Skip to content
BY 4.0 license Open Access Published by De Gruyter Mouton September 30, 2021

Stress and stem allomorphy in the Romance perfectum: emergence, typology, and motivations of a symbiotic relation

Borja Herce
From the journal Linguistics


Perfective stem allomorphy and stress are morphological traits which interact in complex ways in Romance verbal inflection. This article surveys the whole range of variation of these traits across Romance varieties, typologizes the observed interactions between the two, and examines attested and unattested possibilities. A comparison between the modern-day and the original Latin systems suggests that there is a strong pan-Romance bias against having verbs with a concrete combination of properties: perfective root-stress and no perfective stem alternation. This is a combination of traits that would have frequently resulted in diagonal syncretisms between past and present given the phonological changes attested in the daughter languages. Homophony avoidance (and the adaptive-discriminative role of morphology more generally) are therefore argued to motivate the observed bias.

1 Introduction

The predictability of formatives and morphological properties from other elements in the same word or paradigm has recently (re-)emerged as a crucial domain of study in theoretical (e.g., Blevins 2016) and typological (e.g., Stump and Finkel 2013) morphology. The research of its role in diachrony, by contrast, has lagged behind. In general, more predictability is equated with greater inflectional simplicity and should be expected to constitute a powerful force in diachronic analogical change. Thus, any change where two formerly independent traits become predictive of each other would seem to demand an explanation along these lines (see e.g., Herce 2020a). This is the case of perfective stem allomorphy and stress in Romance, which constitute the topic of the present paper.

Due to their extraordinary importance in the evolution of verbal morphology in these languages, stem allomorphy and stress have been studied frequently and for a long time (see e.g., Maiden 1992, 2018a; Malkiel 1966; Penny 1991). Unlike in Latin, the two properties are purely morphological (i.e., morphomic, see Aronoff 1994) in Romance and tend to go “hand in hand” in many contemporary varieties. The synchronic correlation that holds between stem alternation and stress in the present tense (e.g., Sp. /ˈpwedo/ ‘I can’ vs. /poˈdemos/ ‘we can’) is usually a byproduct of the diachronic causation of the former by the latter, as the difference in stem vowel (e.g., /we/ vs. /o/) was derived from regular sound changes in stressed versus unstressed contexts. In certain Romance varieties, however, stem allomorphies in other parts of the paradigm have also come to be correlated to stress without there being any kind of diachronic causation of one by the other. Thus, it is well-known (see e.g., Esher 2015; Maiden 2018b) that in Italo-Romance, and less robustly in some NW Ibero-Romance varieties, the former perfective stems from Latin have become confined to stressed-root (aka. ‘rhizotonic’) contexts. Consider the Latin and Italian subparadigms in Table 1.

Table 1:

Partial paradigm of two verbs in Latin and Italian (stressed syllable in bold) (Maiden et al. 2010; Pellegrini and Passarotti 2018).

‘make’ ‘say’
Latin Italian Latin Italian
1SG.PRES.IND faciō faccio dico
2SG.PRES.IND facis fai cīs dici
3SG.PRES.IND facit fa cit dice
1SG.PERF.IND feci dissi
2SG.PERF.IND cis facesti xis dicesti
3SG.PERF.IND cit fece xit disse
1SG.PLUP.SBJV cissem facessi xissem dicessi

As this Table shows, the former perfective stems have trimmed their paradigmatic distribution in Italian to appear only in rhizotonic paradigm cells nowadays. What is not sufficiently appreciated, however, is that various other Romance varieties (e.g., Spanish and Portuguese) have undergone a similar development, only at the lexeme level and not at the level of the individual paradigm-cell. In these languages, thus, whenever a lexeme has rhizotonic forms in the preterite it will have a special root in the former perfective tenses and vice versa (consider, for example, Spanish /ˈdixe/ ‘say.1SG.PRET’ and /deˈθia/ ‘say.1SG.IPFV’, vs. /meˈdi/ ‘measure.1SG.PRET’ and /meˈdia/ ‘measure.1SG.IPFV’). The predictability of these two traits was not inherited from Latin either, and is not found in all contemporary Romance varieties. In Romanian and certain varieties of Aragonese, perfective roots can appear in consistently arrhizotonic (i.e., suffix-stressed) verbs. As these facts suggest, the paradigmatic and lexical distribution of rhizotony and perfective allomorphy are highly variable across Romance and deserve to be explored in detail.

This variability is particularly interesting because diachronic stability has often been portrayed as one of the most prominent diagnostic qualities of morphomic traits like these (see Maiden 2011). Surveying this variation and the corresponding traits in Latin may help us understand whether/why these particular traits are unstable and what the forces are that have shaped their diachronic (co)evolution. Thus, although considerable research has been undertaken on individual varieties (particularly in those like Italian where the two traits have become coextensive in the paradigm), we still lack a thorough family-wide picture of perfective stem allomorphy and rhizotony, as well as a typology of the possible associations between the two. Causal explanations of change patterns in this domain are also hard to find (see Esher 2015 and Maiden 2018b for this kind of work in concrete varieties).

The research questions of the present paper are thus: To what extent and in what ways is preterite stem alternation correlated to stress in Romance? Why did this connection become established in different varieties? Was this correlation inherited to some extent or did it evolve separately in different languages? To answer these, Section 2 will present in detail the distribution of rhizotony and perfective stem alternants in the Latin lexicon and paradigm. Section 3 will present these same morphological traits in maximally diverse modern Romance varieties and will typologize the possible relationships and implicatures between stress and perfective stem allomorphy both at the lexical and the paradigmatic level. Section 4 will discuss the overall findings. By contrasting the attested (vs. unattested) property combinations in Romance, and comparing these with Latin, possible morphological change biases will be identified such as morphome orthogonality, predictability, homophony avoidance, etc. Finally, Section 5 will recapitulate the findings and present the conclusions and their implications for theoretical morphology.

2 The inherited system: rhizotony and perfectum stems in Latin

To a Roman grammarian, it would be quite striking that all Romance languages, in stark contrast to their ancestor Latin, are characterized by both free stress and by frequent alternations between the form of the stem found in different person-number cells of the paradigm. Furthermore, they would notice that there is often a symbiotic relationship, as it were, between stress and stem alternations. This is so, of course, because the former was often the source of the latter, particularly in the present subparadigm (see Table 2).

Table 2:

Present tense of the Latin verb computāre ‘calculate’ and its Spanish descendant (Maiden et al. 2010; Pellegrini and Passarotti 2018).

Latin computāre Spanish contar
1SG ˈkomputo: ˈkomputem ˈkwento ˈkwente
2SG ˈkomputa:s ˈkompute:s ˈkwentas ˈkwentes
3SG ˈkomputat ˈkomputet ˈkwenta ˈkwente
1PL kompuˈta:mus kompuˈte:mus konˈtamos konˈtemos
2PL kompuˈta:tis kompuˈte:tis konˈtais konˈteis
3PL ˈkomputant ˈkomputent ˈkwentan ˈkwenten

Latin was characterized by a phonologically predictable[1] assignment of stress and by an invariable stem (e.g., komput-) regardless of person-number inflection.[2] Later, however, sound changes would disrupt this harmonious picture. The deletion of certain unstressed segments and the loss of vowel length (e.g., /kompu'ta:re/ > /kon'tar/) rendered the location of stress unpredictable from a phonological or phonotactic point of view. Other sound changes like diphthongization created a divergence between the quality of a stressed and an unstressed vowel (e.g., /we/ vs. /o/ in the paradigm in Table 2). As a result of these changes, stem alternation and rhizotony became correlated morphological properties in the verbal inflection of most Romance languages.

In these highly frequent tenses and persons (i.e., SG+3PL of the present + 2SG imperative), rhizotony and stem allomorphy proved to be notoriously resilient and somewhat productive even in the absence of an extramorphological correlate.This emergent morphological entity has come to be recently known as the N-morphome.[3] The evolution of this stem allomorphy and others, the grammatical status of the emerging paradigmatic patterns, and the exact relation of rhizotony and stem allomorphy in different varieties has been a recurrent topic for research (see Anderson 2011; Herce 2019; Maiden 2011; O’Neill 2014, etc.).

There are, however, two important caveats to everything that has been mentioned here so far. The first one is that Latin was characterized by the presence of other rhizotonic forms in the paradigm beyond the present-tense ones shown in Table 2. In addition to this, and despite what was mentioned before (i.e., that Latin lacked stem allomorphy based on person), stem alternation did have a prominent role in the marking of other grammatical categories (see Table 3).

Table 3:

Partial paradigm of the Latin verb coquere ‘cook’ (Pellegrini and Passarotti 2018).

Latin Infectum Latin Perfectum
1SG ˈkokʷo: ˈkokʷam ˈkoksi: ˈkokseram kokˈsissem
2SG ˈkokʷis ˈkokʷa:s kokˈsisti ˈkoksera:s kokˈsisse:s
3SG ˈkokʷit ˈkokʷat ˈkoksit ˈkokserat kokˈsisset
1PL ˈkokʷimus koˈkʷa:mus ˈkoksimus kokseˈra:mus koksisˈse:mus
2PL ˈkokʷitis koˈkʷa:tis kokˈsistis kokseˈra:tis koksisˈse:tis
3PL ˈkokʷunt ˈkokʷant ˈkokserunt[4] ˈkokserant kokˈsissent

The paradigm of ‘cook’ (and most other verbs) can be said to be “split in half” along the two values of the feature aspect (i.e., imperfective vs. perfective [shaded cells in Table 3]). In the first and fourth conjugations, the perfect was formed usually without stem alternations, by means of a /w/ suffix after the theme vowel (e.g., ama:- vs. ama:w- ‘love’). In second and third conjugation verbs, by contrast, stem alternations were the rule. These were very diverse morphologically, involving vowel lengthening (e.g., wen- we:n- ‘come’), vowel apophonies (ag- e:g- ‘do/drive’), reduplication (e.g., mord- momord- ‘bite’), the loss of a ‘nasal infix’ (e.g., skind- vs. skid- ‘cut’), a /w/ augment (e.g., wol- wolw- ‘want’), an /s/ augment (e.g., kokʷ- vs. koks- above), suppletion (e.g., fer- tul- ‘carry’) and often more than one of these concurrently (e.g., po:n- posw- ‘put’, kad- kekid- ‘fall’). Notwithstanding these facts, a non-negligible number of verbs displayed no marking of perfectivity whatsoever and a single stem was used in perfective and imperfective aspects alike (e.g., de:fend- ‘defend’, wert- ‘turn’ etc.).

Rhizotony also occurred in the perfective aspect in most second and third conjugation verbs. In the first and fourth conjugation, by contrast, due to the presence (usually) of the long thematic vowels /a:/ and /i:/ respectively in the perfective, the prevailing pattern was for all word forms to be arrhizotonic (e.g., /aˈma:wit/ love.PERF.3SG, /auˈdi:wi:/ hear.PERF.1SG). This may be the origin of the strong general tendency throughout Romance for preterite endings to become stressed, as in the most productive conjugation, the first.

