2.1 From Latin to Romance

For reasons of space, we cannot discuss all aspects of the evolution of the Romance CC systems. What should become clear is the fact that Spanish, Catalan and Italian (and other Romance languages) have reduced the number of Latin CCs, maintaining at the same time without any doubt ThVs in their verbal inflection. In Spanish, for instance, Latin ThVs -ā- / -ē- / -ī- developed, as a rule, to Spanish -a- / -e- / -i- respectively, with some exceptions for -ē- and -ī- (cf. Figure 1 where the dotted lines illustrate the exceptions and the solid lines the rule). In contrast, the Latin consonantal conjugation, i.e. the one we call athematic, as well as the mixed conjugation, got lost and the corresponding verbs (if they survived at all) were integrated – again with minor exceptions – into the 2^nd CC of Spanish. This means that, in Spanish, all CCs are thematic CCs, even though we find athematic forms due to phonological reasons (e.g. canto ‘I sing’ vs. *cantao). Suppletive verbs like e.g. ir ‘to go’ may be an exception to this generalization (cf. Pomino & Remberger 2022a/b).

Figure 1

Reduction of the CC system from Latin to Spanish.

More precisely, the Spanish CCs can be identified by the infinitival form, which has either of the ThVs -a-, -e- and -i-. There is, however, also a number of inflected forms which surface without a ThV: the 1^st person singular present indicative (e.g. cant-o ‘I sing’, beb-o ‘I drink’ and part-o ‘I leave’), all persons of the present subjunctive (e.g. cant-e, beb-a and part-a)⁸, the 1^st person singular indefinido (i.e. the simple past as derived from the old Latin perfect, e.g. cant-é ‘I sang’, beb-í ‘I drank’ and part-í ‘I left’) and the 3^rd person singular indefinido of the first CC (e.g. cant-ó ‘she sang’). Even though Spanish has athematic forms, the CC system is nevertheless clearly defined by ThVs.

Another interesting aspect of the Spanish CC system (valid in similar vein for Catalan and other Romance languages) is the fact that in many tenses the 2^nd and 3^rd CC are neutralized, that is they both either show -e- (or rather -ie-) or -i-, e.g. beb-e-n ‘they drink’ / part-e-n ‘they leave’, beb-ie-ron ‘they drank_indefinido’/ part-ie-ron ‘they left_indefinido’, beb-í-a-n ‘they drank_imperfect’/ part-í-a-n ‘they left_imperfect’. In other words, only in 16 out of 59 synthetic forms that a Spanish (non-defective) verb has the two CCs show different ThVs. In all other cases, they are not distinguishable by the ThVs. This is completely opposed to the 1^st CC, in which all thematic forms have the ThV a or, put differently, which is never neutralized with respect to another CC. What is more, the 1^st CC stands out also due to its productivity, the very small number of irregular verbs (out of 9986 verbs only 333 are considered to be irregular, cf. Alcoba 1999: 4936)⁹ and the number of verbs it contains – according to Alcoba (1999: 4936), 90.75% (i.e. 9986 verbs) belong to the 1^st CC, 4.75% to the 2^nd and 4.6% to the 3^rd CC.

These observations lead us to the assumption that even though all CCs are thematic in Spanish and other Romance languages, the CCs differ with respect to their markedness. In all these languages the respective CCs are not equally productive, not equally regular and they also have different dimensions as far as the number of members is concerned. What is more, CC oppositions may be neutralized. Oltra-Massuet (1999; 2020) argues based on this line of reasoning that ThVs are bundles of subatomic abstract features organized in a markedness hierarchy or feature geometry, cf. Figure 2:

Figure 2

Conjugation Classes (regular verbs): Spanish (a), Catalan (b) (Oltra Massuet 1999; 2020: 287) and Italian (c) (based on Napoli & Vogel 1990).

In French, ThVs, in contrast, have developed in a different way (cf. Table 2) and, up to now, there is no consensus in the literature with respect to whether Modern French has ThVs and how the CCs are to be described.

