chapter 2violagmiglio.net/violas_site/papers_files/fincap2.doc · web viewthe distribution to be...

CHAPTER 2

Mantuan Vowel Deletion

2.1 Introduction

Mantuan is a dialect of Italian whose syllabic structure differs significantly from the standard: rather than favouring the Italian CV syllable, it privileges a CVC pattern. A re-ranking of certain constraints explains the difference in syllabic structure between the two languages, in particular, a constraint penalizing consonant clusters is found to be higher ranked in Italian than in Mantuan (see Appendix). Mantuan, on the other hand, has fewer syllables in words cognate with Italian words, often deleting unstressed vowels, which results in complicating syllable structure.

The first phenomenon to be explained is an alternation between the presence of a vowel in the final syllable of certain words, and its absence once a morpheme is added to them. The stress shifts from the syllable preceding the final syllable to the syllable following what used to be the final syllable: pádar - pad<>r+ón ‘’father - master’. The non-derived environments where the vowel appears have traditionally been analyzed as giving rise to epenthesis in related dialects. Both the deletion and the epenthesis solution in OT are evaluated. Different rankings are found to adjust either to the deletion hypothesis or to epenthesis: both are shown to be possible explanations for the alternations in this language.

However, deletion of vowels is proven not to be a uniform phenomenon, depending on the original position of the deleted vowel. Thus, “deletion” subsumes two different phenomena traditionally termed syncope (word-medial deletion) and apocope (word-final deletion). These two phenomena are caused by the ranking of different constraints.

The features described above make Mantuan a typical representative of the phonological characteristics of Northern Italian dialects, and the phenomena of

19

20 Markedness and Faithfulness in Vowel Systems

vowel deletion/epenthesis can be found in other languages of the world, as in Arabic dialects such as Maltese (Hume 1994).

2.2 Data on Syllable Structure

Observe the following data from standard Italian and their cognates in Northern Mantuan.

Gloss Italian Nth Mantuan1

Simple onsets/codas

peach pés.ka [email protected] [email protected] [email protected] die mo.rí.re [email protected] a.ni.má.le ni.málfriend a.mí.ko a.míkmouse sór.tS o só.rakold vE@@k.kjo vE@tSegg wçè.vo O@f

Complex Onsets bald pe.lá.to plákick pe.dá.ta psá.daready prón.to prúntpot pi.¯át.ta p¯á.tachewy té.ne.ro t¯Íswindow fi.nE€s.tra [email protected]

1 ?Northern Mantuan refers to the dialect spoken in the village of Commessaggio (ca. 1000 inhabitants) and the surrounding municipalities, which are administratively in the Mantua province, north of the Po river, but border on the Cremona province. Part of the Mantua province lies south of the Po, thus the Mantuan dialect spoken there will be referred to as "Southern Mantuan", even if in geographical terms, Western and Eastern Mantuan would perhaps be more correct. The difference between the two varieties is not relevant to syllable structure concerns, but it will become an issue in chapter 4. To facilitate reading, stress is marked directly on the vowel.

Mantuan Vowel Deletion 21

Complex Codas_______________________bald kál.vo kálfwarm kál.do káltdisease mç@r.bo mç@rbMarch már.tso márs

In comparison with standard Italian, Mantuan has a more a complex syllabic structure, favouring unusual consonant clusters, especially as onsets, and discouraging otherwise unmarked sequences of CV syllables.

Mantuan shows a high tolerance for word-final codas: by and large, all sonorants and all voiceless obstruents can be codas in Mantuan; this language does not support voice distinctions word-finally, similarly to German (cf. Lombardi 1995). Unlike Italian (cf. Appendix), Mantuan also has word-final complex codas, although they are somewhat restricted: no obstruent can be in the first position of a complex coda, only a sonorant can. This is to be expected, since Mantuan faithfully respects the sonority scale sequence (for sonority scale, cf. Clements 1990, Zec 1995). No affricates are allowed in complex codas either.

Particular attention must be paid to simplex and complex codas word-finally in Mantuan, since there is a correspondence between words that end in -(C)CV# in Italian, and in -(C)C# in Mn. It seems, therefore, that there cannot be a contrast between monomorphemic Mantuan words shaped C(C)VCC ~ C(C)VCCV (such as kard ~ kardu for instance). This is confirmed by the fact that whenever standard Italian words are imported into Mantuan and have that shape, the final vowel is always dropped. Compare Italian and Mantuan cognates in the following table:

(1) Word-final syllable structure correspondences: Italian ~ Mantuan


Standard Italian Mantuan Gloss

pár.ko párk “park”pár.te párt “part”mç@r.te mç@rt “death”ár.ko árk “bow”pE@ s.te pE@st “plague”mós.to móst “grape juice”Sé.mo sém “idiot”stú.pi.do [email protected] “fool”pE@@l.le pE@@l “skin”mar.tE@@l.lo mar.tE@l “hammer”

Therefore it is important to make sure that, whatever the input, either C(C)VCC# or C(C)VCCV#, the outcome is always C(C)VCC#. This phenomenon is both a diachronic and a synchronic characteristic of many Northern Italian dialects and is referred to as apocope.

Even in Mantuan, there are few word internal tautosyllabic clusters in codas (only in onsets, such as in pi.vrón), although it can allow different consonants in word-internal simplex codas (cf. (2) below). The distribution of word-internal codas in Italian allows for sonorants, /s/ (and [z]) (provided one accepts the analysis where /s/ is in the coda, cf. Morelli 1995), whereas obstruents can be found in a word-internal coda only when they are part of a geminate:

(2) Word-internal codas in Mn1) Nasals: /iM.fü.rjá/ "furious", /lám.pa.ra/ "lamp", /áN.ka/

"too", /án.dZul/ "angel", /kan.tón/ "corner";2) /r/: /dZur.ná.da/ "day", /ar.ti.t∫çèk/ "artichoke", /ar.gáj/

"hoarseness";3) /l/: /pal.món/ "lung";4) /s/ & /z/: /pus.tÍn/ "postman", /az.nón/ "big fool".

There is a fifth class of word-internal codas that includes /t/ and /d/ (and possibly even /v/); however, I would be reluctant to include it in the generally accepted codas due to the small number of items found: /sot.kú.a/ "slap, spanking", /skud.máj/ ‘nickname’ are compounds and could obey different restrictions; /kód.ga/ (It. /kó.ti.ka/) ‘pig skin’ (not a compound in Mantuan, probably from a Lat. *[kútik(ul)a], diminutive of cutis ‘skin’), /[email protected]/2

2 ? But cf. /[email protected]/ ‘briar’, without vowel syncope.

mailto:mar.tE@l


(It. /[email protected]/) ‘pole’ and /t∫áv.ga/ (It. /kjá.vi.ka/) ‘dam’ are the only ones that genuinely seem to have obstruents in word-internal codas. They would have to be categorized as exceptions.

However, there are bimorphemic words that are problematic in that they present unexpected obstruent-sonorant clusters word medially: pivrón, padrón, pytlón, and pigrón are all fine word-medial clusters (only [kr] and [gl] don’t seem ever to appear word-medially). Although syllabification by native speakers is by no means a safe test, these words seem to have the cluster as the onset of the second syllable, rather than syllabifying the consonants heterosyllabically, thus: [pa.drón], [py.tlón], rather than *[pad.rón], *[pyt.lón].

It is the result of the diachronic deletion of unstressed vowels3 that has differentiated Italian and other dialects favouring a CV syllable, and in Mantuan syllable maximization (Selkirk 1981) is more important than having a simplex coda or no coda at all. It is to be noticed that late spoken Latin has been argued not to have any word final clusters (Tekavcic, 1980:175), thus it is to be assumed that all those found in Mantuan arose following vowel deletion processes such as syncope, or apocope. It is interesting to notice that these phenomena are still synchronically active, for instance when standard Italian words are imported into Mantuan (see below).

To summarize, possible syllables in Mn are as follows:

3 ? Tekavcic (1980:84) states that: "1) Rumanian and Central and Southern Italian (including therefore the Tuscan dialect and the literary standard) are rather conservative (particularly the South of Italy); 2) Ibero-romance languages tend to implement syncope more frequently; 3) the Gallo-romance territory, as well as certain Northern Italian dialects and the alpine territories, are extremely prone to syncope."


