1. Introduction
Some of the most fundamental questions in linguistic theory have to do with the architecture of grammar: how many components does the grammar have; what are their properties, atoms, and elementary operations; and why? These overarching questions naturally take more specific forms in particular domains of inquiry. In morphosyntax, for example, two major questions along these lines are 1) how many grammatical components the phenomena traditionally referred to as ‘morphosyntactic’ are distributed over and 2) what the division of labor is between those components, if there is more than one. Different answers to these questions have been given in different morphosyntactic frameworks. In Distributed Morphology, for instance (Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993, Reference Halle and Marantz1994; Harley & Noyer Reference Harley and Noyer1999; Embick & Noyer Reference Embick and Noyer2001; Arregi & Nevins Reference Arregi and Nevins2012, a.m.o.), there is argued to be a (sometimes rich and finely articulated) postsyntactic morphological component that takes as its inputs the structures built up in the syntax and subjects them to a range of morphological operations distinct from the narrow-syntactic operations. In the various versions of Morphology as Syntax, by contrast, either the existence of this postsyntactic morphological component is denied altogether or, at the very least, its role is sharply reduced, with many phenomena that might be given postsyntactic analyses in Distributed Morphology attributed instead to interactions between elementary narrow-syntactic operations such as (External and Internal) Merge and Adjoin (Koopman Reference Koopman2020; Zyman Reference Zyman2020, Reference Zyman2025a; Zyman & Kalivoda Reference Zyman and Kalivoda2020; Julien Reference Julien2022; Ntelitheos Reference Ntelitheos2022; Collins & Kayne Reference Collins and Kayne2023; Crippen Reference Crippen2023; and references therein; see also Bruening Reference Bruening2018a,Reference Brueningb).
Naturally, the choice between these two frameworksFootnote 1 is not an all-or-nothing matter: it may be that the morphosyntactic framework that achieves the highest degree of empirical and explanatory adequacy lies somewhere between the most ‘radical’ versions of Distributed Morphology and Morphology as Syntax.Footnote 2 In order to come closer to answering these larger questions, we need to subject particular empirical phenomena to close scrutiny, prioritizing those that seem especially likely to shed light on the division of labor between the syntax and the (putative) postsyntax.
Against this backdrop, the present article investigates theme vowels. Theme vowels are particularly germane in this context because they have been repeatedly claimed to be purely morphological – i.e., of no syntactic relevance. Thus, Halle and Marantz (Reference Halle, Marantz, Hale and Keyser1993: 135) write:
In many languages—for example, Spanish, Russian, Latin, Latvian—word stems must have a Theme suffix, which has no syntactic […] role […] It is natural to assume that such affixes are introduced by the rules that relate [S-Structure] to M[orphological] S[tructure]. By placing them in this part of the grammar, we account for their lack of effect in the syntax…
Similarly, Aronoff (Reference Aronoff1994: 46) says of Latin theme vowels that they have ‘a use in the language, but that use is purely morphophonological’ and that they are ‘thus never referred to’ by any syntactic rule. Likewise, Embick (Reference Embick2010: 75) assumes ‘that Theme morphemes are, in general, “ornamental” pieces of morphology, items that are apparently relevant for morphological well-formedness, but not part of syntax’ (see also Embick & Noyer Reference Embick, Noyer, Ramchand and Reiss2007: Section 9.4.1, esp. 305–306, and Oltra-Massuet Reference Oltra-Massuet2020: Section 2.1).
In view of such claims, it is unsurprising that theme vowels have been argued to be the realizations of nodes that are created postsyntactically (Halle & Marantz Reference Halle and Marantz1994; Oltra-Massuet Reference Oltra-Massuet1999a,Reference Oltra-Massuetb; Embick Reference Embick2000: 196 fn. 23, Reference Embick2010: Section 3.1.2; Embick & Halle Reference Embick, Halle, Geerts, van Ginneken and Jacobs2005; Oltra-Massuet & Arregi Reference Oltra-Massuet and Arregi2005; Calabrese Reference Calabrese2015; Gribanova Reference Gribanova2015; Pomino & Remberger Reference Pomino and Remberger2022a,Reference Pomino and Rembergerc, Reference Pomino and Remberger2025; Grestenberger Reference Grestenberger2023; see also Pomino & Remberger Reference Pomino and Remberger2019, Reference Pomino, Remberger, Janebová, Emonds and Veselovská2022b; Dolatian Reference Dolatian2022; and Dolatian & Guekguezian Reference Dolatian and Guekguezian2023). But a rival strand of work contends instead that theme vowels are, or are the realizations of, terminal nodes (or larger structures) that are present in the narrow syntax (Picallo Reference Picallo1991; Jabłońska Reference Jabłońska2004, Reference Jabłońska2007; Svenonius Reference Svenonius2004a,Reference Svenoniusb; Wiland Reference Wiland2009; Spyropoulos, Revithiadou & Panagiotidis Reference Spyropoulos, Revithiadou and Panagiotidis2015; Kayne Reference Kayne2018: Sections 14–15, Reference Kayne2019; Fábregas Reference Fábregas, Caha, De Clercq and Vanden Wyngaerd2019, Reference Fábregas2022; Taraldsen Medová & Wiland Reference Lucie and Wiland2019; Bertocci & Pinzin Reference Bertocci and Pinzin2020, Reference Bertocci, Pinzin and Rodríguez2021; Manzato Reference Manzato2023, q.v. for discussion of the debate and additional references; Storment Reference Storment2025; see also Mateu Reference Mateu, Fernández-Soriano, Castroviejo and Pérez-Jiménez2017 and Myler Reference Myler2025).Footnote 3
This article will argue for the latter view, proceeding as follows. Section 2 introduces the narrow-syntactic phenomenon of lexical selection (L-selection), and uses it to formulate a diagnostic for the derivational origin of theme vowels: if a theme vowel is L-selected (to the exclusion of others) by a higher head, then it is the realization of a head present in the narrow syntax – but if not, the test is inconclusive. Section 3 presents evidence from Latin showing that, in fact, it is possible for a higher head to L-select one theme vowel to the exclusion of others. Section 4 develops an analysis of this phenomenon, couched in a version of Morphology as Syntax, that relies crucially on the premise that theme vowels are syntactic heads and can therefore be L-selected in a fine-grained way, just like Ps.Footnote 4
Section 5 turns to the question of whether the relevant empirical observations could also be accounted for if theme vowels were taken to be the realizations of nodes created postsyntactically. To do this, it combines the analysis of Latin theme vowels given by Embick (Reference Embick2010: Section 3.1.2) with Embick’s (Reference Embick2000: 188) principle of Feature Disjointness. It is shown that the analysis is unsuccessful. An attempt is then made to strengthen that analysis by dropping Feature Disjointness, thereby bringing it closer to Embick’s (Reference Embick2010: Section 3.1.2) analysis. It is argued that, with the addition of one unconventional assumption, the analysis can be made to account for the relevant empirical observations – but only in a way that seriously undermines the conceptual rationale for postsyntactic theme-node creation and hence brings the postsyntactic analysis very close to being a notational variant of the syntactic analysis. (The supplementary material discusses and evaluates three additional alternative analyses.) Section 6 concludes the article by summarizing its main arguments and considering their broader implications for morphosyntactic theory and for our understanding of grammatical architecture.
