1.1 Outline

The present study discusses the distribution of overt subjects and subject-related inflections in non-finite embedded clauses in Meadow Mari (Uralic; henceforth, Mari). Previous research has documented that certain non-finite clauses in Mari allow overt referentially independent subjects and/or person-number marking on the non-finite form corresponding to the embedded subject (Kangasmaa-Minn 1966, 1970; Kokla 1986; Matsumura 1986, 1999; Andreeva 2008; Serdobolskaya 2008a, 2008b; Serdobolskaya et al. 2012; Georgieva 2016; Riese, Bradley & Yefremova 2022, i.a.; also most recently Kashkin et al. 2023 on the closely related Hill Mari). Overt subjects can be genitive or morphologically unmarked, sometimes described as nominative (I use the more neutral term unmarked throughout the paper).¹ Subject-oriented person-number marking, when allowed, has been described as optional and is often referred to as subject agreement. However, as I will demonstrate, its distribution differs from that of regular agreement in other contexts.

Aside from some general observations, comprehensive research on the two phenomena across various syntactic structures remains limited. The grammars provide only a handful of examples, and the more theory-oriented papers also leave gaps in the description. At the same time, the presence of overt subjects and subject agreement in non-finite environments challenges fundamental assumptions in modern syntactic theory regarding the “defectiveness” of non-finite domains (going back to Chomsky 1965). The Mari data deserve to be carefully documented and discussed to determine whether they can be accommodated within the existing theoretical framework.

From an empirical perspective, the goal of the paper is to thoroughly examine the distribution of referentially independent genitive and unmarked subjects and subject-oriented person-number inflections (INFLs) in non-finite contexts in Mari, specifically in participial clauses, event nominals,² and temporal gerunds. The following research questions are to be addressed:

1. I.In what constructions are overt subjects and INFLs attested and is there a strong correlation between the two phenomena?
2. II.Does the choice of the subject case marking correlate with any syntactic properties of the non-finite construction and/or properties of the subject itself?
3. III.Does the presence/absence of an INFL correlate with any syntactic properties of the non-finite constructions and/or properties of the whole sentence?

From a theoretical perspective, I propose that although on the surface the availability of overt subjects and INFLs in non-finite contexts in Mari supports the idea that the non-finite Tense/Inflection head is not feature-deficient compared to the finite one, in reality INFLs do not pattern with the usual subject agreement (the one attested in finite clauses) and both INFLs and overt subjects are licensed from outside non-finite TPs. I hope that the present paper will help call attention to the Mari data, show why they are relevant to the general discussion of finiteness, and further stimulate theoretical research on Uralic languages.

1.2 The data and methodology

This paper aims to address gaps in existing descriptions of the Mari constructions under consideration, focusing on comparing the distribution of overt subjects and INFLs across various non-finite clauses. To my knowledge, such a comprehensive comparison has not been previously made. Several separate studies have examined participles in Meadow and Hill Mari, drawing on limited primary data from different dialects: Brykina & Aralova (2012), Volkova (2017), Shagal & Volkova (2018), and Pleshak (2022). The most detailed of the existing descriptions of event nominals rely on corpus data: Serdobolskaya (2008a, 2008b), Serdobolskaya et al. (2012). While these works provide valuable insights, they primarily focus on what is possible, often leaving open the question of whether certain alternatives are ungrammatical. At the same time, temporal gerunds have largely remained understudied.

The present study is based on primary elicited data and offers a comprehensive examination of the distribution of overt subjects and apparent subject agreement in several non-finite constructions for a consistent group of speakers. The focus on I-languages (Chomsky 1986) is justified and advantageous due to the high degree of inter-speaker microvariation in modern Mari. The adopted methodology allows us to identify several subtle and important differences between the individual grammars of the native speaker consultants. Such nuances will be overlooked in a larger-scale corpus study that combines data from multiple sources, obscuring individual patterns. While corpus approaches have their merits, the present study's goal – to uncover correlations between different patterns of subject marking, the (un)availability of INFLs, and various syntactic and semantic properties of sentences – is better served by an in-depth analysis of individual grammars.

Unless otherwise specified, the examples presented in this study were collected between 2022 and 2024 through elicitation sessions with two native speakers of the Morkinsko-Sernur variety of Meadow Mari. Consultant A grew up in the village Kugener (Sovetsky district) and moved to Yoshkar-Ola at the age of 18. Consultant B grew up in Yoshkar-Ola, but her family used a mix of Yoshkar-Ola and Morkinsky varieties of the language. Both consultants currently reside in Hungary. More information is presented below:

• Age range: 38–45 years

• First language: Mari

• Second language: Russian (acquisition began at age 3)

• Education: university degree in linguistics/philology

• Language use: regular use of Mari in daily communication with friends and family and as primary language when visiting Mari El republic

Each type of examples presented in this paper was verified multiple times with both consultants in various settings to ensure reliability and consistency. The consultants were presented with translation (from Russian to Mari), judgment, and explication by paraphrase tasks.

