2.1 A sketch of the Sanskrit verbal system

Sanskrit verbs are traditionally organized into three different systems: the present, perfect, and aorist systems, which are distinguished in terms of aspectual features.² Each system includes mood distinctions: indicative, subjunctive, optative, imperative. Additionally, the indicative mood has tense distinctions between a present and a past.³ The endings display person, number, voice, and tense distinctions.⁴ The paradigm of the verb kar-nau- ‘make’ in Table 1 reports the basic structuring of the different verbal forms for the three systems.

The forms of this table display a complex surface allomorphy: the root appears as [kar/kr̥/kr] and the internal suffix as [ṇo/ṇu/ṇav]. In order to understand this pattern, one needs to look into Vedic Sanskrit morpho-phonology, and in particular the phenomena of accentuation and [a]-syncope. A typical sample case is the alternation between kr̥-ṇó-t-i/ kr̥-ṇu-yā́-t which also displays the mobile accentual pattern that is typical of this language. Here, the vowel -a- of the suffix -nau- (which surfaces as [no] by monophthongization and [nav] by resyllabification) is syncopated when an accented syllable such as -yā́- follows. The alternant that displays the vowel (-náu- > -nó-) is traditionally called the “full grade”, and the alternant where the vowel is syncopated (-nu-) is called the “zero grade”.⁷ The relationship between accentual shifts and full/zero grade is rather complex, and cannot be dealt with here (but see Calabrese & Grestenberger 2023 for a recent in-depth analysis). Since accentual mobility is irrelevant to our analysis, we will just focus on zero grade without considering its prosodic conditioning. For the sake of simplicity, we will assume that zero grade is triggered by certain specific suffixes such as the OPT morpheme -yā-, the 2pl.act morpheme -tha, etc. As we will see further below, the one crucial detail here is that while the vowel of some suffixes (“thematic”) never undergoes zero grade, the vowel of other suffixes (“athematic”) does.

Investigation of the cooccurrence restrictions among the different pieces already identified in Table 1 leads to the analysis in (1) where the different pieces appear in their underlying forms, i.e., with the subtraction of the effects of morpho-phonological processes such as monophthongization, accent, zero grade, and resyllabification. Given (1), -nau- (and the relative surface representations: [no], [nav], [nu]) is the exponent of present (imperfective aspect); a_prefix- of past tense; -a- of subjunctive mood, and -yā- of optative mood; and -t of the 3^rd person singular agreement ending.⁸ Otherwise, there are null exponents (see below for a more detailed discussion of the feature assignments of the different pieces).⁹

We are now ready to turn to the exponence of the root-adjacent nodes across systems. In this section, we introduce the exponence of the present system; a more in-depth analysis of this and the other systems is given in the following sections. The grammatical tradition recognizes ten verb classes, depending on the form of the present stem (MacDonnell 1927); they are traditionally divided into two broad groups: thematic (Table 2A) and athematic (Table 2B).

The difference between the athematic suffixes (Class V -nau-, Class VII -na-, and VIII -au-) on one side, and the thematic one (Class I -a-, Class VI -a-, and Class -ya-) follows from their zero-grade properties.¹³ The vowel of thematic suffixes -a- and -ya- never undergoes zero grade. On the contrary, the vowel [a] of as the athematic suffixes -au-, -nau-, -na-, may undergo zero grade when followed by a zero-grade triggering suffix. For example, the 2pl.act ending -tha triggers zero grade in the latter suffixes (2b), but not in the former (2a).¹⁴

In this paper, we also deal with ‘secondary’ verbs, i.e. verbal forms that are composed of roots and additional verbalizing suffixes (Burrow 1955: 353). For the current purposes, we focus on the passive, causative, desiderative, and denominative forms,¹⁵ which we will eventually labeled as VP-shell derivative verbal forms. The passive voice forms are made by adding -ya- to the zero-grade stem (3a). The causative forms add -ay(a)-¹⁶ to the root, which undergoes vowel lengthening (3b). The desiderative is formed by adding -sa- to a reduplicated, zero-grade root (3c).¹⁷ The denominative form add the suffix -ya- to nouns (3d).

Notice that the final [a] of all of these morphemes is “thematic” in not undergoing zero grade before accented endings: cf. (3d) vs (4), while keeping in mind that 2pl.act -tha is a zero-grade triggering suffix (2).