TABLE II: Mantuan Syllable Inventory1) V a.mi.ga "friend";2) CV a.mi.ga; ska.pi.n .li "bottom parts of a sock";3) CVC sty.pit "fool"; v.duf "widower"; ros "red"; lu.dar

"pig"; ra¯ "spider";4) CCV tra "bring here"; dre "behind"; t¯Is "chewy";5) CVCC rçsp "toad"; kulp "shot";6) CCVC grçs "big"; trçp "too much"; mnEs.tra "soup";

fnEs.tra "window";7) CCVCC prunt "ready";8) VC as.kulta "listen";9) sCCVC strap "rip";4

2.3 Stress

The assumption adopted here concerning stress is that it is contrastive (lexical) in this dialect as it is in Italian (Nespor 1993:65). For instance, only stress distinguishes the meaning of atravérsa (“cross!”) from atraversá (“to cross”), of spása from spasá (imperative and infinitive of “to sweep” respectively, cf. Fig. 2, the same holds for most verbs)5.

Stressed vowels in the same (Gallo-romance) linguistic area have been very tenacious historically (Repetti 1995, 1996, and forthcoming), as opposed to the unstressed ones (cf. Lat. /au.gús.tus/ "august" > Fr. /u/, Mn. /a.góst/).

Fig. (1) Examples of stress in Mantuan.

‘pø .lak a.’mi.ga a.tra.’ver.sa a.tra.ver.’sa

“flea” “female friend” “cross!” “to cross”

4 ? Notice that types 4, 5, 6, 7, and 9 are only word-initial. Word-medially, clusters are very restricted (cf. above). As for two word-medial consonants: it is sometimes difficult to ascertain whether they are syllabified as C.C or .CC.5 ? It is true that this kind of stress seems to operate on a morphological basis. Contrasts within the same syntactic category are much more limited than in standard Italian (in some Mantuan dialects pápa ‘pope’ and papá ‘daddy’, however, do contrast for instance).


Some derivational suffixes bear primary stress, i.e. when they are affixed to the word, the stress of the derived word shifts from the root onto the suffix:

(3) Root Suffix sO@ka + E@l6 = sykE@l ‘pumpkin’ > ‘small

pumpkin’váka + ína = vakína ‘cow’ > ‘little cow’óm + ón = umón ‘man’ > ‘big man’

This is not unheard of in Salish languages for instance (Urbanczyk 1995), but also in Italian standard derived words: Nespor (1993:98) states that when two morphemes bearing primary stress as isolated words7 merge into a compound, one of the stresses will be “stronger” than the other one, resulting in a primary stress (ibid.). As a Romance language, she continues, in Italian the stronger stress will be the rightmost in the compound, cf. (4), whereas in Germanic languages it is generally the leftmost (5):

(4) cápo + stazióne > capostaziónevérde + máre > verdemárebátti + scópa > battiscópa

(5) bláck + bírd > bláckbirdschoól + bús > schoólbusalárm + clóck > alármclock

(from Nespor, ibid, only main stress shown here)This is so, because one foot in the phonological word is strong and all

others are weak: in Italian (as in other Romance languages and Turkish) it is the

6 ? Two vowels meeting across a morpheme boundary cause the first one not to surface, provided this is unstressed, a phenomenon well attested in Italian standard as well (Nespor 1993:83). It will not be discussed at length in the present paper. The change in vowel quality is discussed below.7 ? Although Nespor marks primary stress on bound morphemes too (p.98, ex. # (68)), as separated units before merging into one polimorphemic word, she is careful not to say that suffixes have stress underlyingly. I’d rather not discuss the matter in these terms: most suffixes are indeed stressed, but not ALL of them, cf. those adjectives derived from verbs: {ama}+{bile} literally “lov+able”, {mangia}+{bile} “edible” are stressed as: amábile, mangiábile, and so are their correspondents in Mantuan: amábil, mangiábil.

rightmost, in Czech, Finnish and Germanic languages it is the leftmost. Italian feet are trochaic (ibid, p.175, so the prominent element is on the left syllable within the foot), therefore main stress in Italian will fall on one of the last three syllables of the word depending on the nature of the foot: on the last if it is degenerate, on the second-last if disyllabic, and on the third to last in a trisyllabic last foot (ibid).

I assume that Nespor’s analysis works for Mantuan as well.There is another phenomenon to be observed, which happens in Mantuan

and in standard Italian (Nespor 1993): when two vowels come in contact across a morpheme boundary, the first one deletes as in the first two examples in 3). This happens when the first of the two vowels is unstressed, as the phenomenon is blocked in a word such as It. virtù + oso > virtuóso “virtuous”. In the case of segment deletion, violations of a constraint assessing the correspondence of elements between the input and the output will be necessary. This is taken from Correspondence Theory (McCarthy and Prince 1995). MAX is defined as the following faithfulness constraint:

(6) MAXEvery element of a string S1 has a correspondent in a string S2, where a correspondence relationship S1U S2 holds between the two.

This requires input segments to correspond faithfully to segments in the output. However, since it is more important to be faithful to stressed vowels than to unstressed ones, it must be inferred that there are two types of Max constraints: one assessing faithfulness to stressed vowels and the other one to unstressed vowels. The first must outrank the second as set off in the following tableau (Mn ranking is, in this circumstance, the same as the Italian ranking):

(7) Segment deletion in Mn: importance of faithfulness to stressed vowels as opposed to unstressed ones./váka + ína / MAX-’V MAX-V1.vaka<í>na *!2. vak<a>ína *

The first candidate loses because the stressed vowel is deleted, whereas the winning candidate deletes an unstressed one. The phenomenon should be driven by a higher ranked constraint against the coming together of two vowels across a morpheme boundary, which would have to be placed between the two


constraints analyzed above, in order for a word like virtuóso to surface with no vowel deletions.

The same ranking would hold for Italian too:

(8) tç€.ro + E@llo = [email protected] “bull” > “young bull”pík.ko.lo + íno =pik.ko.lí.no “small” > “tiny”

It should be remembered that, unless otherwise stated, in this chapter the vowels undergoing deletion or feature changes are always unstressed, and that there is, both in Mantuan and Italian a higher ranked constraint (crucially higher than the simple Max-V) against lack of input-output correspondence of stressed vowels. This constraint will therefore not be seen in the rest of the tableaux in the first part of the chapter.

Another phenomenon should be mentioned. There is also a high-ranked constraint (discussed below) against deleting semantically significant morphemes at the right word edge. This would be the case of word-final -[a]#, the feminine sg.marker, or [i]#, the fem./masc. pl. marker, and -[u]#, which marks masculine proper names in Mn.

2.4 V ~ 0 Alternations in Syllables Preceding Stress

The tendency to have complex consonant clusters and fewer syllables in Mantuan conflicts with the need to produce acceptable consonant clusters. Consider the last syllable of the following lexemes from Mantuan and their Italian counterpart, as well as the augmented form in Mantuan. The Mantuan forms have been traditionally analyzed as having an epenthesized vowel in the last syllable (Rohlfs 1966, and Repetti 1995, 1996, for related Emilian and Romagnol dialects), since that vowel disappears in derived forms (second column in Table III), and since the Latin etyma and the Italian cognates bear no trace of it.

TABLE III - Vowel ~ 0 AlternationsMantuan Italian gloss

1) Word-final CaC# (or [r]+[n] )

i. káran karnína kárne meat ii. fúran furnÍn fórno oven


2) Word-final CuC# (or [r]+[m]) iii. vE@rum vermÍn vE@rme worm iv. férum firmÍn férmo still

3) Word-final CaC# (or [r(s)]+[k] ) v. sórak surgÍn sórt∫o mouse vi. bórak burgÍn bórgo burg

3 consonant cluster vii. pE@rsak persgÍn p’ska (L.pE€rsikum) peach

4) Word-final clusters that would go against sonority viii. pévar pivrón pépe (L.piper) pepper ix. lábar labrón lábbro lip x. nádar nadrÍn ánatra duck xi. kwátar kwatrÍn kwáttro four xii. négar nigrÍn négro black xiii. ázan aznón ázino ass

3 consonant clusters xiv. sémpar - sémpre always xv. déntar - déntro inside xvi. Lisándar Sandrón Ales'sándro Alexander

In these preceding examples, it should be noticed that all the words in the first Mantuan column, as well as in the Italian one, are stressed on the first syllable (unless marked otherwise), and end in a stressless syllable whose structure is either CaC# or CuC#.

