3.1 Morphologically segmentable converbs in Uralic

In Section 1, we saw an example of the morphologically decomposable converb in Mari (cf. (1)). As already said, this converb suffix contains the non-finite verb form -mE, which is productively used in participial relative clauses and nominalizations (see Serdobolskaya et al. 2012; Brykina & Aralova 2012; Volkova 2017; Voznesenskaia 2018; Pleshak 2022), and the illative case. There is one more converb that is clearly morphologically decomposable on the synchronic level (see Isanbaev 1961; Pengitov 1961; Alhoniemi 2010; Riese, Bradley & Yefremova 2022). This suffix is comprised of the participle -šE and the comparative case -la. Functionally, the -šə̑la-clauses express temporal overlap, as illustrated in (3). (The descriptive literature also lists the converb -meke; I will discuss it in Section 6.)

It is noteworthy that Isanbaev (1961, 71–72) argues that the suffix -šə̑la is not directly the combination of the present-day participial suffix and the comparative case but rather preserves the Proto-Uralic deverbal noun *ś, which is the origin of the Mari participle -šE, and the old allative case, which later gave rise to the comparative case (Galkin 1964, 56, 163; but see also Bereczki 2002, 36–37 on the origin of the case marker). It is also important to mention that the present-day participle -šE does not form non-finite argument clauses, unlike the suffix -mE in -me-ške. The case of -šə̑la highlights another problem that arises when one tries to determine the converb status of a given verb form, namely, diachrony. Specifically, the question is whether we find evidence that the relevant verb form, albeit having diachronically emerged from the combination of a non-finite verb plus a case marker, is to be analyzed synchronically as a morphologically non-decomposable converb. In the case of -šə̑la-clauses, Isanbaev (1961) points out that person agreement appears between the non-finite suffix and the case marker (4). This can be taken as evidence that we are dealing with a morphologically decomposable case-marked form rather than with a morphologically opaque converb suffix.

Burukina (2024) discusses not only the position of agreement in these clauses but also the general parallelism between event nominals and the alleged converbs with respect to the use of agreement and the subject's case marking (unmarked or genitive). She concludes that all three ‘converbs’, -meške, -šə̑la and -meke, are best analyzed as case-marked nominals, i.e., PPs. As said, her arguments come from the internal structure of these clauses. In Section 4, I will provide arguments that are related to their syntactic distribution.

Udmurt presents a similar picture: two of the converb suffixes discussed in the descriptive literature are clearly morphologically decomposable. These are the suffixes -(e)m-en and -(e)m-i̮ś. They both contain the non-finite suffix -(e)m, the cognate of Mari -mE, which heads participial relative clauses and nominalizations (see Dékány & Georgieva 2020 and the references therein), combined with the instrumental and the elative cases, respectively. Both clause types function as cause/reason adverbials, as illustrated in (5) and (6).

The converb status of -(e)m-en and -(e)m-i̮ś has been debated in the literature. They have been considered to be converbs (Kel'makov & Hännikäinen 1999; Winkler 2011), borderline cases between converbs and case-marked nominalizations, that is, an instance of an incomplete grammaticalization (Fokos-Fuchs 1958) or case-marked nominalizations (Georgieva 2018). Fokos-Fuchs's (1958) arguments are semantic rather than syntactic; we will return to the applicability of semantic criteria in Section 6. It should be noted that the position of agreement, i.e., whether it can interleave the two components of the alleged converb suffix, has been put forward as a diagnostic for Udmurt as well (Fokos-Fuchs 1958; Georgieva 2023). However, in the case of -(e)m-en and -(e)m-i̮ś, this test is not applicable since the agreement always follows the instrumental and the elative cases, on both noun phrases and -(e)men and -(e)mi̮ś-clauses, cf. (6). In addition to the morphologically decomposable converbs, Udmurt also employs a converb type that is morphologically opaque, the one expressed with the suffix -sa (see Georgieva 2018 for discussion).

Finally, Ylikoski's (2003) study is also relevant, as he discusses the definitions of different non-finite forms in light of Uralic data. His proposal is couched in the typological approaches that take syntactic distribution to be the decisive criterion. Ylikoski (2003, 203–206, 211–212) discusses the suffixes of the non-finite forms that are referred to as ‘third infinitives’ in Finnish descriptive grammars. They are listed in (7). These non-finite forms are comprised of the non-finite suffix -mA, which goes back to the Proto-Uralic deverbal noun *-ma, and a case suffix. The relevant case suffixes are the adessive (-llA), abessive (-ttA), illative (-An), elative (-stA) and inessive (-ssA). Crucially, the forms in (7) are all morphologically decomposable. Illustrative examples of the clauses formed with them are provided below.

Despite their morphological segmentability, Ylikoski argues that the adessive and abessive forms in (8) are to be treated as converbs and that the illative, elative and inessive forms in (9), (10) and (11), respectively, are best analyzed as infinitives. He motivates this proposal with the different function and syntactic distribution of these non-finite clauses. The ones in (8) are used as clausal modifiers (adjuncts). In contrast, the ones in (9)–(11) occur in argument position. It should be mentioned that Vainikka (1989, Ch. 5.2) investigates the argumental ones ((9)–(11)) and argues that these are locative PPs. This is in line with the proposal of the present paper. Note, however, that Vainikka's (1989) account focuses on the internal syntax of these non-finite clauses.

In sum, the Uralic data clearly show that many converbs are actually morphologically segmentable. Next, let us now take a closer look at the Turkic languages.

