The problem in (a) refers to the following generalisation:

In (8) we can see the present indicative conjugation of three regular verbs, one for each of the three conjugation classes in Spanish. As can be seen, the morpheme traditionally called ‘ThV’ is present between the root and the morpheme that marks subject agreement in 5 of these 6 forms. The only form that lacks a ThV is the 1sg form.

The reader might wonder if one could assume that the ThV is present in these forms, but it gets phonologically erased in contact with the 1sg marker, which is -o, in order to avoid a hiatus. I will reject this explanation for two reasons. First, in the third conjugation class, the ThV is -i-, which could form a diphthong with -o. However, *vivio is ungrammatical as a present indicative form.

Second, there is at least one verbaliser which consists in fact of a hiatus, -ea(r), as in (9), where -a- is the traditional ThV. No phonological operation erases the ThV in these cases, despite the hiatus.

These facts make me disfavour a phonological solution for this problem, as it would be both insufficient to explain why the ThV -i disappears and too strong in predicting that any hiatus involving the ThV is resolved by removing the ThV.

The second problem listed above, which I will relate to the first one, is illustrated in (10) and (11). Some verbs use an irregular exponent ending in -g for the 1sg form, here illustrated with sal-i ‘go out’.

The present indicative is, then, fully regular for these verbs except for the 1sg form.

Importantly, the same irregular g-marked exponent used for the 1sg present indicative form is used in all forms of the present subjunctive (11).

This relationship between the irregular exponent used in the 1sg form in the present indicative and the one used in the present subjunctive is a particular instance of a well-known irregularity pattern, identified in Maiden (1992) (see also Maiden 2018, where he advocates for a morphome-based analysis in terms of Aronoff 1994). The irregularity pattern constitutes an L-pattern of irregularity, given the shape adopted by the distribution of the irregular exponent (12).

A fact that, to the best of my knowledge, has not been noted before is that the verbs that follow this L-pattern using a g-marked irregular exponent share another irregularity: they have an irregular 2sg imperative form (which I assume to be the unmarked imperative, cf. Biezma 2010). Specifically, they lose the traditional ThV in the 2sg imperative, as illustrated in (13).

In contrast to these verbs, the regular 2sg imperative form in Spanish keeps the traditional ThV.

In this article, I will relate these three irregularity facts together, and connect them with the property in section 2.1, namely that 1sg present indicative forms lack a traditional ThV.

The third problem that I will address here is the existence of an allomorph -oy /oj/ for the 1sg in a number of highly functionalised verbs in Spanish, such as the copulative verbs in (15).

In a nutshell, I will argue that all the facts that I have introduced in section 2.1–section 2.3 can be connected in an explanatory way. The problem in all these cases reduces to this: the object traditionally called ‘Theme vowel’ in Spanish is the spell out of a Voice-head (which I label, following Ramchand 2018, Evt for ‘Event’), and these three instances of apparent irregularity involve situations where that head is spelled out by the verb stem. What happens in the 1sg form of the present indicative is that the exponent -o is the only verbal agreement marker that also spells out Evt. At the same time, in irregular verbs like sal-i-r ‘go out’, the exponent sal- spells out the verb together with Evt, while the irregular exponent salg- is used when the Theme Vowel is not spelled out. The 2sg imperative is plain sal because the verb projects only up to Evt. In contrast, 1sg forms require salg- because -o spells out Evt. The same exponent salg- has to be used in the subjunctive because the subjunctive marker, like 1sg -o, also consumes Evt and forces the same verbs to use the syntactically smaller salg- exponent. Finally, verbs that use -oy for the 1sg are verbs which must spell out Evt, forcing insertion of a different 1sg exponent that does not include Evt.

But first, I will present the assumptions that I adopt in my analysis.

As I have advanced, in my view (see also Fábregas 2022: 35–40), the object that has been traditionally called the ThV in Spanish – highlighted again in (17) – is the spell out of the head that builds a complete eventuality1 out of the eventuality-descriptive heads.