The fact that there is a vowel there at all should be seen as remarkable, since there is no trace of a vowel interrupting the equivalent clusters in the Italian cognates (excluding example xiii) or in the corresponding Latin etyma (except for vii). The Italian forms, however, have not undergone the apocope of the final vowel, which they have retained from Latin, unlike Mantuan.

The second remarkable thing to notice should be evinced from the second column (in boldface). These data from Mantuan show that the same forms listed in the first column do not always have that [a] or [u] breaking up the consonant cluster: when a derivational suffix (augmentative, diminutive, pejorative) is added to the root, the vowel in what was the final syllable disappears, and the cluster is broken up between the coda of the first syllable and the onset of the

second one. When it is possible to resyllabify the resulting word-final cluster - as would happen in phrasal phonology - the incriminated vowel disappears again, and the cluster is distributed over coda and onset, or just made into the onset to the following syllable: cf. the adverbs vi) and vii) in Table III: 'always on (her) feet' sempar in pe would become ['sem.prin.'pe], 'four eggs' kwatar Of would become in fast speech [kwa.trO@f].

It should also be mentioned that only three vowels can surface in stressless final syllables (post-tonic position): [a], [i], and [u]. In stressless (pre-tonic) word-initial syllables, [y] can also be found (similarly in Occitan, Wheeler 1988:147). [a] seems to be the most common of all vowels. The more marked mid vowels (except for [e] in some cases, see chapter 4 below), as well as high lax vowels, never surface in stressless positions. Mantuan shares this phenomenon of reduction with the other Romance languages, such as Catalan and Portuguese, that will be analyzed in chapter three.

The distribution to be explained is then that of a vowel between two consonants in word final position, which is missing in the Italian cognates. Two are the possible solutions: the vowel8 is either there underlyingly, but has no correspondent in a pretonic position, when the resulting consonant cluster is not disallowed, or it is inserted epenthetically in the base form to break up an awkward cluster. An analysis of Mantuan syllable structure will prepare the ground to explore the deletion and epenthesis hypothesis in turn.

2.5 An OT Account of Mantuan Syllable and Word Structure

2.5.1 Basic Syllable structure Constraints and Their Ranking

The framework chosen to model these data is Optimality Theory (OT) as laid out by Prince and Smolensky (1993), McCarthy and Prince (1993, 1995) and subsequent implementations and refinements of the model (see Kager 1999).

As is well known, OT conceives of grammars as different rankings of universal constraints, thus, by examining the appropriate constraints and re-ranking them, it should be possible to capture the facts about Mantuan syllable structure

8 ? [a] or [u] in the cases at hand; [i] can also appear in word-final stressless syllables (very rarely, though) in a word such as [ándZil] ‘angel’, or [sty@pit] ‘stupid’ , for instance, but these words do not participate in the alternations with and without vowel: 'fool' = [sty@pit], but 'big fool' = [stypidón] not * [stypdón].


The first constraints involved in explaining this part of Mantuan grammar are:

(9) NOCODASyllables must not have a coda (Prince and Smolensky 1993:85, as "-Coda")

(10) *COMPLEXNo more than one C or V may associate to any syllable position node (Prince and Smolensky 1993:87)

(11) *STRUC()“No syllable”.

This last constraint requires, in practice, to construct fewer syllables and is assessed with one mark per each syllable a candidate has. It is based on Selkirk's Syllable Minimization Principle (1981), and has recently been invoked on various occasions: within OT, for instance, by Alderete for Winnebago (1995:37), and Zoll for Chaha and Yawelmani (1993, 1994). It is generally justified by a natural tendency for "brevity"9, acknowledged by many other authors, working in and outside OT (cf. Goldsmith 1990:118; Vennemann 1988:2).

A tendency to minimize syllables has to be pitted against the other constraints that govern the well-formedness of syllables in the language. The material in the input has to correspond to elements in the output, which is established by Max (6) repeated as (12) here, and by Dep, in (13); MAX is a constraint against segment deletion, and it ensures that the syllables be as large as possible, encompassing all input material:

(12) MAXEvery element of a string S1 has a correspondent in a string S2, where a correspondence relationship S1U S2 holds between the two (McCarthy and Prince 1995).

Conversely, if material that is not in the input nevertheless surfaces in the output, a violation of another constraint, DEP, is incurred. This is the constraint against epenthesis:

99 Vennemann (1988:2) states for instance that "…syncope and apocope…are

manifestations of the preference for briefness" in language.


(13) DEPEvery element of a string S2 has a correspondent in a string S1, where a correspondence relationship S1 U S2 holds between the two (McCarthy and Prince 1995).

The present analysis uses the MAX and DEP version of these concepts laid out in McCarthy and Prince 1995 as part of Correspondence Theory. As is common in the literature (McCarthy and Prince 1995, Appendix A), these two constraints will be further specified as Max-V and Max-C (against the deletion of vowels and consonants respectively), as well as Dep-V and Dep-C (against the insertion of vowels and consonants respectively):(14) MAX-C

Every consonant of a string S1 has a correspondent consonant in a string S2, where a correspondence relationship S1 U S2 holds between the two.

(15) MAX-VEvery vowel of a string S1 has a correspondent vowel in a string S2, where a correspondence relationship S1 U S2 holds between the two.

In order to see the interaction of the constraints on syllable structure and their ranking in Mn, consider the word for "thistle": Mantuan [kard] (or, more precisely, [kart]10, to avoid distraction from the main issue, [kard] will be used). The constraint *STRUC() will be assessed against *COMPLEX in (16), in order to establish the relative tolerance of Mn towards consonant clusters.

10 10 Codas are marked positions; thus, word-finally, a featural distinction of segments is sacrificed, namely [voice]. The process applies only to obstruents, sonorants remaining voiced even word-finally. The singular /dZáld/ "yellow" surfaces as [dZált], whereas the plural form is /dZál.di/ = [dZál.di]; cf. with /prúnt/ "ready" = [prunt], fem. pl. (masc. pl. is invariant) /prún.ti/ = [prún.ti], as well as /vE€duv/ "widower" = [vE€duf], but cf. /vE€d.va/ = [vE€d.va] "widow". The process can be explained with the interaction of 3 constraints (Lombardi 1995): IDOnsLar, requiring that voicing be parsed faithfully in onsets; *Lar, a constraint reflecting the universally marked character of voiced obstruents; and IDLar, a constraint requiring that all laryngeal features present in the input be parsed in the output. Crucially, IDOnsLar ranks highest, followed by the markedness constraints, and by the requirement for faithful parsing, just as in German:/vE€duv/IDOnsLar*LarIDLar1. vE€.duv*!**2. vE€.duf***3. fE€.duv*!***


(16) Mantuan relation between consonant clusters and number of syllables.

input /kárd/ *STRUC() *COMPLEX

1. .kár.d[u]. *!*2. .kárd. * *

Here we have a ranking argument, since, if *COMPLEX were to dominate *STRUC(), the wrong candidate would win, namely # 1.

2.5.2 Importance of NoCoda

It should also be evaluated whether the constraint against codas is relevant in Mantuan syllable structure:

(17) Placement of NoCoda.input /már/ *STRUC() *COMPLEX NOCODA

1. .má.r[V]. *!*2. .már. * *

Tableau 17 establishes the ranking of *Struc() above NoCoda, so that candidate 2 wins over the disyllabic candidate 1. However, no ranking can be established between *Complex and NoCoda.

The ranking between these two constraints can be seen in tableau 18, where the winning candidate 2 has two violations of NoCoda and still surfaces.

In tableau 18, in fact, given two possibilities to epenthesize and somehow break up a three-consonant cluster in both candidates 2 and 4, the winning candidate is the one that has two codas, but no violation of *Complex (candidate 2). Candidate 4 violates NoCoda only once, but still does not surface. Notice that there must be a constraint against three-consonant clusters word-finally in the language, since they are not attested. This constraint has been called “*CCC#” in tableau 18. This is meant only descriptively, although it seems a fairly common restriction in the world’s languages (cf. Kiparsky 1973), but whatever it stands for, it weeds out the faithful candidate 1. It must be placed at least above *STRUC().


(18) Ranking argument for *Complex and NoCoda./pE@rsk/ ‘peach’ *CCC# *STRUC() *COMPLEX NOCODA

1. .p E@rsk. *! * * *2. p [email protected][a]k ** **3. p E@r[a]sk ** *! *4. p [email protected][a] ** *! *

Candidate 3 confirms the ranking of *Complex dominating NoCoda, since if these two were reversed, 3 would win and 2 would wrongly lose; if they were unranked, 2 and 3 would wrongly tie as optimal candidates.