3.2 Morphologically segmentable converbs in Turkic

Many of the Turkic languages employ morphologically opaque converbs, the most common being -Ip, but there are also several morphologically decomposable forms. A distinction similar to the one made in the present paper is proposed for the Turkic languages by Johanson (2021, Ch. 46), who lists ‘primary converbs’ and ‘secondary converbs’, the latter being morphologically complex. His classification is largely influenced by diachronic factors, though, and, importantly, it does not make predictions about the syntax of these clauses. Washington, Tyers & Salimzianov (2022, 717) argue that converbs (‘verbal adverbs’ in their terms) should not be transparent case-marked deverbal nominals, specifically, “a verbal adverb cannot appear to be composed of a verbal-noun suffix that occurs in at least one other case form (morphologically marked or not) and a case-like form that occurs productively in the language”. This study thus takes morphological criteria into account and excludes the morphologically decomposable forms from the discussion of converbs. However, Washington, Tyers & Salimzianov (2022, 729–731) make a remark on the diachronic changes, pointing out that it is not always easy to decide if a certain case-marked form has become a converb. Let us now discuss two Turkic languages, Turkish and Mishar Tatar, in more detail.

In Turkish, we find several adverbial clauses that feature non-finite forms: the factive nominalization -DIK, the action nominal -mA, and the infinitive -mAk (see Kornfilt 1997). They can combine with various semantic cases and syntactically independent postpositions; these uses are labelled as ‘converbial’ in Göksel & Kerslake (2005, 85–88). One such form is the combination of -DIK with the locative case, which is used as a ‘when’-clause, as shown in (12). Cause/reason clauses can be encoded by the combination of -DIK with the ablative case, as illustrated in (13).

To the best of my understanding, Göksel & Kerslake (2005) do not provide explicit arguments why these clauses should be considered to be ‘converbial’. Kornfilt (1997, 66ff.) treats (12) and (13) as case-marked forms and points out their similarity to postpositional phrases. At the same time, Kornfilt (1997, 72) lists another morphologically transparent form, -DIKçA, as a converb (‘gerund’ in her terms). This suffix also contains the factive nominalization, combined with the adverbializer -çA (Göksel & Kerslake 2005, 57–58, 191). As far as I can tell, this is because the -DIKçA-clauses never show agreement (unlike (12) and (13)) and their subject, if overt, is unmarked rather than genitive-marked. (14) provides an illustrative example of -DIKçA-clauses (on their meaning see also Göksel & Kerslake 2005, 413, 418).

Mishar Tatar also employs a plethora of morphologically decomposable converbs. They contain the participle/nominalization -GAn. This non-finite verb form is used across the Turkic language family, and it can form argument, relative and adverbial clauses (see Sahan 2002; Kornfilt 2005, 2015; Asarina 2011; Ótott-Kovács 2018; Laszakovits 2019, a.o.; I gloss this non-finite form uniformly as ptcp). For example, -GAn can occur with the locative case; the combination of the two has the meaning of temporal overlap, as shown in (15).

Additionally, the same non-finite form can combine with the special marker of comparison -daj or the adverbilizer -ča, giving rise to comparative clauses, as shown in (16). The same participle can also appear with the dative case (Tatevosov, Pazelskaya & Suleymanov 2017, 350). This clause type encodes cause/reason adverbial relations, as exemplified in (17).

These non-finite forms are treated as converbs in Tatevosov, Pazelskaya & Suleymanov (2017), although their morphological segmentability is indicated in the glossary (Tatevosov, Pazelskaya & Suleymanov 2017, 20). Earlier grammars consider these forms to be based on -GAn (Zakiev 1992, 348–352) and list very few non-finite verb forms as converbs (Zakiev 1993, 223–230). But apart from the morphologically non-decomposable -Ip, the latter grammar mentions certain converbs that diachronically go back to case-marked non-finites and are partly segmentable even synchronically, e.g., the suffix -GAn-čA ‘before/until’ (see also Section 6). Generally, these sources do not emphasize morphological segmentability as an important factor. The description of converbs in Zakiev (1993) focuses on the function of these clauses (adjunct modifiers of the main clause) as well as certain properties related to the internal structure of converb clauses, such as the absence of negation, voice distinctions and agreement, and the interpretive properties of the subjects of these clauses. In my view, however, these properties do not convincingly prove that the relevant non-finite forms are to be analyzed as converbs.

3.3 Interim summary

The take-away from the previous two subsections is that several ‘converbs’ in Turkic and Uralic are morphologically segmentable: they are case-marked non-finite verbs. Whether they are treated as converbs in the existing literature, largely depends on the criteria proposed by individual authors for individual languages. As mentioned above for Finnish, syntactic distribution might be such a criterion. Agreement is another – not only its availability, as in Turkish and Tatar, but also its position: whether it interleaves the non-finite suffix and the case marker, as in Mari. However, often no explicit criteria are proposed, in which case the term ‘converb’ is by and large equivalent to ‘non-finite adverbial clause’. The situation is further complicated by diachrony: whether or not a certain case-marked non-finite form has turned into a morphologically opaque converb suffix.

In my view, the discussion in the preceding two subsections highlighted that explicit criteria are needed in order to determine if the relevant non-finite forms should be considered to be converbs. These criteria would most likely be language-specific, but the cross-linguistic comparison may also be important: as the summary above reveals, certain diagnostics have been mentioned for different languages, thus, comparing both the methods and findings can be informative.

