2.1 Well-formedness conditions and verb structure

Oltra-Massuet (1999 et seq.) assumes that the verbal structure in Catalan is derived in syntax by movement of the root to the category-assigning head v° that further moves to T° (which, as a general rule, can encode tense, mood and aspect conjointly). This cyclic head-to-head movement results in the structure [[[ √root ] v°]_vP T° ]_TP, which is post-syntactically further modified by two well-formedness conditions: (i) finite T° requires an Agr node (see also Marantz 1991) and (ii) ThV positions have to be adjoined. The first well-formedness condition may be implemented in several ways depending on how subject-verb agreement is conceived. Based on Marantz (1991), one could assume that the φ-features of the subject DP (the c-commanding DP in Spec,TP) are post-syntactically copied and adjoined to T° (Figure 2a). If one assumes instead that the presence of agreement features on the verb is syntactic, the φ-features are already part of T° in the syntactic output. In this case, one may assume that the φ-features must receive a separate slot for realization, i.e. they are split off from T° (Figure 2b). Essentially, the output will be the same in both cases.

Figure 2: (a) Post-syntactic agreement vs. (b) syntactic agreement.

In those languages that have ThVs, the well-formedness condition for adding them obviously plays a central role, i.e. it is a language-specific condition. Despite many diachronic and synchronic differences, Spanish, Portuguese, Catalan and Italian have clearly preserved thematic CCs.² ThVs serve to distinguish between CCs and are considered by many linguists – as in Oltra-Massuet (1999) – to be mere ornamental elements without any effect on syntax and semantics (see however Kayne 2016; 2019 and Fábregas 2017; 2022, who analyze ThVs as verbalizers), which is why they are added only post-syntactically. For Romance verbs, it is standardly assumed that a ThV is added to the verbal root to form a stem. Oltra-Massuet (1999) instead proposes that every functional head receives a ThV position, i.e. not only v° but also T° (and other functional categories such as Fut°). The ThV of these other functional heads are, however, not sensitive to CC features, but depend on morpho-syntactic and sometimes on phonological features (Oltra-Massuet 1999). The morpho-syntactic structure after the application of both well-formedness conditions is as illustrated in Figure 3 (according to Oltra-Massuet 1999 et seq.; see also Pomino 2008 for Spanish).

Figure 3: Morpho-syntactic structure after application of the well-formedness conditions (Oltra-Massuet 1999).

The six terminal elements contained in Figure 3 may be realized with their respective phonological material at Vocabulary Insertion. This is the case for atom-iz-á-b-a-mos, for example. Not all verbal forms have an exponent for each slot, however (see Table 1): crucially, apart from the root, all other terminal nodes may remain unrealized, depending on the specific verb and on specific verbal forms.

Table 1: Selected (regular) verbal forms (Spanish).

The mismatches between the number of terminal elements and the respective realizations can be explained in multiple ways. For the present tense, Oltra-Massuet (1999) and Arregi (2000) propose that morphological complexity is directly related to the syntactic-semantic features of the respective forms. They argue that T° fuses with φ whenever it encodes a semantically unmarked tense feature, i.e. a tense feature that is also never realized morpho-phonologically (see Figure 4).

Figure 4: (a) split T and φ (imperfect) vs. (b) fused T/φ (present tense) in Spanish and Italian.

Fusion thereby cancels or reverses the well-formedness condition by which the position for φ and the ThV of T° were introduced. Trying to reduce the post-syntactic processes to a minimum, we argue against such redundant processes and instead propose that the φ-features encoded syntactically under T° do not receive a separate slot when T° encodes a semantically unmarked tense feature. Or, alternatively, T° and φ features receive separate slots only when T° encodes a marked feature value. From this perspective, the structure in Figure 4b is not the result of a fusion rule that reverses the well-formedness condition, but the result of the rule not applying when unmarked features are encoded in the syntactic output element (e.g. present indicative). Something similar holds for all verbal forms of the 3^rd person singular: the φ-features of these forms are never realized separately (at least in Spanish), irrespective of tense (e.g. 3^rd person am-a, am-a-b-a vs. 2^nd person am-a-s, am-a-b-a-s). More precisely, T°/φ does not split in the present indicative, since the tense value is unmarked, nor in all forms of the 3^rd person singular, because of the unmarked φ-value.³ This means that the φ-features also play a role in determining whether or not they receive a separate slot. We thus suggest the following slight modification of the well-formedness condition proposed by Oltra-Massuet (1999) and Arregi (2000), which makes the fusion rule discussed above superfluous.

