Summary and Conclusions This thesis focuses on the accentuation of systems with lexically determined stress. Two claims are central in this study. First, prosodic structure serves a as parsing cue for morphological structure in lexical accent systems and second, words have unpredictable stress but predictable prosodic shape. Both proposals are developed based on the examination of Greek, Russian, and four languages of the Salish family, namely Thompson, Spokane, Moses-Columbia and Lillooet Salish. In lexical accent systems, morphemes are equipped in the lexicon with an autosegment called lexical accent or simply mark. Conflicts between lexical accents arise when morphemes are combined to form complex expressions and the language requires one prosodically prominent element within the word. Such conflicts are resolved through morphology. This statement implies that first, prosody has access to morphological information and second, prosodic structure is formed in parallel with morphological structure. The principle that enables the prosody-morphology interface is prosodic compositionality. This principle allows prosody to peek into morphology and become sensitive to relations that hold between morphemes. The function that maps morphological structure into prosodic structure is expressed as head dominance: accents sponsored by morphological heads dominate other accents in the word. Head dominance takes the form of the constraint ranking HEADFAITH >> FAITH. The constraint FAITH states that an input accent must have a correspondent in the output, whereas the constraint HEADFAITH confines this statement to accents sponsored by morphological heads. In general, HEADFAITH >> FAITH is a form of positional faithfulness ranking in which the more specific constraint HEADFAITH must be ranked above the more general constraint FAITH. Other accents have a chance to surface only when the head is accentless, whereas words that do not display any inherent accentual properties
324 SUMMARY AND CONCLUSIONS are stressed by the lower ranked DEFAULT constraints. Under the term default are grouped structural constraints that determine the way input forms are footed, the directionality of prominence, and so on. To summarize, accentuation in lexical accent systems is dependent on morphological headedness but it is not totally controlled by it. Accents sponsored by other constituents or rhythmic principles also play a role in accentuation. This is the reason that lexical accent systems in this study are also called head-dependent systems with lexical accents. The ranking schema in (1) summarizes the accentuation in head-dependent systems. (1) ranking schema for head-dependent systems with lexical accents HEADFAITH >> FAITH >> DEFAULT The distribution of lexical accents is not only morphologically based but also prosodically controlled. Only a few positions in the word are targeted by lexical accents. In Greek and Russian, lexical accents occur in positions that ensure that a given structure will have a strictly binary prosodic shape. More specifically, accented words in these languages draw their accentual shapes from the following pool: {(σ#σ)(σσ),(σσ)(σ#σ),(σ#σ)σ,σ(σ#σ)}. In simple words, lexical accents limit their arbitrariness by restricting themselves to a confined set of syllabic positions that guarantee well-formed prosodic patterns. I propose that restricted lexical contrasts arise when constraints on word-form exercise influence on the exact position of a lexical accent. That is to say, next to a constraint that demands faithfulness to the autosegment accent there is another constraint that demands faithfulness to the association of a lexical accent with its underlying vocalic peak. The former constraint is undominated but the latter constraint is under the spell of constraints that determine the prosodic shape of a word and guarantee prosodic wellformedness. A positive outcome of this approach is that unattested accentual patterns are eliminated by constraint-evaluation and not by stipulating restrictions on underlying representations. The ranking in (2) formalizes the claim just presented. (2) restricted accentual contrasts FAITH TO LA >> WORDFORM >> FAITH TO POSITION OF LA In conclusion, words with lexical accents reflect morphological headedness in their prosody and have predictable (maximally binary) prosodic patterns. This characteristic brings marked words in a better position than unmarked ones. The latter have predictable stress on some syllable but variable prosodic shape and mobile stress within the paradigm. In Greek, marked four-syllable words are
SUMMARY AND CONCLUSIONS 325 binary, (servi)(tóros) waiter, but unmarked words are not, a(stráγa)los ankle. Moreover, marked words have columnar paradigmatic stress, klívanos, klívanu kiln, whereas unmarked words have mobile stress, án7ropos, an7rópu man. The accentual alternations of the latter example are closely related to the presence of an accentless head. When the root/head lacks inherent accentual properties, accentuation is determined by simple faithfulness or the default constraints depending on whether the inflectional suffix is accented or not. Optimality Theory offers an explicit framework to account for conflicting demands. Constraint ranking successfully expresses the idea that there are hierarchically ordered preferences in accentual systems. More specifically, it makes explicit why priority is given to head constraints over other constraints of