These suffixes appear to be associated with the present system and are usually morphologically incompatible with other aspectual markers. Combinations of VP-shell derivatives are quite rare, and the ones listed below are apparently the only ones attested (Whitney 1889: 238–239).¹⁸ For simplicity, (5) also reports intermediate forms, before morpho-phonological processes—accent, zero-grade, resyllabification, etc.— apply. From these forms, the passive seems to be in an outer position with respect to the desiderative (5a) and the causative (5c), which in turn are in an outer position with respect to the denominative (5e-f). The causative and the desiderative can instead appear in different scopal positions between each other (5b, d).

In the next section we derive the morpho-syntactic structure of the Sanskrit verbal forms within the DM framework adopted here, which will also allow us to illustrate the basic aspects of the model.

2.2 A DM analysis of the Sanskrit verbal system

DM (Halle & Marantz 1993; 1994) is a syntax-driven theory of morphology claiming that the distribution of morphological pieces in word formation is syntactically driven. The framework rests upon three main tenets. First, morphemes (roots and affixes), as complex bundles of morpho-syntactic and semantic features, are combined into larger syntactic structures, complex heads. Second, phonological spell-out applies cyclically morpheme-by-morpheme bottom up and includes cyclic Vocabulary Insertion (VI) that assign exponents, i.e., phonological indices, to the terminal nodes of the complex heads, and morpho-phonological and phonological processes that modify the phonological shape of these exponents. Third, such morphological and morpho-phonological operations are governed by locality principles sensitive to these structures (Embick 2010; 2013; Bobaljik 2012). The ultimate goal of this framework is to define the grammatical conditions that constrain morphological exponence under the assumption that morphemes are structure-dependent, syntactic entities. This results in two main corollaries. Firstly, given an independently motivated morpho-syntactic structure, changes in word exponence can only involve changes in the exponents of the terminal nodes of this structure. Furthermore, the structure constrains the analytical choices that can be made insofar as the nature of each morphological piece in the surface string must be assessed in terms of the overall—independently motivated—morpho-syntactic structure of the word. This approach to analytical choices will be crucial in this paper.

We derive the morpho-syntactic structure of the verbal systems of the languages considered here, by assuming the verbal functional structure in (6), along the lines of Wurmbrand (2015) and Cinque (1999).

As standard in DM, we also adopt the split VP hypothesis (see Harley 2017 and references therein), and with it a complex structural organization of the VP-shell, which consists of: vP, which assigns the verbal category to the category-unspecified root head of its complement; as such, it can also be a denominative/deadjectival verbalizer; CausP, which introduces causative syntax and semantics; and VoiceP, which controls the presence/absence of the external argument (and consequently assignment of accusative case; Marantz 1991). In addition, the Sanskrit VP-shell may also include a Desiderative⁰ head (and the relative maximal projection, DesP), which we assume entails a low aspectual head (possibly, akin to a conative head in the parlance of Cinque 1999). As illustrated earlier in (5), it may appear both above and below the causative with different scopal effects. The resulting structure is shown in (7). We will refer to the head nodes in (7) and their exponents as the VP-shell functional nodes and the VP-shell exponents, respectively.

The universal hierarchical structure formed by the combination of (6) and (7) is then mapped onto units—X⁰-complexes—which correspond to morphological words (Embick & Noyer 2001). X⁰-complexes, are generated by head movement operations. Along the lines of Calabrese & Pescarini (2014), we assume that morphological operations and syntactic derivation are cyclically interleaved. So, head-movement operations, during what we call morphological spell out (Calabrese 2019), may first generate X⁰-complexes, which can then be targeted by other syntactic head-movement operations such as V-to-C movement. The word-forming head movement operations are the only one of relevance here. As in Calabrese (2019), we assume that the affixal properties of functional heads during morphological spell-out follow from a morphological principle requiring that each functional head X⁰ in an extended projection, must be adjoined to a root or to a Y⁰ complex including a root. In this system, syntactic representations in violation of this principle are repaired through operations such as head raising (see Harizanov & Gribanova 2018; for the sake of simplicity, the alternative repair operation of head lowering is not covered in this paper since it is not directly relevant to the analysis developed here). We will not linger further over the technical aspects of head-movement and their justification here. Such a procedure will create the structure in (8a), which omits VP-shell derivatives such as the desirative and the causative just for the sake of graphic simplicity. The node Agr is then adjoined to the highest X⁰ in the complex verbal head (8b; Halle & Marantz 1993; Bobaljik 2000).

