Navigation – Plan du site

From pure phonology to pure morphology the reshaping of the romance verb

Martin Maiden
p. 45-82


L’article étudie une différence morphologique très importante entre le latin et les langues romanes. Celles-ci se distinguent du latin par l’abondance des alternances du radical qui se manifestent surtout dans le verbe, où elles sont sensibles à des combinaisons arbitraires de traits de personne, de nombre, de temps et de mode. Bien qu’elles résultent d’effets réguliers du changement phonologique, je soutiendrai que ces alternances ont rapidement perdu leur motivation phonologique, pour devenir ‘morphomiques’ au sens d’Aronoff (1994) ; c’est-à-dire qu’elles manquent, sur le plan synchronique, de conditionnement soit phonologique soit morphosyntaxique. Ces alternances constituent en effet des schémas paradigmatiques abstraits, d’une importance remarquable pour l’histoire des langues romanes puisqu’elles constituent le modèle pour une série de changements ultérieurs au travers des langues romanes, dont le contenu phonologique se révèle souvent très hétérogène. Cependant, beaucoup de lingistes cherchent à en rendre compte en faisant appel à un conditionnement purement phonologique. Je passerai en revue quelques-unes de ces tentatives d’explication phonologique sur le plan synchronique : tout en défendant la nature ‘morphomique’ des alternances en question, je reconnaîtrai que la frontière entre la ‘morphocité’ et le conditionnement phonologique est peut-être plus floue qu’on ne l’a reconnu jusqu’à présent.

Haut de page

Texte intégral

1. Introduction

  • 1  Space precludes discussion of nouns and adjectives. These too generally had invariant roots. In Ro (...)

1There is a simple morphological distinction between Latin and Romance, rarely directly acknowledged by Romance linguists. In Latin, as in Romance, the inflectional paradigm of the verb1 typically comprises a root identifying the lexeme, followed by material mostly expressing grammatical properties. The root in Latin is overwhelmingly invariant: that is to say that lexical meaning is represented by an invariant form; in Romance it shows extensive allomorphy: lexical meaning is represented by multiple forms. Naturally these statements need qualification. Latin had some root-allomorphy, and Romance languages have it to varying degrees, although at one extreme (Romanian) absence of root-allomorphy is rare. Yet the distinction is broadly true, and it means that the Romance languages have become in this respect relatively ‘unnatural’. It is intuitive that the optimal relationship between form and meaning should be maximally transparent: ‘one meanin – one form’ (this notion is fundamental to ‘Natural Morphology’ – see, e.g., Dressler et al., 1987). There is apparently no language in which lexical meaning is always represented by multiple forms, while there are many where the opposite is usually true.

2What was virtually unknown in Latin was root-allomorphy correlated with person, number, mood or tense. The exceptions were esse ‘be’, its derivative posse ‘be able’, ire ‘go’, and uelle ‘want’. Uelle has no direct continuant in Romance; esse was erratic and often suppletive, and much of its eccentricity survives; I return to ire and posse later. The crucial point is that none of these directly foreshadows the general patterns of root-allomorphy characteristic of Romance languages. A great deal of it in Romance is the regular outcome of historical sound changes, more or less widespread and ancient, of which three seem ancestral to all (or nearly all) Romance languages. I claim that these changes had profound, chronic repercussions for the organization of Romance morphology, producing a ‘template’ to which subsequent sources of alternation repeatedly conform, and which are autonomously morphological (‘morphomic’in the sense of Aronoff, 1994).

3Aronoffian ‘morphomes’ involve systematic mappings between phonological heterogeneity and morphosyntactic heterogeneity. One example is ‘past participle’in a number of western European languages, a ragbag of forms irreducible to any phonological common denominator each of which serves two disjunct functions, those of passive and perfect; these functions are always expressed by the same form (e.g., Italian Ha/È scritto/visto/lavato/scelto/preso ‘He has/is written/seen/washed/chosen/taken’). The past participle is ‘morphomic’ in that it can be reduced neither to morphophonology nor to morphosyntax: it is ‘purely morphological’ in that it lies between the two. Aronoff’s book is explicitly an invitation (1994:169), to explore morphomic phenomena further, and he focuses (1994: 28) just on what he considers the ‘clearest examples of the purely morphological’, principally in synchrony. In the present study I address some slightly less clear-cut cases, where the possibility lingers that phenomena which are apparently autonomously morphological might in reality retain some of their original phonological conditioning. Some of the arguments have been given before (e.g., Maiden, 2005b), but in 4.1 I adduce other evidence, and address some new arguments which favour a ‘phonologizing’ account. What emerges as a particular problem for the latter is the sheer range of phenomena which can be subject to morphomic distribution within paradigms, and in section 4.2 I emphasize the extraordinary qualitative gap between the common-or-garden phonologically induced segmental allophony at the historical origin of our Romance morphomes, and the types of alternation which have emerged in Romance languages as a result, where the alternants can be segments, lexical roots, whole word-forms, absence of whole word-forms (i.e., defectiveness), inflection class membership (heteroclisis) and even entire ‘slabs’of the inflectional paradigms of individual lexemes.

2. Sound change and the morphology of the verb

2.1. ‘L-pattern’ and ‘U-pattern’

4My (purely arbitrary) label ‘L-pattern’ describes alternation such that some alternant is shared distinctively by the present subjunctive and 1sg present indicative. ‘U-pattern’is the same, except that the distinctive alternant also occupies 3pl present indicative. The L-pattern occurs throughout Romance (its U-pattern variant is restricted to parts of Italy, and Romanian), and arises from two sets of phonological changes. The first is the ‘yod-effect’ (YE), involving original palatalization and/or affrication of consonants immediately preceding yod. The second, common to all Romance except Sardinian, is palatalization and affrication of velar consonants (PAV) before front vowels. It appears (Väänänen, 1963: §§95-100) that YE was operating by the second century AD, and PAV by the fifth.

5The principal source of proto-Romance yod was Latin unstressed, prevocalic, e or i. In second, fourth and some third conjugation verbs, prevocalic e/i appeared after the root in 1sg present and throughout present subjunctive; prevocalic i, but not e, also occurred in this position in 3pl indicative. Consider present indicative (top) and present subjunctive (bottom) of Latin tenere ‘hold’, facere ‘do’ and uenire ‘come’, bearing in mind that where e/i precedes a vowel, it is destined to become yod:

(Table 1)

















































6In Ibero- and Gallo-Romance, yodless 3pl -ent seems to have extended into the other third person plurals (creating L-pattern distribution). The exception is central Italo-Romance, which generally extends the reflexes of 3pl type -iunt, with yod, into the 3pl of verbs originally in -ent, thereby introducing U-pattern distribution. The subsequent history of consonant + yod sequences is intricate. Rather than review the details here (see for example Lausberg, 1976: §§451-78; also Maiden, 2005b; forthcoming b), we may examine some regular results of these changes in various Romance verbs. Examples from Portuguese and old Tuscan:

(Table 2)

Portuguese (nh=/ɲ/; j=/ʒ/; ç=/s/)

tenho ‘hold’












vejo ‘see’











faço ‘do’












meço ‘measure’












caibo ‘fit’












Old Tuscan (gli=ʎʎ, gn=ɲ, ggi=/ddʒ/, cci= /tt/)

vaglio ‘be worth’












rimagno ‘stay’ 












veggio ‘see’ 












piaccio ‘please’ 












muoio ‘die’ 












7In Latin non-first conjugation verbs, the root-final consonant is immediately followed by a front vowel except in 1sg and 3pl present indicative, and the entire present subjunctive, where a non-front vowel follows the root (from dicere ‘say’, crescere ‘grow’):

(Table 3)

























8PAV yields in Romanian and central Italo-Romance a U-pattern. In central and upper southern Italy this exactly replicates the distribution also created by yod. Elsewhere, the 3pl inflection was -ent, and as a result PAV also occurs in the 3pl present indicative, giving rise to a further L-shaped pattern.

(Table 4)





































crezc áis  


  • 2  For the aberrant status of the 1pl and 2pl roots in the subjunctive, see 3.4.

Italian (before i and e, c= [t], g= [dʒ], sc= []; gl= [λλ] before [i])

























  • 3  From coll (i) gere ‘gather’: note that /1g/ + front vowel yields /λλ/.













  • 4  Romanian replaces subjunctive by present indicative forms, save in third person.

Romanian4 (c and g= [t] and [dʒ] before i and e)


















10The L/U-pattern arises from separate and phonologically distinct sound changes, which means that the resulting alternants involve considerable phonological heterogeneity. With YE, the fact that the triggering yod was frequently absorbed into the preceding consonant means that the conditioning environment for the alternations disappears.

2.2. The N-pattern

11‘N-pattern’ is an again arbitrary label denoting an alternation, recurrent across Romance, such that present tense 1/2/3sg + 3pl, and 2sg imperative, share a root distinct from the rest of the paradigm. The source of the N-pattern is quality differentiation between stressed and unstressed vowels. In general, the range of vowel qualities in Romance stressed syllables is greater than in unstressed. This fact has particular impact on verbs, all of which were characterized by stress alternation between root and ending. In Latin the position of stress depended on principles of prosodic phonology, usually falling on the antepenultimate syllable, but on the penultimate if the latter was ‘heavy’. In the present, stress generally fell on the root in the three persons of the singular and in 3pl (in the third conjugation it fell on the root throughout the present indicative — as it still does in Romanian). The predominant pattern is that 1sg, 2sg (+ imperative), 3sg and 3pl present roots are opposed to the rest of the paradigm.

  • 5  Except Logudorese Sardinian where, significantly, there is no evidence for creation of novel N-pat (...)