Be that as it may, concerning the role of rhizotony in the perfectum it is important to note the big difference in comparison to the infectum. In the latter we saw (Table 2) that sound changes (happening after Classical Latin) were often the source of stem vowel alternations. Unlike in the infectum, stress-related sound changes did not leave much trace in Romance in the perfectum. Thus, for example, although short stressed /o/ and /e/ diphthongized in the infectum in many varieties (e.g., Spanish /'kokʷit/ > /'kweθe/), it did not do so in the perfectum (or this left, at any rate, no trace, i.e., /'koksit/ > Sp. /ko'θjo/, cf. expected */’kwexe/). The generalized arrhizotony of many lexemes in the perfectum, along with the lower frequency of occurrence of perfective tenses compared to the infectum (see Figure 1), probably explains the failure of stress-based sound changes to generate person-number stem alternations in the perfectum subparadigm, in stark contrast, as I say, with the infectum.

Figure 1: 
Percentage of lexemes with rhizotonic forms in infectum and perfectum (left) and relative frequency of use of these two categories (right).

Figure 1:

Percentage of lexemes with rhizotonic forms in infectum and perfectum (left) and relative frequency of use of these two categories (right).

Looking at the 300 most frequent Latin verbs in LatInFlexi1.1 (Pellegrini and Passarotti 2018), all of them[5] have rhizotonic word forms in the infectum, while only 69.1% (N = 186) have rhizotonic forms in the perfectum. These differences are statistically significant according to a chi-square test of independence (X 2 = 8.6, p < 0.01). At the same time, perfectum tenses are much less token frequent (N = 2,127,179, 21.4%) than infectum ones (N = 7,814,867, 78.6%), with these differences again highly statistically significant (X 2 = 2,251,082, p = 0).[6] Because of these stark asymmetries, thus, even when rhizotony was preserved in some cells (for example, in Spanish, in 1SG and 3SG of the preterite), and had the right vowel to create stem vowel apophonies, this did not lead to person-based stem alternations in the preterit.

Based on the Latin facts, one has to note that rhizotony and stem[7] alternations were in principle orthogonal phenomena. Perfectum rhizotony and stem alternation occurred in a subset-superset relation in the paradigm, but both could appear independently of each other. All four logically possible combinations of these 2 binary properties, are therefore attested (see tetrachoric Table 4). Petere can serve as an example of a verb without perfective rhizotony and without perfective stem allomorphy. Dicere can provide an example of a verb with both perfective rhizotony and perfective stem alternation. Quaerere in turn, is a verb with perfective stem allomorphy but no rhizotonic cells. Last (but definitely not least, as will become clear later in the discussion in Section 3), vertere is a verb with rhizotonic cells in the perfective but with no stem allomorphy. Although all combinations of properties existed (see Table 4), they were not equally common.

Table 4:

The relation between perfective rhizotony and allomorphy.

Stem allomorphy
yes no
Rhizotony yes 57.6% e.g., dīcere 7 11.5% e.g., vertere
no 0.7% e.g., quaerere 30.1% e.g., petere

At the level of the lexeme, perfective rhizotony and perfective stem allomorphy occur together very often. Among the 300[8] most frequent Latin verbs (as reported in Pellegrini and Passarotti’s 2018 LatInFlexi1.1), more than half of them (57.6%, N = 155) fall in this category. Next in frequency (30.1%, N = 81) come verbs which have neither perfective rhizotony nor stem allomorphy. This class includes the productive conjugation -āre and would thus be much more frequent if more (uncommon) verbs were surveyed. Not rare either (11.5% N = 31) are verbs that have rhizotonic cells in the perfective but lack stem allomorphy. Verbs with perfective stem allomorphy but no rhizotonic cells are, by contrast, extremely rare (quaerere and requīrēre are the only ones found to have these traits). These differences are highly statistically significant according to a chi-squared test (X 2 = 154.66, p = 0).

A first and very important thing to note about rhizotony and perfective stem allomorphy in Latin is therefore that, although the two traits are orthogonal from a logical point of view, this does not translate into an independence of the two traits at the statistical lexical level. Thus, if a verb lacked rhizotonic forms in the preterite, it almost certainly lacked a perfective stem alternant. If it had rhizotonic forms in this perfective swath of the paradigm, chances are it would also have a dedicated stem. The orthogonality of these two traits, thus, was already very imperfect in Latin because two combinations of properties (rhizotony+allomorphy and arhizotony+no-allomorphy) were much more common than the others. This must have been perceived by language users from the very beginning, particularly after stress became morphological and had to start being learnt independently of the phonotactic structure of the word. It might thus not be surprising that some Romance varieties have strengthened these tendencies into more robust implicative relations between rhizotony and perfective allomorphy. The following section presents the paradigmatic distribution of perfective stem alternation and rhizotony across modern Romance offshoots and typologizes the possible predictability relations between these two morphological traits.

3 The modern systems: rhizotony and perfectum stems in Romance

Both rhizotony and perfective stem allomorphy have been comparatively unstable traits in Romance. This is often in stark contrast to comparable traits in other languages. Consider, for example, antepenult stress in the Greek past tense (Bubenik 1979), or the stem alternations in the aorist (Bubenik 1997). Consider also the diachronic resilience of stem alternations between present and past in Germanic strong verbs (Hewson 2001), or between perfective and imperfective aspects in Slavic (Wiemer and Seržant 2017). Although we lack studies that precisely quantify change in these domains, it seems clear that Romance perfective allomorphy has been more unstable and has lost ground in the inflectional system of Romance at a much faster rate compared to what we find in these other Indo-European branches.

Unlike these various comparable traits in Greek, Germanic, and Slavic, rhizotony and stem allomorphy became exclusively morphological phenomena in early Romance. The perfective tenses of Latin are no longer aligned to this semantic value in modern varieties and rhizotony does not correlate either to any phonological or semantic property. Although these inflectional traits in different IE branches are of course different in other respects, one cannot help wondering whether it is the failure of these traits to align to functional categories in Romance that has undermined their diachronic stability. In any case, the comparative instability of these traits, at both the lexical and the paradigmatic levels, has resulted in considerable variability and thus in an interesting history in general. This section will explore the fate of these morphological traits in modern Romance varieties. To do this, I use data from the Oxford Online Database of Romance Verb Morphology (Maiden et al. 2010).

3.1 The domain of perfective stem allomorphy

The overall tendency in Romance has been to restrict the domain of perfective stem allomorphy over time. This has taken place particularly at the lexical level. Thus, whereas a considerable number of Latin verbs displayed it (see Table 4), the number and percentage of verbs that display this allomorphy in modern Romance varieties has been greatly reduced. Although some verbs that used to lack a perfective stem have acquired one analogically (e.g., Lat. volvere > volvī has become It. volvere > volsi), this happens infrequently, and it is the prevalent trend everywhere to reduce the number of verbs that are subject to this stem allomorphy. Although the reasons for this must be complex, the lack of productivity of stem alternation must be largely due to its low occurrence in the -are conjugation.

Something not entirely dissimilar holds at the paradigmatic level. Although analogical extensions of the perfective root to parts of the paradigm where it did not appear in Latin are attested (consider the extension of this root to the gerund [see Pato and O’Neill 2013] or to the 1PL and 2PL present subjunctive [see Maiden 2012] in some Ibero-Romance varieties), such developments are rare. The prevalent trend is, once again, for the paradigmatic domain of perfective allomorphy to be reduced. One of the ways in which this may happen is, trivially, through the loss of tenses.

As Table 5 shows, the number of (perfectum) tenses has tended to decrease in the history of Romance. This loss of erstwhile perfective tenses is weaker in some varieties (e.g., Portuguese and Spanish) and stronger in others (e.g., Alpago Italian [Zörner 1997] and Nuorese Sardinian [Pittau 1972]). When new tenses were created (e.g., the synthetic future and conditional) these have not been based upon the perfectum root, which means that its overall spread within the inflectional paradigm has been declining.

Table 5:

3SG forms of ‘make/do’ of the perfectum tenses in Latin and their descendants in some Romance varieties (labels refer only to the values in Latin) (Herce 2020b: 150).

Latin Portuguese Spanish French Alpago Italian Nuorese Sardinian
PAST.IND fēcerat fizera hicieɾa
PAST.SBJV fēcisset fizesse hiciese fît ˈfese
PRES.IND fēcit fez hizo fit
PRES.SBJV fēcerit fizer
FUT.IND fēcerit

Another different process that achieves the same result is the eviction of the perfective root from paradigm cells where it used to occur in Latin. In Italo-Romance varieties (see Table 6), the former perfective stem no longer occurs in all the word forms where it used to appear in Latin. In those verbs where this root has not been simply lost everywhere, it tends to be preserved only in some cells (3SG, 3PL, 1SG, 1PL) of the former perfect indicative. Similar cases are also found beyond Italo-Romance.

Table 6:

Remnants of the perfectum root in various Italo-Romance varieties.

Sicilian ‘have’

(Maiden et al. 2010)
Italian ‘cook’

(Maiden et al. 2010)
Calvello Italian ‘say’

(Gioscio 1985)
1SG ˈappɪ aˈvissɪ ˈkɔssi kwoˈʧessi rəˈʧjettə rəˈʧessə
2SG aˈviʃtɪ aˈvissɪtʊ kwoˈʧesti kwoˈʧessi rəˈʧistə rəˈʧissə
3SG ˈappɪ aˈvissɪ ˈkɔsse kwoˈʧesse ˈressə rəˈʧessə
1PL ˈappɪmʊ aˈvissɪmʊ kwoˈʧemmo kwoˈʧessimo rəˈʧɛmmə rəˈʧessəmə
2PL aˈviʃtɪvʊ aˈvissɪvʊ kwoˈʧeste kwoˈʧeste rəˈʧistəvə rəˈʧessəvə
3PL ˈappɪrʊ aˈvissɪrʊ ˈkɔssero kwoˈʧessero ˈressərə rəˈʧessərə

As Table 7 illustrates, other varieties have also trimmed the inherited distribution of perfective root allomorphy. Whereas in some of these (Oscos Galician), it has retracted, like in Italo-Romance, to a subset of the cells of the perfect indicative, in others (Sassarese Sardinian, Panticosa Aragonese, and also in Aromanian [see Caragiu-Marioţeanu and Saramandu 2005]) the retraction has been to that tense as a whole.

Table 7:

Remnants of PYTA root in various non-Italo-Romance varieties.

Oscos Galician ‘put’

(Maiden 2018a: 76)
Sassarese Sardinian

(Bazzoni 1999)
Panticosa Aragonese ‘want’

(Nagore Lain 1986)
1SG ˈpuʃeŋ poˈɲɛse ˈfesi faˈdzissi(a) kiˈsje keˈɾese
2SG poˈɲitʃe poˈɲɛses ˈfesi faˈdzissi kiˈsjos keˈɾeses
3SG ˈpuʃo poˈɲɛse ˈfesi faˈdzissi(a) kiˈsjo keˈɾese
1PL poˈɲɛmos poˈɲɛsemos ˈfesimi faˈdzissiami kiˈsjemos keˈɾesemos
2PL poˈɲɛstes poˈɲɛseðes ˈfesiddi faˈdzissiaddi kiˈsjojs keˈɾesejs
3PL poˈɲɛroŋ poˈɲɛseŋ ˈfesini faˈdzissiani kiˈsjon keˈɾesen

Before the relation to stress is addressed, it should be mentioned that there appears to be a certain systematicity with regards to the paradigmatic loci where perfective allomorphy is preferably lost/preserved. Where two cells behave differently, the most frequent of them will be the conservative one and preserve the inherited allomorphy.