Modern French verbs are traditionally divided into three CCs (or four, if verbs ending in -oir are not included in the 3^rd CC on -re) based on the form of the infinitive (cf. e.g. Meyer-Lübke 1908: 202): Verbs of the 1^st CC have an infinitival form ending in -er [-e^ʁ], e.g. aimer [ɛme] ‘(to) love’. This CC does not only include 90% of all French verbs, but it is also highly productive. Apart from aller ‘to go’, which shows suppletion and therefore doesn’t belong to the 1^st CC since it unites forms going back to verbs of different CCs, most verbs ending in -er are classified as being regular, in the sense that they have only one single stem for all tense forms. Only a few verbs have stem allomorphy, but – as shown e.g. by Meunier & Marslen-Wilson (2004) – these all count as phonologically predictable changes. The stem allomorphy found in lever [lǝ.ve^ʁ] ‘to lift’ vs. lève [lɛv] ‘(I) lift’, for instance, is explainable based on the distinction between open and closed syllables. In sum, due to the mentioned characteristics, the 1^st CC is the unmarked CC. The 2^nd CC is characterized by infinitives ending in -ir [-iʁ] and the verbs of this class have a short and a long or extended stem, e.g. finis [fini] ‘(I) finish’ vs. finissons [finisɔ̃n] ‘(we) finish’. This CC is also considered to be fully regular; according to Gertner (1973: 19) the number of regular verbs within this CC is even higher than the one of the 1^st CC. However, the 2^nd CC entails only 2,8% of all French verbs and it is generally – with a few exceptions, including some recent newcomers (cf. §3.2) – considered to be no longer productive, a fact that makes it more marked than the 1^st CC. The 3^rd CC is in essence a smorgasbord of all other verbs that do not belong to the 1^st or the 2^nd CC. Verbs belonging to this class are, for instance, romp-re ‘to break’, pein-dre ‘to paint’, voul-oir ‘to want’, dorm-ir ‘to sleep’ (without -ss-extension) and also all-er ‘to go’. This CC contains many irregular verbs with stem allomorphy and also several suppletive verbs (where the whole inflected form is affected by irregularity, not only the root/stem). It goes without saying that this CC is unproductive and counts as the most marked class.¹⁰

However, especially for French there are many other proposals for the morphological classification into different verb classes. For instance, Dubois (1967) turns away from the diachronically motivated classification along the form of the infinitive and proposes instead a classification based on the number of stems a verb allows: one stem (e.g. chanter ‘to sing’: chant-), two stems (e.g. écrire ‘to write’: écri-, écriv-), three stems (e.g. plaindre ‘to complain’: plain-, plaign-, plaind-), four stems (e.g. connaître ‘to know’: connai-, connaiss-, connaît-, conn(u)-), five stems (e.g. vouloir ‘to want’: veu-, voul-, veul-, voud-, veuill-) and six stems (e.g. pouvoir ‘can’: peu-, pouv-, peuv-, puiss-, pour-, p(u)-). Other researchers try to reduce the number of different stems to an absolute minimum and to give a rather rule-based explanation. El Fenne (2020), for instance, does not assume two stems for écrire (écri- and écriv-) as Dubois (1967) does, but rather one single stem that has a floating consonant (i.e. écri^v-). Floating consonants are overtly realized only if they can occupy the onset position of a following syllable (cf. é.cri.vons vs. é.crit) (cf. §2.2 for more details).

Now, when looking at the infinitival forms of French verbs, there are different proposals in the literature for possible segmentations: The “outcome” of the former Latin ThVs is either segmented as belonging to the verb ending (cf. (3a)), or it is added to the root to form a thematic stem (i.e. root + ThV; cf. (3b)), or it is seen as part of a non-segmented stem (cf. (3c)). In two of these three possibilities the vowel is no longer conceived as ThV (cf. (3a,c)), whereas in the remaining third case (3b) the notion of ThV is even expanded by including also theme consonants and/or theme diphthongs (cf. Table 6 further below).

In sum, some linguists follow the rather traditional classification and consider -r to be the infinitival ending (cf. (3b,c)), whereas the preceding vowel is classified as theme vowel/theme diphthong or seen as part of a non-segmented stem. Others propose instead to analyse -er, -ir, -oir and other endings as purely infinitival, i. e. inflectional. Note, however, that in some proposals the segmentation chosen is not the same for all verb types, but may differ from verb type to verb type (cf. Table 3).¹¹

As (3) and Table 3 show, the assumption of ThVs for French and thereby the classification of the CC system along ThVs has been put into question, not only for the infinitival forms, but also for many finite forms since they are, at least superficially, athematic (cf. Table 4). One question to be answered, however, is whether in the case of French we are dealing with athematic forms or completely athematic conjugations. As illustrated above, also in Latin and Spanish thematic CCs have athematic forms.