12The Romance of most of Italy and the remainder of western Europe (except Sardinia) developed a system with seven vowels [i e ɛ a ɔ o u] in stressed syllables and five in unstressed [i e a o u]. The ‘deficit’ in the unstressed vowels arises because continuants of Latin short ĕ and ŏ remain distinct from those of long ē and ō. In stressed syllables the former yield [ɛ] and [ɔ], the latter [e] and [o]; unstressed they merge as [e] and [o]. This differentiation is well preserved in modern Italian. There were countless other subsequent, local, stressed-based vowel differentiations in all5 Romance languages (see Lausberg, 1976: §§154-296). Some regular effects are illustrated in Table 5 (I give just the present indicative), here just from Romanian and medieval French, although other equally rich examples could be found in many Romance languages.

(Table 5)


mor ‘die’






vin ‘come’






plac ‘please’






mănânc ‘eat’






usuc ‘dry’






iau ‘take’






13Medieval French

lef ‘wash’






crief ‘burst’






peis ‘weigh’






parol ‘speak’






manju ‘eat’






2.3. The import of the patterns

14As I show in section 3, the alternations created by these sound changes form a kind of morphlogical template for subsequent morphological changes across Romance. As a preliminary example, consider the fate of posse ‘be able’, one of the few Latin verbs displaying root allomorphy (poss- vs. pot-) correlated with person, number and tense. Of Romance varieties retaining reflexes of poss- and pot-, none preserves the original distribution (with poss-. throughout the present subjunctive and in 1sg, 1pl and 3pl present indicative). Rather, they are redistributed in a way that perfectly replicates the locally prevalent L- or U-pattern:

(Table 6)

Old Tuscan


























15One repeatedly encounters phenomena which replicate the L/U-pattern and/or the N-pattern, yet lack any resemblance to the alternants created by sound change. My claim is that the patterns, phonological in origin, are ‘morphomic’, and that novel creation of alternations having the same distributional pattern exemplifies ‘morphomically-driven’ morphological change. But is what we observe really ‘morphomic’? In Aronoff’s definition, some phenomenon is morphomic to the extent that it is not synchronically conditioned by any factor outside the morphology. We have therefore to exclude morphosyntactic, semantic and phonological conditioning.

16The charge that our ‘morphomes’are somehow still phonologically conditioned requires a detailed answer, which I give in section 4. As for morphosyntactic conditioning, it is relatively easy to rule out simply because the array of morphosyntactic cells of the paradigm with which the alternants are correlated is irreducible to any coherent feature or set of features. The N-pattern specifies singular and third person, but only in the present; it specifies imperative, but only in the singular; and it specifies present, but not in 1pl and 2pl. Clearly the N-pattern need to be stated in terms of sets of cells each of which has its own morphsoyntactic specification, but the point is that the domain of our patterns cannot be correlated with any feature, or coherent set of features. The domain of the L/U-pattern specifies subjunctive, but only present subjunctive; it also specifies present, but only the subjunctive and 1sg indicative (+ 3pl in U-pattern varieties).

17It is tempting to seek some unifying principle at a more abstract level. To take the N-pattern, could ‘markedness’ not offer an explanation? Singular is less marked than plural; present is less marked than other tenses; third person is generally less marked than other persons. Since the N-pattern precisely distinguishes singular and third person present forms from the rest of the paradigm, surely the pattern is motivated by ‘markedness’? Yet it should be obvious that markedness is inadequate as an explanatory principle, for the range and pattern of intersection of the parameters of markedness involved remains irreducibly arbitrary. Subjunctive is more marked than indicative, yet the N-pattern is insensitive to mood; there are complex relations of relative markedness among different tense-forms yet the N-pattern privileges that between present and other tenses over other possible divisions; plural is more marked than singular, but in the third person (present) the distinction is irrelevant to the N-pattern; it is unlikely that 1pl and 2pl share exactly the same markedness value, but they are generally treated identically. Finally, if the N-pattern diagrams some ‘natural’, and presumably universal, markedness relationship, how is it that this pattern apparently does not recur repeatedly in other languages? This is a point for empirical verification, but to my knowledge nothing like it obtains in other Indo-European varieties with similarly structured verb systems.

18As for the L/U-pattern, it is principally manifested in the subjunctive, which is more marked than the indicative; but it is not manifested in the imperfect subjunctive, which is more marked than the present; and its domain includes the 1sg present indicative (and 3pl in U-pattern varieties), which is less marked than the 1pl present indicative, which lies outside the pattern. It is in fact probably fruitless to try to rationalize our patterns in terms of morphosyntactic or semantic conditioning, because they are the arbitrary and accidental result of ‘blind’ sound change: that is all.

3. Diachronic evidence

19The chronological, geographical and structural range of changes replicating these patterns is vast, and far exceeds the confines of this study. Many of these are reported, for example, in Maiden (2004b; 2005b; forthcoming b). I focus here mainly on phenomena not dealt with elsewhere, or whose theoretical implications were less fully considered before. I have usually restricted exemplification to present indicative and subjunctive forms (arranged one above the other). It is important to underscore from the outset, in the case of the ‘N-pattern’, that alternation is between singular and third person forms of the present, on the one hand, and (as a rule) the whole of the rest of the paradigm, on the other. The 1/2pl present are offered as representative of the ‘rest of the paradigm’. Note that in examples from Italo-Romance, Romansh or Gallo-Romance, the predicted L-pattern alternants are often absent from 1/2pl present subjunctive: this development is interpreted in section 3.4.

3.1. Lexical incursion (suppletion)

  • 6  See Corbett (2007).
  • 7  The ‘complement’ is those cells other than singular + third person present.
  • 8  Exceptions typically involve locally general syncretisms between cells: e.g., in parts of Liguria (...)

20‘Lexical incursion’ is the introduction of forms distinctive of one lexeme into cells of the inflectional paradigm of another6. Incursion affecting the present indicative and present subjunctive of Romance verbs (i.e., those parts of the verb where the L/U-pattern and/or N-pattern are positively manifested) is virtually always sensitive to one of these two morphomes, especially the latter. Examples of lexical incursion differentiating between any of the singular and third person cells of the present indicative and present subjunctive are very rare. If incursion affects ‘N-pattern complement’7 cells, then it affects them en bloc: 1/2pl present indicative and present subjunctive are nearly always8 treated identically, together with non-present-tense forms (although the future and conditional, forms derived from Latin perfectives, and past participles, sometimes behave differently — for the independent ‘morphomic’ status of these see Maiden, 2004a; 2005; forthcoming a).

21In the verb ‘go’, most Romance varieties suppletively conflate two, or three, etymologically different lexemes, from ire ‘go’, uadere ‘make one’s way’, ambulare ‘walk’. Repeatedly, such conflation takes an N-pattern distribution (see Aski, 1995). A common scenario is that continuants of ire are replaced by uadere just in N-pattern cells. Thus old Tuscan gire ‘to go’ :

(Table 7)













22This pattern remains in central and southern Italy, and is extensively attested in Ibero-Romance. A subsequent development in Italy, Catalan, Gallo-Romance and western Romansh, is the encroachment of ambulare (> Fr. aller; It. andare). This time the N-pattern complement cells are affected, so that reflexes of ire are generally ousted. For example Italian :

(Table 8)













In Surselvan, ambulare is continued by forms with initial m- (see Decurtins, 1958). Thus Sedrun :
(Table 9)

vɔn (d)











23These Surselvan data show an interesting additional local development. The presence of/ɔn/within the root of the 1sgand subjunctive forms is phonologically inexplicable in a reflex of uadere. In fact, as Decurtins demonstrates, it reflects the influence of the stressed root of ambulare, so that what has taken place is a ‘blending’ between the root of ambulare and that of uadere, characterized by initial v-. This dialect (and others like it) shows the effects of further lexical incursion of the root of ambulare into part of the root of uadere, according to an L-pattern distribution.

24In many Romansh dialects the verb ‘pull’ comprises reflexes of trahere and *ti'rare, with the latter in N-pattern distribution (I return later to the significance of the infinitive and past participle forms also given here):

(Table 10)








25Infinitive tre
Past participle trac

  • 9  The present subjunctive has donare in all cells. This seems to manifest the tendency to invariance (...)

26Romance languages generally inherit Latin dare as the verb ‘give’, but in Gallo-Romance in particular this lexeme is replaced (see Maiden, 2006) by reflexes of donare, originally ‘donate, grant, bestow’. There are at least two areas of Italy in which the innovatory and the older expressions have come into contact: in the north-west, at the frontier between Gallo-Romance and Italo-Romance varieties, and in Sicily and Calabria, almost certainly as a result of contact with Norman French in the Middle Ages. Thus in Limone, on the Liguria-Piemonte border (Schädel, 1903:108), and in Occitan dialects of the Po valley (Zörner, 2008:158f.),9 we find reflexes of dare in the N-pattern cells but (apparently) donare elsewhere. The same occurs in the southern Italian dialects, but with the lexical distribution reversed, so that donare occurs in N-pattern cells, and dare in the N-pattern complement. For Modica in Sicily (where initial/d/> /r/), Leone (1980:142; 144) gives

(Table 11)







27The examples shown so far involve mutual incursion, such that historically distinct lexemes have been conflated into a single paradigm, in which they are in complementary distribution. Maiden (2006: 91-93) considers some evidence that in old Sicilian the incursion of donare on dare might have been ‘asymmetrical’, the former suppleting the latter, while surviving with a complete paradigm of its own uninfluenced by dare, and retaining the additional (inherited) meaning of ‘donate’. A clear example of asymmetrical suppletion, sensitive to the N-pattern, is the reflex of Latin debere ‘owe, must’, in some varieties of Romansh. What is involved, diachronically, is the intrusion into the paradigm of debere of forms from another verb, whose etymology is reconstructible as *sto'pere, and whose ultimate source is the Latin phrase est opus ‘it is necessary’. The conflation of these lexemes is asymmetrical, in that the descendant of *sto'pere (> stueir or similar) maintains a complete paradigm, but parts of debere (> dueir or similar) are replaced by forms of *sto'pere in what is, in effect, an N-pattern distribution (allowing for the fact that in the relevant dialects the present subjunctive has an invariant stem – see section 3.4); moreover, the two lexemes involved are not, in fact, perfectly synonymous. Consider the Surmiran dialect of Savognin (Signorelli, 2001 ; Anderson, 2008) :

(Table 12)















  • 10  In the context of Romansh, characterized as it is by extreme and idiosyncratic patterns of vocalic (...)
  • 11  See, for example: Vieli & Decurtins (1962: 219; 701).