Perfectly matching the relative frequencies of the different values in Figure 2, a perfective stem alternant tends to be lost in 1PL before 1SG, in 3PL before 3SG, in 2 before 1, in 1 before 3, and in the pluperfect subjunctive (also in the perfect subjunctive in Aromanian) before the perfect indicative. Thus, even if the importance of token frequency has been sometimes played down (see e.g., Albright 2010), the observable diachronic changes in Romance argue in favor of this being a force of the utmost importance in morphological evolution.

Figure 2: 
Frequency (total tokens) of some paradigm cells in LatInfLex1.1 (logarithmic scale).

Figure 2:

Frequency (total tokens) of some paradigm cells in LatInfLex1.1 (logarithmic scale).

3.2 The domain of perfective rhizotony

The previous section has surveyed the state of perfective allomorphy in modern Romance with particular attention to paradigmatic distributional changes to the inherited status quo. This section will do the same regarding stress in that domain. Similarly, to the situation with regards to stem allomorphy, rhizotony has also been in decline at the level of the lexicon. The most notable development here is, thus, that many 2nd and 3rd conjugation verbs have been made arrhizotonic in Romance. In Latin, these verbs were almost invariably rhizotonic in the perfectum (see e.g., ‘say’ in Table 8) because they lacked a thematic vowel in that part of the paradigm. Because of this, it is challenging (and the topic has inspired an extensive literature, see e.g., Bonfante 1941; Wheeler 2012; Williams 1930), to ascertain which lexemes and forms might have provided an analogical model for this “regularization”.

Table 8:

Some perfectum tenses of ‘say’ in Latin and their Portuguese reflexes.

Latin ‘say’

(Pellegrini and Passarotti 2018)
Portuguese ‘say’

(Maiden et al. 2010)
1SG ˈdi:ksi: ˈdi:kseram di:kˈsissem ˈdisɨ diˈsɛɾɐ diˈsɛsɨ
2SG di:kˈsisti: ˈdi:ksera:s di:kˈsisse:s diˈsɛʃtɨ diˈsɛɾɐʃ diˈsɛsɨʃ
3SG ˈdi:ksit ˈdi:kserat di:kˈsisset ˈdisɨ diˈsɛɾɐ diˈsɛsɨ
1PL ˈdi:ksimus di:kseˈra:mus di:ksisˈse:mus diˈsɛmuʃ diˈsɛɾɐmuʃ diˈsɛsɨmuʃ
2PL di:kˈsistis di:kseˈra:tis di:ksisˈse:tis diˈsɛʃtɨʃ diˈsɛɾɐjʃ diˈsɛsɐjʃ
3PL ˈdi:kserunt ˈdi:kserant di:kˈsissent diˈsɛɾɐ̃ũ diˈsɛɾɐ̃ũ diˈsɛsɐ̃ĩ

Similarly, to what we mentioned for the decline of perfective allomorphy, the reasons for the decline of the perfective rhizotony must be many, but one of the main ones must be that the most productive conjugation in Romance (i.e., the Latin first) was characterized by arrhizotonic word forms throughout (e.g., 1SG.PRET /labo:'ra:wi:/ 2SG.PRES /labo:ra:'wisti/). Thus, and although exceptions can always be found (i.e., originally arrhizotonic verbs becoming rhizotonic, like Lat. /kʷai’si:wi:/ > Sp. /’kise/ 1SG.PRET.want), the prevalent trend is towards the loss of rhizotony.

This trend is also found at the paradigmatic level. Trivially, again, rhizotonic forms are obviously lost when whole tenses are lost (e.g., pluperfect indicative and perfect subjunctive) which contained them (see Table 5). However, rhizotony is often lost as well from word forms and tenses that have been preserved. Consider the paradigms in Table 8.

Table 8 shows how root-stress has been lost in Portuguese from the 1PL and 3PL of the preterite, as well as from the SG+3PL of the pluperfect indicative (also from the SG+3PL of the future subjunctive, e.g., 3SG ˈdi:kserit > diˈsɛɾ). Within tenses other than the perfect (and in the infectum other than the present indicative and subjunctive) there is a strong drive to level stress and fix it on the same syllable regardless of person-number. Arhizotony is consistently favored in these cases.

Because rhizotony has disappeared everywhere from perfectum forms outside the preterite, it is the latter tense that shows the most interesting developments for the present discussion. Compare the extent of perfect rhizotony in various Romance varieties In Table 9.

Table 9:

Rhizotony in Romance preterite I.

Sicilian ‘have’

(Maiden et al. 2010)
Italian ‘cook’

(Maiden et al. 2010)
Calvello Italian ‘say’

(Gioscio 1985)
Spanish ‘put’
1SG ˈappɪ ˈcossi rəˈʧjettə ˈpuse
2SG aˈviʃtɪ cuoˈcesti rəˈʧistə puˈsiste
3SG ˈappɪ ˈcosse ˈressə ˈpuso
1PL ˈappɪmʊ cuoˈcemmo rəˈʧɛmmə puˈsimos
2PL aˈviʃtɪvʊ cuoˈceste rəˈʧistəvə puˈsistes
3PL ˈappɪrʊ ˈcossero ˈressərə puˈsieron

From the Romance varieties above, Sicilian is the only one preserving the original paradigmatic distribution of rhizotony, all others having made the 1PL arrhizotonic. In addition to this, some of them lost rhizotony in the 1SG or 3PL as well. This, as well as the loss of rhizotony in other tenses (see Table 8) resembles the evolution of stem allomorphy in that, may be as expected from a comparatively irregular morphological trait, less frequent word forms are more prone to losing rhizotony than more frequent ones. That said, and in a way not unlike what was presented regarding stem allomorphy, root-stress is occasionally extended analogically to the originally arrhizotonic (i.e., 2) cells of the preterite (see Table 10).

Table 10:

Rhizotony in Romance preterite II.

Panticosa Aragonese ‘know’ (Nagore Lain 1986) Aromanian ‘do’ (Caragiu-Marioţeanu and Saramandu 2005) Romanian ‘put’ (Maiden et al. 2010)
1SG ˈsupe ˈfetʃu puˈsei
2SG ˈsupos fɨˈtseʃ puˈseʃ
3SG ˈsupo ˈfe̯atsi ˈpuse
1PL suˈpjemos ˈfe̯atsimu ˈpuserəm
2PL suˈpjois ˈfe̯atsitu ˈpuserətsʲ
3PL ˈsupon ˈfe̯atsirɨ ˈpuserə

In the varieties where this happens, it is due to the changed forms (e.g., 2SG /ˈsupos/, 2PL /ˈpuserətsʲ/) having become derived from or based upon a rhizotonic form (3SG /ˈsupo/, 3PL /ˈpuserə/) in the same tense. Extensions of rhizotony to new paradigm cells remain, in any case, less frequent than the loss of this trait from originally rhizotonic word forms.

3.3 The different (cor)relations between stem allomorphy and rhizotony

Already in Latin, as shown in Table 4, there existed a tendency for perfective stem allomorphy to occur in lexemes with some rhizotonic forms in that part of the paradigm. Stems without perfective allomorphy, in turn, strongly tended towards becoming arrhizotonic everywhere through the perfectum. Thus, although no perfect predictabilities existed in either direction, it can hardly be said that rhizotony and perfective stem allomorphy were completely independent from each other in Latin. This tendency, however, is still far from the exceptionless predictive relations that can be found in Romance synchronically. This section is intended to survey the different interactions between perfective stem allomorphy and rhizotony in the family.

3.3.1 One-directional implicature at the lexeme level

Even the least restrictive systems in Romance with regard to the combinability of these two morphological traits have increased their predictability compared to Latin. These are systems where all verbs that have rhizotonic word forms in the preterite have special perfective stems, but not all verbs with special perfective stems have rhizotonic word forms.

Table 11 shows the three classes of verbs (out of the logically possible 4, see Table 4) that exist in Panticosa Aragonese regarding these two morphological traits. The paradigm of saliɾ ‘exit’ represents the productive class, which lacks both rhizotony and perfective allomorphy. The verb keɾeɾ ‘want/love’, represents a verb that also lacks rhizotonic word forms but has perfective allomorphy, with a stem kis- (instead of keɾ-) in the preterite. Finally, the verb saper ‘know’ has both rhizotonic forms (in bold in Table 11) and perfective allomorphy. These same three classes are found in Romanian as well (see Table 12).

Table 11:

Partial paradigms of three Panticosa Aragonese verbs (Nagore Lain 1986).

saˈliɾ ‘exit’ keˈɾeɾ ‘want/love’ saˈpeɾ ‘know’
1SG saˈliβa saˈlje keˈɾeβa kiˈsje saˈpeβa ˈsupe
2SG saˈliβas saˈljos keˈɾeβas kiˈsjos saˈpeβas ˈsupos
3SG saˈliβa saˈljo keˈɾeβa kiˈsjo saˈpeβa ˈsupo
1PL saˈliβamos saˈljemos keˈɾeβamos kiˈsjemos saˈpeβamos suˈpjemos
2PL saˈliβais saˈljois keˈɾeβais kiˈsjois saˈpeβais suˈpjois
3PL saˈliβan saˈljon keˈɾeβan kiˈsjon saˈpeβan ˈsupon

Table 12:

Partial paradigms of three Romanian verbs (Maiden et al. 2010).

dorˈmi ‘sleep’ veˈde̯a ‘see’ ˈziʧe ‘say’
1SG dorˈme̯am dorˈmii veˈde̯am vəˈzui ziˈʧe̯am ziˈsei
2SG dorˈme̯ai dorˈmiʃ veˈde̯ai vəˈzuʃ ziˈʧe̯ai ziˈseʃ
3SG dorˈme̯a dorˈmi veˈde̯a vəˈzu ziˈʧe̯a ˈzise
1PL dorˈme̯am dorˈmirəm veˈde̯am vəˈzurəm ziˈʧe̯am ˈziserəm
2PL dorˈme̯aʦʲ dorˈmirəʦʲ veˈde̯aʦʲ vəˈzurəʦʲ ziˈʧe̯aʦʲ ˈziserəʦʲ
3PL dorˈme̯au dorˈmirə veˈde̯au vəˈzurə ziˈʧe̯au ˈziserə

In a system entirely parallel to the Aragonese one in Table 11, Romanian has verbs that lack both perfective rhizotony and stem alternation (e.g., ‘sleep’), verbs with stem alternation but no rhizotony (e.g., ‘see’) and verbs with both stem alternation and rhizotony (see ‘say’). What neither of these two varieties (or any other variety) has retained is the fourth logically possible combination of properties. There are no verbs whatsoever, therefore, with rhizotonic forms in this part of the paradigm but without stem allomorphy. No Romance variety, as far as I know, has preserved a class of verbs continuing the Latin one represented by vertere (Table 4). The disappearance of this option has meant that all Romance varieties have increased the predictability of one morphological trait from the other, since rhizotony has become a perfect predictor of stem alternation.

3.3.2 Bi-directional implicature Bi-directional implicature at the lexeme level

The one-directional predictive relation of the varieties in Section 3.3.1 has been taken one step further into others by turning it into a bi-directional one. These are systems in which all rhizotonic verbs have special perfective stems, and all verbs with special perfective stems have rhizotonic forms. Consider the paradigms in Table 13.