In the next section I will propose an analysis for the morphological segmentable ‘converbs’. This analysis takes into account morphological segmentability but, importantly, it is also supported by additional diagnostics that focus mostly on the external distribution of the relevant clauses.

4.1 Proposal

In Section 3, I showed that many of the ‘converbs’ utilized in Uralic and Turkic are morphologically segmentable into a non-finite verb form and a semantic case suffix.⁵ It is received wisdom in the literature that semantic cases are P heads, and differ from syntactically independent adpositions in terms of being morphologically bound. This view is shared by both typologists (see Malchukov & Spencer 2009; Moravcsik 2009) and generative syntacticians (see the contributions in Asbury et al. 2008; Cinque & Rizzi 2010, a.o.). Thus, PPs can be headed by either a syntactically independent P or a morphologically bound one, i.e., a case suffix. This has been proposed for some individual Uralic languages, too (see Asbury 2008; Dékány 2012; den Dikken & Dékány 2018; Dékány & Hegedűs 2021 on Hungarian, Finnish and Estonian; Usacheva 2012; Simonenko & Leontyev 2012; Arkhangelskiy & Usacheva 2015; Pleshak 2019; Burukina 2023b on Udmurt and Meadow/Hill Mari). Further arguments can be brought up from a comparative perspective as well: for instance, in several Uralic languages, spatial adpositions and case suffixes show a tripartite division into Goal, Source, and Location (see Kittilä, Laakso & Ylikoski 2022 for general overview). An example of this is the triplet of the postpositions elé ‘to the front’, elől ‘from the front’, előtt ‘in front’ and the triplet of the illative, elative, and inessive cases in Hungarian. Similarly, in the Turkic languages, the Goal, Source, and Location semantics is expressed by the dative, ablative and locative cases, respectively; various postpositions, which consists of case-marked relational nouns, also show this tripartite division (see Kornfilt 1997, 242–256 on Turkish; some of these postpositions will be illustrated in Section 5.2 below). In Mari, on the other hand, case suffixes are utilized to express Location (inessive) and Goal (illative, see below in (20)), while the Source semantics is expressed with a postposition (gə̑$\acute{\stackrel{ˇ}{\mathit{c}}}$ ‘from’). This supports the idea that adpositions and case suffixes are similar from yet another angle: the spatial element, being morphologically bound or syntactically independent, heads a postpositional phrase, whose function is to encode spatial semantics. Thus, the Uralic and Turkic languages utilize both types of P heads: the syntactically independent adpositions and the morphologically bound case suffixes.

Against this background, I propose that the converb clauses discussed in Section 3 are postpositional phrases headed by a morphologically bound P head. This P head combines with a non-finite clause. This is schematized in (18).

In this sense, converbs are structurally parallel to the non-finite adverbial clauses headed by syntactically independent Ps. This line of thinking is supported by the general patterns of adverbial subordination in the investigated languages. For example, we saw that cause/reason clauses can be expressed with the combination of a non-finite form and a case suffix: the instrumental or the elative in Udmurt, the ablative in Turkish and the dative in Mishar Tatar (see (5)–(6), (13) and (17), respectively). In all three languages, cause/reason clauses can also be expressed with syntactically independent postpositions selecting the same non-finite clauses. This is exemplified in (19) for Turkish. It seems plausible to assume that (13) and (19) are parallel both with respect to their function, i.e., expressing cause/reason adverbial relations, and their structure, i.e., being PPs.

Two remarks are in order regarding the structure schematized in (18). First, throughout this paper I will remain agnostic about the precise structure of the P layer. It is well-known that PPs can have complex internal structure, comprising designated projections for path- and place-denoting elements and/or hosting elements that are at the intermediate stage between relational nouns and adpositions (see Jackendoff 1983; Zwarts & Winter 2000; van Riemsdijk & Huybregts 2002; Terzi 2005; Svenonius 2006, 2010; Cinque 2010; den Dikken 2010; Koopman 2010). We thus might expect to find variation across the languages under consideration with respect to the internal composition of the P layer. Second, the structure in (18) states that the complement of the P head is a non-finite clause, a term that I use pre-theoretically; I will not discuss the internal structure of these clauses in detail. This topic has received a lot of attention in the literature on the individual languages (see the references in Section 3). It is clear that there is a great deal of variation both within and across these languages in this respect. It seems plausible to assume, though, that the P head in (18) selects for a nominalized clause. However, its nominalized status is rather difficult to detect, as standard diagnostics like adjectival modification and pluralization are generally disallowed with event nominals in the languages under consideration. Thus, these diagnostics fail for independent reasons. Agreement and subject case assignment are not always reliable criteria either; but see Burukina (2024), who shows the parallelism between event nominals and the morphologically decomposable converbs in Mari, and suggests that converbs do contain a nominal projection.

For the reasons outlined above I consider it more fruitful to focus on the external syntax of the morphologically decomposable converbs, as it provides more straightforward evidence for the PP status of these phrases. Observe, for example, that distributionally, the relevant case markers can also select a nominal dependent. For instance, the illative case found on the Mari temporal ‘converb’ in (1) can take a DP complement (20). Likewise, the syntactically independent postposition için ‘for’, illustrated above in (19), can also combine with a noun phrase (21).

In what follows I will refer to the morphologically decomposable converbs as ‘clausal PPs’, since the complement of the P head is an extended verbal projection (cf. (18)). The aim of the next subsection will be to provide further evidence that clausal PPs show parallel behaviour to ‘phrasal PPs’, i.e., PPs, in which the P head selects a nominal dependent, as in (20)–(21).

