Unpredictable Accent Patterns in Correspondence Theory January 2002 Diana Apoussidou Projekt C10 Heinrich-Heine-Universität Düsseldorf SFB 282 ‘Theorie des Lexikons’, Nr. 120: Phonologische Überlegungen zu Spezifikationen im Lexikon: Wortakzent – Silbenanlaut - Silbenschnitt 1 1 Introduction Lexical stress can be encoded in different ways and I will discuss two proposals in an Optimality Theory framework in this section: (i) exceptional marking of an accented vowel in the lexicon, and (ii) exceptional marking of a prosodic constituent (syllable, foot or prosodic word) in the lexicon. Canclini (1999), Gerlach (2001) and Popescu (2000) claim that some segments in Italian, European Portuguese and Romanian, respectively, may be lexically marked for a so-called 'prosodic head' or 'peak'; i.e. a segment in the lexical entry (or 'input') may bear main stress and this segment is the called the stress-peak. They furthermore maintain that in languages where stress is not always predictable, lexical stress is preserved in output forms due to a highly ranked prosodic faithfulness constraint which requires that a segment with a stresspeak in the lexical entry must surface with a main stress peak in the output (PEAK-MAX-IO).1 Consider in this respect that main stress in Italian is assigned to the penultimate syllable in the first person singular form coltivo [kol.ti.vo] 'I cultivate land', whereas it is assigned to the antepenultimate syllable in agito [a.di.to] 'I shake'. Also consider that main stress is on the final syllable of the stem in most Romanian verbs (adún ‘I collect’) and on the penultimate syllable in some verbs (ápr 'I defend'). For Canclini and Popescu, this implies that most stems are not marked for stress in the lexicon (e.g. Italian /koltiv/ and Romanian /adun/), whereas stems like Italian /ádit/ and Romanian /ápr/ are marked for lexical stress on the vowel of their penultimate syllable. The highly ranked constraint NONFINALHEAD (NONFINHEAD) ensures that main stress is not assigned to the final syllable in Italian; PEAK-MAX-IO says that a vowel which is marked for stress in the input should also be stressed in the output and ALIGN-HEAD-RIGHT (called EDGEMOST-RIGHT in Canclini 1999) requires that the syllable with main stress is as close as possible to the right word edge: (1) Default stress assignment for Italian according to Canclini (1999) koltiv +o NONFIN HEAD PEAK-MAX-IO ALIGN-HEAD-RIGHT a. kol.ti.vó *! b. kol.tí.vo * c. kól.ti.vo * *! (2) Lexical stress in Italian according to Canclini (1999) ádit +o NONFIN HEAD PEAK-MAX-IO a. a.di.tó *! * b. a.dí.to *! c. á.di.to ALIGN-HEAD-RIGHT * ** The analyses of Canclini, Gerlach and Popescu are closely related to the accounts of main stress provided by McCarthy (1995) and Kager (2000), but differ from them in one important 1 Canclini furthermore distinguishes between a faithfulness constraint for main stress only (PEAK-MAX-IO) and one for main or secondary stress (MAX-STRESS) and Popescu (2000) limits the scope of the prosodic faithfulness constraints MAX-STRESS and DEP-STRESS to stems and affixes. For Modern Hebrew exceptional stress patterns, Graf (1999) assumes underlying metrical structure and she adopts separate faithfulness constraints that refer to prosodic structures in input stems and affixes, respectively. 2 respect: due to the observation that main stress can exceptionally be assigned to another syllable than in regular cases, Canclini, Gerlach and Popescu claim that stress can be a lexical property (i.e. a property of input representations). Even though this is an advantage, their proposal suffers from the same drawback as McCarthy's and Kager's analyses in that the interpretation of PEAK-MAX-IO implies that stress is a property of a segment, rather than a syllable (Liberman & Prince 1977). Moreover, Canclini assumes a separate constraint MAX-STRESS which says that every peak in the input should have a correspondent in the output; this particular constraint is only violated when a vowel which is specified as 'peak' in the input is realized without (primary or secondary) stress in the output (Canclini 1999:86). It is curious to assume two constraints for lexical stress and below I will consider a possible analysis of Modern Greek (henceforth MG) stress alternations in which main stress is not considered a property of a single segment, but rather a relational property of a syllable (one syllable within a foot is more prominent than the other syllable in that foot and all unfooted syllables do not have stress). In Modern Greek, the