Table 13:

Partial paradigms of two Spanish verbs.

saˈlir ‘sleep’ saˈβeɾ ‘say’
1SG saˈlia saˈli saˈβia ˈsupe
2SG saˈlias saˈliste saˈβias suˈpiste
3SG saˈlia saˈljo saˈβia ˈsupo
1PL saˈliamos saˈlimos saˈβiamos suˈpimos
2PL saˈliais saˈlisteis saˈβiais suˈpisteis
3PL saˈlian saˈljeɾon saˈβian suˈpjeɾon

The system in Spanish, concerning the morphological traits surveyed here, is one where the four classes of Latin (Table 4) have been reduced to only two. One (the productive class) is characterized by no stem alternation and no rhizotony (see e.g., salir). The other class of verbs is characterized by a perfective stem alternant, different from the one in the former infectum tenses, as well as by the presence of rhizotonic word forms in the preterite (see e.g., saber). The two formerly independent and cross-classifying morphological traits have thus become aligned in the Spanish lexicon in a way that they are now mutually predictive.[9] The same system is found across Portuguese-Galician varieties. Bi-directional implicature at the word form level

A somewhat different system is found in varieties where the same bidirectional implicative relation that has been described in the previous Section occurs not at the lexeme level, but at the level of the individual word form. In these systems, all rhizotonic word forms of the former perfectum have a special stem, and all word forms with a special stem in that part of the paradigm are, conversely, rhizotonic.

Similarly to Spanish in Table 13, two classes of verbs can be identified in Sicilian (Table 14) regarding these two morphological traits. The productive one (e.g., ‘sing’) is characterized by the lack of perfectum stem alternation and the lack of rhizotony. The other class (e.g., ‘have’) is characterized by both stem alternation and rhizotony. In contradistinction to the system presented in the previous Section, arrhizotonicotonic word forms lack perfectum allomorphy also in those verbs which do have rhizotonic forms and stem alternation in their paradigm. Thus, the domain over which the mutual predictability holds between rhizotony and stem alternation has narrowed down in these varieties to the individual word form. Systems of this type are found across Italo-Romance, as well as in some western Ibero-Romance varieties. They are likely to have evolved from the system of Section

Table 14:

Partial paradigms of two Sicilian verbs (Maiden et al. 2010).

kanˈtarɪ ‘sing’ aˈvɪrɪ ‘have’
1SG kanˈtava kanˈtavʊ aˈvɪva ˈappɪ
2SG kanˈtavɪ kanˈtaʃtɪ aˈvɪvɪ aˈviʃtɪ
3SG kanˈtava kanˈta aˈvɪva ˈappɪ
1PL kanˈtavamʊ kanˈtamʊ aˈvivamʊ ˈappɪmʊ
2PL kanˈtavavʊ kanˈtaʃtɪvʊ aˈvivavʊ aˈviʃtɪvʊ
3PL kanˈtavanʊ kanˈtarʊ aˈvivanʊ ˈappɪrʊ

Such an analogical development (i.e., from Standard Galician to Oscos Galician, see Table 15) could be understood as a further strengthening of the predictive relations between rhizotony and perfective stem allomorphy. In a system like the Standard Galician one, the two traits are coextensive at the lexical but not the paradigmatic level, since word forms exist which lack rhizotony but do have a perfectum stem (see e.g., /souˈβɛtʃes/). It is an understandable simplification to further align the domains of the two traits. Rhizotony and stem allomorphy have thus become coextensive also at the paradigmatic level in varieties like Oscos Galician and in Italo-Romance (see Table 9).

Table 15:

Partial paradigms of ‘know’ in two Ibero-Romance varieties (Maiden et al. 2010).

saˈβeɾ ‘say’ Standard Galician saˈβeɾ ‘say’ Oscos Galician
1SG saˈβia ˈsouβeŋ saˈβia ˈsouβeŋ
2SG saˈβias souˈβɛtʃes saˈβias saˈβitʃe
3SG saˈβia ˈsouβo saˈβia ˈsouβo
1PL saβiˈamos souˈβɛmos saˈβiamos saˈβɛmos
2PL saβiˈaðes souˈβɛstes saˈβiaðes saˈβɛstes
3PL saˈβian souˈβɛɾon saˈβiaŋ saˈβɛɾoŋ
3.3.3 Loss of trait(s)

These are systems in which one or both morphological traits (i.e., perfective allomorphy and/or rhizotony) have disappeared from the language. As a result, the predictability of the traits has become, trivially, perfect.

The verbs in Table 16 are among the ones most commonly associated with perfective rhizotony and stem alternation in other Romance varieties, but they lack both morphological traits in Friulian and Romansh. This same fate is shared by all verbs in these varieties and in many Gallo-Romance ones.[10] Since variation (between rhizotonic and arrhizotonic, or between allomorphic and nonallomorphic) has disappeared entirely from this former perfectum paradigmatic domain, predictability is, trivially, unproblematic under these systems.

Table 16:

Partial paradigms of two verbs in two Romance varieties.

aˈvɛr ‘have’ Engadine Romansh (Scheitlin 1980) di: ‘say’ Friulian (Zof 2000)
1SG aˈvɛva aˈvɛt diˈzevi diˈzei
2SG aˈvɛvast aˈvɛtast diˈzevis diˈzeris
3SG aˈvɛva aˈvɛt diˈzeve diˈze
1PL aˈvɛvans aˈvɛtans diˈzevin diˈzerin
2PL aˈvɛvas aˈvɛtas diˈzevis diˈzeris
3PL aˈvɛvan aˈvɛtan diˈzevin diˈzerin

3.4 Interim discussion

Each of the systems that was presented through Section 3.3 (see also all the information in Appendix II) constitutes a simplification of the previous one because the systems represent progressively higher levels of inter-predictability between stress and stem allomorphy. In addition, as mentioned in Section 3, every single Romance variety has increased the predictability of these morphological traits compared to their ancestor Latin.

Reductions in the orthogonality or the cross-classification of two formatives constitute improvements of predictability (see Herce 2020a). It is also well-known that unpredictable variation is dispreferred in language (see e.g., Smith and Wonnacott 2010). In a perfectly orthogonal system (i.e., where two dichotomous variables combine into the logical maximum of 4 classes, see Latin in Table 17) no trait can be used to predict the other with absolute certainty. This is unfortunate and it may thus come as no surprise that Romance varieties have largely “remedied” this situation. Thus, in Romanian-type systems, within the former perfectum tenses, rhizotony predicts stem alternation. Spanish goes further and the reverse also holds (i.e., stem alternation predicts rhizotony too). In other varieties like Friulian, the ultimate predictable system has developed in which there is no variation whatsoever.

Table 17:

The relation between perfective rhizotony and stem alternation in the lexicon.

Latin Romanian, Pant. Aragonese Spanish, Oscos Galician Friulian, Romansh
Allomorphy + + + +
Rhizotony +

These same properties can be surveyed at the lexical (Table 17) and at the word-form level (Table 18). At this latter level, rhizotony in a given perfectum word form in Spanish predicts the use of a special stem allomorph, while the reverse does not hold. Some varieties improve this predictive relation by making it hold in the opposite direction as well. In Oscos Galician (see Table 15), the use of a perfectum stem allomorph in a particular word form predicts that this form will be rhizotonic.

Table 18:

The relation between perfectum rhizotony and stem alternation in the paradigm.


Stand. Galician
Oscos Galician

Allomorphy + +
Rhizotony +

In his analysis of morphome interactions, Herce (2019) argues that morphomic exponents in disjoint set relations sometimes simplify their distribution by reshaping their paradigmatic distributions into subset-superset relations. This paradigmatic arrangement allows one to predict the presence of the superset trait from the subset one. A further simplification (e.g., from Standard Galician to Oscos Galician) is to rearrange exponents into identical sets (Table 19).

Table 19:

Partial paradigms of ‘put’ in two Ibero-Romance varieties (Maiden et al. 2010).

poˈɲer ‘put’ Standard Galician poˈɲeɾ ‘put’ Oscos Galician
1SG poˈɲia ˈpuʃen puˈʃɛse poˈɲia ˈpuʃeŋ poˈɲɛse
2SG poˈɲias puˈʃɛʧe puˈʃɛses poˈɲias poˈɲiʧe poˈɲɛses
3SG poˈɲia ˈpuʃo puˈʃɛse poˈɲia ˈpuʃo poˈɲɛse
1PL poɲiˈamos puˈʃɛmos puˈʃɛsemos poˈɲiamos poˈɲɛmos poˈɲɛsemos
2PL poɲiˈaðes puˈʃɛstes puˈʃɛseðes poˈɲiaðes poˈɲɛstes poˈɲɛseðes
3PL poˈɲian puˈʃɛɾon puˈʃɛsen poˈɲiaŋ poˈɲɛɾoŋ poˈɲɛseŋ

Under this configuration, both traits can predict the other straightforwardly. It thus becomes a matter of definition whether we are dealing with two or just one morphome in these cases.[11] The ultimate simplification, of course, is once again for the variation to be lost entirely and to have uniformly arrhizotonic and nonalternating paradigms à la Friulian.

4 Pressures, biases and motivations

Earlier sections have presented the different types of predictability (one-directional vs. bidirectional, at the lexical vs. at the paradigmatic levels) that are found synchronically in Romance between perfective stem allomorphy and rhizotony. Given the fact that no implicative relations were present in Latin and given that every single Romance language has moved to a greater or lesser extent toward more predictability (see Table 17), one may feel tempted to identify this as the (sole) driving motivation.

However, given the fact that the surveyed morphological traits (i.e., rhizotony and perfective allomorphy) have been in decline everywhere, and given that not all the property combinations were equally frequent in Latin (see Table 20), it might not be unexpected that, even if the evolution of the two properties had been completely independent of each other, some implicature could have emerged by pure chance. The combinations of properties that tend to be lost in Romance (see Table 17) are, after all, the least common in Latin:

Table 20:

The relation between perfective rhizotony and allomorphy.

Stem allomorphy
yes no
Rhizotony yes 57.6% e.g., dīcere 11.5% e.g., vertere
no 0.7% e.g., quaerere 30.1% e.g., petere

Table 20 (repeated from Table 4) shows the relative size of each of the classes. The class represented by quaerere is the smallest one (with only two members among the 300 most frequent Latin verbs). If these verbs changed (any of) their properties, a one-directional implicature would already be established at the lexical level. This one-directional implicature would be different, however, from the one that is found across Romance. Throughout the family, the combination of properties that is more prone to being lost is not the least frequent one in Latin (+allomorphy, −rhizotony), but rather a much more common one (−allomorphy, +rhizotony) represented among others by vertere. This demands an explanation.

It is a pervasive diachronic pan-Romance trait that perfectum allomorphy and rhizotony constitute irregular, largely unproductive, recessive traits (see Sections 3.1 and 3.2). Because of this, the class of petere is expected to attract new members from the other three classes. One might thus expect, depending on the speed at which verbs change class, systems like Friulian (see Table 17) and systems like Spanish (Table 13). What one would never expect to emerge by chance is systems like the Romanian and Aragonese ones (i.e., with no vertere class, but with a quaerere class).

Given the initial size of each of the classes in Latin (11.5% of the disappeared one vs. 0.7% of the preserved one), a random “desertion” of verbs from the unproductive classes would be most unlikely to result in systems like the ones in Table 21. There seems to be, therefore, a relatively strong bias only against the vertere class, which has disappeared in every single Romance variety despite its comparatively large size in Latin.

Table 21:

Perfective rhizotony and allomorphy in Aragonese and Romanian.