4.2 Case studies

In this section I will present language-specific arguments supporting the PP analysis. I will zoom in on Turkish and Mari; specifically, I will discuss the non-finite clauses formed with -DIK-DA and -DIK-çA and with -me-ške and -šə̑-la, respectively.

The first diagnostic that will be applied is coordination. It will be tested whether the morphologically decomposable converbs, which I analyze as PPs, can be coordinated with phrasal PPs. The underlying assumption here is that coordination is possible between constituents of the same type. This idea is commonly referred to as the ‘Law of Coordination of Likes’ (LCL) in the generative syntactic literature (Chomsky 1957; Williams 1981); similar observations are also made in the typological studies (see Haspelmath 2007, 3). Since its original formulation, there has been a debate about the nature of the LCL: whether it is to be understood as a syntactic or rather as a semantic constraint (see Sag et al. 1985; Pullum & Zwicky 1986 and, more recently, Bruening & Al Khalaf 2020; Patejuk & Przepiórkowski 2023 for discussion). The empirical controversy is related to a limited number of well-defined classes of (apparent) exceptions to the LCL. For instance, it has been shown that PPs can be coordinated with ly-adverbs: We walked slowly and with great care (Sag et al. 1985, 140). Similar examples of coordination seem to be possible in the languages under consideration, too, as shown in (22) for Mari.⁶

It is unclear, though, whether such examples are genuine instances of coordination of unlikes. This is because of the morphological marking of adverbs in Mari. There are very few bare adverbs in Mari; the temporal ones, even the more lexicalized ones, contain a semantic case, the manner adverbs like the one in (22) are marked with the so-called instructive suffix -ə̑n, which is formally identical to the genitive case (Riese, Bradley & Yefremova 2022, 304ff). In other words, it might turn out that these adverbs are underlyingly PPs, and there is thus no violation of the LCL. Therefore, in this paper I will go with the stronger hypothesis that a syntactic constraint on the types of the constituents in coordination is indeed operative in the languages under consideration; consequently, coordination can be used as a diagnostic to determine the PP status of the converbs in question. Coordination will also be shown to obey semantic constraints, in accordance with the cross-linguistic observations (see the discussion in Johannessen 1998, Ch. 7, for example).

The second diagnostic to be discussed is suspended affixation (SA). This is a process that deletes certain bound morpheme(s) on all conjuncts other than the final one, with the suspended affixes maintaining their semantic scope over all conjuncts. This is illustrated in (23): the locative case and the plural marker have been suspended on the first conjunct and appear only on the final conjunct.

Starting with Lewis (1967), SA has attracted a lot of attention in the literature on Turkish: the phenomenon has been discussed extensively from both empirical and theoretical perspectives (see Orgun 1995; Kabak 2007; Broadwell 2008; Kornfilt 2012; Akkuş 2016, a.o.). Mari is also a language that has been argued to utilize SA; the phenomenon is much less investigated, though, with the only detailed study being Guseva & Weisser (2018).

SA can be used in order to show whether the case suffix on morphologically decomposable converbs realizes an independent functional head. If it is possible to suspend it to the exclusion of the non-finite suffix, this proves that the combination of the two is not morphologically opaque. As will be shown below, SA as a test is partly applicable to the two alleged converbs in Turkish. Nevertheless, the results make an interesting empirical contribution to the discussion of SA in Turkish, since, to the best of my knowledge, the two types of adverbial clauses have not been discussed in the studies on SA. As for Mari, it will be shown that this test is of limited applicability in the case of the two converb clauses. The purpose of reporting the findings is partly to highlight the issues related to SA that need to be further investigated: for example, it will be shown that much more interspeaker variation in SA is observed than previously discussed in Guseva & Weisser (2018). More generally, Mari is an endangered and understudied language and its morphosyntax is still a rather unexplored territory, thus, the results reported here can be informative for future studies, even though they did not prove particularly useful for the purposes of the present investigation.

4.3 Interim summary

To recap, the cases studies dealing with Turkish and Mari focused on two diagnostics. The first one was coordination. The results showed that the converb clauses formed with -DIK-DA, -DIK-çA, -me-ške and -šə̑-la can be coordinated with phrasal PPs. This constituted an argument for analysing them as PPs. The second test was suspended affixation. Together with the possibility of agreement interleaving the non-finite suffix and the case marker, SA can be used to determine whether the case marker spells out a separate functional head. This diagnostic proved to be less informative for the present study, though, mostly due to independent factors. The results are summarized in Table 1.

In the next section, I will explore some general predictions of the PP analysis.

5.1 Subcategorized PPs

Prediction 1 states that the converb clauses analyzable as PPs can be subcategoirzed by the matrix verb, on a par with phrasal PPs. Below I will present two pieces of support in favour of this.

Let us first discuss the Udmurt emen- and emi̮ś-clauses. We have already seen that they can be used as cause/reason adjuncts (see (5) and (6) above). But these clauses can also be arguments. For example, the verb gerʒ́aśki̮ni̮ takes an instrumental-marked argument, which can be a clausal one, as shown in (45). In this example, the non-finite clause does not have a cause/reason meaning; rather, the sentence means ‘something is connected with/related to something’. Similarly, emi̮ś-clauses can be selected by verbs that take an elative-marked argument, e.g., ‘something stems/arises/comes from something’ or ‘to be afraid of something’, the latter being illustrated in (46).