In sum, the morpho-syntactic structure of the present (indicative) is less complex than that shown in Figure 3 in the sense that it has only 4 (not 6) terminal elements to be realized. The reverse is true for future tense and conditional: Both tenses have, according to Oltra-Massuet (1999) and Arregi (2000), one additional functional head. This is visible in the conditional forms, which are clearly more complex than the imperfect forms (Arregi 2000: 6), compare e.g. cant-á-b-a-mos vs. cant-a-r-í-a-mos.⁵ Following Oltra-Massuet (1999: 66), Arregi (2000) assumes for Spanish that /i/ and /a/ each realize the ThV of different functional heads, namely Fut° and T°: cant-Ø_v°-a_VTh-r_Fut°-í_Th-Ø_T°-a_Th-mos (Oltra-Massuet 1999 proposes a ModP that encodes [future] as a modal feature). The structure of conditional verb forms is thus more complex and the conditional is interpreted as “future (or posteriority) in the past” (as in for example María dijo que Juan cantaría una canción ‘Mary said that John would sing a song’). In other words, posteriority (or future) is detached from T° and has its own functional head, i.e. Fut°. Analogically, future tense does also consist of T° (= simultaneity) and Fut° (= posteriority). Further evidence can be found in diachrony: Roberts & Roussou (2003) discuss, for example, a comparable verb structure in their diachronic analysis of the future tense (which is derived from periphrastic cantare + habeo_{[present], [obligation]} > cantaré) and the conditional (derived from periphrastic cantare + habebam_{[imperfect], [obligation]} > cantaría). The features, formally encoded in the auxiliary, stem from two different functional heads, Mod° for [obligation] and T° for [present/imperfect].⁶

In summary, after the well-formedness condition discussed above is applied, future tense and conditional have the morpho-syntactic structures shown in Figure 5. Note that for the future tense (Figure 5b), T° encodes an unmarked tense relation (i.e. [present]), since the future meaning is a proper functional category derived from present tense mood, and thus, as discussed above, the φ-features do not receive a separate slot.

Figure 5: Conditional and future tense.

Let us turn now to the forms of the Spanish indefinido, which, rather like the present tense, are morpho-phonologically less complex than the imperfect forms (see Table 2); e.g. cant-á-b-a-mos vs. cant-a-mos). The verb forms of the indefinido are considered to be particularly problematic in Romance morphology as far as their segmentation is concerned. Ambadiang (1993), for example, claims that there are three different stems for the regular Spanish verb cantar ‘sing’ and divides the ending into four parts assuming several zero morph(eme)s, while Schpak-Dolt (1999) argues for one root which is followed by a ThV – the ThV of the 1^st person singular is atypically -é (not -a) – and a complex (or fusioned) ending. Despite many differences, both proposals reflect the fact that in the indefinido there is no clear exponent of T° and/or φ.

Table 2: Two different proposals for segmentation of the indefinido.

In the framework of DM, Arregi (2000) again postulates a fusion operation to account for the short forms of the indefinido and the irregular endings: “we assume that T and Agr fuse in the perfective […]. Given that T and Agr fuse, we explain why there is only one slot for both morphemes and why Agr looks so different from other contexts” (Arregi 2000: 18f). The resulting structure (and segmentation) is then comparable to the proposal by Schpak-Dolt (1999) in Table 2 and similar to that for the present indicative given in Figure 4b.

The main objection to Arregi’s (and Oltra-Massuet’s) analysis is the lack of motivation for the fusion rule that they postulate. For present indicative, both authors assume that it is the unmarked tense value that causes fusion, but this cannot be the reason for fusion in the indefinido, since this tense value is semantically not unmarked. Fusion remains thus unmotivated. For this reason, we want to argue that the morpho-syntactic structure of the indefinido is fundamentally the same as for the imperfect. The difference is, however, that in the indefinido the available VIs do not realize each slot one-by-one but cumulatively (see our analysis based on spanning in §4). The cumulative exponence in this tense is motivated diachronically by the loss of the Latin aspect marker, which in some cases led to root allomorphy, and subsequent phonological fusion processes (e.g. -avit > -aut > -ót > -ó for 3^rd person singular; Alvar & Pottier 1983: 237). Before going into more detail on this point, we must first further explore ThVs and the question of which functional category they belong to. Note that only the ThVs of v° depend on CC features and are, thus, ThVs in a strict or traditional sense. When we speak of athematic verbal forms in what follows, we mean that the ThV of v° is not morpho-phonologically realized.

2.2 The notion of theme vowel (ThV): From Latin to Romance/Spanish

ThVs are a particular property of some Indo-European languages, among them Latin and its Romance descendants, which have preserved or further transformed them, depending on the language under investigation. They are commonly defined as vowels that appear after the root (including derivational affixes) and the inflectional suffixes. In traditional Latin grammar, root and ThV build the stem of the verb. In some Romance languages, as shown in Section 2.1, they are the marker of CC and, in the DM framework, they are inserted by a well-formedness condition. ThVs were already CC-markers in Latin, although the Latin CC system was different from that of Romance, not only because of the loss of phonematicity in vowel length in Romance (see Table 3).⁷