The resulting structure undergoes VI, i.e. the procedure of association between (bundles of) morpho-syntactic features (i.e., functional nodes) and phonological content (exponence), which is assumed to occur cyclically inside out (Bobaljik 2000). Now there is a problem. The X⁰-complex generated in (8) should include at least six affixes in addition to the root; only three pieces maximally appear in the actual forms in Table 1, as also shown in (9).

A first possibility is that the lack of overt realization of certain nodes is simply due to syntactic inactivity, i.e., the nodes that not overtly marked are simply absent in the morpho-syntactic representation. Embick & Halle (2005) and Halle (2018) proposed that unmarked features specifications are just syntactically absent in morpho-syntactic representations. A striking case discussed by Calabrese (to appear a) is the exponent of the Latin 1SG.ACT AGR node, which, putting aside the perfect /-ī/ (cf. laudāvī), is -ō in present and future indicative forms (laud-ō ‘I praise’, laudāb-ō ‘I will praise’, laudāver-ō ‘I will have praised’), otherwise -m (laudāba-m ‘I used to praise’, laudāvera-m ‘I had praised’, laude-m ‘(that) I praise (subj.)’, laudāre-m ‘(that) I praise (subj.)’, laudāvisse-m ‘(that) I used to praised (subj.)’, laudāveri-m ‘(that) I praised (subj.)’). The only way to characterize the distribution of -ō is to refer to the feature specifications [–PAST, -IRREALIS]. These are clearly unmarked, yet they must be present when Agreement exponence is inserted. This could not be accounted for if the relevant nodes were not syntactically projected. Zero-exponence (conventionally signaled as -Ø-) seems therefore unavoidable in situations where distributional properties of exponents can only be accounted for by referring to phonologically silent (i.e., null) feature specifications (see Calabrese to appear a, for further arguments). On the other hand, adoption of zero-exponence leads to theoretical underdetermination, which allows for a proliferation of equally possible theories (among others, Anderson 1992; Marantz 1997; Trommer 2012; Dahl & Fabregas 2018). To exemplify, consider the Sanskrit present suffix -nau-. It appears in root-adjacent position and competes with other pieces occurring in the same position; thus not only with the other present system markers in Table 2, and the VP-shell derivative suffixes in (3)–(5), but also the aorist -Ø- and -s-, as we will see below. Its distribution can then be characterized by the features [–DEN, –PASS, –CAUS, –PFV]. There is no other piece whose distribution depends on one of them. Once the featural distribution of an item such as -nau- is established, one must determine its structural distribution: the insertion site of the exponent. If the specification(s) belong(s) to a single terminal node, there is no issue: the exponent is inserted into this terminal node. The situation is more complicated if the Vocabulary Item includes specifications belonging to multiple terminal nodes. This is the case of -nau-. In this form, at least three nodes are realized by the single piece -nau-; so, potentially only one node is overtly realized; the other two are silent. At its current version, DM in principle allows the practitioner to analyze -nau- as realizing any of the existing nodes above the root (10), with no independent way to prefer one solution to the others (Pullum & Zwicky 1991):

To account for situations like this, an operation of node fusion, which bundles nodes together, was proposed in early models of DM (e.g., Halle & Marantz 1993).

Fusion, however, suffers from a look-ahead problem insofar as it must occur before VI so to allow insertion of the relevant fused exponent (Chung 2007; 2009). An alternative operation that can directly connect fusion and zero exponents is null-node pruning. This procedure consists of delinking nodes where Ø’s are inserted when independently-motivated terminal nodes lack a phonological realization. As Calabrese (2019) shows, null-node pruning is independently required to (a) simplify the phonological realization of morphosyntactic structures; and (b) account for the convergence of complex morphosyntactic structures and their simplex phonological realization. Additionally, (c) it explains why phonologically null exponents — regardless of their marked/unmarked status — appear not to act as interveners for morphological locality. Pruning applies as in (12). First, the terminal node Y⁰ is assigned a Ø during VI. As such, Y⁰ is detached from its immediately dominating node, but is not deleted. Its features become floating and are merged with an adjacent higher terminal node if there is one. The lower adjacent node X⁰ has already undergone VI – in (12), Φ is an overt exponent that is already inserted; the higher adjacent node Z⁰ is instead yet to be realized, thus being the available docking site. As a result, the nodes Y⁰ and Z⁰ fuse into Y⁰ + Z⁰, which undergoes VI again and get assigned the exponent Ω.