28The lexeme continuing debere has long been in retreat in Romansh, having all but disappeared in the Engadine, and seems moribund in many other places, having been replaced by alternative expressions; the continuants of *sto'pere seem, in contrast, to be in rude health everywhere. There is absolutely nothing in the phonological or morphological structure of debere to explain its disappearance, still less its partial disappearance in the present. Some dialects retain a full paradigm (see Decurtins, 1958:152f.), and the rhizotonic forms (usually de-) are robustly attested into the 18th century10. Wherever debere and *sto'pere coexist, moreover, they retain a slight semantic contrast11. debere is generally associated with (moral) obligation, *sto'pere with absolute necessity. The former is inherently weaker than the latter (in the sense, precisely, that assertion of an obligation is no guarantee that the action will be carried out), and yet also potentially face-threatening: to assert of someone, especially in the present tense, that they are obliged to do something effectively imposes the obligation and creates the possibility that they might fail; none of this is true if one asserts that some action is a matter of absolute necessity, that the subject ‘has no choice’. I propose this as a possible motivation for asymmetrical incursion of one verb into the other in the present tense, but it is not enough to motivate the actual pattern of alternation observed: why not the whole present tense? and why not, in particular, in both second person forms (it is surely in direct address that the ‘threat to face’would be strongest)? That the incursion avoids 1/2pl present, but occurs systematically and uniformly in the singular and third person forms, indicates that the morphomic N-pattern has also played a role.

29Some central and Surselvan varieties (see Decurtins, 1958) show clear signs of influence of reflexes of facere on those of dicere. In the relevant dialects, the root-final consonant in the present 1sg and subjunctive of dicere (i.e., in the L-pattern cells) is not the expected phonetically regular palatal plosive /c/, but the affricate /t∫/ regularly found as the root-final consonant of facere (the regular reflex of facio > *'fakjo > fat, faciam > *'fakja > *'fata, etc.). Thus the dialect of Pignia :

(Table 13)

























30As the above examples suggest, there is no interference of facere upon dicere in any other part of the paradigm. In many other central and Surselvan dialects, we find the same pattern of interference, but in reverse. The regular expected \t\ of the present 1sg and subjunctive of facere, is replaced by the \c\ of dicere. Again, the influence of one verb on the other is restricted to the L-pattern. Thus the dialect of Almen :

(Table 14)





fa'ɟe :s








31A different example is offered by a case where the source of the lexical incursion has a different paradigmatic distribution from that displayed in the lexeme undergoing the incursion. Specifically, part of a root-allomorph limited to the present subjunctive of the source, assumes L-pattern distribution in the target verb. In various Romansh varieties, certain verbs, pirncipally reflexes of facere, display a non-etymological root vowel /ɛ/. The consensus (e.g., Decurtins, 1958: 22f.; Prader Schucany, 1971: 231) is that the source of this vowel is dare and/or stare, with root dɛt-\stɛt-, limited to the present subjunctive. In some dialects there has been extension of this analogically imported vowel from the subjunctive into the 1SG present indicative, thereby achieving an L-pattern distribution for the vowel not found in dare/stare themselves: see Table 13, for Pignia.

32Some central Romansh dialects inflect reflexes of laxare ‘let’in a manner that clearly shows the influence of the ‘rhyme’of the root of the verb ‘be’, limited just to the L-pattern (the root-initial //, which also has an L-pattern distribution, has a different explanation). Compare the two verbs in Razen :

(Table 15)

ʃund ‘let’












sund ‘am’












3.2. Heteroclisis

  • 12  For evidence of incursion of fourth conjugation morphology into the N-pattern complement of third (...)

33Heteroclisis is ‘the property of a lexeme whose inflectional paradigm contains forms built on stems belonging to two or more distinct inflection classes’(Stump, 2006: 278). There is at least one12 branch of Romance in which heteroclisis shows sensitivity to the N-pattern, namely Romanian.

34A subset of Romanian verbs which otherwise manifest fourth conjugation morphology, acquire analogical first conjugation morphology in a domain bounded by the N-pattern. Briefly, in the normal case the fourth and first conjugations are inflectionally distinct outside the present and subjunctive, as they are in the 1pl, 2pl, 3sg and 3pl present and subjunctive, and the imperative. Fourth conjugation verbs are divided into two varieties, here ‘a’ and ‘b’. In class ‘a’, comprising the overwhelming majority of fourth conjugation verbs, singular, third person and 2sg imperative forms are also characterized by an ‘augment’ (a semantically empty formative between root and inflectional ending, which originated as an affix marking ingressivity, and has in Romanian assumed an N-pattern distribution: see Maiden, 2004b; 2005b). The much smaller ‘b’ class lacks an augment, and is inflectionally non-distinct from the first conjugation in 1sg and 2sg present, and subjunctive. There is one further small, but significant, difference between first and fourth conjugation class ‘b’ verbs, regarding the 2sg imperative: in all first conjugation verbs, and transitive fourth conjugation class ‘b’ verbs, this imperative is identical to 3sg present indicative; but in intransitive fourth conjugation class ‘b’ verbs, it is identical to 2sg present indicative.

(Table 16)

First conjugation cânta ‘sing’

Pres. ind.








Pres. subj.  










35Fourth conjugation class ‘a’ iubi ‘love’

Pres. ind. 







Pres. subj. 










36Fourth conjugation class ‘b’ dormi ‘sleep’

Pres. ind. 







Pres. subj. 










37*Compare this with transitive imperative of the class ‘b’verb simţi ‘feel’, namely simte (identical to 3sg present)

38There is a further division within both class ‘a’verbs and class ‘b’verbs, best understood in historical phonological terms (see Iordan, 1935; Lombard, 1955:750-52): in the environment of a historically preceding /rr/, front vowels and diphthongs often ‘centralized’: i /i/ > â/î /ɨ/; e /e/ >ă /ə/; ea /ea/ > a /a/. The result is a parallel series of fourth conjugation verbs in which ‘centralized’, rather than front, vowels follow the root (e.g., urî ‘to hate’, 3sg pres. ind. urăşte, 3sg pres. ind. urască). Interestingly, class ‘b’ verbs behave unexpectedly in this respect. Take coborî ‘come down’: the underlined forms are surprising, since they should not display first conjugation endings.

(Table 17)

Pres. ind. 







Pres. subj. 










  • 13  The sound change also produced identity to the first conjugation in the imperfect.
  • 14  Maiden (2009) shows that in certain S.W. Romanian dialects a rule of centralization after /s/ has (...)

39Centralization in the 3sg present produces an ending identical to that of the otherwise distinctively first conjugation 3sg present. This much is purely phonological in origin, but the resultant partial and accidental overlap has promoted further adjustment to first conjugation morphology, bounded13 by the domain of the N-pattern. Coborâ is principally an intransitive verb, so the 2sg imperative should be identical to the 2sg present. But in fact it follows first conjugation intransitives (and transitives) in showing identity to 3sg present. In fourth conjugation verbs there is always syncretism between 1sg and 3pl present, whilst in the first there is syncretism between 3sg and 3pl present: coborâ shows the latter kind of syncretism. Moreover, the distinctive marker of first conjugation 3sg and 3pl subjunctive, -e, appears instead of expected -ă. Given that 1sg and 2sg present subjunctive are already non-distinct with regard to conjugation class in first and fourth conjugation class ‘b’ verbs, we can say that we have a conjugation shift confined just to the N-pattern cells. A small number of other verbs behave similarly in respect of their N-pattern cells, even thought they were not subject to centralization: e.g., acoperi ‘cover’, suferi ‘suffer’, sprijini ‘support’, absolvi ‘absolve’ and in some varieties curăţi ‘clean’, gâdili ‘tickle’ (see Lombard, 1955: 746f.). These verbs presumably owe something to the model of coborâ, but what is interesting about the last two is that they also exist in first conjugation variants (curăţa, gâdila). It is also significant that neologisms in general, if they are ascribed to to the fourth conjugation, are ascribed to class ‘a’, with the ‘augment’. That the N-pattern cells of these verbs show no augment is a sign, in addition to their first conjugation inflectional endings, that the N-pattern cells of their paradigm have been wholly assigned another conjugation.14

3.3. Defectiveness

40‘Defectiveness’ is the situation whereby speakers are conscious that for certain cells of the paradigm, exceptionally, no form exists. The type of defectiveness of interest here is that for which there is no independent motivation — where a form could exist, but arbitrarily does not. In Ibero-Romance, such ‘paradigm gaps’ are sensitive to morphomic structure.

41Albright (2003; 2006) shows that the behaviour of native speakers with regard to defectiveness is often more ‘gradient’ than prescriptive grammars indicate. Moreover (especially for Portuguese) grammarians disagree over which verbs are defective and over what pattern of defectiveness a particular verb has. Yet, perhaps surprisingly, I maintain that observation of what grammarians prescribe can give us in this case a clearer insight into native-speaker competence than observation of actual usage. The grammarians are themselves native speakers of the languages they treat, who impose on themselves the task of systematizing gaps whose distribution may be ‘fuzzy’ in actual usage: yet they do this along lines which correspond not to any conventional category (e.g., ‘present’, ‘subjunctive’), but to the N-pattern or the L-pattern (or to the union of the two).