This means that the traditional Theme Vowel in (17) is the spell out of a syntactic head, whose role is to turn the lexical verb into an eventuality that has time and world parameters, and, as such, the traditional ThV is nothing but the spell out of part of the verb’s functional structure that projects it into a clause. Let me be more specific about this claim by labelling the projection that I take to correspond to the traditional Theme Vowel in (17). The basic structure that I assume for (17) is represented in (18). Here, the theme vowel corresponds to Ramchamd’s (2018) Event Phrase, and the verbal stem is the spell out of the event-descriptive heads Init, Proc and Res.

Following Ramchand (2018), I assume that the event-descriptive heads Init, Proc and Res are responsible for introducing a partial description of an eventuality that lacks time and world parameters. Init, Proc and Res define different subphases from an eventuality. In the specific case of cant- ‘sing’, I assume that the verb defines a Proc(ess) part consisting of a dynamic extended subphase, and an Init(iation) part which defines a causing subevent. In terms of argument structure, Proc selects a possible incremental theme that measures the dynamic part of the event and Init defines an external argument of the eventuality as an agent. (19) represents this part of the verbal structure.

I assume that in a regular verb (19) is spelled out as cant-, that is, as the verbal stem without the traditional ThV (see section 3.3 below for a detailed presentation of my assumptions about spell out).

(19) does not correspond, however, to an eventuality in the way in which it has generally been interpreted since Davidson (1967). The reason is that (19) defines theta roles and Aktionsart – that is, it has provided a description of what the eventuality should look like in the real world – but lacks the parameters of time and world that allow this eventuality to be anchored to tense and be modified by mood and aspect. In other words, (19) cannot be projected into a clause because it lacks the time and world dimension that is required to combine with AspectP, MoodP and TenseP.

In Ramchand (2018), the head that adapts (19) – a predicate – so that it can project into a clause is the head Evt. This head acts as a function that takes an event description and gives as a result an eventuality that is tagged for time and world (see also Fábregas & González Rodríguez 2020), and can therefore further combine with Asp, T, and the rest of the functional nodes that build a clause in Spanish.

Evt does not have any descriptive property, and simply provides syntactic space for the external argument, whose entailments depend on the projection that Evt takes as complement. In (19), that external argument in spec, EvtP, is interpreted as an initiator – agent, causer or instrument – because the InitP layer imposes that interpretation on the specifier of EvtP, but in its absence the external argument would be interpreted as a patient, as ProcP or ResP would be the complement of Evt.

In this sense, EvtP has a role similar to VoiceP in Harley (2013) or Alexiadou, Anagnostopoulou & Schäfer (2015): a head that profiles the eventuality providing space for a participant that is promoted to the subject position, but which does not provide descriptive content to the eventuality. The crucial difference between Ramchand (2018) and these other works is that the equivalent of little v in Ramchand (2018) lacks information about time and world, so that in her theory, the equivalent to VoiceP is the head that provides this information to the lexical verb.

Then, my answer to (16a) is summarised in (21).

The role of Evt in Ramchand (2018) is very similar to what copulative verbs are generally taken to be: as copulative verbs, Evt takes a predicate that is not defined as a full verb yet –because it lacks time and world parameters – and provides it with what is missing so that a clause can be built on top of it. Unlike what happens with the copula in Russian or Arabic, the Spanish copula must be present in any tense, aspect and mood, irrespective of polarity. My claim is that the traditional ThV in Spanish does the same as the copulative verb in (22b). (22a) is as ungrammatical as a (regular) verb in Spanish without the ThV.

Let me summarise some of the evidence that I present in Fábregas (2022: 40–45) in favour of this idea that the traditional Spanish ThV performs the same role as the copulative verb. In essence, I claim that both copulative verbs and the traditional ThV in Spanish perform the same function – adding time and world parameters to a predicate that lacks them – and occupy the same position, Evt.

This means that copulative verbs and traditional ThVs occupy the same position in Spanish, which means – in particular, if the traditional ThV is the direct manifestation of a syntactic head – that copulative verbs must be verbal forms without a traditional ThV –because the verb is the spell out of the same head that in a regular verb is spelled out as the ThV.