position of word stress is unpredictable: a language learner cannot decide by default which syllable in a word should be stressed. In other languages, main stress can always be assigned to the same position in a word. In Icelandic, for instance, main stress is always on the first syllable of the word, and in French, words have main stress on the last syllable. In MG, any of the last three syllables may be stressed. This kind of irregular stress assignment may be attributed to specifications in the lexicon, where morphemes are marked for stress (stafía 'raisin-Nom.sg.'). Conflicts arise, when more than one morpheme in a word is specified for stress, since MG allows only one main stress per word (spírto + kutí = spirtokúti, *spírtokutí 'matchbox-Nom.sg.'). Previous analyses of MG (e.g. Revithiadou 1998) show that the stress patterns in MG can be explained in terms of Optimality Theory (OT) and its successor Correspondence Theory (McCarthy & Prince 1995). Faithfulness constraints referring to specifications in the lexicon as well as prosodic markedness constraints interact with each other and account for stress placement. My aim is to show that no reference to the morphological class of a word is necessary to explain the facts (in contrast to Revithiadou 1998). Rather, the grammar of MG is only sensitive to certain kinds of metrical information in the lexicon. The following properties of MG are relevant for my analysis. First, syllable weight does not play a role in stress assignment. In other words, MG is a quantity insensitive language. Second, main stress is always limited to one of the last three syllables in a word (this is also known as 'the trisyllabic window'), and only one main stress peak per word is permitted (see for example Joseph & Warburton 1987; Malikouti-Drachman & Drachman 1980).2 Syllables are parsed (or grouped together) into feet and each foot has one prominent syllable at its left edge (trochaic foot) or its right edge (iambic foot). MG has binary trochaic feet; i.e. each foot has two syllables and the left syllable in a foot is prominent. Finally, MG has fusional morphology: a word minimally consists of a root and an inflectional ending (it should be noted, however, that there are a few exceptions, e.g. monomorphemic nouns like fós 'light-NOM.sg.', and function words). After the presentation of the data in section 2, the optimality theoretical analysis will follow in section 3, and I present concluding remarks in section 4. 2 I do not assume secondary or rythmic stress in MG, based upon phonetic anaysis from e.g. Arvaniti (1992). So all syllables which are not contained in the stressed foot are unfooted. 3 2 Modern Greek Stress Data The discussion on MG stress is limited here to one productive class of nouns, since they display all patterns of stress assignment attested in this language. 3 The data in (3) and (4) reveal some interesting properties. First of all, any of the last three syllables in a word may bear stress. Moreover, stress may either be fixed on a certain syllable, (as in klívanos - klívanu and stafía - stafión in 3b-d and 4b-d), or stress may alter its position from antepenultimate to penultimate or ultimate syllable (as in ánropos - anrópu and álasa - alasón in 3a and 4a). (3) masculine nouns in -os (NOM.sg.) and -u (GEN.sg.): a. ánrop-os anróp-u but: b. klívan-os klívan-u c. fantár-os fantár-u d. uran-ós uran-ú (4) feminine nouns in -a (NOM.sg.), -on (GEN.pl.): a. álas-a alas-ón but: b. ónol-a ónol-on c. stafí-a stafí-on d. for-á for-ón 'man' 'kiln' 'soldier' 'sky' 'sea' 'gondola' 'raisin' 'turn' Some preliminary conclusions can be drawn: a fixed stress pattern arises, when the root contains an inherent specification for accent, so stress remains on the same position regardless of the inflection (e.g. klívanos - klívanu), indicating that the root bears a mark for stress on a special position within a morpheme. In contrast, stress becomes mobile, when the root itself is unmarked, so that the specification of an inflectional suffix may appear on the surface (for example álas-a - alas-ón). If the word does not have any inherent specification, stress is assigned by default to the antepenultimate syllable due to general prosodic principles (e.g. ánropos, álasa), which constitute the default mechanism in the language. In the next section, I will turn to the analysis. 