2. L-selection as a diagnostic for the syntactic status of particular empirical phenomena
The main question about theme vowels that we want to answer, then – in order to come closer to settling the more overarching theoretical questions discussed above – is whether they are realizations of nodes (or larger structures) that are present in the narrow syntax or of nodes that are created postsyntactically. Those two analyses of the derivational origin of theme vowels will be referred to below for convenience as the Syntactic Analysis and the Postsyntactic Analysis, respectively. Crucially, the terms ‘Syntactic’ and ‘Postsyntactic’ in those names allude to the grammatical components in which the nodes (or larger hierarchical structures) are created that theme vowels realize. The question of whether the Syntactic or the Postsyntactic Analysis is superior is thus orthogonal to the question of whether phonological content, including that of theme vowels, is inserted late, in the postsyntax (in keeping with work in Distributed Morphology) or early, during the narrow-syntactic derivation (Collins & Stabler Reference Collins and Stabler2016, Bruening Reference Bruening2019b, Kayne Reference Kayne2019–2020, Collins & Kayne Reference Collins and Kayne2023).
One phenomenon that can shed considerable light on whether the Syntactic or the Postsyntactic Analysis is preferable is L-selection (Pesetsky Reference Pesetsky1991: Section 1.3; Merchant Reference Merchant2019), exemplified below:
In (1a), the verb depend takes an on-PP as its complement, and the result is acceptable. Example (1b) shows that the complement of depend (in this use) is obligatory: it cannot be omitted. Example (1c) shows that not just any complement will do: the complement of depend cannot be a nominal phrase. Crucially, (1d) shows that not even all PPs are acceptable as the complement of depend. Rather, depend imposes a much stricter requirement: its PP complement must be headed by on (compare (1d) with (1a)).Footnote 5
L-selection is extremely local: for two lexical items X and Y, if X L-selects Y, then (a projection of) Y merges with (a projection of) X (see also Landau Reference Landau2007: 488–489). Thus, the on-requirement imposed by depend cannot be met just anywhere in the structure. If on is to satisfy this requirement, it cannot appear too high, as in (2a) (which is unacceptable regardless of whether depend is given a structural complement or not). Likewise, on cannot appear too low (2b).
The strict locality of L-selection provides some of the strongest evidence that Internal Merge is involved in deriving structures like (3):
For formal definitions of Merge aimed at accounting for the locality of selection, see Collins and Stabler (Reference Collins and Stabler2016: 63–64), Merchant (Reference Merchant2019: 326), and Zyman (Reference Zyman2024, esp. Section 4; Reference Zyman2025b: 11, 36). L-selection is discussed further in Haegeman, Jiménez-Fernández and Radford (Reference Haegeman, Jiménez-Fernández and Radford2014: 87), Brodkin (Reference Brodkin2021: Section 2.2), Zyman (Reference Zyman2021: 515; Reference Zyman2022: Sections 6–7), Hewett (Reference Hewett2023: 352, Reference Hewett2025), and Serova (Reference Serova2023: Section 3.1).
L-selection is relevant here because it is a fundamentally syntactic phenomenon, a conclusion argued for by Merchant (Reference Merchant2019: 332) and further argued for extensively by Zyman (Reference Zyman2024: supporting information, esp. 5–8). That being so, L-selection makes available a diagnostic that can help us choose between the Syntactic and the Postsyntactic Analysis of theme vowels:
Notably, what is crucial here is whether a higher head can L-select a particular theme vowel to the exclusion of others, not whether a theme vowel can L-select a lower head – in the most widely discussed cases, a root – to the exclusion of others. That is because what a defender of the Syntactic Analysis might analyze as L-selection by a theme vowel of a root could alternatively be analyzed by a defender of the Postsyntactic Analysis as postsyntactic contextual allomorphy – specifically, different roots differentially conditioning the realization of (postsyntactically created) Theme nodes (see the references cited for the Postsyntactic Analysis in the introduction).
With the diagnostic in (4) established, we can now proceed to consider an empirical phenomenon to which we can apply it.
3. The phenomenon: Latin synthetic causatives and their interactions with theme vowels
Almost all Latin verbs fall into a handful of conjugation classes, each associated with a particular theme vowelFootnote 7:
The first-, second-, and fourth-conjugation theme vowels – -ā, -ē, and -ī – shorten to -a, -e, and -i, respectively, under phonologically predictable circumstances.
With (5) established, we can now turn to the main phenomenon of interest: synthetic causatives formed with the root fac- ‘make’. The term ‘synthetic’ here is used in a traditional and purely impressionistic and descriptive sense, alluding to the fact that the causative verbal structures in question have each traditionally been considered one ‘word’ – but no commitment is intended to the hypothesis that there is any such thing as a syntactic (or, for that matter, a morphological) ‘word’. (See Svenonius Reference Svenonius2018: Section 1 for an overview of various theoretical positions on ‘wordhood’.)
The phenomenon of synthetic causatives with fac- ‘make’ – henceforth ‘fac- causatives’ – is illustrated in (6)–(7). Example (6) shows that calēre ‘be warm’ is a second-conjugation verb, and accordingly features the theme vowel -ē. Example (7) shows that from cal-ē ‘be.warm-th.II’ (a subpart of (6)) can be derived the fac- causative calefacere ‘warm [something]’.Footnote 9
Particularly relevant for us is the empirical generalization – argued for meticulously by Hahn (Reference Hahn1947) – that there is a strong link between being a second-conjugation verb (with theme vowel -ē) and being able to form a fac- causative. This generalization emerges clearly from the literature on (or mentioning) fac- causatives taken as a whole: Corssen (Reference Corssen1870: 514), Paucker (Reference Paucker1872: 50 fn. 46), Schmidt (Reference Schmidt1883: 397), Lindsay (Reference Lindsay1894: 490), Skutsch (Reference Skutsch1901, Reference Skutsch1914, esp. Chapters 30, 37), Sonnenburg (Reference Sonnenburg1901), Stowasser (Reference Stowasser1901), Sommer (Reference Sommer1902: 102–103; Reference Sommer1914: 144–145), Güntert (Reference Güntert1917: 19–24), Kretschmer (Reference Kretschmer1929: 53–56), Meillet and Vendryes (Reference Meillet and Vendryes1968: Section 435), Fruyt (Reference Moussy2001; Reference Fruyt2005: 238 fn. 7), Brucale (Reference Brucale2012, esp. 113), and Brucale and Mocciaro (Reference Brucale, Mocciaro and Poccetti2016: 282–283), among other works; see in particular Gradenwitz (Reference Gradenwitz1904: 370–371 and also 372). To begin to see this, consider (8): in every subexample, the (pre-r ‘stem’ of the) second-conjugation verb on the left-hand side of the arrow forms the fac- causative on the right-hand side.Footnote 10
In fact, (8) considerably understates the productivity of the pattern, since it excludes verbs formed with P-like prefixes (see Hahn Reference Hahn1947: 303–304).
