Romance Root Suppletion and Cumulative Exponence: Fusion, Pruning, Spanning

Abstract

This paper discusses verbal stem allomorphy in Romance within the framework of Distributed Morphology (DM). We will present several technical instruments provided by the framework, applying them to an analysis of Romance verbal forms, with a particular focus on stem suppletion with the verb go. We conclude that the best solution to the problem of form–function discrepancies, as they appear in suppletion (but not only), is a spanning approach. This approach operates at Vocabulary Insertion only, without any need for the assumption of further, often critically discussed, morphological processes, such as fusion or pruning.

1. Introduction

The main aim of this research paper is the analysis of irregular verbal inflection in Romance within the framework of Distributed Morphology (DM; cf. Halle and Marantz 1993). In DM, morphosyntactic processes derive hierarchical structures combining roots and functional elements, and the outcome of syntax is then mapped onto morphophonological realizations (=vocabulary items) by a process called Vocabulary Insertion (VI). We will focus on verbal inflection, in particular on the forms of Romance go, in order to discuss possible problems in the context of VI where the mapping process includes allomorphy and cumulative exponence, as is the case with root suppletion.

Romance go-suppletion is one of those cases where locality effects and intricate morphological variation are found in the verbal domain, since there are cases where the suppletive pattern is dependent on the syntactic context (e.g., agreement, as in French, Italian; “contextual suppletion”, Hippisley et al. 2004) and cases where it is dependent on functional features inherent to the verb (e.g., TAM, as in Spanish; “categorial suppletion”, Veselinova 2006, 2017). In order to explain the differences in the patterns and structural interpretation of go-suppletion, we will explore different theoretical approaches that offer tools for the modeling of suppletion within DM and discuss their pros and cons. In line with Haugen and Siddiqi (2013, 2016), we argue that the assumption of nonterminal VI—i.e., VI that targets units taller than one syntactic terminal element—is the best option for explaining root allomorphy, since form alternation is, as a general rule, accompanied by a reduction in segments (cf. e.g., Vanden Wyngaerd 2018).

The overall aim is to investigate the cumulative exponence of the verbal forms concerned and the contextual conditions for root suppletion. It is commonly assumed that root allomorphy can only be triggered by elements that are (linearly) adjacent to each other. The question is which of the approaches available accurately predicts the distribution of forms without being too unrestrictive as far as context conditions are concerned. Our research questions in particular are:

• Research Question 1 (RQ1). How can TAM-triggered root suppletion be modeled?
• Research Question 2 (RQ2). How can suppletion triggered by φ-features be implemented?

The paper is structured as follows: in Section 2, we report on materials and methods used. In Section 2.1, we present the data, starting from the paradigm1 of go in Latin, extending the view to Romance. In Section 2.2, we introduce the DM framework and discuss several methodologically possible analyses for cumulative exponence and contextual conditions for root allomorphy (Section 2.2.1), embedded in different stages of theoretical thinking in DM, namely fusion (Section 2.2.2), pruning and allomorphic closeness (Section 2.2.3), and spanning (Section 2.2.4). In Section 3, we set out our analyses for several instances of Romance go within the framework of DM, presenting our results for each language in single subsections (Section 3.1 Spanish, Section 3.2 Italian, Section 3.3 French). In Section 4, the advantages of a spanning approach to Romance verbal inflection are highlighted, in particular insofar as this approach can model and maybe also explain the differences and properties in common of the Romance languages discussed.

2. Materials and Analyses: The Romance Data and the Instruments of DM

In this core section of the paper, we will first present an overview on the morphological shape of the movement verb go and its development from Latin to Romance; we will also discuss different types of suppletion (Section 2.1), which will serve as the basis of our analysis in Section 3. In Section 2.2, after an introductory part on cumulative exponence and contextual conditions (Section 2.2.1), we discuss three possible approaches to the problems posed by contextual suppletion within the framework of DM, namely fusion (Section 2.2.2), pruning (Section 2.2.3), and spanning (Section 2.2.4).

2.1. The Data: Go from Latin to Romance

The paradigm of the Latin verb go is quite regular (cf.

Table 1. Latin ī-re ‘(to) go‘: present and perfect system (cf. Leumann et al. [1965] 1972, [1926–1928] 1977).

Tense/Person	1sg	2sg	3sg	1pl	2pl	3pl
present indicative	e-ō	ī-s	i-t	ī-mus	ī-tis	e-unt
imperfect indicative	ī-bam	ī-bās	ī-bat	ī-bāmus	ī-bātis	ī-bant
future	ī-bo	ī-bis	ī-bit	ī-bimus	ī-bitis	ī-bunt
present subjunctive	e-am	e-ās	e-at	e-āmus	e-ātis	e-ant
imperfect subjunctive	ī-rem	ī-rēs	ī-ret	ī-rēmus	ī-rētis	ī-rent
imperative I		ī			ī-te
imperative II		ī-tō	ī-tō		ī-tōte	e-untō
perfect	i-(v)ī	ī-(vi)stī	i-(v)it	i-imus	ī-stis	i-ērunt
future II	i-erō	i-eris	i-erit	i-erimus	i-eritis	i-erint
pluperfect	i-eram	i-erās	i-erat	i-erāmus	i-erātis	i-erant
perfect subjunctive	i-erim	i-eris	i-erit	i-erimus	i-retis	i-erint
pluperfect subjunctive	ī-ssem	ī-ssēs	ī-sset	ī-ssēmus	ī-ssētis	ī-ssent

) despite the following facts: First, Latin īre stems from one of the athematic verbs in Proto-Indo-European, i.e., many forms have no theme vowel (Th) between the stem and the verbal ending; and second, the lack of the Th is responsible for (stem) allomorphy (i.e., the so-called Wurzelablaut or root apophony; i.e., a sound change that conveys grammatical information as e.g., Engl. sing in the present tense vs. sang in the imperfect; cf. Leumann et al. ([1926–1928] 1977, p. 522ff) for the original distribution of the Ablaut of īre). This stem allomorphy can be grasped by the following rule: In the present system the stem is /i:/ before consonant and /e/ before vowel (except e, see the active participle īens). Before /t/, as in the third-person singular, this long I at some stage is shortened (it, but īt would be the older form, cf. Julia 2016, p. 62). In the perfect system it is /i/, as general rule. Forms with long [i:] in the perfect system are the result of phonological adjustment (e.g., [i:] < [i] + [i]).

In Romance instead, all languages and varieties have started with the loss of verbal forms of Lat. īre or have even completely lost it, substituting it by other roots. Verbal forms of Romance go stem thus from different Latin verbs: īre ‘to go’, ambīre/*ambitāre ‘to go around’ or ambulāre/amblāre ‘to go, to walk’ (>*andāre ‘to go around’, *allāre ‘to walk’), vādere/*vādēre ‘to go’, venīre ‘to come’ and esse ‘to be’.2 It is important to note that these Latin verbs belong to different Latin conjugation classes (CC) and some verbs are thematic (e.g., ambulāre > *allāre), whereas others count as athematic (e.g., īre and vādere). This is a fact that is still reflected in the modern Romance languages, as we will show in Section 3.

There is no Romance variety that retained the full paradigm of forms of īre. However, the Romance varieties reached different solutions. As can be seen from

Table 2. Different data samples for Romance go-suppletion3.

