Irregular Morphology and Athematic Verbs in Italo-Romance

The article deals with irregularities in the morphological make-up of Italian verbal forms, focusing on perfect and past participle forms. It aims to account for root-based contextual allomorphy in the framework of Distributed Morphology (Halle & Marantz 1993). Building upon the generalisation that morphological irregularities result whenever the thematic vowel is absent, the article provides a synchronic account and a diachronic analysis by means of a restricted set of morphophonological rules, thus challenging both the traditional view, according to which morphological irregularities follow from stress positioning, and paradigm-based accounts.


Introduction
In a series of recent papers (Calabrese 2012(Calabrese , 2013(Calabrese , 2015)), I have investigated the morphophonology of Italian irregular perfect and past participle forms * .In the present article, I clarify, revise, simplify and integrate the main aspects of the different proposals made in them.
When we talk of irregular morphology, we are dealing with morpheme-specific morphology, i.e., with situations in which morphological operations are dependent on morpheme specific information.In (1) I contrast a case of irregular morphology with a case of regular morphology.On one hand, we have the Italian Imperfect marker, which is regular in being always the same across verbs.On the other hand, we have the Italian perfect marker /s/, which appears only with certain verbal roots.In the case of this marker, we need a special vocabulary item that includes reference to root information in the structural description.No such contextual restrictions are needed for regular morphology. (1) Regular morphology Irregular morphology Italian Imperfect marker Italian perfect marker /s/ amavo/battevo/partivo persi /-v-/ ↔ [+imperfect] /-s-/ ↔ [+perfect]/ root s ___ (root s =perd, etc.) One of the most typical case of irregular morphology involves morphological operations dependent on root specific information.Morpheme exponence dependent on root specific information can be referred to as root based contextual allomorphy.Rootbased contextual allomorphy is accounted for by: vocabulary items (VI) and morphophonological (MP) rules 1 including root-information in their structural description.
In my work on irregular Italian perfect and past participle forms, I have observed a striking correlation between presence vs. absence of regular morphology and presence vs. absence of thematic vowels, 2 respectively.Root based contextual allomorphy occurs only when the thematic vowel is absent.
[ [[[perd] root __ ] V -s-] T ] o] AGR [[[part] root -i TV -] V -t-] T -o] AGR perso 'lose-PST.PRT-MSG' partito 'leave-TV-PST.PRT-MSG' Athematic Thematic It thus appears that root-based contextual allomorphy is observed only in athematic morphology.I have accounted for this basic fact by assuming, following Embick (2010), that the transmission of information necessary for morphological operation application, and more generally any morpheme-to-morpheme interaction, can occur only in a local configuration, where locality involves linear adjacency, as stated in the principle below: 3(3) a.
α, β are local if no overt node intervenes (linear adjacency).Now, whereas in the case of thematic perfect/ past participle we have the structure in (4), in the case of the athematic perfect/past participle we have the structure in (5), where I assume that the Thematic vowel has not been inserted (see below for discussion).Given (3), the tense morpheme can interact with the root in the structure in (5) but not in the structure in (4).As shown below, root information can be accessed only in the former structure (see (6)a) but not in the latter (see (6)b).Given that root information cannot be transmitted across the thematic vowel, only default, regular morphology can appear in this case.In this paper, I will revise the analysis of the irregular morphology characterizing these forms as it appears in my previous articles (Calabrese 2012(Calabrese , 2013(Calabrese , 2015) ) and investigate the development of Latin irregular perfects and participles into Italo-Romance varieties.The paper is organized as follows.After a brief introduction of Distributed Morphology (Section 2), I provide an analysis of Italian regular morphology (Section 3), thus introducing the basic conceptual tools that will be used in the following sections.Then I deal with irregular perfect forms (Section 4).As I already mentioned above, they are athematic because they do not undergo a rule inserting the verbal thematic vowel.I introduce the special VIs and MP rules that account for the allomorphy of the irregular perfect forms.In section 5, the allomorphy in the past participle forms is investigated.Section 6 will summarize the main aspects of the locality-based account of these forms.Section 7 deals with person-based allomorph alternations in the Italian perfect.In section 8, I will deal with alternative analyses of allomorphic alternations found in these irregular verbal forms.Discussion of the development of the Latin perfects and past participle into Italian and Italo-Romance varieties (Sections 9 and 10) will provide further evidence for the analyses provided here.
The paper shows that an adequate analysis of the allomorphy found in Italian irregular perfect and past participle forms, and of its historical development, can be achieved by using simple and motivated morphosyntactic structures, standard morphological segmentation, morphemes (vocabulary items) and morphophonological rules.This provides evidence for models using morphemes organized in a syntactic structure where locality principles can govern morpheme interactions.