Thus, what matters is the pattern of overt exponence. Given an independently motivated morpho-syntactic structure such as the hierarchy in (8b) and the distribution of overt exponents, zero exponent insertion and bottom-up cyclic pruning can generate the relevant simplified morpho-syntactic structures. Once null exponents are identified, Øs can be considered as devices triggering pruning and subsequent fusion and we will be using them with this function. The resulting bundles will include features of nodes that do not have overt exponence despite being morpho-syntactically active.²¹ Pruning could be thought of as triggered by the UG filter in (13) which disallows a terminal node exhaustively dominating a null exponent, i.e. phonologically realized as Ø (see Arregi & Nevins 2012; Calabrese 2019, for a theory of morphology including morphological filters and repairs):

The constraint target morpho-syntactic configurations. Null exponents are independently needed as argued earlier to account for the silence associated with certain morphosyntactic features. Null exponents can be cyclically inserted during Vocabulary Insertion as motivated earlier. Once these null exponents are inserted in terminal nodes, they violate (13). But then they are immediately removed by pruning as a repair. This theory has consequences on the structure of the VIs and on the feature specification to be used in them. In fact, due to (13), the use of Ø is costly. We propose a principle governing VI composition:

It follows that given an overt exponent whose distribution involves the features of more nodes, it must be postulated in such a way that it maximally expones the entire sequence of nodes and must therefore be inserted in the highest node.²² Thus, in the case of -nau- whose distribution is characterized by the feature specifications [–DEN, –PASS, –CAUS, –PFV] belonging to different nodes, the most economical VI requested by (14) will be (15). Zeros, on the other hand, are assigned only by an universal default elsewhere operation which applies to all nodes not receiving overt exponence.²³

Let’s see how this works for Sanskrit. In the morpho-syntactic structure in (8) overt exponents are cyclically inserted bottom-up, with Ø’s being inserted at terminal nodes with no overt phonological realization. Taking into consideration the Sanskrit forms in (1), one observes that only Asp⁰ and Mood have overt exponence (16), whereas the other nodes are assigned zeros (16). These zero-exponed nodes are then pruned; feature floating and docking will generate the cyclic derivation in (17), where v⁰ and Voice⁰ are pruned and fused with Asp⁰; and T⁰ is pruned and fused with Mood⁰ (post-VI morpho-phonological processes are not considered here).

The resulting structure after insertion of an overt exponent in AGR is (18):

If also Mood⁰ has zero exponence as for example in indicative forms where only Asp⁰ has overt realization (e.g., kar-ṇáu-ti > kr̥nóti, ‘s/he makes’), the resulting structure is derived as in (19) and (20). (Again post-VI morpho-phonological processes are glossed over here.)

There are cases in which Asp⁰ is non-overt (e.g., dviṣthá ‘you (pl) hate’), therefore all of the functional nodes are fused together under Mood⁰ as in (21):

In the next section, we assume the basic structures in (18), (20), and (21), which will then be revised later on, when the secondary verbs are treated (sec. 2.3.1). This short theoretical digression served as an introduction to the theoretical framework the remainder of the paper couches in. From our perspective, what matters is the pattern of overt exponence. Given an independently motivated morpho-syntactic structure such as that in (8) once the distribution of overt exponents has been established, insertion of zeros and cyclic pruning will generate the relevant simplified morpho-syntactic structures.

2.3 The functional status of the Sanskrit root-adjacent exponence

Let us now consider the exponence of the root-adjacent nodes in more detail in the present, aorist, and perfect systems. As already mentioned above in Section 2.1, Sanskrit root-adjacent exponence of the present system of the primary verbs consists of the suffixes: -a-, -ya-, -nau-, -na-, and -Ø- (we will discuss the secondary verbs in the next section). As observed above, the suffixes -a- and -ya- are traditionally called “thematic” suffixes, insofar as they never undergo zero grade (22A). The suffixes -nau- and -na- are instead called “athematic”, insofar as they may undergo zero grade (22B). Strikingly, all of these suffixes never co-occur together, which suggests that they all compete for the same structural position.

