1.1 Overview of the argument

Adjunct clauses that modify events and situations are widespread across the world’s languages and raise an interesting question: How do they work? Given that finite and many types of non-finite clauses are fully saturated propositions, how do they function as modifiers, that is, predicates? Moreover, where do the various flavors of modification (temporal, causal, rationale, etc.) come from?

A common strategy for overcoming both problems is to embed a clause as a dependent of a special P head, as in [_PP P [_CP/TP … ]]. The resulted PP can be used as a VP/VoiceP/TP-level modifier due to the inherently predicative nature of adpositions, and the adposition further determines the specific interpretation of the adjunct. For non-finite clauses, an elaborate version of this analysis has been recently proposed by Landau (2021), who shows that the P-approach successfully accommodates the relevant data from English as well as Romance languages and Hebrew. The approach also leaves room for phonologically null adpositions and thus accommodates “unmarked” clausal adjuncts. It is tempting to further extend it to other languages and consider it a universal adjunction strategy.

The present paper focuses on infinitival rationale clauses in Mari, a Uralic language. A rationale infinitive bears a marker lan, identical to the affixal dative postposition, which at first glance fits the P-approach perfectly. What makes Mari interesting is that such ‘dative-marked’ rationale clauses also contain a complementizer manən, and the surface order of lan and manən indicates the structural sequence [_CP [_PP [_TP … T] P] C]. This pattern poses a problem for the theory of categorial selection (Grimshaw 1991, i.a.) and, consequently, for the P-approach to clausal adjunction, because a projection of category P appears to intervene between two functional layers of an extended verbal projection.

I examine several ways to reconcile the P-approach with the Mari data: (i) re-analyzing infinitives as deverbal nominals; (ii) treating manən as a lexical verb; and (iii) taking the underlying structure to be [_PP [_CP [_TP … T] C] P] and postulating a postsyntactic operation responsible for the reordering of lan and manən. I demonstrate that all these options face further problems.

To account for the behavior of Mari rationale clauses without undermining the theory of grammatical categories and categorial selection, I propose an alternative Mood-approach, whereby lan spells out not a P but a functional head at the periphery of the non-finite clause, namely Mood. I work out a syntactic analysis at the core of which lies the idea that rationale clauses contain a modal operator Mod_Rat, structurally represented as the head of MoodP (Nissenbaum 2005; Grosz 2014; Dąbkowski & AnderBois 2023).

The Mood-approach allows us to maintain a clear distinction between lexical and functional categories. It makes postulating an additional P head on top of a rationale clause unmotivated and redundant: it is the clause-internal Mood head itself that turns the rationale infinitive into a predicate and yields the desired purpose interpretation, obviating the need for a clause-external adposition.

Thus, Mari rationale infinitives provide no support for the P-approach to clausal adjunction and indicate that an alternative may be required, at least for some languages and some types of clausal modifiers. I do not claim that the Mood-approach can or should substitute the P-approach universally; rather, the two strategies complement each other and are both needed to capture the full range of adjunct clauses cross-linguistically.

From an empirical perspective, the paper fills in a gap in the description of Mari, an understudied language. While non-finite rationale clauses in English and some related Indo-European languages (e.g., German) have received much attention in the literature (see Faraci 1974; Jones 1991; Johnston 1994; Whelpton 1995, to name a few), adjunct infinitives outside the Indo-European family and in many Uralic languages specifically have remained understudied. The present work draws attention to the peculiar construction attested in Mari, which so far has only been briefly mentioned in grammars (Riese et al. 2022). From a theoretical perspective, the study contributes to the discussion of modifying clauses by offering a viable alternative to the P-approach and demonstrating that the distribution of infinitival adjuncts can be regulated from inside the clause, without a clause-external adposition. It also addresses broader issues of word categories, categorial features, and categorial reanalysis (section 5). Ultimately, the paper proposes that clausal adjunction is not regulated by a single universal mechanism, as two strategies are available cross-linguistically. Some clauses become modifiers by combining with a lexical P head with a specific interpretation (“power from outside”), while some clauses are turned into modifiers by inserting a modal in the clause-internal Mood head (“the power from within”), with the result structure functioning as a predicate with a rationale/purpose/goal reading.