I believe that this prediction is borne out. Spanish traditionally has two copulas. The verb ser ‘to be’, used in general for identificational and individual level predicates (Carlson 1977), has a present indicative form where it is impossible to identify any of the three theme vowels in Spanish, /a/, /e/ or /i/.

In the present indicative form of this verb, when we segment the agreement markers, the remaining material does not provide evidence for the presence of a theme vowel. My claim is that this verb lacks a traditional ThV. The reason is that ser is already introduced in EvtP, in order to adapt the predicate so that a clause can be built on top of it. As the ThV performs the same function, this verb does not have a ThV because – so to say – the verb itself is a theme vowel.

The verb estar ‘to be’ is the second Spanish copula, generally used for stage level predicates and spatial locations (Carlson 1977; Arche 2006). Despite appearances, in Fábregas (2022: 43) I show that the right segmentation is not (24a), with a ThV, but (24b), where the final /a/ is part of the verbal stem, and thus that this verb lacks a theme vowel.

My first piece of evidence is that the perfective form of this verb is not the one that one expects from a 1st conjugation verb in /a/. If we compare (25a) to (25b) and (25c), we see that the verb inflects in the perfective as a verb from the 2nd or 3rd conjugation.3

This contrasts with a verb like andar ‘to walk’, which in theory has the same (normative) perfective form (26), but where speakers have a very strong tendency to regularise to a form parallel to (25b), the one in (27a). Importantly, to the best of our knowledge the regularised form for estar is unattested (27b), showing that speakers do not treat the verb as a 1st conjugation form at any level of analysis.

Secondly, in the present indicative, the verb estar does not inflect as a verb with a ThV. Note in particular the stress pattern of estar in contrast with a regular verb of any conjugation (28). The singular forms and the 3pl form of estar have stress on the last syllable, in contrast to regular verbs.

The pattern in (28) supports an analysis where the final /a/ of estar is not a ThV. In their analysis of verbal stress in Spanish, Oltra-Massuet & Arregi (2005) propose that in the present indicative form of the verb, stress is assigned by projecting the right boundary of a metrical foot to the left of the tense morpheme, which in the present tense is a zero exponent (29a), and building an iambic foot from it (29b).

However, this rule does not predict the position of stress for the singular forms and the 3pl form of a regular verb like cantar, as it would predict forms with stress on the last vowel, such as *cantá, *cantán. To solve this, Oltra-Massuet & Arregi (2005: 60–62) propose that Spanish has an idiosyncratic rule that erases the prosodic unit corresponding to the theme vowel in the singular forms and the 3pl form of the present indicative. Then, the theme vowel does not project into a metrical grid, and stress is shifted one position to the left, producing the attested stress forms.

I take this rule to be correct, as it explains in an elegant way the stress pattern of the vast majority of verbs in the present indicative. However, this rule does not apply to estar, where forms like está, están are the correct ones, as shown in (28). Why? My proposal is simply that this rule does not apply to estar because the rule erases the position related to ThVs, and the final -a of estar does not correspond to a ThV. Therefore, the prosodic position related to this -a is not erased, there is no stress shift, and stress always appears – as the algorithm predicts – in the segment immediately preceding the agreement marker.

I believe that there are also strong empirical reasons to also treat estar as a verb that lacks a traditional ThV. In my analysis, this is explained because traditional ThVs and copulative verbs compete for the same position, which is Evt.4 (31) represents my structure for copulative verbs. I assume with Baker (2002) that non-verbal predicates introduce their subject through functional structure, specifically Bowers’ (1993; 2001) Predication Phrase.5

I will now address questions (16b) and (16c), which have to do with how grammatically homogeneous ThVs are in the language in question. Let me start by saying that I believe that it is quite evident that the objects that have been labelled ‘Theme vowels’ across languages do not display a homogeneous set of empirical properties.

Taraldsen Medová & Wiland (2019) have argued that thematic formatives in Czech and Polish can be matched with aspectual properties, and Garet (2021) has shown that French -i- interacts with prefixes in order to define a coherent class of lexical predicates with a defined set of scalar, aspectual and argumental properties. This has nothing to do with the behaviour of (traditional) ThVs in Spanish, where the same ThV is compatible with verbs of every aspectual or argumental class (see 32, as an illustration).