The distinction between subcategorized and non-subcategorized emen- and emi̮ś-clauses is not merely a semantic one. As was shown in my earlier work (Georgieva 2018, 78–80), corpus data reveal that there is a strong correlation related to the subject's case-marking: adjunct emen- and emi̮ś-clauses tend to have nominative subjects, whereas the argumental ones have genitive subjects (compare (5) and (6) with (45) and (46), respectively). It has also been shown that nominative case-marking is the common pattern in non-finite adverbial clauses; in contrast, the subjects of non-finite argument clauses have been argued to be genitive-marked (see Dékány & Georgieva 2020). Thus, the difference between (5)–(6), on the one hand, and (45)–(46), on the other, is very much in line with the observations regarding the subject's case-marking: the former are adjunct PPs and thus have nominative subjects, whereas the latter are argumental PPs and have genitive-marked subjects.¹¹

This peculiar case-alternation raises at least two questions, as pointed out by the anonymous reviewers. First, as far as clausal PPs are concerned, one might wonder how to account for the subject's case-marking. It could be hypothesized, for example, that the raising to Spec,DP that the external argument undergoes in argument nominalizations (see Dékány & Georgieva 2020) is unavailable when the nominalization is used in adjunct position and so nominative is assigned lower in the verb phrase (as a default case) (see the discussion in Georgieva 2018, Ch. 4 about the status of the nominative case (being default or structural case) in the Udmurt non-finite adverbial clauses). However, I would like to emphasize that in this paper I present the facts related to the subject's case-marking as a merely descriptive generalization, without making a specific claim about how case is assigned to the subject. Second, the question is whether we see the same case alternations in phrasal PPs in Turkic and Uralic. Indeed, the nominal dependents of Ps can be either nominative- or genitive-marked. However, this is most likely governed by other factors. Specifically, it has been shown that the nominal dependent can either sit in the complement position of P or be introduced as a possessor of a (silent) nominal complement (see the references listed at the beginning of Section 4). A recent analysis of phrasal PPs in Mari has tied the nominative/genitive alternation precisely to these two structural possibilities (Burukina 2023b). But since the case alternations in clausal PPs may be due to different factors (e.g., the (un)availability of movement to Spec,DP), it is not expected that phrasal and clausal PPs should show the same case alternations. As said above, the nominative/genitive alternation was brought up here primarily to support the distinction between subcategorized and non-subcategorized emen- and emi̮ś-clauses and the present analysis does not predict that phrasal PPs should show the same case alternation.

I would also like to address two more general objections raised by anonymous reviewers. The first one is that the argument about the PP status of the clauses in (45) and (46) would be more convincing if the predicates in question selected only a PP, but not other types of arguments. To the best of my knowledge, this does hold for gerʒ́aśki̮ni̮. The verb ki̮škani̮ ‘to be afraid’, however, can select various types of arguments: nominalizations, infinitives or finite complement clauses (see Serdobolskaya et al. 2012), just like in English. But importantly, the clausal PPs in (46) differ from finite CPs and infinitives in their formal properties: from the former by lacking a complementizer and from the latter by showing agreement. Thus, it seems implausible that all of these complements are constituents of the same type. Moreover, there is compelling cross-linguistic evidence that even if a predicate can take various complements, they are not alike, both formally and semantically, e.g., forget to vs. forget + -ing, thus, they instantiate different complementation strategies (see Wurmbrand & Lohninger 2023 for discussion). Semantic differences between the various types of complement clauses have been reported for Udmurt as well: regarding their temporal orientation, factivity, and pragmatic properties (Serdobolskaya et al. 2012; Klumpp 2016; Georgieva 2018).

The second objection concerns the morphologically non-decomposable converbs. Consider the Mishar Tatar converb -Ip, which was already mentioned in Section 3. Apart from its use as a manner/temporal adjunct, it can be selected, typically by phasal predicates, as in (47), and can also participate in various aspectual constructions (Tatevosov, Pazelskaya & Suleymanov 2017; Grashchenkov 2017).

The morphologically non-decomposable converbs in Udmurt and Mari, -sa and -n, respectively, can also be subcategorized (Serdobolskaya et al. 2012). Importantly, they can only be complements of certain phasal and modal verbs, e.g., bi̮dti̮ni̮ ‘to finish’ in Udmurt. Obviously, this does not (necessarily) mean that the morphologically non-decomposable converbs are also PPs; the two predictions explored in this section focus on the morphologically decomposable ones, in the case of which we also found independent syntactic evidence for their PP status. But I would like to emphasize that selection highlights another difference between the two types of converbs. Specifically, unlike the morphologically non-decomposable converbs, the clausal PPs are arguments only of predicates standing higher on the complementation scale, e.g., emotives like ‘be afraid’.¹² Thus, the two types of converbs have distinct distribution. This fact does not directly provide support for the PP analysis of morphologically decomposable converbs but it can be thought to indicate differences in clause size comparing to the morphologically non-decomposable converbs. It is well-known that phasal and modal predicates take complements that are small in size, while emotives take bigger complements (a similar situation is also observed with the different types of English infinitives; see also Wurmbrand & Lohninger 2023, 212ff. for a broader discussion). In sum, the fact that morphologically non-decomposable converbs can also be selected is not problematic for the present proposal.