Distributed Morphology
The theory of Distributed Morphology (DM; Halle & Marantz 1993) proposes a piecebased view of word formation, in which the syntax/morphology interface is as transparent as possible.It crucially incorporates hierarchical structure into morphology; essentially, it assumes the input to morphology to be syntactic structure, as in (7).
Features (or feature bundles) are distributed over nodes forming morphemes, which in turn are subject to Vocabulary Insertion rules that add phonological material (exponents) to these morphemes.
Contextual allomorphy is accounted for by vocabulary insertion, on one hand, and through the application of MP rules, morpho-syntactically conditioned phonological rules, and plain phonological rules, on the other. (7) The Grammar Syntactic Derivation

PF LF
Allomorphic interactions are constrained by the manner in which Vocabulary Insertion operates, and by the interaction of linear and cyclic locality conditions.Two different conditions are relevant in this paper.The first of these conditions (8) enforces "inside out" cyclicity (e.g.Halle andMarantz 1993, Bobaljik 2000): Vocabulary Insertion proceeds cyclically from the lowest element in the structure outwards.

Basic Properties of Italian Verbal Morphosyntax
The basic morpho-syntactic structure of Italian verbs is given in (10).It is generated by verb raising to T, morphological merger between V and T and AGR insertion (see Halle and Marantz (1993): As proposed by Oltra-Massuet (1999), Oltra-Massuet and Arregi (2005), Embick and Halle (2005), every functional/lexical projection in Latin and Romance has a 4 A morphosyntactic change occurred in the development of the Romance languages as can be seen in (i), where I compare the Latin pluperfect subjunctive in (ia) with the form that historically derived from it in Italian, i.e., the Imperfect subjunctive (ib): (i) a. laud -a: + u-i + s + s-e: + mus 'praise-PRF-PST-SBJ-1PL' b. lod -a-+ss-i-+ mo 'praise-IMP.SBJ-1PL' In Italian, functional categories such as aspect, tense and mood are no longer represented as independent morphological pieces as they were in Latin.Instead, a single morpheme appears in their place.I will simply assume that the Asp, Tense and Mood nodes are fused together in Italian (i.e., Tense=Aspect+Tense+Mood).Thematic Vowel.Thematic Vowels (TV) are special morphological elements adjoined to certain functional heads in morphological structure by the rule in (11): (11) X 0  X 0 X 0 TV After thematic vowel insertion, the structure in ( 10) is changed into that in (12). ( (12) accounts for the morphological structure of the imperfect forms in ( 13): batt-e-v-o batt-e-v-i batt-e-v-a batt-e-v-a-mo batt-e-v-a-te batt-e-v-a-no PARTIRE 'leave': part-i-v-o part-i-v-i part-i-v-a part-i-v-a-mo The vocabulary items needed to account for verbal inflections in the Italian present and imperfect tenses of regular verbs are given below.(In the case of the thematic vowels in (15) the Head can include a root or the head of a functional projection.This accounts for the parallel behavior between roots and inflectional heads; so both the root /am-/ and imperfect tense have /-a-/ as a thematic vowel): (15) A phonological rule that is very important in accounting for the surface distribution of the Thematic Vowels is (20), which deletes a TV before a suffixal vowel: Let us turn to the perfect of regular verbs.Also in this case, no overt exponent for Tense and the associated thematic vowel are present.
(22) AMARE 'love': am-a-i am-a-sti am-ò am-a-mmo am-a-ste am-a-ro-no BATTERE 'beat': batt-e-i batt-e-sti batt-è batt-e-mmo batt-e-ste batt-e-ro-no TEMERE 'be afraid' tem-e-i tem-e-sti tem-è tem-e-mmo tem-e-ste tem-e-ro-no PARTIRE 'leave': Given that both the present and the perfect display a null exponent, we can assume that this is the elsewhere tense VI: The vocabulary items for the AGR terminal node in the Perfect are given in ( 25): (25) VIs for the AGR morpheme in the Perfect.
Given that stress plays an important role in analyses of verbal forms in Italian, I provide a brief discussion of it.
Stress in verbal forms in Italian is predictable from the morpho-syntactic structure of the string (cf.Otra-Massuet and Arregi (2005) on Spanish): if one put aside special AGR endings such the future/conditional (batterò, batteréi), stress falls: either on the TV preceding the AGR suffixes of 1 st and 2 nd Pl. (battiámo, battevámo) or on the TV preceding Tense (battéva, battésse, battè).Otherwise stress falls on the root: i. if the TV is absent (pérsi, pérsero); ii. in the present (both indicative and subjunctive: báttono, teléfonano) (see Calabrese (2012) for more detail and a formal analysis of the stress patterns in Italian verbs)