Language	TAM	Verbal Forms						Latin Etyma
		1sg	2sg	3sg	1pl	2pl	3pl
Spanish	present	voy	vas	va	vamos	vais	van	vādere/*vādēre
	imperfect	iba	ibas	iba	íbamos	ibais	iban	īre (elsewhere)
	future	iré	irás	irá	iremos	iréis	irán	īre
	indefinido	fui	fuiste	fue	fuimos	fuisteis	fueron	esse
	imperative		ve			id		vādere/*vādēre + īre
Italian	present	vado	vai	va	andiamo	andate	vanno	vādere/*vādēre + *andāre
	imperfect	andavo	andavi	andava	andavamo	andavate	andavano	*andāre (elsewhere)
	future	andrò	andrai	andrà	andremo	andrete	andranno	*andāre
	imperative		vai			andate		vādere/*vādēre + *andāre
French	present	vais	vas	va	allons	allez	vont	vādere/*vādēre + *allāre
	imperfect	allais	allais	allait	allions	alliez	allaient	*allāre (elsewhere)
	future	irais	iras	ira	irons	irez	irons	īre
	imperative		va		allez			vādere/*vādēre + *allāre
Lombard (Monza)	present	vo	vet	va	vem	andé	van	vādere/*vādēre + *andāre (elsewhere)
Old Tuscan	present	vado	vai	va	gimo	gite	vanno	vādere/*vādēre + īre (elsewhere)
Engadinish	present	veñ	vaš	va	yáin	yáivat	van	Venire+ vādere/*vādēre + īre (elsewhere)

, suppletion with go varies in Romance with respect to (a) the number of verbs that are combined in the respective suppletive paradigm—it ranges between two verbs (e.g., Standard Italian) to three verbs (e.g., in some Rhaeto-Romance varieties)—and (b) the language-specific distributions of these verbs within the suppletive paradigm.

As can be seen from this sample, suppletion is sensitive to the grammatical context. In Spanish for example, go-suppletion is TAM-sensitive (also called categorial or inherent suppletion), but not φ-sensitive4: in the present tense, va- (< athematic CC) has taken over; in the imperfect, the future and the conditional forms derived from īre (< athematic CC) are found, but in the indefinido and categories pertaining to it, forms of the verb be (Lat. esse = athematic CC) appear. This is a case of overlapping suppletion (cf. Juge 2000), i.e., two different verbs share the same stems in certain cells of the paradigm; in Spanish, overlapping suppletion is also TAM-conditioned.

In Italian, in the present tense the singular (including the imperative) and the third-person plural have the suppletive form va- (< athematic CC), whereas all other cells of the paradigm have taken over the form and- (< thematic CC) and no trace of a stem derived from īre is left; i.e., Italian go-suppletion is sensitive to φ-features and to TAM.

In French, go-suppletion is both sensitive to φ-features and to TAM. In the present tense we find the same conditions as in Italian, but contrary to Italian, French has not extended all- (< thematic CC) to all of the remaining cells of the paradigm, but still retains forms derived from īre (< athematic CC) in the future and the conditional. This is noticeable, since as general rule, the conditional and the future (at least historically) go back to the infinitive (cf. Roberts 2009 for a discussion of the traditional accounts—e.g., Benveniste 1968; Rohlfs 1969; Tekavćič 1972—and for a formal account of the grammaticalization process). This is, however, not the case synchronically in modern French where the infinitive of go is aller.

Interestingly, suppletion sensitive to φ-features is only found in the present tense in Romance. There we find also other patterns, different from Italian and French, as e.g., in the Lombard dialect of Monza, where the second plural is the only form to have an and- stem in the present tense. Old Tuscan, in the present tense, has the same pattern as Old Italian, with the difference that instead of and- forms there still are traces of īre in the first and second plural (cf. again Table 2).

Another case of φ-sensitivity is found, e.g., in Rhetoromance, here Engadinish, where we have a distribution of forms derived from īre and and- forms as in Old Tuscan, with the exception of the first-person singular, where a verbal form of the verb come, namely veñ, is found (another case of overlapping suppletion). The distribution of the sample in Table 2 can be illustrated as in Figure 1a,b:

Of course, there are far more distributional patterns of go-suppletion in the Romance varieties and languages (cf. Pomino and Remberger 2019), which we cannot represent here. Furthermore, the use of go as a lexical verb encoding movement versus the use of go as an auxiliary (as in the Spanish progressive future, cf. e.g., Gómez Torrego 1999, or the Catalan preterite, cf. e.g., Detges 2004; Oltra-Massuet 2013) must be distinguished. However, since our paper is mainly concerned with the interpretation and analysis of root allomorphy (such as suppletion) and cumulative exponence (nonsegmentability) in Romance, in what follows, we will concentrate only on Italian, Spanish and French and we will consider only the use of go as a lexical verb.

To sum up, Spanish has only TAM-sensitive suppletion, which is also called categorical (or inherent) suppletion, since it depends on categories inherent to the verb; Italian and French have additionally noncategorial (or contextual) suppletion, since φ-features are not primary categories of the verb, but contextually derived from agreement processes during the syntactic derivation (for the terminology, cf. Hippisley et al. 2004, as well as Veselinova 2006, 2017). Note again that noncategorial (contextual) suppletion, i.e., suppletion conditioned by φ occurs only in the present tense. Figure 2 shows the corresponding distributional patterns for go-suppletion:

The three languages, however, have in common that the respective paradigms of go contain roots going back to different Latin CCs, which are either thematic or athematic.

2.2. Instruments of DM for Cumulative Exponence and Contextual Conditions: Fusion, Pruning, Spanning

In DM (cf. Halle and Marantz 1993; for an overview, cf. Bobaljik 2017) a set of postsyntactic operations can alter the syntactic output before the vocabulary items (i.e., the morphophonological material) are inserted. Some of these processes are considered necessary to cope with different mismatches between syntax and morphology (cf. Haugen and Siddiqi 2016 for more details and arguments against these processes): (1) Fusion is assumed, for example, for portmanteau morphemes (e.g., French du, instead of *de le ‘of the’) and other cases where the syntactic output is more complex than the morphological formatives; (2) Fission, on the other hand, is proposed to account for more or less the opposite situation, i.e., for those cases where the form is more complex than the syntactic output, e.g., as in the Italian passato remoto in the third-person plural of the regular verbs, where the φ-node seems to be filled twice, i.e., is submitted to fission, which splits the φ-node into two positions, cf. irregular diss-e-ro ‘they said’ (where we have √say-Th-φ) vs. regular cant-a-ro-no ‘they sang’ (with two person–number suffixes, i.e., √say-Th-φ1-φ2) (cf. Calabrese 2015b); (3) Impoverishment is an operation that does not alter the number of syntactic terminal nodes but only the features contained therein, and has been used to explain, for instance, repairs of Romance Clitic Clusters (e.g., Spanish *le_3.sg.dat lo_3.sg.acc where the dative clitic le surfaces as (spurious) se, i.e., se lo) (cf. Bonet 1991); (4) Pruning is another process introduced to account for locality effects (on allomorphy): nodes that are not exponed with phonological material are removed from the structure directly after VI with a direct effect on (linear) adjacency (Embick 2010). In the following subsections we will exemplify and critically discuss these processes along Romance verbal inflection.