Codas are often present in Mantuan syllables, thus NOCODA is frequently violated and is at the bottom of the hierarchy; it is more important not to violate a constraint against clusters, than not to have codas.

2.5.3 Interaction with Faithfulness: No Insertion (Dep-V) and No Deletion (Max-V)

The losing candidate 1 in t.16, kard[u], also incurs a DEP-V violation not evaluated in that tableau. If Dep-V were to be evaluated, candidate 1 would lose because of its violation. It must be placed at least as high as *STRUC(), in order for the right candidate to surface.

(19) Mantuan syllable structure constraints and Faithfulness.input /kárd/ Dep-V *STRUC() *COMPLEX

1..kár.d[u]. !* **2. .kárd. * *

In fact, it can be ascertained that Dep-V overlaps with the role of *Struc() in most cases, to the point where Dep-V is redundant, whereas the opposite would not be not true, i.e. *STRUC() could not be substituted by Dep-V on all occasions (cf. section 3.2). Because of its redundancy, Dep-V will not be included in the remaining tableaux.

As mentioned on p. 3 of section 1, Mantuan does not contrast between monomorphemic words such as kard and kardV. This is also proven by the fact that imports from standard Italian typically undergo apocope: the Italian for “television”, [te.le.vi.zió.ne], appears in Mantuan as [te.le.vi.zión]. Apparent exceptions such as contrastive dZald ~ dZald-i, “yellow” masc. sg. and

masc./fem. pl., are in fact to be analyzed as a monomorphemic and a morphologically complex form respectively.

Apocope can be explained by the ranking of *Struc() above Max-V, favouring the deletion of vowels. Consider the following tableau 20, as long as Max-V is below *STRUC(), the right result will be obtained, i.e. candidate 2 will surface. However, there is no argument for a domination of *Complex over Max-V: very often in fact, where one is violated, the other is too, since eliminating a vowel often gives rise to consonant clusters.

(20) Syllable structure constraints and Max-V placement.input /kárdu/ *STRUC() *COMPLEX Max-V1..kár.du. *!*2..kárd. * * *

Here, *STRUC() is seen as driving apocope of the word-final vowel, since it is better to have fewer syllables, if the surfacing candidate only violates lower-ranked constraints.

2.5.4 Prevention of Apocope

This constraint hierarchy shows that words ending in -(C)CV# / -(C)C# do not contrast in Mantuan, all else being equal, and would both surface as -(C)C#. This is a good result, given what we know about imports from Italian.

As previously mentioned, inputs such as dZald ~ dZaldi ~ dZalda , on the other hand, cannot all surface with the same form, because {-i}# and {-a}# are distinctive morphemes (pl. and fem. sg. respectively). It should also be mentioned that {-u}# is also a morpheme with limited distribution, compared to the other two, it is a proper name masculine marker: thus Italian proper names such as Roberto and Riccardo are imported as [(ru).bér.tu], [(ri).kár.du]. In order to keep morphologically complex words from losing their word-final morpheme, a higher ranked morphological constraint should be posited.

Abu-Mansour (1995:2) uses a constraint called "Parse-morph" for Arabic, which states that a morphologically significant segment should not be deleted. From this, one may formulate the following constraints in Correspondence Theory terminology:

(21) DEP-morph11: A morphologically significant segment should not be epenthesized.

and (22) MAX-morph:

A morphologically significant segment should not be deleted.

This may of course be short-hand for something else, since, presumably, what makes a segment "morphologically significant" is also where it is located, calling for alignment considerations. The concept of Dep-morph is particularly problematic and would perhaps require further elaboration to be tenable. Precise definition of these constraints, however, goes beyond the scope of the present discussion. For Mantuan, at any rate, the relevant one is always Max-morph.

Thus, word-finally, only -a, -i, (-u) can ever appear unstressed, and these are all semantically significant morphemes. This means that all inputs C(C)VC(C) do contrast with inputs C(C)VC(C)V, where the final vowel is one of the three indicated above.

In tableau 23, there is an example of how the higher ranked Max-morph constraint can stop apocope from happening: candidate 2, in fact, fares better under the *STRUC() constraint, but deletes a significant morpheme and does not surface.

(23) *STRUC() drives apocope, but Max-Morph stops it./pált-a/“mud” sg.fem.

MAX-{MORPH}

*STRUC() *COMPLEX Max-V

1..pál.ta. **2..pált<a>. *! * * *

2.5.5 Deletion of Consonants in Mantuan: Max-C

At this point, the importance of faithful surfacing of consonants will be evaluated. The tendency is to save parts of the consonantal input in Mn, even if, by not parsing them, a violation of *COMPLEX could be avoided, cf. 24):

(24) Max-C higher than *Complex in faithful surfacing of consonants.

11 ? In reality, epenthetic material is not morphologically affiliated, so that this constraint penalizes the insertion of an epenthetic segment homophonous with an existing morpheme [a, i, u].

/káld/ 'warm' MAX-C *COMPLEX

1.káld *2.kál<d> !*

This must be the right ranking hierarchy between these two constraints, since the opposite would produce the wrong result. Candidate 1 saves the word-final cluster even at the price of contravening *Complex and surfaces, whereas not parsing one of the consonants results fatal to candidate 2. The proposal is then to posit a Max-C constraint dominating *Complex, independent of Max-V below it in the hierarchy, since deletion of consonants is unattested in the language.

Max-C dominates *Complex, which in turn has been seen to be equally ranked with respect to *STRUC(). However, Max-C must also be ranked higher than *STRUC(), or the tendency would be for this language to have only monosyllables, all else being equal.

The ranking governing Mn syllable structure could then be expressed as follows:

(25) (DEP)/MAX-{MORPH}, *CCC# >> MAX-C, *STRUC() >>*COMPLEX, MAX-V >> NOCODA

To summarize the findings in this section, the following input tokens will be evaluated in one tableau: kárd, kárdV, pálta, and pE€rsk. “V” in kardV means any vowel that is not a significant morpheme. The outcome of the following tableau proves that the hierarchy established in (25) is correct and that Mantuan inputs do not contrast in shape between C(C)V(C)C and C(C)V(C)CV, where this last vowel is not interpreted as a morpheme. Apocope is driven by a constraint requiring the minimization of the number of syllables in a word.

(26) Summary tableau for basic syllable structure constraints hierarchy/kárd/ sg.m. MAX-

{MOR}*CCC# MX-C *STRC

()*CPX MX-

VNOCODA

1.kárd * * *2.kár.d[V] *!* *3.kár<d> *! * */kárdV/ sg.m.1. kár.dV *!* *2. kárd<V>

* * * *

/pált-a/ sg. f.1. pál.ta ** *2.pált<a> *! * * * */pE@rsk/ sg.m.1. .pE@rsk. *! * * *2. pE€r.s[a]k

** **

3. pE€r[a]sk ** *! *4. [email protected][a] ** *! *

2.6 Mantuan Alternations as Deletion of Pretonic Vowels

2.6.1 *STRUC() drives deletion

Vowel ~ 0 alternations in Mantuan can be analyzed either as deletion of a word-medial vowel in the morphologically complex forms, or as epenthesis of a vowel to break up a disallowed word-final cluster in the base forms: the deletion and the epenthesis hypotheses will be explored in turn.

The alternations in question are the V ~ 0 ones in examples such as the following, where the vowel attested in the base form disappears in the derived form made up of the root and an augmentative morpheme (it could equally be exemplified by a pejorative or diminutive morpheme):

sórak ~ sur<>g+ónpádar ~ pad<>r+ónpévar ~ piv<>r + ónv‘rum ~ ver<>m+ ón

Such alternations are not attested in forms such as the following:

tSákara ~ tSakar +ónbágula ~ bagul +ónsty@pit ~ stypid + ón

At the end of the section, a summary tableau of two of the basic forms presented in t. 26 and of some morphologically derived forms will be given, in order to assess the correctness of the newly attained hierarchy.