1.3 The scope of the study

The paper focuses on overt genitive and unmarked subjects in non-finite clauses, possibly accompanied by corresponding person-number markers on the non-finite form. To determine the range of contexts for examination, let us first consult the most recent grammar of Meadow Mari by Riese, Bradley & Yefremova (2022). Table 1 summarizes the information in it, with two important remarks regarding its presentation. First, participles used as prenominal modifiers are listed separately from event nominals used as arguments. Riese, Bradley & Yefremova (2022) do not describe event nominals as a separate category but rather as participles used as verbal nouns – an approach also adopted by Alhoniemi (1993) and Savatkova (2002) for Hill Mari. Nevertheless, the grammar acknowledges distinct properties of verbal nouns compared to modifying participles, which justifies keeping them apart. Second, in the table, “No” indicates that the grammar explicitly states that using an overt subject or an INFL is not allowed, and “No?” indicates that the (un)grammaticality of a certain pattern is not mentioned and there are no relevant examples. Additionally, it should be noted that in all constructions, the embedded subject may remain unexpressed.³

The paper proceeds as follows. Taking Riese, Bradley & Yefremova (2022) as the starting point and keeping in mind the research questions outlined in section 1.1, I examine the distribution of subjects and INFLs in the three types of constructions: participles, event nominals, and temporal gerunds. Section 2 discusses modifying participial clauses. Section 3 and section 4 examine the distribution of genitive/unmarked subjects and INFLs in event nominals. Section 5 focuses on temporal gerunds. Section 6 outlines a theoretical analysis of the data within the generative framework. Section 7 concludes.

2.1 The reported patterns

Mari has four types of participles:

1. (i)active participles ending with -šE:

   • pričastij-əm šəmlə-še jəlməze ‘participle-acc study-ptcp.act linguist’⁴

2. (ii)non-active (passive) participles with -mE:

   • pu dene čoŋə-mo pört ‘wood with make-ptcp.nact house’

3. (iii)future participles with -šaš:

   • šəndə-šaš peledəš-βlak ‘plant-ptcp.fut flower-pl’

4. (iv)negative participles with -dəmE:

   • poč-dəmo okna ‘open-ptcp.neg window’

As reflected in Table 1, Riese, Bradley & Yefremova (2022, 255–262) do not provide examples of participles used as modifiers with either a referentially independent subject or an INFL. Similarly, Matsumura (1986, 1999) observes that modifying participial clauses are subject- and INFL-less. The incompatibility of active -šE, -šaš, and -dəmE participles with an independent subject is unsurprising, as the participial clause necessarily contains a subject gap linked to the modified noun. However, that non-active -mE, -šaš, and -dəmE forms are equally incompatible with an embedded subject is less expected. Despite the name ‘passive participles’ used by Riese, Bradley & Yefremova (2022), going back to Pengitov (1951), these items do not bear passive morphology and, arguably, do not involve complete removal of the Agent. They typically contain a gap in the logical object position linked to the modified noun, which leaves the Agent/logical subject potentially available.

Brykina & Aralova (2012) and Volkova (2017) report that participial relative clauses in Meadow Mari allow embedded subjects. Focusing on non-active -mE participles, they observe the following. The choice between genitive and unmarked subjects depends on the DP's position in the animacy hierarchy. Personal pronouns as subjects must be marked genitive. [+human] subjects tend to be genitive. [−human] and [−animate] subjects can be either genitive or unmarked. In addition to this, both papers mention that no INFL can be added to the participle itself; however, a possessive suffix cross-referencing the embedded subject and morphologically identical to INFLs may appear on the relativized (modified) noun.⁵ It further follows from their data that the possessive suffix is optional.

Similarly, Shagal & Volkova (2018) and Pleshak (2022) report for Hill Mari (a sister-language of Meadow Mari) that non-active participial clauses allow genitive and unmarked overt subjects. Genitive subjects are optionally cross-referenced by a possessive marker on the relativized (modified) noun. Interestingly, while Shagal and Volkova note that modifying participles can bear no person-number inflection, Pleshak mentions that INFLs (in her terminology, subject agreement morphology) “can occur not only on head nouns, but also on participles” (Pleshak 2022, 157). She illustrates this with the following example:

Two remarks should be made regarding this example. First, note that the embedded subject – a personal pronoun – is unmarked. This is already unusual, since personal pronouns in this function are normally genitive, both in Meadow Mari and in Hill Mari. Second, the 2sg inflection/possessive suffix is also used as a discourse marker (Savatkova 2002, 115). Since Pleshak does not provide a context for this sentence, it is impossible to determine what exactly -et contributes to its interpretation. As stated in the next section, my consultants rejected examples with inflected participles in Meadow Mari, with and without an overt pronoun.

2.2 The observed patterns

The data from my native speaker consultants largely corroborate the observations made by Brykina & Aralova (2012) and Volkova (2017). First, all types of participles are strictly incompatible with an INFL, regardless of whether it cross-references the embedded logical subject or the modified noun. The examples in (2) and (3) are originally from Riese, Bradley & Yefremova (2022, 255) but the (un)grammaticality judgments are from my consultants.