1.2 The data and methodology

Meadow Mari (henceforth, Mari) is one of the two literary norms of Mari, Uralic; see Saarinen (2022) and Bradley & Luutonen (2023) for recent descriptions. Speakers of Mari primarily live in the Mari El Republic, Russian Federation, with other relatively large communities in Tatarstan, Bashkortostan, and Udmurtia. The status of Meadow Mari is described as “threatened” (Campbell et al. 2022): according to the 2010 Russian Census, there were at that time about 370,000 speakers of Meadow Mari, but the number of speakers registered in 2020 is lower – approximately 260,000.

The primary data presented in the paper were collected in 2020–2025 from two native speakers of the Morki(nsko)-Sernur dialect in individual online elicitation sessions. The consultants are from the same age group (38–45 y.o.), both grew up in the Mari El Republic and have a higher education degree; they are bilingual in Mari and Russian and use Mari on an everyday basis in communication with their friends and relatives. All the judgments on the data considered in the paper were robust and confirmed multiple times. They also do not contradict the information provided in the most recent grammar of literary Meadow Mari by Riese et al. (2022) and the data in the Korp corpus of Mari and the Meadow Mari social media corpus (Arkhangelskiy 2019).

1.3 Outline of the paper

The paper proceeds as follows. Section 2 presents the T-P-C puzzle, which challenges the adpositional approach to clausal adjunction. Section 3 considers and refutes two possible ways to avoid the C-over-P problem while still analyzing lan in infinitival rationale clauses as a postposition: reanalyzing the complementizer as a verb (3.1) and introducing a postsyntactic reordering operation (3.2). Section 4 outlines an alternative Mood-analysis that successfully captures the morphosyntax of Mari rationale clause. Section 5 discusses some implications of the proposal and concludes.

2.1 The T-P-C sandwich

Mari is a head-final language with basic SOV word order and optional subject pro-drop.¹ It has several types of subordinate clauses, both finite and non-finite. The latter group includes aš infinitives, participial and nominalized clauses, and so-called converbs (gerunds); see Burukina (2024a) for a recent discussion of the latter three groups.

Infinitives typically appear as complements (Riese et al. 2022: 251) and are rarely used as adjuncts, with a notable exception being the rationale clauses exemplified in (1). Rationale clauses (RatCls) express a reason behind some action: the action of the main clause serves as an instrument to bring about the action of the adjunct clause (Faraci 1974; Farkas 1988).

As shown in (1), infinitival RatCls in Mari have either a referentially independent subject⁴ (1a, 1b) or a silent non-obligatorily controlled subject (1c, 1d), marked here as PRO, similarly to the English She worked hard in order (for her children) to go to school.

Rationale infinitives in Mari are accompanied by the suffix lan, glossed as neutral lan throughout the paper. The suffix is identical in its form to the dative marker (e.g., rβeze-βlak-lan boy-pl-dat), which belongs to the class of affixal postpositions (Luutonen 1997; Burukina 2023b; also Serebrennikov 1967 and Majtinskaja 1982 on Uralic).^5,6 Given that dative in Mari is used to mark a nominal purpose or cause (2, Riese et al. 2022: 76) and that cross-linguistically, adjunct clauses are often introduced by an adposition (cf. in English She did this while/after/before/despite doing something else), it seems obvious that one should follow traditional grammars (e.g., Luutonen 1997) and assume that lan in (1) spells out the same syntactic element as in (2), namely a dative P head.

In addition to lan, rationale infinitives also allow a complementizer manən. This complementizer is not unique to RatCls: it appears in embedded finite clauses (3a) and with some infinitival complements (3b).

While lan and manən cooccur in (1), it is also possible to leave either or both out. To the best of my knowledge, their surface presence/absence does not change the interpretation or any syntactic properties of the sentence, and I assume that both items have phonologically null variants. The only restriction that lan and manən fall under is that, when present overtly, lan always attaches to the infinitive and manən is always clause-final. There is some inter-speaker variation, with one consultant preferring to use either lan or manən rather than both in the same clause; however, importantly, both consultants regularly accepted sentences with the combination lan manən and produced such examples in a translation task (from Russian to Mari). Furthermore, examples with both lan and manən are also found in the Meadow Mari social media corpus (4), confirming the pattern.⁷