The generalization that there is a tight link between being a second-conjugation verb and being able to form a fac- causative is further strengthened by the additional examples of the -ēre
$ \to $
-efacere pattern in (9), which are adapted from Hahn (Reference Hahn1947: 306). Hahn notes (see also Brucale Reference Brucale2012: 113; Brucale & Mocciaro Reference Brucale, Mocciaro and Poccetti2016: 283) that the verbs on the right-hand side of the arrow in (9) have the intriguing characteristic that the fac- within them ‘does not render them causative, or, indeed, alter their meaning in any way, so far as can be judged’ (on non-valency-increasing causatives, see Tyler Reference Tyler2023). Naturally, this attribute of the fac- verbs in (9) does not at all undermine the claim that (9a–h) provide further evidence of the productivity of the -ēre
$ \to $
-efacere pattern.
We see, then, that a great many verbs with -ē as their theme vowel form fac- causatives. This generalization could be accounted for in a simple and straightforward manner by positing that fac- L-selects -ē. Footnote 13 However, in order to show that fac- truly L-selects -ē, as opposed to c-selecting theme vowels in general, we must show that fac- cannot immediately follow theme vowels other than -ē/-e.
For the first-conjugation theme vowel, -ā, this is straightforwardly surface-true: there are no Latin verbs of the form in (10) (where ‘V’ stands for a verbal ‘stem’, descriptively speaking, which could consist of a simplex root or be internally complex). There are also no verbs of the form in (11), which speaks strongly against the possibility that fac- can immediately follow the first-conjugation theme vowel but the latter must shorten in that circumstance (for whatever reason).Footnote 14
What is more, verbs instantiating such structures are, in my (admittedly nonnative) judgment, unacceptable: contrast the attested example in (12) with the unacceptable constructed example in (13).Footnote 15 (The symbol ‘[C]’ means ‘constructed’.)
Turning to the fourth-conjugation theme vowel, -ī, there are also no verbs in which this theme vowel immediately precedes fac-, i.e., no verbs of the following form:
There are some verbs of the form X-i-facere (or X-i-fierī: fierī is the passive of facere), in which the ‘make’-element, fac-/fī-, is immediately preceded by a short -i (Hahn Reference Hahn1947: 301 fn. 1; see also Gradenwitz Reference Gradenwitz1904: 371–372). In all these verbs, however, X is nonverbal: aedifacere ‘build’ (cf. aedēs ‘house’); compendifacere ‘shorten, abridge’ (cf. compendium ‘savings, profit’); floccifacere ‘consider unimportant, care nothing for’ (cf. floccus ‘lock of wool, trifle’); levifacere ‘consider worthless, make light of, pay little attention to’ (cf. levis ‘light, unimportant’); lucrifacere ‘gain (as profit)’ (cf. lucrum ‘profit’); lūdifacere ‘make a game of, make fun of’ (cf. lūdus ‘game, play’); magnifacere ‘make much of, think much of’ (cf. magnus ‘big’); multifacere ‘make much of, esteem highly’ (cf. multus ‘a lot’); parvifacere ‘make light of, despise’ (cf. parvus ‘small’); sūm(p)tifacere ‘lay out, spend, expend’ (cf. sūm(p)tus ‘expense’); turpifacere (‘do dirty/nasty things’, according to Forcellini Reference Forcellini1875: 218 and Commelerán y Gómez Reference Commelerán y Gómez1907: 1434; cf. turpis ‘ugly, shameful’).Footnote 17
Since the X in these X-i-facere verbs is nonverbal, the -i immediately following it is not a verbal theme vowel. More specifically, it is not a shortened version of the fourth-conjugation theme vowel -ī, nor is it the theme vowel of the third conjugation (‘plain’ subtype or -i-stem subtype).Footnote 18 Therefore – contrary, perhaps, to initial appearances – there are no verbs with the following structure:
Recapitulating (14)–(15): there are no fac- causatives in which fac- is immediately preceded by the theme vowel of the fourth conjugation or of the third conjugation (‘plain’ subtype or -i-stem subtype). Here too, verbs instantiating these structures are, in my (admittedly nonnative) judgment, unacceptable: contrast the attested (a)-examples in (16)–(18) with the unacceptable constructed expressions in the corresponding (b)-examples.
3.1. Interim summary
Surveying the landscape of fac- causatives, then, we find:
There is, then, a strong prima facie case that fac- does indeed L-select -ē (qua structurally lower element) to the exclusion of the other theme vowels. If this is correct, then theme vowels must be present in the syntax, since L-selection is a fundamentally syntactic phenomenon (see Section 2) – i.e., the Syntactic Analysis of theme vowels is to be preferred to the Postsyntactic Analysis.
But in order to be able to adjudicate confidently between the Syntactic and the Postsyntactic Analysis, we need to build each of them up explicitly so that we can see how successfully each of them accounts for the generalizations in (19). In particular, we must determine whether the Postsyntactic Analysis can account for them in terms of some mechanism other than L-selection, and thus remain a viable alternative to the Syntactic Analysis.Footnote 21
4. Developing the Syntactic Analysis
We will begin by developing the Syntactic Analysis explicitly and in detail. Because the Syntactic Analysis relies crucially on the assumption that theme vowels are present in the syntax, fleshing it out will require us to have a clear idea of what the clause structure of Latin is.
4.1. A brief primer on Latin clause structure
A great deal of work has converged on the conclusion that the derivation of an ordinary clause in Latin involves movement of a relatively large (phrasal) verbal category to a fairly high specifier position (Bailey Reference Bailey2010; Danckaert Reference Danckaert2012, Reference Danckaert, Lahousse and Marzo2014, Reference Danckaert2017a,Reference Danckaertb; Gianollo Reference Gianollo2016; for discussion of such derivations crosslinguistically, see many of the papers in Carnie & Guilfoyle Reference Carnie and Guilfoyle2000 and Carnie, Harley & Dooley Reference Carnie, Harley and Dooley2005). The present article will adopt the version of this analysis put forth by Zyman and Kalivoda (Reference Zyman and Kalivoda2020) (henceforth Z&K) and further developed by Zyman (Reference Zyman2025a), to which we now turn.