When it comes to the distribution of overt embedded subjects, both consultants allowed their use in participial relative clauses. Personal pronouns and [+human] DPs used as subjects must be marked genitive. Consultant A occasionally allowed examples with an unmarked [+human] non-pronominal subject, but only as marginal. Consultant B rejected such sentences.

According to Consultant A, animate [−human] subjects and inanimate subjects can be either genitive or unmarked. Consultant B occasionally preferred the unmarked option of [−human] animates (5) and consistently preferred unmarked inanimates (6).⁶

When the embedded subject is genitive, the modified head noun can bear the corresponding possessive marker, morphologically identical to INFLs. According to Consultant A, such marking is optional and its presence together with an overt embedded subject gives rise to a more emphatic interpretation. Consultant B strongly preferred examples with a possessive suffix.

The data are summarized in Table 2.

2.3 Analyzing genitive subjects as possessors

Leaving unmarked subjects aside for a minute, section 2.2 showed that using a participle together with a genitive DP⁸ is allowed. In principle, such genitive DPs could be analyzed as either an embedded subject of the participial clause or as a possessor of the relativized (modified) noun. This potential ambiguity is acknowledged already in Brykina & Aralova (2012, 488); however, to the best of my knowledge, the only attempt to distinguish between the two analytical options by applying syntactic diagnostics has been made by Pleshak (2022) for Hill Mari. She reported variation among her native speaker consultants: in group 1 genitive DPs co-occurring with participles patterned with clause-external possessors, while in group 2 they showed properties of clause-internal embedded subjects. In what follows I discuss the results of Pleshak's tests when applied to the Meadow Mari data under consideration. As we will see, Meadow Mari shows similar microvariation.

To begin with, it should be noted that the genitive DP does not have to be interpreted as an Agent of the underlying event and can denote an actual possessor (owner) of the object referred to by the head noun. This is shown in (8), which has a felicitous reading ‘Masha's shirts washed by someone.’

However, at least in some contexts, the genitive DP must be analyzed as an argument of the participial predicate. First, Mari nominal phrases can generally have only one genitive dependent; as shown in (9), the speakers rejected examples with multiple possessors.

At the same time, in sentences with participial modifiers, Consultant A allowed for the modified noun to be marked with a possessor suffix that indexes an unpronounced possessor even in the presence of a non-coreferential genitive DP. The latter is unambiguously interpreted as the subject of the participial clause. Consultant B rejected all attempts to indicate an independent possessor in the presence of a participial clause with a genitive DP.

Second, Consultant A allowed the genitive subject DP to be an NPI if the participle itself was in the negative form and in the absence of a matrix negation. This is illustrated in (11). Consultant B was again more restrictive and evaluated such examples as marginal at best but mostly unacceptable.

The inter-speaker variation is further noticeable when we examine the relative position of a genitive DP interpreted as the embedded subject and other dependents in the participial clause. Consultant A allowed a flexible word order, with the genitive DP being either followed or preceded by the clause-internal material. Consultant B only accepted and produced sentences where the genitive DP was at the left edge of the participial clause.

2.4 Interim summary

Modifying non-active participles in Meadow Mari allow embedded subjects and prohibit INFLs. The distribution and morphological marking of embedded subjects reveal significant inter-speaker variation.

Both genitive and unmarked subjects are attested in these constructions, but their behavior is not uniform. The unmarked option is predominantly reserved for non-human animate and inanimate subjects, while the genitive option is obligatory for personal pronouns and [+human] DPs. Microvariation is evident in the licensing conditions for genitive subjects. Consultant A allows genitive subjects independently of the nominal context outside the participial clause. In Consultant B's grammar genitive subjects are essentially treated as possessors of the modified head noun, with stricter constraints on their distribution.

3.1 Restricting the distribution of unmarked subjects

As in participial clauses, morphologically unmarked subjects and genitive subjects are allowed in the same types of nominalized clauses; however, the distribution of the former is more restricted than that of the latter. I will discuss the nom/gen alternation in nominals in detail in section 4. At this point, it is important to mention the following two properties that distinguish unmarked and genitive overt subjects. These are strict rules with no exceptions, and they hold regardless of the grammatical function of the nominal (i.e., a subject, a direct object, an indirect object, or a dependent within a PP).

First, as with participles (section 2.2), personal pronouns used as embedded subjects can never remain morphologically unmarked and must always bear a genitive suffix. This is illustrated below for affirmative nominals (14a), negative nominals (14b), and future nominals (14c), with 1^st and 2^nd person singular and plural pronouns.⁹

Second, in all three types of event nominals unmarked subjects are strictly incompatible with an INFL, as shown in (15).

In other words, an INFL can only appear with either an overt genitive-marked subject, as in (16) parallel to (15), or a silent subject (pro), whose case marking is impossible to identify.