First, we want to point out that the assumption of these processes has been met with some criticism on the basis of the principle that (given equal explanatory adequacy) a smaller number of processes is preferable to a larger number (see Ockham’s razor); several efforts have thus been made to eliminate these processes (cf. Haugen and Siddiqi’s 2016 arguments in favor of a Vocabulary-Insertion-only programme for nonlexicalist realizational models of morphology). Fission can be avoided, for example, by allowing insertion into one and the same syntactic node more than once (cf. Halle 1997); impoverishment is not necessary, if insertion of zero exponents is allowed (cf. Trommer 1999), and fusion as well as pruning can be avoided, if VI is not limited to terminal elements, as in the nonterminal spellout of Nanosyntax (cf. Starke 2009) and the spanning approach of Svenonius (2012), among others. In what follows, we will not only show that fusion and pruning can be avoided, but also that they are not adequate for the analysis of suppletion in Romance. Before that we will introduce basic aspects of contextual conditions of allomorphy in the framework of DM.

2.2.1. Cumulative Exponence and Contextual Conditions

After the syntactic derivation, in DM, terminal nodes are handed over to VI. Vocabulary items are form-meaning correspondences and their insertion may be contextually conditioned by adjacent terminal nodes. Any context-induced surface variation of a morpheme could thus, in principle, be explained with VI. The contextual features triggering allomorphy are either phonological or morphosyntactic depending on whether the realization of the terminal node is inward- or outward-sensitive (cf. Bobaljik 2000). This idea is illustrated in Figure 3: when it comes to the realization of the terminal X, its form may depend on the phonological features of the vocabulary item previously inserted (and/or on following grammatical features). This is a case of inward context sensitivity, since the realization of X depends on the phonological shape of the root. In contrast, the realization of the root is outward sensitive. That is, the contextual feature [F1] of X, a grammatical feature, may condition root allomorphy (or suppletion) in (b).

In this sense, suppletion of verbal roots is a case of outward context sensitivity, because the features triggering suppletion are the agreement features (in cases of noncategorial suppletion) or the TAM-features encoded in T° (in case of categorial suppletion). Yet are these features locally close enough to the root in order to trigger allomorphy? In other words, the question that arises is how we can correctly predict contextually conditioned allomorphy and restrict the conditions for the occurrence of suppletive roots. In particular, how far apart can the √root and the functional head encoding TAM or φ be to have a conditioning effect? Can T°, for example, and the grammatical features it contains trigger root allomorphy/suppletion, although it is not a direct sister to √root?

There are several proposals in the literature to grasp the locality conditions under which contextual allomorphy is possible. It is stated, for example, that two terminal nodes, α and β, are local enough if (a) no XP intervenes (Bobaljik 2012; Bobaljik and Harley 2017); (b) no overt node intervenes (Embick and Halle 2005; Embick 2010; Calabrese 2015a); (c) they belong to the same phase (Moskal 2015; Embick 2010); or (d) if they form a span (Svenonius 2012, 2016; Merchant 2015). In what follows, we will discuss some of these possible approaches to the problem at issue and show which additional postsyntactic processes are to be assumed. We will have a closer look at three processes, namely Fusion (Y can trigger root allomorphy only if it fuses with X and appears thus closer to the root), Pruning (Y triggers root allomorphy only if X is nonovert and is thus deleted, i.e., pruned) and Spanning (Y triggers root allomorphy only if it is in an adjacent span).

2.2.2. Fusion

As said before, fusion is assumed for those cases where the syntactic output is more complex than the morphological formatives. Depending on the theoretical assumptions, the present tense forms of Romance are one example for this kind of mismatch between (morpho)syntactic structure and (morpho)phonological realization. We will therefore begin with the syntactic derivation and the input nodes it provides for VI of a finite verbal form in Spanish and Italian. After the syntactic derivation where movement of the root to v° and of the √-v°-complex to T° takes place, a complex T-node serves as the input structure to VI, cf. Figure 4a. In line with Oltra-Massuet (1999), we may assume that in Spanish and Italian there are, at least, two well-formedness conditions to be fulfilled before VI: As general rule, the φ-features that T° has acquired through person–number agreement with the subject in syntax receive their own slot for VI in Romance. In other words, finite T° postsyntactically requires an appropriate φ-feature position. Second, as has been claimed by Oltra-Massuet (1999) for Catalan, functional categories need to be equipped with a position for theme vowels. She proposes that a theme vowel position will be added to all syntactic functional heads, in our case v° and T°.6 That v° and the theme vowel are different VI can be seen in derivative forms such as Sp. atomizabamos or It. atomizzavamo ‘we atomized’, where after the realization of the root, the derivational suffix Sp. -iz- or It. -izz- is the vocabulary item for v° followed by the Th a, which marks the first CC.7

As a result of applying the mentioned well-formedness conditions, the syntactic output structure consisting of three terminal nodes (cf. Figure 4a) is augmented to a structure of six terminal nodes (cf. Figure 4b) to be realized by VI.

Figure 5 illustrates the realization of this structure for Spanish and Italian (note that French behaves differently with respect to theme vowels as we will show in Section 3; cf. also note 7). In the derived forms in the imperfect (cf. Figure 5a), morphological complexity stands in direct relation to the syntactic-semantic features of the respective forms (Oltra-Massuet 1999; Arregi 2000). Each input node is realized by a vocabulary item: In addition to the already mentioned realization of the root, v° and the theme vowel, imperfect T° is realized by Sp. -b- or It. -v-, again followed by a theme vowel selected by the vocabulary item for T°, and the φ-position is filled by Sp. -mos or It. -mo, since the agreement gave result to first-person plural features for φ. In the forms without a derivative affix, little v° is empty.

In the present tense, however, different from most other tenses, T° has no exponent, nor is there a theme vowel for T° (cf. Figure 5b).8 There is thus a mismatch between the number of terminal nodes and the available phonological material for the realization of this structure that has to be explained somehow.

Oltra-Massuet (1999) argues that this mismatch is not accidental, but has to do with the semantically unmarked tense feature. She proposes that heads containing only unmarked features are subject to fusion. She assumes that in the present tense, T° fuses with φ since it encodes a semantically unmarked tense feature, i.e., a tense feature that is also morphophonologically never realized. Note that in a certain sense, in the present tense, fusion reverses the well-formedness condition that consists in the addition of a separate φ-position proper to T°. After fusion of the present tense T° (which has no exponent) with the added φ-position, the φ-features appear under the fused T°/φ-head (cf. Figure 6).

Before we question the necessity of such an operation, let us come back to suppletion: As mentioned before, noncategorial (or contextual) suppletion is restricted to the present tense in Romance, and it is exactly in this tense where we find fusion, according to Oltra-Massuet (1999). We could be tempted to assume, thus, that suppletion/stem allomorphy is possible in this context because φ fuses with T°, becomes a sister to v° and thus can influence the selection of the VI for the root. In this sense, in the present tense, φ can locally condition the selection of different vocabulary items for the root.9

But, strictly speaking, even after fusion the φ-features are neither structurally nor (in most cases) linearly adjacent to the root, since there is always v° and the Th-position intervening. The only locality condition that is met is that there is no intervening XP (cf. Bobaljik 2012; Bobaljik and Harley 2017). What is more, why does φ trigger allomorphy only in the third-person plural and in the singular, but not in the first and second plural? In the fusion approach this would be mere coincidence, since fusion takes place in all persons. And finally, fusion is per se rejected by many linguists. First, as said before, in Oltra-Massuet’s analysis, fusion just reverses the well-formedness condition, i.e., a node is added to T° to ensure that some features of T° (the φ-features, which, as said above, are not primarily inherent to T° but derived from agreement processes) are realized independently from T° and shortly afterwards this position is fused with T° and they appear again where they originally were encoded. Would it not be more efficient to simply assume that in contrast to other tenses, in the present tense T° does not receive a separate slot for the realization of its agreement features? Second, a more general problem with the fusion approach is that it remains unclear what it is triggered by and how it is restricted. Oltra-Massuet (1999) also assumes fusion in other tenses where the semantic content of tense is certainly quite specific. In sum, fusion does not seem the appropriate means to alter the closeness of φ to other elements and to explain how root allomorphy can be triggered by φ-features.