Finnish provides further support for the distinction between adjunct and argumental PPs. As was shown in Section 3, the so-called third infinitives in Finnish are non-finite clauses topped off by a semantic case. Nevertheless, based on their syntactic distribution, some of these infinitives were analyzed as arguments, while others were argued to be converbs (Ylikoski 2003). I suggested that all of these non-finite clauses are PPs. Furthermore, they can be subcategorized or non-subcategorized. Toivonen (1995) demonstrates that the two groups show different behaviour with respect to (sub)extraction.¹³

These data show that (sub)extraction is banned from the adjunct PPs in (48), but it is allowed from the argumental ones in (49). In my view, the pattern observed cannot be satisfactorily accounted for if the non-finite forms in (48) are simply labelled as ‘converbs’ and the ones in (49) as ‘infinitives’; under such an account, the contrast would be stipulated at best. The unacceptability of (48) is rather to be explained as an adjunct island violation. In fact, the (im)possibility of extraction shown in (48)–(49) very much resembles the facts observed for English. Consider the contrast between (50) and (51):

What is puzzling is the acceptability of (51) – as it is well-known that extraction from PPs is generally not permitted – and it has been explained with the fact that the PPs are subcategorized. Thus, it seems that in both Finnish and English extraction is permitted from subcategorized PPs, as in (49) and (51), respectively, but it is banned from non-subcategorized ones, as evidenced by (48) and (50), respectively. The parallel with English is also important because the gerunds in (50) and (51) are complements of syntactically independent Ps; thus, the PP status of these constituents is indisputable. The difference is that the P head is occupied by a case suffix in Finnish. Further support for the PP analysis of the Finnish forms comes from the fact that the same verbs can subcategorize for either a phrasal or a clausal PP, as shown in (52) (see Vainikka 1989, 249–253 for more examples).

To recap, the existing literature has reported evidence for certain morphologically decomposable converbs of individual languages that suggests that they can be used as arguments. This conforms to the general behaviour of PPs and thus provides support for the PP analysis of the morphologically decomposable converbs.

5.2 PP subjects

In this subsection I will provide evidence that Prediction 2 is also born out. According to this prediction, clausal PPs are not likely to occur as subjects, unless phrasal PP subjects are also allowed. Again, the parallelism between phrasal PPs and clausal PPs will play the key role here. We thus expect that if a language disallows phrasal PPs as subjects, it will also prohibit clausal PPs to appear in subject position.

The Uralic and Turkic languages do provide empirical support for this prediction. Specifically, I will discuss two of the morphologically segmentable converbs from the case studies on Turkish and Mari, -DIK-DA and -šə̑-la, respectively. I will present further evidence that these two clause types have the syntactic distribution of PPs. It will be shown that they are disallowed in subject position, on a par with phrasal PPs. These two clause types are ideal candidates to test the validity of Prediction 2 because they feature Ps with static semantics, which gives rise to the meaning of ‘at’, ‘as’, ‘during’ in temporal clauses – as was shown above for English, PP subjects are typically attested in copular clauses and the PP has locative/temporal meaning. Furthermore, I will also discuss some Finnish data that also align with Prediction 2.

Let us begin with Turkish. The baseline is that subjects must be noun phrases (in the nominative case); the locative case-marked PPs with locative or temporal semantics are disallowed. This is evidenced by the difference in grammaticality between (53) and (54). The unacceptability of (54) is also in sharp contrast with the English PP subject construction (cf. (42) above).

Similarly to the locative case-marked PPs in (54), -DIKDA-clauses are also banned in subject position (55). In order to express the intended meaning, the factive nominalization is used as a relative clause modifying the noun zaman ‘time’ (56), thus, we have a complex noun phrase in subject position.

Next, I turn to the Mari šə̑la-clauses. Recall that they can encode temporal overlap (‘when’- or ‘as’-clauses) (cf. (3) and Section 4.2). The corresponding phrasal PPs with the comparative case mean ‘like/as’ (Riese et al. 2022, 87–90). According to the judgements of my consultants, neither phrasal nor clausal PPs with the comparative case can stand in subject position, as evidenced by (57) and (58), respectively.^14,15

Subjects in Mari need to be noun phrases in the nominative case and so the judgements improve if the intended meaning of (57) and (58) is rendered as (59a) and (59b), respectively. In these examples, the subject is a deverbal nominal modified by the phrasal PP (59a) or a complex event nominal (59b). These examples were judged as not perfectly natural by the speakers, though. The most natural ways of expressing the intended meanings is by using an infinitive, with a PP modifier, and an adjectival predicate, as in (60).¹⁶

Based on the Turkish and Mari data, I conclude that similarly to their phrasal counterparts, -DIKDA-clauses and -šə̑la-clauses cannot occupy the subject position. The anonymous reviewers correctly point out that this cannot be taken as direct evidence for the PP status of these clauses, as other types of constituents cannot be subjects either. Indeed, for languages, in which PP subjects are generally disallowed, Predicition 2 clearly demonstrates the parallelism between the alleged converbs and the respective phrasal PPs but cannot be taken as a definitive argument for the PP status of the former. Recall from Section 3 that it is possible to imagine a diachronic scenario where the original PP has turned into a(n extended) verb phrase, i.e., a converb. Hence, one might want to contrast the alleged converbs with CPs, for example: if the latter can be subjects, this could be taken as further evidence that the alleged converbs pattern with PPs, and not with CPs. At this point let us take a look at Mari. In Mari, finite subordinate clauses are introduced by the complementizer manə̑n (originally a converb form of the verb ‘say’). According to Toldova & Serdobolskaya (2014), these CPs can occur with evaluative adjectives like saj ‘good’ as well as with emotive verbs like kelšaš ‘to appeal (to someone)’; examples supporting this claim are given in Toldova & Serdobolskaya (2014, 78). However, it seems that there is interspeaker variation in the acceptability of CP subjects, as reported by Burukina (2023a, 93). My consultants either completely rejected such examples or judged them as very degraded. Thus, the comparison with finite CPs may not prove to be useful here, especially in light of the judgements of my consultants. It might be applicable in the case of speakers who accept finite CPs in subject position, but not converbs. But even in that case, the comparison with CPs might not be informative altogether – if the erstwhile PPs have been reanalyzed as verb phrases smaller than CPs, for example. For these reasons, I conclude that the findings related to Prediction 2 are valid, although not watertight. What I would like to emphasize again is the alikeness of the phrasal and clausal PPs, which is clearly confirmed by the data presented above. In my view, combined with the diagnostics discussed in the case studies in Section 4.2, the fact that the ‘converbs’ in question conform to Prediction 2 constitutes a further argument for analyzing them as PPs.