The data in Matsumura (1986, 1999) also suggest that unmarked subjects cannot co-occur with an INFL. Serdobolskaya et al. (2012) report that 86% of the nominals with an unmarked subject in their corpus of the Staryj Torjal variety of Meadow Mari (app. 1200 sentences with an event nominal) were not cross-referenced by an INFL marker.¹⁰ In the Morkinsko-Sernur variety of Meadow Mari discussed in the present paper all examples with an unmarked subject and an INLF were deemed unacceptable by the consultants.

3.2 The optionality of gen and INFLs

Another general rule concerns the mutual distribution of genitive subjects and INFLs. As mentioned above, genitive subjects (but not unmarked subjects) are compatible with an INFL. Importantly, neither of these elements is obligatorily present. In other words, a subject can remain silent (e.g., a pro) when an INFL is present and an INFL on the nominal can be omitted when there is an overt genitive subject.

When the embedded subject is intended to be a personal pronoun, it is preferably left silent and only a matching INFL on the event nominal is used (to license the silence and/or facilitate the interpretation). The speakers also allow for the subject to be an overt pronoun marked genitive in the absence of an INFL. Although such examples receive lower acceptability scores compared to their INFL-only counterparts and are not produced in the translation tasks, they are not rejected; hence, the question mark in (17b). The two versions of a sentence – with a sole INFL or with a sole gen – receive the same interpretation.

When the embedded subject is a non-pronominal DP marked genitive, an INFL can also be absent.¹¹ Such sentences with an overt subject and without a person-number marker on the nominal are interpreted as neutral with respect to the information structure. This is illustrated in (18) for an animate and inanimate embedded subject.

It is also possible for a genitive subject (either a personal pronoun or a referential DP) to co-occur with an INFL. When asked whether such examples (19) differ from those with the genitive subject alone (18) and, in case of a pronominal subject, with an INFL alone (17), both consultants suggested that the doubly-marked sentences were perceived as more emphatic, with the embedded subject being contrasted to the possible alternatives (see also section 2.2 on a similar interpretational pattern in participial clauses). Notably, an overt pronominal subject on its own is not necessarily interpreted as emphatic (17b). Future work is yet to determine the exact range of possible information structure readings for such examples (for instance, whether they always involve a contrast).

From a morphosyntactic perspective, it is important to highlight the difference between INFLs in non-finite constructions and the subject agreement marking in finite clauses. The former are optional. The latter is always obligatory regardless of whether the subject is overt or silent, contrasted or not. This will be relevant when I sketch out a possible theoretical account of the phenomena under discussion (section 6).

4.1 The observed patterns

This section summarizes the main empirical generalizations behind the alternation between genitive and unmarked subjects in event nominals. Due to space limitation, I will only provide examples with affirmative -mE nominals. There are some slight differences between the grammars of the two native speaker consultants, which resemble the differences observed in participial clauses (section 2).

Consultant A produced sentences with genitive subjects, regardless of the type of DP (animate or inanimate). Likewise, they preferred genitive subjects over unmarked subjects when asked to evaluate and compare examples in Mari. They also readily accepted examples with unmarked inanimate subjects. Finally, they sometimes allowed examples with unmarked non-human animate subjects but mentioned that such sentences sounded “more artificial”. They did not themselves produce sentences with unmarked [+human] subjects and evaluated the corresponding constructed examples as marginal (but not absolutely unacceptable).

At the same time, Consultant B preferred animate subjects to be marked genitive and inanimate subjects to remain unmarked, matching the pattern reported by Riese, Bradley & Yefremova (2022). They allowed inanimate genitive subjects and occasionally (less frequently) allowed non-human animate unmarked subjects. As a rule, they did not allow unmarked [+human] subjects.

The data to illustrate these observations are given in (20)–(22). (20) shows the general preference for genitive [+human] subjects and the marginal (A)/unacceptable (B) status of unmarked [+human] subjects.

The sentences in (21) demonstrate that unmarked non-human animate subjects were (marginally) accepted by the native speakers, who preferred and produced the genitive options.

Finally, examples in (22) illustrate the variation among the speakers when it comes to inanimate subjects: both allowed unmarked subjects, and Consultant A frequently allowed and produced sentences with a genitive subject.

The animate/inanimate contrast in event nominals was also noted, for instance, by Serdobolskaya (2008a). She examined corpus data from the Staryj Torjal variety of Meadow Mari (app. 1200 sentences) and reported a general tendency (but not a strict rule) to use genitive animate subjects, especially [+human] ones, and unmarked inanimate subjects. It should be noted that Serdobolskaya grouped together proper names and personal pronouns, while my data indicate that their behavior is not identical, with the former occasionally staying unmarked but the latter being strictly genitive (section 3.1). She also showed no absolute numbers but only provided the percentages (Table 3).

Interestingly, when asked to compare parallel equally acceptable examples with an unmarked subject and a genitive subject, both consultants independently suggested that there was a correlation between the presence/absence of a genitive suffix and the referential status of the subject DP. All else being equal, unmarked DPs tend to be interpreted as non-specific and genitive DPs tend to be interpreted as specific. This is demonstrated by the contrast in the translations in (23) and (24).