Stem allomorphy Stem allomorphy
yes no yes no
Rhizotony yes saˈper ˈzitʃe
no keˈɾer saˈlir veˈde̯a dorˈmi

The reason for this bias is to be found, I believe, in the syncretism/homophony patterns that such a class of verbs would have given rise to in the paradigm. Consider the verbs in Tables 22 and 23.

Table 22:

Partial paradigms of some Italian verbs.

porre ‘put’ piacere ‘please’ fare ‘do’ cantare ‘sing’
1SG ˈpongo ˈposi ˈpjatʧo ˈpjakkwi ˈfatʧo ˈfeʧi ˈkanto kanˈtai
2SG ˈponi poˈnesti ˈpjaʧi pjaˈʧesti ˈfai faˈʧesti ˈkanti kanˈtasti
3SG ˈpone ˈpose ˈpjaʧe ˈpjakkwe ˈfa ˈfeʧe ˈkanta kanˈtɔ

Table 23:

Partial paradigms of some Spanish verbs.

caber ‘fit’ decir ‘say’ tener ‘have’ amar ‘love’
1SG ˈkepo ˈkupe ˈdiɣo ˈdixe ˈtengo ˈtube ˈamo aˈme
2SG ˈkabes kuˈpiste ˈdiθes diˈxiste ˈtjenes tuˈbiste ˈamas aˈmaste
3SG ˈkabe ˈkupo ˈdiθe ˈdixo ˈtjene ˈtubo ˈama aˈmo

Segmentally homophonous suffixes are found frequently in Romance (e.g., -e and -i in Italian, see Table 22, -e and -o in Spanish, see Table 23) in both the present tense and the preterite. They cannot, thus, by themselves, disambiguate between the word forms (e.g., 1SG.PRET vs. 2SG.PRS, 1SG.PRET vs. 3SG.PRS, and 1SG.PRS vs. 3SG.PRET) where they occur. A perfectum stem alternation (see e.g., ˈkabe vs. ˈkupe) or a difference in stress (e.g., ˈamo vs. aˈmo) preserves these morphological distinctions in the classes of dīcere and petere, and the two traits (would) do so redundantly in the class of quaerere (see Table 20). In the missing class of vertere, by contrast, none of these traits will reliably salvage the formal distinction,[12] and thus some forms would often become homophonous when no other formative or stem alternation pattern (accidentally) prevents it.

Although some (most notably Maiden 2018b: 23) have argued against a reasoning along these lines to explain related phenomena (namely the failure of stem levelling to apply to rhizotonic cells in cases like Italo-Romance and Oscos Galician [see Table 19]), I believe this homophony explanation may not be on the wrong track here. Maiden’s main concern is that, in his opinion, the amount of homophony that would result would not be sufficient or disruptive enough because of the occasional presence of orthogonal formatives (e.g., diphthongization and velar stem augments, see tener in Table 25) which in practice would prevent whole-word homophony in many cases (cf./ˈdiɣo/ vs. /ˈdiθo/).

Table 25:

Partial paradigms of some Pseudo-Spanish verbs, with levelled stem and stress.

caber ‘fit’ decir ‘say’ tener ‘have’ amar ‘love’
1SG ˈkepo ˈkabe ˈdiɣo ˈdiθe ˈtengo ˈtene ˈamo ˈame
2SG ˈkabes kaˈbiste ˈdiθes diˈθiste ˈtjenes teˈniste ˈamas aˈmaste
3SG ˈkabe ˈkabo ˈdiθe ˈdiθo ˈtjene ˈteno ˈama ˈamo

When applied to our broader Pan-Romance phenomenon (i.e., the absence of the vertere class), this reasoning faces two problems. The first is that it only considers the issue from one side: what would happen with the highly irregular contemporary rhizotonic verbs if they did not exhibit a stem alternation, and ignores the alternative perspective: what would happen with contemporary regular non-stem alternating verbs (e.g., amar) if they were rhizotonic in the perfective. Although the acquisition of perfectum rhizotony by these regular verbs nowadays would be, of course, completely unmotivated, one has to keep in mind that many of these contemporary regular arrhizotonics [e.g., Spanish /entenˈdi/ < /inˈtendi:/, /berˈti/ < /ˈwerti:/, /responˈdi/ < /resˈpondi:/, etc.] are analogically derived from vertere-type verbs. It is thus natural to ask what their paradigm would look like had they remained in their original class.

The second problem is that the basic language-processing nature of homophony avoidance is missed. Some hypothetical non-stem-alternating but rhizotonic 3SG forms like */ˈkabo/ */ˈdiθo/ */ˈteno/ */konˈduθo/, etc. may not be strictly homophonous with the 1SG.PRS forms /ˈkepo/ /ˈdiɣo/ /ˈtengo/ /konˈduθko/, etc. but they would be the 1SG.PRS forms of these verbs if the usual productive rule (i.e., add -o to the default stem) had applied in these verbs. Because of this, and even in the absence of true surface homophony, forms like 3SG.PRET *konˈduθo might still be subject to some clash with these potential (albeit unrealized) 1SG forms.[13]

Be that as it may, Maiden (2018b) targets a more localized paradigmatic change: that from a Spanish-type (see Section to an Italian-type (see Section perfectum-stem system. His explanation relies on the structural model provided by the highly frequent 3SG.PRET, which would have provided an island of conservativeness and autonomy in the paradigm à la Bybee (1988). This could well be right and sufficient to explain this paradigmatic change but does not address the Pan-Romance lexeme-level realignments (see Table 17) that concern us in this paper. Even if homophony-avoidance were relatively unimportant to account for the emergence of the Italo-Romance paradigm-cell-level realignment, it may still constitute a promising explanation for the pervasive absence of the vertere verb class throughout Romance. Although the dislike for this combination of traits appears to be still in place in some contemporary varieties (see Footnote 12), this proposed bias might have originated/operated long ago, for example in Vulgar Latin, when the loss or weakening of some word-final suffixes/segments (e.g., -m, -t) and of (unstressed) vowel distinctions (e.g., -i: -i -e: -e) would have threatened to generate syncretisms between various cells in the paradigm (e.g., between 3SG.PRS /wertit/, 1SG.PRET /werti:/, 2SG.IMP /werte/, etc.).

Languages can tolerate relatively high levels of ambiguity. It is true, as Maiden (2018b) points out, that Romance languages (like e.g., Spanish) tolerate PRS-PRET formal identities seemingly unproblematically. The form amamos, for example, is both the 1PL.PRS and the 1PL.PRET of the verb ‘love’. There are also recurrent 1SG/3SG formal syncretisms in the language (e.g., ame 1/3SG.PRS.SBJV, amaba 1/3SG.IPF etc.). These syncretisms, however, are of a quite different nature than the hypothetical ones in Table 25. A form like amamos can still be described (and presumably processed) as a 1PL form and could thus be understood simply as uninformative/unspecified for tense. Forms like ame fail to single out a concrete person value but do signal tense and number unmistakably and could thus be analyzed as similarly unspecified (for person). A form like the hypothetical /ˈkabe/ in Table 25, by contrast, would not allow such analysis, as it would have incompatible/irreconcilable feature values. This would be unfortunate from a processing perspective, as homonymy has been shown to be costlier than polysemy (Klepousniotou 2002; MacGregor et al. 2015).

There is a frequent syncretism in Spanish which involves similarly incompatible values, thus seemingly weakening the hypothesis that “diagonal” patterns of formal conflation are particularly dispreferred. As Table 26 shows, the 2SG imperative and the 3SG present indicative are regularly syncretic in Spanish. The two forms became homophonous in most cases due to regular sound changes, [14] which caused the loss of the 3SG ending -t, the loss of vowel length as a distinctive feature, and the merger of several vowels (e.g., amat vs. amā > ama vs. ama, currit vs. curre > corre vs. corre). It is revealing, however, that despite this massive syncretization in the largest inflectional classes, the two cells have often resisted systematic formal identity in various ways.

Table 26:

Partial paradigms of some Spanish verbs.

amar ‘love’ caber ‘fit’ decir ‘say’ tener ‘have’
2SG ˈama ˈamas ˈkabe ˈkabes di ˈdiθes ten ˈtjenes
3SG ˈame ˈama ˈkepa ˈkabe ˈdiga ˈdiθe ˈtenga ˈtjene

Under regular sound change, for example, 2SG.IMP and 3SG.PRS of tener should be *tjen (cf. tenet vs. tenē). Due to sound change, word-final /e/ was lost in certain contexts (cf. pan < pane < pānem ‘bread’, sal < sale < salem ‘salt’). This (should have) applied regularly, affecting both cells or neither depending on the phonotactics of the word. However, what we find is that deletion has only managed to survive in modern Spanish in the 2SG.IMP of some verbs, while is has been always analogically “turned back” in the 3SG.PRS (i.e., *sale/sale > *sal/sal > sale/sal ‘exit’, *pone/pone > *pon/pon > pone/pon ‘put’ etc.). Morphological change, thus, has sometimes come to the rescue of this morphosyntactic distinction.[15]

Syntactic changes have also contributed to the discriminability of the two forms, as the imperative, unlike other finite forms, has preserved in Spanish a postposed ordering of object clitics. That is, original da=me ‘give=1SG’ or mata=lo ‘kill=3SG’ were ambiguous as for their 2SG imperative (i.e., ‘give me!’, ‘kill him!’) or 3SG present (i.e., ‘gives me’, ‘kills him’) values. After the change in the syntagmatic ordering of clitics in finite non-imperatives, however, word order has come to disambiguate the values that had fallen together by sound change. In contemporary Spanish, thus, me=da must mean ‘gives me’ and da=me ‘give me!’, while lo=mata must mean ‘kills him’ and mata=lo ‘kill him!’.

One could argue that, although the formal identity of these two diagonally syncretic values seems not to be actively pursued or defended against disruptive changes (which would already be revealing), the former changes may not necessarily be due to homophony avoidance. There is, however, one last piece of evidence that still in contemporary Spanish something is indeed being actively avoided in some cases. Thus, the verb estar ‘be’ and others are defective in the 2SG.IMP (*¡está callado! ‘be quiet!’ / está callado ‘he is quiet’). This is not due to this verb’s semantics, as comparable verbs like ser ‘be’, in which the two forms are not syncretic (see Footnote 15), do not share this same quirk (¡sé bueno! ‘be good!’ / es bueno ‘he is good’). The near-synonymous verb estar=se is normally recruited to fill the gap missing from estar (¡está=te callado! ‘be quiet!’). This confirms that whatever prevents the existence of 2SG.IMP está is not operative in verbs where 2SG.IMP and 3SG.PRS are morphologically or syntactically distinguished.

Thanks to cases of defectiveness similar to this one, homophony avoidance has proved to be a valid (if also non-deterministic) pressure in morphological organization (Baerman 2011). This pressure must be particularly relevant in those cases where the syncretized forms have conflicting feature values (i.e., in cases of diagonal syncretism) and thus cannot be reconciled in a single lexical entry. This force is likely to be still stronger in those cases where, in addition, a big frequency asymmetry exists between the two forms:

As Table 27 shows, perfective forms in Tamashek differ from the corresponding short imperfective forms in their lack of prefixes. This results in some formal conflations, which are tolerated where the resulting form can be analyzed as unspecified (3SG.M+3SG.F = 3SG) but result in defectiveness when the collapsed forms have conflicting values (3SG+1PL). In these cases, the least frequent value is the missing one.[16] This is not unexpected from the perspective of processing, since the faster lexical access to the more frequent sense/lexical entry will give it a competitive advantage over the more infrequent one.