3 Analysis Morphemes may be specified for stress in the shape of an already existing foot structure in the lexicon (I follow an idea from Inkelas 1998, who stipulated inherent feet for exceptional stress in Turkish). Since feet are trochaic in MG, the leftmost syllable constitutes the phonological head of a foot, dividing it into a prominent part (named foot head) and a weak part (foot tail): (s w) or (s)4. 3 4 The stress patterns of adjectives and verbs are more restricted. 's' for 'strong' and 'w' for weak'. 4 MG stress assignment can be explained in considering only parts of feet as specifications in the lexicon, in such a way that a morpheme may be marked for either the head of a foot (s) or the tail (w) (cf. Revithiadou 1998). In (5-8) some examples are given. In case that a word consists of unmarked morphemes (which constitutes the default pattern), only prosodic principles are responsible for stress assignment (5). (5) Unmarked case (default pattern): a. alas + a Foot / \ s w (á.la.)sa Foot / \ s w b. anrop + os (án.ro.)pos The representation of marked morphemes is different. Let us have a look at words consisting of unmarked roots and marked suffixes. The examples in (6) indicate that the inflectional ending -on (GEN.pl., 6a) is specified for a strong part of a foot and attracts stress to itself. The GEN.sg. ending -u (6b) seems to carry a foot tail and surfaces as a prestressing element: (6) Marked morphemes (stressed and prestressing suffixes): s s a. alas + on a.la.(són) w b. anrop + u s w an.(ró.pu) Roots, on the other hand, seem to be specified only for foot heads, as the examples in (7) show. I claim that a root like uran- (7b) is marked on its last segment (in this case the consonant -n-) for part of a foot. The idea is that the morpheme causes the next syllable to be prominent. In this way, this phenomenon can be explained with the same principles as those for other marked roots.5 (7) Marked morphemes (stressed and poststressing roots): s s w s s w s s / a. klivan + os (klí.va.)nos b. fantar + os fan.(tá.ros) c. uran + os u.ra.(nós) If marked roots combine with marked suffixes, a stress clash arises. Since MG allows only one stress peak per word, only one specification may survive in the surface representation. In those cases root stress surfaces (8). (8) Marked morphemes (two specifications within a word): s s s w s+w s s w / a. stafi + on sta.(fí.on) b. uran + u u.ra.(nú), *u.(rá.nu) 5 In contrast to Revithiadou (1998), who assumes a floating accent for such cases. 5 3.1 Analysis of the default pattern To explain phonological stress assignment (the default pattern), a mechanism is needed that assigns stress to the antepenultimate syllable. This can be achieved by treating the last syllable in a trisyllabic word as extrametrical (Drachman & Malikouti-Drachman 1996). How can this be done in terms of OT? In OT the grammar of any language consists of a set of universal constraints, which impose restrictions on a set of candidates. Constraints may be violated in favour of satisfying a higher ranked constraint. The differences between languages is due to different rankings of these constraints. In the successor of OT, Correspondence Theory, the constraints compare two strings which stand in a corresponding relation to each other. In the case of MG stress, input and output are the corresponding elements. Lexical specifications are marked in the input, and highly ranked constraints are responsible for their appearance in the output. In this section I want to discuss the case when no specification for stress is available in the input. The following constraints have to be high ranking: FOOTBINARITY ensures that two syllables in a word group together as pairs, and thus build a binary foot (9). TROCHEE causes the left syllable in a foot to be prominent (10). (9) FOOTBINARITY (FTBIN) Feet are binary, i.e. consist of two syllables: (). (10) TROCHEE Feet are left-headed: (), (). NONFINALITY guarantees that the last syllable in an at least trisyllabic word is counted as extrametrical (11). On the other hand, stress has to be hampered from moving further to the left than to the antepenultimate syllable. This is due to the constraint ALIGN-FOOT-RIGHT (12). These two constraints are responsible for the fact that the scope of stress is limited to the three-syllable-window. (11) NONFINALITYFoot (NONFINFT) A foot shall not stand in final position in a word. (12) ALIGN-FOOT-RIGHT (ALIGN-FT-R) The right edge of a foot is aligned with the right edge of the prosodic word.6 In (13-15) it is shown how the correct output form can be selected from a possible candidate set. In a bisyllabic word without a lexical specification (13), there is no antepenultimate syllable, where the default pattern could apply. So there are two possibilities left: either stress could be placed on the first or the second syllable. The evaluation in (13) shows that stress is assigned to the first syllable. A candidate like (13c) realises an iambic foot and thus violates the constraint TROCHEE, which demands the left part of a foot to be prominent. Candidate (13b) leaves a syllable unparsed, so that it violates FTBIN. Even though all candidates violate NONFINFT in that they all parse the last syllable into a foot, candidate (13a) is the winner, since it doesn't violate any of the other relevant constraints. 