Consider the following verb form:
On the analysis adopted here, (20) is derived as in (21). (The nominals are analyzed as K(ase) P(hrase)s; see Lamontagne & Travis Reference Lamontagne, Travis and Crowhurst1987; Bittner & Hale Reference Bittner and Hale1996; and Norris Reference Norris2021 and Reference Norris, Alexiadou, Kramer, Marantz and Oltra-Massuet2022: Section 2.3, a.m.o. EA and IA stand for ‘external argument’ and ‘internal argument’.)
In (21), Voice bears two unvalued probe features: a Person feature and a Number feature. (Following Heck & Müller Reference Heck, Müller, Bainbridge and Agbayani2007, unvalued probe features are notated [
$ \ast $
F:
$ \square $
$ \ast $
], and probe features that have successfully probed for a value are notated
$ \left[\ast \mathrm{F}:\boxed{\mathrm{VAL}}\ast \right] $
.) Voice values these two features by entering into an Agree relation (dotted arrow) with the structurally most prominent nominal within its c-command domain. Finally, two further operations are crucially involved in giving rise to the constituent/morpheme order observed: 1) Asp undergoes head movement to T, and 2) vP moves to [Spec,TP].Footnote 23
Before proceeding, let us see how one more verb form is derived:
In (22), the passive Voice morpheme -ur occurs farther away from the root than does the element -ē, a portmanteau exponing present tense and subjunctive mood – even though Voice is crosslinguistically lower syntactically than Tense(/Mood). The form in (22) therefore appears – along with many other passive verb forms in Latin – to violate Baker’s (Reference Baker1985: 375) Mirror Generalization (Cinque Reference Cinque1999: 197, citing Calabrese Reference Calabrese1985; Embick Reference Embick2000: 196–199; Baker Reference Baker2014: 8–9; Calabrese Reference Calabrese2019: Section 3.4.4.2).
On the present analysis, however, the derivation of (22) is virtually identical to that of (20):
This derivation has the same three main steps as the one in (21): 1) the unvalued
$ \varphi $
-features of Voice are valued under AgreeFootnote 24; 2) Asp undergoes head movement to T (the former happens to be silent in this form); and 3) vP moves to [Spec,TP].Footnote 25
Z&K argue that this type of derivation – whose most crucial components are Asp-to-T head movement and vP-movement to [Spec,TP]; see their pp. 7–8 for discussion – successfully accounts for the morpheme order in all Latin finite verb forms, and extends straightforwardly to nonfinite and even nonverbal forms, regardless of whether they obey the Mirror Generalization (like laud-ā-v-era-t ‘he/she/it had praised’ in (20)) or superficially appear to violate it, like the form in (22)/(23). See Z&K (appendix) on how the analysis generates all these forms; Z&K (Section 5) for several bodies of independent syntactic evidence for the analysis; and Zyman (Reference Zyman2025a: Section 4.1) for further discussion of the analysis.
4.2. The Syntactic Analysis of theme vowels and their interactions with fac- causatives
With those basics of Latin clause structure established, we can now develop in detail the Syntactic Analysis of theme vowels and how they interact with other subconstituents of fac- causatives. Consider (24):
On the Syntactic Analysis, (24) is derived as follows. First, the structure in (25) is built:
In (25), the V cal- ‘be warm’ takes as its complement the KP āera ‘the atmosphere’, forming a VP. This VP in turn serves as the complement of the second-conjugation theme vowel -ē/-e, which is crucially a syntactically present head (the central claim of the Syntactic Analysis), here taken to be a v head.Footnote 26 The reason that the v head -ē/-e is able to externally merge with [VP āera cal-] ‘the atmosphere be warm’ is that the former L-selects the head of the latter. (More precisely, -ē/-e L-selects disjunctively: it must externally merge with [a constituent headed by] one of the Vs that it L-selects. These include not only cal- ‘be warm’ but also doc- ‘teach’, mon- ‘warn’, and many others. For legibility’s sake, the disjunctive nature of -ē/-e’s L-selectional requirements is not shown in (25); disjunctive selection is discussed further in the works cited in fn. 13. To be a second-conjugation V, then, just is to be a V that is L-selected by -ē/-e.) Following Heck and Müller (Reference Heck, Müller, Bainbridge and Agbayani2007), selectional features – i.e., features that trigger (External or Internal) Merge – are notated as bullet features [
$ \bullet \mathrm{F}\bullet $
]. Satisfied selectional features are struck through:
$ \cancel{\left[\bullet \mathrm{F}\bullet \right]} $
. For much more explicit and formally rigorous discussion of the satisfaction of selectional features, see Zyman (Reference Zyman2024, esp. Section 4; Reference Zyman2025b, esp. Sections 3–4 and appendix).
The next step in the derivation is the most important for present purposes: the constituent headed by -ē/-e, [ v P āera cale-] ‘the atmosphere be warm’, merges with the V fac- ‘make’ and becomes its complement (building in part on an idea from Corssen Reference Corssen1870: 514). This application of External Merge, just like the previous one discussed, is triggered/licensed by an L-selectional requirement: fac- ‘make’ L-selects -ē/-e,Footnote 27 so it is able to merge with a constituent headed by -ē/-e. Crucially, fac- does not L-select any other theme vowel – first-conjugation -ā/-a, fourth-conjugation -ī/-i, or the -i that is the theme vowel of both subtypes of third-conjugation verbs (19) – so it cannot merge with a vP headed by any of those.
The syntactic operations that build the rest of the structure in (25) are routine. The constituent [VP āera calefac-] ‘make the atmosphere warm’ merges as the complement of the third-conjugation (-i-stem subtype) theme vowel -i – also taken here to be a v head – and the external-argument KP lūx ‘the light’ merges in as the specifier of this v. Footnote 28 The adverbial nōndum ‘not yet’ adjoins to vP. Finally, Voice merges in and values its unvalued Person and Number features under Agree, as usual.
The derivation continues as follows:
In (26), Asp undergoes head movement to T, and vP moves to [Spec,TP], as usual. The final portion of the derivation then unfolds as follows:
In (27), [KP lūx] ‘the light’ is topicalized, yielding the constituent order observed in the example at hand (24).Footnote 29 Topicalization is taken here for concreteness to involve movement to the specifier of a left-peripheral Top(ic) head, à la Rizzi (Reference Rizzi and Haegeman1997: 286) and, for Latin, Devine and Stephens (Reference Devine and Stephens2006: 25–26, 28), Danckaert (Reference Danckaert2012), and Z&K (Section 5.1.1), but that is not crucial: essentially any approach to topicalization would be compatible with the overall analysis developed here.