Further support in favour of the validity of Prediction 2 comes from Finnish. This will be an indirect argument, though, as it shows that PPs in Finnish can be used as non-canonical subjects rather than canonical ones, but it nevertheless very nicely highlights the parallelism between clausal and phrasal PPs.

One of the most important properties of Finnish syntax is that finite clauses, with the exception of impersonal sentences, require that a phrase occurs in the preverbal T-field, satisfying the EPP; the verb-initial order in (61a) is ruled out in neutral sentences (Vainikka 1989; Vilkuna 1989; Holmberg et al. 1993; Holmberg & Nikanne 2002). This phrase can be of different categories: an expletive, a subject, a direct object or a (locative or temporal) adverbial. Example (61b) illustrates that an adessive-marked PP can occupy the preverbal position, thereby satisfying the EPP. Importantly, since the XP in preverbal position functions as a topic, it has to be referential, and this excludes manner and sentence adverbials, for example.

Holmberg & Nikanne (2002, 84–89) argue that (i) a phrase must fill the position (Spec,FP) in the preverbal T-field (ii) both subjects and non-subjects (including PPs) move to the same position and (iii) subjects undergo A-movement, while non-subjects undergo movement that shows mixed A- and A′-properties. What is important for the present discussion is that locative and temporal PPs like the adessive-marked PP in (61b) move to Spec,FP similarly to subjects.¹⁷

With this in mind, we can test whether the clausal PPs, i.e., the non-finite clauses formed with the suffixes listed in (7), pattern with phrasal PPs. As shown below, the adessive-marked ‘converb’ in (62a) can appear preverbally, in Spec,FP, just like the adessive-marked phrasal PP in (61b). That the PP is in Spec,FP, and not higher, e.g., in Spec,CP where contrastive XPs are found, is confirmed by the fact that it follows the sentence adverbial ilmeisesti ‘apparently’ (sentence adverbials are argued to be adjoined to FP, see Holmberg & Nikanne 2002, 85ff.). The subcategorized -ma-clauses, i.e., the ones inflected with the illative, elative, and inessive cases, can also undergo the same type of movement, as illustrated for the illative-marked one in (62b).¹⁸

What these examples show is that phrasal and clausal PPs show parallel behaviour: they can both undergo the same type of movement and appear as non-canonical subjects. Together with the facts presented above for Turkish and Mari, these data provide support for the PP analysis of morphologically segmentable converbs proposed in the present paper.

5.3 Interim summary

To recap, this section explored two predictions of the PP analysis. According to the first one, clausal PPs can be arguments of the matrix verb. According to the second one, they are not likely to occur in subject position. The preceding two subsections gathered evidence from the existing literature and also presented new findings, confirming that many of the morphologically decomposable non-finites that earlier literature has treated as converbs conform to either of these two predictions. This means that these clauses pattern with phrasal PPs in terms of their syntactic distribution.

6.1 Implications for typology

Starting with the general discussion, I would like to once again stress the importance of morphological segmentability, which was the point of departure in the present paper. As was shown in Section 3, the Uralic and Turkic languages utilize ‘converbs’ that are morphologically decomposable into a non-finite suffix and a case marker. As these languages exhibit transparent syntax-morphology mapping, it seemed plausible to take morphological segmentability as an indication that we are not dealing with a morphologically opaque converb suffix. This initial finding needed to be further corroborated by additional evidence. Section 4 presented two case studies, on Turkish and Mari, in which two language-specific diagnostics were discussed at length: coordination and suspended affixation. The former dealt with the external distribution of the phrase, providing support for its PP status; the latter aimed at probing the syntactic independence of the case marker. Finally, Section 5 made two predictions based on the general properties of PPs: (i) they can be arguments and (ii) they are not likely to occur in subject position. This section also presented compelling evidence from several Uralic and Turkic languages, according to which the morphologically decomposable converbs indeed conform to either of these predictions. This confirmed that their syntactic distribution matches the one of phrasal PPs.

In the present paper I would like to emphasize the importance of the cumulative evidence based on the arguments discussed: morphological segmentability, language-specific diagnostics and conformity to either of the two predictions of the PP analysis. Hence, the first important take-away is precisely this methodology that helps to determine what kind of a syntactic phrase a given non-finite form corresponds to.