the system and how restricted accentual contrasts arise. Besides these two central issues, there are some other questions related to lexical accent systems. These questions are: i. Where exactly on the stress map do lexical accent systems stand? ii. What is the phonological identity of a lexical accent? iii. How do constraints that refer to lexical accents interact with other constraints of the grammar? iv. Is prosody sensitive to different types of morphology and, if yes, how is head dominance expressed in different morphological constructions? The first question is addressed in Chapter 1 where I present a typology of accentual systems. I argue that next to systems with fixed stress there is a group of languages whose stress behavior can be best understood if seen from the perspective of the prosody-morphology interface. Such systems are called interface systems and are grouped into different categories depending on how they rank structural constraints (STRUCTURAL), constraints referring to lexical accents (FAITH) and constraints referring to morphological headedness (HEADFAITH and HEADSTRESS). Lexical accent systems constitute a specific class of interface systems where, as mentioned above, priority is given to marked morphological heads but where constituents other than heads can also reveal their marking properties. There are, however, related languages that are more forceful in promoting one-to-one correspondence between morphological and prosodic heads, and languages that do not exploit morphological headedness for accentual phenomena. All varieties arise from a reranking of the aforementioned archetypical constraints. A factorial typology of stress systems is given in the Appendix of Chapter 1.
326 SUMMARY AND CONCLUSIONS The second question is addressed in Chapter 2. On the basis of empirical evidence, I argue that a lexical accent is an autosegmental feature, an abstract entity which provides no cues about its phonetic manifestation. If included in the prosodic word, it is assigned phonetic make-up which varies from pitch to stress depending on the language. In addition, lexical accents are divided into strong and weak. A strong accent corresponds to a prosodic head and is phonetically realized as stress in languages with dynamic stress or high tone in pitch-accent languages. A weak accent avoids prosodic prominence either by being in a weak prosodic position (i.e. foot-tail), or by hosting a low tone, or by having duration but no loudness. This chapter is completed with a presentation of other theories of marking. The third question is addressed in Chapter 3, which focuses on the prosodic aspects of two lexical accent systems, namely Greek and Russian. Two main points are made in this chapter. First, lexical marking is not tantamount to exceptional stress. Second, the distribution of lexical accents is not free. Faithfulness constraints urge an inherent accent to be realized in the output form. Structural constraints on foot-form, on the other hand, enforce more rhythmic patterns. The effects of marking become apparent when FAITH outranks FOOTFORM constraints (i.e. DEFAULT constraints). As already mentioned above, lexically assigned metrical information in Greek and Russian is restricted by constraints that define the prosodic shape of the word, namely word-form constraints. These constraints dominate faithfulness to the underlying position of a lexical accent. Consequently, well-formed prosodic structures in marked words emerge from the ranking: FAITH TO LA >> WORDFORM >> FAITH TO POSITION OF LA >> FOOTFORM. The fourth question is central in this thesis and is addressed in the last two chapters. Chapter 4 accounts for competing accents in two languages with fusional morphology: Greek and Russian, and Chapter 5 accounts for competing accents in four Salish languages with polysynthetic morphology. In Chapter 4, the theory of head dominance is tested in inflected and derived constructions of Greek and Russian. In both languages, a marked root attracts stress from a marked inflectional suffix but it loses stress from a marked derivational suffix. The explanation is simple: inflectional suffixes cannot change the syntactic category of the base they attach to. Roots and derivational suffixes can do this; they are the morphological determinants (i.e. heads) of the word. In this chapter it is further shown that the theory of head dominance voids the need for the complex derivational machinery of cyclic and non-cyclic levels. Dominance is not an attribute of cyclicity but the result of being a morphological head and being accented. It is not necessary to motivate cyclic
SUMMARY AND CONCLUSIONS 327 and non-cyclic strata with independent function in order to derive the correct stress results. There is one function (one ranking) that is sensitive to the structural roles of morphemes and not to the scope in which phonological operations take place. In addition, the theory of head dominance offers a convincing counterproposal to the metaconstraint ROOTFAITH >> SUFFIXFAITH (McCarthy and Prince 1995). It restates this metaconstraint as a type of positional faithfulness ranking. When the root is the head of the word both head dominance and the metaconstraint make the same predictions, but when derivational suffixes are heads, then only the former ranking makes the right empirical predictions. The theory of head dominance developed in Chapter 4 is extended to polysynthetic languages in Chapter 5. Following