Table 27:

Partial paradigm of Tamashek ‘be black’ (Baerman 2011:15).

Short imperfective Perfective
1 ikwɑl-æɣ n-ikwɑl kæwɑl-æɣ ??
2.M t-ikwɑl-æd t-ikwɑl-æm kæwɑl-æd kæwɑl-æm
2.F t-ikwɑl-æd t-ikwɑl-mæt kæwɑl-æd kæwɑl-mæt
3.M ikwɑl t-ikwɑl-æn kæwɑl kæwɑl-æn
3.F t-ikwɑl t-ikwɑl-ænt kæwɑl kæwɑl-ænt

The importance of word discriminability has been well established in the lower levels of language. Segments with a high functional load, thus, have been shown to be phonetically enhanced (i.e., the /p/ in ‘pill’ is pronounced longer than the one in ‘pipe’ because of the need to keep the former distinct from ‘bill’, see Hall et al. 2018). At a more abstract level, phonemic distinctions with a high functional load (i.e., those which distinguish a large number of minimal pairs) are more resilient and tend to be preserved over those with a lower functional load (Wedel et al. 2013). It is hardly a far-fetched suggestion that more abstract levels of language like morphology might also be sensible to this same pressure to keep semantically irreconcilable word forms distinct.

Proponents of abstractive models have indeed already defined morphology as an “adaptive discriminative system” (Blevins et al. 2016). Under such a conception of the place of morphology in grammatical architecture, it is to be expected that inflectional systems, traits and changes will be dispreferred which result in deleterious homophony. The absence of rhizotonic non-stem-alternating preterite forms throughout Romance (see Table 21) would thus be motivated (partially) by the avoidance of accidental diagonal syncretisms (see Tables 24 and 25) between cells with incompatible feature values and with widely different token frequencies. This bias would never be deterministic, of course, (we know systems with these syncretism types are not unlearnable) but if it provides even a slight synchronic competitive advantage to certain traits over others, this may lead eventually (i.e., over long periods of time) to sizable asymmetries between them, and to the success of some traits and the extinction of others.

Table 24:

Partial paradigms of some Pseudo-Italian verbs, with levelled stem and stress.

porre ‘put’ piacere ‘please’ fare ‘do’ cantare ‘sing’
1SG ˈpongo ˈponi ˈpjatʧo ˈpjaʧi ˈfatʧo ˈfaʧi ˈkanto ˈkantai
2SG ˈponi poˈnesti ˈpjaʧi pjaˈʧesti ˈfai faˈʧesti ˈkanti kanˈtasti
3SG ˈpone ˈpone ˈpjaʧe ˈpjaʧe ˈfa ˈfaʧe ˈkanta ˈkanto

The morphological properties that have been surveyed in this paper (i.e., rhizotony and former-perfective stem alternation) are morphomic in the sense of Aronoff (1994). They are thus meaningless, purely morphological traits when one considers their contemporary distribution in the paradigm. They do not realize any value whatsoever and are therefore problematic for a strictly realizational conception of morphology that sees sub-word morphological units as just a vehicle for the direct expression of syntactico-semantic values. Under an abstractive conception of morphology, however, sub-word morphological units are not expected to bear meaning by themselves. Content differences can be signaled by whichever means are available in the paradigm to generate whole-word contrasts.

After the loss (due to sound changes) of some of the morphological material that used to encode tense distinctions reliably in Latin (most notably perfective /w/), Romance varieties sometimes “had to” reinvent themselves. In complex inflectional systems like these, there are many ways to potentially avoid person-number-tense ambiguities, both syntactically (e.g., adding compulsory subject pronouns) and morphologically. In this latter domain, suffixes were sometimes extended analogically to reliably signal values (e.g., -g has been extended in Occitan and Catalan across the preterite, see Lief 2004; and -r, and -t have been extended with the same value in Provençal and other varieties, see Bybee and Brewer 1980).

Along with meaningful morph(eme)s, however, meaningless traits like stress or stem differences can also be co-opted for the signaling of value differences in concrete (e.g., problematic) cases. If the hypothesis I put forward in this paper is on the right track, those morphological traits and trait combinations that failed to disambiguate diagonal syncretisms like those in Tables 24 and 25 would have been at a (slight) disadvantage against other competing morphological alternatives.

Thus, for example, a +Rhizotony −PerfectiveStem form like Latin 1SG.PRET /abs'kondi:/ ‘hid’ would have been expected to develop into Italian /as'kondi/, and into Spanish /es'konde/. These are not the forms we encounter synchronically in those two languages, however, where /as'kosi/ and /eskon'di/ are found instead. This lexeme (and every other one with the same properties in Latin)[17] has thus been transferred to another class, in this case to the +Rhizotony +PerfectiveStem one in Italian, and to the −Rhizotony −PerfectiveStem one in Spanish. In either case, a morphological trait (stem alternation in Italian, arhizotony in Spanish) has been analogically extended to this verb in a way that the regularly expected syncretism has been avoided with 2SG.PRS /as’kondi/ in Italian and with the 3SG.PRS /es’konde/ in Spanish.

Because competing patterns of preterit formation were readily available in earlier Romance, a suboptimal morphological strategy (+rhizotony −PerfectiveStem) could be locally outcompeted and seems to have disappeared eventually. As expected from the role attributed to morphology in abstractive models, any available morphological trait (in this case the stress and stem differences that were already present in that part of the paradigm) could be exapted to perform this whole-word-form discriminative role. This seems to have led, over time, to the complete disappearance of the dispreferred type.

5 Conclusions

This paper has surveyed the diachronic evolution of two morphological traits (rhizotony and stem alternation) in the former Latin perfectum tenses in Romance. Although deterministic predictive relations between the two traits were absent from Latin, they are found to have emerged in every single contemporary Romance variety. A typology of the different types of predictive relations encountered through the family has been presented next, with one-directional and bidirectional predictabilities on the one hand, and lexical-level and word-form-level predictabilities on the other. The types of systems found (e.g., Spanish, Friulian) are often not unexpected from the original size of the different verb classes in Latin, as smaller classes have generally a higher chance of disappearing from a given variety. These systems are thus compatible with a “random drift” understanding of morphological change in this domain. There is, however, an important exception to this. Whereas, by sheer size of the class in Latin, verbs without rhizotony but with stem allomorphy (0.7%) would have been expected to be the most likely to disappear, Romance languages show that, instead, the much larger class of verbs with rhizotony but without stem alternation (11.5%) is the most vulnerable one diachronically, as it has been universally eradicated.

An explanation for this, I argue, is to be found in the whole-word-form discriminatory role of morphology. Arhizotony and stem alternation can (single handedly or concurrently) distinguish certain preterite and present forms from each other. In their absence, some diagonal syncretisms (i.e., accidental homophonies) would emerge in the paradigm that would be particularly uncomfortable from a processing perspective since they involve not only cells with incompatible feature values (e.g., 3SG.PRS vs. 1SG.PRET), but also with highly asymmetric token frequencies. These are traits that have been found to correlate with homophony-avoidance clashes in other languages. My claim, thus, is that the traits (i.e., arhizotony or a distinct stem alternant) that disambiguate between incompatible word forms would be preferred at the whole-system level (i.e., through the former perfectum like in Spanish), and maybe particularly in the cells that risk becoming syncretic (e.g., 1SG.PRET and 3SG.PRET in Oscos Galician) or in the tense that contains them (see Sassarese Sardinian and Panticosa Aragonese in Table 7). This fits nicely with the developments observed across Romance, as well as with the literature on language processing, which has firmly established the greater cost of semantically unrelated homophones, as well as the faster lexical access to their most frequent meaning. These findings are also aligned with the literature on homophony avoidance at the phonetic and phonological levels, as well as with the discriminative (i.e., not realizational) role attributed to morphology in abstractive models.

Corresponding author: Borja Herce, Department of Comparative Language Science, Distributional Linguistics Lab, University of Zürich, Thurgauerstrasse 30, 8050 Zürich, Switzerland, E-mail:

Funding source: NCCR Evolving Language, NCCR Evolving Language, Swiss National Science Foundation

Award Identifier / Grant number: Agreement Nr. 51NF40_180888


Early versions of this paper were presented at the IVS Research Colloquium at the University of Zürich (23 September 2020), and at Going Romance 34 (27 November 2020). I thank the audiences at those events for stimulating questions and ideas. I also would like to thank Volker Gast and 4 anonymous reviewers of Linguistics for their very insightful feedback that has helped improve this paper substantially. All remaining errors are, of course, mine.

  1. Research funding: This research was made possible by funding from the NCCR Evolving Language, Swiss National Science Foundation (Agreement Nr. 51NF40_180888).


Appendix I The 300 most frequent Latin verbs (LatInFlex1.1) with perfectum rhizotony and stem

