Abstract

The uniformity of stress assignment across inflectional forms in Albanian leads to massive phonological opacity, which seems to lend itself either to paradigmatic output-output constraints (Benua 1997, McCarthy 2005) or to a stratal organization of phonology (Kiparsky 2000, Bermúdez-Otero 2008) where inflected word forms preserve the stress assigned to stems at an earlier stratum. In this article, I show that a detailed analysis of Albanian morphology provides strong evidence for a stratal account: stress position in inflected word forms is correctly predicted by their stems, but not by ( partially defective) paradigms.

1 Introduction

In Albanian, word forms lacking overt inflection (e.g., nominative indefinite nouns) have final stress if the final syllable is closed (1a) or ends in a nonmid vowel (1b–c), while stress falls on the penultimate syllable if the final syllable ends in a mid vowel (1d–e) (Bevington 1974, Trommer, to appear).1 In forms with inflectional affixes (e.g., accusative definite nouns), stress falls on the same syllable as in corresponding forms without overt inflection, even when the phonology would trigger a different stress position. For example, for adetin we expect final stress, but stress falls on the penultimate syllable. In other words, stress in forms with inflectional affixes is phonologicallyopaque.

(1) Opaque stress in Albanian nominal inflection

graphic

In this article, I argue for a stratal analysis of the Albanian data, where stress is assigned at the stem level and retained by input-output faithfulness at later levels. I show that paradigmatic output-output faithfulness is neither necessary nor sufficient to capture stress uniformity in the language. Importantly, it turns out that subtle details of morphological segmentation are crucial for the evaluation of alternative phonological analyses, an aspect largely neglected in the ongoing debate about whether morphologically mediated phonological opacity should be captured by constraints on paradigmatic uniformity (Benua 1997, Kager 1999, Downing, Hall, and Raffelsiefen 2005, Albright 2008b) or by partially derivational levels of computation (Kiparsky 2000, Collie 2007, McCarthy 2007, 2010, Bermúdez-Otero 2008). Opacity in Albanian stress assignment also adds to our understanding of phonological systems where nouns and verbs seem to differ substantially in their phonological behavior (Albright 2008a, Bobaljik 2008, Smith 2011): whereas nouns and other parts of speech in Albanian show the highly differentiated stress system illustrated in (1), verbs have consistent stem-final stress. Thus, a verb form such as kalon ‘you/(s)he passes’, which according to traditional analysis consists of the stem kalo and the inflectional suffix -n, would be expected to show the penultimate stress of ˈba.bo-n (1e) if nominal and verbal morphophonology follow the same principles. Here, I show that all differences in stress between Albanian nouns and verbs follow from morphologically conditioned opacity and the fact that all relevant verb stems in the language are underlyingly consonant-final (hence pattern with a.ˈdet ‘habit’ (1a)). Finally, the data from Albanian inflection provide evidence against both symmetric and asymmetric output-output approaches to paradigm uniformity. Asymmetric approaches, which construct paradigms based on single privileged base forms (Benua 1997, Kager 1999, Hall 2007, Albright 2008b), are shown to face specific problems with defective paradigms, where the relevant base form is missing. On the other hand, the symmetric Optimal Paradigms approach proposed by McCarthy (2005) cannot account for the differential treatment of nouns and verbs, or for the dependence of stress in inflectional paradigms on morphologically uniform base forms.

The article is organized as follows. In section 2, I introduce in more detail the phonological factors determining Albanian stress assignment and possible approaches to stress uniformity. In section 3, I discuss an apparent problem that verbal paradigms pose for a stratal analysis and show that this problem dissolves once an appropriate morphological segmentation of verb forms is adopted. In section 4, I argue that verbs actually pose a problem for an asymmetric paradigmatic account since there are paradigmatically defective verbs that do not provide adequate base forms to anchor appropriate output-output faithfulness constraints. In section 5, I introduce cases of defective nouns and stress-shifting inflection in nouns that further support a stratal analysis of stress. In section 6, I discuss the specific problems the Albanian data pose for Optimal Paradigms Theory. Section 7 contains a short summary.

2 Albanian Word Stress and Different Approaches to Paradigmatic Uniformity

In this section, I introduce the crucial phonological factors responsible for stress assignment in Albanian (section 2.1), recapitulate the optimality-theoretic analysis of word stress in base forms from Trommer, to appear (section 2.2), and introduce possible implementations of the stress uniformity effect (sections 2.3 and 2.4). In section 2.5, I discuss criteria for choosing between stratal and paradigmatic approaches.

2.1 Phonological Determinants of Stress Assignment

Albanian word stress is highly systematic. As shown in Trommer, to appear, the position of stress in base forms (stems without overt inflection for nouns, adjectives, prepositions, and adverbs, and the present indicative active 3sg form for verbs) can be predicted by purely phonological factors.2 In all overtly inflected forms, the stress pattern of the base form is retained. Stress assignment in a base form is sensitive to both the weight and the vowel quality of its final syllable. In a base form with a final open syllable, stress is determined by the quality of the final vowel. If this is a nonmid vowel (a, i, or u), it attracts stress, and the final syllable is stressed (2b). If the final vowel is instead a mid vowel, the vowel repels stress; hence, the penultimate syllable is stressed (2a).3

(2) Vowel quality and stress

graphic

If the final syllable of a base is heavy and contains a full vowel, it is stressed. Thus, as shown in (3b), all base-final syllables with coda consonants receive stress even when the syllable contains a mid vowel that in open final syllables repels stress (cf. (2)). The only exception to this generalization is a final syllable with a schwa nucleus, which is systematically unstressed even though the syllable is closed (3c). Crucially, sensitivity to syllable weight is restricted to the final syllable. Whether a nonfinal syllable is closed or has a diphthong does not influence the stress pattern of a given word. As shown in Trommer, to appear, weight and vowel quality of the final syllable account for the stress assignment of roughly 97% of the Albanian lexicon.

(3) Stress and the syllable weight of the final syllable

graphic

The phonological regularity of stress assignment is also evident from the fact that derivation by suffixation leads to phonologically predictable stress shift, as shown in (4): stress migrates from derivational bases to suffixes, reflecting the generalization that closed syllables and syllables ending in i are generally stressed in word-final position.

(4) Stress shift in derived nouns

graphic

2.2 An Optimality Theory Analysis of Stress Assignment in Base Forms

This section summarizes the optimality-theoretic analysis of base stress assignment in Albanian proposed in Trommer, to appear.

The fact that stress in Albanian always falls on the ultimate or the penultimate syllable is derived from crucially undominated ALL-FEET-RIGHT and RHYTHMIC-TYPE = TROCHEE (ALL-FT-RT, TROCHEE; Kager 1999), which have the effect that every prosodic word contains only one (monosyllabic or bisyllabic) foot at its right edge. Lower-ranked constraints basically select between these two alternatives under appropriate phonological conditions. The effect of vowel quality on stress is mediated by the ranking of the constraints in (5), which penalize specific vowels in the weak position of a foot.

(5)

graphic

In combination with FOOT-BINARITY (FT-BIN; Prince and Smolensky 1993:50), which requires that feet contain either two syllables or two moras (or both), and the ranking shown in (6), these constraints allow stress assignment to be derived in base forms with a final light syllable. If this contains a nonmid vowel, high-ranked *{a,i,u}w ensures a monosyllabic foot because the final vowel would otherwise constitute an unstressed foot vowel.

(6) Input: bari ‘lawn’

graphic

In base forms with a final mid vowel, FT-BIN ensures a bisyllabic final foot and hence penultimate stress.

(7) Input: babo ‘midwife’

graphic

(8) Input: balë ‘ball’

graphic

For base forms with a final consonant, an additional constraint becomes active. *Cμ (Rosenthall and Van der Hulst 1999)4 penalizes moraic coda consonants, capturing the fact that Albanian is generally not weight-sensitive. This constraint prevents a closed final schwa syllable from forming a foot on its own, resulting again in penultimate stress.

(9) Input: afër ‘near’

graphic

However, if a final closed syllable contains a full vowel, the coda consonant is assigned an exceptional mora in a monosyllabic foot to satisfy FT-BIN (10a), (11a). This violates *Cm, but avoids constructing a bisyllabic foot involving a violation of higher-ranked *{a,i,u}w or *{e,o}w (10b), (11b). Thus, Albanian exhibits ‘‘coerced weight’’ (see Morén 1999 and references cited there).

(10) Input: besnik ‘true’

graphic

(11) Input: patok ‘gander’

graphic

2.3 A Stratal Account of Stress Uniformity

We are now in a position to characterize the problem of stress uniformity in Albanian more concisely. Consider for example the nominative indefinite form ˈba.bo ‘midwife’, which corresponds to the bare noun root. As noted earlier, roots ending in a closed mid-vowel syllable have final stress (e.g., paˈtok ‘gander’). However, adding the accusative suffix -n(ë) to ˈba.bo results in ˈba.bo-n, not in *ba.ˈbo-n, which would be expected from the word’s phonological makeup. The penultimate stress of the nominative base form is retained even though the form itself seems to require final stress. Stress uniformity hence renders stress positioning phonologically opaque.

The Optimality Theory (OT) literature offers two major approaches to morphologically mediated opacity of this type: the stratal approach and output-output faithfulness. In the following paragraphs, I will show how stress uniformity for overtly inflected nouns can be derived under both approaches.

In Stratal OT (Kiparsky 2000, Bermúdez-Otero 2008), there are two levels of affixation and phonological evaluation: the stem level and the word level. In Albanian, the stem level roughly corresponds to stems ( possibly containing derivational affixes), and the word level to inflected words. If stress is assigned on phonological grounds at the stem level and input-output faithfulness constraints for stress are ranked high at the word level, stem (base) stress is retained independently of the phonological consequences of inflectional affixation. This is shown schematically for the nominative indefinite and the accusative definite forms of babo in (12).