6 Violations are counted by the number of syllables between the foot and the right word edge. 6 (13) krátos 'state-NOM.sg.' krat-, -os TROCHEE a. (krátos) b. kra(tós) c. (kratós) *! FTBIN *! NONFINFT * * * ALIGN-FT-R In a trisyllabic word consisting of unmarked morphemes (14), stress is assigned by default to the antepenultimate syllable. The winning candidate in (14a) builds a perfect binary trochee and leaves the last syllable unparsed. It only violates ALIGN-FT-R, but this is not crucial, since this constraint is lowly ranked. A candidate like (14b), which realises stress on the same position as the winning candidate in (13a), has to fail, since now there is a better candidate that fulfils NONFINFT. (14) ánropos 'man-NOM.sg.' anrop-, -os TROCHEE a. (ánro)pos b. an(rópos) c. anro(pós) FTBIN NONFINFT *! *! * ALIGN-FT-R * The low ranking of ALIGN-FT-R becomes crucial in the case of words with four syllables (15). To prevent stress from moving to the pre-antepenultimate position, each violation of ALIGNFT-R counts (compare candidate 15a with candidate 15b). In this way, it is guaranteed that stress is kept within a certain scope. (15) astráalos 'ankle-NOM.sg.' astraal-, -os TROCHEE a. (ástra)alos b. a(stráal)os c. astra(álos) FTBIN NONFINFT ALIGN-FT-R **! * *! We thus arrive at the following ranking for the phonological default pattern in MG: TROCHEE, FTBIN, NONFINFT >> ALIGN-FT-R. At this point of the analysis, only the ranking between ALIGN-FT-R and the other constraints is established. In the next section it will be shown that other constraints come into play as soon as a word has a lexical specification. 3.2 Analysis of words with lexical specifications As soon as morphemes with a specification for stress are part of a word, additional constraints become relevant that have no impact on words without such a specification. These constraints demand (i) that lexical specifications are realised in the output, (ii) that the specifications do not change their position, and (iii) that they retain the metric information of specification. Faithfulness constraints like MAX ensure that a lexical specification will surface in the output (16), while markedness constraints such as ANCHOR-POSITION ensure maintenance of position (17). 7 (16) MAX(HEAD/TAIL) A specification for a foot head in the input surfaces as a foot head in the output. A specification for a foot tail in the input surfaces as a foot tail in the output. (17) ANCHOR-POSITION(Foot, Foot, Initial/Final) (ANCHOR, McCarthy 2000: 161) An element that is initial/final in a foot in the input is initial/final in the corresponding foot in the output. These constraints have to be ranked higher than the constraints which are relevant for the default pattern (FTBIN in 9, NONFINALFT in 11, and ALIGN-FT-R in 12), because otherwise the default pattern would always surface.7 Let us have a closer look at words with one marked morpheme. In (18), the word fantáros contains a marked root, i.e. the second syllable is specified for foot structure. A candidate like (18c) that does not realise a stress peak at all would be rejected, since words in MG have to have a prominence peak. It would also violate the MAXconstraint, since one of the morphemes of the word has a specification which is not realised in the output. Candidates like (18a) realise stress on a wrong position, since it moves one syllable to the left, and is thus ruled out by ANCHOR. The winning candidate (18b) parses the last syllable into a foot, violating NONFIN. However, this is not crucial, since it is a lowly ranked constraint. So in the end we have a word that is parsed into a binary trochaic foot which is anchored on a certain position (18b). (18) fantáros 'Soldier-NOM.sg.' fan(tar-, -os MAX(HEAD) a. (fánta)ros b. fan(táros) c. fantaros *! ANCHOR *! FTBIN NONFINFT * The ranking established up until now is as follows: MAX(HEAD), ANCHOR >> FTBIN, NONFINFT. In words with an unstressable (or poststressing) root