Table 3: Latin CCs.⁹

In Latin, there was a further distinction between CCs with and without ThVs, and this division led some scholars to distinguish, in contrast to Table 3, only two Latin CCs, one thematic and one athematic (Kühner & Holzweissig 1912: 659).¹⁰ The athematic class is older and comprises Latin verbs like ferre ‘to carry’, esse ‘to be’, velle ‘to want’ etc. The Latin third CC, which is also called the consonantal conjugation, contains many irregular verbs and has many forms in the paradigm where there is no ThV, namely in the perfectum.¹¹ A visible vowel after the root is found instead in the infectum, e.g. in leg-e-re ‘to read’ or leg-i-mus ‘we read’. In the perfectum, there are several possible types of inflectional marker, but the ThV is generally visible only in the I and the IV CC, so we find am-ā-v-it ‘s/he loved/has loved’ and aud-ī-v-it ‘s/he heard/has heard’ with the respective ThV ā and ī, but mon-u-it ‘s/he warned/has warned’, scrip-s-it ‘s/he wrote/has written’, vēn-it ‘s/he came/has come’ etc. without the ThVs ē and ĕ (see also Calabrese & Petrosino 2023: 38). Furthermore, even in the most regular verbs, there are forms without a visible ThV, as the first person singular of the I CC, am-ō ‘I love’ without ThV¹² vs. am-ā-s ‘you love’ with ThV.

In any case, the function of the ThVs¹³ is highly controversial, but here we want to emphasize that beyond the distinction between Latin thematic and athematic CCs, an important distinction must also be drawn between thematic verbal forms and athematic verbal forms. Furthermore, as will be shown in what follows, not every vowel following a root is automatically a ThV (in the sense defined at the end of §2.1).¹⁴

A further distinction concerning ThVs comes into play when we return to the claim made by Oltra-Massuet (1999) that every functional projection is subject to the well-formedness condition that there must be a syntactic slot for a ThV projected. That means that not only v°, but also T° and other verbal functional categories have ThV-positions (or are licensed by a ThV). For Romance, this has already been illustrated in Section 2.1, but it can also be shown for Latin, as in Table 4, where T/Asp°¹⁵ has a ThV, independent of CC (note that the ThV is always the same, since it is the ThV conditioned by T°/Asp°[perfect]):

Table 4: Selected verbal forms in the Latin perfect.

There can even be three ThVs when further functional projections are contained in the syntactic input, as shown in Table 5:

Table 5: Selected verbal forms in the Latin pluperfect Latin.

In Table 5, the ThV of the T/Asp° head of the pluperfect is /e/, whereas the ThV of pluperfective T° is always /a/, again, independent of CC. That is, verbal forms in Latin have several types of ThV, and only the ThV of v° is a CC-marker. Moreover, the CC-marker is visible in the verbal forms built with the perfect root allomorph only in the I and IV CC, whereas all the other verbs lack a CC-marker in these tenses.

If we adopt this analysis for the successor forms of Latin in Spanish, we will see the same effect: In the root allomorph used for the indefinido, the past subjunctive (which is derived from the Latin pluperfect, either the indicative or the subjunctive), and the future subjunctive (derived from the Latin future II), the vowel following the root allomorph is the ThV of v° only in the first Spanish CC in -ar, whereas in all the other cases the vowels are ThVs of other functional categories; see Table 6 for 3^rd person singular:

Table 6: Selected verbal forms from Latin to Spanish 1.

The same pattern can be observed for the Latin pluperfect subjunctive, which became an alternative form for the past subjunctive in Spanish, and for the future subjunctive, which stems from the Latin future II; see Table 7 for the 3^rd person singular.

Of course, in Spanish in particular several verbs change CC and the supposedly five CCs reduce to three; however, many verbs of the III CC are also irregular in Spanish, and new irregularities also developed under analogical pressure. What can be said is that in the tenses and moods that feature the Latin stem known as the perfective, only the I CC exhibits a ThV connected to v° in the resulting form in Spanish. The other ThVs, including those that survived the phonological reductions in their development into Spanish, belong to other functional categories, which can clearly be seen from the fact that they are only dependent on CC distinctions insofar as they might have disappeared in the first CC, but are the same in the second and third CC in Spanish. To illustrate this, Table 8 shows the strong indefinido forms of the verbs andar (1^st CC), saber (2^nd CC) and decir (3^rd CC):

Table 7: Selected verbal forms from Latin to Spanish 2.

Table 8: Strong indefinido forms.

Although these three verbs belong to different CCs, they all show the same ThVs – this is especially surprising for the first CC which is never usually neutralized with other CCs in Spanish. The element -uv- in anduve is a remnant of the Latin -u-preterite that was preserved in a handful of commonly used verbs. Interestingly, in those cases where this remnant is maintained, verbs of the first CC also have -e-, -i- or -ie- as the ThV (and never the default -a-). On the basis of the Latin forms exemplified above, it is our opinion that this clearly shows that these ThVs are not related to CC features, i.e. they are not ThVs of v°. This observation will be relevant in the next section, which looks at irregularities.

In summary, ThVs are different kinds of elements and it is important to distinguish between:

• – verbs that belong to a CC that is specified by a certain ThV;

• – verbal forms that require the realization of a ThV; and

• – the functional elements that the ThV is required by.

The next section takes a closer look at irregular verbal forms (in Spanish) and makes a further distinction with respect to the presence of ThV slots and ThV realizations.