2.2.3. Pruning

Pruning is a process introduced by Embick (2010) to account for certain locality effects on allomorphy: nodes that are not exponed with phonological material are removed from the structure after Vocabulary Insertion with a direct effect on linear adjacency (Embick 2010). To understand the necessity of this process, we will first discuss Embick’s assumption with respect to the locality conditions under which contextual allomorphy is possible after all. He assumes that allomorphy is restricted in the following two ways:

• Phase domains: Morphemes can interact for allomorphy etc. only when they are in the same phase domain.
• Linear adjacency: Morphemes can interact for allomorphy only when they are immediately linearly adjacent, i.e., concatenated: X⁀Y (Embick 2010).

As for (1), domains correspond to phases: allomorphy can only occur when triggered in the same phase, i.e., in a locality condition. Even if linearly adjacent, elements belonging to different phases cannot trigger allomorphy. This is illustrated by the English examples destruction and destroying in Figure 7 (cf. Ingason and Sigurðsson 2015, p. 4; also Marantz 2013; Embick 2014):

In the case of destruction, the root and the nominalizer n° are in the same domain and thus they can trigger allomorphy on each other (cf. destruction not *destroytion). However, in the gerund, the root and the nominalizer are not in the same domain, since the phase is closed with the phase head v°. Therefore, root and nominalizer cannot condition allomorphy on each other—see the gerund destroying (not the hypothetical allomorphic form *destructing or something similar)—since here v° is an intervening phase head.

As for (2), according to Embick, allomorphy can only be triggered by linearly adjacent elements. Yet, linear adjacency, if not given, can be achieved through pruning: VI takes place at the linearized structure and nodes that are not exponed with phonological material are removed from the structure with a direct effect on linear adjacency (Embick 2010):

3.

Pruning rule: √root⁀[x,-Ø], [x,-Ø]⁀Y → √root⁀Y (‘if x is not realized, it is pruned so that √root and Y become linearly adjacent to each other’)

At first glance, this proposal seems to work out for Romance suppletion, as illustrated with the French example allons in

Table 3. Derivation French allons based on Embick’s assumptions (simplified structure).

		allons
(a)	Linearization	√root⁀[v], [v]⁀T[pres], T[pres]⁀φ[1pl]
(b)	VI at v°	√root⁀[v,-Ø], [v,-Ø]⁀T[pres], T[pres]⁀φ[1pl]
(c)	Pruning	√root⁀[v,-Ø], [v,-Ø]⁀T[pres], T[pres]⁀φ[1pl] → √root⁀T[pres], T[pres]⁀φ[1pl]
(d)	VI at T°	√root⁀T[pres,-Ø], T[pres-Ø]⁀φ[1pl]
(e)	Pruning	√root⁀T[pres,-Ø], T[pres,-Ø]⁀φ[1pl] → √root⁀φ[1pl]
(f)	VI at √root	[al-]⁀φ[1pl]
(g)	VI at φ	[al-]⁀[-ɔ͂z]]
		allons [alɔ͂z]

: Since v° and T° (and their possible theme vowels which we have omitted here for sake of simplicity, cf. also note 7) are not exponed in the French present-tense forms (cf. Table 3—b,d), they are pruned (cf. Table 3—c,e) and thus φ results to be linearly adjacent to the root (cf. Table 3—f), making φ-triggered root allomorphy possible, according to Embick (2010).

In sum, to make Embick’s proposal work, instead of a fusion rule one has to postulate a pruning rule. However, also here we encounter several problems, since there are additional claims to be made: Note that it is standardly assumed that VI starts in the most embedded element, i.e., the √root. But in Embick’s approach, v° (and also T°) has to be subject to VI and to pruning before the √root is phonologically realized, since otherwise, φ could no longer trigger allomorphy on the √root (once it is realized). Note, furthermore, that grammatical features are no longer available after VI. Therefore, the √root must be subject to VI before φ is realized since the φ-features must still be morphologically present in order to trigger the correct realization of the √root. In sum, the postsyntactic derivation must follow several ordered steps which appear to be somehow circular (see also below).

If we look at the Italian present tense forms in

Table 4. Present tense and imperfect of go in Italian.

	1st sg	2nd sg	3rd sg	1st pl	2nd pl	3rd pl
pres.	va-do	va-i	va	and-ia-mo	and-a-te	va-nno
impf.	and-a-v-o	and-a-v-i	and-a-v-a	and-a-v-a-mo	and-a-v-a-te	and-a-v-a-no

, one could think that the first-person plural and second-person plural trigger allomorphy on the root since all the other forms are based on va-. In Embick’s view this would not be possible, because in these cases φ and the √root are not linearly adjacent to each other: -ia- and -a- intervene between them (see also the imperfect forms in Table 4). Yet one could also assume that and- is the default realization for √go (as it is; see the other tenses) and that φ triggers allomorphy on the root in the first-, second- and third-person singular and the third-person plural, where √go and φ are linearly adjacent to each other. Note that in these va-based forms there is no intervening exponent and φ and the √root are linearly adjacent. However, singular and third-person plural do not represent a natural class (for a solution to this problem, see Section 3).

But the pruning process itself has to be likewise regarded with a critical eye: Let us have a look at the single steps for the realization of Italian vai vs. andate. As you can see in

Table 5. Processes of pruning in single steps.

		va-i	and-a-te
(a)	Linear.:	√go◠v, v◠Th, Th◠T, T◠Th, Th◠φ[2sg]	√go◠v, v◠Th, Th◠T, T◠Th, Th◠φ[2pl]
(b)	VI at v:	√go◠v:Ø, v:Ø◠Th, Th◠T, T◠Th, Th◠φ[2sg]	√go◠v:Ø, v:Ø◠Th, Th◠T, T◠Th, Th◠φ[2pl]
(c)	Pruning:	√go◠Th, Th◠T, T◠Th, Th◠φ[2sg]	√go◠Th, Th◠T, T◠Th, Th◠φ[2pl]
(d)	VI at Th:	√go◠Th:Ø, Th:Ø◠T, T◠Th, Th◠φ[2sg]	√go◠Th:a, Th:a◠T, T◠Th, Th◠φ[2pl]
(e)	Pruning:	√go⁀T, T⁀Th, Th⁀φ[2sg]	---
(f)	VI at T:	√go⁀T:Ø, T:Ø⁀Th, Th⁀φ[2sg]	√go◠Th:a, Th:a◠T:Ø, T:Ø◠Th, Th◠φ[2pl]
(g)	Pruning:	√go⁀Th, Th⁀φ[2sg]	√go◠Th:a, Th:a◠Th, Th◠φ[2pl]
(h)	VI at Th:	√go⁀Th:Ø, Th:Ø⁀φ[2sg]	√go◠Th:a, Th:a◠Th:Ø, Th:Ø◠φ[2pl]
(i)	Pruning:	√go⁀φ[2sg]	√go◠Th:a, Th:a◠φ[2pl]
(j)	VI at √go:	√go: va⁀φ[2sg]	√go: and◠Th:a, Th:a◠φ[2pl]
(k)	VI at φ:	√go: va⁀φ[2sg]: i	√go: and◠Th:a, Th:a◠φ[2pl]: te
		[vai̯]	[andate]