What is the structure of Mari RatCls and how do they fit into the broader typology of clausal adjuncts? Presumably, aš spells out the infinitival T, the dative lan spells out P, and the complementizer manən is in C. In adherence to the Mirror Principle (Baker 1985), the surface order of these elements indicates that the underlying structure of RatCls is [_CP [_PP [_TP [… verb ] T=aš ] P=lan ] C=manən]. However, this T-P-C sequence of projections is problematic from the point of view of categorial selection, as it is extremely unlikely for a clausal C head to select a PP. As stated already by Grimshaw (1991), in an extended projection the lexical head and all the functional heads must belong to the same category (see also Van Riemsdijk 1990). The only exception are the so-called mixed projections—e.g., deverbal nominals and adjectives—in which the category is changed by a special nominalizing or adjectivizing head. Importantly, after the category is changed, no more projections of the core category can be built on top of the resulting structure; this is why, for instance, externally nominal ing gerunds in English are incompatible with T/C, as in *that/to John’s running of the marathon.⁸

Analyzing lan in the aš-lan rationale clauses as the dative marker faces further challenges. While cases of an adposition combining with an infinitival TP/FinP have been discussed in the literature, they appear to have rather limited distribution. For instance, aside from purpose/goal infinitives, most of the English data discussed in Landau (2021) involves gerunds (arguably, mixed verbal-nominal projections, see Bresnan 1997; Borsley & Kornfilt 1999; Alexiadou 2001, i.a.); similarly, in Turkic languages adpositions combine with clausal nominalizations (Kornfilt 1997), while in Slavic languages they take headless or light-headed relative clauses (p.k.).

These observations suggest that Ps in Mari and cross-linguistically may be limited to selecting a dependent of category N and incompatible with extended verbal projections altogether (Marcel den Dikken p.c.). If true, this restrictiveness prevents adopting the P-approach to clausal adjunction as universal and brings us back to the question posed in the introduction: What makes clauses work as adjuncts (i.e., modifiers) and determines their specific adjunct interpretation?

2.2 Infinitival, not nominal

One may argue that Mari aš items under discussion are not infinitives, i.e., verbal forms, but rather event nominals; this would explain the aš-lan sequence (a nominal + dative), if not the aš-lan manən pattern (a nominal + dative + a complementizer). This idea finds support in the history of Mari, as researchers generally agree that aš items stem from either nominalizations or headless participial relative clauses (Rätsep 1954; Galkin 1964; Bereczki 2002). Nevertheless, extending such an analysis to modern Mari is rather problematic, as the distribution of aš infinitives is strikingly different from that of regular event nominals derived with mE or maš (Riese et al. 2022; also Burukina 2024a). On the one hand, infinitives are allowed with various modals and other control verbs, such as ‘decide’ (lit. ‘think’), yet nominalizations are ruled out in these contexts (5).

On the other hand, dative event nominals can appear in various argument positions, as shown in (6a), yet they never alternate with aš-lan items under consideration (6b), and the latter can only be used as rationale clauses. If all the relevant forms in (6)—lijməlan, ulməlan, lijašlan, ulašlan—were dative-marked nominalizations, we would expect them to be interchangeable in at least some contexts, contrary to the facts.

Overall, event nominals share the distribution with other nominal phrases and can bear semantic and structural case marking (e.g., comitative or accusative, (7a, 7b)) or be embedded under an adposition (7c, 7d).

In contrast, infinitives in Mari generally do not combine with structural and semantic cases (Lavrentjev 1972; Riese et al. 2022). Several unsuccessful attempts to combine an aš form with a case marker are given in (8).

In fact, lan is the only case-like marker that can appear on an infinitival aš form (Riese et al. 2022: 252). Independent (i.e., non-affixal) postpositions also do not allow infinitival dependents, as shown in (9).

These discrepancies make it difficult if not impossible to analyze aš forms as nominal, but they are unsurprising if aš forms are, indeed, extended verbal projections (clauses). This brings us back to the original puzzle. If we want to keep the P-approach to adjunction and analyze the aš-lan manən sequence in rationale clauses as resulting from the underlying TP-PP-CP structure, both manən and the dative lan end up having a surprisingly limited and generally unusual distribution: lan is now the only adposition that can combine with infinitival TPs/FinPs (and only in one specific context) and allows adding a complementizer otherwise restricted to clauses. In the next section, I explore several possible ways to resolve this conundrum while still analyzing lan as a P head and salvaging the P-approach. Section 3.1 attempts to reanalyze manən as a lexical verb. Section 3.2 considers treating the surface aš-lan manən as a result of some morphological (i.e., postsyntactic) operation. All these attempts are shown to face certain challenges. After that, section 4 presents my original analysis, whereby lan is not a P head but a Mood head at the periphery of an infinitival clause.