Irregular Perfect forms
We can now discuss irregular perfect forms.They show an overt morpheme for this tense.
( As discussed in Calabrese (2012Calabrese ( , 2015)), the crucial aspect of the structure in ( 27) is the absence of the Verb Thematic Vowel.5 (28) Regular vs.
I assume that they are exceptions to the TV insertion rule. 6These roots are assigned a special diacritic [-TV] indicating that the rule in (11) does not apply when verbal head contains one of these roots.Thus, no verbal TV is inserted.
In addition to /-s-/ (cf.val-e/val-s-e 'be worth pres/Perf'), irregular forms of the perfect display other exponents ( 29)-( 30).( 29 e, the regular exponent=( 23)) (31)b) triggers gemination.The skeletal position is filled in by the preceding consonant.The floating secondary labial articulation is attached to the place node only when the preceding consonant is dorsal ( cf. tatʃeva/takk w i) The surface phonological shape of the roots in the irregular forms of the Perfect also requires the application of MP Rules .
( These changes are implemented by dedicated MP Rules, for example, ( 37) and ( 36) are needed to account for ( 33) and ( 34), respectively (see Calabrese (2013) for discussion of the other rules needed to account for the irregular perfects allomorphy) (37) Coronal stop deletion: Coronal stop deletion applies before Consonantal assimilation.Sample derivations for some 3 rd pl.irregular perfect forms: (39) a.
[ [[[ val ]  The complex allomorphy of the Italian irregular perfect forms can, therefore, be synchronically derived in very simple ways.

Past Participle
The class of verbs that display irregular morphology in the Perfect may display it also in the past participle.Also in the case of the Past Participle the irregular forms are athematic.The regular exponent for the Participial morpheme is /t/ : Irregular forms may display /-t-/ (see (41)a) or they may display /-s-/ (see (41)b): (41) a.All roots that are athematic in the past participle are also athematic in the Perfect.The reverse does not hold.Thus, there are roots that are athematic in the Perfect but not in the Past Participle.For example, all the athematic roots that take the geminating exponent in (31)b) are systematically thematic in the past participle.
( The few athematic roots that have the exponent Ø in the perfect such as feci, vidi have an athematic past participle with /-t-/ (the /s/ of visto is due to a MP rule discussed below): (44) fetʃi fatto vidi visto If we exclude the roots in ( 43) and ( 44), we can postulate that if a root is athematic in the past participle, regardless of whether the exponent of the PP is /-t-/ or /-s-/, then it will have /-s-/as the exponent of the Perfect (cf.Vogel (1994), Calabrese (2015)): (45) a.
valere 'to be worth' PP: valso PA valsi 'I was worth, b.
scuotere 'shake' PP: skosso PA: skossi 'I shook Specifically, if a root takes /s/ in the Past Participle, one predicts that it will take /s/ also in the Perfect.Simply, the roots that take /-s-/ in the Past Participle are a subset of those taking /-s-/ in the perfect.
(46) kyudeva/kyuso/kyuse 'know' fondeva/fuso /fuse 'melt' korreva/korso/korse 'run' mwɔveva/mosso/mɔssi 'move' Note that once the past participle takes the suffix /-s-/, then it will also have the same allomorphy as the Perfect: The similarity in distribution shown above between the Perfect and the Past Participle excludes the possibility that there is simply an accidental homophony between the stem of these forms.There must be a deeper relation between them.
Thus, as proposed in Calabrese (2015), I assume, following Ippolito (1999), that the Past Participle has the structure in (47) with an aspectual [+perfect] feature.
A crucial feature of her analysis of the participle involves reference to the morphosyntactic context in which T occurs.In particular, she proposes that a [+perfect] T has finite morphology, i.e., it is a "perfect" form when it is dominated by CP; otherwise it acquires participial morphology.For the sake of simplicity, I will assume the rule in (48), which assign the feature [+participle] to a [+perfect] T when not dominated by CP.
(49) T The diacritic [-TV] blocks the insertion of the verbal TV also in the irregular past participle.Insertion of the TV in regular ones generates (50): TV +perfect +participle The Vocabulary Items relevant for the Past Participle are as follows: (51) a. s ↔ [+perfect, (+participle)] T /Root S __{Root S = val, scriv, muov, etc.} b. t