Beyond the fact that ThVs have been defined as morphemes that attach to verbs, I do not think there is any cross-linguistic grammatical property that allows us to say that the label ‘Theme vowel’ has been applied to a grammatically-defined natural class of elements.

If we now look internally to Spanish, however, there is some generalisation that can be stated. As the reader will have noticed, I have been using the term ‘traditional ThV’ to refer to the element that appears in the infinitival form. The reason is that I follow Oltra-Massuet (1999) in her claim that the element traditionally classified as ThV in a language like Spanish is only one of the members of a larger set of elements that appear in verbal forms. I assume with Oltra-Massuet (1999) that an inflected verbal form may have more than one ThV. In particular, I accept Oltra-Massuet’s claim that in a conditional form like (33) there are three theme vowels that can be isolated.

In Oltra-Massuet (1999), the traditional theme vowel is part of a set of morphemes whose role is to mark the functional structure of the verb. In her analysis, the traditional ThV is the marker of the verbal categorising head, but heads corresponding to (grammatical) aspect, tense and mood can also be tagged with ThVs. I accept Oltra-Massuet’s claim that theme vowels appear also as manifestations of tense, mood and grammatical aspect. In fact, in my analysis of forms like sal-e-s ‘you exit’ (cf. section 5.1), I will argue that the ThV -e- that is visible in the verb is a projection of T, not Evt. Note also that in my proposal it is crucial that other verbal heads beyond Evt introduce ThVs in Spanish, because forms that do not correspond to the present indicative of the copulative verbs ser and estar exhibit clear instances of ThVs (34). I take these ThVs to be the spell out of higher clausal projections corresponding to tense, aspect and mood.

That said, I disagree with Oltra-Massuet’s (1999) treatment in two important senses. Pursuing further my claim that the traditional ThV is the spell out of the lowest head used to adapt a verbal predicate so that it projects as a clause, I will answer that Spanish ThVs spell out clausal nodes, starting from Evt and going up to TP. The second way in which I disagree with Oltra-Massuet (1999) is, of course, the nature of the ThV itself. As I noted in section 1, in her analysis – as well as in other analyses put forward within Distributed Morphology or Lexicalism – the ThV is a morphological property belonging to a (postsyntactic) morphological component. Specifically, ThVs are introduced after syntax as dissociated morphemes next to specific functional nodes. In my analysis, however, I will argue that the ThV is the direct spell out of a syntactic head. Therefore, a ThV has a syntactic and semantic role, and it establishes syntactic relationships with other elements in the derivation. This means that, in contrast to Oltra-Massuet (1999), I decompose the conditional in (33) as a syntactic tree – not a morphological tree – and place each ThV in a distinct syntactic head. I represent this for the sake of clarity in (35), although I have not fully developed this analysis yet.

As can be seen in (32), I assume the standard C-T-Asp-v ordering for the clause. The only non-standard claim that I am making is, following Cardinaletti (2004), that I assume a Subject Phrase above TP as the designated position for the subject and the locus of subject agreement.

Unfortunately, at this stage of my research I am not in a position to develop a full analysis of the verbal inflection of Spanish. What is relevant from (35) in this article is that I am assuming that in a regular verb, the traditional ThV is the lowest member of this set, corresponding to EvtP, and that this element is the direct spell out of a syntactically active head.

Let me end this section by answering questions (16b), (16c) and (16d), as a summary. Question (16b) is whether ThVs form a grammatically homogeneous set of elements or not, and question (16c) asks whether ThVs can be reduced to an independently needed set of grammatical categories. In my proposal these two questions are tightly related; (36) provides my answer to them.

Question (16d) asks which components are involved in defining ThVs. My answer to this question is formulated in (37).