3.1 Option 1: manən is a lexical verb

To solve the C-over-P problem, one may propose that manən in RatCls is not a complementizer but a lexical predicate.¹⁰ The clausal adjunct then has the structure [_VP [_PP [_TP/FinP … T] P] V], where the P-headed non-finite clause is a dependent of the verb. In what follows I provide evidence against this approach.

As mentioned earlier, the complementizer manən is identical in its form to and has been grammaticalized from the non-agreeing converb of the verb manaš ‘say, tell’ (Isanbajev 1961; Galkin 1964). Such converbs are typically used to express an action simultaneous with the main event; the use of manaš as a lexical predicate is illustrated in (10).

This makes analyzing manən in RatCls as V and not C appealing; however, upon a closer examination, this approach can hardly be successful. One challenge is to determine the role of the dative PP dependent. While Mari verbs of saying can co-occur with dative DPs, those generally refer to a Goal of communication, that is, a preferably [+human] participant that the speaker is talking to. These verbs can also embed an infinitival clause expressing an order or a suggestion, but they are strictly incompatible with infinitival complements marked with lan, unless those are interpreted as rationale modifiers. For instance, the infinitival clause in (11) cannot be understood to be an argument of the lexical manaš ‘say’ and describe the Lexical Material, that is, what was being said.

One may argue that the lan infinitive in (11) is indeed a rationale modifier. A hypothetical account will then proceed as follows. The rationale constructions with manən are biclausal: manən is the main lexical predicate and it is modified by a rationale (dative) PP with an embedded infinitive. This is schematized in (12) corresponding to (13). Under this approach, manən is a converb, hence the denotation CvbP, and pro_LM stands for the null object of manən referring to the Lexical Material. The structure yields the literal translation of (11) along the lines of ‘You broke the fence, while saying (something) in order for us to get into the yard.’

The structure in (12) raises more questions than it answers, as it struggles to account for several important empirical observations concerning the behavior of manən in rationale clauses, which are easily accommodated under the manən-as-a-complementizer approach. Below I expand the argumentation from Burukina (2023a), who shows that manən in certain clausal dependents must be a complementizer.

First, manən in rationale clauses is desemanticized, that is, it is used even when no speech act can possibly be implied. This becomes especially clear in examples with a matrix subject referring to an inanimate object that cannot speak, such as (14).

One may suggest that the examples in (14) still involve the structure in (12) with the silent Agent of manən ‘saying’ being interpreted either as an arbitrary PRO or as a discourse-linked pro (e.g., ‘they’), yielding the literal reading ‘the table stands in this room while/as they are saying something in order to receive guests’ and ‘the window is big while/as they are saying something in order to better see the mountains.’ First, it is not immediately clear how such literal readings can give rise to straightforward rationale interpretations ‘X is in order for Y’. Second, Mari converbs ending with n enforce obligatory control: the antecedent (controller) of the implicit embedded subject must be in the immediately dominating clause and c-command it. This is illustrated in (15). In (15a) the silent subject of the converb can only be interpreted as coreferent with the c-commanding subject of the closest finite clause—Masha, but not Vasja—and this ban on non-local and extra-linguistic antecedents further renders (15b) semantically awkward. Going back to (14), if manən in these examples is a converb, ‘the table’ and ‘the window’ must control its implicit Agent, resulting in an infelicitous interpretation.

If manən in the sentences under consideration is a lexical predicate, we further expect it to alternate with converbs derived from synonymous speech-act verbs, such as kutəraš ‘tell, speak’ or kalasaš ‘say, tell’. This prediction is not borne out (16). Likewise, it is not possible to substitute manən with man-de ‘say-cvb.neg’, the negative converb form of manaš.

In addition to this, given the structure in (12), one expects some overt lexical material to be able to intervene between lan and manən. As shown in (17), this is not the case: even examples in which manən could be interpreted as referring to an additional speech act are evaluated as marginal at best, and the speakers suggest that only the non-speech-act rationale reading is salient.

Taking these data into account, I conclude that analyzing manən in all rationale infinitives as a lexical predicate is not feasible. I do not exclude a possibility that some rare sentences may indeed have the structure as in (12), with manən introducing a speech event, yet in the majority of sentences with an infinitival rationale clause manən is a complementizer inserted in the C head.