42Albright (ib.) convincingly demonstrates that ‘low speaker confidence’plays a central role in determining defectiveness in low-frequency verbs, due to lack of sufficiently robust models provided by other verbs in the language. The canonical example is Spanish abolir ‘abolish’, for which, in the singular and third person forms of the present indicative and subjunctive at least two possible stem-forms would be possible, e.g., 3sg present indicative *abuele or *abole, 3pl present indicative *abuelen or *abolen, 3sg present subjunctive *abuela or *abola, and so forth. This reflects the fact that many Spanish verbs with /o/ show an ‘N-pattern’ alternant /we/, whereas others do not alternate. However, there are important respects in which appeal to ‘uncertainty’ is not enough. Consider the paradigm of abolir :

(Table 18)



pres. ind.

pres. subj.

impf. ind.



imp. subj.



























  • 15  This fact is in itself remarkable, since (excepting oír ‘hear’) all other Spanish verbs of the -ir(...)

43Note that the pattern of defectiveness is the union of the N-pattern and the L-pattern: for there are also gaps in the 1pl and 2pl present subjunctive. This last fact is harder to explain in terms of ‘low speaker confidence’. It is true of virtually all Spanish verbs (and all non-defective verbs in -ir) that the stem vowel of 1/2pl present subjunctive is that of the preterite and the imperfect subjunctive. Abolir has a full set of such forms, and they contain /o/:15 there should therefore be no doubt about the identity of the 1/2pl present subjunctive (*abolamos, *aboláis). Even if one wished to claim that ‘N-pattern’ defectiveness is a kind of epiphenomenon of ‘low confidence’in the relevant cells, no such argument can apply to the rest of the present subjunctive. The missing forms are unambiguously predictable. That they are not there is a morphomic generalization: if there are gaps in some L-pattern cells (1sg present indicative and subjunctive, etc.), then the domain of the gap must be all L-pattern cells.

44The morphomic nature of the gaps (N-pattern + L-pattern) is even more inescapable in verbs like blandir ‘brandish’ (and various other defectives with roots in /a/, such as garantir ‘guarantee’), which share the same pattern of defectiveness as abolir. Here there is no possible reason for ‘low confidence’: no Spanish verb ever shows alternation for root /a/ in any relevant part of the paradigm. Forms such as the missing *blando, *blanda, *garanto, *garanta, are perfectly predictable and wholly unproblematic.

45My interpretation of these facts is that they do betoken a kind of ‘low confidence’, actually one that borders on ‘paranoia’: for defectiveness occurs even where there is no room for doubt. It is actually a striking general characteristic of defective verbs in Spanish that they never show any kind of allomorphy: apart from their gaps, they have invariant roots. The lexemes involved are always marginal (usually learned) and very often of relatively recent origin. What is at work (see Maiden and O’Neill forthcoming) is at base a strategy for avoidance of allomorphy in verbs transmitted into Ibero-Romance from other languages (principally Latin) with a single root allomorph. What appears to be directly relevant, however, is speakers’sense that there are domains within the verb in which allomorphy characteristically occurs, namely the N-pattern and the L-pattern. Speakers avoid using the relevant forms not because of real potential for allomorphy, but because of awareness, at a more abstract level, that the domains in question are those where allomorphy often does occur.

3.4. ‘Clash’ of morphomes

46There is both ‘overlap’ and ‘mismatch’ between the paradigmatic domain of the N-pattern, and that of the L/U-pattern. They overlap in respect of the 1sg present and singular and third person forms of the present subjunctive; there are discrepancies in relation to the second person singular and third person forms of the present (N-pattern but not L-pattern), and in respect of the 1/2pl present subjunctive (L-pattern but not N-pattern); in U-pattern verbs the overlap, of course, includes 3pl. Evidence for the role of both patterns in diachronic change is the widespread tendency to ‘accommodate’ one pattern with the other. The predicted distribution of L/U-pattern allomorphs is – with an exception to which I return later – robustly maintained in Ibero-Romance, including Catalan, as also in Sardinian. Elsewhere we witness, to a greater or lesser degree, a general elimination of the discrepancy in the 1/2PL present subjunctive, which takes one of two forms. In ‘Type A’, the original L/U-pattern allomorph is lost from 1/2PL present subjunctive, the domain of the L-pattern allomorphs becoming a proper subset of that of N-pattern allomorphs; in Type B the discrepancy between L/U-pattern and N-pattern cells in the 1/2pl present subjunctive is eliminated by generalizing root-uniformity to all cells of the present subjunctive. Whereas in typical L/U-pattern allomorphy a relatively small number of verbs displays identical segmental (not prosodic) structure in all roots of the present subjunctive, in Type B the property of root-uniformity is hypercharacterized and generalized: verbs tend to generalize and reinforce root-uniformity throughout the present subjunctive, even in respect of stress. This usually means that the root of the singular and third person forms of the present subjunctive is introduced in the 1/2pl as well, so that the root is also stressed throughout the present subjunctive. The distribution of (original) N-pattern alternants is changed, and the distribution of L/U-pattern root allomorphs is consequently a subset of the N-pattern.

47For the most widely encountered type, (A), further generalizations can be made. If any Romance dialect has lost L/U-pattern root allomorphy in the present subjunctive, it will have lost it in the 1/2pl. Crucially, this allomorphy is overwhelmingly lost only in the 1/2pl present subjunctive, although in Romanian L/U-pattern root allomorphy is lost from 1/2sg forms as well. Loss of the root allomorph in the 2pl and 1pl is virtually always mutually implicational. Outside Romanian, if the L/U-pattern allomorph survives in any of 1sg, 2sg, 3sg or 3pl present subjunctive, then it survives in all of them. In short, the typical development in Romance outside Ibero-Romance, Sardinian and Romanian, is for the root allomorph to assume an N-pattern distribution within the present subjunctive, remaining in place in the singular and third person, but being expelled from 1/2pl. Such ‘expulsion’ may affect only some lexemes, as in Tuscan, or it may be general. In addition, ‘expulsion’ may involve just the root allomorph, leaving distinctive present subjunctive inflections in place, or the entire word-form, so that the 1/2pl present subjunctive acquires forms syncretic with the present indicative. Thus Tuscan :

(Table 19)

muoio ‘die’ 












nasco ‘be born’ 












48In French, expulsion is much more lexically widespread, leaving just a few verbs, such as faire ‘do’ (fassions fassiez) and pouvoir ‘be able’ (puissions puissiez) untouched. In contrast western Romansh (Surselvan, together with nearby central Romansh varieties — see Decurtins, 1958) never has the L-pattern allomorph in 1/2pl present subjunctive. The root that appears in these cells is always identical to that of 1/2pl present indicative, and bears no distinctive resemblance to that of the rest of the present subjunctive:

(Table 20)


fɛt∫ ‘do’  












krec ‘believe’ 









kar'te :ɟan  

kar'te :ɟas  

kar'te :ɟan 

'∫tɔ ‘must’  












50In all of the foregoing the replacement is limited to the root, the subjunctive desinence remaining intact. Over a wide area of central and northern Italo-Romance, however, the entire word-form is replaced by that of the 1/2pl present indicative. This syncretism is found in all verbs, not only those with subjunctive root-allomorphs. This occurs in several areas, for example in Istria (see Cernecca, 1974), and in central Italy. Thus Ascrea (in Lazio ; Fanti, 1939) :

(Table 21)

'krɔpo 'cover' 












'sacco 'know' 












51‘Type B’, with introduction of root-identity throughout the present subjunctive, crops up in at least three areas: Corsica, parts of northern Spain (with some New World varieties of Spanish) and Engadine (with Surmiran) Romansh. Such root-identity is normally total, although examples exist involving only the segmental content of the root. Romansh splits into a ‘western’ type, already illustrated, in which all trace of distinctive present subjunctive root allomorphy is ejected from 1/2pl present subjunctive, and an ‘eastern’ type, in which not only is present subjunctive root allomorphy strictly preserved, but all variation between the 1/2pl root and that of the rest of the present subjunctive, is eliminated, including differences of stress:

(Table 22)

  • 16  Signorelli (2001); also Anderson (2008).


53The last example in Table 22 illustrates the ‘augment’, a formative appearing after the lexical root and before the inflectional ending, and showing in most Romance languages an ‘N-pattern’ distribution (see Maiden, 2004b; section 3.2 below). Similar developments are observable notably in Piedmont and Liguria. In Corsica we find the following (Alfonsi, 1932: viiif.):

(Table 23)

'kredu ‘believe’ 



'kri dimu 

'kri dite 








54Some dialects of northern Iberia show the same tendency. Thus Somiedo in western Asturias (Cano González, 1981), where root-uniformity in the present subjunctive is mainly confined to non-first conjugation verbs, and to verbs which have distinctive subjunctive root-allomorphs:

(Table 24)

cómu ‘eat’ 












fágu ‘make’












4. Theoretical implications

4.1 Phonological conditioning ?

  • 17  Pace, e.g., Burzio (2004) who asserts that « whatever identity relations have a statistical presen (...)

55The phonological conditioning environment for some alternation often persists more or less intact long after the original phonological process which gave rise to it has ceased to operate. In such circumstances it may seem obvious that the phonological environment continues to play some role in ‘conditioning’ the alternation. Yet the word ‘condition’ is treacherous, because it can denote both ‘causation’ and ‘circumstance’: while it may be that some phonological environment is a recurrent ‘circumstance’ of an alternation, it does not follow that a statistically significant ‘circumstance’ noticed by linguists is equally noticed by speakers, let alone that the latter assume a causal relationship between it and the alternation.17 There are really only two ways to test for the psychological reality of claims that surviving phonological environments continue to condition (‘cause’) alternation: direct psycholinguistic experimentation, and the ‘natural experiments’ carried out by language change. In particular, signs that the original phonological process remains productive across the grammar lend support for phonological conditioning. In contrast, evidence that the phonological process fails to occur when independent changes create the environment for the change counts against a phonological analysis.

