are predictable on the basis of rule type, rather than the domain in which the rule applies, as Borowsky and Myers suggest. Specifically, they argue that assimilation rules --- rules which create linked structures -- may create novel segments even within the lexicon, in the same way that proposed constraints governing the operation of phonological rules to linked structures (e.g., the Linking Constraint, Hayes 1986) allow two-tiered representations (e.g.., long vowels and consonants) to violate phonological rules which refer to single tiers. Thus, according to MacFarland and Pierrehumbert, the only rules which should be subject to Structure Preservation are rules which do not create linked structure. In principle, it is possible that, say, both Borowsky' s and MacFarland and Pierrehumbert' s relaxations of Structure Preservation might be correct. In that case, we would expect to find level one assimilatory rules which are non-structure preserving and word level non-assimilatory rules which are non-structure preserving, · Certainly the latter condition is found. Borowsky' s examples of allophonic rules which must apply at the word level are at best randomly divided between assimilatory (producing multiply linked structures) and non-assimilatory types (not producing multiply linked structures). In some cases, there is not enough information to tell if a rule is assimilatory. (5)
Belfast Dental Assimilation Aitken's Law London vowels before l Northern Irish E: English re tensing Adelaide Vowel Rounding London o allophones English pre-l breaking S. American i allophones
assimilatory (if [r] is dental) non-assimilatory unclear non-assimilatory non-assimilatory unclear non-assimilatory non-assimilatory? non-assimilatory?
On the other hand, we have not found any clear-cut examples of assimilatory level 1 rules which evade Structure Preservation. 1 The examples of assimilatory rules that do not preserve structure are typically word level rules. _____ F_o_r _i_n_stan_ ce, at the oral presentation of this paper, Andrew Spencer 1 At the oral presentation of this paper, Abigail Cohn reminded us· of her analysis of Nasal Spread in Sundanese (see Cohn 1989 and references cited there). 1Jlis rule does indeed seem to be a lexical rule which creates novel nasalized segments in linked structures. Cohn shows that Nasal Spread both precedes and follows a morphological rule, so we conclude that it cannot be a word level rule in most current models, certainly not in Borowsky's (1993). We are uncertain how much weight to lay upon this one example. We would hope to find particular facts about this rule, the segment inventory, or the lexical level(s) in Sundanese that might render the violation of Structure Preservation explicable without invoking the Linking Constraint.
reminded us that Rubach (1984), in his analysis of Polish, allows the lexical pr?duction of a non-underlying palatalized r' via spreading of the feature [-back]. Is this an example of the sort that would support MacFarland and Pierrehumbert? Not necessarily. This allophonic segment is produced at the last lexical level -- the word level in Borowsky's model, the postcyclic level in Rubach's analysis. 1.4Summary
Various weakenings of Kiparsky's Principle have been recently proposed: Borowsky Myers MacFarland and Pierrehumbert
original
Structure Preservation
SP ceases to apply at word level where SP ceases to hold marked by individual languages SP does not hold of assimilation rules
In the following sections, we first examine McFarland and Pierrehumbert' s objection to an analysis of German Fricative Assimilation by Hall (1989). We conclude that the reason Fricative Assimilation is not structure preserving is because it is a word-level rule, not because it is an assimilatory rule. This will be our general approach to the question of systematic exceptions to Structure Preservation, in fact: level 1 assimilatory rules are indeed structure preserving, but word level rules are not, regardless of their assimilatory or non-assimilatory nature. The available evidence is most consistent with Borowsky' s hypothesis. 2
2. Structure Preservation and German Fricative Assimilation 2.1 The basic facts
McFarland and Pierrehumbert's 1991 article centers around an analysis by Hall (1989) concerning the distribution of the German fricatives [x] and [c;]. As is well known, [x] appears after back vowels, while [
eve
FILL
*-CC-
[miml]V[qacaj[n]
*
*
[mimVl][qaca][n]
*
*
[mimVl]
*
[miml]V
*
[mimVl]V
*! *
Candidates
MI
Archangeli, D. 1984. Underspecification in Yawelmani phonology and morphology. Unpublished PhD dissertation, Massachusetts Institute of Technology. [Publishedby Garland Press, 1988]
*!