(12) Stress uniformity under a stratal analysis

graphic

I will formalize faithfulness for main stress using the constraint , which requires correspondence between foot heads in line with the same constraint proposed for heads of prosodic words by McCarthy (2000:183).

(13)

  • If α is in Hˈ(Foot) and ot).

is violated once by every output foot whose stressed (head) vowel is not in correspondence with a stressed vowel in the input. Adopting the ranking introduced in section 2.2 for the stem level of Albanian, we must assume that is dominated by all relevant markedness constraints, because otherwise lexically prespecified prosodic structure could survive in stems, predicting unattested variation in stress positioning. Hence, the complete stem-level ranking for Albanian is as follows:

(14) Albanian stem-level ranking

graphic

At the word level, must be ranked above all other relevant constraints to ensure that only vowels corresponding to stressed (foot head) input vowels can be stressed (foot heads).

(15) Albanian word-level ranking

graphic

This is illustrated for babo-n (see (1e)) in (16). (16a) is excluded because it contains a foot head that does not correspond to an input foot head. Since lexical words in Albanian contain at least one foot, the underlying foot is retained (apart from resyllabification), and o is stressed (note that (16a) would win without high-ranked

(16) Input: (ˈba.bo)-n (word level, acc.def.sg ‘midwife’)

graphic

(17) shows a slightly more complicated example, the dative/ablative definite plural of the noun mollë ‘apple’, which demonstrates that must be ranked above ALL-FT-RT at the word level, since otherwise the erection of a new foot at the right word-edge, disrupting stress uniformity, would be preferred (17c).

(17) Input: (ˈmo.llëz)-a-ve (word level, dat/abl.pl ‘apple’)

graphic

As this analysis predicts, stress uniformity in Albanian holds without exception for the inflection of lexemes with morphologically invariant stems (see section 5.2 for stems with suppletion, which exhibit variable stress positions). This is illustrated by the two representative and exhaustive noun paradigms in (18) and (19).

(18) Complete inflectional paradigm of patok ‘gander’

graphic

(19) Complete inflectional paradigm of babo ‘midwife’

graphic

Many of the forms in (18) and (19) are predicted to have a different stress pattern by the phonological constraints governing stress in base forms (see section 2.2); without faithfulness to stems, forms ending in consonants and nonmid vowels (e.g., patokun and babon) would be incorrectly assigned final stress, and forms ending in mid vowels (e.g., baboje) would be incorrectly assigned penultimate stress. The data in (18) and (19) also illustrate an exceptionless generalization that derives straightforwardly from the stratal analysis: stress in Albanian never falls on a vowel that is part of an inflectional affix. This follows because stress is computed for stems in the absence of inflectional affixes at the stem level and is then transmitted faithfully to the corresponding stem vowels at the word level.

2.4 An Output-Output Account of Stress Uniformity

There are two major approaches to paradigmatic uniformity in the OT literature. Under the asymmetric approach (Benua 1997, Kager 1999, Albright 2008b), faithfulness between morphologically derived forms and their base forms is taken into account in the phonological evaluation of the derived forms, but not of their bases. Under the symmetric approach (Kenstowicz 1996, McCarthy 2005), entire paradigms are evaluated, and no single form has a privileged status. Here I will discuss a typical asymmetric approach. A detailed discussion of McCarthy’s Optimal Paradigms Theory is given in section 6.

Like the stratal approach, the output-output approach developed by Benua (1997) (called Transderivational Correspondence Theory, TCT) allows a straightforward extension of the analysis in section 2.2 to capture stress uniformity. Simplifying slightly, Benua assumes that phonological evaluation of a morphologically derived form F happens recursively. The first step is to evaluate the base of F; the second is to evaluate F itself, which is then subject to output-output faithfulness requiring identity to the output of base optimization.5 The resulting system is asymmetric because the derived form is forced to reflect properties of the base but not vice versa, hence allows both base and derived form to be derived by the same constraint ranking. (20) shows schematically how this system works for ˈba.bo-n.

(20) Stress uniformity under a paradigmatic analysis

graphic

The constraint ranking necessary to achieve this is exactly like the one in (15) except that is replaced by the output-output constraint , which assigns a constraint violation to every foot head H of a candidate C for a morphologically derived form F if H does not correspond to a foot head in EVAL(Base(F)).

(21) TCT constraint ranking

graphic

(22) and (23) illustrate the application of this ranking with babo-n. In the evaluation of the base (22), does not induce any constraint violations because babo does not have a morphological base (it is not a derived form). Hence, (22b) is selected on phonological grounds just as in (7).

(22) Input: babo (evaluation of the base)

graphic

In the evaluation of the derived form, (23a) is excluded because the foot head o does not correspond to a foot head in (22b) (the winning candidate of the base evaluation); hence, stress on a is retained even though (23a) would be more harmonic for the lower-ranked markedness constraints.

(23)

graphic

An important point to keep in mind is that the output-output relation invoked by must be restricted to specific affixes—namely, the inflectional affixes of Albanian—because derivational affixes do not show any relevant uniformity effect (see the examples in (4)). Thus, in a sense, the TCT approach also separates the affixes of the language into different strata with different phonological effects. Since Benua’s (1997) model also effectively invokes consecutive evaluation procedures for base and derived forms, the crucial difference between the models lies in the fact that Stratal OT uses abstract (not necessarily surfacing) stems as bases, whereas bases in TCT are always independently occurring words.6

2.5 Choosing between the Stratal and Paradigmatic Accounts

The cases we have discussed so far can be analyzed equally well under both the stratal and the paradigmatic accounts. The dilemma shown schematically in (24) is that stress that is opaque in a specific output form O (e.g., the accusative definite ˈba.bo-n) can be motivated either by a stem contained in O for which stress assignment is phonologically transparent (e.g., the root ˈba.bo) or by a paradigmatically related form for which stress assignment is phonologically transparent (e.g., the nominative indefinite ˈba.bo).

(24) Stress uniformity under an output-output analysis

graphic

Since stem and base form are usually the same, both approaches make identical predictions and are equally well-supported by the data. Hence, we are most of all interested in cases where the neat picture in (24) breaks down—that is, where stems and related output forms lead to different predictions either because transparent stem and transparent output form are nonidentical or because transparent output form or stem is lacking. Albanian apparently exhibits both phenomena. The first case appears to hold in the major class of verbs represented by formoj (here called simply class I verbs to avoid commitment to a particular morphophonological analysis). According to standard analyses of many verb classes (25), there is no stem that predicts the correct stress position. Thus, interpreting formoj as the stem formo plus the suffix -j, we expect penultimate stress since formo ends in a mid vowel, while stress assignment based on the full form formoj generates correct stress, which can then be enforced by output-output constraints on other forms such as formoni.

(25) Present indicative active of class I verbs

graphic

The second relevant case—lack of an appropriate output form—can be found in deponent verbs such as pendohem ‘I regret’, which lack active forms but show exactly the same stress patterns as nondefective verbs in their nonactive forms. Under the assumption that stress assignment in verbs is generally motivated by bases that are active forms, the lack of the latter ones undermines the claim that stress uniformity is motivated by output forms.

Finally there is a somewhat ambiguous case—namely, a minor class of nouns that show suppletion in the plural and at the same time seem to constitute the only exception to the generalization that stress is uniform across inflectional forms (e.g., njeˈri (sg) – ˈnjerëz ( pl) ‘human(s)’). Morphologically these plural forms can be analyzed either as cases of word suppletion, which would imply that they lack a separable stem, or as cases of stem suppletion (in the case of njerëz with a ∅-plural marker), which would mean that the base form njeˈri and the stem (ˈnjerëz) are distinct.

In the rest of the article, I will show that under appropriate morphological analysis the apparent evidence for a paradigmatic account is illusory and that the Albanian data provide a strong case for a stratal analysis of stress uniformity.

3 Verb Stress Can Be Analyzed Stratally

Final stress on class I verbs leads to a striking asymmetry between nouns and verbs in Albanian: whereas the position of stress in Albanian nouns and other parts of speech is highly heterogeneous (see the nouns in (1)), all Albanian verbs (and verb forms) have stress on their stem-final syllable. This follows straightforwardly in a paradigm-based account from the assumption that the base of the verbal paradigm is one of the present indicative singular forms, since these have either consonant-final stems and are affixless (see kullot in (26)) or are class I, which means that they bear final j in the 1sg and n in the 2sg/3sg, independently of whether these are part of the stem or suffixes. On the other hand, the stratal analysis predicts that any verb containing a polysyllabic stem ending in a mid vowel should have penultimate stress, just as noun stems do (e.g., ˈba.bo ‘midwife’).

Thus, the argument that Albanian verbs provide against a stratal analysis of stress uniformity stands and falls with the claim that class I verbs are vowel-final in their underlying lexical representation (e.g., /formo/). Since mid-vowel-final stems should induce penultimate stress in the stem, we would incorrectly expect *ˈformoj instead of forˈmoj. In this section, I will show mainly on the basis of morphological arguments that the underlying forms of class I stems are actually consonant- (or diphthong-) final. Under a stratal analysis, the stem is assigned final stress, which is then transferred to inflected forms just as in consonant-final nouns and verbs.

Albanian has two major classes of verbs, differing inter alia by their inflection in the present indicative active. The most productive class is the one represented by formoj (25), repeated in (26); it is traditionally taken to be the main class of vowel-final verb stems. In addition, there is a smaller class of consonant-final stems apparently taking slightly different affixes (class II verbs). (26) shows the morphological segmentation of the present tense paradigms into stems and inflectional suffixes that is assumed in most descriptive work on Albanian (e.g., Newmark, Hubbard, and Prifti 1982, Camaj 1984, Buchholz and Fiedler 1987, Agalliu et al. 2002).