, the form vai is easily derivable, since pruning allows the root to be adjacent to the φ-features (see the line highlighted in bold) and thus φ can condition the selection of va-. However, with andate, at a certain point, there is a theme vowel (highlighted in bold in the table), which disallows pruning. That is, the selection of the allomorphic root and- cannot be conditioned by φ. It is, indeed, the default (elsewhere condition). However, why is v°-Th realized by /a/ in one case and not in the other? When it comes to the realization of the theme vowel (cf. line (d) in Table 5), there are only two elements that can impinge on its realization: T° and √go. T° can be discarded immediately, since vai and andate have the same TAM-features. One could thus think that the phonological shape of the root triggers the respective realization of Th: Th is realized as Ø when preceded by /va/ and by /a/ when preceded by /and/. Yet, VI at √root is one of the latest steps (cf. line (j)) and is thus unable to influence the selection of the theme vowel.

A solution to this dilemma is proposed by Calabrese (2019): He claims that suppletion, in the present tense, is not conditioned by the φ-features, but rather by the present-tense feature, and that the apparent φ-sensitivity is a “secondary” effect of an Impoverishment rule (cf. Halle 1997). To make his idea work, he introduces two kinds of pruning rules: a pruning rule that operates before VI (=Pre-VI-pruning) and another one which operates, as assumed by Embick (2010), after VI. The Pre-VI-pruning is not sensitive to the phonological null status of the pruned category, it is rather triggered by a diacritic [+suppletive] encoded in some special roots (e.g., √go[+suppl]). Pre-VI-pruning applies, for example, with vai, leading to a structure where v° and its Th-position are simply not present for VI. The nonappearance of the va-forms for the first- and second-person plural is due to an additional impoverishment rule: in these contexts, the diacritic [+suppletive] is deleted. As a consequence, Pre-VI-pruning cannot apply (Calabrese 2019). However, how the present-tense forms andiamo and andate are derived is not explained by him. Furthermore, there is no explanation for the observations that his special pruning is obviously restricted to the present tense. In addition, this approach is very technical, and one remains left with the impression that many assumptions are just restating the facts. In particular, the assumption of a diacritic and its deletion seem problematic to us. Another problem, among others, is that in the case of pruning nodes must be somehow “foreseen” by VI, since it is the realization of the root that tells us whether a verb is thematic (/and/) or athematic (/va/), but the realization of the root is conditioned by φ (which is preceded or not by a Th).

However, Calabrese (2015a, 2019) puts forward one innovative idea that we will follow in our analysis: there seems to be a link between athematicity and irregularity.

2.2.4. Spanning

In the standard version of DM, VI operates from left to right, realizing step by step each single terminal node. Yet, it has been shown in the last years that many poststyntactic operations can be avoided if VI is not limited to terminal elements (cf. e.g., Haugen and Siddiqi 2016). Within the spanning approach, VI operates over the hierarchical structure allowing to insert phonological material not just in one terminal node at a time, but also in spans of terminal nodes that are in a complementary relation with each other (Williams 2003; Svenonius 2012; Merchant 2015; recently also Bonet 2017 for Spanish).10 Figure 8 illustrates this hypothesis: The nodes of the complex head may be realized separately or in spans. <√root,X,Y> as well as <√root,X,Y,Z>, etc., represent possible spans, but *<√root,Y> and *<√root,X,Z> do not.

Now, with respect to the locality conditions on allomorphy, Merchant (2015) assumes that only span-adjacent elements can trigger allomorphy on each other (cf. (4)). In other words, nonadjacent nodes and their features cannot trigger allomorphy on each other unless the nonadjacent heads form a span that is adjacent to the node (or span) with the allomorphy-triggering condition.

4.

Span-Adjacency Hypothesis: Allomorphy is conditioned only by an adjacent span. (Merchant 2015, p. 294)

Thus, given the derivation in Figure 9, Y can trigger allomorphy on the root if, e.g., <√root,X> is realized as a span, since in this case Y would be an adjacent span. Something similar holds for Z: Z could trigger root allomorphy if, e.g., <√root,X> and <Y,Z> were realized as span.

We will apply the spanning approach to suppletion in the following Section 3 and we will identify different spans for VI for suppletive patterns of Romance √go, depending on language and syntactic context. Since the exact conditions and rules for spanning are still to be investigated, we hope to contribute to the issue by our analyses.

3. Results: Analysis and Outcome

In what follows, we will propose a spanning hypothesis for Spanish, Italian and French go-suppletion, which clearly shows the advantages of this approach from both a grammar-theoretical point of view as well as the perspective of cross-linguistic differences.

3.1. Synchronic Analysis: Spanish

As we saw in Section 2, Spanish only has categorial and no contextual suppletion, i.e., stem allomorphy is conditioned by categories inherent to the verb (and in the indefinido and related forms we find overlapping suppletion), cf.

Table 6. Spanish: present tense and imperfect of go and some other verb forms.

	1st sg	2nd sg	3rd sg	1st pl	2nd pl	3rd pl
pres.	v-o-y	va-s	va	va-mos	va-is	va-n
impf.	i-b-a	i-b-a-s	i-b-a	í-b-a-mos	i-b-a-is	i-b-a-n
pres.	cant-o	cant-a-s	cant-a	cant-a-mos	cant-á-is	cant-a-n
impf.	cant-a-b-a	cant-a-b-a-s	cant-a-b-a	cant-á-b-a-mos	cant-a-b-a-is	cant-a-b-a-n
pres.	d-o-y	‘(I) give’
pres.	s-o-y	‘(I) am’
pres.	est-o-y	‘(I) stay’
pres.	h-a-y	‘there is’	(originally: ‘have.3sg-there.cl’)

, where we find the segmented forms for go in the present tense and the imperfect.

In Table 6, the present tense form of the regular verb sing, cantar, can clearly be segmented into root, theme vowel and agreement features, i.e., √ + Th + φ; the imperfect form can be segmented in even more exponents, with the root, a theme vowel, an overt exponent for T° and a following theme vowel and the agreement features, i.e., √ + Th + T° + Th + φ. In contrast to Oltra-Massuet (1999), we will assume that T°-[present] is not subject to the well-formedness conditions that postsyntactically add positions on T° for the realization of the theme vowel and φ-features (cf. Figure 4, Section 2.2.2). The φ-features remain encoded under the T°-node and will be realized there if the language at issue has an appropriate vocabulary item for it. In contrast, in the imperfect form, T° receives a theme vowel position and a position for the realization of its φ-features.

In case of a thematic verb such as cantar, we assume that the span size of the vocabulary item /kant-/ is <√sing, v°>; the following theme vowel position is realized instead by the default theme vowel /a/ (cf. Figure 10).11 The span size of the vocabulary item /kant-/ as well as the following realization of the theme vowel is dependent on the CC features of the root. /kant-/ is specified as belonging to the first CC, which is without any doubt thematic in Spanish.