As can be seen in 27, a trisyllabic word such as /suragón/ syncopates to two syllables, as long as deleting a vowel does not cause any disallowed clusters to surface, a case of conspiracy (related examples are, for instance, in Kiparky 1973). Candidate one is therefore the winner, since deleting the pretonic vowel only augments the number of codas, violating the lowest-ranked constraint. The second, faithful candidate has one syllable too many; the deletions in 3 and 4 produce extra *Complex violation, as the candidates would surface with unattested clusters, and are therefore weeded out. Clearly, candidate 4 also deletes the stressed vowel, incurring a violation of the higher ranked constraint not evaluated in this tableau against deletion of stressed vowels.

(27) *STRUC() driving word-medial deletion./sórag+ón/ ‘big mouse’ *STRUC() *COMPLEX MAX-V NOCODA

1.sur<a>.gón ** * **2.su.ra.gón **!* *3.sra.gón ** !* * *4.su.rag<ó>n ** !* * *

The ranking of *Struc() above *Complex has the effect of encouraging deletion wherever possible, since the lowest number of syllables is always privileged, as long as the complexity of the surfacing clusters does not violate sonority or *CCC. Let us examine further consequences of this ranking.

2.6.2 Consequences for Other Derived - Alternating Forms

In the alternating V ~ 0 examples, the following happens: an input /pévar+ón/ “bell pepper” will surface as [pivrón] (candidate 1) in spite of the Max-V violation, because the faithful candidate (2) has one syllable too many (cf. Tableau 28), and the emerging word-medial cluster in 1 is allowed.

(28) Deletion at work. (Only relevant constraints shown)/pévar+ón/ *STRUC() *COMPLEX MAX-V1.pi.v<a>rón[pi.vrón]

** * *

2.pi.va.rón[pi.va.rón]

**!*

On the other hand, an input /pévr+ón/ will also surface as [pivrón] (candidate 1 in 29 below), showing that these two types of words cannot contrast in this language. *Complex is in fact too low to force a *Struc() violation, which is fatal to candidate 2 below:

(29) Faithful surfacing./pévr+ón/ *STRUC() *COMPLEX MAX-V1.pi.vrón[pi.vrón]

** *

2.pi.v[a].rón[pi.va.rón]

**!*

2.6.3 Disallowed Clusters in Alternating Forms

However, it is not sufficient just to assess the syllabic shape of the word, neglecting segmental content. Words such as tSakarón, bagulón, stypidón do not syncopate, because if they did, a disallowed cluster would emerge. Clusters such as [kr], [gl] and [pd] never appear word-medially, a fact that the present hierarchy fails to capture. This hierarchy would never allow a candidate as [t∫a.ka.rón] to surface. This is obviously a bad result, since t∫akarón is an actual Mantuan word, but the input in t. 30 could never surface faithfully, since it is better to delete a vowel than to incur an extra *Struc() violation.

(30) Wrong result: this word should fail to syncopate.

/t∫ákara+ón/ *STRUC() *COMPLEX MAX-V1.t∫a.k<a>r<a>ón[t∫a.krón]

** * **

2. t∫a.ka.r<a>ón[t∫a.ka.rón]

**!* *

In order to work, this solution requires that a constraint ranked higher than *Struc() be responsible to weed out the unattested clusters. It is not clear what the language has against these clusters: [pd] may be a case of a disallowed sonority plateau in a cluster, whereas [kr] and [gl] may be out because of a dorsal component that the liquids seem to have in this language (Miglio 1996), that prevents them from surfacing in a cluster with another dorsal. It should be remarked that all three clusters are possible in word-initial position, but an anchoring constraint may force them to surface there. A more detailed treatment of this issue goes beyond the scope of the present discussion.

Thus, a portmanteau constraint *Dis(allowed)-cl(uster) subsuming the three above-mentioned clusters will be introduced:

(31) The *Dis-cl constraint at work./t∫ákara+ón/ *DIS-CL *STRUC() *COMPLEX MAX-V1.t∫a.k<a>r<a>ón[t∫a.krón]

*! ** * **

2.t∫a.ka.r<a>ón[t∫a.ka.rón]

*** *

With this new constraint in place, the input of tableau 31 surfaces faithfully, unlike in t. 30, and no input with a disallowed cluster is ever allowed to surface in the language.

It is even better to epenthesize than having a disallowed cluster, as can be seen with the different input in t. 32. The new constraint dooms candidate 1 because of a disallowed consonant cluster and candidate 2 surfaces.

(32) The *Dis-cl constraint at work, different input./t∫ákra+ón/ *Dis-cl *STRUC() *COMPLEX MAX-V1.t∫a.kr<a>ón[t∫a.krón]

*! ** * *

2.t∫a.k[a].r<a>ón[t∫a.ka.rón]

*** *

With the placement of *Dis-cl above *Struc(), the language requires that no contrast be possible between inputs /t∫ákara.+ón/ and /t∫ákra.+ón/: they would both surface as [t∫akarón].

2.6.4 A Different Syllabification Problem

It should also be mentioned that the /t∫ak<a>rón/ input can never surface with a different syllabification such as [t∫ak.rón], as [sur.gón] does, or it would be violating one of the laws proposed by Vennemann for syllable structure (Vennemann 1988:40):

(33) Vennemann's "Contact Law"

In a word-medial syllable contact, the sonority of the consonant in the coda (A) must be higher than the sonority of the consonant in the following onset (B).

This "law" was adapted to the OT framework by Repetti (1995:4, see also Davis 1997, Hieronymous 1999) who uses it for similar environments in Emilian and Romagnol dialects.

Thus, a constraint against "bad sonority contacts", Syll-Contact, should be introduced in the tableaux 35) and 36), essentially reproducing Vennemann's law12

:

(34) Syll-Contact is violated each time that, in a word-medial syllable contact, the sonority of the consonant in the coda is lower than the sonority of the consonant in the following onset.

Consider therefore tableau 35): clearly the new sonority-related constraint must be above the top-ranked *STRUC(), or 1 would lose in favour of 2.

12 ? Vennemann’s original name for this law/constraint, also used by Repetti, is “A$B”. It is substituted here with “Syll-Contact” for clarity.

(35) Syll-Contact as top-ranked./tSakara+ón/ Syll-Cont *DIS-CL *STRUC() *COMPLEX MAX-V1. tSa.ka.r<a>ón

*** *

2.tSak<a>.r<a>ón !* ** **

Here the sonority related constraint is able to suppress the candidate violating it and gets the right candidate to surface, i.e. 1. The same ordering works for /sórag+ón/, in 36), where Syll-Contact is vacuously satisfied (the sonority of the coda is indeed higher than that of the following onset); the first candidate does not surface because of one syllable too many, the second wins even if it does not respect faithful correspondence of one vowel (because Max-V is low-ranked).

(36) Syll-Contact as top-ranked./sórag+ón/ Syll-

Cont*DIS-CL *STRUC() *COMPLEX MAX-V

1.su.ra.gón **!*2. sur<a>.gón ** *

If Syll-Contact is ranked anywhere but above *STRUC(), the mini-grammar exemplified in 35-36 would produce unattested results in Mantuan, since it would pick the right candidate in 36), i.e. [surgón], but the wrong candidate in 35), i.e. [tSak.rón].

2.6.5 Summary

It is important to have the right sonority sequencing in Mantuan syllable contacts, i.e. word-medially, which explains why word-medial codas are limited. Max-V is low ranked, and this is to be expected if the V ~ 0 alternations are to be ascribed to deletion. Both apocope and syncope are seen to be driven by *Struc(). This ascribes to the same mechanism two phenomena that have been traditionally treated separately.

This solution accounts for all the facts and alternations. It also shows that inputs such as /pi.va.rón/ and /pi.vrón/ cannot contrast and both surface as [pi.vrón]. It requires, however, that an extra constraint be introduced, in order to rule out disallowed clusters such as [kr]: in this case inputs such as /tSakra+ón/ and /tSakara+ón/ are shown not to contrast, since they both would surface as [tSakarón].

This lack of contrast makes of Mantuan a “conspiracy” case in the sense of Kisseberth (1970), where constraints conspire to produce a certain surface form, i.e. a syncopated or apocopated one, provided other considerations of well-formedness do not intervene to stop vowel deletion from happening.

The ranking required by this solution is:

(37) (DEP) / MAX-{MORPH}, *CCC#, *DIS-CL, Syll-Contact >> MAX-C >> *STRUC() >> *COMPLEX, MAX-V >> NOCODA

2.7 An Alternative with Alignment and DEP-V Driving Deletion

Given that Mantuan has been seen not to distinguish between inputs such as kard and kardV, where the last vowel is not a significative morpheme, an alignment constraint could be introduced to force apocope, requiring that a consonant be aligned with the end of the Prosodic Word: Align-R, C]PrWd.