56Both L/U-pattern and N-pattern, especially the latter, preserve a good deal of their original conditioning environment intact. N-pattern alternations are overwhelmingly correlated, across Romance, with the presence of stress on the root. In Romance languages with relatively conservative unstressed vowel systems, the conditioning environment for palatalization of velars — namely inflectional front vowels — is well preserved. In contrast, the conditioning environment for alternations originally triggered by yod completely disappeared, by ‘absorption’ of yod into the preceding consonant, with resultant creation of new palatal/affricate phonemes. Are the major Romance morphomes still phonologically conditioned?

  • 18  An earlier attempt is Fanciullo (1998), for a response to which see Maiden (2001b) and Pirrelli (2 (...)

57The suspicion that synchronic survival of a portion of the historical conditioning environment for allegedly ‘morphomic’ alternations cannot be accidental, and must play some role in ‘syntagmatically’ conditioning alternations, is most recently18 articulated by Burzio (2004), in his critique of Pirrelli and Battista’s autonomously morphological treatment (2002) of Italian verb morphology. I limit my observations here specifically to what Burzio says about (in my terms) ‘L/U-pattern’ alternations.

58Burzio’s approach is grounded in Optimality Theory, and invokes the notion of violability of constraints to accommodate the possibility that phonological conditioning may operate, being in certain contexts ‘outranked’ by other constraints. He questions the assumption (Pirrelli & Battista, 2002: 323) that “All alternating stem roots which are not accountable in terms of exceptionless phonological rules of Italian are to be considered as independent B[asic] S[tems] in Aronoff’s sense”. The existence of some exceptions clearly does not wholly rule out phonological conditioning a priori, so there is a case to answer; but there need to be ground rules about how to answer it. Mere synchronic correlation, however statistically dominant, does not of itself constitute evidence of ‘conditioning’: it may enhance the plausibility of the phonological environment triggering the alternation, but it does not prove it.

  • 19  See Bybee & Pardo (1981: 956f.) for analysis of similar data in Spanish which speak against the ‘s (...)

59There is an overwhelming dearth of evidence for the thesis that alternations in the Italian verb stem are syntagmatically conditioned by the phonological content of the inflectional endings.19 Italo-Romance is simply littered with accumulated phonological counterexamples to the alleged conditioning. Velar consonants occur wholesale before front vowels (e.g., stan[k]i ‘you tire’, stan[k]e ‘tired fpl’, fun[g]i ‘mushrooms’, al[g]e ‘seaweed’); palatal consonants are similarly followed by ‘non-palatalizing’ vowels on a massive scale ([ʎ ʎʎ]a ‘straw’, catt∫]a ‘hunting’,
ra[ɲɲ]o ‘spider’, fa[∫∫]a ‘strip’). The sound changes which historically gave rise to the modern ‘U-pattern’ alternations had taken place towards the middle of the first millennium and on all available evidence had ceased to operate by the end of that millennium. The onus is on those who believe in the continued phonological causation of the relevant alternation types to find any example in the attested history of the Italian language which indicates that the appearance of new forms constituting the relevant conditioning environment then triggered the expected alternation. I do not know of any. There is massive, and millennial, counterevidence to the ‘syntagmatic’ account.

  • 20  The /j/ survives (but not in the indicative) in the subjunctives abbia ‘have’ and sappia ‘know’. T (...)
  • 21  Even if this were possible, the paradigmatic distribution of the alleged i would in turn need expl (...)

60Assessment of Burzio’s arguments is hampered by a problematic synchronic phonological claim. The alternations found, for example, in volere ‘want’- voglio (etc.), piacere ‘please’ vspiaccio (etc.), are allegedly due to a ‘pre-affixal insert i’, which is supposedly in competition with the ‘insert g’ found in valere ‘be worth’- valgo (etc.). This i (= /i/, /j/?) is ‘not observable directly, but rather only via the palatalization and/or gemination effects it produces’(Burzio, 2004:24). In other words, it is a phantom: its existence is purely orthographic (and diacritic), and the last evidence20 for the presence of a sound of this kind with the relevant distribution in the relevant phonological environments dates from, perhaps, 1500 years ago. In verbs such as volere and piacere there is simply alternation between /vol/, /pjatS/, on the one hand, and /vOʎʎ/, /pjattS/, on the other. The alternants are disparate members of an even bigger class of disparate alternants having the same distribution (the valerevalgo type itself, potere ‘want’– posso, etc.), and should not be spuriously reduced to a phonological regularity.21

61Burzio’s treatment of some glaring counterexamples to his ‘syntagmatic approach’ within the Italian verb system itself is also problematic. With regard to ‘U-pattern’ alternation, these are, (i), the incidence of the ‘palatalized root alternant in the environment of the past participle ending -uto (with back vowel). For example infinitive cre[]ere ‘grow’, present indicative cre[sk]o, cre[]i, etc., present subjunctive cre[sk]a, etc., but past participle cre[]uto; and, (ii), the existence of a speech variety where the first conjugation present subjunctive inflection -i has extended to other conjugations, yet without palatalizing root-final consonants (so that ven[g]a,ven[g]ano ‘come’ gives way to ven[g]i,ven[g]ino).

62Burzio (2004: 34f.) deals with (i) by invoking an Output-to-Output Faithfulness constraint relating past participles to their infinitives, which outranks palatalization. He adduces ‘independent evidence that past participles are in strong correspondence with their infinitives’. There are at least two problems here. First, the root [kreSS] is found in most cells of the paradigm of the verb, and no reason is given for privileging the relation of the past participle to the infinitive. Second, Burzio’s independent evidence boils down to the observation that third conjugation past participles tend to share root-stress with infinitives (e.g., infinitive vìncere ‘win’, past participle vìnto). But they do not, in any significant way, share segmental content with the infinitive; indeed, the majority of Italian rhizotonic past participles have segmentally different roots from the infinitive (e.g., prendere ‘take’– preso, trarre ‘draw’– tratto, correre ‘run’– corso, rompere ‘break’– rotto, fondere ‘melt’– fuso). One might say with greater justification that there is an ‘Output-to-Output Faithfulness constraint’ among present tense stems in the great majority of Italian verbs, since they usually do not have alternating stems, yet ‘palatalization’ is curiously not overridden in this part of the paradigm. In reality, the emergence and persistence of cre[]uto and forms like it are indicators that speakers do not analyse the alternation as syntagmatically conditioned by the phonological environment. It is likely that the historically underlying form was *kres'kuto (cf. Romanian crescut): the most probable explanation for the disappearance of the velar root is purely morphological, the fact that, in the overwhelming majority of verbs, the velar alternant is characteristic just of certain cells of the present indicative and subjunctive, and not encountered elsewhere. The behaviour of these past participles is perhaps linked with a fact about modern Italian verb morphology that does at first sight seems to lend some weight to the ‘phonologizing’ hypothesis, namely that velar alternants never appear before the 1sg and 2sg present subjunctive -iamo and -iate. We always have cresciamo cresciate, veniamo veniate, doliamo doliate etc. Any argument for phonology playing a role here is rather contradicted, though, by the fact that the same alternant appears before a back vowel in past participles [kreS'Suto], [ve'nuto], [do'luto]. Diachronically, both the past participles and the present subjunctive forms are probably to be explained in a unified way in purely morphological terms, as I have argued in 3.4.

63The phenomena described in (ii) are attested since the Middle Ages, across the dialects of southern Tuscany, and of Umbria and Lazio west of the Tiber and even in some dialects of Corsica: cf. Rohlfs, 1968: 297-99); Hirsch, (1868: 416f.); Bianchi (1888: 50); Bianconi (1962: 110); AIS map 1695. They long predate, pace Burzio (2004: 36), the existence and influence ‘standard Italian’, and his claim (2004: 38) that the phenomenon ‘constitutes a type of language change from the standard, which must therefore have provided the input data’, is curious. Even if we interpret his analysis simply as a claim that an older model survives alongside, and influences, an innovatory type with subjunctive -i, it still has to be recognised, as Burzio does, that “paradigmatic relations enter into the mental computation”. But when he also claims that: the fact that “syntagmatic relations do not obtain in this case […] does not mean […] that they do not exist, but rather only that they are outranked”, we must ask ‘On what evidence?’. What we have here is another robust piece of evidence against the notion that speakers are analysing the alternation in terms of phonological conditioning, and it needs to be recognised squarely as such. The case for phonological conditioning locally outranked by other constraints can only stand when there is substantive diachronic (or psycholinguistic) indication that innovations creating the alleged phonological environment for the alternation actually can trigger it. The phonological conditioning of the alternations is not ‘outranked’ in any sense of interest to historians of Romance languages; it is not there.

64Burzio (2004: 32-34) also observes that ‘when affixes level, so do stems’. He is referring to the fact that the modern Italian non-first conjugation present subjunctive has endings which create only two different environments for stems, involving a back (more precisely, non-front) vowel (-a), or i + vowel, and that the present subjunctive has accordingly a maximum of two stems corresponding to the two environments (e.g., ven[g]a, veniamo). This is locally true for modern Italian, but as a synchronic observation it proves nothing one way or the other; it also overlooks the historical evidence. The historical ending of the 2sg present subjunctive in all verbs was a front vowel (originally -e in some cases, later uniformly -i): this is attested in medieval Tuscan Italian, and persists in many modern Italo-Romance dialects. This vowel is the product of regular sound changes affecting reflexes of the Latin 2sg present subjunctive inflection -as (see Maiden, 1996). Nowhere in the historical or comparative record does this front vowel ever trigger the ‘phonologically’ predicted outcome. Wherever, for example, dire ‘say’ preserves present subjunctive morphology and 2sg -i, we find exactly the same alternant in the 2sg as before inflectional -a: thus old Tuscan 1sg. present indicative dico; 3pl. present indicative dicono; 1/2/3sg present subjunctive dica, dichi, dica ([ 'dika 'diki 'dika]). I find no evidence anywhere for ['dika *'ditSi 'dika] or the like. See also AIS maps 1653, 1654, 1695.