like in (19) the same principles are valid. The specification for stress is anchored to the last segment of the root, so that the trochaic foot starts with this segment, as in candidate (19c). The other possibilities, namely moving the specification to another position, would violate ANCHOR in both cases, which is worse than violating the lower ranked prosodic constraints FTBIN and NONFIN. (19) uranós 'sky-NOM.sg.' ura(n-, -os MAX(HEAD) a. (úra)nos b. u(rános) c. ura(nós) ANCHOR *! *! FTBIN NONFINFT * * * Tableau (20) shows a word containing a prestressing suffix. This suffix is specified as a weak part of a foot. The rest of the MG grammar (especially TROCHEE) seeks to build the rest of the 7 The constraint TROCHEE is considered to be undominated inMG, so it does not appear in the following tableaux. 8 foot, so the preceding syllable gets the strong or prominent part of the foot, and thus is stressed. Other candidates which have another prominent position fail since they shift the stress peak (20a) or change the lexical specification (20c). (20) anrópu 'man-GEN.sg.' anrop-, -u) MAX(HEAD) a. (ánro)pu b. an(rópu) c. anro(pú) MAX(TAIL) ANCHOR *! *! FTBIN NONFINFT * * * Up to now, the ranking MAX(HEAD), MAX(TAIL), ANCHOR >> FTBIN, NONFINFT (i.e. between the constraints responsible for the default case and the constraints responsible for the marked case) is established. In words where only one of the morphemes has a lexical specification for stress, this morpheme can realise its stress specification. But what happens, when more than one morpheme in a word is marked? Only one stress peak is permitted in MG, so where in the word does stress surface? In tableau (21) you see a word containing two specifications for a prominent part of a foot, namely in the root stafi- and in the inflectional suffix -on. All candidates in (21) violate MAX, since they realise only one specification, even though two specifications are available in the input. The actual output (21b) is the one which does not change the position of an underlying specification and builds a prosodically wellformed foot. (21) stafíon ‘raisin-GEN.pl.’ sta(fi-, -(on MAX(HEAD) a. (stáfi)on * b. sta(fíon) * c. stafi(ón) * MAX(TAIL) ANCHOR *! FTBIN NONFINFT *! * * Consider now tableau (22): it shows that it is more important to realise a specification for a strong part than to realise a weak part of a foot, and to keep the position of a specification. The candidates which change the position (22b and 22c) are prosodically more wellformed than the winning output, but that is less important. (22) uranú 'sky-GEN.sg.' ura(n-, -u) MAX(HEAD) a. ura(nú) b. u(ránu) c. (úra)nu ANCHOR MAX(TAIL) * *! *!* FTBIN * NONFINFT * * So the ranking is MAX(HEAD), ANCHOR >> MAX(TAIL), FTBIN >> NONFINFT. So far it is striking that prosodic wellformedness is subordinate to lexical specifications. When derivational suffixes with a lexical specification come into play, this is confirmed. According to Revithiadou (1998), derivational suffixes are dominant (since they constitute the morphological head of the word) and may assign stress by means of HEADFAITHFULNESS. I want to propose a different solution to show that it is not necessarily the morphological head 9 of a word which assigns stress, but rather another kind of faithfulness, namely a constraint ensuring that a lexical mark in the input doesn't change its specification in the output: (23) INPUTHEAD # OUTPUTTAIL A foot head in the input may not become a tail in the output. This constraint does not refer to a morphological class, but only to the specification that a morpheme may have. A specification for stress in the lexicon does not “know” whether it belongs to a root or a kind of affix, but rather to a syllable or a segment. So in the end there are only prosodic principles that are in competition with restrictions referring to lexical specifications. This is shown in tableau (24). The candidate in (24a) represents the same pattern as the output in (14), that is the default pattern. But in this case, another candidate is selected as the optimal output, namely the one which parses the last two syllables into a foot (24b). (24) stafiáki (stafíRoot + ákDerivSuffix + iInflecSuffix) 'small raisin-NOM.sg.DIM' sta(fi-, -(ak-, -i MAX I-HEAD ANCHOR MAX FTBIN (HEAD) #O-TAIL (TAIL) a. sta(fía)ki * *! b. stafi(áki) * NONFINFT * The constraint ranking that I propose for MG is as follows: MAX(HEAD), I-HEAD#O-TAIL, ANCHOR >> MAX(TAIL), FTBIN >> NONFINFT. The intuition behind this ranking is that in the unmarked case stress