Baker (1988), who claims that morphological structure in these languages is built in the syntax, I argue that the (morphosyntactic) head is also accentually prominent. Interestingly, stem-level and word-level stress is pursued in a somewhat different fashion in Salish. The major construction of stem-level morphology is lexical suffixation, a form of incorporation. In such formations, the root is the head of the VP into which the lexical suffix incorporates. The root/head is prosodically dominant, if accented; otherwise, the (incorporated) lexical suffix is given a chance to reveal its inherent accent and determine stress. In other words, the ranking HEADFAITH >> FAITH is also central for the accentuation of stem-level morphology in Salish languages. At the word-level, however, the picture is somewhat different. Aspectual and modal markers, which head aspectual and modal phrases, respectively, are accentually dominant, regardless of whether they bear an accent or not. In other words, word-level morphology reveals a stricter form of head-dependence, expressed with the ranking HEADFAITH >> HEADSTRESS. This domination order guarantees a one-to-one correspondence between heads and stress. The split in Salish accentuation can be attributed to morphosyntactic differences between lexical and grammatical suffixation. Lexical suffixation is a lexically flavored syntactic operation. The incorporated element is a suffix and not an independent noun. Lexical suffixation is needed to establish the appropriate configuration in which complex expressions can be licensed. On the other hand, grammatical markers have a certain degree of autonomy. Whether they incorporate to the root or not is decided by syntactic rules during word formation and not in the lexicon. This thesis makes clear that specific aspects in the morphological component of the grammar are important for phonology. There is a group of accentual
328 SUMMARY AND CONCLUSIONS systems, the interface systems, in which prosodic structure is a parsing cue for morphological structure. There are many ways in which this prosodymorphology interface is expressed. Some languages choose to assign a special prosodic status to specific morphemes or grammatical markers (morphologydependent systems), whereas some other languages choose to promote elements that are structurally important to prosodically important positions (headdependent and head-stress systems). It is interesting that the prosody-morphology interface in the systems examined here is always interpreted as head prominence. It has been argued by Dresher and Van der Hulst (1997) that heads allow richer contrasts and a greater degree of complexity than non-heads. This claim is reinforced by recent studies (Elordieta 1997, 1998) which show that other phonological phenomena as, for instance, vowel assimilation (in Basque), are sensitive to morphosyntactic relations holding among syntactic heads. We assert from the above that the existence of head constraints in the grammar cannot be refuted. One may wonder, however, whether there are systems in which non-heads take up a prominent prosodic role. It is well-known that constituents above the level of the word are not strictly head-oriented. For example, stress in phrases and in (phrasal) compounds is often hosted by the non-head constituent. The asymmetry between word and phrase level accentual phenomena must be explored in the future. Another asymmetry that needs to be closely looked at is the one between accent systems and harmony systems. Vowel harmony does not express asymmetrical dominance the way accent does. In Turkish, for instance, derivational suffixes are as sensitive to harmony as inflectional suffixes are. For these cases McCarthy and Prince s metaconstraint seems to gain ground, even though it still falls short in explaining why roots, and not other morphemes, behave in a special way. Perhaps a way to account for these facts is to investigate what types of head constraints exist in Universal Grammar besides the head constraints proposed here, which mainly refer to morphological headedness. This study focuses on head-dependent systems with lexical accents, but, as mentioned earlier, there are many varieties of head-based systems and many different ways in which prominence is phonetically realized. Further research should shed more light on the accentual behavior and other phonological properties of these systems. A final issue that should be investigated more in the future is whether the degree of head-dependence is related to the type of rules that participate in word formation. We have seen in Greek, Russian and part of Salish that morphologically oriented structures promote heads but not all the way;
SUMMARY AND CONCLUSIONS 329 accentuation is partially controlled by other constituents (i.e. non-heads). On the other hand, structures which are derived by purely syntactic operations such as the Salish transitive and intransitive formations, are more eager to accomplish head-to-stress correspondence. This study shows that all lexical accent systems share one property: dependence on morphological headedness. Head dominance, expressed with the ranking HEADFAITH >> FAITH, is the central component of accentuation in the grammar of these languages. It can interact with prosodic constraints but it can never be dominated by them. So, even though Greek, Russian and Salish differ in their morphological and rhythmic make-up, they are in principle head-based.