Things are quite different for the Spanish verb ir: The above segmentations are not possible for go neither in the present tense nor in the imperfect; if we compare the present tense forms of go in Table 6 with those of the regular verb cantar, we observe two differences, namely first that also in Spanish—as in Italian—in the va- forms there is no theme vowel, i.e., √ + φ; and second, that the first-person singular is extended by the ending -y [j], parallel to other very common short verbal forms in the first-person singular, such as doy, soy and estoy, as well as maybe the existential hay.13 In the imperfect, the verbal form can be segmented into more exponents; however, again there is no theme vowel following the stem, i.e., √ + T° + Th + φ. We see, thus, that both roots are athematic, just like the Latin etyma were athematic (cf. vadere with epenthetic /e/ and īre). This fact is still reflected in the modern Spanish verb.

We will argue that the differences in the behavior of suppletive go and regular cantar are due to different sizes of “spanning” of the relevant vocabulary items: being athematic, the vocabulary item realizing the root va- has a larger spanning size (cf. Figure 11) than cant-, which is thematic, and whose span size is only <√, v°>. Again, the span size depends on the respective CC features; without going into details (cf. e.g., Pomino and Remberger, forthcoming, submitted, for a detail discussion), all athematic (or mixed) CCs have a larger spanning size than thematic verbs (e.g., cant-a-r, beb-e-r, part-i-r) and may show idiosyncratic irregularities. Consider Figure 11: in the present tense, <√go, v°, Th> forms a span that is realized by /ba/ (va-) since it is context-sensitive to the following present-tense feature. T° does not receive any exponent associated with tense, but its φ-features are realized (note that as in the regular forms in the present tense, there is no exponent for the third-person singular). In the other tenses (with the exception of the indefinido, which is a case apart; cf. Figure 12), T° has a proper exponent and a separate slot for the realization of the φ-features. The relevant point is, however, that in the imperfect, for example, the span <√go, v°, Th > receives a default realization, which is athematic /i/ in Spanish.

With respect to the principles governing VI, we follow in essence the traditional proposal of Halle (1997, p. 428), cf. (5), with one addition proposed by Radkevich (2010, p. 8) necessary for integrating spans into the framework of DM), cf. (6): that is, a vocabulary item is inserted into the minimal node (=span) enclosing all the features of the vocabulary item. With this, it is guaranteed that the exponents do not realize a span too large or a too small.

5.

The Subset Principle (Halle 1997, p. 428): The phonological exponent of a vocabulary item is inserted into a morpheme […] if the item matches all or a subset of the grammatical features specified in the terminal morpheme. Insertion does not take place if the vocabulary item contains features not present in the morpheme.

6.

The Vocabulary Insertion Principle (Radkevich 2010, p. 8): The phonological exponent of a vocabulary item is inserted at the minimal node dominating all the features for which the exponent is specified.

We will illustrate this in what follows with the Spanish indefinido. As we have already mentioned, the morphosyntactic structure of these tense forms is less complex than in other tenses—not only for go but also for regular verbs; see

Table 7. Spanish: The indefinido for go and regular verbs.

	1st sg	2nd sg	3rd sg	1st pl	2nd pl	3rd pl
go	fue	fui-ste	fue	fui-mos	fui-steis	fue-ron
cantar	cant-é	cant-a-ste	cant-ó	cant-a-mos	cant-a-steis	cant-a-ron

:15

Both paradigms lack a separate exponent for T° and have a set of exponents for φ that is different from the present tense. We take this as an indicator that independent of the realization of the root and following elements, T°, Th and φ are realized as one span in the indefinido.16 This explains the exceptional shortness of these forms. Now, with respect to the VI, we assume the following: The vocabulary item /s/ for the second-person singular cannot be inserted in the indefinido, since there is a vocabulary item that is specified for more features, i.e., /ste/ for <T:indef, Th, 2sg>. This vocabulary item, however, cannot be inserted into the φ-slot (since it has features not encoded there), but it has to be inserted in those minimal nodes dominating all the features for which the vocabulary item is specified. Due to the tense feature included in its context of insertion, the vocabulary item /ste/ thus spans over T°, Th and φ. This is common to both verbs—regular/thematic cantar and irregular/athematic ir.

Yet, in the verbal forms of cantar there still is a preceding theme vowel in some of the forms, whereas for the go-forms, we again have an athematic root that spans over the theme position, as can be seen in Figure 12. Note that in this tense form, there are thus two adjacent spans, which trigger allomorphy on each other. This is, in contrast, not the case with thematic cantar since the theme vowels intervene.

Figure 12. The indefinido of Spanish go.17

Figure 12. The indefinido of Spanish go.17

To sum up, we have argued that Spanish has in essence two different verbs: thematic verbs (e.g., cantar = first conjugation, beber = second conjugation, partir = third conjugation) and athematic verbs (e.g., ir). This can be explained by assuming a fourth CC for athematic verbs. The main point is, however, that vocabulary items specified for belonging to one of the thematic conjugations have the span size <√go, v°>, whereas those specified for belonging to the athematic class have a larger span size which includes the theme vowel position, i.e., <√go, v°, Th>. With this, T° is an adjacent node and may trigger “root allomorphy” in the latter case, but not in the former one. Furthermore, also the verbal endings may exceptionally be realized as a span, as in the indefinido. This leads to shorter forms, which are, as general rule, also irregular.

The most important point is, however, that our analysis is predominantly based on the general process of VI to correctly derive the Spanish forms. We additionally need the well-formedness conditions, which add what are called ornamental morphemes (e.g., theme vowels) to the syntactic derivation, and regular phonological rules. We have thus shown in this section that integrating spanning into DM allows us to reduce the postsyntactic processes to a minimum. There is no need for fusion or different pruning rules (nor for impoverishment); root suppletion is instead explained in essence via VI. The spanning size of the vocabulary items realizing the roots is motivated by the (a)thematicity of the roots. In what follows, we will show that this approach is also valid for Italian and French, which, in contrast to Spanish, have noncategorial suppletion in the present tense.

3.2. Synchronic Analysis: Italian

In Italian, suppletion only concerns the present tense, where it is context-sensitive. In all the other verbal categories the root /and-/ is generalized, i.e., it is the default realization for √go (cf.

Table 8. Italian: present tense and imperfect of go and regular sing.

	1st sg	2nd sg	3rd sg	1st pl	2nd pl	3rd pl
pres.	va-do	va-i	va	and-ia-mo	and-a-te	va-nno
impf.	and-a-v-o	and-a-v-i	and-a-v-a	and-a-v-a-mo	and-a-v-a-te	and-a-v-a-no
pres.	cant-o	cant-i	cant-a	cant-ia-mo	cant-a-te	cant-a-no
impf.	cant-a-v-o	cant-a-v-i	cant-a-v-a	cant-a-v-a-mo	cant-a-v-a-te	cant-a-v-a-no

). Since andare is formally a first-conjugation verb, it is equipped with the theme vowel /a/ in all the cells of the paradigm where it appears, parallel to the regular verb cantare ‘to sing’.18 The only peculiarity, common to all Italian CCs, is that in the first-person plural, instead of /a/, the theme vowel is changed to /ia/ (which is a feature originally belonging to the subjunctive but then extended to the first-person plural in Florentine Tuscan) by a readjustment rule (cf. cant-ia-mo ‘we sing’, ved-ia-mo ‘we see’, fin-ia-mo ’we finish’). The suppletive forms we are interested in are the singular forms and the third-person plural in the present tense, a group of paradigmatic cells that—as mentioned before—do not belong to a natural class, whereas the first- and second-person plural would clearly represent such a natural class, characterized by the feature [+participant, plural]. Again, it is important to note that in contrast to and-, the va-based forms are athematic.