2.3 Irregular verbs in Spanish

The irregularities found in Spanish verbal inflection exhibit a certain systematicity.¹⁹ Alcoba (1999) distinguishes between the external and the internal distribution of these irregularities: the external distribution focuses on which verbs are affected by an irregularity, while the internal distribution highlights which forms of a verb exhibit the corresponding irregularity. This subsection focuses on the latter, which is more systematic than the former and reveals an astonishingly regular pattern. Of the five internal distribution patterns described by Alcoba (1999: 4952ff.) three are related to athematic verbal forms (in the sense discussed in §2.2) and are thus the focus of our analysis in Section 4.

As shown in (4a), the root allomorphy that is found with strong indefinido forms internally spreads over the imperfect subjunctive and future subjunctive (Alcoba 1999). As shown in Section 2.2, these tense forms all share the property of being athematic: more precisely, they have ThVs for T° (and/or Fut°), but not for v°. What has changed from Spanish to Latin is that in Latin a perfect stem, like scripsit ‘s/he said/has said’ or dixit ‘s/he said/has said’, was still segmentable in that as the /s/ was recognizable as a VI (i.e. as a morpho-phonological realization) for T/Asp°, but in Spanish the indefinido has lost the aspectual meaning of the perfect and it is no longer segmentable. Allomorphic forms such as dij- thus do not exhibit a recognizable realization of T°, but represent a portmanteau-morpheme, i.e. a VI that realizes the root, v° and its ThV slot in the context of a T°_[indefinido], which is itself realized as part of a span (cf. §3).

Something similar is true for the future tense and the conditional in (4b): verbs that show root allomorphy in these tenses leave the ThV of v° unrealized in Spanish (compare *poneré/*ponedré; *ponería/*ponedría, where a morpho-phonological realization ThV of v° leads to ungrammaticality). And finally, (4c) illustrates the internal distribution pattern of consonantal root allomorphy found in the present indicative and the present subjunctive. Note that in the present indicative 1^st person singular is always athematic, including in regular verbs (e.g. beb-o/*beb-e-o²⁰ vs. beb-e-s). This was already the case in CC I and III in Latin, so the athematic form is partially inherited (see §4 for more details). What is more, it is important to note that the vowel -a- in the present subjunctive is not the ThV of v°, but of T°.²¹ These forms also leave the ThV of v° unrealized in Latin.

We have argued form a diachronic and a synchronic point of view that all the verb forms listed in (4) are athematic and we offer an analysis based on spanning for these forms in Section 4.

We want to highlight, however, that there are two other patterns of root allomorphy in Spanish that are not related to athematicity, but to prosodic and/or phonological factors. Consider first the vocalic alternations in (5) and their internal distribution: As is well-known, the common feature of all these irregular verbal forms is that the stress falls onto the root.

From a synchronic point of view, it is not possible to assume a pure phonological process that predicts the appearance of the monophthong vs. diphthong, i.e. we cannot postulate a phonological process that automatically (re)adjusts the output of Vocabulary Insertion.²⁴,²⁵ It is also not possible, however, to assume that certain morpho-syntactic features trigger the respective alternation. That means that one would be forced to postulate a morpho-phonological readjustment rule valid for a list of verbs or a diacritic marker for those verbs that are affected by the respective rule(s) (Embick 2004, among others), e.g. /pEn-/ ↔ √pensar and a rule that states that /E/ is realized as a diphthong when stressed. In DM it is however also possible to explain the alternation mentioned above via Vocabulary Insertion by assuming one VI that must be inserted whenever the root is stressed (e.g. pien-) and another VI that can only be inserted in an unstressed root (e.g. pen-).²⁶ For these cases of root allomorphy, some kind of listing seems to be inevitable.

Another type of vocalic alternation found in the verbal inflection of Spanish, which always co-occurs with that shown in (5), is the raising of /e/ → /i/ and /o/ → /u/ (see the internal distribution in (7)). In contrast to what we saw above, stress does not fall on the root in these verb forms and only verbs of the third CC (i.e. the “i-conjugation”, e.g. pedir) are affected by this root alternation. There are different traditional explanations for this kind of apophony (see Togeby 1972 for more details and counterexamples): (i) metaphony due to a following yod (e.g. Menéndez Pidal 1904 et seq: §§11, 105, 114) – but why is metaphony restricted to the third CC rather than also affecting verbs of the second CC in the same phonological context?, (ii) monophthongization of /je/ to /i/ (see Malkiel 1966; lat. vesto > *viesto > visto), (iii) dissimilation of /i/ to /e/ under the influence of a tonic i in the following syllable (e.g. dicere > decir not *dicir; ridere > reír not *riir; see Togeby 1972: 258), (iv) analogy among others. Whatever the correct diachronic explanation is, synchronically this alternation can no longer be motivated solely by an active phonological process. Thus, one of the possibilities above (or some other explanation) must be postulated, but this lies outside the scope of our paper.