3.2 Option 2: postsyntactic reordering

Another possible solution for the C-over-P problem posed by Mari rationale infinitives that would allow us to keep the P-approach to clausal adjunction is to reassign it from syntax to morphology. The underlying structure of RatCls (18) may be taken to be [_PP [_CP [_TP T] C] P], with the dative P head selecting a CP as its complement. The order T-P-C is then derived at Spell-Out, as a result of a postsyntactic operation.

Several analyses in terms of postsyntactic reordering have been independently proposed for Mari to account for the distribution of possessive morphology, and it is reasonable to extend one of them to the RatCls instead of introducing a new approach. In possessive constructions and PPs with a possessed nominal dependent the possessive marker appears to the left of a structural case suffix (accusative or genitive) but flexibly to the left or to the right of an affixal P: pört-em-βlak-əm ‘house-poss.1sg-pl-acc’ but not *pört-βlak-əm-em ‘house-pl-acc-poss.1sg’, and either pört-em-βlak-əšt/lan ‘house-poss.1sg-pl-ine/dat’ or pört-βlak-əšt/lan-em ‘house-pl-ine/dat-poss.1sg’.

The following two analyses aim at capturing this variation. The first is the lowering analysis (McFadden 2004). McFadden takes the underlying structure of possessed nouns to always be [_KP [_DP [_NP N] D] K], with D hosting a POSS suffix and K standing for both structural and spatial cases. He then argues that some K-s lower onto the D head via postsyntactic head-movement, à la Embick & Noyer (2001). The second approach is the metathesis analysis (Guseva & Weisser 2018). Guseva and Weisser adopt a similar underlying structure as McFadden, with the D head (poss) base-generated under K/P. Building upon Harris & Halle (2005) they propose that D and K swap places via an operation of Metathesis available after linearization but before vocabulary insertion. Metathesis reduplicates a certain string (e.g., N-|D-K| → N-|D-K|-|D-K|), after which parts of both copies are deleted (N-D-K-D-K → N-K-D).

One may argue that in RatCls the surface sequence aš lan manən is derived from the underlying [_PP [_CP [_TP T] C] P] structure in a similar way: either lan (P) lowers onto manən (C) attaching to the left of it¹¹ or lan and manən change positions via Metathesis. Both options are appealing because of the general parallelism between the CP and DP domain, often remarked upon in the literature. However, these postsyntactic accounts encounter at least two problems.

First, the alternation POSS-P/P-POSS is optional, including the contexts with P_DAT. In contrast, in RatCls, the order lan manən is the only possible one and *manən lan is strictly prohibited. Note that manən is a stand-alone lexical item that bears a phonological stress and could easily host an affixal P. The restriction thus weakens the link between the two phenomena and requires an additional explanation for why Lowering or Metathesis must apply specifically in case of an infinitival dependent but not elsewhere.

The second observation concerns embedded negation. Infinitives, similarly to other non-finite dependents, combine with the negation ogəl. Ogəl is traced back to the combination of the negative verb o in the npst.3sg form and the connegative form of the copula ulaš (Galkin 1964; Alhoniemi 1993; Riese et al. 2022). In modern Mari it is an invariant particle that is used as a constituent negation with nominals, PPs (19), adjectives, and non-finite verbal forms.

As shown by Georgieva et al. (2021), ogəl is a constituent negation that is adjoined to the negated constituent; thus the example in (19) has the structure [_PP [_PP kugət dene] [NEG=ogəl]]. Ogəl combines with RatCls too. Importantly for the present discussion, it can appear only in the following linear positions: (i) after the infinitive and before the suffix lan or the complementizer (20),¹² or (ii) at the edge of the infinitival construction, after lan and/or the complementizer (21). As indicated by the English translations, the two orders yield distinct interpretations: negation that is closer to the infinitive receives a narrow scope reading and negation that is at the periphery receives a wide scope reading.

Assuming that NEG as a constituent negation can attach at different levels, one option is to adjoin it to the TP, as in (22). In this case it does not matter if the reordering is done via Lowering (McFadden 2004) or Metathesis (Guseva & Weisser 2018), since in neither case will it affect ogəl: the expected result is the allowed ogəl lan manən (20b). The same structure (22) with a null P (and no postsyntactic reordering) yields ogəl manən (20a).