↔ [+participle]
(51)a) is the same VI as (31)a).The round parentheses allow the feature [+participle] to be optionally present so that this VI can be in competition, not only with (51)b) but also with the other VI of the Perfect in (31).The fact that when the participle takes the suffix /-s-/, it will also have the same allomorphy as the Perfect readily follows from the fact that MP Rules such as (37) and ( 38) simply apply before suffixal /s/ regardless of whether or not the /s/ is the exponent of the Past Participle or of the Perfect.
(52) Thematic past participle T There are restrictions on the distribution of irregular past participles.The roots with the diacritic L , which are athematic in the Perfect, and therefore must also have the diacritic [-TV] lose this diacritic in the Past participle by means of the impoverishment operation in (54).8Therefore, they are thematic in the past Participle.
(57) vinsi vs. vinto tolsi tolto skonfissi skonfitto This accounts for why the roots that take /-s-/ in the Past Participle are a subset of those taking /-s-/ in the perfect.The morphophonological rules in ( 37) and ( 38) also apply in the past participle, with some peculiarities, for example, rule (37) deleting coronals before /-s-/ is subject to some exceptions in the Past Participle.Thus, it does not apply to met 'put', so that this root undergoes /s/-assimilation in ( 38 Other rules are discussed in Calabrese (2015).

Locality
We can now account for the generalization that the presence or absence of regular morphology correlates with presence or absence of thematic vowels, respectively.As discussed in the introduction, irregular inflectional VIs crucially require access to root information, cf. the VI for the exponent /-s-/ of the Perfect in (31)a), repeated here as ( 62): (62) /s/ ↔ [+Perfect, (+participle)] T / Root s __ (Root s = corr, etc.) Assuming the principles in ( 8) and ( 9), repeated here as ( 63)-( 64), respectively, linear adjacency is required between root and tense morpheme.
(63) Vocabulary Insertion proceeds cyclically from the lowest element in the structure outwards.
Due to these locality requirements, (62) can apply only when the Thematic Vowel is missing.
(65) a. Thematic Perfect/past participle b.Athematic perfect/past participle: In fact, given the two conditions in ( 63) and ( 64), the presence of the thematic vowel prevents application of such VI.When the cyclic application of vocabulary insertion reaches T, root information cannot be accessed because the root is not linearly adjacent to T due to the presence of the thematic vowel.Hence, (64) blocks application of (62).So it can apply in the structure in (66) but not in (68) where the TV is present.Thus forms such as those in (( 66)-( 67) are licit; forms such as those in ( 68)-( 69) are illicit: (66) a.
[ The unmodified form of the root, therefore, appears before the thematic vowel since the rules introducing irregular allomorphy cannot apply.If the TV is present, only regular VI can be inserted. (72) a.