Finally, we want to mention an observation made by Alcoba (1999: 4955) regarding the overlapping of different alternations. The verb tener shows, for example, a vowel alternation and a consonant alternation, the latter being associated with athematicity. The internal distribution of these alternations partially overlaps (see the highlighted cells in (8)). In this case, it is the consonantal alternation that wins in this kind of competition. As we will show in Section 4, our analysis based on spanning naturally explains why the correct root is teng-: teng- is the VI that, due to its spanning size, optimally fits into the context of insertion and blocks any other competing VI.

We have seen, based on Spanish verbal forms, that there are different kinds of alternation with different internal distributions. However, our paper focusses on those that are linked to athematicity, i.e. the absence of a visible ThV, though we acknowledge that there are other alternations, mostly conditioned by stress, that may require other explanations (e.g. readjustment rules, specific VIs as proposed above for /tjen-/).

3.1 Locality conditions on allomorphy

Since DM is a realizational morphological theory, syntactic elements do not carry phonological information and the phonological material is instead provided post-syntactically via Vocabulary Insertion, as explained in Section 1. The realization of the terminal nodes may be conditioned, however, by various features (inward and outward sensitivity, as shown in Figure 1 based on Bobaljik (2000) in §1). The question is under which locality condition a grammatical feature may or may not trigger allomorphy. Consider Figure 6: There seems to be consensus in the literature that, when it comes to the realization of the root, the element X with the feature [F1] can impinge on the form of the root because it is linearly and hierarchically adjacent to the root. One point of controversy, however, is whether or not non-adjacent elements (here: Y and Z) can influence the realization of the root, i.e. trigger root allomorphy.

Figure 6: Root adjacent vs. non-adjacent elements.

Let us illustrate this using further examples: Departing from the morphological structure proposed by Oltra-Massuet (1999) in Figure 3, the following question arises: Under which conditions may T° and/or φ (or other apparently non-adjacent features, e.g. Asp°) trigger root allomorphy? In line with Calabrese (2015a; b), we will argue (see §4) that root allomorphy is possible whenever the ThV of v° is not separately realized. Our assumption is based on Bobaljik’s (2012; 2015) observation that there are certain locality domains within complex words that are determined by domain delimiters.

In Romance, as we want to argue, it seems as if morpho-phonologically realized ThVs also have a domain delimiting function of this type: An overtly realized ThV blocks the interaction between the context to its left and the context to its right; see. Figure 7. Therefore, the default realization of the root cannot spread further than v° and, more importantly for our analysis in Section 4, in these cases, T°, T°-ThV and φ cannot trigger root allomorphy. The situation is different, however, with athematic verbal forms: As soon as there are more specific VIs that contain the ThV of v° (by spanning) such that it cannot be separately realized (or visibly present) the domain delimiting effects of visible ThVs vanish and root allomorphy can be triggered by properties of T° or φ.

Figure 7: Italian ThV as domain delimiter in the sense of Bobaljik (2015) (the structure is based on Oltra-Massuet 1999).

We will illustrate this in the following sections using a spanning approach in order to account for the interplay of athematicity and allomorphy with examples from Romance verbal inflection.

3.2 Suppletion generalization

Work on verbal suppletion has shown that, in addition to domain delimiting effects, there is also a relation between the irregularities of a verbal form and the number of affixes that form has; see the two generalizations in (9).

To illustrate his ideas, Vanden Wyngaerd (2018) refers to Calabrese (2015b: 70) who noted that there is a striking correlation between the presence and absence of regular morphology and the presence and absence of ThVs. This correlation is reflected, for instance, in the past participle forms given in Table 10. In the regular past participles amato, battuto and partito, the √root is extended by a ThV to form a stem. The irregular past participles perso, corso and eccelso are instead athematic, i.e. the √root is not extended by a ThV and thus constitutes the stem in Calabrese’s segmentation. According to Calabrese, the irregularity of the Italian past participles is not reflected in the √root itself, but in the ending (here -s- not -t-).

Table 10: Italian past participles (the segmentation corresponds to that given in Calabrese 2015b).

In contrast to Calabrese, we have argued in Pomino & Remberger (2022a; b; c) that the segmentation given in Table 10 for the irregular past participle, which is based on the Latin sigmatic perfect, is not tenable from a synchronic point of view. We propose instead that Italian pers-, cors-, eccels- etc. should be synchronically represented as allomorphic VIs to perd-, corr-, eccell- etc. and thus should no longer be segmented (see Table 11). In other words, the generalization in (9b) can be paraphrased as follows: The loss of an affix (and its features) is formally reflected in the irregularity of the root (note that this is also a plausible reflex of the underlying diachronic development).

Table 11: Italian past participles (simplified).

As will be shown in Section 4, our proposal can be implemented as follows: the irregular form pers- does not only realize the root slot, but spans over far more than one final element. As a direct consequence, the irregular verbal form will be shorter or have fewer affixes than the regular form.