The second option is to introduce NEG as an adjunct to the CP, as in (23). In this case, Lowering yields acceptable lan manən ogəl, with lan skipping the adjunct ogəl (see McFadden 2004 on adjuncts not intervening in Lowering). The problem with this result, however, is its interpretation: according to the base structure in (23) NEG is within the scope of the rationale P, which is not corroborated by the actual data in (21b). Applying the Metathesis rule is more promising. As formulated by Guseva & Weisser (2018), the operation allows for some material to intervene between the two swapped items; thus, the sequence manən ogəl lan can be turned into ogəl lan manən (20b).

Finally, NEG can also be adjoined to the whole PP, as in (24). The negation is not affected and applying either Lowering or Metathesis results in the same sequence lan manən ogəl, matching (21b) with the correct scope.

The problem with the P-approach is that it undergenerates and cannot produce the commonly attested pattern lan ogəl manən, as in (21a). To get this surface order one can start with a structure in which NEG is adjoined to the TP (yielding ogəl manən lan) and then either apply Lowering twice (breaking the intermediate sequence lan-manən on the way) or allow Lowering to skip a head (C=manən). However, allowing for either of these options causes more problems. First, Lowering of P (lan) needs to result not in left-adjunction to C (manən) but in right-adjunction to T (aš). Second, in this configuration NEG is base generated within the scope of P, thus not matching the actual interpretation of these examples (21a). Most importantly, double Lowering and non-local Lowering are not attested elsewhere; both are banned explicitly by McFadden, following a discussion of lowering being blocked by an intervening head in Embick & Noyer (2001). Likewise, one may postulate additional metathesis operations, but they will lack independent motivation and risk overgenerating.

To summarize, we have seen that, while it is tempting to analyze lan in infinitival rationale clauses as a dative postposion, this approach poses a problem for the theory of categorial selection that cannot be solved by reanalyzing manən as a verb or by assuming that the surface order of the exponents results from a postsyntactic operation. These leaves the original questions without an answer: How should we analyze lan and what enables lan-marked infinitives (and more generally, infinitives cross-linguistically) to serve as rationale modifiers?

I propose an alternative to the P-approach to clausal adjunction—a Mood-analysis, whereby lan in rationale clauses spells out a Mood head in a peripheral yet clause-internal MoodP projection; the surface order of the exponents matches the order of the projections on the clausal spine (25). This approach straightforwardly captures the data in (20) and (21), as schematized in (25). Ogəl NEG can be adjoined to the TP, MoodP, or CP. In the former case NEG scopes below the modal in Mood (spelled out as lan) and in the latter two cases it scopes above the modal, yielding the two distinct interpretations. Importantly, no other combinations are predicted to be possible, and indeed the patterns in (20) and (21) are the only ones allowed by the speakers.

The Mood analysis offers a simple solution to the T-P-C puzzle that does not violate the Mirror Principle and avoids unnecessary stipulations. It predicts all and only the attested combinations with NEG to be possible, utilizing the structural components independently attested in Mari and/or cross-linguistically. The next section discusses the approach in more detail.

4.1 Deriving rationale clauses

I propose that all RatCls in Mari have the structure as in (26).¹⁴

The crucial component of the structure is a teleological Mod_Rat, which makes reference to the goals of the main clause agent or initiator; in this I follow Dąbkowski & AnderBois (2023) on the semantics of rationale clauses in A’ingae, an Amazonian language, who in turn, build upon Grosz (2014). In essence, Mod_Rat is a two-place predicate that requires two propositional arguments of the type <s,t>: one is the embedded TP, and the other is the main TP. The modal quantifies over the set of possible worlds compatible with the matrix agent’s/initiator’s goals in this specific situation.¹⁵

Grosz (2014) and Dąbkowski & AnderBois (2023) concur that, structurally, Mod_Rat is inserted in the Mood head that takes an infinitival TP as its complement; however, they do not focus on the syntax of RatCls. Let us go through the derivation in (26) step-by-step.

• i. A RatCl contains Mod_Rat, a teleological modal, which is inserted in the Mood head. The modal is spelled out as lan, making the suffix homonymous between an exponent of Mood, a functional head in the extended verbal projection, and the dative adposition, an item of category P. Mod_Rat requires two propositional arguments.