That means that in Italian, there is a particular vocabulary item only for the singular forms and the third-person plural in the present tense for the root, whereas the case elsewhere is the insertion of the vocabulary item /and-/, which is a thematic root, i.e., with a span size limited to <√go, v°>. For the athematic va-forms, we assume in line with what has been said for Spanish that the “spanning” size of the relevant vocabulary items concerns <√go, v°, Th>, cf. Figure 13. As can be seen, T° (including its φ-features) is adjacent to the preceding span and may trigger allomorphy in case of the va-forms, but not in case of the and-form, since here there is an intervening theme vowel.

The problem now is how to circumvent the fact that [sg/3pl] is not a natural class, but rather a morphomic pattern (the N-pattern) as identified by Maiden in his work (e.g., Maiden 2004, 2009, 2018).19 We would like to go back to a solution proposed by Trommer (2016) in his “postsyntactic morpheme cookbook”: he proposes that in order to explain morphomic patterns in inflection, there are “hidden parasitic features”, or “meta-features on markedness”. These features have a metalinguistic value, but are indeed observable to have their relevance in the derivation of morphomic rules. What is needed according to Trommer (2016), is a redundancy rule that makes a metafeature [m] visible to VI. In the case of Italian go-suppletion in the present tense, this redundancy rule would look like this:

7.

Redundancy Rule:20

[ ] → [m]/[+sg]

[ ] → [m]/[+3]

We can than reformulate the rule for the insertion of the vocabulary item /va/ in Italian as follows:

8.

Vocabulary Item va:

<√go, v°, Th> ↔ /va//___ [pres, m]

The last two issues that need to be explained are the forms vado in the first-person singular and vanno in the third-person plural. We assume that the insertion of epenthetic /d/ in the outcome /vao/ is just another strategy to avoid the hiatus. Indeed, in colloquial Italian (and Italian from Tuscany), there is the supplementary and regular form vo (cf. Markun 1932, pp. 282, 299; Rohlfs 1949, p. 278), where the strategy to avoid the hiatus is simply monophthongization. As for the geminate /nn/ in vanno, we would like to propose that this is also a readjusted form since tonic syllables have to be bimoraic in Italian, i.e., either the vowel of the tonic syllable is lengthened or the syllable is closed by gemination; since the vocabulary entry in this case contains a short vowel, which somehow seems to resist lengthening, the other strategy, namely gemination, is chosen.

We have shown that the derivation of Italian and Spanish go forms is quite similar in the sense that athematic and thematic VI have the same span sizes in both languages. The main difference relies in the fact that Italian mixes athematic and thematic verbs in the present tense, whereas Spanish does not.

3.3. Synchronic Analysis: French

Even though French differs in its morphology in many respects from Spanish and Italian, we assume essentially the same analysis for French go suppletion. As can be depicted from

Table 9. French: several forms of go and sing.

	1st sg	2nd sg	3rd sg	1st pl	2nd pl	3rd pl
pres. ind.	je vai-s	tu va-s	il va	nous all-ons	vous all-ez	ils von-t
pres. subj.	je aill-e	tu aill-es	il aill-e	nous all-i-ons	vous all-i-ez	ils aill-ent
impf.	j‘all-ai-s	tu all-ai-s	il all-ai-t	nous all-i-ons	vous all-i-ez	ils all-ai-ent
future	j‘ir-ai	tu ir-as	il ir-a	nous ir-ons	vous ir-ez	ils ir-ont
pres. ind.	je chant-e	tu chant-es	il chant-e	nous chant-ons	vous chant-ez	ils chant-ent
pres. subj.	je chant-e	tu chant-es	il chant-e	nous chant-i-ons	vous chant-i-ez	ils chant-ent
impf.	je chant-ai-s	tu chant-ai-s	il chant-ai-t	nous chant-i-ons	vous chant-i-ez	ils chant-ai-ent
future	je chant-e-r-ai	tu chant-e-r-as	il chant-e-r-a	nous chant-e-r-ons	vous chant-e-r-ez	ils chant-e-r-ont

, the inflectional patterns in French clearly show a reduced segmentability, not only for go, but also for regular sing. Note that only the first- and second-person plural still have exponents for φ-features that are not dependent on liaison.21 The exponents for the other persons are, as general rule, /z/ and /t/, which are latent consonants, i.e., their realization depends on a following vocalic onset (otherwise they do not surface).

In contrast to Spanish and Italian, the orthographic form of French verbs and their pronunciation differ greatly. In addition, as just mentioned, not all exponents surface in all contexts. This means that the segmentation of the orthographic form may give a wrong impression with respect to the phonological shape of the French verbs. In what follows, we will indicate the French verbs in their phonic form where superscript elements stand for latent consonants that may not surface. This is especially the case for the TAM and the agreement morphemes.

Thus, compared to Spanish and Italian, French verbs, including the regular ones, are structurally less “transparent”. It is not only the apparent loss of theme vowels, but also the verbal endings that are subject to a sandhi phenomenon such as liaison that makes French look different form other Romance languages, at least superficially. As said before, the first- and second-person plural still have exponents for φ-features that are “hearable” independently from liaison, CC, etc., but all other person–number endings are, in contrast, highly instable in the sense that they surface only in certain strongly restricted contexts. The illustration in

Table 10. French person–number endings (orthographic and phonic forms).

	1st Conjugation	2nd and Most 3rd Conjugation Verbs	Passé Simple	Future
1sg	-e /ə/	-s /z/	-s /z/	-ai /-e/
2sg	-es /əz/	-s /z/	-s /z/	-as /az/
3sg	-e(t) /ət/	-t /t/	-t /t/	-a(t) /at/
1pl	-ons /ɔ͂z/	-ons /ɔ͂z/	-mes /mz/	-ons /ɔ͂z/
2pl	-ez /ez/	-ez /ez/	-tes /tz/	-ez /ez/
3pl	-ent /ət/	-ent /ət/	-rent /rt/	-ont /ɔ͂t/

is based on Schpak-Dolt (1992, pp. 115–16); we have highlighted the forms that surface independently of the phenomenon of liaison. As you can see, person is realized in all tenses only in the first- and second-personplural. What is more, in spoken colloquial French, the first-person plural often is expressed by the impersonal form on chante [ʃãt], on va [va], etc. That means that the only remaining vocabulary item for φ-features in this case is the second-person plural.

In order to keep the focus on the main aim of our paper, we will not enter into all details of French verbal inflection here, and will limit ourselves to the forms of the present tense and the imperfect (but see Pomino and Remberger, forthcoming, submitted). For these two tenses the French verbs aller and aimer can be segmented as given in

Table 11. Selected (liaison) forms of French aller and aimer (provisional).