In the following section we will discuss one proposal with and another one without the consideration of ThVs in order to find out which challenges each approach is confronted with.

2.2 Approaches with and without theme vowels for French verbal inflection

Let us start with approaches where the existence of ThVs is denied and the vowel present in the infinitival form is instead assigned to a non-segmentable stem and/or to the ending (cf. possibilities (3a,c) above). Among the various proposals based on this idea we will discuss the work of El Fenne (1994, 2020). A basic argument for El Fenne’s classification for French CCs is the contrasting behavior with respect to the consonant-zero-alternation found in different roots with apparently identical phonological environments. The root final consonant of viv(re) ‘to live’, e.g., is maintained if there is a possibility for it to appear in a syllable onset, e.g. before a vowel (e.g. nous vivons [vi.vɔ̃(z)] ‘we live’) or a consonant with which it can build a complex onset (e.g. nous vivrons [vi.vʁɔ̃(z)] ‘we will live’), but it is deleted before a following consonant with which it cannot form an onset (tu vis [vi-(z)] ‘you_sg live’). However, the same final consonant of the verb arriv(er) ‘to arrive’, in contrast, is never deleted: tu arrives [aʁiv-(z)] ‘you_sg arrive’, and not *tu arris [aʁi-(z)]. Instead of assuming two different stems for vivre (i.e. viv- and vi-) and one for arriver (i.e. arriv-), El Fenne (2020) proposes that French verbs must be classified based on whether the verb stem ends in a permanent consonant (PC) or in a floating consonant (FC). Stem final PCs are never affected by consonant deletion, i.e. irrespective of the following element, they are phonetically realized, while stem final FCs (annotated as superscript consonants) are realized only if they can be linked with the onset position of a following element: Thus, the stem viv- is represented in the lexicon as having a FC. In case of nous vivons, the FC is linked on the skeletal tier to the open x-slot of the affix -ons, while it will not surface in case of tu vis, since there is no following x-slot available. In contrast, the final consonant of the stem arriv- will always be realized since it has a PC, i.e. it is already associated with its own x-slot in its lexical entry. In case of the verbal form nous arrivons, there will be only “relinking” from the coda position into the onset position. In this analysis, both verbs have only one stem. The realization of the final consonant depends on whether it has a FC or a PC, i.e. the phonological process is triggered by this information alone and does not affect all final consonants, cf. Figure 3. We would like to refer the reader to El Fenne (2020: 35) for a table which contains an inventory of floating consonants of French verbal stems. El Fenne’s inventory of the floating consonants does not exactly correspond to the latent consonants involved in the (word-external) sandhi phenomenon of liason.

Figure 3

Consonant-zero-alternation without ThV.

Now how does the distinction between FC and PC correlate to the CCs? There is no one-to-one correlation between CCs and PC or FC, but a clear tendency can be observed. According to Paradis & El Fenne (1995: 198–199), 94% of the French verbs (11.294 verbs; cf. Bescherelle 1985; 2019 for concrete numbers) belong to the 1^st CC. Out of these, 4% have a stem ending with a vowel and in the remaining 90% the stems end in a consonant that is a PC. There is, however, a small number of verbs (0,3%, i.e. 30 verbs) belonging traditionally to the 3^rd CC that also have a PC ([ʁ] or [j], e.g. courir ‘run’). All other verbs, whether they belong to the 2^nd or 3^rd CC, have FCs. Leaving minor exceptions aside, we can summarize this correlation as given in Table 5.

Table 5

Twofold classification based on the verb stem (El Fenne 1994, 2020).