Booij, G. and Rubach, J. 1987. Postcyclicversus postlexical rules in in lexical phonology. Linguisticlnquiry 18, 1-44. Gussmaun, E. 1980. Abstraction in phonology. Cambridge, MA: MIT Press.
=>
* ltoh, J. 1989. A prosodic theory of epenthesis. Natural Language and Linguistic Theory 7, 217-259.
*! *
Kenstowicz, M. 1979. Chukchee vowel harmony and epenthesis. CLS 15, 402-412. Kenstowicz, M. 1993. A constraint-based analysis of Chukchee syllabificationand stress. Talk presented to ROW-I, 23 October, 1993, Rutgers University.
=>
=>
[wejem]V[lq]V[n]
**
*
[wejem][lVq]V[n]
**
*
[pVkir]V[k]
**
[pVkirJ[k]V
**
*!
*
*
**
*!
[pkir]V[k] V[pkir]V[k]
*!
*!
Table 3: Tableaux for meml(iilqacan, we;em@lq@n, and p@kir(iil,k
The most harmonic candidate (i.e. the wellformed string) is indicated by the symbol=>. Each asterisk in a constraints colnmn represents a violation of the constraint. Where an asterisk bears an exclamation mark, this is a violation which excludes that particular form, because there is another form in the candidate set which either violates only lower ranking constraints, or which violates that constraint fewer times. Strictly speaking, for each fonn the candidate set evaluated by the constraints is infinitely large. For expositional reasons, l have only provided a sample of the theoretically possible candidates. This analysis of a fragment of Chukchee, then, illustrates some of the conceptual and empirical advantages of Optimality Theory over a derivational theory such as Lexical Phonology. It remains to be seen, however, to what extent this success can be maintained by further exposure of the theory to the vagaries of Chukchee phonology.
Kiparsky, P. 1985. Some consequences of Lexical Phonology. Phonology Yearbook 2, 83138. Kiparsky, P. 1993. Blocking in nonderived environments. In: S. Hargus & E. Kaisse (eds.) Contributions to Lexical Phonology. Phonology and Phonetics, vol. 4. Orlando: Academic Press, 277-314. Kodzasov, S. & Muravyova,I. 1980. Slog i ritrnika slova v aljutorskomjazyke. In: Aktual'nye voprosy strukturnoj i prikladnoj lingvistiki. Publikatsii otdelenija strukturnoj i prikladnoj lingvistiki filologi&skogo fakul'teta MGU, vyp. 9, Moscow, 103-128. Krause, S. 1979. Topics in Chukchee Phonology and Morphology. Unpublished PhD dissertation, University of Illinois. Levin, J. 1987. Between epenthetic and excrescent vowels (or what happens after redundancy rules). WCCFL 6, 187-201. McCarthy, J. & Prince, A. (ms) Prosodic Morphology I. University of Massachusetts and Rutgers University. McCarthy, J. & Prince, A. (to appear) Generalized aligrunent. In: G. Booij & J. van Made (eds.) Yearbook of Morphology. Mohanan, K. P. 1991. On the bases of radical underspecification. Natural Language and Linguistic Theory 9, 285-325. Muravyova, I. A. 1988. 0 morfonologiceskixosobennostjaxkompozitov. In: Jazyki narodov SSSR Novosibirsk: Novosibirskijgosudarstvennyj universitet. Prince, A. & Smolensky, P. (ms) Optimality Theory. Rutgers University and University of Boulder. Rubach, J. 1984. Cyclic and Lexical Phonology. The structure of Polish. Dordrecht: Faris.
226
227
Skorik, P. Ja. 1961. Grammatika cukotskogo jazyka I. Moskva-Leningrad: Izdatel'stvo akademii naul