	Present Tense	Imperfect	Present Tense	Imperfect
1sg	vɛ-z	al-ɛ-z	ɛm-z	ɛm-ɛ-z
2sg	va-z	al-ɛ-z	ɛm-z	ɛm-ɛ-z
3sg	va	al-ɛ-t	ɛm	ɛm-ɛ-t
1pl	al-ɔ͂z	al-j-ɔ͂z	ɛm-ɔ͂z	ɛm-j-ɔ͂z
2pl	al-ez	al-j-ez	ɛm-ez	ɛm-j-ez
3pl	vɔ͂-t	al-ɛ-ət	ɛm-t	ɛm-ɛ-ət

. The present tense forms have, at first glance, only two components or even only one, whereas the imperfect forms allow a threefold decomposition. When compared to Spanish and Italian it seems thus as if French verbs were always athematic. In none of the given French forms can we find the overt realization of a theme vowel (cf. e.g., Fr. [al-j-ɔ͂^z] allions vs. It. and-a-v-a-mo).

There are, however, good reasons to assume that French also still has theme vowels. In Pomino and Remberger (forthcoming, submitted), we assume that French has athematic verbs, but also thematic verbs: verbs of the first and second CC (e.g., aimer and finir) are thematic. Evidence for the postulation of theme vowels in French for the first CC come, for example, from the consonant–zero alternation (cf. Schane 1966): The root final consonant of athematic viv(re) ‘to live’ (third CC) is maintained if there is a possibility for it to appear in a syllable onset, e.g., before V (e.g., nous vivons [vi.vɔ͂] ‘we live’) or a C with which it can build a complex onset (e.g., nous vivrons [vi.vʁɔ͂] ‘we will live’), but is deleted before a following consonant with which it cannot form an onset (tu vis [vi-(z)] ‘you_sg live’ not *tu vivs [viv-(z)]). The same final consonant of thematic arriv(er) (first CC) is, in contrast, never deleted: tu arrives [aʁiv-(z)] ‘you_sg arrive’ not *tu arris [aʁi-(z)]. Schane (1966) and others assume that in this case the theme vowel [ǝ], which does not surface, blocks consonant deletion (cf./aʁiv+(ǝ)+(z)/).22 The assumption of theme vowels for the second CC (e.g., finir) is straightforward, since here the theme vowel surfaces as either [i] or [is] (cf. e.g., second-person singular [fi.ni] finis and second-person plural [fi.ni.se] finissez).23

If this is correct, the liaison forms of the French verbs aller and aimer are the ones given in

Table 12. Selected (liaison) forms of French aller and aimer.

	Present Tense	Imperfect	Present Tense	Imperfect
1sg	vɛ-z	al-ə-ɛ-z	ɛm-ə-z	ɛm-ə-ɛ-z
2sg	va-z	al-ə-ɛ-z	ɛm-ə-z	ɛm-ə-ɛ-z
3sg	va	al-ə-ɛ-t	ɛm-ə	ɛm-ə-ɛ-t
1pl	al-ə-ɔ͂z	al-ə-j-ɔ͂z	ɛm-ə-ɔ͂z	ɛm-ə-j-ɔ͂z
2pl	al-ə-ez	al-ə-j-ez	ɛm-ə-ez	ɛm-ə-j-ez
3pl	vɔ͂-t	al-ə-ɛ-ət	ɛm-ə-t	ɛm-ə-ɛ-ət

: We assume that aller as well as aimer (but not the va-forms) have a theme vowel /ə/, which does, as general rule, not surface, but it blocks consonant deletion of the root final consonant.

With the sole exception that there is maybe no or not enough evidence for assuming a theme vowel position for T° in French,24 the verbal structure and the analysis of go-suppletion is in essence parallel to Spanish and Italian, cf. Figure 14.

By assuming theme vowels for French, too, we can better understand the conditions on go-suppletion. Being thematic, the span size of the default realization of French go is <√go, v°>. In contrast, the span size of the va-forms is larger, i.e., it is <√go, v°, Th>, and the allomorphy (cf. /vɛ/, /va/ and /vɔ͂/) is conditioned by the span-adjacent features of T° (which in the present tense contains φ).

4. Discussion: The Advantages of a Spanning Approach

In Section 2, we have shown that some of the postsyntactic processes assumed in DM are disputable as far as our Romance data concern. Fusion, as well as the pruning approaches, face several shortcomings that can be solved when we admit that VI is not restricted to terminal nodes, but is allowed to be inserted in spans. In Romance, in the present tense, fusion seems to be strongly redundant, since it simply reverses the result of posited well-formedness conditions. It is thus better to assume that the respective conditions just do not apply. What is more, fusion does not alter the locality between the root and φ in a way it would be necessary to trigger allomorphy. The pruning approach, in contrast, faces a look-ahead problem: the realization of the theme vowel depends on the realization of the root, but the realization of the root can be determined only after the theme vowel is realized or pruned. Furthermore, in a pre-VI-pruning approach such as the one proposed by Calabrese (2015a), an additional impoverishment rule is necessary in order to delete the diacritic [+suppletive], to avoid a va-form being inserted in the first- and second-person plural in Italian. In sum, both approaches assume additional postsyntactic processes (fusion, pruning, impoverishment) which, in our opinion, cannot even correctly derive the respective forms.

In our analyses in Section 3, we proposed that spanning is the solution to contextual conditioned allomorphy. This approach centers predominantly on the general and well-established process of VI and correctly derives Romance verbal forms. Of course, we still additionally need the uncontroversial well-formedness conditions, which add what are called ornamental morphemes (theme vowels and φ-positions) to the syntactic derivation in other tenses than the present tense, as well as regular phonological-readjustment rules. We have thus shown that integrating spanning into DM allows us to reduce the postsyntactic processes to a minimum. There is no need for fusion or different pruning rules, and root suppletion is instead explained in essence via VI.

For the three Romance languages analyzed here, we have illustrated that the span size of the respective vocabulary items stands in direct relation to the CC they belong to. The CC may be thematic or athematic, and athematicity implies, in our analysis, a larger span size. What is more, allomorphy is restricted to adjacent spans, and thus, in those cases where T° and/or φ trigger allomorphy, the respective verbs must be athematic.

Besides this, the framework of DM with an integrated spanning approach allows us to grasp and model cross-differences on a theoretical level. To illustrate this, we will resume some generalizations that our analyses brought forward:

We have seen that there are some essential differences in the input and outcome of VI as far as tenses are concerned: The present tense is obviously a verb form where T° has no theme vowel nor a separate φ-position, allowing therefore contextually conditioned (noncategorial) allomorphy in all three languages studied here. The imperfect, at least in Spanish and French, is the tense form where we find a maximum of form–function correspondence, including theme vowels for the vocabulary items for functional categories v° and T°. Only v° is not realized except in derivatives, yet it is syntactically (and semantically) present, and visibly followed by a theme vowel. Only the root and little v° therefore are realized by a vocabulary item that builds a span. The presence of the theme vowel prevents stem allomorphy. In French, the presence of theme vowels is less obvious even in the imperfect. It is highly probable that, in French, the well-formedness condition for the insertion of theme vowels holds only for little v°, and there only for certain thematic verbs whose main characteristic is that their stem final consonant is maintained. The tenses derived from the old perfect forms, i.e., the Spanish indefinido or the Italian passato remoto again are structurally less complex, similar to the present tense. However, contrary to the verbal forms in the present tense, we assume that in this tense we have both the realization of T° and its theme vowel in form of a span that includes also φ (an assumption that is also historically justified). This “inflectional” stem then can trigger stem allomorphy, provided that there is no theme vowel, as it is the featural content of T°-[indefinido] to condition the realization of the particular inflectional endings of the indefinido.