(26) Standard segmentation of verbal inflectional suffixes (present indicative)

graphic

However, elsewhere I argue (Trommer 1997) that this analysis should be substantially revised as in (27), where class I stems are reanalyzed as consonant-final, undergoing a morphologically conditioned alternation of the stem-final formative between n and j (the stem of formoj that has underlying final n in the 2pl form /formon-ni/ becomes [formoni] on the surface, owing to a general process of geminate reduction).7 I leave the exact formal character of this alternation open here. A plausible implementation suggested by an anonymous reviewer is that n and j are thematic elements required for stem formation from roots—rather like thematic vowels in Romance languages, which show similar types of allomorphy. This analysis would also predict that n and j are phonologically stem-level affixes according to the principle that root-based affixation is obligatorily stem-level (see Bermúdez-Otero 2007:283, 2010).

(27) Revised segmentation of verbal inflectional suffixes (present indicative)

graphic

The i-formative in the 1pl and 3pl of class II verbs is a plural marker for subject agreement. It also shows up at many places in the paradigm where the person/number suffixes proper would otherwise appear after a consonant in word-final position, as in the imperfect active 1pl/2pl/3pl formon-i-m/formon-i-t/formon-i-n, the imperfect nonactive 1pl/2pl/3pl formo-he-sh-i-m/formohe- sh-i-t/formo-he-sh-i-n, or the optative 1pl/2pl/3pl formo-fsh-i-m/formo-fsh-sh-i-t/formo-fshi- n. I assume that it is a phonologically conditioned allomorph of an otherwise preferred ∅-marker and that it surfaces only to break up word-final consonant clusters, which also explains why it occurs before the 2pl morph -t as in imperfect formon-i-t/*formon--t, but not before -ni (*formoni- ni/*formon-ni).

This reanalysis has four advantages over the traditional account. First, it substantially reduces the number of suffix allomorphs to the alternation between -m/-n and -/-në, which follows from the general process of rhythmic schwa epenthesis (see below for detailed discussion). Second, it provides a unified account for all occurrences of j at the stem-suffix boundary: they are all analyzed as stem-final formatives. By contrast, under the traditional segmentation they are taken as independent inflectional affixes in the 1sg, but as unsegmentable substrings of the 1pl and the 3pl affixes, which is at odds with the observation that 1pl -m and 3pl -n do occur without preceding j in other parts of the paradigm (cf. the corresponding aorist forms thellua-m/thellua-n ‘we/they deepened’). Conversely, j and n do not occur as person/number suffixes anywhere else in the conjugational paradigm of Albanian. Consider the representative forms in (28), which show all relevant allomorphs of singular person/number markers (the last affix in the respective forms).

(28) Realization of 1sg, 2sg, and 3sg in different tenses and moods

graphic

Third, under this analysis, another minor verb class (class III verbs) that exhibits final vowels, but not stem-final j or n, receives the same analysis as class I verbs. Under the traditional segmentation, class III verbs like ha ‘to eat’ must be stipulated to take different affixes than other vowelfinal stems for most person/number categories.8 Under the reanalysis, as shown in (29), the two classes select exactly the same inflectional affixes.

(29) Main paradigm of j/n-final and vowel-final verbs (present indicative)

graphic

Fourth, the reanalysis explains why singular class I forms are not subject to rhythmic schwa epenthesis, a process that inserts schwa after inflectional suffixes consisting of a single consonant attached to a vowel- and stress-final base. Consider the paradigms of feminine nouns in (30). If the noun ends in an unstressed vowel, purely consonantal affixes are added (30a). However, if the noun ends in a stressed vowel, schwa is epenthesized after the suffix to ensure a bisyllabic foot (30b). This must not be confused with schwa insertion before the suffix, which occurs with consonant-final nouns and seems to be motivated by syllable structure well-formedness conditions (30c). Exactly the same patterns are found with verbs (31).

(30) Schwa epenthesis in nouns

graphic

(31) Schwa epenthesis in verbs

graphic

Interestingly, rhythmic schwa epenthesis does not happen in word forms without overt inflection that end in stressed syllables and the same consonants (e.g., pe.ˈlin ‘absinthe’, a.ˈdet ‘custom’, ka.ˈlem ‘pen’), and it also fails to apply if the stem is derived by a derivational suffix (e.g., mendim ‘thought’ from mend ‘mind’; see (4) for more examples). Hence, final schwa epenthesis indicates that the preceding consonant is an inflectional suffix. Crucially, under the assumption that present indicative singular forms bear the affixes -j and -n, they have to be marked idiosyncratically as exceptions to rhythmic schwa epenthesis. On the contrary, under the reanalysis proposed here, these forms are stems, and the nonapplicability of schwa epenthesis is straightforwardly predicted.9

The price to pay for the reanalysis of Albanian verb forms is the assumption that class I verbs show stem-final allomorphy between a nasal and a glide according to the following affix. However, this move adds nothing new to the morphology of Albanian because there are other verb classes that show idiosyncratic alternations of stem-final consonants (and preceding vowels) in present tense forms, as shown in (32).

(32) Main paradigm of j/n-final and s/t-final verbs (present indicative)

graphic

The pervasiveness of stem-final consonant alternation can also be seen in aorist and participle forms, where particular verbs show further specific alternations. Thus, for arrij, hyj, and gjej, the stem-final j is replaced by t (see (33e), (34e), (34g)), r (see (33d)), or nd (see (34g)). For rroj (see (33f ), (34f )), t is added to the j-final stem.

(33) Further stem-final mutation in participles

graphic

(34) Further stem-final mutation in aorist forms

graphic

The intuitive appeal of analyzing class I stems as basically vowel-final in the literature is probably due to the fact that their paradigm exhibits many forms that superficially suggest such an analysis. Consider, for example, present nonactive forms where the nonactive suffix takes the shape -e after class II stems, but -he after class I stems.

(35) Present nonactive forms of class I and class II verbs (formoj ‘form’, hap ‘open’)

graphic

Accepting the premise that the underlying stem for formoj in all these forms is /formoj/, we must assume that there is a process that consistently deletes stem-final j in these forms. In fact, there is independent evidence that such a (morpho)phonological process is at work. Thus, class III stems with a stem-final diphthong, such as shpie, and class I stems with a diphthong preceding j/n, such as ndiej, show systematic monophthongization before h in nonactive forms (36), and it is natural to assume that the process responsible for this reduction also reduces the diphthong like structure oj to o in this context. Similar patterns can be observed with optative and aorist forms.

(36) Diphthong-final stems in nonactive forms

graphic

Another group of class I forms is superficially vowel-final because the forms take j- or ninitial suffixes (e.g., formo-ni, underlyingly /formon-ni/). Similarly, in the imperfect 1sg and 2sg forms (e.g., imperfect 1sg [hap-ja]/[formo-ja], underlyingly /hap-ja/, /formojja/), the stem-final segment is also obscured by reduction due to an identical suffix segment. Finally, imperative forms of a number of class I verbs are strongly suggestive of a vowel-final analysis (e.g., formo! ‘form!’) since the imperative of virtually all consonant-final stems consists of the bare stem (or root) without a suffix (e.g., hap! ‘open!’). However, many class I stems actually exhibit a j in the imperative (e.g., laj! ‘wash!’, blej! ‘buy!’). We might either assume that j in these cases is yet another homophonous j-affix that is lexically restricted to specific class I roots, or assume that j is simply the general stem-final formant in all class I verbs and is deleted by a lexically restricted morphophonological process.10 Since these alternatives are empirically equivalent, the imperative data provide no decisive argument about whether class I verbs should be characterized as vowel- or consonant-final.

I conclude that class I stems are consistently consonant- (or glide-) final, which means that only class III stems are truly vowel-final. However, all attested class III stems are either monosyllabic (e.g., pi ‘to drink’, ‘to put’) or derived by prefixation from monosyllabic roots (e.g., përzë ‘to persecute’). Since prefixes are generally invisible to stress assignment and monosyllabic stems receive stress by default, class III verbs provide no evidence for the correct analysis of stress and stress uniformity. Crucially, all verb stems with theoretically relevant stress patterns are underlyingly consonant-final. Since these stems never contain schwa, the stem-level phonology of the stratal analysis correctly predicts stem-final stress for all verbs.

4 Verb Stress Cannot Be Analyzed Paradigmatically: Defective Verbs

I turn now to evidence against a paradigmatic output-output analysis of stress uniformity. My first argument comes from deponent verbs, which have a defective paradigm and hence lack appropriate base forms to motivate stress placement.11 In section 4.1, I introduce the empirical data and the basic problem for a paradigmatic approach. In section 4.2, I discuss in detail different ways to detect alternative bases for deponent verbs, and I show that these are highly problematic. (We will see in section 6 that the Optimal Paradigms approach can handle stress for deponent verbs, but at the cost of making wrong predictions for nouns.)

4.1 The Nonactive and Deponent Verbs in Albanian

Albanian verbs have an all-pervasive grammatical category called nonactive in the Albanianist literature. Morphosyntactically, nonactive is expressed according to verbal categories such as tense and mood by the suffix -e (e.g., vrit-e-m, kill-NAC-1SG, ‘I am being killed’), by the clitic u (e.g., uvra-v-a, NAC kill-AOR-1SG, ‘I was killed’), or by the alternation of auxiliaries (e.g., ka-m vrar-ë ‘I have killed’ vs. ja-m vrar-ë ‘I have been killed’). From a semantic point of view, nonactive serves three different functions (see Kallulli 2006 for detailed discussion). Nonactive forms are interpreted as either reflexives (e.g., lahem ‘I wash myself’), unaccusatives (e.g., kthehem ‘I return’), or passives (e.g., vritem ‘I get killed’). For all these nonactive forms, there are corresponding active forms: laj ‘I wash’, kthej ‘I turn around’, vras ‘I kill’. However, there is also a class of deponent verbs that have only nonactive forms with a fixed lexical meaning (Camaj 1984:192–193, Buchholz and Fiedler 1987:190–191, 510, Agalliu et al. 2002:272). (37) shows representative examples.