I have not said anything about (16e), which is the question of whether the classes of verbs that emerge when one considers ThV marking can be described in this proposal with the tools that the theory makes available. The rest of this article can be read as an illustration of how this proposal works in practice. However, before I move to this part of the article, I need to present in some detail my assumptions about how spell out works, as this is an important ingredient of my analysis.

Let me now make explicit my assumptions about how the structure is spelled out. At the level of spell out, I am following standard nanosyntactic proposals (Caha 2009; Starke 2009; see Baunaz & Lander 2018 for an overview). I am assuming the Phrasal Spell Out principle, which I take from Caha (2009). This principle states that exponents are introduced in the highest phrasal node that contains all the nodes that match the exponent’s lexical entry. This has as a practical consequence that exponents can spell out a syntactic constituent, and are not restricted to insertion in terminal nodes.

For illustration, assume that (38) is the lexical entry of the exponent cant- ‘sing’.

The lexical entry of this exponent matches all the nodes contained in the tree representation in (19) – repeated here as (39) – once the DP arguments (by assumption) have been syntactically evacuated from the constituent.

This makes it possible to insert cant- in the InitP node, in order to lexicalise the eventuality descriptive heads that provide lexical content to an argument structure and an Aktionsart representation to the verb. I assume with Caha (2009) that copies left behind after movement and material that has been previously spelled out do not count for purposes of defining the insertion of an exponent at spell out.

Nanosyntax, moreover, requires that each feature present in the syntax is spelled out, forbidding operations that remove any syntactic feature previous to lexical insertion. This means that nanosyntax does not allow any operation like Distributed Morphology’s Impoverishment operation (Bonet 1991; Noyer 1997), which removes features from the syntactic structure so that particular exponents can spell out the structure.

The principle that states that every syntactic feature must be identified by an exponent is known as Exhaustive Lexicalisation (Fábregas 2007). This principle has a consequence for how exponents are chosen in syncretism situations. When there is no exponent that is a perfect match to spell out a syntactic constituent, the Superset Principle (Caha 2009) applies. This principle is formulated in (41).

In practice, (41) means that an exponent is introduced in a tree structure provided that the exponent has in its lexical entry all the syntactic constituents that the tree has, even if the exponent has in its lexical entry additional features. (41) also guarantees that exponents are only introduced to spell out trees if the features that its lexical entry matches form a constituent. In other words: if the lexical entry of the exponent is as in (42a), the exponent can be used to spell out (42b), which is a perfect match to its lexical entry, or (42c), because (42c) is a constituent within the lexical entry of the exponent.6

However, it will not be able to spell out (43), because (43) is not a constituent within the lexical entry of the exponent in (42a).

(43) does not satisfy the definition in (41) because it is not true that the node where the exponent would be introduced (XP) is identical to a node S in its lexical representation, and all nodes contained under S are also identical, as W is missing. This corollary to the Superset Principle as defined in (41) is known as the Anchor Condition (Abels & Muriungi 2008).

At this point, we have presented the system that we assume for how ThVs work, and we are ready to show how this approach works in addressing the empirical problems that we presented in section 2 above.

Imperatives have been analysed as truncated clauses where the functional structure between the lexical verb area and the CP area is radically impoverished (Zanuttini 1997; Biezma 2010; Isac 2015). The intuition is that imperatives express eventualities that are not instantiated in time periods, and therefore tend to appear without any tense or aspectual marking, and where the information that they transmit – an order given by the speaker to a single addressee – is satisfied already by the illocutionary force in CP. Biezma (2010) proposes the following structure for 2sg imperative forms.

I assume that this truncated structure is in essence correct for 2sg imperative forms, which – following Isac (2015) – I take to be unmarked imperative structures (see Isac 2015 for the more complex structure that might be needed for imperatives involving negation or plural subjects). (53) involves a basic structure where tense, aspect and mood have disappeared but the verb has been projected fully. (54) adapts (53) to the notation that I have used in this article.

Note that the presence of Evt is justified in the imperative because imperatives are modalised, so the verbal predicate requires a world parameter that allows the verb to be modally anchored. Moreover, imperatives require verbs with an agentive external argument, and EvtP is in Ramchand (2018) the projection that provides syntactic space to external arguments.