Allomorphic alternation in the perfect forms
The irregular perfect verbal forms are characterized by distinctive pattern of alternations: ( We observe a regularity in the irregularity: In the 1 st , 3 rd sg and 3 rd pl there is an athematic form of the verb, thus the root appears in a idiosyncratic irregular form.In the 1 s , 2 nd pl. and 2 nd sg, there is a thematic form of the verb, thus the form of the root is that regularly appearing with other tenses (76) Irregular: [[[perd] root ] V -s-] TNS ] i] Agr vs.Regular: [[[perd] root -e TV -] V -Ø-] TNS -i] Agr Athematic Thematic 10 There is a single exception to this generalization in Italian: vissuto, the past participle of vivere 'live' (cf.perf.vissi) where a irregular stem form is found in a thematic context.See below for discussion.
I have account for this alternation as follows.An important operation in Distributed Morphology is Impoverishment (Bobalijk 2003, Bonet 1991, Halle 1997, Halle and Marantz 1993, Harley 2008, Nevins 2011, Noyer 1992, 1998(but also Calabrese 1994, 2008 for critical discussion of some of its uses).Impoverishment deletes or removes features from the morpheme in a terminal node.By doing this, it blocks the insertion of a more specific Vocabulary item and allows the insertion of a less specified one.This results in what we can call the retreat to the general case In Calabrese (2012), I proposed that impoverishment can also delete diacritic indices thereby preventing the application of morphophonological rules.Specifically, I proposed that in the second singular and plural and in the first plural, the special root index blocking the insertion of the TV by rule ( 11) is deleted by the impoverishment rule in (77).Once the diacritic [-TV] is deleted, the TV insertion rule is no longer blocked from applying, and the verbal thematic vowel is regularly inserted in the structure.The presence of the thematic vowel prevents the application of the special VIs and morphophonological rules.Only regular unmarked morphology will occur.
We can now account for the difference between the 1 sg.and the 1 st pl. of the verb /scriv/ as follows.In ((79)a), /scriv/ has the diacritic [-TV] that blocks the application of the TV insertion (11) as in (79)ii).The exponent /-s-/ for the T +perfect can be inserted by ((31)a) and rule (34) can apply so that we get (79)iii ).
[ [[ skris ] root s ] T i ] AGR skriss i In (80) impoverishment deletes the diacritic [-TV] as in (ii).TV /e/ is inserted, the regular exponent of the past tense must be inserted and no MP rules may apply.

[[ skriv ] root e] Ø ] tns mmo] AGR skrivemmo
In Calabrese (2012) I proposed that impoverishment in this case, as a historical innovation, is due to the principle of compensation (Brøndal 1940(Brøndal , 1943)).This principle disfavors the cumulation of idiosyncratic exponence in words containing marked categories.This is the case in the morphological context [+perfect, +participant, +plural].Special vocabulary items and readjustment rules create idiosyncratic exponence in the Perfect, a marked morphological category.By removing the lexical diacritic blocking the insertion of the thematic vowel, impoverishment prevents the appearance of idiosyncratic exponence in the 1st and 2nd singular and plural, which are also marked, both phonologically (the geminate [mm] in the 1plur and the cluster [st] in the 2 sg and pl) and morphologically (see discussion of the markedness of the 1 and 2 pl in Calabrese 2011Calabrese , 2012)).
Thus, only regular morphology appears in these morphological contexts.
It is to notice that the alternations we observe in the perfect forms are part of one of the most characteristic general patterns governing Italo-Romance verbal morphosyntax: the tendency to avoid idiosyncratic, "marked", exponency in the first and second plural.Exponents in these two persons tend to be syncretic, to disappear (=be defective), or to display regular morphological behavior.This is what we can call the 1 st and 2 nd Plural conspiracy (Calabrese 2011(Calabrese , 2012): (81) a.
In the many Italian dialects, the exponent of the 1 st plural oblique clitic was replaced by the exponent of a locative.The same thing happened for the 2 nd plural one.b.
In the other Italian dialects, the exponent of the 1 st Plural oblique clitic was replaced by the exponent of a partitive ( < Latin INDE).e.
In the Campidanese dialect of Sardinian, the exponent of the reflexive clitic /si/ has also become the exponent of first and second plural oblique clitics.c.
In the Tuscan dialect of Lucca, the exponent of the reflexive/impersonal clitic /si/ has also become the exponent of 1 st plural oblique clitic.d.
In Tuscan the 1 st plural subject inflectional suffix is replaced by the reflexive/impersonal clitic /si/.The verb appears in the 3 rd sg.f.
First and second plural subject clitics undergo syncretism or are missing in most varieties with subject clitics.g.
In all the paradigms characterized by irregular stem allomorphy, regular stem allomorph are found in 1 st and 2 nd plural verbal forms.For example in the following forms, special morphological operation apply in all persons except 1 st and 2 nd (and in the forms that are always regular: the imperfect and the infinitive): morphophonological rules in (82)a-b, insertion of extension /-isk-/ in(82)c, suppletion in (82)d), irregular deletion of imperfect marker ((82)e).
( As discussed in Calabrese (2011Calabrese ( , 2012)), the 1 st and 2 nd plural conspiracy effects can be accounted for by assuming that i) these two persons are marked 11 (see Calabrese   11   This does not mean that there is no markedness difference between the 1 st and 2 nd plural.In fact, the markedness effect seems to be stronger with the 2 nd plural.For example, there are cases in which it is only the 2 nd plural that displays regular allomorphy: e.g., voglio, vogliamo, volete 'want-PRS.1SG/1PL,2PL(cf.voleva/volere want-IMP3SG/INF), posso, possiamo/potete (cf.poteva, potere) 'be able'.See also the irregular perfect in some southern Italian varieties where the 1PL displays irregular morphology like the 1SG, 3SG and 3PL (Mussomeli: sappi, sappi, sappimu, sappiru 'know-PRF1SG/3SG/1PL, 3PL', but not the 2PL (and SG.): sapistivu/sapisti (2011) for arguments) and that idiosyncratic exponence (including contextual allomorphy) for marked categories tend to be removed in historical changes (see Calabrese (2008)). 12Different repairs may implement this removal: syncretism and deletion account for (81)a-f.(seeCalabrese (2011) for more discussion); the cases in (81)g can be instead accounted for by assuming that they involve a different type of repair, the one discussed in this paper: impoverishment of the diacritics triggering contextual allomorpy: MP rules in (82)a-e, and (82)g, and suppletion in (82)f (Calabrese 2012). 13