This would have to be crucially ranked above *Complex and it would substitute *STRUC() in this respect:

(38) Alignment drives apocope./kárdu/ ALIGN-R, C]PW *COMPLEX Max-V1.kár.du. *!2. .kárd. * *

In this case, the faithful candidate would be weeded out by the requirement that words end in a consonant where possible, and in spite of an extra Max-V violation, candidate 2 surfaces. This is not unlike one of the requirements imposed by incomplete-phase morphology in Rotuman (McCarthy 1994); the constraint McCarthy uses to drive apocope in some cases is purely morphological (short hand for “comply with Incomplete Phase requirements”), and requires that incomplete phase disyllabic words undergo apocope to make the word a heavy monosyllable (cf. McCarthy 1994:11 and ff., and p.18): Complete Phase to{kíri}, Incomplete Phase to{kír}.

Dep-V would also need to be evaluated, cf. 39:

(39) Dep-V works for some derived forms./ladr+ón/ ALIGN-R, C]PW DEP-V *CPX MAX-V NO CODA

1. la.drón * *

2. la.d[a].rón *! *

In 39, the faithful candidate 1 surfaces and 2 is weeded out by the insertion of an extra vowel. Both inputs in 38 and 39 are therefore seen not to contrast on the surface. No ranking can be established between alignment and Dep-V.

However, a form showing alternations between V and 0, (ignoring the input-output vowel feature differences for the moment), will set off the limitations of this approach (only relevant constraints shown), where 1 is the desired winner (marked by ):

(40) Wrong result: this hierarchy is not sufficient to explain the alternations.

/sórag+ón/ ‘big mouse’

ALIGN-R, C]PW DEP-V *COMPLEX MAX-V NOCODA

1. sur<a>.gón *! **2. su.ra.gón *3. sra.gón *! * *4. su.ra.g<ó>n *! * *

Candidate 2 wrongly surfaces as the winner, and is the candidate segment-wise most faithful to the input. The first candidate loses because of an extra Max-V violation, although that should be the actual winner. The third has an extra word-initial cluster that dooms it, as well as the last one (which also incurs a higher ranked violation against deleting stressed vowels, not shown here).

It is interesting to assess that, whereas in the previous two tableaux on monomorphemic words, *Struc() was overlapping with the Alignment and Dep-V constraints, here the right result is only obtained with *Struc() in place, as there is nothing in (40) to drive the deletion of a word-medial vowel (syncope). Thus, the *Struc() analysis in section 3.1 is to be preferred.

2.8 Epenthesis in (C)C^C# environments

An alternative account of the V ~ 0 alternations will now be explored, since in all fairness to traditional philology (Rohlfs 1949 passim, specifically on Lombardy dialects) and to other researchers (Repetti 1995, 1996, 2000) on regionally close dialects), epenthesis is what has been proposed to explain the base forms.


These word final environments have in fact been treated as having a disallowed cluster in the underlying representation, say /padr/ or /vrm/, giving rise to the epenthetic insertion of a vowel. This works beautifully for those dialects that present a single vowel, most often a [´] as in Repetti's dialects, in those environments. A few problems arise in the case of Mantuan; they can, however, all be solved.

Please refer to an excerpt of Table III repeated below as Table IV.

TABLE IV - Vowel ~ 0 Alternations (excerpt)Mantuan Mantuan deriv. Italian gloss

1) Word-final CaC# (or [r]+[n] ) i. káran karnína kárne meat ii. fúran furnÍn fórno oven

2) Word-final CuC# (or [r]+[m]) iii. vE@rum vermÍn vE@rme worm iv. férum firmÍn férmo still

3) Word-final CaC# (or [r(s)]+[k] ) v. sórak surgÍn sórt∫o mouse vi. bórak burgÍn bórgo burg

3 consonant cluster vii. pE@rsak persgÍn pE@ska peach

(L.pE@rsikum)

4) Word-final clusters that would go against sonority viii. pévar pivrón pépe pepper ix. lábar labrón lábbro lip x. nádar nadrÍn ánatra duck[...]

3 consonant clusters xiv. sémpar - sémpre always xv. déntar - déntro inside xvi. Lisándar Sandrón Ales'sándro Alexander Mantuan has been seen to have [a] and [u] in the root-final syllables, if these are epenthetic, there must be a way to choose between the two. However, the words


in this category all end in uC#, where C = [labial]; this could make predictable which vowel to epenthesize in which environment.

If assessed more closely, the (C)C^C# environments where the presumed epenthetic vowel surfaces are really somewhat different:

a) The consonants under 4) invert sonority, therefore presenting a good reason to epenthesize (a constraint against sonority inversion must be posited, at least as high as Syll-Contact in the hierarchy).

b) The 3-consonant clusters under 4) invert sonority in the last two consonants, but present no Syll-Contact violations when a new syllable is created by epenthesis. This explanation for epenthesis is also plausible (see also fn. 13 below).

c) The words under 1) and 2) do not invert sonority and still trigger epenthesis. This can be explained as the desire of Mantuan to avoid sonority plateaus in clusters13.

d) The words under 3) do not invert sonority, and do not have two sonorants in the cluster. This seems to show a Mantuan dislike of [rk] clusters, whereas [rsk]# can be considered problematic because it is a three-consonant cluster14.

In spite of the complications mentioned above, there could be a case for epenthesis. The constraint hierarchy will enable one to assess the claim that these alternations are a function of epenthesis, rather than deletion.

13 ? Liquids are more sonorous than nasals, considering the following sonority scale: vowels - glides - liquids - nasals - obstruents (Kenstowicz 1994a:254 among others). However, Mantuan seems to need a separation of more than one degree in sonority for the sequencing principle to be observed, so that a sequence of a liquid and a nasal may be considered a plateau and be ill-formed. This would be similar to what happens in Greek (Steriade 1982). 14 ? As high ranked as Syll-contact or Son, there must be a constraint banning three-consonant clusters from surfacing faithfully in word-final position in this language, since they are never seen. Thus, *CCC# ranked as high as Syll-contact / Son is needed here too, in order to rule out the faithful surfacing of a candidate /pErsk/ as in tableau 16. As for the ban on [rk] clusters, this may again be due to the dorsal secondary component of liquids in this language (Miglio 1996) and will be subsumed under the “Disallowed cluster” constraint for present purposes.


Since insertion of material which is not in the input is going to be assessed, violations of the DEP-V constraint are going to be important. No “motor” for syncope (i.e.*Struc()) is to be expected here, since the alternation to be explained is: [sorak]~[surgon], coming from a base form /sork/: without the medial vowel, there is no need to syncopate in the derived form. But there is still the need to “clip” monomorphemic forms ening in -(C)CV#. So *Struc() will be substituted by the alignment and Dep-V constraints analyzed and subsequently rejected for deletion in 2.6.

The ranking for Mantuan is then:

(41) MAX-{MORPH}, *CCC# >> ALIGN-R, C]PW, MAX-C, DEP-V >> *COMPLEX, MAX-V >> NOCODA

2.8.1 Introduction of Sonority constraint

However, this ranking would not be able to stop an input such as /ládr/ from surfacing, therefore, since [dr]# violates sonority sequencing, this requires the introduction of another constraint, SON:

(42) SONClusters must respect the sonority scale and avoid sonority plateaus.

This is only a shorthand version of whatever constraints govern sonority (cf. Smolensky 1995, Hironymous 1995). The results in tableau 43) imply that reversing the sonority scale must be at least ranked higher Dep-V, and its violation dooms the 2nd candidate. See the results of the introduction of the new constraint in the following tableau:

(43) Introducing Son. (Only relevant constraints shown)./ládr/ 'thief' *CCC# SON AL-R,C]PW MAX-C DEP-V *CPX MAX-V NO

CODA

1. lá.d[a]r * *2.ládr *! * *

A candidate such as [ladr[a]] is never posited in these tableaux because it would violate a high ranked Dep-morph constraint. This should explain why no candidates for masculine singular inputs in these tableaux ever end in [a] ({-a}#


= fem. sg.) or [i] ({i}# = masc./fem. pl.).15 Alternatively, Alignment can prevent word-final epenthesis.