65Finally, Burzio (2004: 30-32) claims it as an advantage for his phonologizing approach that it predicts the non-existence of patterns of alternation which are not attested in the Italian verb, whilst nothing in the approach of Pirrelli and Battista could rule these out. The trouble is that it is an empirical question, requiring proof, whether the unattested patterns could not exist. That they do not exist is simply a chance product of phonological history (whence the seductive amenability of the existing patterns to ‘syntagmatic’phonologically-stated generalizations); their unattestedness is not necessarily a consequence of some kind of ill-formedness requiring synchronic explanation.

  • 22  Conde Saiz (1978 : 177) ; Arnal Purroy (1998 : 355 ; 362) ; Alvar et al. (1995).
  • 23  See Quint (1998: 55); Bendel (1934:97f.); Ronjat (1937:245).
  • 24  The last two if the penult has a consonantal coda.

66Matters are quite different where stress and the N-pattern are concerned. Virtually all Romance N-pattern alternations are still systematically correlated with their original conditioning environment, although there are examples in modern dialects of northern Spain22 and Occitan23 of N-pattern vocalic alternants displaying paradigmatic distribution independent of stress. Yet there are several objections to treating the N-pattern as conditioned by stress. One is that in the history of the Romance languages the position of stress has not generally been phonologically predictable. While Latin stress was phonologically determined, in Romance it came to fall unpredictably on any of the last three24 syllables of a word. This means that in the verb the distribution of root stress itself requires specification of the N-pattern : it falls on the root in singular and third person forms of present indicative, present subjunctive and imperative ; elsewhere it falls on the ending. There is therefore no advantage in analysing the N-pattern as conditioned by stress rather than morphology, because the morphological formulation is independently required to account for the distribution of stress.

67A more specific reason for considering stress as incidental to the N-pattern regards the suppletive replacement of Romance *ire (< ire) by reflexes of uadere (Table 7). The latter did not replace the former in the infinitive, or in the 1/2pl (< imus, itis). Yet the stressed vowel seems itself to be the root, and if it is the suppletion cannot be sensitive to stress. The only way to salvage a stress-conditioned account is to say that the infinitive consists of a zero root followed by a stressed ending. It is counterintuitive to make such a claim, although not inconceivable, and it would mean that speakers analysed Latin 2/3sg present indicative is, it as ‘stressed root + ending’, but infinitive and 1/2pl present indicative ire ; imus itis as ‘zero root + ending’ : an analysis which, in effect, presupposes the N-pattern. A similar and more modern case is Romansh suppletive replacement of reflexes of Latin trahere (> Romance *'tragere) by reflexes of *ti'rare, in an N-pattern distribution, shown in Table 10. Trahere was a third conjugation verb and in consequence preserves root-stress in the infinitive ; yet tirare, whose root only occurs in the relevant varieties when stressed, never affects the root-stressed infinitive (nor the past participle). This implies that stress is irrelevant to the process of suppletion.

68The accommodation between N-pattern and L-pattern of ‘Type B’ (3.4), such that the root of the 1/2pl present subjunctive becomes identical to that of the singular and third person, also presupposes that stress is sensitive to morphosyntactic conditioning. As I suggested above, this change is a kind of analogy : just as the L-pattern involves identity of the root-final consonant throughout the present subjunctive, so the root is remodelled so that it is identical throughout the present subjunctive in respect of vocalism, stress and any other distinctive characteristics. This usually involves the extension of what were originally N-pattern alternants into 1/2pl present subjunctive. Now if N-pattern allomorphy were triggered by stress, this kind of change could only be explained in one of two ways. Either the 1/2pl present subjunctive inflectional endings first became unstressed, causing the roots to be stressed, or stress moved onto the 1/2pl present subjunctive root, thereby triggering the appearance of N-pattern allomorphs. The first hypothesis is completely ad hoc. The second is almost as bad : the L-pattern is in origin a matter of segmental allomorphy : that it should cause just stress to shift its position, and that subsequent segmental adjustments should simply be a secondary effect of that stress-shift, seems mysterious, not to say highly implausible, and makes the resultant root-uniformity an accident. What has actually happened is that the root-allomorphs of the N-pattern, segmental and prosodic, have been generalized together, and on this account stress, as much as any segmental characteristic, is a function of the morphomic N-pattern. In fact, in many Romansh (but by no means all Romance — cf. Tables 22-24) examples of Type B, the role of morphological conditioning is patent, for root identity is achieved by syncretism of whole word forms, the 1pl and 2pl word-forms being taken over respectively from the 3pl and 2sg.

69In general, phenomena such as the emergence of Type B accommodation between the N-pattern and the L-pattern, or the the need to refer to the union of the N-pattern and the L-pattern to explain, for example, defectiveness (3.3), are simply incompatible with a stress-conditioned account of the N-pattern. If the N-pattern is phonologically conditioned, and the L-pattern (as I have argued) is morphologically motivated, then there is no ‘common ground’in terms of which their interactions can be understood. Even if the L-pattern were itself phonologically conditioned, the heterogeneity of the alleged phonological conditioning of it and the N-pattern would make accommodations and unions between them difficult to understand. If they are both viewed as defining a set of paradigmatic cells which already substantially overlap, then such interactions are more readily intelligible.

70Another objection to stress-conditioning is the sheer phonological implausibility of the alternations involved. No plausible stress-conditioned process could effect multiple, disparate consonantal and vocalic alternations, including suppletion and heteroclisis. The morphomic distribution of defectiveness seems particularly problematic for the hypothesis of phonological conditioning. I have argued that speakers are effectively avoiding word-forms in cells in which root-allomorphy might potentially occur. We can hardly appeal to ‘stress’ where there is no phonological content in the first place ; at best we would have to invoke a more abstract notion of ‘cells of the paradigm in which stress could have the effect of producing allomorphy’— in other words, we would need to specify those cells in terms of their morphosyntactic content.

71In other kinds of N-pattern alternation, some scholars effectively ‘take the bull by the horns’, acknowledging that the alternants themselves cannot be derived by rule, and must be somehow listed in the lexical specification of each verb, but claiming that their distribution can still be effected phonologically, as a function of stress. Carstairs (1988 : 17 ; 1990 : 20), claims that various N-pattern phenomena in Italian, including the distribution of the augment, involve ‘phonologically conditioned suppletion’. ‘Phonologically conditioned allomorphy’ is possible, and Carstairs adduces persuasive examples from various languages, but for Italian one still has to meet the objection that stress itself is sensitive to the N-pattern. The fact that many of the wide range of alternants implicated are idiosyncratically unique to the verb, and unparalleled elsewhere in the phonological system of Italian, also make it hard to see why appeal to purely phonological conditioning is preferable. We are dealing with something irreducibly arbitrary and idiosyncratic, and in these circumstances it is at best unclear why a phonological account is preferable to a morphological one.

  • 25  Anderson claims that the observed distribution of alternants involves a different from my ‘morphom (...)
  • 26  Although, as he crucially demonstrates, it is synchronically impossible to reduce these alternants (...)

72A slightly different approach, but in a similar vein, is the synchronic study by Anderson (2008) of the Surmiran Romansh of Savognin (see Table 22), which argues that the N-pattern allomorphy (actually, a ‘Type B’ modification, such that the N-pattern root is found throughout the present subjunctive)25 of that dialect is a matter of ‘phonologically conditioned allomorphy’, such that lexically specified vocalic root allomorphs are distributed according to stress. Anderson’s argumentation is detailed and impressive, and deserves much more detailed critical discussion than I can offer here. Briefly, in this dialect the position of stress has apparently become predictable on purely phonological grounds, so that the argument that the N-pattern is independently needed is not available. Anderson’s analysis relies on maximizing the phonological plausibility of what is involved : he deals only with binary vocalic alternants which are historically the product of stress-related vowel differentiation,26 and particularly emphasizes that these have a generality in the grammar which extends beyond the verb and is widely represented, for example, in derivational morphology. This approach has the drawback, however, of loss of generality in another direction. For the same vocalic alternants also belong to a larger series of disparate alternant types having the same distribution within the verb and which, taken together, make highly implausible candiates for any kind of phonological conditioning. These include suppletion (cf. incursion of *sto'pere on debere described in 3.1), and the distribution of the ‘augments’, one of many possible illustrations of which is the last example in Table 22. Anderson obviates this problem by effectively turning these other cases into epiphenomena of phonologically conditioned vocalic allomorphy : the augment -esch- following the lexical root is alleged to appear as a kind of repair strategy precisely when there is no ‘stressed’vocalic alternant available for the root, and the suppletion is a response to the absence of an appropriate ‘stressed’ alternant for dueir. This seems uncomfortably circular. One might equally say that the augmented root simply is the ‘stressed’ alternant of the relevant verbs ; as for dueir, it is certainly not ‘defective’, as Anderson claims, and I doubt that it ever has been. If it lacks a stressed root etymologically related to it (this is the only sense in which it ‘lacks a stressed alternant’), it is because one has been provided by its rival and near-synonym stueir. Anderson’s approach also runs into difficulties where there is more than one root-allomorph (allegedly) associated with stress — i.e., the kind of ‘intramorphomic allomorphy’ that I discuss in 4.3. For example, pudeir ‘be able’ has root allomorphs pud-, poss- and po-, poss- appearing in the 1sg present indicative and through the present subjunctive, and po- in 2sg, and third person present indicative (see Table 22). Actually, pos- and po- are both N-pattern alternants, but Anderson selects one (poss-) as the stressed alternant of pud-, leaving the forms post, po, pon to be distributed in terms of purely morphological specification. A morphological statement of their distribution obviates this difficulty. The same problem arises with the suppletive incursion of stueir into dueir. As Anderson points out, ‘it is not just a single stem, but the full range of irregular forms of stueir (ia stò, te stost, el stò, els ston ; Subjunctive ia stoptga, etc.) that replaces those of dueir where stress would fall on the stem’. In fact, we need to be even more precise : ‘the first and second person singular, and third person, forms of the indicative of stueir are copied onto the first and second person singular, and third person, forms of the indicative of dueir, and the present subjunctive cells of stueir are copied onto the corresponding person and number cells of dueir’. Only in this way do we get the observed distribution, but what we have had to specify is nothing less than the union of the N-pattern and the L-pattern.