3.3 Grammatically determined athematicity in DM approaches

In this section, we will briefly discuss how the absence of a ThV can be implemented in the framework of DM (see Pomino & Remberger 2022a; b; c for more details). There are several proposals in the literature regarding the analysis of athematicity, but it can be reduced to three major types (we will leave aside proposals in which verbal forms are not segmented into √root + ThV, e.g. Bermúdez-Otero 2013).²⁸ One option is to assume that a terminal does not surface because it is realized by Ø (zero exponence with pruning of Ø) (e.g. Calabrese 2012). Another option would be to postulate that a terminal node does not surface since it fuses with another terminal before Vocabulary Insertion (see §1 for the unmarked present tense and the unmarked 3^rd person singular). A third possibility is to assume that a terminal node does not surface because it is realized cumulatively with other features (encoded in other terminal elements): this is the proposal that we will follow here. These three possibilities are exemplified in Figure 8: the proposal in Figure 8a is based on Calabrese (2012 et seq.) and exemplifies zero exponence; Figure 8b illustrates fusion (in this case of the √root and X); Figure 8c shows the solution based on the mechanism of spanning. In what follows, we will briefly highlight some shortcomings of Calabrese’s analysis from 2012, based on zero exponence (Figure 8a) with subsequent pruning (Embick 2010), and will propose our own analysis based on spanning in Section 4 (Figure 8c).

Figure 8: Zero exponence, fusion and spanning (Pomino & Remberger in submitted)

Pomino & Remberger (2022a; b; c) have repeatedly shown that an analysis based on zero exponence is far too complex and makes use of several, in our opinion unnecessary, processes. Calabrese (2012; 2013; 2015a; b) posits well-formedness conditions, exceptions to these conditions, zero exponence, different diacritic features, impoverishment (of diacritic features), listing of roots, highly context-sensitive VIs (that are soon afterwards deleted), and so on (Pomino & Remberger submitted).

Fusion also seems not to be the correct explanation, however: In an earlier paper, Pomino & Remberger (2019) did indeed argue in favor of an analysis based on fusion (Figure 8b) which they later had to reject. The reason for this change of mind is threefold: (i) as discussed in Section 2.1, fusion is even in the present tense a redundant process, since it simply reverses the well-formedness condition that introduces positions for φ (and ThV) and thus seems to be unnecessary, (ii) fusion, if assumed at all, is related to semantically unmarked tense features (as in the present tense) and it is thus not motivated for other tense forms such as the indefinido (the forms of which are as short as the present tense forms, e.g. cant-o and cant-é), and (iii) fusion, if one is willing to admit it in the indefinido (as e.g. Arregi 2000), would not alter the locality condition between the root and T°/φ in the necessary way in the case of quis-i-mos, for example. If one assumes fusion of T° and φ for the indefinido, the -i- cannot be interpreted as the ThV of T°; it has to be the ThV of v°. Apart from the fact that we have already shown in 2.2 that -i- is not the ThV of v°, we would generate a structure (see Figure 9) where allomorphy could not be motivated, that is it would be blocked because the root and the T/φ-features would be neither structurally nor (in most cases) linearly adjacent to the root, since there is an intervening ThV.

Figure 9: Spanish indefinido quisimos if fusion were an option (an approach that we do not follow).

The third possibility, which we will pursue in the next section, is the non-realization of certain terminals by spanning, which can insert phonological material in more than one terminal node at a time (Figure 8c). Spanning is a way to formalize cumulative exponence.²⁹

4.1 Grammatically-determined athematicity

Consider again the irregular forms of the Italian past participles in Table 10–11 in Section 3.2: Calabrese (2015) proposes that Italian past participles do not have a ThV position for Asp°. Although there is never³¹ an overt exponent for a ThV for this functional head in the sense of Oltra-Massuet (1999), we would like to retain her proposal, and assume that every functional head receives a ThV (see Figure 11). That is, although we do not see a phonologically realized ThV of Asp° in the Italian past participle, we still assume – for reasons of parallelism – that there is a position for it, even if it is realized only within a span (even without the assumption of a ThV for Asp° our proposal would work in the same way). For the time being, we represent the participle Asp° with a structural position for ThV; see the structure given in Figure 11 for Italian past participles:

Figure 11: Morpho-syntactic structure of Italian past participles.

If in this structure the ThV of v° is overtly realized, it functions as a domain delimiter and it will block root allomorphy triggered by Asp°.³² The feature Asp°-[participle]³³ cannot impinge on the form of √root. Instead, if the ThV is within the span containing the root, allomorphy is possible. Moreover, the original status of /s/ as an exponent of Asp° is now reflected in a more widely spanning item, namely a VI that spans over <√root, v°, Th, Asp°-[participle], Th>. Our analysis is illustrated in Figure 12 with the participles temuto and perso.

Figure 12: (a) regular temuto ‘feared’ vs. (b) irregular perso ‘lost’ (Pomino & Remberger submitted).