However, El Fenne’s classification with respect to the verb stems is problematic with respect to the number of the infinitival endings: Although she assumes only two types of stem, she has to posit six different endings for the infinitive (as Table 5 shows). Yet, from her analysis, it is not clear how the stems select their corresponding endings. Probably she has to assume special (readjustment) rules that would be extremely complex and unmotivated. What is more, also for finite forms, El Fenne (2020) proposes a segmentation into only two constituents. For this, she posits further verbal stems for the future tense and the conditional and cumulative inflectional endings for tense and agreement (e.g. -ions [jɔ̃] for the 1^st person plural imperfect indicative), which she terms a general affix (GA). Figure 4 contrasts the present tense form for the 1^st person plural with the corresponding future form, the conditional, and the imperfect form as they are analysed by El Fenne. Due to the rather poor segmentation of French verbal forms into only two parts, El Fenne (2020) has to posit specific stems for the future (and conditional) and a specific GA for the imperfect (and conditional), even for the completely regular 1^st CC which is characterized by having no stem (or root) allomorphy (apart from regular phonological changes). Both her assumptions are highly disputable: El Fenne (2020) argues that the future and the conditional are built on the basis of the infinitive form. Yet, the infinitive aimer is [ɛme^ʁ], not [ɛmə^ʁ].¹² That is, she must either explain the vowel reduction or assume a specific stem for the future (and conditional). A more serious problem is, in our opinion, the GA [jɔ̃] for the imperfect (and conditional) form. Except for the passé simple – a tense which has many idiosyncrasies in all Romance languages – the 1^st person plural is always realized as [ɔ̃] in French – so why should we assume such a GA and dispense with an obvious regularity. One could also associate [j] with the stem, but even this would again lead to a higher number of stems and an unnecessary complication of the inflectional system.

Figure 4

Comparison of different tense forms based on El Fenne (2020).

In sum, in El Fenne’s approach, the reduction of the structural complexity to only two constituents goes hand in hand with an increase of allomorphy. We also think that, if there is something really regular within French inflection, it is the person-number suffix [ɔ̃] for the 1^st person plural (and [e(z)] for the 2^nd person plural). The segmentation proposed by El Fenne would interpret this regular element as an irregular element.

Let us now turn to an analysis that posit ThVs also for French. As we have already seen, there is no consensus at all in the literature with respect to the shape of French infinitival suffixes. There are several proposals and researchers count with up to six different infinitival endings. It goes without saying that, depending on which segmentation one is willing to accept, we are faced with different problems to be solved. Those approaches who assume that the ThV is part of the stem have proposed a binary opposition for the infinitive ending (e.g. Martinet 1969; Van den Eynde & Blanche-Benveniste 1970; Pinchon & Coute 1981; Plénat 1981). For them, the infinitive of the 1^st CC ends in [-e] (or rather [-e^ʁ])¹³, whereas in all other cases it is marked by [-ʁ]. In these approaches, the verb stem (highlighted in grey in Table 6) is generally further segmented into √root and ThV. Standardly, verbs of the 1^st, the 2^nd and some of the 3^rd CC count as being thematic in many traditional works. Other 3^rd CC verbs (e.g. dire, vivre) are analyzed as athematic. Finally, some researchers also argue for the presence of theme consonants and theme diphthongs, as given in Table 6. The challenge for those who maintain this segmentation lies in explaining, among other things, how which theme element is combined with which √root.¹⁴

Table 6

Classification based on the infinitive ending (IS = inflectional suffix).¹⁵

Evidence for assuming ThVs in French for the 1^st CC comes from the consonant-zero-alternation mentioned before and again illustrated in Figure 5: As said before, the root final consonant of viv(re) ‘to live’, which in these approaches belongs to an athematic CC, is maintained, if there is a possibility for it to appear in a syllable onset, otherwise it is deleted. The same final consonant of arriv(er) ‘to arrive’, which is thematic, is never deleted. Schane (1966; 1968) and others assumes that in this case the ThV /ǝ/, which does not surface, blocks consonant deletion (cf. /aʁiv- (ǝ)+(z)/).¹⁶ In this case, both verbs have just one root/stem (/viv-/ and /aʁiv-/) and the realization of the final consonant depends on the following element.¹⁷

Figure 5

Consonant-zero-alternation with ThV.

The assumption of the presence of a ThV for the CC of finir ‘to finish’ is straightforward for many linguists, since here the ThV surfaces as either [i] or [is].¹⁸ With respect to the [i]-[is]-alternation, Schwarze (2009) assumes that the underlying form of the ThV is, in all cases, /is/. The /s/ of this theme element surfaces only when it can occupy an onset position in the syllable structure (cf. the plural forms of finir in Table 7), otherwise it is deleted. As can be seen from this example, it is sufficient to assume only one root/stem and one ThV; the corresponding surface forms result from regular phonological processes.