Alternative accounts
In this section, I will address alternative accounts of the distribution of irregular stem allomorphs in perfect and past participle forms, and show that they cannot be maintained.
I will begin with the traditional phonological account (Buchholtz 1889, Meyer-Lübke 1972, Lausberg 1976, Rohlfs 1966, Tekavcic 1980) of the development of person-based alternations in Italian perfect forms.This account assumes that the irregular perfect stem alternants must be stressed on the root, and derives this property from the development of the Perfect exponent /-u-/ of Latin in postconsonantal position, i.e., in athematic stems.In particular, it is hypothesized that this /u/, after becoming the labio-velar glide [w], triggered gemination of the preceding consonant.Crucially according to this account, this occurred only after stressed syllable, the labio-velar glide was instead deleted in pretonic positions: Therefore, when the roots of the athematic Perfect were not stressed, they became identical to the basic stem allomorph found in the imperfect, in the infinitive, and other regular forms: (84) ven-ísti cad-ésti vol-ésti ven-ívo cad-évo vol-évo ven-íre cad-ére vol-ére The verbs in which this occurred were not many (venire, cadere, volere, avere, sapere, nuocere, tacere) There are several problems with this account.In the first place, as Maiden (2000) observes, it is unclear why the labio-velar glide should cause gemination only after stressed vowels but not after unstressed ones.There are no phonological reasons for such asymmetry.In the same way, it is unclear why there should be deletion of the labio-velar glide after unstressed vowels.Observe at this regard that in nouns, differently than in verbs, consonants are geminated before a labiovelar glide also after unstressed vowels: (87) januárius gennáio 'January' manuária mannáia 'cleaver' The evidence for lack of consonant gemination after unstressed vowels is therefore provided only by the evolution of the irregular perfect forms.The traditional explanation, therefore, is fundamentally circular (Maiden (2000).
Also notice that if the changes accounting for the development of the forms in (83) were only phonological, we should expect Italian cadisti, volisti (cf.venisti) with preservation of the vocalic quality of the Latin Perfect Thematic Vowel, instead of the actual cadesti, volesti where we see the appearance of the Verb Thematic Vowel.This can be explained only by assuming some form of morphological restructuring, which involves a complication of the purely phonological account that is assumed for the development of the forms in (83).
Furthermore, note that the 1 st pl. is stressed on the root vowel in Latin.Therefore, according to the traditional account, one should expect gemination in this case in Italian, contrary to the facts.
As in the preceding case, here too, we need to resort to a morphological account, which, therefore, undermines the assumption that the development of the forms in (83) was purely phonological.
Finally, assuming that the phonological account assumed traditionally is correct, it is unclear why a few alternations such as those in (83) could lead to such a robust generalization that caused analogical restructuring in all other verbs.One can wonder why analogical pressure did not work in the opposite way so as to remove the few marginal alternations by extending to them the irregular stem form.In conclusion, the traditional phonological explanation cannot be maintained.
Let us again consider the rhizotonic nature of irregular stems.As discussed earlier, traditional analyses of verbal irregular morphology postulate that stress plays a crucial role in the selection of the irregular stem allomorph: irregular stems/roots are inherently stressed.The same postulation, or a variation of it, is found in the most recent accounts of Italian verbal morphology (cf.Burzio (1998), Maiden (2000Maiden ( , 2010)), Pirelli and Battista (2000), Vogel (1994)).
Here I want to argue that the fact of being stressed is not the identifying property of irregular stems/roots, and that this property cannot be used to account for their distribution.The point is that verbal stress is fundamentally predictable from morphosyntactic composition of the string: as already mentioned in section 2, if one put aside the future/conditional where stress falls on the AGR suffix, it is a TV that is assigned stress: either the TV preceding the AGR suffixes of 1 st and 2 nd Pl. (battiámo, battevámo) or the TV preceding Tense (battéva, battésse, battè).Otherwise stress falls on the root: i. if the TV is absent (pérsi, pérsero); ii. in the present (both indicative and subjunctive: báttono, teléfonano) where one can argue that unmarked present Tense is actually missing (see Calabrese (2012) for more detail and a formal analysis of the stress patterns).Therefore, the fact that the irregular perfect and past participle stems are stressed follows directly from the fact that in the case of these stems the verbal Thematic Vowel, i.e., TV preceding Tense, is missing.Both the accentual and the morphological properties of these stems then follow from the fact that they are athematic.More generally, it is a fact that all generalizations regarding irregular verbal stem allomorphy in Italian crucially refers to the property of being athematic.For example, as discussed in section 4, the distribution of irregular stem forms in the past participle can be readily captured when one observes that all roots that are athematic in the past participle are also athematic in the Perfect.Notice that the generalization does not refer to the actual phonological shape of the stem which can be quite different: feci/fatto, tolsi/tolto, sconfissi/sconfitto, valsi/valso, but to an abstract property: presence vs. absence of the TV.What unifies these pairs is the fact that they are athematic, and therefore accessible to contextual allomorphy, which may generate the same stem forms (corsi/corso), but does not necessarily do so (tolsi/tolto).