One might argue that, on their own, the speakers' intuitions are rather unreliable, and I agree that further research should be done on the referential status of DPs in Mari in general. Yet, the same unmarked/non-specific vs. genitive/specific tendency was previously mentioned by Serdobolskaya (2008a) and Serdobolskaya et al. (2012). Serdobolskaya (2008a) illustrated it with the pair of examples in (25).

It is important to acknowledge that the pair in (25) is not minimal – in addition to the genitive suffix on the embedded subject, (25a) also contains an INFL on the event nominal. As was mentioned in section 3.2, the presence of an INFL together with an overt subject prompts the latter to be interpreted as contrastive, emphatic, and may make the specific reading more salient. However, together with the minimal pairs in (23) and (24), elicited from my native speaker consultants, these data lend credence to the general correlation pattern.

The correlation between the marking of a DP and its specificity has been observed in other constructions as well. For instance, both Meadow Mari and Hill Mari allow differential object marking (the term goes back to Bossong 1985) in non-finite contexts (Tužarov 1986; Alhoniemi 1993; Serdobolskaya 2005; Riese, Bradley & Yefremova 2022; Pleshak & Sirotina 2023, i.a.). Unmarked objects are usually indefinite unmodified nominal phrases, which parallels the tendencies characteristic of unmarked subjects in event nominals. (See section 6 for possible analyses of unmarked dependents.)

The correlation may also help account for the fact that, for both consultants, an unmarked inanimate subject often becomes dispreferred when it is modified by an adjective:¹⁴ the presence of a modifier makes a specific reading more salient. This is illustrated in (26) and (27). There is no obvious correlation with other properties of the embedded subject, for instance, its thematic role: an inanimate Cause argument (26) behaves the same way as an inanimate Patient (27).

Importantly, there is no ban on co-occurrence of unmarked inanimate subjects with adjectival modifiers: for example, both speakers agreed that ‘big box’ in (28) can be either genitive or caseless.

In general, unmarked subjects do not have to be bare nominal heads; for instance, they are not restricted in terms of plurality, as shown in (29).

The next section discusses the tendencies behind the genitive/unmarked alternation that have been mentioned in the existing literature but that do not find sufficient support in data from the variety of Mari under consideration.

4.2 The disputable patterns

4.3 Interim summary

To summarize, event nominals in Mari allow overt referentially independent subjects, which are either marked genitive or morphologically unmarked. A genitive subject can, but does not have to, be accompanied by an inflection on the event nominal matching its person and number features. Having an INFL is also possible when the subject of an event nominal is silent. In this case, the presence of an INFL facilitates interpretation. When both a genitive subject and a matching INFL are present, the subject is perceived as being emphasized.

There are several strict rules that regulate the genitive/unmarked alternation: (i) unmarked subjects are impossible with an INFL, and (ii) all overt pronominal subjects must be morphologically marked (cf. section 2 on participles). In addition to this, several tendencies should be pointed out: unmarked subjects tend to be inanimate and non-specific, while genitive subjects tend to be animate and specific. The marking of the subject does not correlate with its thematic role and the syntactic function of the event nominal.

The next section focuses on the distribution of overt subjects and INFLs in temporal gerunds. As I show, it is virtually identical to that in event nominals, which is not necessarily expected if the two are separate categories. I attempt to explain the similarities in section 5.2, where I propose to analyze gerunds as postpositional phrases with an event nominal dependent.

5.1 Overt subjects and INFLs in temporal gerunds

Mari has three types of temporal gerunds:

1. (i)gerunds of prior action ending with -mek(e)

   • cf. in English after having done something …

2. (ii)gerunds of future action with -meš(ke)

   • cf. before doing something …

3. (iii)gerunds of simultaneous action with -šəla

   • cf. while doing something …

The three constructions are exemplified in (44).

The three types of gerunds all allow INFLs, but only gerunds of prior and future action are used productively with overt subjects. According to Riese, Bradley & Yefremova (2022), gerunds of simultaneous action are compatible only with silent subjects that are coreferent with the matrix subject. The native speaker consultants occasionally accepted examples with an overt embedded subject (45) but did not produce such sentences themselves.¹⁵ (Notice how in (45b) an INFL is positioned inside šəla, resembling an interfix; this will become important in section 5.2.)

According to Riese, Bradley & Yefremova (2022, 268–273), overt subjects in temporal gerunds can be either genitive or unmarked. The grammar further reports that animate nouns show a preference for the genitive marking. In addition to this, it is stated explicitly that all personal pronouns must be marked genitive when used as an embedded subject. The elicited data corroborate these observations.

Unmarked subjects are incompatible with INFLs (46).

Pronominal subjects must bear a genitive suffix (47).

Similarly to event nominals, a pronominal subject in a temporal gerund can appear without a matching INFL (47). It can also remain silent when an INFL is present (48). When both a genitive subject and an INFL are realized overtly the former receives a more emphatic interpretation, as indicated in the translation in (47).