There is thus evidence for assuming ThVs for the CC of type aimer and the CC of the type finir. However, things are different for verbs belonging to the remaining verbs since these do not form a class but are completely heterogeneous with respect to potential ThVs, as the examples c.-f. in Table 6 show. If we were to analyse these forms as containing ThVs, we would either be forced to build subclasses and motivate these subclasses, or again to assume special (readjustment) rules and motivate these. In other words, while the combination of the root and the ThV is straightforward for verbs of the type of aimer and finir, it is not clear how this can be done for the remaining verbs unless one is willing to propose unmotivated diacritic features and/or (readjustment) rules.

To sum up, both approaches discussed in this section have, in our opinion, the same basic problem: Approaches like El Fenne (1994; 2020) try to keep the stems as regular as possible, which means that they have to assume various allomorphic suffixes for the same verbal form (e.g. the infinitive). The other approaches strive for the minimal allomorphy of inflectional endings (e.g. only -e and -r for the infinitive), but then have to shift irregularity to the stem (cf. Figure 6).

Figure 6

Problems for approaches with and without thematic elements.

The framed elements in Figure 6 are the ones that contain the irregularity and in both general approaches it remains unclear how the correct combination of stems and suffixes is achieved: What prevents [vu^l-] to be combined with [-iʁ], for example? We instead treat ThVs as autonomous elements that are situated between a stem and the inflectional ending. We also believe that the infinitival forms in French are exceptional (since some of them, e.g. partir, are neutralized in some forms of the paradigm, cf. §3.2 below) and do not give us a clear picture of ThVs. In our analysis, we will follow a mixed approach, considering ThVs relevant, but only for some CCs. Some of the arguments given in the literature will be used in our analysis, which is mainly based on the assumption that also in French verbs can be segmented into root, ThV, and ending.

In the approaches presented above, there are, however, important differences with respect to the segmentation of the verbal forms. In more traditional approaches, the stem is further segmented into root and ThV (or Th-diphthong or ThC), whereas it is one single unit in the proposal by El Fenne (2020). In her proposal, what we might analyze as ThV, is either part of the stem (cf. fini-r) or part of the ending (cf. dorm-ir). The question is whether there is any theory-external evidence for a fine-grained segmentation for French verbal forms or whether they are only segmentable in a stem and an ending. To pursue this question, we will make a small digression into psycholinguistic studies at the beginning of the next section.

3.1 Psycholinguistics evidence for the segmentation of French verbs

As shown in Section 2, there is no consensus in the literature with respect to the segmentation of French verbs. A verb like partir ‘to leave’ could be segmented e.g. in the following three ways: part-i-r (with autonomous ThV), parti-r (where the vowel is part of a no further segmented stem) or part-ir (where the vowel is part of the verbal ending). In this subsection we briefly discuss psycholinguistic studies concerning the mental representation and processing of verbal stems and conjugation classes in French in order to find out whether these provide evidence for one of the three possibilities or whether we can, at least, exclude one of the possibilities.

Estivalet & Meunier (2015a/b; 2016) conducted two priming experiments – a cross-modal priming experiment (CMPE) and a masked priming experiment (MPE) – to find out how French verbs are processed. The basic assumption behind both experiments is that the response with respect to the target is (positively or negatively) influenced by the prime. The authors considered two conditions with different priming predictions and an additional test condition: (a) In the identity condition, prime and target are identical (e.g. parlons ~ parlons) and full priming is predicted. (b) No priming is predicted in the control condition where the prime is the infinitive form of one verb, while the target is an inflected form of another verb (e.g. aimer ~ parlons). (c) In the test condition, the prime and the target belong to the same verb, but the prime is given in the infinitive and the target is an inflected form (e.g. parler ~ parlons). With respect to this test condition, Estivalet & Meunier (2015a/b; 2016) posited the following three hypotheses (RT = Reaction Time).

The experiments had the following results: For verbs of the type lever (with phonologically conditioned allomorphy, cf. je lève), parler and dormir, vendre, Estivalet & Meunier (2015a/b; 2016) found a significant difference between the control and test conditions, but no difference between the test and identity conditions. That is, the targets are fully primed. In contrast, verbs like écrire and joindre show only partial priming.²⁰ According to the authors, these findings suggest that French parler, dormir etc. have a root-based inflection, while écrire and joindre have a stem-based inflection. In other words, full priming in parler ~ parlons is only possible if the verbal forms are fully decomposed (cf. the left side in Figure 7), since only then we have identical root segments that leads to full priming. In contrast, in case of stem-based inflection (cf. the right side in Figure 7) there are similar, but not identical “stem/root” elements and thus only partial priming.²¹

Figure 7

Full vs. partial priming (adapted from Estivalet & Meunier 2014; 2015a/b).