Development of Italo-Romance perfects
We can now turn to the development of Italian perfects.The morpho-syntactic structure of Latin Verb (Embick and Halle (2005) is given in (90).A crucial assumption is that rule (11) inserting TV is active in Latin, as in Italian, the only difference between Italian and Latin, other than in the shape of exponents, is that Latin has a more complex verbal structure than Italian, aspect, tense and mood are fused in a single node (see note 3): Here I will focus only on the perfect morphology.The regular perfect forms are thematic: (98) Cluster simplification after liquid codas algeo alsi 'feel cold' mergo mersi 'merge' spargo sparsi 'spread' An important difference between the Latin Inflectum and Perfectum involves the infix /-n-/ which characterizes the forms of the inflectum in many verbs.Since this infixal /n/ may be absent in the perfect we have alternations like those in (99): (99) vinco vīci 'win' fundo fūdi 'pour/melt' rumpo rūpi 'break' frango frēgi 'break up' After this brief description of Latin perfect morphology, we can turn to the development of Italian perfect forms.We need both phonological and morphological changes.
Some of the phonological changes, which are not relevant to the analysis here, are the following: 1.
Consonant assimilation removed many consonant cluster of Latin which became geminate in Italian: Example: scripsi → scrissi, etc. 'write'

2.
In Latin, prevocalic, post-consonantal high vocoid /u/ (together with /i/) was syllabified as a syllabic nucleus (e.g.ve.nu.i, ka.du.i)--although not after /r/ (e.g.par.wi).A fundamental change of Proto-Romance was the elimination of hiatus configuration by glide formation which lead to the resyllabifications and gemination (see Calabrese (2013) for an account of this gemination process) Example: venui/kadui/takui → venni/kaddi/takk w i 'come/fall/be silent' We can now address the morphological changes, the main topic of discussion in this section: 1.
Reduplication was eliminated, i.e., the diacritic triggering Reduplication was impoverished.The null exponent characteristic of the reduplicated forms was replaced by /s/ or /v /.
( The application of morphophonological rules was extended to new verbs.For example, vowel fronting was extended to sap-u-i>sep-u-i> It. seppi.