(37) Deponent nonactive verbs

graphic

A stratal analysis for stress assignment in verbs such as pendohem is straightforward. Since verb stems are always consonant-final, the final syllable of the stem pendoj is stressed at the stem level and retains stress after further affixation of -he and -m at the word level (recall from section 3 that the stem-final oj in pendoj is reduced to o by a general morphophonological process). However, an output-output analysis is problematic: in the inflectional paradigm of pendohem, there is simply no corresponding active form such as *pendoj or *pendon that would phonologically motivate final stress. Since pendohem would hence be effectively freed from the burden of output-output faithfulness, it should be subject only to the general phonological constraints of Albanian, which require final stress for a word ending in a closed nonschwa syllable. Hence, an output-output analysis incorrectly predicts *pen.do.ˈhem, with final stress.

4.2 Potential Alternative Bases

A possible escape hatch for an output-output analysis would be to look for appropriate nonactive base forms that might motivate stem-final stress for deponent verbs. Here, I will go through all possible base forms for a standard class I deponent verb and show that they either predict the wrong stress location or are implausible base forms for the paradigm. (38) contains the full paradigm for the deponent verb pendohem.12

(38) Exhaustive paradigm of pendohem ‘I regret’

graphic

(39) lists all analytic nonactive forms of Albanian that have a suffix with two syllabic nuclei such as the present 2pl form pendo-he-ni ‘you ( pl) regret’. To see that none of these is an appropriate base, recall that phonologically conditioned stress in Albanian always falls on one of the last two syllables, which for these forms would mean after the stem. Assuming that any of these forms is the base would predict incorrectly that verb stress falls after the stem whenever possible. Thus, the imperfect 2sg base pendoheshe would induce the prosodification *pen.do.- (ˈheshe), resulting in the present 1sg *pen.do.(ˈhem).13

(39) Nonactive forms of class I verbs with 2-σaffixes

graphic

Next, consider forms with a monosyllabic inflectional suffix complex (e.g., the aorist 1sg u pendo-va ‘I regretted’). Here the situation is slightly more complicated. If the suffix is a closed syllable containing a full vowel or if it ends in a, i, or u, we again incorrectly expect post-stem stress. Correct stem-final stress is predicted if the suffix ends in e or a schwa syllable.

(40) Nonactive forms of class I verbs with 1-σaffixes

graphic

Finally, the correct stress position is also predicted for potential bases whose final syllable has a diphthong and that either lack an overt affix (e.g., aorist 3sg u pendua ‘he regretted’) or have a suffix consisting only of a consonant (e.g., the participle penduar),14 since for these forms the stem-final syllable is heavy and correctly attracts stress.

(41) Nonactive forms of class I verbs with no syllabic affix

graphic

This leaves us with the aorist 2sg (40), the optative 3sg (40), and the forms in (41) as potential bases for the correct prediction of stress. We can immediately eliminate the participle from this list because it would make the wrong prediction for class II verbs, which take the participle suffix -ur (e.g., hap-ur ‘opened’). Hapur is incorrectly predicted to have stress on its post-stem syllable since this syllable is heavy (cf. a.ˈhur ‘barn’, fla.ˈmur ‘flag’). Hence, we are left with the aorist and optative forms in (42).

(42) Potential alternative bases

graphic

Let us now check whether any of these forms constitutes a good base according to standard criteria for basehood.

Basehood by containment (Benua 1997, Kager 1999): Perhaps the most obvious approach to basehood, at least in derivational morphology, is the idea that the base is a morphological unit that is ‘‘contained’’ in the dependent word or word form (i.e., the base contains a subset of the morphemes contained in the dependent form) or from which the dependent word is ‘‘morphologically derived.’’ Thus, Benua (1997:28–29) states that ‘‘the base is the independent word identified with the string that undergoes morphological derivation; in affixation the base is the word identified with the string adjacent to the affix.’’ While this idea has intuitive plausibility for derivational morphology, McCarthy (2005:172) shows that it is not in general applicable to the relation between inflected forms, and Benua herself admits that containment is a highly problematic criterion that ‘‘does not always hold’’ (Benua 1997:29). A slightly different type of containment is invoked by Kager (1999), who claims that a base is a word that contains a subset of the grammatical features of the derived form. However, it is highly implausible that any of the Albanian forms in (42) would contain a subset of the features in a present indicative form. Even if we were to assume a privative feature system for tense and mood, the relevant features would be [aorist] and [optative], which are semantically marked and overtly marked by affixation in Albanian, not [present] and [indicative], which are unmarked in both ways.

Basehood by markedness or frequency: Another possible approach is to equate the least marked or the most frequent member of a paradigm with its base. While this position has not been explicitly argued for in the recent literature on phonological paradigm effects, there is some plausibility to it because it has been repeatedly claimed that the unmarked or most frequent form of a morphological paradigm serves as a base for historical paradigm leveling (Mańczak 1958, Greenberg 1966). Since markedness and frequency are highly interrelated concepts, I will discuss these approaches together.

It is quite obvious that neither aorist nor optative is the unmarked paradigm of Albanian in terms of morphosyntactic features; the semantically and morphologically unmarked paradigm is the present indicative. There are no reliable frequency data for Albanian verb forms, but the use of the optative in the language is marginal at best, and it has been shown for a number of Indo- European languages that past tense forms are generally less frequent than present tense forms (Greenberg 1966, Haspelmath 2008). This observation is very likely to carry over to Albanian, where the aorist competes with two other past tenses, the imperfect and the perfect. Sometimes ‘‘unmarked form’’ is also interpreted in the sense of unaffixed form (Bybee 1985). Under this view, the present singular forms could be identified as bases for nondeponent verbs because, as argued in section 3, they bear no affixes; and the aorist 3sg might predict the correct stress pattern for deponents because it also bears no overt affix. However, it is at best unclear whether the aorist nonactive 3sg is really unaffixed in the relevant sense; this is because it is obligatorily accompanied by the nonactive clitic u, which has a rather affix-like behavior (it always occurs adjacent to the verb or to auxiliaries that in turn must be adjacent to the verb, and in postverb position it precedes the inflectional 2pl suffix -ni; as shown above, it is also in complementary distribution with the affixal realization of nonactive in the present and imperfect of the indicative). Moreover, taking unaffixed forms as bases raises the question of why Albanian does not choose other affixless forms as bases for the active of nondeponent verbs—especially the imperative, which would predict different stress patterns (recall formo!, which should have initial stress). Finally, Albright (2002) has shown that crosslinguistically, it is not generally the case that inflectional bases are unaffixed forms.

Basehood by predictability: Using historical evidence, Albright (2002) argues that the base of an inflectional paradigm is the form that allows the best-informed predictions for the shape of other forms in the paradigm. For example, he claims that the base for verb inflection in Yiddish is the present 1sg because it exhibits many contrasts that are neutralized in other forms. For example, 1sg forms maintain stem-final voicing contrasts (e.g., lib ‘I love’ vs. zip ‘I sift’), which are neutralized in the 2sg, 3sg, and 2pl by voicing assimilation to following obstruents (e.g., lipt ‘he loves’, zip-t ‘he sifts’). As an example of morphological informativeness that determines inflectional basehood, he adduces the nominal paradigm in Latin, for which he shows that oblique forms exhibit distinctions between arbitrary declension classes that are not visible in the nominative singular (e.g., corp-oris, genitive of corpus ‘body’; domin-i, genitive of dominus ‘lord’).

Many varieties of Standard Albanian also exhibit assimilation of voiced obstruents to following voiceless obstruents and final devoicing of obstruents at word boundaries (e.g., dri[ð]-ni, Prs.2pl vs. dri[θ]-te, Imf.3sg and dri[θ], Prs.3sg, ‘make tremble’). Hence, aorist 2sg/aorist 1pl/ aorist 2pl/aorist 3pl would make good bases for deponent verbs (e.g., u dri[ð]e, Aor.2sg) because here the voicing of the final obstruent is distinctive, while it is neutralized in the optative forms (e.g., dri[θ]-të, Opt.3sg). However, under this criterion most of the present and imperfect forms would also make perfect bases because the stem-final consonant in these forms is also intervocalic (e.g., dri[ð]et, Prs.3sg) and voicing distinctions are maintained. However, these potential bases would receive final, hence post-stem, stress.

There are other morphophonological alternations that might also be relevant for predicting forms in Albanian. Since Albright’s conception of predictability is based on effectiveness over the entire morphology of the language, making crucial reference to the frequency of specific patterns, here I consider only alternations that occur in relatively productive patterns, that is, verb classes comprising 20 or more verbs.

As discussed above, the final stem vowel of class I verbs and many related minor verb classes alternates between a diphthong and a monophthong variant. For many forms, this leads to a certain indeterminacy in class membership. For example, verbs that inflect like shkruaj ‘to write’ have exactly the same distribution of o and ua in aorist and past forms as standard class I verbs, but differ in present tense and imperfect forms. Similar pairs of classes exist for the monophthong/diphthong alternation e/ye and i/ie (the parts of the paradigm that do not allow verb classes to be distinguished are shaded in (43) and (44)).

(43) Diphthong/Monophthong alternations in verbs

graphic

Crucially, for (43a) and (43b) present indicative (active or nonactive) forms are much better indicators of classhood than aorist or optative forms. Thus, from the present nonactive 3sg it can be inferred with high security that a verb ending in -ohet is class I and a verb ending in -uhet is of the shkruaj class, whereas the aorist nonactive 3sg is compatible with both classes. Another possibly relevant pattern is the alternation between s and t in specific consonant-final verb classes. For example, mat takes t throughout, but zgjat replaces stem-final t by s in specific forms. Here, the present active makes a good predictor for classhood but neither optative, aorist, nor present nonactive forms do.