We take the argumentation put forward in Estivalet & Meunier (2015a/b; 2016) as a clear argument against all those proposals where the (theme) vowel is analyzed as a non-segmentable part of the verbal stem (e.g. dormi- or fini^s-) (cf. possibility illustrated in (3c)), because full priming in prime-target pairs like partir ~ partons, dormir ~ dormons would otherwise remain unexplained.

However, it is possible to raise some objections with respect to another claim put forward by the authors. According to Estivalet & Meunier (2015a/b; 2016) stems that contain a ThV are fully decomposed into a root and the ThV. However, the priming effects observed by Estivalet & Meunier (2015a/b; 2016) can also arise if prime and target are decomposed in other ways. In our opinion, both options of segmentation in Figure 8 would predict full priming. Yet, in the segmentation given at the right-hand side, there is no separate ThV, but this element is rather part of the infinitival ending.

Figure 8

Full vs. partial priming (alternative possibility).

The argument for ThVs in Estivalet & Meunier (2015a/b; 2016) is not as clear as we hoped it would be. Only a segmentation like the one given in (3c) seems not to be supported by their experiments. In the next subsection we will therefore discuss another phenomenon, the CC neutralization, that could give us more insights.

3.2 A feature geometry for French conjugation classes

Let us start the discussion by having a look at the French CC opposition and neutralizations exemplified in Figure 9 (for ease of simplicity our proposal is restricted to four verb types, i.e. the type aimer, the type finir, the type partir and the type vendre; the table is divided in two parts, one representing the orthographic version of the verb forms, cf. Figure 9a, the other one the phonetic transcription, including the segmentation proposed by us, cf. Figure 9b; the shaded cells highlight the distribution of the ThV /i(s)/). We argue that aimer and finir represent different thematic CCs with different ThVs, whereas partir and vendre belong to two different athematic CCs that behave differently with respect to neutralization of the CCs. As can be seen, in the future and the conditional the ThV -i- of the type finir is also found in the CC of the type partir, whereas the CC of the type aimer (usually called the 1^st CC) maintains its ThV /ə/ and the CC of the type vendre has no vocalic element between stem and ending in these tenses. Moreover, in the passé simple and in the past subjunctive, the ThV -i- of the CC of the type finir is not only found in the CC of the type partir, but also in the type vendre, whereas the CC of the type aimer has another ThV. Therefore, we can say that the ThV of the type finir spreads over other CCs in some tenses, leading to a neutralization of CC oppositions between the CCs of type finir, partir and vendre. In our way of reasoning, the neutralization process which consist in the spreading of one ThV into other CCs clearly shows that ThVs are autonomous elements and not non-segmentable parts of a stem or an ending.

Figure 9a

CC oppositions and neutralizations.²²

Figure 9b

CC oppositions and neutralizations (segmented phonetic form before re-syllabification; elements that do not surface are given in brackets).

The CC of the type aimer is, however, never affected by this neutralization process. This means that, in some tenses, we have a twofold opposition (i.e. 1^st CC vs. other CC), in others a threefold opposition (i.e. 1^st CC vs. the CC of type finir vs. other CCs) etc. What is more, partir and vendre have thematic forms only in those cases where we find the mentioned CC neutralization. This leads us to assume that the ThV is taken over from the thematic CC of the type finir.

Based on the observations made so far and in contrast to many other proposals, we argue that only the CC of the type aimer and finir are thematic in French. All other CCs are athematic. More precisely, we will assume that also partir (as well as dormir, cf. above) belongs to an athematic CC. The fact that this verb (apparently) has a ThV in some forms is related to the above-mentioned neutralization of CCs (which in the case of partir, parallel to the forms of the future and conditional, also affects the infinitive). This assumption is crucial for the discussion of the nature of ThVs. In our opinion ThVs are indicators for CCs (cf. §2 and fn. 3).