Lexeme Frequency PFV_stem PFV_rhizotony Lexeme Frequency PFV_stem PFV_rhizotony
abscondo/-or 27,435 no yes libero/-or 34,598 no no
absoluo 27,656 no yes liceo/-eor/-et 55,291 yes* yes
accedo 33,203 yes yes loco 46,917 no no
accido1 91,696 no yes loquor/-o 124,993 unclassifiable: defective
accido2 91,778 no yes malo 24,620 yes* yes
accipio 152,516 yes yes mando 82,223 no no
addo 37,576 yes yes maneo 50,688 yes yes
aduenio 23,713 yes yes mentior/-io 29,680 no no
aduerto 79,126 no yes meo 47,973 no no
afficio 34,119 yes yes mereo/-eor 159,967 yes* yes
ago 258,211 yes yes metior/-io 73,222 unclassifiable: defective
aio 56,926 no yes miror/-o 23,216 no no
alo 48,703 yes* yes mitto 94,322 yes yes
ambulo 23,140 no no morior/-io 149,538 unclassifiable: defective
amitto 22,530 yes yes moueo 125,130 yes yes
amo 69,097 no no multo 42,018 no no
animo 74,883 no no muto 32,674 no no
aperio 53,929 yes* yes nascor/-o 98,097 unclassifiable: defective
appareo 35,347 yes* yes nego 36,121 no no
appello1 32,333 no no neo 104,981 unclassifiable: no stem vowel
appeto 26,890 no no nescio 41,790 no no
armo 23,174 no no nolo 44,801 yes* yes
ascendo 46,330 no yes nomino 28,807 no no
assumo 31,761 yes yes nosco 97,475 yes yes
audio 151,329 no no noto 44,886 no no
aufero 31,399 yes yes obicio 48,726 yes yes
benedico 31,548 yes yes obseruo 24,016 no no
beo 135,844 no no occi’do 38,415 no yes
bibo 25,288 no yes occi\do" 48,982 no yes
cado 27,803 yes yes occulo 40,108 yes* yes
capio 62,133 yes yes offero 44,278 yes yes
careo 28,085 yes* yes operor/-o 71,748 unclassifiable: defective
caueo 25,076 yes yes oportet 38,530 unclassifiable: defective
causor/-o 67,479 unclassifiable: defective oppono 31,589 yes yes
cieo 24,787 yes yes ordino 57,462 no no
clamo 22,627 no no orior 48,164 unclassifiable: defective
claudo 26,923 yes yes oro 47,290 no no
claudo/-or 25,624 unclassifiable: defective ostendo 101,865 no yes
coepio 36,151 no yes paciscor/-o 23,746 unclassifiable: defective
cogito 37,880 no no pando 27,813 no yes
cognosco 93,916 yes yes pareo 64,102 yes* yes
cogo 25,943 yes yes pario 90,083 yes yes
colligo2 27,621 yes yes paro 39,312 no no
colo 41,196 yes* yes pateo 30,443 yes yes
committo 32,652 yes yes patior/-io 84,853 unclassifiable: defective
compleo 24,257 no no patro 29,878 no no
compono 46,049 yes yes pecco 536,156 no no
concedo 34,186 yes yes percutio 22,460 yes yes
concipio 26,237 yes yes perdo 29,064 yes yes
condo 25,717 yes yes perficio 121,264 yes yes
confero 32,208 yes yes pertineo 43,185 yes* yes
confiteor 35,526 unclassifiable: defective peruenio 33,381 yes yes
congrego 23,462 no no peto 48,941 no no
coniungo 31,292 yes yes placeo 37,287 yes* yes
consentio 27,369 yes yes pono 173,601 yes yes
consequor 71,591 unclassifiable: defective porto 26,453 no no
considero 64,756 no no possideo 24,797 yes yes
constituo 70,443 no yes possum 412,886 yes yes
consto 30,241 yes yes poto 25,733 no no
consulo 28,504 yes* yes praecedo 30,718 yes yes
contemno 30,790 yes yes praecipio 123,165 yes yes
contendo 25,056 no yes praedico1 64,269 no no
contineo 54,088 yes* yes praedico2 79,159 yes yes
contingo1 29,242 unclassifiable: defective praefero 22,763 yes yes
contingo2 37,229 yes yes praemitto 26,982 yes yes
conuenio 66,608 yes yes praesto 25,188 yes yes
conuerro 38,025 no yes praetereo 41,315 no no
conuerto 64,640 no yes priuo 28,217 no no
corrumpo 24,893 yes yes probo 51,322 no no
credo 161,915 yes yes procedo 39,459 yes yes
creo 152,621 no no profero 28,898 yes yes
cupio 23,171 no no proficiscor/-o 55,239 unclassifiable: defective
curo 31,718 no no prohibeo 24,591 yes* yes
curro 30,879 yes yes promitto 46,089 yes yes
custodio 28,115 no no propono 74,854 yes yes
damno 29,314 no no prouideo 23,520 yes yes
debeo 148,061 yes* yes puto 60,055 no no
decerno 30,449 yes yes quaero 103,326 yes no
decresco 26,096 yes yes queo 567,263 unclassifiable: no stem vowel
dedo 29,119 yes yes recipio 54,648 yes yes
deficio 43,035 yes yes reddo 54,395 yes yes
delinquo 40,744 yes yes redeo 25,649 no no
demonstro 28,441 no no refero 43,285 yes yes
descendo 39,849 no yes regno 29,272 no no
desero1 48,297 yes* yes rego 124,013 yes yes
desidero 29,452 no no relinquo 62,425 yes yes
determino 31,576 no no remoueo 28,069 yes yes
dico1 68,813 no no reor/-eo 63,452 unclassifiable: defective
dico2 1,437,559 yes yes requiro 23,220 yes no
differo 44,490 yes yes respicio 31,134 yes yes
diligo 96,239 yes yes respondeo 82,018 no yes
dimitto 34,471 yes yes reuertor/-o 29,466 no yes
dispono 25,409 yes yes saluto 24,994 no no
distinguo 35,426 yes yes sancio 498,809 yes yes
diuerto 121,268 no yes sapio 136,225 no no
diuido 44,072 yes yes scio 177,315 unclassifiable: no stem vowel
diuino 33,191 no no scribo 315,457 yes yes
do 254,389 yes yes secerno 40,646 yes yes
doceo 66,275 yes* yes secundo 41,625 no no
dono 35,796 no no sedeo 40,964 yes yes
duco 50,206 yes yes sentio 117,520 yes yes
efficio 76,190 yes yes separo 37,645 no no
egredior 24,133 unclassifiable: defective sequor/-o 128,850 unclassifiable: defective
eligo 66,504 yes yes sero1 108,765 unclassifiable: no stem vowel
eo 214,966 unclassifiable: no stem vowel serpo 31,484 yes yes
erro 22,590 no no seruio 43,255 no no
excipio 28,312 yes yes seruo 50,640 no no
exeo 31,240 no no significo/-or 67,218 unclassifiable: defective
exerceo 42,113 yes* yes signo 30,417 no no
expono 33,357 yes yes sino 110,705 unclassifiable: no stem vowel
exprimo 22,987 yes yes sisto 40,792 yes yes
facio 705,366 yes yes soleo 45,310 yes* yes
fallo 106,860 yes yes solo 37,932 no no
fero 66,257 yes yes soluo 50,407 no yes
fido 39,195 no yes spero 27,413 no no
finio 49,206 no no spondeo 59,589 yes yes
fio 98,458 unclassifiable: defective statuo 36,507 no yes
for 24,694 unclassifiable: defective sto 79,637 yes yes
formo 40,434 no no subdo 37,300 yes yes
fruor 58,713 unclassifiable: defective subeo 41,456 no no
fugio 39,972 yes yes subicio 102,107 yes yes
gaudeo/-eor 31,582 unclassifiable: defective sufficio 34,354 yes yes
genero 33,973 no no sum 2,706,431 yes yes
geno 51,666 yes* yes sumo 48,122 yes yes
gero 62,628 yes yes suo 290,218 no yes
gigno 32,476 yes yes suscipio 59,261 yes yes
habeo 475,932 yes* yes sustineo 24,053 yes* yes
habito 29,722 no no taceo 25,386 yes* yes
ignoro 41,330 no no tango 28,031 yes yes
impero 24,426 no no tego 26,990 yes yes
impleo 38,903 no no teneo 65,079 yes* yes
incipio 32,322 yes yes timeo 49,430 yes* yes
indo 28,657 yes yes tollo 34,988 yes yes
infero 24,216 yes yes trado 46,498 yes yes
ingredior 27,337 unclassifiable: defective traho 25,292 yes yes
inquam 71,842 no yes transeo 39,720 no no
instituo 38,523 no yes tribuo 29,497 no yes
intellego 139,707 yes yes ualeo 39,188 yes* yes
intendo 31,793 no yes ueneo 69,912 no no
interrogo 23,228 no no uenio 197,463 yes yes
intro 32,643 no no uerto 32,114 no yes
inuenio 107,691 yes yes uideo 351,881 yes yes
iubeo 47,727 yes yes uinco 49,593 yes yes
iudico 63,153 no no uito 48,086 no no
iungo 22,345 yes yes uiuo 119,028 yes yes
laboro 29,729 no no uoco 79,531 no no
lateo 25,331 yes* yes uolo1 173,537 yes* yes
laudo 40,799 no no uoluo 54,272 no yes
lego1 66,052 no no uoueo 54,531 yes yes
lego2 123,473 yes yes utor/-o 94,008 unclassifiable: defective

Appendix II Attested combinations of rhizotony and perfective stem in Romance varieties

Variety Lexemes Comment −Rhizot. −PerfStem +Rhizot. +PerfStem −Rhizot. +PerfStem +Rhizot. −PerfStem Aligned in paradigm
Classical Latin 231 petere dicere quaerere uertere
Catalan Eastern Alguerès 17 no PRET cantare
Catalan Eastern Balear Mallorquí 19 cantare
Catalan Eastern Central Barceloní 19 no PRET cantare
Catalan Eastern Rossellonès 18 no PRET cantare
Catalan Western Lleida 18 no PRET cantare
Catalan Western València 18 cantare
Dalmatian Vegliote Veglia/Krk 40 no PRET cantare
French Francoprovençal Lyonnais Vaux 56 no stress
French Valaisan Val_dIlliez 65 no stress
French Franc-Comtois Pierrecourt 62 no stress
French Lorrain Ranrupt 55 no stress
French Modern Standard 66 no stress
French Norman Jersey 27 no stress
French Norman Guernsey 23 no stress
French Picard Mesnil-Martinsart 49 no stress
French Poitevin-Saintongeais Vendéen 52 no stress
French Wallon Namur 57 no stress
Friulian Friulian 29 coquere
Friulian Western Maniago 23 no PRET clamare
Galego-Portuguese Galician Xermade 15 cantare dicere traces
Galego-Portuguese Galician Lubián 15 cantare dicere no
Galego-Portuguese Galician Fisterra 15 cantare dicere no
Galego-Portuguese Galician Dodro 15 cantare dicere no
Galego-Portuguese Galician Oscos 15 cantare dicere yes
Galego-Portuguese Galician Cualedro 15 cantare dicere no
Galego-Portuguese Portuguese 47 cantare dicere no
Italian Central Laziale Ascrea 51 mandare morire yes
Italian Central Marchigiano Macerata 25 capere mouere morire yes
Italian Central Modern 46 cantare dicere yes
Italian Central Tuscan Corsican 25 cantare dicere yes
Italian Northern Emilian Travo 27 no PRET cantare
Italian Northern Ligurian Genoese 25 no PRET cantare
Italian Northern Piedmontese Montenotte 21 no PRET partire
Italian Northern Piedmontese Basso 34 no PRET posse
Italian Northern2 Veneto Istrioto 20 no PRET dicere
Italian Northern2 Veneto Alpago 18 no PRET dicere
Italian Southern Lucano Archaic 26 coquere dicere habere yes
Italian Southern Lucano Papasidero 17 cantare facere yes
Italian Southern Lucano Calvello 22 fatigare dicere exire yes
Italian Southern Molisano Casacalenda 35 no PRET dicere
Italian Southern2 Sicilian Central 18 cantare facere yes
Ladin-Dolomitic Atesino Val_Badia 49 no PRET dicere
Occitan Northern Auvergnat Gartempe 44 no stress
Occitan Northern Limousin St-Augustin 27 amare placere
Occitan Northern Vivaro-Alpin Seyne 39 mittere currere
Occitan Southern Languedocien Graulhet 34 parabolare parescere
Occitan Southern Provençal Nice 46 cantare
Romanian Aromanian 48 dormire coquere stare no
Romanian Istro-Romanian Šušnjevica 32 no PRET posse coquere
Romanian Megleno-Romanian 37 cantare fervere no
Romanian Modern Standard 58 dormire dicere videre no
Spanish Aragonese Panticosa 24 coquere facere quaerere no
Spanish Aragonese Ansotano Ansotano 19 quaerere facere
Spanish Astur-Leonese Asturian Somiedo 16 comedere dicere traces
Spanish Astur-Leonese Asturian Parres 22 subire tenere no
Spanish Modern Standard 43 cantare tenere no
Romansh Puter Engadine 64 coquere
Romansh Surmiran Bivio-Stalla 22 no PRET sapere
Romansh Surselvan 103 no PRET cantare

Appendix II shows the explored morphological trait combinations in the Romance varieties in the Oxford Online Database of Romance Verb Morphology (Maiden et al. 2010). 10 varieties with less than 15 lexemes documented have been excluded, as have been 4 varieties where the relevant forms (i.e., perfectum tenses) were insufficiently documented: Ladin-Dolomitic Atesino Val Gardena, Italian Northern Emilian Romagnol, Italian Central Marchigiano Servigliano, and Sardinian Nuorese. The remaining varieties are shown in the table.