(44) s/t-alternations in verbs

graphic

Considering all three processes (voicing, diphthong/monophthong, and s/t-alternations), the predictability account partially is consistent with the bases considered in (42) (though it does not uniquely designate them above other potential bases), and partially disfavors them. I conclude that this account also does not motivate the correct bases for stress assignment in Albanian deponent verbs. Since the predictions of an asymmetric paradigmatic account of uniformity crucially depend on a principled selection of bases, the Albanian data are highly problematic for such an account.

5 Noun Stress Cannot Be Analyzed Paradigmatically: Defective and Stress-Shifting Nouns

Albanian nouns pose substantial problems for a paradigmatic analysis, too. Pluralia tantum lead to complications similar to those of deponent verbs since a considerable number of them lack singular base forms that would motivate the correct position of stress (section 5.1). Stress-shifting nouns have singular forms, but singular forms with different stems that fail to predict the correct stress for plural forms (section 5.2). Again, a stratal analysis based on morphological stems provides a unified account of all stress patterns.

5.1 Pluralia Tantum: Paradigmatically Defective Nouns

Like many other languages, Albanian has nouns that occur only in the plural, never in the singular: pluralia tantum. (45) shows some representative examples (Buchholz and Fiedler 1987, Agalliu et al. 2002).

(45) Pluralia tantum

graphic

All cases in (45) are equally unproblematic for either a stratal or a paradigmatic account. Under a stratal analysis of mustaqe, stress is assigned to the final syllable of the stem mustaq because stem-final closed syllables containing full vowels generally attract stress, which is then transmitted by faithfulness constraints to the plural form. Under a paradigmatic account, mustaqe has no base and receives penultimate stress because words ending in open mid vowels receive penultimate stress. What turns out to be problematic for the paradigmatic account are pluralia tantum that take the plural suffix -a.

(46) Pluralia tantum with suffix -a

graphic

These nouns have penultimate stress even though a in a word-final syllable attracts final stress in (nondefective) bare stems (e.g., ha.ˈta ‘doom’). Thus, penultimate stress in all forms in (46) is unexpected under a paradigmatic account. Under a stem-based account, this pattern follows directly from the analysis developed here, just as for nondefective plural forms: a stem such as pemurin is assigned final stress because the last syllable contains a full vowel and is closed; stress is then inherited by the inflected plural form. Again there are no alternative forms in the paradigm that would motivate the observed stress pattern without resorting to stems and strata. Consider for example the full paradigm in (47) for pemurina, where all forms have at least one variant that predicts stress on -a: nominative indefinite and accusative indefinite for the reasons already mentioned; ablative indefinite, nominative definite, and accusative definite because the final syllable is closed; and the t-less variants of dative definite and ablative definite because words ending in an open mid vowel receive penultimate stress. The dative/ablative variants with final t would predict stress in the syllable following -a, which is closed.

(47) Exhaustive inflectional paradigm of pe.mu.ri-na ‘fruit’

graphic

Crucially, no form in the paradigm predicts stress on the syllable preceding the plural marker. An anonymous reviewer suggests that the stress assignment in defective nouns might be captured by assuming a high-ranked morphophonological constraint (say, *STRESS/PLURAL) that blocks positioning stress on plural affixes. Although this would predict the correct outputs for pluralia tantum, it would do so at the cost of a suspicious constraint conspiracy (see Kisseberth 1970, Kager 1999, Pater 1999, on rule conspiracies). Whereas HEAD-MATCH directly derives the fact that plural affixes (and inflectional affixes in general) in the nominal paradigm are never stressed from the fact that the base is a form without overt inflection, *STRESS/PLURAL would ensure the same outcome for pluralia tantum by a morpheme-specific constraint, clearly missing the completely general uniformity principle governing the morphology-phonology interface in Albanian stress assignment.

5.2 Stress-Shifting Nouns

The final evidence against an output-output account and for a stratal approach to stress uniformity in Albanian comes from a small set of irregular nouns that—together with a handful of adjectives exhibiting similar patterns—form the only exception to the statement that stress is immovable across inflectional paradigms. (48) gives some representative examples.

(48) Stress-shifting nouns

graphic

The crucial observation is that these nouns are exceptional both morphologically and phonologically. They are exceptional phonologically because they violate the uniformity of stress position that is pervasive in regular nouns. They are irregular morphologically in the sense that none of the affixes used in the plural forms is productive, and plural suffixes generally do not replace full stem-final vowels (compare (48a) with qershi (sg)/qershi-a ( pl) ‘cherry/cherries’). Ideally, a single explanation should account for both types of irregularity in a unified way.

5.2.1 Problems for a Paradigmatic Approach

An output-output analysis could account for the cases where the plural suffixes replace the stressed part of the singular form, since suppression of the corresponding stressed vowel makes satisfaction of HEAD-MATCH impossible and thus ‘‘frees’’ the plural form to choose a purely phonologically motivated stress pattern. Thus, assuming that nje.ˈri receives final stress by regular phonological stress assignment and that high-ranked IDENT blocks correspondence between the ë in the plural and the i in the singular (49c), the candidates with initial (49a) and final (49b) stress both violate because the stressed vowels do not correspond. (49a) wins since the phonological constraints active in Albanian favor penultimate stress for a word with a final schwa syllable.

(49)

graphic

However, this analysis does not extend to (48c–e) because for these forms the singular stress could be transferred to the plural without any independent faithfulness violations. This is illustrated for (48e) in (50). The candidate with initial stress satisfies both IDENT and , hence is incorrectly predicted to become optimal.

(50)

graphic

5.2.2 A Stratal Analysis

A stratal analysis of irregular stress-shifting nouns is possible if we assume that they exhibit stem suppletion, hence that the irregular plural forms contain an irregular stem followed by a plural affix as in (51).

(51) Stress-shifting nouns

graphic

Under this morphological segmentation, stress assignment is straightforward. Kalli receives final stress at the stem level because word-final i attracts stress, and kallëz receives stress on the first syllable because word-final schwa syllables trigger penultimate stress (see section 2.1). Finally, as argued in section 2.3, stem-level stress is retained at the word level, resulting in kaˈlli and ˈkallëza. -a is one of the productive plural affixes in Albanian (e.g., shtëpi/shtëpi-a ‘house(s)’, kumbull/kumbull-a ‘cucumber(s)’) and is common for noun stems ending in ëz (e.g., lidhëz ‘fiber’, mollëz ‘cheek’); hence, the only idiosyncratic fact about kalli is that the stem shape varies in the singular and the plural. The phonological nonuniformity of stress follows straightforwardly from this morphological anomaly. All plural suffixes in (51) are independently attested in other nouns without suppletion and stress shift: - (e.g., ergjëz ‘young louse (sg/pl)’, punonjës ‘worker (sg/ pl)’) and -ë (e.g., detar (sg) – detar-ë ( pl) ‘sailor(s)’, tipar (sg) – tipar-ë (pl) ‘feature(s)’) are quite productive, while -j is restricted to a small number of items (e.g., kufi (sg) – kufi-j (pl) ‘border(s)’, kalama (sg) – kalama-j ( pl) ‘child(ren)’).

At this point, it is important to highlight why a suppletion analysis cannot be simply transferred to a paradigmatic approach. In an output-output analysis, faithfulness constraints are essentially blind to the derivational history of inflected forms. Thus, in (50) is predicted to compare stress in singular and plural forms by virtue of their paradigmatic relatedness independently of whether they share a stem or not. Of course, one could posit a principle to the effect that forms with different suppletive constituents are exempt from output-output faithfulness, but this would stipulate a mechanism that follows in the stratal account from the architecture of the grammar itself: stress in the inflected form of a suppletive stem maintains the stress of the stem because the suppletive stem is the input of the form. In other words, the stratal analysis only has to assume the virtually obvious morphological suppletion; the paradigmatic approach must assume suppletion and a special relation between suppletion and output-output faithfulness. Moreover, a principle that would block under suppletion would incorrectly predict that the plural of (48b) kallëza should receive final stress since word-final a regularly attracts stress in base forms.

Note finally that there are interesting subpatterns in stem formation that seem to involve further internal structure and show the independence of stem formation patterns and plural inflection. Thus, njeri and kalli show exactly the same alternation if the variable stem-final part of the forms is separated from the invariable initial stem portion and the plural affixes.

(52) Morphological structure of kalli/kallëza and njeri/njerëz

graphic

This might well motivate a further decomposition of irregular stems into roots and stem extensions, which would not change the phonological analysis of stress placement. Crucially, the suppletion account of morphological irregularity also accounts for the observed stress shift without any additional stipulation.15

6 Paradigm Uniformity in an Optimal Paradigms Account

A version of output-output faithfulness theory that radically eliminates the search for the correct base in forms related by inflectional processes is Optimal Paradigms Theory (McCarthy 2005; OP). In OP, it is not single inflectional forms, but entire paradigms, that are evaluated. Output-output faithfulness constraints hold between all members of a paradigm in a completely symmetric way. This offers an attractive approach to stress uniformity in deponent verbs (section 6.1), but proves problematic for capturing differences between nouns and verbs (section 6.2) and for capturing the nominal paradigm itself (section 6.3). In section 6.4, I discuss a variant of OP developed by Cable (2004), which solves some of these problems, but still makes the wrong predictions for stress-shifting nouns and pluralia tantum.