INFLs are also optional when embedded subjects are referentially independent DPs. The interpretational contrast between sentences without an INFL and those with both an INFL and a genitive subject is shown in (49).

5.2 The genitive/unmarked alternation in temporal gerunds

The genitive/unmarked alternation in temporal gerunds is regulated by the same generalizations as that in event nominals. Animate subjects, especially [+human] ones, are preferably marked genitive, as shown in (50). I observed the same difference between the consultants' individual grammars as in the case of event nominals, with Consultant A occasionally allowing (though not producing) sentences with an unmarked [+human] subject and Consultant B being more restrictive in this respect.

Inanimate subjects can stay morphologically unmarked (51). Consultant A generally preferred sentences with a gen embedded subject, regardless of the animateness of the latter. Consultant B strongly preferred inanimate subjects to remain unmarked.

As in event nominals, inanimate embedded subjects in temporal gerunds do not have to be bare nominal phrases. For instance, they allow modification by an adjective, as in (52).

5.3 Temporal gerunds as PPs

Comparing the data from section 4 and section 5 reveals a striking similarity in the distribution of subjects and INFLs between event nominals and temporal gerunds in Mari. One might suggest that subject DPs and INFLs behave the same way in all non-finite contexts, but this is not true. As discussed in section 2, Mari participial modifiers reject INFLs. Similarly, -aš infinitives disallow genitive and unmarked subjects and have more restricted INFL usage (Burukina 2023b). Affirmative instructive gerunds with the suffix -n also reject genitive subjects and INFLs (Riese, Bradley & Yefremova 2022, 263–264). This highlights the unexpected uniformity between event nominals and temporal gerunds, especially if they are considered distinct grammatical categories.

I propose that gerunds are, in fact, event nominals combined with a postposition or a lexical case, the latter being analyzed as a suffixal P (on affixal Ps see Riemsdijk & Huybregts 2002; Asbury et al. 2007; Simonenko & Leontyev 2012; Pleshak 2019; Burukina 2023a on Mari in particular). While a diachronic link between event nominals and gerunds is well established (Bereczki 2002), synchronic descriptions of Meadow and Hill Mari typically treat these constructions as separate categories (e.g., Alhoniemi 1993; Savatkova 2002; Riese, Bradley & Yefremova 2022; Kashkin et al. 2023). To my knowledge, both descriptive and theoretical work on Mari (and other Uralic languages) draws a line between the two types of constructions without much justification. Event nominals are often described as nominal uses of participles and temporal gerunds as broadly defined converbs and dejepričastija. I argue for extending the historical PP analysis to modern Mari, as it better captures the semantic and syntactic behavior of these forms.¹⁶

From a diachronic perspective, the temporal gerund markers go back to combinations of affixal postpositions. The marker of prior action, meke, stems from the combination of the event nominalizer m and the old lative -ke (Bereczki 2002, 136; see also Isanbaev 1961).¹⁷ Bereczki (2002, 31) argues that the old lative marker is a continuation of a prolative suffix. This allows us to hypothesize a connection between lative and the prior action reading: ‘somebody did Y after (lit. through/by) doing X’. The marker of future action, meške, consists of the nominalizer m and the illative -ške. The combination straightforwardly derives the desired interpretation ‘somebody did Y before (lit. into) doing X’. Finally, the marker of simultaneous action, šəla, goes back to the combination of šə and la. The former has been described as either a nominalizing suffix or a participle marker (cf. məla forms used instead of šəla forms in some North-Western dialects of Mari), while the latter is a so-called modal case; in modern Mari la is a clitic productively used, for instance, in comparative constructions. (Cf. in English temporal uses of as meaning ‘while’, e.g., They arrived as we were leaving.)

The historical data show that, without a doubt, temporal gerunds in Mari originated as PPs with an embedded event nominal. I propose that they have preserved the original complex structure and should still be analyzed as postpositional phrases in modern Mari. This is indicated by the glosses in (53), revised from (44).

The proposed PP approach has the following advantages. First, it yields compositionally transparent semantics for gerund markers: through doing X for gerunds of prior action, into doing X for gerunds of future action, and as doing X for gerunds of simultaneous action. Second, it straightforwardly accounts for the fact that temporal gerunds can only be used as modifiers (Georgieva 2024), PPs by their nature being predicates (see e.g., Den Dikken 2006). Third, the analysis explains the observed similar distribution of embedded subjects and INFLs in temporal gerunds and event nominals.

Additionally, the PP approach predicts that another marker may potentially intervene between the two parts of the gerund “suffix” (i.e., the nominal mə/šə and an affixal postposition). The prediction is confirmed by examples of gerunds of simultaneous action (44) where an INFL is inserted between šə and la, consistent with the tendency of person-number markers to precede the modal/comparative la (Riese, Bradley & Yefremova 2022, 88).

6.1 Licensing unmarked and genitive subjects

The data obtained from two native speaker consultants are summarized in Table 4.