tends to be more leftwards in a word, while in the marked case it prefers the right edge. 4 Conclusion Lexical stress or accent can be explained via specifications in the lexicon. This accounts for the unpredictable stress pattern in MG, as already shown in previous analyses (e.g. Revithiadou 1998). These specifications are represented in the lexicon as parts of foot structure, i.e. morphemes may be marked for a strong or a weak part of a foot. This contrasts with Inkelas' treatment of exceptional stress patterns in Turkish (Inkelas 1998), where she assumes entire feet to be specified in the input of morphemes. Her account would not work in MG since too many conflicts would arise under this assumption, which could not be solved. So the distinction between a strong or a weak part of a foot has to be made. Furthermore, the accounts proposed by Canclini (1999), Gerlach (2001) and Popescu (2000) cannot cope with the MG situation, too. The assumption of exceptional marking of a single vowel in the lexicon fails to explain phenomena like unaccentable or preaccenting morphemes, since in this case there is no corresponding vowel available in the lexicon that could be marked. To realize lexical stress, faithfulness constraints referring to corresponding specifications in the lexicon (markings for foot structure) in interaction with prosodic markedness constraints are required, and there is no need to assume constraints referring to floating features nor to classes of morphemes (like in Revithiadou 1998). The kind of metrical information, namely if it is a head or a tail of a foot, is crucial for the placement of accent, regardless of whether the corresponding morpheme is regarded as dominant or not. 10 References Arvaniti, Amalia (1992). Secondary Stress: Evidence from Modern Greek. In: Papers in Laboratory Phonology II (Gesture, Segment, Prosody). Ed. by Gerard J. Docherty and Robert Ladd. Cambridge University Press. Canclini, Emanuela (1999). Die italienische Verbflexion: Eine morphophonologische Analyse. MA thesis, Heinrich-Heine-Universität Düsseldorf. Drachman, Gabriel and A. Malikouti-Drachman (1996). Greek word accent. In: Word Prosodic Systems in the Languages of Europe. Ed. by Harry van der Hulst. Berlin & New York: Mouton de Gruyter. Gerlach, Birgit (2001). Clitics between syntax and lexicon. Doctoral dissertation, HeinrichHeine-Universität, Düsseldorf. Amsterdam: John Benjamins. Graf, Dafna (1999). The Metrical Structure of Modern Hebrew Nominals. MA thesis Heinrich-Heine-Universität, Düsseldorf. Inkelas, Sharon (1998). Exceptional stress-attracting suffixes in Turkish: representations versus the grammar. In: The Prosody-Morphology Interface. Ed. by René Kager, Harry van der Hulst, Wim Zonneveld. Cambridge University Press. Joseph, Brian D. and Irene Warburton (1987). Modern Greek. London: Croom Helm. Kager, René (2000). Stem stress and peak correspondence in Dutch. In: Optimality Theory: Phonology, Syntax, and Acquisition. Ed. by J. Dekkers, F. van der Leeuw, and J. van de Weijer. Oxford University Press. Liberman, Mark & Alan Prince (1977). On Stress and Linguistic Rhythm. In: Linguistic Inquiry, 8: 249-336. MS, University of Massachusetts: Amherst. Malikouti-Drachman, Angeliki and G. Drachman. 1980. Slogan chanting and speech rhythm in Greek. In: Phonologica 1980. Ed. by W. Dressler, O. Pfeiffer, and J. Rennison. Innsbruck: Insbrucker Beiträge zur Sprachwissenschaft. McCarthy, John J. (2000). Faithfulness and Prosodic Circumscription. In: Optimality Theory. Phonology, Syntax, and Acquisition. Ed. by Joost Dekkers, Frank van der Leeuw, and Jeroen van de Weijer. Oxford University Press. McCarthy, John J. & Alan S. Prince (1998). Faithfulness and identity in Prosodic Morphology. In: The Prosody-Morphology Interface. Ed. by René Kager, Harry van der Hulst, Wim Zonneveld. Cambridge University Press. -(1995). Faithfulness and Reduplicative Identity. MS, University of Massachusetts, Amherst and Rutgers University. -(1993). Prosodic Morphology: Constraint Interaction and Satisfaction. MS, University of Massachusetts, Amherst and Rutgers University. 11 Popescu, Alexandra (2000). Rumänische Verbextensionen. Paper presented at the Phonology Circle, 10 and 17 April 2000, Heinrich-Heine-Universität Düsseldorf. Prince, Alan & Paul Smolensky (1993). Optimality Theory: Constraint Interaction in Generative Grammar. MS, Rutgers University, New Jersey. Revithiadou, Anthi (1998). Headmost Accent Wins. Head Dominance and Ideal Prosodic Form in Lexical Accent Systems. Holland Academic Graphics: The Hague. 12