For the verb perdere ‘(to) lose’ we assume (at least³⁴) the VIs in (10): the irregular form pers-, which spans over <√lose, v°, ThV, Asp°-[participle], ThV> leading to an athematic form and the (regular) default realization perd- (for reasons of space we do not list VIs that are not crucial for the argumentation; see Pomino & Remberger submitted for more details). Note that (10a) is a highly specialized item, which realizes several syntactic input nodes at once:

The form of the Italian suppletive verb andare ‘(to) go’ can be analyzed in a similar vein (see Pomino & Remberger 2022c for more details). In Italian, suppletion of the verb go only occurs in the present tense, where it is context-sensitive. In all other verbal categories, the root /and-/ is generalized, i.e., it is the default realization for √go (see Table 12).

Table 12: Italian present tense and imperfect of go.

As can be seen from this partial paradigm, since andare is regular and formally a first CC verb, it is equipped with the ThV /a/ in all the cells of the paradigm where it appears, parallel to other regular verbs like cantare ‘to sing’. In contrast, the suppletive forms based on va- are all athematic.³⁵ In line with what has been said before, we assume that the spanning size of va- is <√go, v°, Th>, whereas and- is the default realization with the spanning size <√go, v°>. Again, as can be seen in Figure 13, T° (including its φ-features) is adjacent to the preceding span and may trigger allomorphy, or function rather as context for insertion, in the va-forms, but not in the and-forms, since here there is an intervening ThV.³⁶

Figure 13: Italian suppletive verb (slightly modified from Pomino & Remberger 2022c: 17).³⁷

This analysis is not only applicable to Italian verbal inflection, but also to other Romance languages. Consider the Spanish verbal forms in Figure 14 and Figure 15: The irregular future form pondremos ‘(we) will put’ in Figure 14 as well as the irregular indefinido³⁸ form in Figure 15 differ from the respective imperfect form in the spanning size of the VI for the realization of the root. In both cases, it spans over the ThV and is adjacent to T°. That is, the features encoded in T° serve as a context of insertion.

Figure 14: (a) Spanish imperfect vs. (b) future tense.³⁹

Figure 15: Spanish imperfect vs. indefinido (= indef).⁴⁰

In the next section we will discuss the second type of athematicity that is not triggered by grammatical features.

4.2 Athematic conjugation classes (French)

The distinction between thematic and athematic CCs is crucial for our analysis of French verbal forms (Pomino & Remberger 2022b). Modern French verbs are traditionally divided into three CCs based on the form of the infinitive (see e.g. Meyer-Lübke 1908: 202): Verbs of the first CC have an infinitival form ending in -er [-e^ʁ], e.g. aimer [ɛ'me] ‘(to) love’. This CC does not only include 90% of all French verbs, but is also highly productive. Apart from aller ‘to go’, which shows suppletion and therefore only belongs in part to the first CC since it unites forms derived from verbs from 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.⁴¹ The second 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 and the proportion of regular verbs within this CC is even higher than in the first CC. However, the second CC includes only 2.8% of all French verbs and it is generally – with a few exceptions, including some recent newcomers – considered to be no longer productive, a fact that makes it more marked than the first CC. The third CC is in essence a smorgasbord of all other verbs that do not belong to the first or second CC. Verbs belonging to this class are, for instance, romp-re ‘to break’, peind-re ‘to paint’, voul-oir ‘to want’, dorm-ir ‘to sleep’ (infinitive in -ir but without -ss-extension). This CC contains many irregular verbs with stem allomorphy. It goes without saying that this CC is unproductive and counts as the most marked class.

There is no consensus in the literature with respect to whether or not French still has ThVs, and proposals for the classification of French CCs diverge considerably. We cannot discuss all the insights here (Pomino & Remberger 2022b), but we assume that only the first and second CCs are thematic in French. 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]. With respect to the [i]-[is]-alternation, Schwarze (2009) proposes 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 (as in the plural forms of finir in Table 13); 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.

Table 13: Forms of finir (in Pomino & Remberger 2022b adapted from Schwarze 2009).

Evidence for the existence of ThVs in French for the first CC comes from the consonant-zero-alternation mentioned in Table 9 (iv): The root-final consonant of viv(re) ‘to live’, which belongs to an athematic CC, is maintained if there is a possibility for it to appear in a syllable onset; as final coda consonant it is instead deleted (e.g. vivons [vi.vɔ̃] in onset position vs. vis [vi] in coda position, not *[viv]).⁴³ The same final consonant of arriv(er) ‘to arrive’, which is thematic, is never deleted: arrivons [a.ʁi.vɔ̃] in onset position and arrives [a.ʁiv] in coda position (not *[a.ʁi]). Schane (1966; 1968) and others assume that in this case the ThV /ǝ/ (e.g. /aʁiv+(ǝ)+(z)/), which does not surface since it is deleted at some point in the derivation,⁴⁴ blocks consonant deletion. Schane’s (1966; 1968) conclusion can be easily translated into our analysis of spanning.