A possible account for the distribution of unmarked subjects in nominals has been sketched out by Georgieva (2016) for Besermyan Udmurt. Following the discussion in Fejes (2005), she proposes that unmarked subjects are in fact nominal heads that form compounds with the nominalized predicate. This explains the many restrictions that unmarked subjects in non-finite clauses in Besermyan Udmurt must comply with: they are restricted to R-expressions (i.e., personal pronouns in the subject position must be morphologically marked) and cannot be pluralized or modified. They must also be adjacent to the verbal nominal; that is, no modifier, clitic, or particle can intervene between the two. Georgieva tentatively extends the analysis to Mari, however, she admits that not all instances of unmarked subjects should necessarily be analyzed as non-heads of compounds. She does not offer an account for either unmarked non-compound “exceptions” or genitive subjects and INFLs.

While compounding may account for some instances of unmarked subject-like dependents in Mari clausal nominals (cf. in English reading books vs. book reading), it cannot accommodate all relevant data. Recall the examples with non-adjacent, modified, and plural unmarked subjects in section 3 and section 4: these cannot be analyzed as bare nouns adjoined to the head of the event nominal.

I propose a complementary account within the Dependent Case Theory (DCT) framework (Marantz 1991), building on Pleshak's (2022) analysis of Hill Mari participial modifiers. DCT states that case is assigned configurationally and not under agreement with a dedicated functional head and posits the following hierarchy of cases: (i) lexical (tied to a specific thematic role), (ii) dependent (assigned in the presence of another DP in a c-command relation with the target DP), (iii) unmarked (assigned to any DP that has not received a lexical/dependent case), and (iv) default. For Hill Mari participles, Pleshak (2022) argues that morphologically unmarked subjects receive nominative case in the domain where it is the unmarked case, i.e., within the complement of a C head. Genitive subjects are in the domain where genitive is the unmarked case, i.e., within the complement of a D head; see also Burukina (2023a) on genitive being an unmarked case in Meadow Mari DPs. Pleshak proposes that examples with a participial clause and a genitive DP have two possible underlying structures, depending on whether the genitive DP is a true embedded subject or a possessor of the modified noun. In the former case, Pleshak assumes that the participial clause lacks a CP layer, which makes the participial TP a part of the higher nominal domain from a case assignment perspective. The three derivations are schematized in (55), corresponding to the examples in (54). (Pleshak's analysis does not address the presumed absence of an operator in (55b) compared to (55a, c), and its implications for relative clause formation.)

I propose to extend the Dependent Case Theory approach to Meadow Mari with some modifications. First, I assume that all participial relative clauses in Mari are of the same structure and size (CPs). Second, non-finite TPs are deficient compared to finite clauses, unable to independently license DP subjects (cf. accounts of licensing via agreement). However, structurally smaller subjects – nPs and NPs – are allowed in non-finite context. Thus, morphologically unmarked subjects in non-finite clauses are caseless nPs/NPs, not nominative DPs. That nominal dependents smaller than DPs can serve as arguments has been independently argued for PPs by Simonenko & Leontjev (2012) and Volkova (2014), i.a. The nP/NP status of unmarked subjects explains their incompatibility with INFLs, assuming that only full DPs can be probed for agreement/feature-sharing. It also captures the correlation between the marking and specificity of embedded subjects, the DP layer being generally associated with specificity and definiteness (and potentially animacy/Person features). Furthermore, as personal pronouns are DPs (e.g., Déchaine & Wiltschko 2002), they are predicted to be illicit in “unmarked” contexts, which is confirmed by the data: personal pronouns as subjects must be marked genitive.

As for the genitive DPs occurring with participial clauses, recall from section 2 that they exhibit properties of embedded subjects in Consultant A's grammar, while in Consultant B's grammar they pattern with clause-external possessors. To explain this variation, I follow Pleshak (2022) in assuming that participial clauses differ in their opacity. I hypothesize that Consultant A allows CPs with multiple specifier positions; one of them can serve as a landing site for the embedded subject. This movement places the DP in the higher nominal domain, where it receives genitive case. In Consultant B's grammar, CPs have only one specifier position, which in relative clauses is occupied by the operator. The embedded subject remains inside the participial TP. However, a separate DP possessor can be base generated in a higher nominal projection.

The applicability of this approach to event nominals and the status of INFLs is addressed in the following sections. Section 6.2 discusses the structure of event nominals and their relation to participles, while section 6.3 examines the status of INFLs.

6.2 Event nominals as mixed projections

Event nominals show some verbal properties: they denote events, have an argument structure, and are compatible with aspectual modifiers (section 3.1). They are highly productive and show no idiosyncrasy in their internal morphosyntax. It is therefore reasonable to assume that they are formed in syntax and include a verbal core (a TP) as well as some nominal projections. This aligns with previous analyses of deverbal nominalizations built in syntax (Grimshaw 1990; Moulton 2014) and work on mixed verbal-nominal projections, including English gerunds (Bresnan 1997; Borsley & Kornfilt 2000; Alexiadou 2001, i.a.). In a simplified way, their structure can be represented as [_[N]P … [_[V]P … ] ], where [V]P is an extended verbal projection and [N]P is an extended nominal projection.