Considering the paradigms in Figures 9a and 9b, we assume that also in French the CCs are hierarchically structured according to diacritic features. As shown by Oltra Massuet (1999) for Catalan, the respective features allow us to represent the neutralizations of the CC mentioned before and also the differences in markedness (i. e. productivity, irregularity, size). We propose the hierarchy in Figure 10:

Figure 10

Feature geometry for the French conjugation classes.

The 1^st CC, type aimer, the default (illustrated here with [–α]²³), is the biggest, the most regular and also the most productive CC in French. It is characterized by a ThV ([ə]) and is thus thematic. Type [+α, –β, –γ] represents the CC with the augment -ss- (finir), which is highly regular, once this augment is explained as being part of the theme [is] which has a phonologically conditioned ThV variant [i] (cf. Schwarze 2009 and Table 7). Apart from being rather regular, this class also has some recent newcomers (deadjectival verbs like vioquir – from vioque, argot ‘vieux’ – meaning ‘to become old’ with a strongly pejorative connotation, cf. Boyé 2000: 23 as well as Schwarze 2009: 38). The athematic CCs of the type vendre and partir have the common features [+α, –β, +γ] and are conceived as being more marked than the other two regular CCs. Most other irregular verbs (e.g. écrire, valoir, faire, avoir etc.) are also athematic and have the features [+α, +β] (note that this group can be further divided in several even more marked sub classes).

The CCs with the features [+α, –β, +γ] are the ones showing CC neutralization, i.e. if there appear ThVs in some verbal forms, this is only due to the neutralization process where the ThV is taken over from the “dominating” CC of type finir [+α, –β, –γ]: (i) in the passé simple and the past subjunctive the unproductive type vendre (i.e. [+α, –β, +γ, –δ]) and the unproductive type partir (i.e. [+α, –β, +γ, +δ]) are neutralized, (ii) in the future tense and the conditional only type partir is affected from CC neutralization (cf. the rule in (6) and Figure 13 below).

3.3 Our proposal based on a mixed CC system

In this section, we propose an analysis within the framework of Distributed Morphology, where syntax generates hierarchical structures, which are then morphophonologically realized by Vocabulary Insertion (VI). Since we consider ThVs to be mere ornamental elements, they do not impinge on the syntactic derivation and are thus added post-syntactically by a well-formedness condition (cf. Oltra-Massuet 1999). We assume that neither the syntactic derivation nor the post-syntactic well-formedness conditions are affected by the CC features presented in Section 3.2, cf. Figure 11. This means, on the one side, that all CCs have the same morphological structure for VI and, on the other side, that athematicity does not (yet) affect the morphological structure in French, but only the exponence of this structure.

Figure 11

Syntactic output and morphological structure including ThV.²⁴

The exponence of this structure is sensitive to the CC features and therefore differs considerably from one CC to another, as illustrated in Table 8: CC of the type aimer (cf. atomiser,²⁵ aimer) and CC of the type finir have a regular root followed by three inflectional affixes in the 1^st plural imperfect. The CCs that are marked by two and more [+]-diacritics (cf. vendre and partir), instead, allow less affixes after the root and the ThV is missing, except in the cases where neutralization takes place (note, again, that many of the irregular verbs are also marked by more [+]-diacritics than just [+α, +β] once subclassification is taken into consideration; we will not analyse these irregular verbs here).

Table 8

Relation between morphological structures and CC (1pl. imperfect).²⁶²⁷

We assume that the CC feature [+γ] (and [+β]) are responsible for the non-realization of the ThV position, making these verbs athematic. The group marked with [+γ], i.e. the types vendre and partir, is additionally affected by CC neutralization. More precisely, the features [+γ, (+/–δ)] are deleted in certain contexts and, due to this, the opposition between these two CCs as well as the opposition between the two CCs and the one of the type finir are neutralized. The CC neutralization has the effect that also the athematic CCs have realized ThVs, e.g. partirons (first person plural future) and vendissions (first person plural imperfect subjunctive). To grasp this idea, we will illustrate in what follows the realization of the morphological structure first with and then without CC neutralization.