Before we conclude this section, a few brief remarks are in order about the categorial feature of the syntactic node that a theme vowel realizes, on the analysis just developed. Such nodes were taken above to be v nodes. This idea – along with the Syntactic Analysis more generally – is sometimes objected to (Oltra-Massuet Reference Oltra-Massuet1999a: Section 3.3, Reference Oltra-Massuet1999b: Section 3.3; Pomino & Remberger Reference Pomino and Remberger2022a: 3 fn. 3; see Z&K: 8 fn. 12 and Bertocci & Pinzin Reference Bertocci, Pinzin and Rodríguez2021: 19–20 fn. 12 for references and discussion) on the grounds that theme vowels in Latin can cooccur with overt verbalizers such as -fic ‘make’, as in ampli-fic-ā-re ‘widen’ (see Aronoff Reference Aronoff1994: 46; Calabrese Reference Calabrese2019: 9; and Calabrese Reference Calabrese2023: 422 for more examples), and there is something suspect about positing a double-vP configuration. This objection has little force, however, for several reasons. First, as alluded to by Z&K (p. 8 fn. 12), it is not obvious that overt verbalizers such as -fic are v heads; they might instead be Vs (cf. Bruening Reference Bruening2016), in which case verbs such as ampli-fic-ā-re ‘widen’ would not involve a double-vP configuration. Secondly, if overt verbalizers in Latin are v heads, leading us to posit double-vP configurations for the relevant verbs after all, that may conceivably not dovetail very naturally with highly rigid approaches to clause structure (Sportiche Reference Sportiche, Campos and Kempchinsky1995, Starke Reference Starke, Cardinaletti and Guasti1995, Cinque Reference Cinque1999), but it would be entirely unproblematic for more flexible ones (Wexler Reference Wexler, Hornstein and Lightfoot1994; Rizzi Reference Rizzi and Haegeman1997: 314–315; Starke Reference Starke and Belletti2004: Sections 2.1–2.2; Adger Reference Adger, Kibort and Corbett2010; Bošković Reference Bošković2016: 42, Reference Bošković2022: 8–9, Reference Bošković, Grohmann and Leivada2023, Reference Bošković2024a,Reference Bošković, Autry, de la Cruz, Figueroa, Mihajlovic, Ni, Smith and Harleyb; Erlewine Reference Erlewine2016: 475; Bertocci & Pinzin Reference Bertocci and Pinzin2020: 33 fn. 13; see also Z&K: 17), particularly approaches on which ‘functional sequences’ (Starke Reference Starke and Belletti2004) or ‘extended projections’ (Grimshaw Reference Grimshaw, Coopmans, Everaert and Grimshaw2000; Brodkin Reference Brodkin2023, Reference Brodkin2025: Chapter 5) are emergent effects of (categorial or lexical) selection (Di Sciullo & Isac Reference Di Sciullo and Isac2008; Lebowski Reference Lebowski2018, Reference Lebowski2025: Section 3.2.2; Bruening Reference Bruening2019a; Zyman Reference Zyman2021: 536 fn. 29; Hewett Reference Hewett2023: 354; see also Surányi Reference Surányi and Putnam2010: 174–175, esp. fn. 12, for a distinct but related view). Third and finally, if overt verbalizers in Latin are v heads rather than Vs and positing a double-vP configuration turns out to be problematic for whatever reason, we can simply posit that the syntactic head that a theme vowel realizes is not a v but instead bears some other categorial feature – say, [cat Th]. Such a move would spur us to revisit and possibly rethink the geometry of the verbal domain lower than the theme vowel, but it would not at all threaten the idea at the core of the Syntactic Analysis of theme vowels defended here: that theme vowels realize heads that are present in the narrow syntax and can hence be L-selected by higher heads.
Recapitulating, this section has developed the Syntactic Analysis of theme vowels in detail, and shown that it can indeed successfully account for the generalization that, in a fac- causative, fac- can be immediately preceded by the second-conjugation theme vowel -ē/-e but not by any other theme vowel (19). On the Syntactic Analysis, this is a straightforward effect of L-selection, entirely parallel to L-selection for a particular P (1).
5. Developing the Postsyntactic Analysis
Things look good for the Syntactic Analysis, then. But we cannot yet conclude confidently that it is correct, because we have not yet developed the Postsyntactic Analysis in detail and evaluated it. It is to this task that we now turn.
This portion of the investigation will proceed in stages, because there are two different versions of the Postsyntactic Analysis worth considering.
5.1. The Postsyntactic Analysis, version 1
Embick (Reference Embick2010: Section 3.1.2) presents a partial analysis of Latin verbal morphology that is a version of the Postsyntactic Analysis – i.e., it incorporates the assumption (from Oltra-Massuet Reference Oltra-Massuet1999b) that theme vowels realize nodes added postsyntactically. This assumption is motivated by Embick’s view that ‘Theme morphemes are, in general, “ornamental” pieces of morphology, items that are apparently relevant for morphological well-formedness, but not part of syntax’ (p. 75).
Embick (p. 76) illustrates this portion of the analysis using a subpart of (forms of) the fourth-conjugation verb audīre ‘hear’ (theme vowel -ī). On Embick’s analysis, the (relevant portion of the) syntactic structure that is transferred to the postsyntactic morphological component initially looks as follows (the feature [IV] is a conjugation-class diacritic):
Following Oltra-Massuet (Reference Oltra-Massuet1999b), Embick assumes that a Th(eme) element is added to the constituent in (28) postsyntactically, yielding (29):
Embick posits that the structure in (29) is linearized before Vocabulary Insertion. The (null) terminal v node is pruned, and the root and Th are concatenated; once this happens, the Th node is adjacent to an element bearing [IV] (namely, the root). In this derivation, Th is realized by the following Vocabulary Item:
The realization of Th, then, is conditioned by the conjugation-class feature on the root.
Before we attempt to extend this analysis to account for the interactions between theme vowels and fac- causatives, we need to address an important question about Embick’s conjugation-class diacritics: are they features of roots themselves (qua abstract feature bundles present in the narrow syntax), as in
$ {\sqrt{\mathrm{AUD}}}_{\left[\mathrm{IV}\right]} $
, or do they belong to exponents inserted into the structure by Vocabulary Insertion, as in aud-
[IV]? (Call these the Diacritics-on-Roots Hypothesis and the Diacritics-on-Exponents Hypothesis, respectively.Footnote 30) Embick himself (p. 76) makes the former assumption – but both possibilities are worth considering, particularly in view of the following hypothesis, which appears to play a prominent role in much reasoning within Distributed Morphology:
Embick (Reference Embick2010: 76) states that a conjugation-class feature like [IV] is not ‘an uninterpretable feature in the sense familiar from syntactic theory. Rather, it is a diacritic with effects that are seen in the PF derivation’. If so, Feature Disjointness (31) suggests it should not be present in the syntax but only in the postsyntactic morphological component, where it presumably ‘comes in’ as part of an exponent like aud- [IV]. This subsection will provisionally adopt that conclusion (the Diacritics-on-Exponents Hypothesis),Footnote 31 particularly because Embick (Reference Embick2000: 215 fn. 50) argues that ‘Feature Disjointness should be maintained to the fullest possible extent’.Footnote 32 The alternative – the Diacritics-on-Roots Hypothesis (Embick Reference Embick2010: 76) – is considered in Section 5.2.