4.2. What can be sensitive to morphomic structure ?

73There are profound qualitative differences between the origins of our morphomic patterns and their diachronic repercussions. In origin they involve binary segmental alternations of root vowels (in the N-pattern), or of root-final consonants (in the L/U-pattern). These constitute something novel in the verb system : allomorphy correlated with tense, person, number and mood, indeed correlated with a heterogeneous and arbitrary combination of these categories. The relevant cells are bound together, and distinguished from the rest of the paradigm, by a common and distinctive segmental identity in the root. What links these alternations with later manifestations of the L/U-pattern and the N-pattern, and what connects them synchronically in individual languages, is no more than this ‘lowest common denominator’ : a distributional pattern of distinctive identity within the paradigm. In earlier studies (e.g., Maiden, 1992 ; 2005b) I emphasized what I term ‘coherence’ as both diagnostic and proof of the role of morphomes in diachrony. By this I mean that alternants having a morphomic distribution maintain it in the face of potentially disruptive diachronic changes. For example, as observed above, certain lexical incursions and other changes affect all and only N-pattern or L-pattern cells en bloc. But here there is an apparent contradiction : many cases of lexical incursion actually result in allomorphy within the relevant morphomic domain. The result may, indeed, be allomorphy where there was none before.

74The L-pattern infringement of forms of the verb ‘be’ on the verb ‘let’ in, for example, Table 15, shows this very clearly : forms of the former appear in the present subjunctive and 1sg present indicative of the latter, yet preserve the unique allomorphy found in the verb ‘be’, such that 1sg present indicative and present subjunctive have near-suppletive stems. The same applies where *sto'pere impinges on debere in Romansh, complete with its own almost unique type of allomorphy (in the case cited, between sto- and stoptg-). The intrusion of reflexes of donare into dare in southern Italy can be seen to comport an L-pattern palatal alternant (duɲɲ-) in the 1sg present indicative of the verb, alternating with dun-. It is possible (albeit difficult to demonstrate conclusively) that the root allomorphy in reflexes of uadere (e.g., Italian vad- va- ; French vai- va-), which appears to be of considerable antiquity, predates the intrusion of this verb into ire : if so, the suppletive change not only created novel allomorphy, but brought allomorphy with it. Lexical incursion is ‘coherent’, not at the level of any phonologically concrete form, but at the level of the paradigm itself. An entire morphomically-defined ‘slab’ of the paradigm of one lexeme can replace that of another — bringing with it its own internal patterns of allomorphy.

75The notion of ‘morphome-internal-allomorphy’ brought to light by the evidence of lexical incursion is also suggested by what we observe in the case of ‘Type A’ accommodation between morphomes, where the L-pattern restricts itself to a subdomain of the N-pattern. Speakers are analysing L-pattern alternants as occupying a subset of the domain of the N-pattern. Here one morphome (defined as [1sg + subjunctive]) is now ‘nested’ within a larger one defined as [present [singular + third person]].

  • 27  Compare also Stewart & Stump (2007: 408). What I lack is a Romance example involving periphrases c (...)

76Morphomic defectiveness and morphomic heteroclisis are also phenomena of a quite different kind from the original segmental alternations. In the case of defectiveness, what is ‘defective’ is not a segment, nor a root, but a whole word-form : the morphome defines a class of cells which is simply not ‘realized’ in any way at all. Some cases of ‘Type A’ accommodation bear not on roots but on whole word-forms : in Italo-Romance dialects of the kind illustrated in Table 21, inflectional endings (as well as any root alternants) are accommodated to the N-pattern. In heteroclisis, the locus of the novel alternation is not in the root at all, but in an inflection-class marking formative (in our Romanian examples, the first conjugation ending -ă). There is even some evidence for morphomic distribution of the distinction between synthetic and periphrastic27 forms comprising entities larger than the word, as illustrated by an example of suppletion in in the verb ‘go’ in some Transylvanian dialects of Romanian (cf. Maiden, 2004b : 240-44) where synthetic word-forms of one lexeme show N-pattern alternation with forms of an inherently reflexive lexeme comprising a reflexive clitic pronoun (e.g., present indicative mә duk te duc sә 'duce 'mErem 'mErets sә duk).

5. Conclusion

  • 28  Indeed, the fact that both N-pattern and L/U-pattern cause a particularly ‘salient’ fracture, invo (...)

77It is a commonplace that sound change gave rise to allomorphy in the Romance verb (and the noun-adjective), but the novel nature of that allomorphy has not usually been properly appreciated, in two respects. First, the inherited transparency of the lexical root is shattered, through introduction of allomorphy along intersecting parameters of tense, mood, person and number.28 It also inflicts a profound fracture on the paradigmatic organization of the verb. The morphologically arbitrary and incoherent effects of some perfectly ordinary sound changes create abstract patterns of intraparadigmatic sameness and difference, ultimately statable only in terms of an opposition between an arbitary array of paradigm cells and its complement : this is the only historical continuity between the source of the morphomes and their subsequent manifestations in Romance.

Haut de page


AIS Jaberg, Karl ; Jud, Jakob (1928-1942). Sprach- und Sachatlas Italiens und der Südschweiz. Zofingen : Ringier.

Albright, Adam (2003). A quantitative study of Spanish paradigm gaps. In Garding, Gina ; Tsujimura, Mimu (eds.). West Coast Conference on Formal Linguistic 22 Proceedings. 1-14. Somerville : Cascadilla.

Albright, Adam (2006). Lexical and morphological conditioning of paradigm gaps [draft ms.].

Alfonsi, Tommaso (1932). Il dialetto còrso nella parlata balanina. Livorno : Giusti.

Alvar, Manuel ; Alvar, Carlos ; Mayoral, José Antonio (1995). Atlas lingüístico y etnográfico de Cantabria. Madrid : Arco Libros.

Anderson, Stephen (2008). Phonologically conditioned allomorphy in the morphology of Surmiran (Rumantsch)’. Word Structure 1 : 109-34.

Arnal Purroy, María Luisa (1998). El habla de la Baja Ribagorzana occidental. Zaragoza : CSIC.

Aronoff, Mark (1994). Morphology by Itself. Cambridge, MA : MIT Press.

Aski, Janice (1995). Verbal suppletion : an analysis of Italian, French, and Spanish ‘to go’. Linguistics 33 : p. 403-32.

Bendel, Hugo (1934). Beiträge zur Kenntnis der Mundart von Lescun (Basses Pyrénées). Biberach : Biberacher Verlag.

Bianchi, Bianco (1888). Il dialetto e la etnografia di Città di Castello. Città di Castello : S. Lapi.

Bianconi, Sandro(1962). Ricerche sui dialetti d’Orvieto e di Viterbo nel medioevo. Studi di linguistica italiana 3 : p. 3-175.

Burzio, Luigi (2004). Paradigmatic and syntagmatic relations in Italian verbal inflection. In Auger, Julie ; Clancy Clements, J. ; Vance, Barbara (eds.). 17-44. Contemporary Approaches to Romance Linguistics. Amsterdam : Benjamins.

Bybee, Joan ; Pardo, Elly (1981). On lexical and morphological conditioning of alternations : a nonce-probe experiment with Spanish verbs. Linguistics 19 : p. 937-68.

Cano González, Ana María (1981). El habla de Somiedo (Occidente de Asturias). Santiago : Universidad de Santiago.

Carstairs, Andrew (1988). Some implications of phonologically conditioned suppletion. Yearbook of Morphology 1 : p. 67-94.

Carstairs, Andrew (1990). Phonologically conditioned suppletion. In Dressler, Wolfgang et al. (eds). Contemporary Morphology. p. 17-23. Berlin-New York : Mouton – De Gruyter.

Cernecca, Domenico (1974). Morfologia del dialetto di Valle dell’Istria : il verbo e l’avverbio. Studia romanica et anglica zagabriensia 37 : p. 205-46.

Conde Saiz, María Victoria (1978). El habla de Sobrescobio. Mieres del Camino : Instituto « Bernal de Quirós ».

Corbett, Greville (2007). Canonical suppletion, typology and possible words. Language 83 : p. 8-42.

Decurtins, Alexi (1958). Zur Morphologie der unregelmässigen Verben im Bündnerromanischen. Bern : Francke.

Dressler, Wolfgang ; Mayerthaler, Willi ; Panagl, Oswald ; Wurzel, Wolfgang (eds.) (1987). Leitmotifs in Natural Morphology. Amsterdam : Benjamins.

Eichenhofer, Wolfgang (1999). Historische Lautlehre des Bündnerromanischen. Tübingen : Francke.

Elvira, Javier (1998). El cambio analógico. Madrid : Gredos.

Fanciullo, Franco (1998). Per una interpretazione dei verbi italiani a ‘inserto’ velare. Archivio glottologico italiano 83 : p. 188-239.

Fanti, Renata (1939). Note fonetiche e morfologiche sul dialetto di Ascrea . L’Italia dialettale 15 : p. 101-33.