This ranking for Mantuan shows that inputs ladr / ladar cannot be contrastive, and ladar should always be the surfacing candidate, epenthesizing in the first case, and surfacing faithfully in the second.

Syll-Contact is confirmed to be as high ranked as in the deletion section, on a par with Son. In t. 44, candidate 3 fails to surface because of it (syllabification is different from candidate 1). Candidate 2 does not surface because of a Dep-V violation.

(44) Augmented form./ladrón/ 'thief' (augment.)

SYLL-CONT

SON DEP-V *CPX MAX-V NOCODA

1. la.drón * *2. la.d[a].rón *! *3.lad.rón !* **

Notice however that an input /ladarón/ would surface faithfully with this ranking, without resorting again to the introduction of *Struc(), the syncope mechanism. This would imply that both ladarón and ladrón are possible Mantuan surface forms. However, forms such as [ladarón] are not attested in the language, unless they had disallowed word-medial clusters (cf. t. 46 below). This is not likely to be an accidental gap, given that the syncopated version of these forms is quite common in the language. Their non-existence could point to an inadequacy of the epenthetic solution.

This ranking also caters for inputs such as /kárn/ > [káran], where two sonorants come in contact in the same cluster, cf. (45), violating the sonority plateau constraint. If they were syllabified heterosyllabically, the constraint would not be active, and an input /kar.n+í.na / would surface faithfully. In t. 45, however, candidate one inserts [a] epenthetically and avoids a fatal violation of SON, although it violates a number of lower-ranked constraints. Candidate 2 violates SON and fails to surface.

15 ? Thus, for instance, [sór.ga] with epenthetic word-final /a/ is blocked by the morphosemantic interpretation of /a/# as a feminine marker, which would be turn the input "mouse" into "female mouse". This consideration holds for all cases of this shape where a gender alternation is possible (if only for comical purposes), and since Mantuan has grammatical gender, this means quite a few items.

(45) Avoidance of sonority plateaus./kárn/ 'meat' SYLL-

CONTSON DEP-V *CPX MAX-V NO

CODA

1. ká.r[a]n * *2. kárn *! * *

2.8.2 Dealing with disallowed clusters

The same constraint used in the ranking for deletion to ensure that clusters such as [kr] do not surface will also be needed, since an input /tSakara+ón/ will surface faithfully as [tSakarón] even without it, but an input /tSakra+ón/ would wrongly surface faithfully in 46 if *Dis-cl were not introduced.

(46) Input without underlying vowel: importance of *Dis-cl./tSakra+ón/ 'thief' (augment.)

*Dis-cl SYLL-CONT

SON DEP-V *CPX MX-V NOCODA

1.tSa.kr<a>ón *! * * *2. tSa.k[a].r<a>ón * * *

This constraint will also take care of forms such as /sórk/, which are prevented from surfacing faithfully because of the bad cluster involving two dorsal elements that would otherwise surface (cf. Section 2.6.3). Since the constraint seems to apply simply to a sequence of [kr] or [rk], and considering that [r] in this language seems to have a dorsal component, the constraint should probably be stated in terms of an OCP restriction on sequences of dorsal elements in Mantuan. This constraint must be at least as high-ranked as Syll-Contact and Son, but it is unranked with respect to them.

(47) The /sórk/-type of forms weeded out by *Dis-cl. /sórg/ 'mouse' *Dis-cl SON DEP-V *COMPLEX MAX-V NOCODA

1. só.r[a]k * *2. sórk *! * *

Here, finally, the candidate with the epenthesized vowel wins, because it saves the input from surfacing with a cluster contravening *Dis-cl. This would happen if the second candidate, the most faithful one to the input, were to surface.


The final Mantuan ranking for the epenthesis hypothesis is therefore

(48) MAX-{MORPH}, *CCC#, SON , *DIS-CL, SYLL-CONTACT >> ALIGN-R, C]PW, MAX-C, DEP-V >> *COMPLEX, MAX-V >> NOCODA

This solution accounts for all the data, although some further mechanism is required in order to explain which epenthetic vowel is to appear (only mentioned, but not actually implemented in the ranking). Word-final apocope is driven by alignment (an example of how it would act upon base forms can be seen in tableau 38 above “kardu”, although that example belongs to a type of deletion, the ranking would be the same as in the epenthesis solution).

Furthermore, words such as ladr / ladar cannot be contrastive as inputs, and ladar is always the surfacing candidate. Likewise, in this solution, the outcome [tSakarón] can surface from an input with an underlying vowel, i.e. /tSakarón/, or without, as /tSakrón/. However, both inputs /ladrón/ and a plausible but unattested /ladarón/ will surface faithfully. This could point to the real limitation of the epenthesis hypothesis for Mantuan, since the non-existence of the trisyllabic forms, as well as the great number of the “syncopated” forms, seems to support the fact that the lack of forms such as [ladarón] cannot be an accidental gap.

2.9 Conclusion: Assessment of Deletion and Epenthesis

From this study, it could be said that - at least synchronically - there are two independent phenomena in the language and it is important that output configurations accommodate both. Word-internal vowel ~ 0 alternations can be explained EITHER 1a) by syncope word-medially in derived forms, OR 1b) by epenthesis word-finally in base forms. Whichever explanation applies word-medially is independent from apocope, which occurs word-finally in all relevant cases, and constitutes the second phenomenon.

In the deletion hypothesis, the V ~ 0 alternations are ascribed to deletion (word-medial syncope) in derived environments (such as diminutives) and apocope word-finally. Both these phenomena are driven by a high-ranked placement of *STRUC(), and avoided when disallowed clusters would arise.

In the epenthesis hypothesis, the two phenomena are word-final epenthesis in base forms and word-final apocope. Notice that the first phenomenon is driven by sonority concerns or other well-formedness constraints, and the second by alignment. Apart from being a less economic solution which does not connect the motivation for both phenomena, it also fails to account for the lack

of non-syncopated forms such as [ladarón], unattested in Mantuan. The hypothesis should therefore be rejected.

This however does not mean that Mantuan has no epenthesis as a mechanism embodying a “repair strategy”, as will be proven by the following tableau, but only that *STRUC(), as the motor behind word-medial syncope, is needed as an active force in the language to explain the V ~ 0 alternations.

Consider the following tableau of a word, “pediatrician”, that can and is commonly imported from standard Italian (/pediátra/ masc. sg.) into Mn ([pediátar] masc. sg.). Of the hierarchy established for deletion, only the relevant constraints are shown; incidentally, this proves that *Complex outranks Max-V, rather than being unranked with respect to it (cf. 49). If it were unranked, candidate 3 would tie with 1 over the hierarchy *Complex, Max-V and 3 would wrongly lose because of a lower ranked NoCoda violation that 1 does not incur. This new ranking has no consequence for all previous tableaux.

(49) Epenthesis AND deletion in Mn./pediátra/ MAX-

MORPHSON *STRUC() *CPX MAX-V NO

CODA

1. pe.diá.tra *** *!2. pe.diátr<a> *! ** * * *3. pe.diá.t[a]r<a> *** * *

The faithful candidate does not surface because of an extra *Complex violation. Candidate 2 reverses sonority and is doomed, notice that it does not incur a Max-morph violation because the noun is masculine singular and word-final -a# is only a significant morpheme in Mantuan in feminine singular nouns. Candidate 3 surfaces in Mantuan with both an epenthetic segment and a deleted one.

The result would be different if the input were “female pediatrician”: in this case, as shown in t. 50, the input has the same form, but candidates 2 and 3 delete a significative morpheme and therefore the faithful candidate is in a position to surface.

(50) Importance of the significative morpheme -{a}#./pediátra/ f. sg. MAX-

{MORPH}SON *STRUC() *CPX MAX-V NO

CODA

1. pe.diá.tra *** *2. pe.diátr<a> *! * ** * * *3. pediát[a]r<a> *! *** * *

Thus, both epenthesis and deletion are seen in Mn in different environments, but it is the action of the *STRUC() constraint that explains Mn V ~ 0 alternations.

Appendix to Chapter II: A Digression on Italian Syllable Structure

1. Italian syllable structure ranking

Crucially, Italian must have the reverse order of what we have seen to be the most accepted syllable structure constraints for Mantuan, where *STRUC() >> *COMPLEX; if it were the same, the wrong candidate would win.