On this slightly modified version of Embick’s analysis, then – which crucially incorporates the Diacritics-on-Exponents Hypothesis – the derivation of a form such as cal-e-fac-ere ‘warm [something]’ will unfold in such a way that, at some point, the morphological component will receive (at least) the following structureFootnote 33:
The highest terminal node in (32) will eventually be realized as fac-. It could alternatively have been represented as a root – but because it is unclear whether Embick’s (Reference Embick2010) system would allow a root to take a complement and project (or, more theory-neutrally, to appear in the relevant position in (32)), this element is analyzed here as a causative v, v[caus], on analogy with Embick’s (Reference Embick2010: 77) treatment of the Latin morpheme -ess as a desiderative v, v[des] (though see fn. 32).
Following Embick (Reference Embick2010: 76), a Th node is attached to the maximal projection of every terminal v node in (32), yielding (33):
The structure is linearized; the (null) v that is sister to
$ \sqrt{\mathrm{CAL}} $
is pruned; and Vocabulary Insertion occurs from the inside out (Embick Reference Embick2010: 42), yielding (34). Following Embick (Reference Embick2010: 77), when a terminal node X is realized by an exponent Y, the result is notated [X, Y]:
In (34),
$ \sqrt{\mathrm{CAL}} $
is realized by the exponent cal-
[II]. Since this exponent bears the conjugation-class feature [II] (in keeping with the Diacritics-on-Exponents Hypothesis), it causes the lower Th node to be realized by the second-conjugation theme vowel -ē-/-e-. That is, the conjugation-class feature on the exponent of the root conditions the realization of the Th node adjacent to the root node. The v[caus] node is realized by the exponent fac-
[III(i)] (recall from fn. 9 that fac- itself belongs to the third conjugation, -i-stem subtype). This exponent conditions the realization of the right-adjacent Th node, which is realized as -i-, the theme vowel of the third conjugation (-i-stem subtype) according to Embick’s analysis (p. 76). This -i- will surface unchanged in many forms (e.g., calefacit ‘he/she/it is warming [something]’), but it will become -e when immediately preceding -r, as in the infinitive calefacere ‘warm [something]’.
So far, so good: our slightly modified version of Embick’s (Reference Embick2010: Section 3.1.2) analysis, which crucially incorporates the Diacritics-on-Exponents Hypothesis (in keeping with Feature Disjointness), successfully generates fac- causatives such as calefacere ‘warm [something]’.
There is a serious problem, however. In a fac- causative, as established in Section 3, fac- cannot be immediately preceded by any theme vowel other than second-conjugation -ē/-e. But the version of the Postsyntactic Analysis developed in this subsection has the consequence that, when the narrow syntax is merging in (or not merging in) the lexical item v[caus], which will eventually be realized as fac-, no information is yet available about what theme vowel will eventually immediately precede it. The Th node has not yet been added, let alone realized, since both of those steps occur postsyntactically. And the root, though present syntactically, gives no indications as to what conjugation class its eventual exponent will belong to, since, by hypothesis, conjugation-class features belong to (postsyntactically inserted) exponents, not to syntactically present abstract feature bundles. Thus, the version of the Postsyntactic Analysis developed in this subsection massively overgenerates; in particular, it generates fac- causatives such as [*]mūtafaciēbat ‘was making [something] change’ (13), [*]bullifacere ‘make [something] bubble’ (16b), [*]flectifaciunt ‘make [someone] veer [somewhere]’ (17b), and [*]fugifaciō ‘I make [someone] flee’ (18b).
Recapitulating, this subsection developed a version of the Postsyntactic Analysis that was closely based on Embick’s (Reference Embick2010: Section 3.1.2) analysis of Latin theme vowels, but departed from it in that it crucially incorporated the Diacritics-on-Exponents Hypothesis. This seemed like a natural and conceptually appealing move (in the Distributed Morphology context), because it was directly motivated by Feature Disjointness (31). Unfortunately, the analysis turned out to be empirically untenable.
5.2. The Postsyntactic Analysis, version 2
All is not lost for the Postsyntactic Analysis, however. A straightforward response to the problem encountered in Section 5.1 is to modify it by pivoting from the Diacritics-on-Exponents Hypothesis to the Diacritics-on-Roots Hypothesis (despite the former’s theoretical attractiveness in the Distributed Morphology context), thereby incidentally bringing the analysis closer to Embick’s (Reference Embick2010: Section 3.1.2) original analysis of Latin theme vowels. Let us do that, then – i.e., assume that conjugation-class diacritics belong not to postsyntactically inserted exponents of roots but rather to the abstract syntactic feature bundles constituting roots in the narrow syntax.Footnote 34
If we make that assumption, and hold constant all our other relevant assumptions from Section 5.1, then, at some point before the Th nodes are added and the exponents inserted, we will have (35):
As above, the terminal node v[caus] will eventually be realized as fac-.
Crucially, although Vocabulary Insertion has not occurred yet, the root
$ \sqrt{\mathrm{CAL}} $
‘be warm’ already bears the conjugation-class diacritic [II], because we are adopting the Diacritics-on-Roots Hypothesis. That is, we are assuming, with Embick (Reference Embick2010: 76), that conjugation-class diacritics are present on roots from the beginning of the narrow-syntactic derivation (indeed, in the presyntactic lexicon, List A; see Harley & Noyer Reference Harley and Noyer1999: 3). This is auspicious, because it brings us a step closer to being able to account for the tight link between being a second-conjugation verb and being able to form a fac- causative. The natural way to do this is to follow the Syntactic Analysis developed in Section 4 (in this respect) and posit that the link is an effect of L-selection: v[caus] L-selects lexical items that bear the conjugation-class feature [II]. (This is analogous to the ability of the finite C that to L-select any T that bears the finiteness specification [fin:+] – i.e., TPRES, TPAST, or any of the modals – but not the T to, which bears [fin:–], at least on one analysis of these lexical items and the relations thereamong. On L-selection for particular features as opposed to an entire lexical item, see Merchant Reference Merchant2019: 325–326 and Hewett Reference Hewett2023: 240–241, 475.)
There is a problem, though: given the structure in (35), v[caus] cannot L-select certain roots (those bearing [II]) to the exclusion of others, because v is in the way. (Embick Reference Embick2000: 201 and Embick Reference Embick2010: 71 take Latin verb words – such as (35) would be a subconstituent of – to be complex heads, contra the analysis adopted in Section 4.1. But the point remains, because, in the underlying structure from which the complex head in (35) would be assembled on Embick’s analysis, v would structurally intervene between v[caus] and the root.) There is thus a locality-of-selection problem (see Section 2 and references cited there for discussion).