Event nominals in Mari are derived using the same suffixes as participles: -m for affirmative nominals and non-active participles, -dəme for negative nominals and negative participles, and -šaš for future nominals and future/irrealis participles. While one could argue that this is mere homonymy, with -m spelling out distinct functional elements (a nominalizer and a participle head), a more unified syntactic analysis capturing the similarity between these constructions is worth pursuing. An attempt to unify participles and nominals has recently been made by Dékány & Georgieva (2020) for Udmurt. They outline possible analytical options, which I list below, adapted to Mari data. In all three scenarios the relative clause is fully saturated and does not contain an operator, and the relevant suffix (e.g., -m) spells out the participial head T (55).

1. I.Event nominals are nominalized participles

A participial clause is selected by a nominalizing head that is responsible for the categorial shift from the verbal category to the nominal category. This is schematized in (56), with the nominalizing head labeled n.

1. II.Event nominals are participial clauses topped by a single nominal projection (DP)

A D head takes an extended verbal projection directly as its complement; cf. Kornfilt & Whitman (2011) and Pietraszko (2019).

1. III.Event nominals contain a silent N

Event nominals are full DPs whose lexical head, N, is an unpronounced semantically bleached noun (e.g., a silent FACT). The participial clause is merged in the complement position of the silent N.

The three approaches make different predictions. Option I leaves room for accommodating modifiers of the types restricted to the nominal domain yet associated with the nominalized predicate. This approach is challenged if event nominals are incompatible with numerals, quantifiers, and adjectival modifiers. Indeed, event nominals in Mari appear not to allow numerals (59) and modification by adjectives (60).

Option II and option III can both explain the ban on numerals and adjectives with event nominals. Under the second approach, such nominals contain only the DP layer, but none of the lower functional projections NumP, nP, or NP. Under the third approach the modifiers are associated not with the embedded clause but with the silent N head, and the semantic deficiency and phonological silence of the latter can explain the restrictions on modification.

II and III make contrasting predictions when it comes to the compatibility of event nominals with such predicates as true/false. As discussed by Moulton (2020), these predicates require an NP dependent and are not compatible with true mixed projections (as in (56) and (57)). The data below demonstrate that in Mari ‘truth’ can be predicated of an event nominal.

Taking these data into account, the third analysis, whereby the nominal distribution of -m items is due to the presence of a covert N head, is the most promising. It accommodates the relevant data and makes predictions that are borne out.

The structure in (62) reproduces what have been said so far regarding the internal syntax of Mari event nominals, in comparison to the modifying participles.

When it comes to the distribution of overt subjects in event nominals, the same licensing mechanisms are at play as in participial clauses. The only difference, I suggest, is that a saturated participial dependent of the silent N head in (62b) does not contain a CP layer that would host an operator. This renders the subject DP part of the higher nominal domain. Consequently, it becomes eligible for genitive case assignment, a pattern consistently observed in the individual grammars of both consultants. Alternatively, the subject can be a smaller nP/NP, which remains unmarked. (See section 6.1 on syntactic and semantic restrictions imposed on non-DP subjects.)

6.3 The source of INFLs

Turning our attention to INFLs, one might argue that they spell out the phi-features acquired by a non-finite T via agreement with the embedded subject, in parallel to the subject agreement in finite clauses. However, the following considerations undermine an approach whereby an INFL is directly associated with the embedded T. First, INFLs are not allowed in participial modifiers and most of infinitival clauses,¹⁸ yet these arguably contain a T head. If INFLs come from the non-finite T why is their distribution so restricted? Second, INFLs are always optional, and their presence/absence correlates with the information structure properties of the sentence and pro-drop. This is not characteristic of the subject agreement in finite clauses, which is always obligatory, regardless of whether the subject is overt or covert, contrastive or not. In addition to this, morphologically, INFLs attested in non-finite clauses are not identical to the verbal inflections in finite clauses (Riese, Bradley & Yefremova 2022); instead, the non-finite INFLs perfectly match the possessive agreement markers, as shown in Table 5.¹⁹ INFLs are not fused with the marker of the non-finite category (e.g., -mE or -mek(e)) but always attach to it, which contrasts with the behavior of subject agreement in finite clauses, where the person-number marking is regularly fused with the tense marking. While these problems are not insurmountable, they need to be carefully addressed before a T-agreement analysis is adopted.

At the same time, these issues are easily resolved if INFLs are taken to be exponents of a higher head F of the category N associated with the possessive structure (see Dékány 2011, 2018, going back to Mel'chuk 1973, on splitting a possessive phrase into a PossP and an FP). This also explains why INFLs are morphologically identical to the possessive morphemes (section 2.1). Following Burukina (2023a), I suggest that F may probe either only for A features, as in ordinary possessive phrases, or for both A features and A-bar features, as in the case of event nominals (see Lohninger 2022; Lohninger, Kovač & Wurmbrand 2022 on mixed A/A-bar probes). In the latter case it may probe exclusively DPs associated with a contrastive feature.²⁰