• ii. Mod_Rat combines with a saturated embedded TP of the type <s,t>, thus filling in one of its argument slots.¹⁶ Rationale infinitives with an overt subject are evidently saturated, but one may ask whether this holds for the TP in RatCls with a controlled subject. This is a fair question: if RatCls with a PRO subject involved obligatory non-exhaustive control (OC), they could plausibly be analyzed as predicates, to be predicated of a controller in the main clause (Landau 2015). In this case we would need to postulate another Mod_Rat that could combine with a complement of the type <e, <s,t>>. This is not an insurmountable problem; for instance, Landau (2021) postulates homonymous adpositions that introduce either NOC or OC clausal adjuncts. Yet, such an approach may not be necessary for Mari: as shown in Burukina (2024b), in the variety of Mari under discussion, all RatCls with a controlled subject instantiate non-obligatory control and are accommodated by the structure in (26) without change.

• iii. At the MoodP level, Mod_Rat still misses one propositional argument. This slot is to be filled in by the main TP: the embedded CP is a one-place predicate (Williams 1980) predicated of the main TP.

• iv. The MoodP is merged with the complementizer manən. Mari has several lexical entries for manən: one is used in finite CP complements (section 2) and subject clauses and selects a saturated TP/FinP, another appears in non-finite clausal dependents, for instance, those that instantiate control, where it combines with a property-type TP/FinP.

• v. The result rationale CP is adjoined to the main TP.

The central part of the proposal that distinguishes it from the P-approach is that infinitival RatCls are built around a teleological modal that is responsible for their interpretation and determines their syntactic distribution, i.e., allows for them to be used as modifiers/predicates.

One may attempt to bring the P-approach and the Mood-approach together by analyzing lan in Mari rationale clauses as Mood and postulating a silent P that takes the whole CP as its complement, as in [_PP [_CP [_MoodP [_TP … ] Mood=lan ] C=manən ] P_Ø]. However, it is virtually impossible to prove that such silent P is structurally present. One may suggest that it is the covert preposition that carries the rationale semantics, while Mood (spelled out as lan) is merely interpreted as irrealis. This, however, predicts that, when combined with a constituent negation, the aš lan ogəl manən pattern with the underlying structure [_PP [_CP [_MoodP [_MoodP [_TP … ] Mood=lan ] Neg=ogəl ] C=manən ] P_Ø] should be interpreted with negation being in the scope of the rationale modal, contrary to the facts ((21a) in section 3.2). With P being semantically vacuous and unpronounced, its presence remains unproven and unmotivated other than by the desire to have a uniform analysis for all clausal adjuncts: Mood already does all the necessary job turning an infinitival clause into a rationale modifier.

Before I discuss some implications of the Mood-approach, the position of RatCls in the main clause needs to be addressed. That RatCls are adjoined at the TP level is not self-evident, as some types of clausal adjuncts have been shown to be lower VoiceP/vP-level modifiers (see Landau 2021 and references therein). The next section discusses this issue.

4.2 Rationale clauses as TP-level adjuncts

The following observations support the analysis of RatCls in Mari as TP adjuncts: (i) their behavior under ellipsis, (ii) their interpretation in sentences with a matrix negation, and (iii) the surface word order.

First, RatCls cannot be elided together with the main VP and excluding the other material, which proves that they attach above the VP level. This is shown in (27), where the elided part can only be reconstructed as ‘(you) showed the recipe (to Semjon)’, without the RatCl.

Second, RatCls always scope above the matrix negation (28), regardless of whether they are sentence-initial (28a) or preceded by other main-clause material (28b).

Syntactically, NegP in Mari is inserted between vP/VoiceP and TP and typically takes the highest scope (Georgieva et al. 2021). Therefore, for a RatCl to scope over the matrix negation it must be merged above the NegP, at the TP level. Under the compositional semantics approach, at that stage a modifier can be a predicate of propositions (naturally combining with the matrix TP). However, it is very unlikely to be a predicate of events: the event variable must be existentially closed before the negation is merged to get the desired interpretation ‘there is no such event e’ and not ‘there is an event that is not e’.

Finally, the surface position of RatCls in Mari also indicates that they are adjoined to one of the higher projections. Their natural placement is at the left edge of the sentence, in front of all other items of the main clause (29a). Occasionally, a RatCl can be preceded by some material; however, the latter is then interpreted as a topic or focus: for instance, in (29b) joča-βlak-lan is naturally interpreted as an aboutness topic (as indicated by the English translation).

Taken together, the data above strongly suggest that Mari RatCls are TP-level adjuncts and corroborate the analysis in (26).