There is a logically possible, if unconventional, way to solve this problem: to get rid of the structural intervener, v, by positing that it is simply not there (it is never merged in in the first place). If that is so, then (35) should be revised as follows:
Just what is unconventional about (36) will be discussed in detail below. But first, let us see explicitly how the rest of the derivation unfolds. When (36) reaches the postsyntactic morphological component, one Th node is attached to the root and another one to the maximal projection of v[caus]. (In attaching the hierarchically lower Th node to the root rather than to a noncausative v, this analysis departs from Oltra-Massuet’s [Reference Oltra-Massuet1999b] and Embick’s [Reference Embick2010: 76] original analyses – in the obvious and essentially necessary way, since we have posited that there is no noncausative v in (36).Footnote 35) This yields (37):
The structure is linearized, and Vocabulary Insertion occurs from the inside out, producing (38):
In (38), the root is realized by the exponent cal-. The lower Th node is right-adjacent to an element bearing the conjugation-class feature [II] (namely, the root), and is accordingly realized as -ē-/-e-, the second-conjugation theme vowel. The terminal node v[caus] is realized by the exponent fac-, which bears the conjugation-class feature [III(i)]. (This is analogous to Embick’s treatment of his head v[des], on which it is realized by the exponent -ess
[III]; see fn. 32. Alternatively, v[caus] could be replaced by a root
$ {\sqrt{\mathrm{FAC}}}_{\left[\mathrm{III}\left(\mathrm{i}\right)\right]} $
, in which case the relevant conjugation-class feature would already be present in the narrow syntax and indeed in List A.) Finally, the higher Th node, being right-adjacent to an element bearing the conjugation-class feature [III(i)], is realized as -i-, the theme vowel of the third conjugation (-i-stem subtype). This derives the desired linear output: cal-e-fac-i-, as in calefacit ‘he/she/it warms [something]’. As mentioned under (34), the theme vowel -i becomes -e when immediately preceding -r, as in the infinitive calefacere ‘warm [something]’.
This version of the Postsyntactic Analysis, then – like the one developed in Section 5.1 – correctly derives fac- causatives like calefacere ‘warm [something]’. But it improves significantly on the previous version of the Postsyntactic Analysis in that it can account for the link between being a second-conjugation verb and being able to form a fac- causative: v[caus] (what will be realized as fac-) only L-selects roots bearing the conjugation-class feature [II], so fac- will only ever immediately follow the second-conjugation theme vowel -ē-/-e-, and never any other theme vowel (19).
5.3. Evaluating the Postsyntactic Analysis (version 2)
Despite this success, the version of the Postsyntactic Analysis developed in Section 5.2 has at least two features that might give one pause. The first is an analytic oddity whose significance is not entirely clear, but the second is a major conceptual shortcoming. Let us take up each of them in turn.
It was mentioned above that the structure in (36) – repeated below – is unconventional:
This is because there is fairly good reason to think that, insofar as the structure in (39) is correct, the root in this structure is not categorized. It might initially seem as though it is, because it is sister to v[caus], whose name suggests it is a subtype of v. But ‘v[caus]’ is just a convenient label, modeled after Embick’s ‘v[des]’ (see fn. 32). As mentioned above, we have no more reason to think fac- realizes ‘v[caus]’ than that it realizes a root
$ {\sqrt{\mathrm{FAC}}} $
‘make’.
But suppose we tried to reconcile (39) with the view that all roots must be categorized (Embick & Noyer Reference Embick, Noyer, Ramchand and Reiss2007: 296) by positing that v[caus] (the element ultimately realized as fac-) is a categorizer, and it is what categorizes the root in (39). A consequence of this hypothesis is that, in both cale-fac-ere ‘warm [something]’ (cf. calēre ‘be warm’) and satis-fac-ere ‘satisfy’ (cf. satis ‘enough’), fac- is the lowest element that verbalizes the structure. (Alongside satisfacere ‘satisfy’ here can be cited the -i-facere verbs listed under (14), which also display no evidence of verbal structure below fac-.) That is, there is no sense in which cale-fac-ere ‘warm [something]’ is deverbal but satis-fac-ere ‘satisfy’ is not. This seems like a highly counterintuitive consequence.
That being so – i.e., since analyzing v[caus] in (39) as a categorizer does not seem to buy us much – the advocate of the Postsyntactic Analysis (and hence of (39)) might well be inclined to bite the bullet and simply accept the idea that the root in (39) is not categorized, contra Embick and Noyer’s (Reference Embick, Noyer, Ramchand and Reiss2007: 296) assumption that all roots must be.Footnote 36 Whether this conclusion is problematic and, if so, how it might be avoided in a principled and otherwise satisfactory way are questions that would be best answered by advocates of the Postsyntactic Analysis (and, plausibly, of Distributed Morphology more broadly). On the Syntactic Analysis, by contrast, this issue does not even arise: what fac- L-selects is -ē/-e itself, which is possible because theme vowels realize syntactically present nodes – and because those nodes were analyzed in Section 4 as v nodes, there is no need to posit that the innermost root in a fac- causative is never in a local configuration with a categorizing head.Footnote 37
Now, let us leave that somewhat inconclusive discussion behind and turn our attention to something much clearer and more concrete.
As is emphasized by advocates of the various versions of Morphology as Syntax (see the works cited in the introduction, especially Collins & Kayne Reference Collins and Kayne2023: Section 4.5 and Zyman Reference Zyman2025a), any putative grammatical operation that overlaps in its effects with (External or Internal) MergeFootnote 38 is suspect on grounds of theoretical unparsimoniousness. Such operations immediately raise the question of whether they could not in fact be eliminated, and their putative effects attributed to Merge instead. They therefore bear a heavy burden of proof.
The putative operation of insertion of dissociated or ornamental morphemes (Embick Reference Embick1997) – commonly taken in Distributed Morphology to be responsible for the existence of theme vowels (see the introduction for references) – is, as noted by Collins and Kayne (Reference Collins and Kayne2023: Section 4.5), an operation of just this sort. This putative operation combines elements that will (sometimes) be overt by derivation’s end – as does Merge. The question, then, is whether there are strong arguments for incorporating this operation into the theory alongside Merge. Focusing on theme vowels specifically, the main motivation for analyzing theme vowels as realizations of postsyntactically added dissociated nodes seems to be the analytic intuition that theme vowels are purely morpho(phono)logical, and hence of no syntactic significance (see the quotes reproduced in the introduction). It is worth noting that even this consideration does not strongly support the Postsyntactic Analysis: even if theme vowels appeared to be inert syntactically, it would not follow that they could not be introduced in the narrow syntax – so we would not have a decisive, or even a particularly strong, argument for positing dissociated-node insertion in addition to Merge.