We now expand on the root-adjacent exponent of the perfect and the aorist systems. The perfect system is characterized by root partial reduplication (not discussed here) and has a suffixal null exponent (23).

The aorist system may be formed in a number of ways; here we only consider a subset of them for reasons of space.²⁴ The athematic root aorist does not have an overt aspectual exponent (24a). The “thematic” aorist is formed with suffixal -a- (24b). The s-aorist has the aspectual exponent -sa- (24c) or -s- (24d).

Perfect and aorist exponents may never co-occur with one another, and with any of present system exponents. The affixes occurring in the present system indeed never occur with affixes of the other two systems in the same verb (25).

Accordingly, potential forms showing affixes of the two systems such as ^✘bhi⟨n⟩d-sa-, ^✘bhi⟨n⟩d-a- (25a); ^✘kr̥-nu-sa-, ^✘kr̥-nu-a- (25b); ^✘vr̥k-a-sa-, ^✘jan-a-sa- (25c, d), etc., are not attested. It follows that the present, aorist, and perfect suffixes appear to compete for the same morpho-syntactic position. Given their semantics, we suggest that all of them are exponents inserted under Asp⁰ as in (26):

One can therefore propose that Asp is realized by (27).²⁵ Notice that -a- (i.e., what the tradition labels as the “thematic vowel”) is here treated as the elsewhere exponent of the functional Asp (27f).^26,27

It is worth underlining that, in this analysis, Ø is treated as the default exponent for any functional node lacking any more specified exponence rule (27g), as suggested by its wide distribution across functional heads (Voice, Asp, T, and Mood) and featural configurations. As such, (27g) can be considered an operation of default assignment of null exponence and not a Vocabulary Insertion rule.

3.1 A sketch of the AG verbal system

AG verbal system is organized similarly to Sanskrit, i.e. in the present, aorist, and perfect systems. Comparison of the Sanskrit forms in (1) with the AG counterparts (44) leads one to plausibly assume the common morpho-syntactic structure, repeated in (45) below, in which the VP-shell V⁰ and Asp⁰ nodes on one side, and the T⁰ and Mood⁰ nodes on the other are bundled in two separate nodes as resulting from the null pruning operation proposed above (section. 2.2).³⁴

As we did above for Sanskrit, in the following section, we aim at providing independent evidence for the V⁰+Asp⁰ node bundle by analyzing the morpho-syntactic behavior of all root-adjacent morphemes and their realizational patterns.

3.2 The functional status of the AG root-adjacent exponence

Similar to Sanskrit, the AG present system consists of two conjugations: the thematic and athematic conjugations. The thematic conjugation displays a mid vowel in the present forms, right between the root and the endings. This is often called “thematic vowel”. It surfaces as -o- in front of nasal-initial endings (46b), and as -e- elsewhere (46a). The athematic conjugation lacks such vowel (47), regardless of the shape of the rightward suffix.

Verbs are classified in classes depending on the form of the verb in the present system (cf. Smyth 1956; Sihler 1994).³⁵

As in Sanskrit, the difference between thematic and athematic conjugation disappears in the aorist and perfect systems. The aorist may have different formations. The athematic root aorist does not have an overt aspectual exponent, and it is essentially conveyed by the full grade of the root (“aorist III”; 48a). The “thematic” aorist is formed with the suffix -e/o- and the zero grade of the root (“aorist II”; 48b). The sigmatic aorist is formed with the suffix -sa- (“aorist I/weak”; 48c).

The perfect is mainly characterized by root partial reduplication and may be realized as -ka-, usually usually associated with full grade—also called normal grade— of the root (49a); or -a-, see (49b) where the root vowel undergoes backing and rounding, the so called o-grade, (leip- > loip-).