Assuming the structure in (54), a verb that is regular in the imperative spells out the relevant heads in the way that we have proposed for the general case: Evt is spelled out by the ThV, and the verb stem spells out the eventuality descriptive heads Init-Proc-Res. I assume that C is spelled out as zero in the imperative.

My proposal is that what is special about the verbs in (51)–(52) above, which lack a ThV in the imperative, is that their verb stem exponent spells out Evt together with Init-Proc-Res. (56) represents my proposal for the spell out domain of these verbs, illustrated with sal- ‘go out’.

If (56) is the lexical entry of the exponent sal-, and that exponent includes the head that is otherwise spelled out as the ThV (Evt), it follows that the reduced structure of the imperative would give as a result a form that lacks a ThV.

Therefore, my claim is that these verbs contain verbal exponents that consume also the Evt head, and when the clausal structure is truncated, they emerge without a ThV.

I propose that the apparent ThV that is visible with these verbs in the present indicative is, in fact, a spell out of the T head9 – following Oltra-Massuet (1999), I assume that T can also be characterised with a ThV, see section 3.2 – which is normally spelled out as zero but with these verbs gets exceptionally spelled out with an -e-/-i- exponent.10 As an illustration, (58) presents my representation for the present indicative sal-e-s ‘you go out’.

We are now in a position to connect the imperative 2sg irregularity with the irregularity exhibited by these verbs in the present indicative, where they use an irregular -g- exponent in the 1sg form.

As I have argued in section 4 above, my proposal is that the 1sg exponent -o, only used in the present indicative, involves the spell out of the EvtP head. In a regular verb, as we saw in section 4, no consequences follow from -o spelling out down to Evt, because the verbal exponent spells out maximally up to Init.

However, this second group of verbs have verbal stems that spell out up to Evt. This means that the exponent -o and the verbal stem exponent sal- will compete to spell out Evt, as both have this head in their lexical entry. Following the Superset Principle and its corollary the Anchor Condition (remember 19 and 20 above, section 3.3), however, the -o exponent is the one that will spell out Evt, as Evt is the lowest head in its lexical entry.

The verbal exponent, in consequence, will have to restrict itself to spelling the Init-Proc-Res heads, as in any regular verb. My proposal is that these irregular verbs store two exponents for the verbal stem: the sal- exponent (59a), as we saw in (57), spells out Evt together with Init-Proc-Res, while the second exponent, salg-, spells out only up to Init (59b).

My claim is that these verbs are irregular because they store two distinct exponents (or more, see footnote 9) for the verb stem, and one of them spells out Evt together with the standard Init-Proc-Res sequence.

In the present indicative, then, when this verb combines with the -o exponent, the second verbal exponent, the one in (59b) that does not include Evt, will have to be used.

In contrast, in any other present indicative form that uses a subject agreement exponent that does not spell out Evt, the sal- exponent in (59a) will be used.

As I noted in section 2.2 above, the irregularity pattern that involves g-marked verbal exponents like salg- is a particular subcase of the so-called L-shaped irregularity pattern identified in Maiden (1992) for several Romance languages. In my analysis, I have provided an account of the g-marked exponents, but I have not addressed the general problem of L-shaped irregularity patterns. Let me explain why.

In my analysis, it is crucial that the 2sg imperative form lacks a traditional ThV (sal ‘go out!’). This is what my analysis expects if the verbal stem exponent sal- spells out Evt, as the g-exponent salg- is used when Evt is not included in the material spelled out by the lexical verb. However, it is not empirically correct to generalise this pattern to all verbs that follow and L-shaped irregularity pattern in Spanish.

The L-shaped irregularity pattern includes three more classes of verbs, none of which have a 2sg imperative form without a ThV:

i) Verbs that contain the verbaliser suffix -ec-e, such as viejo ‘old’ > envejecer ‘to get old’, which take an extra /k/ segment in the irregular exponent.