As illustrated in Figure 12, if the root-slot is realized by a vocabulary item that is specified for the CC [+α, –β, –γ] (e.g. fin-), the feature [+α] will trigger that the following ThV-slot is realized by the vocabulary item /-is-/ (cf. (5b)). If the vocabulary item inserted in the root-slot is, in contrast, specified additionally with the feature [+γ] (cf. vendre and partir) (or [+β] for the irregular verbs) the ThV-slot remains unrealized (cf. (5a)).²⁸ The insertion of /-is-/ is not possible in these cases since there is a vocabulary item, i.e. (5a), which is matching the greatest number of features specified for the context of insertion (cf. the Subset Principle, Halle 1997). The vocabulary item in (5c) is the default, which is inserted in the case of the thematic 1^st CC, i.e. type aimer.

Figure 12

Vocabulary Insertion for thematic and athematic CCs.

Figure 13

Neutralization of the CC oppositions.

Another aspect we want to mention here is that the athematic CCs are the ones with a high percentage of irregular verbs. This may be due to the fact that in these cases T° and/or φ are (linearly) adjacent to the root, since there is no intervening ThV (cf. Figure 12), and can thus trigger root allomorphy, whereas this is not the case in the thematic CCs (cf. Calabrese 2015a/b and Pomino & Remberger 2022a/b for more details).

Let us now look at those cases where the CC opposition is neutralized. Neutralization means that the verbs of the type vendre ([+α, –β, +γ, –δ]) and partir ([+α, –β, +γ, +δ]) get the ThVs of the CC of type finir ([+α, –β]) in particular tenses/moods. In line with Oltra-Massuet (1999: 45), we assume that neutralization can be analysed as arising from impoverishment of marked idiosyncratic features (or rather by cutting off the respective branches of the feature geometry). For the CCs at issue here, we propose the impoverishment rules in (6) that establish that the features [+γ, +δ] are deleted in the context given in (6a), whereas the feature [+γ] (and all branches it dominates) is deleted in the context described in (6b) (cf. also Figure 13).

The impoverishment rules have a direct effect on VI (cf. Figure 14): Rule (6a) will delete, for example, the features [+γ, +δ] of partir in the future, making the insertion of vocabulary item (5a) impossible. Since the feature [+α] remains, however, unaffected by the impoverishment rule, the ThV-slot will be realized by (5b), i.e. by /-is-/. This is how the athematic CCs get the ThV of the CC of type finir.

Figure 14

Vocabulary Insertion after CC neutralization (1pl future).

Our analysis does not only predict the correct forms, it also brings other clear benefits: Assuming a morphological structure which contain ThVs, we are not forced to distinguish between FC and PC, since in case of aim- and fin- the root-final consonant is always “saved” by the ThV, whereas in case of part- and vend- its realization depends on the realization of T° or φ (in case T° is not overtly realized as, e.g. in the present tense), cf. Figure 15. Note that El Fenne’s proposal could nevertheless be easily integrated into our analysis: We could assume without any problem that the root-final consonants are PC in the 1^st CC. This would however mean that we would unnecessarily account twice for the consonant-zero-alternation.

Figure 15

partons (1pl) vs. part (3sg).

With respect to the verbs partir and vendre, our analysis does not diverge much from the proposal by El Fenne (cf. the verb vivre in Figure 3), since we agree with her in that these verbs do not have, as a general rule, (an exponent for) ThVs and therefore the final consonant does not surface if it cannot occupy an onset position. The main difference of our proposal consists in the fact that we don’t assume a general affix (cf. stem + GA) and tense-specific stems but propose a finer decomposition for French verbs into root + (ThV +) T + φ. The reason for that has already become clear in Section 2.2, where we considered the other tense forms in the French paradigm and showed that a finer segmentation of French verbal forms can indeed account for the presence of ThVs in the CCs of the type aimer and finir, including the extension of the ThV of the type finir to athematic CCs in some particular tenses, where CC-neutralisation takes place. The finer segmentation we opt for and which makes the assumption of several tense-specific stems and tense-specific person-number endings unnecessary is given in Table 9:

In a regular thematic 1^st CC verb, like arriver we see that the 1^st person plural ending is regularly the same in most verbal forms, i.e. we just need one vocabulary item for these φ-features. T° is not realized in the present tense, in the same way as it is unrealized in other Romance languages. However, in other tenses, here in the future and the imperfect, we have specialized exponence for T°. The ThV of the 1^st CC is usually deleted, however, its presence and insertion as a vocabulary item prevents the deletion of the stem final consonant. This segmentation gives us a maximally consistent inventory of form-function correspondences, at least for the regular verbs.