If we look at the overall distribution of the suffixal vowel -e/o-, we notice that in the present system it is complementary on one side to that of suffixes such as -(a)no/e- (e.g., amart-ano-men ‘we make mistakes’), *-ye/o- (which surfaces in a number of forms: e.g., *tak-ye/o- > tasse/o-: tass-o-men ‘we order’), and -ske/o- (e.g., eur-isko-men ‘we find’);³⁷ and on the other to that of other suffixes such as -nu- (e.g., deik-nu-men ‘we show’, but ^✘deik-nu-o-men³⁸) and -Ø- (e.g., ti-the-Ø-men ‘we put’, but ^✘ti-the-o-men). This distribution is similar to what we have already seen for Sanskrit, with a crucial difference: in AG zero grade was specific to the root only (rather than being applicable to suffixes too, as in Sanskrit). As a consequence, AG did not adopt the Sanskrit distinction between “thematic” and “athematic”, which essential hinges on the applicability of zero grade onto a given suffix. Rather, “thematicity” in AG ended up referring to the mere presence of the “thematic” vowel -e/o- in the root-adjacent exponent. The parallelism between AG, Sanskrit, and PIE root-attaching affixation is shown in Table 5 below.

In addition, all these suffixes are in complementary distribution with morphemes of the aorist and the perfect systems we have seen above: -(s)a- (50a) and -(k)a- (50b).³⁹

The semantics of all of these suffixes suggests that, as in Sanskrit, they are inserted under the Asp⁰ node.

4.1 A sketch of the Latin verbal system

Latin is traditionally described as having four conjugations characterized by root-adjacent vocalic suffixes: -ā-, -ē-, -ĕ-, and -ī-, to which we add a fifth conjugation characterized by the vowel -ĭ- (Embick & Halle 2004: 18).⁴⁸ These suffixes were specifically referred to as tvs by the grammatical tradition as vowels consistently occurring after the root, and before any other functional (aspect, mood, agreement) morphemes (56). Thematic vowels may however be absent in specific morphological categories such as irregular perfects or past participles, and in specific verbs (e.g., su-mus be-1PL ‘we are’).

The Latin verbal system is characterized by the aspectual opposition between imperfective (īnfectum) and perfective forms (perfectum), which is historically due to the collapse of the Aorist and Perfect of PIE—an issue irrelevant here. Tense distinctions (present, past, future) are found in both aspectual categories in the indicative mood as shown in (57). (For simplicity, in all examples below, we will only use the 1pl.act forms of the indicative mood; see Gildersleeve & Lodge 2000, or any other standard Latin grammar for the complete paradigms. Also, the features used below have merely descriptive purposes.)

A basic segmentation of Latin verbal forms is given in (58) below (see Calabrese to appear a for further details).

In what follows, we provide an in-depth analysis of the distribution of root-adjacent exponence in the Latin verbal system. In particular, we will first engage with the debate on Latin tvs, and argue for their fundamentally ornamental – as opposed to functional – nature on logical and theoretical grounds. We will then delve into discussing the origins of ornamentality in the Latin verbal system.