The same pattern is adopted by a small set of verbs that, diachronically, had the Latin antecedent of -ec-e, such as nacer ‘to be born’, conocer ‘to know’.

ii) Verbs that are formed with the verbal stem -ducir and a prefix (Aronoff 1976) also add an extra /k/ under the same conditions: producir ‘to produce’, reducir ‘to reduce’, conducir ‘to drive’, etc.

iii) A small set of verbs which have an irregular exponent that adds -ig- or alter the root vowel (78).

Despite following an L-shaped irregularity pattern, these verbs have imperative forms that contain a ThV:

This means that in my analysis I am not fully covering L-shaped patterns, but only g-marked exponents. I am aware of the existence of alternative analyses to L-shaped patterns. In fact, there is almost complete agreement in the literature that the existence of irregularity models that define specific patterns across cells in a paradigm support a morphomic approach (Aronoff 1994; Trommer 2016). In a morphomic approach, grammar contains specifically morphological features that have no role in syntax or semantics, and whose only function is to determine the morphological shape of some forms.

The account I have sketched in this article tries to derive the irregularity without appealing to this type of purely morphological features, but I have restricted myself to a subset of the L-patterns, specifically those that mark the irregular exponent with -g-. I cannot, then, claim that my analysis makes morphomes redundant in any L-pattern.

However, I would like to sketch a possible extension of my analysis to cover the additional three groups of verbs that follow the L-shaped pattern. One possible move is to split the Evt head in two distinct layers, one designed to introduce the world parameter (w) and a second one whose role is to introduce the time parameter (t). Instead of one head, as in Ramchand (2018), this proposal would follow Nanosyntactic tenets closer, because it would split the two semantic dimensions of Evt in two distinct heads.

In this extension of the analysis, the sal-i ‘go out’ class of verbs would be verbs whose stem exponent sal- spells out both Evt layers, Evt^t and Evt^w (81a). This results in 2sg imperative forms without a ThV (81b).

In contrast, the other L-shaped irregular stems mentioned in (i–iii) above would be cases where the verbal stem spells out only the lowest layer of Evt, Evt^w, leaving part of the Evt area to be spelled out as a ThV. (82) represents the lexical entry that I propose for the verbal exponent of the three groups of verbs in (i–iii) above, illustrated with ca-e ‘fall’.

When introduced in the imperative structure (83), the verbal stem exponent does not consume Evt^t, where a ThV -e has to be introduced.11

The rest of the analysis for these verbs would be in essence identical to the one proposed in section 5 and section 6 for g-marked exponents – these verbs store a second verbal exponent which only spells out up to InitP, which is caig- in the case of the verb ca-e ‘fall’.

I am aware, in any case, that I have not provided independent arguments to support the splitting of Evt into two layers. Therefore, this potential extension should be understood as a speculation that might be worth exploring in further analysis.

The crucial point that I want to emphasise as a conclusion is that approaches where positions created ex profeso accommodate ThVs cannot connect the irregularities that I have discussed in a principled way – although of course this type of approach can stipulate them as irregular quirks of specific verbs. The reason is that, since a ThV position is created in addition to standard syntactic heads, this type of approach does not predict any type of competition between exponents to lexicalise the syntactic head that spells out as a ThV.

My hypothesis might very well be mistaken, and I have in fact pointed out several complications emerging with it, related to the location of some ThVs in present indicative forms (section 5.1) and the connection between verbs like dar ‘give’ and copulative verbs (section 7). My point is, however, that these complications appear precisely because placing the ThV in the syntax makes rigid predictions about what one expects to find and what should be impossible. A syntactic analysis such as the one I have presented here makes an attempt to relate several patterns of irregularity together, hoping to gain a deeper understanding of the internal structure of verbal inflection. This is, then, opposed to a system that is more powerful in what it can generate because it allows post-syntactic operations that can create additional positions to accommodate morphemes whose syntactic and semantic role is less clear.

My main claim here, perhaps, is that a syntactic analysis of ThVs will always force us to ask deeper questions about the internal structure of inflection. With the shortcomings of my proposal, I hope to at least have shown that this type of approach is worth exploring, because it allows us to tie together a number of properties that would otherwise be treated as unconnected and idiosyncratic irregularities of verbs.