6.1 An Optimal Paradigms Account of Deponent Verbs

Here is how an OP analysis for deponent verbs might work. Assume that the paradigm of pendohem consists just of the aorist 2sg u pendove, the optative 2sg u pendofsh (forms that predict stemfinal stress), and the present 3sg pendohet (a form that predicts word-final stress). These forms are phonotactically representative for the entire paradigm, instantiating nonsyllabic consonantfinal (-fsh), syllabic consonant-final (-het), and syllabic vowel-final (-ve) shapes of suffix sequences. The OP evaluation now uses as candidates miniparadigms of prosodifications for u pendove, u pendofsh, and pendohet as in (53). If we replace by and leave the ranking of the other constraints intact, stem-final stress follows since eliminates all candidates where the forms of the paradigm have stress on noncorresponding vowels (see (53c)). Effectively, only those candidates survive where stress consistently falls on a specific stem syllable (see (53a–b)), and (53b) is eliminated because it violates ALL-FT-RT. Note that ALL-FT-RT in (53b) is violated twice because the violations for pendohet and u pendove are summed. Similarly, is violated four times in (53c), once for every ordered pair of noncorresponding vowels in different forms: ‹u.pen.(ˈdo.ve), pen.do.(ˈhet)›, ‹pen.do.(ˈhet), u.pen.(ˈdo.ve)›, ‹u.pen.(ˈdofsh), pen.do.(ˈhet)›, and ‹pen.do.(ˈhet), u.pen.(ˈdofsh)›.

(53) Input: [u pendove − u pendofsh − pendohet]

graphic

6.2 Problems Facing Optimal Paradigms: Differences between Nouns and Verbs

Obviously, this analysis derives stem-final stress for the complete paradigm of pendohem (and all other deponent and nondeponent verbs) because a paradigmatically consistent rightmost foot will always have its head on the stem-final syllable. However, this success with verbs turns out to be problematic for a unified account of nouns and verbs. The analysis incorrectly predicts that nouns whose stem ends in a mid vowel should also have stem-final stress (54a), since forms with a syllabic affix such as babo-je ‘midwife’ (Dat/Abl.ind.sg; babo-n is accusative definite singular) violate ALL-FT-RT if the penultimate stem syllable is stressed (the head of the foot), as in (54b).

(54) Input: [babo − babon − baboje]

graphic

It might seem that the OP analysis can be saved by ranking FT-BIN above ALL-FT-RT as in (55).

(55) Input: [babo – babon – baboje]

graphic

However, this leads to wrong predictions for verbs. Thus, for a nondeponent verb such as kaloj ‘to pass’, the ranking in (55) incorrectly predicts penultimate stress since the imperative active singular kalo! (see section 3 for more discussion of the imperative) has essentially the same phonological shape as babo (kalon is present 2sg/3sg, and kaloje imperfect 2sg).

(56) Input: [kalo – kalon – kaloje]

graphic

The problem for OP is that the paradigms of babo and kalon are completely isomorphic with respect to the shape of their output forms. They both have bisyllabic vowel-final forms (ba.bo, ka.lo), bisyllabic consonant-final forms (ba.bon, ka.lon), and trisyllabic vowel-final forms (ba.- bo.je, ka.lo.je). The relevant difference between nouns and verbs is basehood: the base of the nominal paradigm is vowel-final (the nominative indefinite singular babo), whereas the base of the verbal paradigm is consonant-final (the present indicative 3sg kalon). In terms of an asymmetric output-output approach (see section 2.4), stress placement is computed locally for these base forms and transmitted by output-output faithfulness to the entire paradigm. Similarly, in the stratal analysis developed in section 2.3, stress is computed locally for the corresponding stems babo and kalon and maintained at later cycles. However, the essential claim of OP is that faithfulness constraints in inflectional paradigms do not refer to stems or to morphologically designated base forms. Constraint evaluation is global for the entire paradigm and symmetric (treating all forms in the paradigm on a par). Thus, it cannot reconstruct the insight of asymmetric paradigmatic and stratal approaches that stress in nouns and verbs depends in a uniform way on morphological base forms (see Hall 2007 and Albright 2008b for related problems facing OP in other empirical domains).

This problem for OP is not due to the restricted choice of forms used in the paradigm tableaux in (54)–(56) or to the specific phonological constraints chosen. Thus, for every verb form of a class I (hence mid-vowel-final) verb whose stem-final and post-stem-final syllable has a specific phonological shape, there is a form of a mid-vowel-final noun that has the same shape. Nonetheless, the noun systematically has different stress.

(57) Phonological shapes of noun and verb forms

graphic

If an OP ranking of FT-BIN and ALL-FT-RT (or any other markedness constraints) enforces stemfinal stress on any of the verb forms in (57), which is then transmitted to the whole paradigm, this ranking will also enforce stem-final stress, incorrectly, on the phonologically corresponding noun form and the nominal paradigm. Thus, the inability of OP to account for the difference between nouns and verbs is independent of the specific constraints that are invoked.

6.3 Problems Facing Optimal Paradigms: Nouns

As an anonymous reviewer remarks, the minimal pairs in (57) would still be consistent with an OP approach under the assumption that nominal and verbal paradigms in Albanian establish distinct phonological subsystems with different constraint rankings (or lexically indexed constraints targeting nominal and verbal forms differentially; see Smith 2011 for discussion). Here, I will show that there is also no consistent OP analysis that would account for all nominal forms. Note, however, that the major piece of evidence that has been adduced for OP in the literature is its ability to derive the distinct phonological behavior of nouns and verbs in specific languages without additional stipulations (Cable 2004, McCarthy 2005). Thus, the problems facing OP encountered up to this point suffice to establish the failure of OP to achieve its original research program for Albanian stress, whereas the stratal analysis succeeds (see Bobaljik 2008 for arguments that Arabic and Itelmen also raise similar problems for OP).

6.3.1 Regular Nouns

Consider first nondefective nonsuppletive nouns, which induce a ranking paradox for FT-BIN and ALL-FT-RT similar to the one we have already seen. Since stress for nonsuppletive stems is perfectly invariant across the entire paradigm, must be undominated, and I will consider here only paradigm candidates without stress shift. Now, to capture the correct stress position for a noun with a final mid vowel, FT-BIN must dominate ALL-FT-RT, as in (58). The opposite ranking would lead to consistent stress on the final vowel, as in (59), since this ranking permits feet that align perfectly with the right word edge in affixed forms (59a), an option excluded for the stress pattern in (59b) under the general ban on ternary feet in Albanian and probably crosslinguistically.

(58) Input: [babo − baboje]

graphic

(59) Input: [babo − baboje]

graphic

However, this leads to a paradox with nouns ending in a nonmid vowel and a penultimate closed syllable such as qerˈshi ‘cherry’ (which has essentially the same inflectional suffixes as babo; see (19)). For these nouns, the ranking necessary for final mid-vowel stems, (58), would incorrectly predict that stress consistently falls on the penultimate syllable of the base, as in (60b), which satisfies FT-BIN in all forms either because this syllable is closed (in the form without overt inflection) or because it is followed by a syllabic suffix.

(60) Input: [qershi − qershie]

graphic

The reason why OP runs into trouble with nouns is again that evaluation is global and symmetric. In the stratal analysis sketched in section 2.3, ALL-FT-RT holds without exceptions at the stem level as in (babo)F and qer(shi)F (and dominates FT-BIN as well as ), although it is violated massively by forms undergoing further affixation at the word level owing to higher-ranked {IMAGE}(e.g., (ba.bo)je)). In other words, violations of ALL-FT-RT are systematically banned for stems—word forms without overt inflection—but tolerated for overtly inflected words, violating it by virtue of adding inflectional affixes. Asymmetric paradigmatic accounts (see section 2.4) can derive this asymmetry from the assumption that base forms are not subject to faithfulness constraints. However, in OP, there is no such asymmetry. If ALL-FT-RT is ranked above FT-BIN as in (59), its effect cannot be restricted to forms without overt inflection such as babo and qershi; it will also maximize rightward stress for affixed forms such as baboje, predicting the incorrect stress pattern for the entire paradigm.

6.3.2 Pluralia Tantum

The problems that pluralia tantum nouns raise for asymmetric approaches fully transfer to OP: a noun such as pe.mu.ˈri.n-a is expected to show stress on final a since all forms of its paradigm motivate stress on post-stem syllables according to the phonological constraints on Albanian stress laid out in section 2.1. That this is not the case is shown by the paradigm in (61), repeated from (47).

(61) Exhaustive inflectional paradigm of pe.mu.ri-na ‘fruit’

graphic

The forms ending in -a (nominative/accusative indefinite) should have final stress, like all bare nouns ending in nonmid vowels (e.g., ha.ˈta ‘doom’), and similarly the (nominative/accusative definite, ablative indefinite) forms ending in -aC ( parallel to all bare nouns with final full vowel + C; e.g., sa.ˈhat ‘hour’) and the forms in -ave (dative/ablative definite), following the general rule for mid-vowel-final nouns (e.g., rru.ˈza.re ‘rosary’). Finally, the variants in -avet would motivate stress on e (e.g., a.ˈdet ‘habit’). Thus, given high-ranked and ALL-FT-RT, an OP analysis incorrectly predicts consistent stress on a, which is the rightmost vowel present in all forms of the paradigm, and at the same time the one that predicts the least marked stress pattern for most of the forms.

6.3.3 Stress-Shifting Nouns

OP also fully inherits the problems that stress-shifting suppletive nouns such as lumë (sg)/lu.ˈmenj ( pl) ‘river(s)’ pose for asymmetric output-output approaches. Consider the full paradigm of lumë in (62).

(62) Exhaustive inflectional paradigm of lumë ‘river’

graphic

The only syllable nucleus present in the entire paradigm is the u of the stem. This predicts—under undominated —that this vowel should be consistently stressed, a prediction borne out only by the plural forms that exhibit stress shift.