4.
The final, and most important, change to account for the non-phonological differences between Latin and Italian, was a redistribution of exponents of the perfect, and specifically a drastic restriction in the use of the null exponent /-Ø-/ in athematic perfects: ( We can now turn to the development of Participle.In Latin the original suffixal exponent for the past participle inherited from IE was /-t-/ which is regularly preserved in thematic roots.A change affected it in roots with athematic past participle: In early Latin, clusters of heteromorphemic coronal stops were affected by a process that changed them into a geminate coronal fricative: tt--> ss.This geminated fricative was degeminated after long vowels: rīd-to  rīssu rīsu but mit-to  missu. At a certain point in preclassical Latin, the shortened /s/ resulting from this process was reanalyzed as a suffixal element.One can speculate that the /s/ of cases such as rīsu was reanalyzed as being the exponent of the Vocabulary Item in (94)b), which was already present in Latin.In fact, in Latin, we find many cases in which we have /s/ in the past participle instead of the etymologically expected /t/ (see also Vincent (1978) The development of the Italian Past Participle simply involves further expansion of the use of the Vocabulary Item in (94)b), to still other roots.Thus, /s/ is reanalyzed as being a single suffix both in Past Participle and Perfect forms, as I proposed in the analysis in (51)a), in section 4.
From what we see in ( 104)-( 106), we can conclude simply that irregular morphology in Italo-Romance varieties is restricted only to athematic constructions, as expected in the theory developed in this paper.Only in these type of constructions, in fact, do we have the local structural configuration that allows root based contextual allomorphy.Therefore, only regular morphology is possible when the verbal thematic vowels is present.
It is important to observe that there are indeed cases in which the special suffix /s-/ is found in thematic constructions.Crucially in this case, it has become the regular exponent of the perfect (Rohlfs (1968)) Thus, the suffix /s/ is found with regular verbs such as 'eat', 'sing ', and 'think' in  This is what one expects in the framework adopted here. 15 However, there are also problematic cases, those in (109).They are restricted to the past participle.In the case of these forms, an irregular stem allomorph is found in a thematic construction.These are exceptions to the generalization that irregular stem forms appear only when the TV is absent.
The misanalysis behind this innovative form, however, involves extension of the stem of the perfect to a thematic context (i.e., __ -u-), in other words, a case in which the complex root + Tense suffix has been incorporated in a suppletive root form: Thus, these forms show that this extension is potentially possible.What is striking is that this type of misanalysis is apparently rare even in dialects like Neapolitan where there appears to be more attestations.Irregular root morphology seems indeed to be associated with lack of the thematic vowel.What happened with these forms, however, needs to be better understood.

Conclusions
As discussed by Kiparsky (1968), historical changes in grammatical structures provide the best "window" on the actual composition of these structure in so far as we expect that the components of the structures play a role in the changes.
The morphological changes, we observe in the historical development of the Perfect and past Participle forms in Italo-Romance are most adequately accounted for when analyzed in terms of the following components: (i) Morphemes such as roots, thematic vowels, and Tense; (ii) Vocabulary Items, rules that add phonological material to morphemes; MP Rules, morphosyntactically conditioned phonological rules, We saw that these components are also crucial to account for the allomorphy of these forms synchronically.Both in the diachrony and in the synchrony of Italo-Romance we observe a correlation between presence vs. absence of regular morphology and presence vs. absence of thematic vowels.This correlation can be easily accounted for if the morphemes are organized in syntactic structures governed by locality principles.

Table of Contents
The distribution of the two exponents of the irregular participle cannot be predicted in phonological terms since both exponents can occur in the same phonological environment: Some additional rules are also needed.For example, the rule of coronal fricativization in (60) is needed to account for forms such as those in (61): . To account for what happened to the other irregular verbs, the traditional account assumes that given the alternations in (83), a generalization was postulated:(85) If the root is unaccented, it is regular If the root is accented, it is irregular According to this view, it was this generalization that lead to a restructuring of all irregular perfect forms by analogical levelling.
).I assume that these cases involve extension of the application of the Vocabulary Item in (94)b), to new roots: The same is found for the participle: There are cases in which a special suffixal form typical of athematic constructions is found in thematic constructions.Crucially again, it has become the regular exponent of the past participle in the e-conjugation.