But we can go beyond that relatively conceptual argument against the Postsyntactic Analysis (and for the Syntactic Analysis) and formulate a more targeted, more empirical one. As just mentioned, the Postsyntactic Analysis is motivated by the analytic intuition that theme vowels are fundamentally morphological and hence syntactically irrelevant. Crucially, though, this section has argued that the only way the Postsyntactic Analysis of theme vowels can be reconciled with the empirical observations about fac- causatives presented here is if the putative dissociated nodes have ‘precursors’ in the narrow syntax, namely conjugation-class features on roots. Differently put, the Diacritics-on-Roots Hypothesis (as distinguished from the Diacritics-on-Exponents Hypothesis) is not an inessential component of Embick’s (Reference Embick2010: Section 3.1.2) analysis of Latin theme vowels: it is indispensable if the Postsyntactic Analysis is to achieve empirical adequacy. But the assumption that roots bear conjugation-class features in the syntax arguably deprives the claim that theme vowels realize postsyntactically added dissociated nodes of much of its empirical content; at any rate, it seriously undermines the conceptual motivation for that claim. The Postsyntactic Analysis is intended to be an outworking of the intuition that theme vowels are fundamentally morphological, not syntactic – but it cannot reflect that intuition, or cash it out, because it needs to posit that theme vowels have ‘precursors’ (conjugation-class features on roots) that are ineluctably narrow-syntactic. The Postsyntactic Analysis thus comes close to being a notational variant of the Syntactic Analysis.
That being so, the natural move, in a minimalist context, is to do away with circuitous derivations involving (syntactically present) conjugation-class features on roots conditioning the realization of postsyntactically added Th nodes, and simply introduce the nodes corresponding to the theme vowels in the narrow syntax to begin with.Footnote 39 That move additionally pays another significant analytic dividend: it enables us to replace dissociated-node insertion, in the analysis, with (External) Merge, a fundamental syntactic operation that it strongly resembles. The elimination of such suspicious duplications in the theory is, of course, in keeping with minimalist goals.Footnote 40
6. Conclusion
A major topic of debate in morphosyntactic theory has been whether there is a (postsyntactic) morphological component distinct from the narrow syntax and, if so, how rich and internally complex it is. In Distributed Morphology, it is crucially assumed that there is such a component, and its internal structure is sometimes taken to be rich and elaborate. In the various versions of Morphology as Syntax, by contrast, either the existence of such a component is denied outright or, at the very least, its role is greatly reduced.
Theme vowels are highly relevant to this debate because there is a persistent analytic intuition that they are fundamentally morpho(phono)logical in nature, and hence syntactically irrelevant; as a result, they are often taken, in Distributed Morphology, to be realizations of dissociated nodes added postsyntactically (the Postsyntactic Analysis). A competing strand of work, however, analyzes theme vowels as realizations of nodes, or larger structures, that are present in the narrow syntax (the Syntactic Analysis).
To help adjudicate between the Syntactic and the Postsyntactic Analysis, this article introduced a diagnostic designed to shed light on the derivational origin of theme vowels – one based on L-selection. The diagnostic is that if a theme vowel is L-selected (to the exclusion of others) by a higher head, then it is the realization of a head present in the narrow syntax (supporting the Syntactic Analysis), since L-selection is a fundamentally syntactic phenomenon – but if not, no conclusion can be drawn (4). It was then argued that the former situation obtains in Latin: in a fac- causative, fac- ‘make’ can be immediately preceded by -ē/-e, the second-conjugation theme vowel, but not by any other theme vowel (19).
A version of the Syntactic Analysis was developed in detail that accounts for this observation. That analysis crucially posits that theme vowels realize syntactic heads; as a result, fac- can directly L-select -ē/-e to the exclusion of other theme vowels in Latin, exactly as a V can L-select one or two Ps in a language to the exclusion of all the others (1).
We then turned our attention to the Postsyntactic Analysis of theme vowels, to determine whether it too could account for the interactions between theme vowels and fac- causatives. An initial version of the Postsyntactic Analysis incorporating the Diacritics-on-Exponents Hypothesis – i.e., assuming that conjugation-class diacritics are borne by postsyntactically inserted exponents – was empirically unsuccessful. A revised version of the Postsyntactic Analysis incorporating the Diacritics-on-Roots Hypothesis – i.e., assuming, with Embick (Reference Embick2010: 76), that conjugation-class features are present on roots in the narrow syntax – was able to achieve empirical adequacy. It required, however, the unconventional assumption that, in a fac- causative, there is no categorizing v head between the innermost root and fac- ‘make’.
More importantly, this revised version of the Postsyntactic Analysis faces a significant conceptual shortcoming. The Postsyntactic Analysis is motivated by the analytic intuition that theme vowels are purely morpho(phono)logical in nature, and hence of no syntactic significance. But the relevant empirical observations about Latin theme vowels and their interactions with fac- causatives (19) require the Postsyntactic Analysis to crucially incorporate the assumption that the putative dissociated nodes, which are supposedly added postsyntactically, have narrow-syntactic ‘precursors’ in the form of conjugation-class features on roots – i.e., one of the ingredients of a theme vowel is inescapably narrow-syntactic. This seriously undermines the conceptual rationale for the Postsyntactic Analysis, because it means that the latter cannot in fact reflect the analytic intuition that theme vowels are purely morphological in nature.
That being so, the Postsyntactic Analysis ends up being little more than an unwieldy notational variant of the Syntactic Analysis. It is then natural, in a minimalist context, to do away with the former’s circuitous derivations – in which syntactically present conjugation-class diacritics on roots condition the realization of postsyntactically added dissociated Th nodes – and simply replace them with the derivations that the Syntactic Analysis posits. These derivations allow the Syntactic Analysis to account directly for the two insights that the Postsyntactic Analysis is forced to account for indirectly: that L-selection plays a crucial role in determining what theme vowels can and cannot immediately precede fac-, and that theme vowels are ineliminably (at least partially) syntactic in nature.
Choosing the Syntactic Analysis also enables us to posit that theme vowels are introduced not by the putative operation of dissociated-node insertion but by the fundamental, and very similar, syntactic operation of (External) Merge.Footnote 41 Insofar as we can refrain from appealing to putative postsyntactic operations such as dissociated-node insertion that closely resemble Merge but are not identical to it, suspicious duplications in the theory are avoided, in line with the minimalist imperative.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S002222672610111X.
Acknowledgments
Many thanks, for valuable discussion, to Dan Brodkin, Alex Chuang, Chris Collins, Anton Fleissner, Bob Freidin, Matt Hewett, Nick Kalivoda, Aditya Krishna, Zach Lebowski, Azuri Lorig, Jason Merchant, Gereon Müller, Ross Rauber, Reagan Sparks, Oliver Sweet, three anonymous JL reviewers, and JL Editor Adam Ledgeway. The usual disclaimers apply.
Funding statement
This work was generously supported by a Faculty Residential Fellowship (2023–2024) from the Franke Institute for the Humanities at the University of Chicago.
Competing interests
The author declares none.
Open access