If one is willing to assume ThVs for the first CC in French,⁴⁵ the alternation between having a root-final consonant or not, i.e. the allomorphy between viv- and vi-, can be explained by assuming that the exponent /viv-/ spans over root, v° and ThV. The final consonant is maintained whenever it can move into the onset position of the VI of the following node. Otherwise, the final consonant is deleted by a phonological rule, i.e. the fact that the final consonant does not surface in some forms is phonologically conditioned. Thus, as illustrated in Figure 16, in vivons the root-final consonant surfaces, because it can move into the onset position of the span-adjacent element (i.e. the exponent -ons of T°/φ). In vis, the consonant never surfaces, because it cannot move into the onset position; it is thus deleted phonologically. Things are different for arriver: The element arriv- spans only over root and v°. This means that the span-adjacent element is a ThV, although later deleted, and thus the final consonant will always be “saved”.

Figure 16: Thematic arriver vs. athematic vivre.

In summary, we assume that the well-formedness condition that adds ThVs is still valid for French, i.e. there is no structural difference between the roots of thematic arriver and athematic vivre. However, there is only one exponent for vivre: viv- (in all cases). The fact that the final consonant does not surface in some forms is phonologically conditioned.⁴⁶ If it is not linked to an onset slot of the span-adjacent element, it is deleted. The essential difference when compared to thematic arriver is that arriver always has an onset slot available that avoids consonant deletion, because of the presence of a realized ThV.

4.3 Inherited athematicity

Let us turn to the last type of athematicity, which we have termed inherited or reanalyzed athematicity. In order to understand our first tentative approach to this kind of athematicity, we will briefly discuss the analysis of van der Spuy (2020) for Latin, in which he argues for a phonological explanation of the Latin verbal inflection. He claims that all regular Latin verbs have the structure [stem]-[tense/mood]-[person/number]. Each of these components may receive a respective phonological realization and the resulting form may be further subject to a reduced number of (morpho-)phonological rules. Two of these rules are the phonological rules affecting ThVs proposed by van der Spuy (2020: 5), as shown in (11).

For instance, 1^st person present indicative is subject to the rule in (11) as illustrated in Table 14, as are all persons of the present subjunctive, since the subjunctive marker is either /e/ or /a/ (see Table 15).

Table 14: 1^st person present indicative laudo and lego according to van der Spuy (2020: 5).

Table 15: 1^st person present subjunctive laudo and lego according to van der Spuy (2020: 3, 5).

However, the rules in (11) postulated by van der Spuy (2020) are not easily traceable in diachrony, since there are simply no attested forms of the type laud-ā-ō (see also fn. 12); furthermore, this /a:/, following the Latin stress rule, would be accented and would constitute a really quite rare case of a stressed long vowel, in addition the central low vowel, which is the most stable in diachrony, being deleted. Let us simply state, then, that certain verbal forms were already athematic in Latin, even in the first CC.

Whatever the explanation is for the verbal forms in Latin, the corresponding verbal forms in Spanish (and in other Romance languages and varieties) are still athematic today; see the highlighted cells in Table 16.

Table 16: Present tense forms of regular comer ‘(to) eat’.

In the analysis we have proposed so far, this means that the span-adjacent element T° (or T°/φ) can condition the form of the root. This is exactly what we observe in the forms of irregular poner in Table 17: In the present indicative we have φ-triggered root allomorphy and in the present subjunctive root allomorphy is triggered by mood encoded in T°.

Table 17: Present tense forms of irregular poner ‘(to) put’.

How can we explain under a synchronic point of view, however, that the highlighted forms are athematic? One possibility, which is pursued by Oltra-Massuet (1999) and Arregi (2000), is to assume some kind of impoverishment rule that deletes the ThV in the respective contexts, as in (12).

Another possibility, which we want to propose despite being unable to trace the details of the diachronic evolution here, is to argue that the athematic Latin forms laudo, lego etc. (whether they are the output of a former Latin phonological rule as in (11) or not;) were reanalyzed as a form that completely lacks a ThV slot in its input form. Synchronically, this would mean that – also for regular verbs – v° lacks a ThV in 1^st singular present indicative (Figure 17) and in all persons in the present subjunctive. These forms, i.e. v° in the context of T°/φ are thus not affected by the well-formedness condition⁴⁸ that adds a ThV to functional heads (Oltra-Massuet 1999). Note that cant- has the same spanning size in both cases.

Figure 17: (a) thematic vs. (b) athematic present tense forms (based on Oltra-Massuet 1999).

If this generalization holds, we can once again use spanning to explain the irregularities in the Spanish verbal inflection that affect 1sg present indicative and all persons of the present subjunctive: pon-e-mos is regular and thematic, whereas pong-o, pong-a, pong-a-s etc. are irregular and athematic. Being athematic, the span-adjacent T°(/φ)-node can trigger root allomorphy; see Figure 18 and the VI in (13a)

Figure 18: Irregular forms of poner ‘(to) put’ (1^st person singular (a) and all present subjunctive forms (b)).

What the irregularities discussed in these subsections all have in common is that the verbal forms analyzed here are all athematic. The reasons for their athematicity are different, however (see Table 9): they may be athematic in the form (see the examples in §4.1) or in the morpho-syntactic structure (see the examples in §4.2, conditioned by CC, and §4.3, inherited conditions).