6.4 Cable’s (2004) Variant of Optimal Paradigms Theory

Cable (2004) proposes a variant of OP that potentially resolves some of the problems with the Albanian data. While McCarthy (2005) assumes that symmetric OP correspondence governs inflectional paradigms in general, Cable (2004) suggests that symmetric evaluation of paradigms as in OP holds only for lexemes that do not have roots occurring as free forms, hence for bound roots such as pendo in pendohem. Lexemes that have bases by virtue of occurring unaffixed in some part of their paradigm (e.g., babo) undergo asymmetric evaluation as in Benua’s (1997) model:

[W]henever the set of words containing a given root lacks an independent base, the O-O relation holding between them ‘‘goes symmetric.’’ Since a privileged member of the set cannot be found, all forms come to influence one another equally. On the other hand, if a base for the set can be found—a word whose phonology forms a part of all other forms in the set—the correspondence relation is the asymmetric Base-Identity relation, facilitating base-derivative correspondence. (Cable 2004:11)

In this system, OP faithfulness constraints such as apply only to paradigms of lexemes that do not have a base in the relevant sense. Thus, assuming that is undominated in Albanian and ranked above the subranking in (21), the evaluation of pendohem and its paradigm would be as in (53) in OP style, whereas the paradigm of babo would be computed by the recursive constraint evaluation demonstrated in section 2.4 (see tableaux (22) and (23)).

It is not obvious what predictions Cable’s proposal would make for nondeponent verbs since the predictions would depend on subtle details of analysis. To be sure, in Cable’s system, a base for a paradigm is not determined by the relatedness of the base form and the other word forms of the paradigm in terms of morphological derivation; rather, it is defined as an input word form that is properly contained in (a substring of ) all other input word forms of this paradigm (Cable 2004:12).16 For most class I verbs, there is a word form that is a likely candidate for meeting Cable’s definition of a base: the imperative 2sg (e.g., kalo! for kalon) that is contained in the bulk of surface forms. The only potentially doubtful cases are forms that exhibit diphthongization (e.g., the participle kaluar and specific aorist forms such as the active 1pl kaluam; also see section 4.2). Diphthongization is completely exceptionless for class I verbs in the relevant forms and also applies to stems ending in e, resulting in ye (e.g., rrëfen ‘to tell’ has the participle rrëfyer and the aorist active 1pl rrëfyem). This alternation could be captured either by assuming that diphthongizing affixes such as the participle suffix -r carry floating phonological material (e.g., the feature [+high]) that triggers diphthongization on roots ending in o/e or by assuming a morpholexical rule of diphthongization. If kalua is the result of affixation to underlying kalo(j), the imperative is a perfect base in Cable’s sense, and the computation of stress should be asymmetrically based on the output form kalo, which predicts incorrect penultimate stress (ˈka.lo in parallel to the noun ˈba.bo). However, if kalua is directly generated by a morphological process (and hence already present in the input to phonology), the imperative is not contained in the inputs to ua-final forms, hence not an asymmetric base, and stress should be computed as for deponent verbs, predicting correct stem-final position.

On the other hand, Cable’s approach clearly leads to wrong predictions for Albanian nouns. Thus, the nominative/accusative of a plurale tantum such as pe.mu.ˈri.na is straightforwardly contained in all forms of its paradigm (see the table in (61)) and should be the asymmetric base of the remaining forms. Since pemurina ends in a nonmid vowel, stress is again incorrectly predicted to be located on the final syllable of the base, which is then transferred to the entire paradigm, as discussed in section 5.1.

For suppletive nouns, Cable’s system predicts different routes for deriving stress in nouns such as at in (63f), where the nominative/accusative singular is a proper substring of the corresponding plural, and in nouns where this is not the case such as ˈdra.për in (63d) ((63) repeats the data from (51)).

(63) Stress-shifting nouns

graphic

Since case morphology in Albanian is strictly concatenative (singular forms add suffixes to singular stems, and plural forms to plural stems (see the paradigms in (18), (19), and (62)), a nominative singular that is contained in the nominative plural is properly contained in all forms of the corresponding paradigm and is therefore an asymmetric base in Cable’s sense. Stress assignment for at (63f ) should therefore be computed locally, with main stress on a, and then transferred by

graphic
to all forms of the paradigm, resulting in incorrect *ˈat.lla.rë. Conversely, the paradigm of nje.ri does not have an asymmetric base: the only substring that is contained in all of its forms is njer, which does not correspond to any form of the paradigm. Thus, stress for its paradigm is purely sensitive to
graphic
. Since this constraint is undominated, we again expect, incorrectly, consistent stress on the rightmost vowel, which is shared by the whole paradigm (*ˈnje.rinje.rëz). Thus, under Cable’s approach, both types of output-output faithfulness lead to wrong results for different suppletive nouns.

Note finally that a potential theoretical proviso exempting suppletive stems from outputoutput faithfulness would be especially problematic for Cable’s approach. As the discussion of verbs has shown, the assumption that class I verbs have alternative stems provided by morphology is the only way to ensure that these verbs undergo symmetric OP-style constraint evaluation. If the existence of alternative morphologically conditioned input stems were to block application of

graphic
, verbs would be predicted to undergo asymmetric output-output faithfulness, resulting in the familiar problem that the verb kalo should have penultimate stress just like the noun ˈba.bo.

7 Summary

Paradigmatic and stratal approaches to the morphology-phonology interface provide closely related accounts of the fact that morphologically related word forms often exhibit phonologically unexpected uniformity. Conceptual claims for the superiority of one approach over the other have proven to be problematic (Bermúdez-Otero 2008, Collie 2007), but the two approaches make substantially different empirical predictions. Under the paradigmatic account, uniformity should be sensitive to the concrete shape of paradigms. Under the stratal approach, only morphological constituency should be relevant. In this article, I have shown that stress uniformity in Albanian favors the latter view: uniformity remains intact with uniform stems even though the relevant paradigms are defective (sections 4 and 5.1); uniformity collapses if stems are nonuniform even though the paradigms are intact (section 5.2). Apparent evidence for the decisive role of paradigms turns out to be illusory (section 3). Adopting a fully symmetric model of paradigm uniformity as in OP solves some of the difficulties of an output-output approach, but leads immediately to empirical problems with morphologically designated bases (section 6). The important methodological point revealed by the Albanian data is that morphological details, especially the morphemic segmentation of inflectional forms, turns out to be crucial for evaluating phonological analyses.

Notes

I am grateful to Ricardo Bermúdez-Otero, Jonathan Bobaljik, Dalina Kallulli, Marc van Oostendorp, and the audience of the GLOW workshop ‘‘Approaches to Phonological Opacity’’ in Barcelona (5 April 2006, GLOW 29) for helpful discussion. All remaining errors are my own.

1 Throughout the article, I will cite Albanian data in Standard Albanian orthography, where ë stands for the central mid vowel [ə] (‘‘schwa’’), ç for [tʃ], r for [ɾ], x for [dz], and the following digraphs are used: dh = [ð], gj = [ɉ], ll = [ɫ], nj = [ɲ], rr = [r], sh = [ʃ], th = [θ], xh = [ʤ], zh = [ӡ]. All other letters and the marking of stress correspond to their usage in IPA.

The following abbreviations are used for grammatical categories: abl = ablative, acc = accusative, act = active, aor = aorist tense, dat = dative, def = definite, imf = imperfect tense, imp = imperative, ind = indicative (for verbs) or indefinite (for nouns), nom = nominative, nac = nonactive, opt = optative, pl = plural, prs = present tense, ptc = participle, sbj = subjunctive, sg = singular, 1 = 1st person, 2 = 2nd person, 3 = 3rd person.

2 I use the term base form in this article as a descriptive label for word forms without overt inflection that are phonologically identical to the stem of the form. Technically, this corresponds to the base forms of inflectional paradigms in asymmetric output-output approaches to opacity (see section 2.4), but to stems in stratal approaches, which deny the relevance of paradigms for phonological evaluation (see section 2.3).

3anë ‘side’, hënë ‘moon’, ehë ‘blade’, pronë ‘possession’, sivë ‘gray’, faqe ‘face’, ëndje ‘appetite’, ende ‘flower’, golle ‘hole’, fice ‘overripe’, babo ‘midwife’, neto ‘net’, loço ‘featherbrained’, birko ‘excellent’, hata ‘calamity’, pastërma ‘salted smoked meat’, xhela ‘shoe polish’, otra ‘bootlace’, rixha ‘prayer’, bari ‘lawn’, gjuhësi ‘linguistics’, qershi ‘cherry’, komshi ‘neighbor’, zili ‘envy’, ashtu ‘this way’, akëku ‘here and there’

4Rosenthall and Van der Hulst (1999) write this constraint *μ/CONS.

5 Strictly speaking, Benua (1997) does not assume sequentially ordered evaluations of base and derived form; rather, she assumes simultaneous evaluation of a paradigm comprising both forms, where base and derived form are ultimately evaluated against different copies of the same constraint hierarchy such that the constraints relevant for the base are ranked higher than those for the derived form. Since this formalization can be straightforwardly translated into a model where evaluation of the derived form follows evaluation of the base for the same constraint hierarchy, I will stay with the simpler version here.

6 See Bermúdez-Otero 2008 for a detailed analysis of the structural similarities between Stratal OT and the output-output model.

7 The analysis sketched here follows the spirit of, but differs in several noncrucial details from, the one proposed in Trommer 1997.

8 The different behavior of ha cannot be reduced to the different root vowel; compare la-j ‘to wash’, which basically inflects like formo-j.

9 There are different ways to implement rhythmic schwa epenthesis. The most straightforward interpretation in a stratal architecture is to analyze it as a strategy for satisfying foot binarity that is only available at the word level owing to the reranking of faithfulness constraints with respect to the stem level (qer.(ˈshiμ.μ) would be favored over qer.(ˈshiμn) and qer.(ˈshiμnμ) because it is the only possibility for forming a binary foot without adding a mora to the underlyingly nonmoraic n). Additionally, faithfulness constraints on moraic weight would also ensure that stem-final heavy monosyllabic feet are not altered by epenthesis (stem-level pe.(ˈliμnμ) is maintained instead of being changed to pe.(ˈlim.m)).