4.2 The ornamental behavior of Latin root-adjacent exponence

In his analysis of Latin morphology, Aronoff (1994) discusses the double nature of Latin tvs. On the one hand, they are legitimate morphemes, as being easily distinguishable from the nearby root or the derivational suffix (on the left side), and the inflectional suffixes (on right side). Moreover, they are in complementary distribution with one another and consistently occur within the same verb paradigm. On the other hand, according to Aronoff, they seem to carry no syntactico-semantic meaning to the word, and are considered to be pure “ornamental” elements. Much of the subsequent literature on Latin, especially in DM, has assumed Aronoff’s analysis and proposed that Latin tvs are the exponents of ornamental structural entities inserted in the post-syntactic component. Aronoff’s analysis has been recently questioned by Bertocci (2017) and Bertocci & Pinzin (2021), who investigated the Latin root-adjacent vocalic pieces trying to assign them morpho-syntactic and semantic properties. They distinguish between the formatives -ā-, -ī-, which are found both in the īnfectum and the perfectum, and the formatives -ē-, -ĕ-, -ĭ-, which are found only in the īnfectum. They argue that ā-conjugation verbs have a causative/agentive semantic Aktionsart due to the piece -ā- as a productive verbalizer. This nature is shown by the residual causativizing function in couples like stative liqu-ē-re ‘to be liquid’ vs. causative liqu-ā-re ‘to make liquid’; plac-ē-re ‘to please’ vs. plac-ā-re ‘to appease’; and sed-ē-re ‘to sit’ vs. sed-ā-re ‘to settle’. From our point of view, however, the presence of sporadic word pairs does not imply the presence of a systematic word-formation process. Furthermore, as also acknowledged by Bertocci & Pinzin, such characterization is not referrable to the set of the primary ā-verbs, which therefore cannot be derived from attested nouns, adjectives or verbs (De Vaan 2012): transitive activity verbs (e.g., amāre ‘to (make) love’, arāre ‘to plough’, dolāre ‘to hew’), as well as intransitives (e.g., fāri ‘to speak’ and volāre ‘to fly’), and unaccusative non-agentive verbs (e.g., cubāre ‘lie down’ and flagrāre ‘to glow’). The diverse nature of the subsets suggests that no common syntactic or semantic features can plausibly characterize the distribution of /-ā-/. The same holds for the ī-verbs: while some of them are indeed clearly related to a nominal/adjectival basis (e.g. finio ‘I limit’ < finis ‘limit’), many others are not (e.g., dormio ‘I sleep, venio ‘I come’). As before, finding a common semantic feature or structural property is not possible in this case. The ā- and ī-verbs are a rather composite class where no clear common features can be established, both in terms of morpho-syntactic and semantic properties. A further problem is that, whatever their etymological or derivational bases, all of the ā- and ī-verbs can undergo suppression or demotion of agentivity in passive constructions: am-ā-tur ‘s/he is loved’, liqu-ā-tur ‘s/he is made liquid’, plac-ā-tur ‘s/he is appeased’, sed-ā-tur ‘s/he is settled’. Despite this, they preserve the -ā- and -ī- markers. This shows that these markers cannot be associated with “agentivity”. In this sense, Aronoff is right in proposing that root-adjacent pieces such -ā- or -ī- do not carry any syntactico-semantic meaning to the word. Furthermore, as shown by Aronoff, vocalic pieces such as -ā- and -ī- can cooccur with iterative (59a), intensive (59b), and causative (59c) markers. If -ā- and -ī- carried a causative/agentive meaning, they should be incompatible with the markers below. Again, in these forms, they appear to be semantically empty, redundant morphological elements.

As for tvs -ē- and -ĕ-, Bertocci & Pinzin propose that they realize functional features related to the higher actional/aspectual domain, since they are consistently absent in perfective forms as shown in (60)–(62).

However, also in this case, as for -ā- and -ī-, it is difficult to find common morpho-syntactic or semantic properties. First, ē-, ĭ-, and ĕ-verbs cannot be characterized in terms of common special aktionsart properties. As a matter of fact, the same aktionsart properties can be found across Latin verbal conjugations, and it has never been possible to classify Latin conjugation classes in semantic terms. Second, the aktionsart properties of the verbs in question do not change with the change of aspectual features, as Bertocci and Pinzin observe. If the root adjacent vocalic pieces are markers of these properties, it is unclear why they disappear when these properties are still preserved. Furthermore, and most importantly, these formatives cannot be characterized as aspectual markers. To show this, let us consider the imperfect indicative forms of the ē-verb mon-ē-re ‘to scold’ (63), the ĕ-verb leg-ĕ-re ‘to read’ (64), and the ĭ-verb cap-ĕ-re ‘to take’ (65).

The distribution of the exponent -b- in such forms requires the feature [–PFV] to appear in the /insertion site of -b- along with the tense feature [+PAST]; therefore, Asp⁰ must be fused with T⁰. This is impossible if we assume that tvs -ē- (63), -ĕ- (64), and -ĭ- (65) are the exponents of [–PFV] Asp⁰. If we assumed this, -b- could only expone T⁰, which is not consistent with the previous requirement. Therefore, the hypothesis that the vocalic pieces -ĕ- ē- and -ĭ- are exponents of [–PFV] is difficult to maintain. Furthermore, root-adjacent, distributionally indistinguishable, vocalic pieces end up being analyzed as having two different statuses: as verbalizers (or as ornamental pieces related to verbalizers) in the case of -ā- or -ī- and as aspectual elements in the case of -ē-, - ĕ-, and -ĭ-. But this violates the analytic criterion of pattern congruity: an identical distribution should entail the same structural position provided that there is no other evidence, which in this case is missing. In other words, the hypothetical structures where -ĕ-, -ē-, and -ĭ- are exponents of [–PFV]_Asp are simply inconsistent with the general patterning we see in the other conjugations (see Calabrese to appear a for further arguments on the ornamentality of such vocalic pieces). If this analysis is on the right track, one must conclude that all root-adjacent vocalic pieces are the same: they are devoid of syntactic and semantic properties and are, therefore, ornamental.