(1) Wrong result: Italian (with wrong domination hierarchy). /kárdo/ *STRUC() *COMPLEX

1. .kárd.<o> * *2. .kár.do. * ! *

If the two constraints were unranked with respect to each other, the two candidates would tie (even NOCODA would not help, since both candidates have one mark there), and 1 would have to lose via a violation of MAX-V. Given that we are assessing the relative tolerance of the two dialects to clusters, this does not seem to be the right way to proceed. The ranking between *STRUC() >> *COMPLEX should instead be reversed:

(1a) Italian relation between consonant clusters and number of syllables. /kárdo/ *COMPLEX *STRUC()1. .kárd.<o> *! *2. .kár.do. **

A ranking between *COMPLEX and NOCODA cannot be established, since in evaluating codas, where there is a violation of *Complex, there is always also a violation of NoCoda. Tableau 2 proves this, since candidate 1 would always be the loser, regardless of the ranking between these two constraints:

(2) Italian relation between consonant clusters and codas. /kárdo/ *COMPLEX NOCODA

1. .kárd.<o> *! *2. .kár.do. *

On the other hand, NoCoda must be ranked higher than *Struc(), as shown in tableau 3, where candidate 2 is the winner although it has two syllables, over shorter candidate 1 with one NoCoda violation:

(3) Italian relation between syllables and codas. /máre/ NOCODA *STRUC()1. .már.<e> *! *2. .má.re. **

Codas in Italian are limited to sonorants, [s], and geminates. The final ranking for Italian is:

(3a) *COMPLEX , NOCODA >> *STRUC()

Lexicon Optimization (Prince and Smolensky 1993) proposes that the shape of the lexical items themselves bears traces of the grammar of a language (i.e. of the ranking of its constraints). This may entail that the input posited for some visible output should be the most harmonic: the winning candidate should deviate the least from the input. Given the actual output [kárdo], an input /kardo/ will not violate either MAX-V or DEP-V, whereas /kárd/ will always surface through a candidate that violates DEP-V. It is most harmonic to posit /kárdo/, therefore, as an input. Along the same lines, /kárd/ will be taken to be the Mantuan input, and not /kárdo/, which would always incur a MAX-V violation, given the actual output [kárd]. However, in Mantuan it is important that whichever the input, the output always be of a C(C)V(C)C# shape. Conversely, in Italian an input ending in -V(C)C# should always surface as VCC[V]#, which could show that in Italian DEP-V is low ranked.

Considering that codas in Italian syllables are mostly restricted to sonorants (Itô 1988, Morelli 1995) the Syll-contact constraint is expected to be high-ranked in Italian as well. MAX-V is also expected to be high ranked, since under normal circumstances the winning candidate is the most faithful to the input. Consider the following loanword16 ('feldspar') and the standard pronunciation [feldspáto]: it can be evinced that Max(-C and -V) is higher than either two of the other constraints, and forces their violation in standard Italian:

(4) Role of Max in Italian standard.

16 ? Loanwords have been chosen because they present more complex clusters than native Italian words.

/felds’pato/ MAX SYLL-CONTACT *COMPLEX

1.fel<d>.spato *! *2.feld.spato * **

Candidate 1. violates the top-ranked constraint and does not surface, candidate 2. is maximally faithful to the input and wins in spite of multiple violations of lower ranked Syll-contact and *Complex. The hierarchy reached for standard Italian is then:

(5) MAX >> Syll-contact, *COMPLEX , NOCODA >> *STRUC()

Compare with Mantuan (only comparable constraints shown):

(6) SYLL-CONTACT >> MAX-C >> *STRUC() >> *COMPLEX >> MAX-V >> NOCODA

For the sake of completeness, it should be added that a common pronunciation heard on the news poses an interesting question for the placement of Max-C in Italian (and of Dep-V, cf. below): it is common to hear, in fact, [felspato]. This must mean that, in this variety (that will be called “central” here, since it is most likely to be equated with the central Italian dialect spoken in Rome and therefore treated as standard for prestige considerations), Max-C is ranked lower than *COMPLEX:

(7) A common non-standard pronunciation (Central Italian)/feldspáto/ SYLL-CONTACT *COMPLEX MAX-C1.fel<d>.spáto * *2.feld.spáto *! **

Independently from how we judge the syllable contacts, candidate 2 always incurs at least one violation of Syll-contact and two of *Complex; according to this micro-grammar, 2 should always fail to surface and provides no ranking argument for Syll-contact with respect to *COMPLEX. Max-C is ranked lower than *Complex in this dialect.

The same dialect presents a parallel issue with respect to Dep-V placement: consider the following pronunciation of the word 'psychology', psicologia, pronounced by some speakers as [pisikolodZía]. In tableau (8), all input segments surface faithfully in both 1. and 2., granted that /psikolodZia/ is the input. But *COMPLEX is high-ranked, and in 1 an epenthetic [i] is inserted to

break up the disallowed cluster. This implies in turn that a DEP-V violation has occurred, but that this constraint is ranked lower. 2 does not surface because of a *COMPLEX violation. 3 shows that faithfulness is ranked high enough for the cluster not to be simplified by not parsing one of its segments (a violation of MAX-C).17 There is also a clear asymmetry between onsets (at least word-initial ones), where all segments are preserved and clusters are avoided by epenthesis, and codas, where it is better to delete an input segment than to epenthesize, in order to avoid a complex cluster (as in [felspato]).

(8) Dep placement in “central” Italian:/psikolodZia/ *COMPLEX MAX-C DEP-V1.p[i]sikolodZia *2.psikolodZia *!3.sikolodZia *!

The ranking in this dialect must have MAX-C higher-ranked than DEP-V. Another tendency of central Italian speakers is to favor CV syllables to

avoid codas in general. This is seen in imperfect second language acquisition of English, where those speakers tend to round off word final (and word internal) codas adding a schwa (or an [e]) and syllabifying the offensive coda as an onset: 'don't' as [don.t´], and 'break-through' as [bre.k´.ru]. It is also noticeable in their pronunciation of loanwords found in Italian: 'ex-' ['e.ke.se].

NOCODA, in reality, must also be higher than DEP-V, or /eks/ would not surface as [ekese], but *STRUC() is unranked with respect to DEP-V: they must both be low since epenthesis is resorted to, and epenthesizing a vowel as exemplified here means de facto adding an extra syllable. Thus:

(9) Central Italian ranking: SYLL-CONTACT, *COMPLEX >> MAX-C >> NOCODA >> DEP-V, *STRUC()

Cultivated speakers from the central region could have both a central and standard Italian grammar and allow for some “free variation” (due to this kind of bilingualism). It would be interesting to assess whether the placement of Max-C is just as high for “aboriginal” clusters in this dialect, or whether it reflects the

17 It should be noticed that a candidate like 3 would win in, for instance, Mexican Spanish given a comparable input /psikoloxia/, entailing that in that language the hierarchy between MAX-C and DEP-V must be reversed: DEP-V >> Max-C (better to delete an extra segment than to insert epenthetically).

partial adaptation of foreign strata of vocabulary into Italian (cf. Fukazawa 1996 for similar issues in Japanese).

In standard Italian, the situation must be as exemplified in (10):

(10) Dep-V placement in standard Italian:/psikolodZia/ MAX-C DEP-V SYLL-CONT *CPX

1.p[i]sikolodZia *!2.psikolodZia *3.sikolodZia *!

It is equally bad to violate Max-C or Dep-V in the standard (no relative ranking is attainable between the two, therefore), as demonstrated by the two losing candidates 1 and 3. On the other hand, candidate 2 is maximally faithful to the input, and surfaces in spite of its violation of *Complex.

Summarizing the ranking for central (11) and standard Italian (12):

(11) Central Italian ranking:SYLL-CONTACT, *COMPLEX >> MAX(-C) >> NOCODA >>DEP(-V), * STRUC( )

(12) Standard Italian :MAX-C, DEP-V >> SYLL-CONTACT, * COMPLEX , NOCODA >> * STRUC( )

This proves that in the standard, maximum faithfulness to the input is very important, followed by the rules governing the propriety of consonant contacts, possibly avoiding clusters. The number of syllables and the existence of codas are ranked lowest. Basic syllable structure constraints are underlined. This compares with significant differences in Mantuan:

Mantuan ranking summary:

(13) SYLL-CONTACT >> MAX-C , (DEP-V) , * STRUC ( ) >> *COMPLEX >> MAX-V >> NOCODA.

chapter 2violagmiglio.net/violas_site/papers_files/fincap2.doc · web viewthe distribution to be...

Documents