10 As an anonymous reviewer points out, the absence of j in many imperative forms is also consistent with analyzing j as a thematic stem formative, given the strong crosslinguistic tendency for imperatives to consist of bare roots (see Jakobson 1985:136).

11 A similar argument against output-output approaches is made by Bobaljik (2008) for Itelmen.

12 I am not considering analytic verb forms that are formed by adding particles to synthetic forms (e.g., the subjunctive) or combine auxiliaries and participles (e.g., the perfect and the admirative). Under this selection, Albanian past tense forms can only have indicative mood, and optative and imperative are always present tense.

13 For nonbase forms with different syllabic affixes, it is unclear to me what the prediction would be. Thus, it depends on the ranking of output-output faithfulness constraints whether [i] in the admirative 1pl affix -im of the optative 1pl u pen.dof.shim corresponds to the vowel of the nonactive affix -he in the base *pen.do.(ˈheshe). If there is no output-output coindexation between the stressed vowel of *pen.do.(ˈheshe) and a vowel in u pen.dof.shim, every stress position in the latter will violate , the constraint will therefore be ineffective, and the predicted stress position will be word-final by the general phonological preferences of Albanian.

14 For pendohem, there are no forms without a syllabic suffix where the final stem vowel is a monophthong.

15 An anonymous reviewer suggests another plausible subanalysis of stems that is consistent with the overall analysis developed here: i and ëz could be irregular plural suffixes since irregular inflectional affixes show a well-known tendency to behave phonologically as stem-level. Compare, for example, the irregular English past tense suffix -t, which causes stem-level closed-syllable shortening (as in dr[ε]m-t) in contrast to the regular past tense allomorph -d (dr[i∶]m-ed).

16 Proper containment is defined in terms of inputs (underlying forms) since otherwise, basehood would be influenced in an erratic way by regular phonological processes.

References

Agalliu
,
Fatmir
,
Engjëll
Angoni
,
Shaban
Demiraj
,
Ali
Dhrimo
,
Enver
Hysa
,
Emil
Lafe
, and
Ethem
Likaj
.
2002
.
Gramatika e gjuhës Shqipe. Vol. 1, Morfologjia
.
Tirana
:
Botimi i Akademisë së Shkencave
.
Albright
,
Adam
.
2002
.
The identification of bases in morphological paradigms
.
Doctoral dissertation, UCLA, Los Angeles, CA
.
Albright
,
Adam
.
2008a
.
How many grammars am I holding up? Discovering phonological differences between word classes
. In
Proceedings of the 26th West Coast Conference on Formal Linguistics
, ed. by
Charles B.
Chang
and
Hannah J.
Haynie
,
1
20
.
Somerville, MA
:
Cascadilla Proceedings Project
.
Albright
,
Adam
.
2008b
.
Inflectional paradigms have bases too: Evidence from Yiddish
. In
Inflectional identity
, ed. by
Asaf
Bachrach
and
Andrew
Nevins
,
271
312
.
Oxford
:
Oxford University Press
.
Benua
,
Laura
.
1997
.
Transderivational identity: Phonological relations between words
.
Doctoral dissertation, University of Massachusetts, Amherst
.
Bermúdez-Otero
,
Ricardo
.
2007
.
Morphological structure and phonological domains in Spanish denominal derivation
. In
Optimality-theoretic studies in Spanish phonology
, ed. by
Fernando
Martínez-Gil
and
Sonia
Colina
,
278
311
.
Amsterdam
:
John Benjamins
.
Bermúdez-Otero
,
Ricardo
.
2008
.
Stratal Optimality Theory
.
Ms., University of Manchester
.
Bermúdez-Otero
,
Ricardo
.
2010
.
Cyclicity
. In
The Blackwell companion to phonology. Vol. 4, Phonological interfaces
, ed. by
Marc
van Oostendorp
,
Colin J.
Ewen
,
Elizabeth
Hume
, and
Keren
Rice
,
2019
2048
.
Malden, MA
:
Wiley-Blackwell
.
Bevington
,
G. L
.
1974
.
Albanian phonology
.
Wiesbaden
:
Otto Harrassowitz
.
Bobaljik
,
Jonathan David
.
2008
.
Paradigms, optimal and otherwise: A case for skepticism
. In
Inflectional identity
, ed. by
Asaf
Bachrach
and
Andrew Ira
Nevins
,
29
44
.
Oxford
:
Oxford University Press
.
Buchholz
,
Oda
, and
Wilfried
Fiedler
.
1987
.
Albanische Grammatik
.
Leipzig
:
VEB Verlag Enzyklopädie
.
Bybee
,
Joan L
.
1985
.
Morphology
.
Amsterdam
:
John Benjamins
.
Cable
,
Seth
.
2004
.
Phonological noun-verb dissimilarities in Optimal Paradigms
.
Ms., MIT, Cambridge, MA. Available at http://people.umass.edu/scable/papers/Phon-N-V-Dissim.pdf
.
Camaj
,
Martin
.
1984
.
Albanian grammar
.
Wiesbaden
:
Otto Harrassowitz
.
Collie
,
Sarah
.
2007
.
English stress-preservation and Stratal Optimality Theory
.
Doctoral dissertation, University of Edinburgh
.
Downing
,
Laura
,
Tracy Alan
Hall
, and
Renate
Raffelsiefen
.
2005
.
Introduction: The role of paradigms in phonological theory
. In
Paradigms in phonological theory
, ed. by
Laura
Downing
,
Tracy Alan
Hall
, and
Renate
Raffelsiefen
,
1
16
.
Oxford
:
Oxford University Press
.
Greenberg
,
Joseph H
.
1966
.
Language universals
.
The Hague
:
Mouton
.
Hall
,
Tracy Alan
.
2007
.
Comparative markedness vs. standard faithfulness theory: A typological comparison
.
Ms., Indiana University, Bloomington
.
Haspelmath
,
Martin
.
2008
.
Frequency vs. iconicity in explaining grammatical asymmetries
.
Cognitive Linguistics
19
:
1
33
.
Jakobson
,
Roman
.
1985
.
On aphasic disorders from a linguistic angle
. In
Roman Jakobson: Selected writings (vol. VII)
, ed. by
Stephen
Rudy
,
128
140
.
Berlin
:
Walter de Gruyter
.
Kager
,
René
.
1999
.
Optimality Theory
.
Cambridge
:
Cambridge University Press
.
Kallulli
,
Dalina
.
2006
.
A unified analysis of passives, anticausatives and reflexives
. In
Empirical issues in syntax and semantics: Proceedings of the Colloque de Syntaxe et Sémantique à Paris 6 (CSSP 2005)
, ed. by
Olivier
Bonami
and
Patricia Cabredo
Hofherr
,
201
225
. .
Kenstowicz
,
Michael
.
1996
.
Sonority-driven stress
.
Ms., MIT, Cambridge, MA. Rutgers Optimality Archive 33, roa.rutgers.edu
.
Kiparsky
,
Paul
.
2000
.
Opacity and cyclicity
.
The Linguistic Review
17
:
351
367
.
Kisseberth
,
Charles
.
1970
.
On the functional unity of phonological rules
.
Linguistic Inquiry
1
:
291
306
.
Mańczak
,
W
.
1958
.
Tendences générales des changements analogiques
.
Lingua
7
:
298
325
and
387
420
.
McCarthy
,
John
.
2000
.
The prosody of phase in Rotuman
.
Natural Language and Linguistic Theory
18
:
147
197
.
McCarthy
,
John
.
2005
.
Optimal Paradigms
. In
Paradigms in phonological theory
, ed. by
Laura
Downing
,
Tracy Alan
Hall
, and
Renate
Raffelsiefen
,
170
210
.
Oxford
:
Oxford University Press
.
McCarthy
,
John
.
2007
.
Hidden generalizations: Phonological opacity in Optimality Theory
.
London
:
Equinox
.
McCarthy
,
John
.
2010
.
An introduction to Harmonic Serialism
.
Ms., University of Massachusetts, Amherst
.
Morén
,
Bruce Timothy
.
1999
.
Distinctiveness, coercion and sonority: A unified theory of weight
.
Doctoral dissertation, University of Maryland, College Park
.
Newmark
,
Leonard
,
Philip
Hubbard
, and
Peter
Prifti
.
1982
.
Standard Albanian: A reference grammar for students
.
Stanford, CA
:
Stanford University Press
.
Pater
,
Joe
.
1999
.
Austronesian nasal substitution and other *NC̥ effects
. In
The prosody-morphology interface
, ed. by
René
Kager
,
Harry
van der Hulst
, and
Wim
Zonneveld
,
310
343
.
Cambridge
:
Cambridge University Press
.
Prince
,
Alan
, and
Paul
Smolensky
.
1993
.
Optimality Theory: Constraint interaction in generative grammar
.
Technical report, Rutgers University Center for Cognitive Science, New Brunswick, NJ
.
Rosenthall
,
Sam
, and
Harry
van der Hulst
.
1999
.
Weight-by-position by position
.
Natural Language and Linguistic Theory
17
:
499
540
.
Smith
,
Jennifer L
.
2011
.
Category-specific effects
. In
The Blackwell companion to phonology. Vol. 4, Phonological interfaces
, ed. by
Marc
van Oostendorp
,
Colin J.
Ewen
,
Elizabeth
Hume
, and
Keren
Rice
,
2439
2463
.
Malden, MA
:
Wiley-Blackwell
.
Trommer
,
Jochen
.
1997
.
Eine Theorie der albanischen Verbflexion in mo_lex
.
Master’s thesis, University of Osnabrück
.
Trommer
,
Jochen
.
To appear
.
Albanian word stress
.
Balkanistica
.