When a variety has lost stress as a morphological property of the word, this is reported in the ‘Comment’ column and no more information is provided about that variety, as the property of rhizotony cannot logically apply in the absence of word-stress. Also reported in that column is the loss of the preterite tense in many varieties. Since this is the central tense of the perfectum (see Figure 2) and the one where rhizotonic forms should be expected (see Table 8), varieties that lose this tense typically lose both rhizotony and the perfectum stem.

Columns 4 to 7 show which of the trait combinations that are explored in the paper are present in a given variety. Existing combinations are indicated by providing the Latin lemma whose descendant has those properties in the daughter language. In the last column, only for those varieties that have preserved both rhizotony and perfectum stems, the information is given of whether the two properties have become aligned at the word-form level or not (see the subtypes identified in Section 3.3.2).


Albright, Adam. 2010. Base-driven leveling in Yiddish verb paradigms. Natural Language & Linguistic Theory 28(3). 475–537. in Google Scholar

Anderson, Stephen R. 2011. Stress-conditioned allomorphy in Surmiran (Rumantsch). In Maria Goldbach, Marc-Olivier Hinzelin, Martin Maiden & John Charles Smith (eds.), Morphological autonomy: Perspectives from Romance inflectional morphology, 13–35. Oxford: Oxford University Press.10.1093/acprof:oso/9780199589982.003.0002Search in Google Scholar

Aronoff, Mark. 1994. Morphology by itself: Stems and inflectional classes. Cambridge, MA: MIT press.Search in Google Scholar

Baerman, Matthew. 2011. Defectiveness and homophony avoidance. Journal of Linguistics 47(1). 1–29. in Google Scholar

Bazzoni, Gian Paolo. 1999. Elementi di grammatica sassarese. Sassari: SIG.Search in Google Scholar

Blevins, James P. 2016. Word and paradigm morphology. Oxford: Oxford University Press.10.1093/acprof:oso/9780199593545.001.0001Search in Google Scholar

Blevins, JamesP., Ackerman Farrell & Malouf Robert. 2016. Morphology as an adaptive discriminative system. In Daniel Siddiqi & Harley Heidi (eds.), Morphological metatheory, 271–302. Amsterdam & Philadelphia: John Benjamins.10.1075/la.229.10bleSearch in Google Scholar

Bonfante, Giuliano. 1941. The Latin and Romance weak perfect. Language 17. 201–211. in Google Scholar

Bubenik, Vit. 1997. From Ancient to Modern Greek. In John Hewson & Vit Bubenik (eds.), Tense and aspect in Indo-European languages: Theory, typology, diachrony, 249–264. Amsterdam & Philadelphia: John Benjamins.10.1075/cilt.145.20bubSearch in Google Scholar

Bubenik, Vit. 1979. Historical development of the Ancient Greek accent system. Indogermanische Forschungen 84. 90–106.Search in Google Scholar

Bybee, Joan L. 1988. Morphology as lexical organization. In Michael Hammond & Michael Noonan (eds.), Theoretical morphology, 119–142. San Diego, CA: Academic Press.Search in Google Scholar

Bybee, Joan L. & Mary Alexandra Brewer. 1980. Explanation in morphophonemics: Changes in Provençal and Spanish preterite forms. Lingua 52(3–4). 201–242. in Google Scholar

Caragiu-Marioţeanu, Matilda & Nicolae Saramandu. 2005. Manual de aromână. Carti trâ înviţari armâneaşti. Bucarest: Editura Academiei Române.Search in Google Scholar

Esher, Louise. 2015. Morphomes and predictability in the history of Romance perfects. Diachronica 32(4). 494–529. in Google Scholar

Gioscio, Joseph. 1985. Il dialetto lucano di Calvello. Stuttgart: Steiner.Search in Google Scholar

Hall, Kathleen Currie, Elizabeth Hume, T. Florian Jaeger & Andrew Wedel. 2018. The role of predictability in shaping phonological patterns. Linguistics Vanguard 4(s2). in Google Scholar

Herce, Borja. 2019. Morphome interactions. Morphology 29(1). 109–132. in Google Scholar

Herce, Borja. 2020a. Alignment of forms in Spanish verbal inflection: The gang poner, tener, venir, salir, valer as a window into the nature of paradigmatic analogy and predictability. Morphology 30(2). 91–115. in Google Scholar

Herce, Borja. 2020b. A typological approach to the morphome. Vitoria & Guildford: University of the Basque Country and University of Surrey Dissertation.Search in Google Scholar

Hewson, John. 2001. Aspect and tense from PIE to Germanic: The systemic evolution. In Sheila Watts, Jonathan West & Hans-Joachim Solms (eds.), Zur Verbmorphologie germanischer Sprachen. Tübingen: Max Niemeyer.10.1515/9783110918656.73Search in Google Scholar

Klepousniotou, Ekaterini. 2002. The processing of lexical ambiguity: Homonymy and polysemy in the mental lexicon. Brain and Language 81(1–3). 205–223. in Google Scholar

Lief, Eric. 2004. Proto-Romance *[w] and the Velar Preterites. In Julie J. Auger, Clancy Clements & Barbara Vance (eds.), Contemporary approaches to Romance linguistics: Selected papers from the 33rd Linguistic Symposium on Romance Languages (LSRL), 257–274. Amsterdam & Philadelphia: John Benjamins.10.1075/cilt.258.14lieSearch in Google Scholar

Lloyd, Paul. 1987. From Latin to Spanish: Historical phonology and morphology of the Spanish language, 173. Americal Philosophical Society.Search in Google Scholar

MacGregor, Lucy J., Jennifer Bouwsema & Ekaterini Klepousniotou. 2015. Sustained meaning activation for polysemous but not homonymous words: Evidence from EEG. Neuropsychologia 68. 126–138. in Google Scholar

Maiden, Martin. 1992. Irregularity as a determinant of morphological change. Journal of Linguistics 28(2). 285–312. in Google Scholar

Maiden, Martin. 2005. Morphological autonomy and diachrony. In Geert Booij & Jaap van Marle (eds.), Yearbook of morphology 2004, 137–175. Dordrecht: Springer.10.1007/1-4020-2900-4_6Search in Google Scholar

Maiden, Martin. 2011. Morphological persistence. In Martin Maiden, John Charles Smith & Adam Ledgeway (eds.), Cambridge history of the romance languages, 155–215. Cambridge: Cambridge University Press.10.1017/CHOL9780521800723.006Search in Google Scholar

Maiden, Martin. 2012. A paradox? The morphological history of the Romance present. In Sascha Gaglia & Marc-Olivier Hinzelin (eds.), Inflection and word formation in Romance languages (Linguistik Aktuell/Linguistics Today 186), 27–54. Amsterdam & Philadelphia: John Benjamins.10.1075/la.186.02maiSearch in Google Scholar

Maiden, Martin. 2018a. The Romance verb: Morphomic structure and diachrony. Oxford: Oxford University Press.10.1093/oso/9780199660216.001.0001Search in Google Scholar

Maiden, Martin. 2018b. New thoughts on an old puzzle: The Italian alternation type dissi, dicesti; feci, facesti. Revue Romane 53(2). 217–260. in Google Scholar

Maiden, Martin, John Charles Smith, Silvio Cruschina, Marc-Olivier Hinzelin & Maria Goldbach. 2010. Oxford online database of Romance verb morphology. Available at: in Google Scholar

Malkiel, Yakov. 1966. Diphthongization, monophthongization, metaphony: Studies in their interaction in the paradigm of the Old Spanish-ir verbs. Language 42(2). 430–472. in Google Scholar

Nagore Lain, Francho. 1986. El aragonés de Panticosa: gramática. Huesca: Instituto de Estudios Altoaragoneses.Search in Google Scholar

O’Neill, Paul. 2014. The morphome in constructive and abstractive models of morphology. Morphology 24(1). 25–70.10.1007/s11525-014-9232-1Search in Google Scholar

O’Neill, Paul. 2018. Velar allomorphy in Ibero-Romance: Roots, endings and clashes of morphomes. In Miriam Bouzouita, Ioanna Sitaridou & Enrique Pato (eds.), Studies in historical Ibero-Romance morpho-syntax: A descriptive and prescriptive analysis, 13–46. Amsterdam & Philadelphia: John Benjamins.10.1075/ihll.16.02oneSearch in Google Scholar

Pato, Enrique & Paul O’Neill. 2013. Los gerundios analógicos en la historia del español (e iberorromance). Nueva Revista de Filología Hispánica 61(1). 1–27. in Google Scholar

Pellegrini, Matteo & Marco Passarotti. 2018. LatInfLexi: An inflected lexicon of Latin verbs. In Cabrio Elena, Alessandro Mazzei & Fabio Tamburini (eds.), Proceedings of the Fifth Italian Conference on Computational Linguistics (CLiC-it 2018). in Google Scholar

Penny, Ralph. 1991. A history of the Spanish language. Cambridge: Cambridge University Press.Search in Google Scholar

Pittau, Massimo. 1972. Grammatica del sardo-nuorese, il più conservativo dei parlari neolatini. Bologna: Pàtron.Search in Google Scholar

Scheitlin, Walter. 1980. Il pled puter: Grammatica ladina d’Engiadin’ota cun lecziuns e register da pleds / Grammatik in Puter (Oberengadin) mit Lektionen und Wörterverzeichnis. Samedan: Uniun dals Grischs.Search in Google Scholar

Smith, Kenny & Elizabeth Wonnacott. 2010. Eliminating unpredictable variation through iterated learning. Cognition 116(3). 444–449. in Google Scholar

Spencer, Andrew. 2012. Identifying stems. Word Structure 5(1). 88–108. in Google Scholar

Stump, Gregory & Raphael A. Finkel. 2013. Morphological typology: From word to paradigm, vol. 138. Cambridge: Cambridge University Press.10.1017/CBO9781139248860Search in Google Scholar

Trommer, Jochen & Sebastian Bank. 2017. Inflectional learning as local optimization. Morphology 27(3). 383–422. in Google Scholar

Wedel, Andrew, Scott Jackson & Abby Kaplan. 2013. Functional load and the lexicon: Evidence that syntactic category and frequency relationships in minimal lemma pairs predict the loss of phoneme contrasts in language change. Language and Speech 56(3). 395–417. in Google Scholar

Wheeler, Max W. 2012. Vies d’analogia i d’explicació en l’evolució del pretèrit feble de la conjugació -e romànica. Estudis Romànics 34. 7–36. in Google Scholar

Wiemer, Björn & Ilja A. Seržant. 2017. Diachrony and typology of Slavic aspect: What does morphology tell us. In Walter Bisang & Andrej Malchukov (eds.), Unity and diversity in grammaticalization scenarios, 239–307. Berlin: Language Science Press.Search in Google Scholar

Williams, Edwin B. 1930. The Portuguese and Spanish preterit. Modern Philology 27. 297–302. in Google Scholar

Zof, Fausto. 2000. Gramatiche pratiche de lenghe furlane. Pasian di Prato: Leonardo.Search in Google Scholar

Zörner, Lotte. 1997. Il pagotto: Dialetto dell’Alpago: descrizione fonologica, storico-fonetica e morfologica. Padova: Unipress.Search in Google Scholar

Received: 2020-03-09
Accepted: 2021-06-25
Published Online: 2021-09-30
Published in Print: 2022-07-26

© 2021 Borja Herce, published by De Gruyter, Berlin/Boston

This work is licensed under the Creative Commons Attribution 4.0 International License.

Scroll Up Arrow