Hirsch, L. (1886). Laut- und Formenlehre des Dialekts von Siena, VIII. Verb. Zeitschrift für romanische Philologie 10 : p. 411-46.

Iliescu, Maria (1990). Les suffixes d'élargissement verbaux. (État de la question. Évolution sémantique de -esc-/-isc.). In Calboli, Gualtiero (ed.). Latin vulgaire — latin tardif II. p. 159-69. Tübingen : Niemeyer.

Iordan, Iorgu (1935). Forme de conjugare mixtă în limba română. Buletinul Institutului de Filologie Romînă “Alexandru Philippide” 2 : p. 47-127.

Lausberg, Heinrich (1976). Linguistica romanza. Milan : Feltrinelli.

Leone, Alfonso (1980). La morfologia del verbo nelle parlate della Sicilia sud-orientale. Palermo : Centro di Studi Filologici e Linguistici Siciliani.

Lombard, Alf (1955). Le Verbe roumain. Lund : Gleerup.

Maiden, Martin (1992). Irregularity as a determinant of morphological change. Journal of Linguistics 28 : p. 285-312.

Maiden, Martin (1996). On the Romance inflectional endings -i and -e. Romance Philology 50 : p. 147-82.

Maiden, Martin (2000). Di un cambiamento intramorfologico : origini del tipo dissi dicesti ecc., nell’italoromanzo. Archivio glottologico italiano 85 : p. 137-71.

Maiden, Martin (2001a). A strange affinity : perfecto y tiempos afines. Bulletin of Hispanic Studies 58 : p. 441-64.

Maiden, Martin (2001b). Di nuovo sulle alternanze velari nel verbo italiano e spagnolo. Cuadernos de filología italiana 8 : p. 39-61.

Maiden, Martin (2004a). Verb augments and meaninglessness in early Romance morphology. Studi di grammatica italiana 22 : p. 1-61.

Maiden, Martin (2004b). When lexemes become allomorphs : on the genesis of suppletion. Folia Linguistica 38 : p. 227-56.

Maiden, Martin (2005a). La ridistribuzione paradigmatica degli “aumenti” verbali nelle lingue romanze. In Kiss, Sandor ; Mondin, Luca ; Salvi, Gianpaolo (eds.). Etudes de linguistique offertes à József Herman à l’occasion de son 80e anniversaire. p. 433-40. Tübingen : Niemeyer.

Maiden, Martin (2005b). Morphological autonomy and diachrony. Yearbook of Morphology 2004 : p. 137-75.

Maiden, Martin. Where does heteroclisis come from ? Evidence from Romanian dialects. Morphology 19 :

Maiden, Martin (forthcoming a and b). “Morphophonological persistence” and “Morphophonological innovation”. In Maiden, Martin ; Smith, John Charles ; Ledgeway, Adam (eds) The Cambridge History of the Romance Languages. Cambridge : CUP.

Maiden, Martin ; O’Neill, Paul (forthcoming). On morphomic defectiveness : evidence from the Romance languages of the Iberian Peninsula. In Baerman, Matthew ; Corbett, Greville ; Brown, Dunstan (eds). Defective Paradigms. Oxford : British Academy and OUP.

Menéndez Pidal, Ramón (1968). Manual de gramática histórica española. Madrid : Espasa Calpe.

Meyer-Lübke, Wilhelm (1972). Grammatica storica della lingua italiana e dei dialetti toscani. Turin : Loescher.

Pirrelli, Vito. (2000). Paradigmi in morfologia Un approccio interdisciplinare alla flessione verbale dell’italiano. Pisa-Rome : Istituti editoriali e poligrafici internazionali.

Pirrelli, Vito ; Battista, Marco (2002). The paradigmatic dimension of stem allomorphy in Italian verb inflection. Rivista di linguistica 12 : 307-80.

Prader Schucany, Silvia, (1970). Romanisch Bünden als selbständiges Sprachlandschaft. Bern : Francke.

Quint, Nicolas (1998). Le Parler occitan alpin du Pays de Seyne, Alpes-de-Haute-Provence. Paris : L’Harmattan.

Rohlfs, Gerhard (1968). Grammatica storica della lingua italiana e dei suoi dialetti. Morfologia. Turin : Einaudi.

Ronjat, Jules (1930-41). Grammaire istorique des parlers provençaux modernes. Montpellier : Société des Langues Romanes.

Santamarina, Antonio (1974). El verbo gallego. Santiago : Verba Anejo 4.

Schädel, Bernhard (1903). Die Mundart von Ormea. Halle : Niemeyer.

Signorell, Faust (2001). Vocabulari surmiran tudestg/Wörterbuch Deutsch Surmiran. Chur : Lehrmittelverlag des Kantons Graubünden.

Stewart, Thomas ; Stump, Gregory (2007). Paradigm-function morphology and the morphology-syntax interface. In Ramchand, Gillian ; Reiss, Charles (eds). Oxford Handbook of Linguistic Interfaces. p. 383-421.Oxford : OUP.

Stump, Gregory (2006). Heteroclisis and paradigm linkage. Language 82 : p. 279-322.

Tekavčić, Pavao (1980). Grammatica storica della lingua italiana. II Morfosintassi. Bologna : il Mulino.

Väänänen, Veikko (1963). Introduction au latin vulgaire. Paris : Klincksieck.

Vieli, Ramun ; Decurtins, Alexi (1962). Vocabulari romontsch-sursilvan-tudestg. Chur : Ligia Romontscha.

Zörner, Lotte (2008). I dialetti occitani della Valle Po. Turin : Valados usitanos.

Haut de page


1  Space precludes discussion of nouns and adjectives. These too generally had invariant roots. In Romance, there is extensive root allomorphy, largely the product of regular sound change. See Maiden (forthcoming b).

2  For the aberrant status of the 1pl and 2pl roots in the subjunctive, see 3.4.

3  From coll (i) gere ‘gather’: note that /1g/ + front vowel yields /λλ/.

4  Romanian replaces subjunctive by present indicative forms, save in third person.

5  Except Logudorese Sardinian where, significantly, there is no evidence for creation of novel N-pattern allomorphy.

6  See Corbett (2007).

7  The ‘complement’ is those cells other than singular + third person present.

8  Exceptions typically involve locally general syncretisms between cells: e.g., in parts of Liguria and Piedmont 2pl is systematically formed from 2sg; in Tuscany, a (reflexive) 3sg form tends to replace the 1pl, and so forth.

9  The present subjunctive has donare in all cells. This seems to manifest the tendency to invariance in present subjunctive roots discussed in 3.4.

10  In the context of Romansh, characterized as it is by extreme and idiosyncratic patterns of vocalic alternation, the fact that the arhizotonic alternant was du-could hardly explain loss of one of the allomorphs.

11  See, for example: Vieli & Decurtins (1962: 219; 701).

12  For evidence of incursion of fourth conjugation morphology into the N-pattern complement of third conjugation verbs in Romansh, see Maiden (2009).

13  The sound change also produced identity to the first conjugation in the imperfect.

14  Maiden (2009) shows that in certain S.W. Romanian dialects a rule of centralization after /s/ has triggered the same pattern of heteroclisis in some third conjugationverbs.

15  This fact is in itself remarkable, since (excepting oír ‘hear’) all other Spanish verbs of the -ir conjugation containing a back vowel in the root, show /u/ in these cells. Abolir flies in the face of this generalization, yet there is no lack of speaker confidence in using the forms in /o/.

16  Signorelli (2001); also Anderson (2008).

17  Pace, e.g., Burzio (2004) who asserts that « whatever identity relations have a statistical presence in the data, also have, ipso facto, a grammatical status, expressible as faithfulness constraints in the O[ptimality] T[heory] formalism ».

18  An earlier attempt is Fanciullo (1998), for a response to which see Maiden (2001b) and Pirrelli (2000: 79f.; 178-84).

19  See Bybee & Pardo (1981: 956f.) for analysis of similar data in Spanish which speak against the ‘syntagmatic’ approach.

20  The /j/ survives (but not in the indicative) in the subjunctives abbia ‘have’ and sappia ‘know’. The fact that it survives after labials in two high frequency verbs hardly justifies setting it up underlyingly for verbs in which in any case it would have a different paradigmatic distribution.

21  Even if this were possible, the paradigmatic distribution of the alleged i would in turn need explanation.

22  Conde Saiz (1978 : 177) ; Arnal Purroy (1998 : 355 ; 362) ; Alvar et al. (1995).

23  See Quint (1998: 55); Bendel (1934:97f.); Ronjat (1937:245).

24  The last two if the penult has a consonantal coda.

25  Anderson claims that the observed distribution of alternants involves a different from my ‘morphomic’ N-patterns. In fact, we find in Savognin a perfectly common version of the N-pattern (see 3.4).

26  Although, as he crucially demonstrates, it is synchronically impossible to reduce these alternants to unique underlying forms. They have to be listed in the representation of each verb.

27  Compare also Stewart & Stump (2007: 408). What I lack is a Romance example involving periphrases comprising anything larger than word-form + clitic (e.g., auxiliary + head verb), but it is a possibility worth looking out for.

28  Indeed, the fact that both N-pattern and L/U-pattern cause a particularly ‘salient’ fracture, involving the morphology of the present tense, helps to explain why they have persisted so long and so widely: the relevant alternations are presumably acquired fairly early by speakers.

Haut de page

Table des illustrations

Fichier image/jpeg, 118k
Haut de page

Pour citer cet article

Référence électronique

Martin Maiden, « From pure phonology to pure morphology the reshaping of the romance verb », Recherches linguistiques de Vincennes [En ligne], 38 | 2009, mis en ligne le 01 octobre 2011, consulté le 30 mars 2017. URL : ; DOI : 10.4000/rlv.1765

Haut de page


Martin Maiden

University of Oxford

Haut de page

Droits d’auteur

© Presses universitaires de Vincennes

Haut de page