In my view, using the tools generative grammar offered a reliable and insightful way of exploring the (morpho)syntax of converbs. The findings of the present study can be fed into the typological studies dealing with converbs. As mentioned in Section 2, the empirical landscape of converb clauses seems to be quite diverse. The PP analysis proposed for morphologically decomposable converbs essentially removes them from the pool of data that should be discussed under label ‘converb’. The approach advocated here is partly aligned with Haspelmath's (1995) definition, as it takes into account morphological evidence. At the same time, my analysis puts very strong emphasis on the syntactic distribution of the clausal PPs. Hence, it is also important to reflect on the other strand of typological research, namely, the approaches that define converbs based on their distributional properties (cf. Section 2). The question that arises at this point is the following: is the PP analysis proposed here similar/identical to the ‘distribution-based approaches’? Specifically, are the same predictions made in these typological works, too? As for Prediction 1, the answer is definitely ‘no’. Recall from Section 2 that the typological studies on converbs definite them as adverbial clauses, that is, adjuncts. Thus, the subcategorized clausal PPs discussed in Section 5.1 remain unexplained (or perhaps would be treated as ‘non-canonical uses’ of converbs). In the PP analysis proposed, however, the facts observed follow naturally from the general distributional properties of PPs. Recall that Nedjalkov (1995, 97), for example, has claimed that converbs are different from both infinitives and gerunds/nominalizations, as they do not occur as arguments, including in subject and object position. This means that in the case of Prediction 2, the distribution-based typological approaches and the PP analysis advocated here converge. However, I would like to emphasize that unlike the distribution-based typological approaches, the PP analysis provides a principled explanation of why these ‘converbs’ do not occur as subjects – this follows from the general unavailability of PP subjects. More broadly, my understanding is that this strand of typological research does take syntactic distribution into account for the purposes of providing a classification of the relevant verb forms but it rarely makes predictions that follow from this classification. The present approach thus clearly differs from the ‘distribution-based approaches’ precisely because it explores the consequences of the proposed analysis.

6.2 (Apparent) counterevidence against the PP analysis

Next, I would like to discuss some issues that can be brought up as arguments against the PP analysis. They can be grouped into two bigger questions: one related to the semantics of converb clauses, and one related to phonological attrition and morphologization. I will argue, however, that these are not serious problems for the present proposal.

Let us begin with the consequences of the PP analysis for the semantics and the potential form-meaning mismatches. An important advantage of the PP analysis is that the meaning of clausal PPs is expected to arise compositionally from the building blocks of the PP. That is, morphological decomposability should go hand in hand with transparent semantics. This is born out by a large portion of the empirical data. For example, based on comparative data from the Turkic languages, Johanson (2021, 909) argues that ‘when’-clauses are often expressed with the locative case (cf. the Turkish and Mishar Tatar examples in (12) and (15), respectively). This is by no means surprising given that the locative case encodes Location (in space or time). Moreover, the phrasal PPs with the locative also had a similar meaning. As Johanson (2021, 914) further argues, cause/reason clauses are frequently encoded with the dative or the ablative (cf. examples (13) and (17)). This is also expected from the general meaning of these case suffixes. Some even more complex cases of how semantics can be derived from the subparts of the PP are discussed in Georgieva (2023). In this paper, I examined the -(o)ńńa-clauses in the Middle Cheptsa dialect of Udmurt and argued that they are not simply temporal clauses but have locative semantics: the events expressed by the main and adverbial clauses must be construed as one event, and thus they must match spatiotemporally. It is further proposed that this meaning can be accounted for based on the structure of the PP: the morphologically complex P layer consists of the inessive case plus the so-called domus suffix, a place denoting spatial element (which is independently attested in this dialect), which takes a deverbal noun as its complement. This structure gives rise to the meaning ‘in the time and place of V-ing’, as in (63).

The question that arises here is how to account for converbs that are morphologically decomposable but their meaning cannot be (straightforwardly) predicted or for cases where the meaning of the clausal PP does not match the corresponding phrasal PP. The former scenario has been said to be observed in Udmurt -emen-clauses (cf. (5)). Fokos-Fuchs (1958, 287) argues that these non-finite forms are segmentable, but their meaning is not transparent: they are translated into German with also ‘(causal) since’ rather than with mit ‘with’, which he interprets as a sign of the grammaticalization of -emen into a converb suffix. However, English with-absolutives can also have causal semantics (cf. With Peter the referee we might as well not play the match, Aarts 1992, 42), thus, this argument is not entirely convincing (see Georgieva 2018, 99 for discussion).

The latter scenario can be argued to be instantiated by the Mari -šə̑la-clauses or the Turkish -DIKçA-clauses. Recall that the comparative case in Mari means ‘as/like’ in phrasal PPs, but the clausal PPs encode a temporal overlap. This might be taken to indicate that we are dealing with different meanings. However, I propose that they can be seen as related to each other, especially in light of the fact that the comparative case is argued to go back to a spatial suffix (see Galkin 1964, 56). A similar argument can be made for Turkish as well: the adverbializer -çA has more lexicalized uses in phrasal PPs, while the respective clausal PPs have broader semantics and preserve the original spatial/temporal semantics of the P head. Similar mismatches are attested in English as well, cf. the meaning of upon, which is different when the P combines with gerunds and derived nominals (upon arriving = ‘as soon as’) compared to nominal dependents (upon his head = ‘on’). However, such mismatches are not really problematic for the analysis proposed. I would like to argue that while ‘converbs’ with compositionally derivable semantics clearly support the PP analysis, clausal PPs with less transparent meaning do not falsify it. Thus, the form-meaning argument is unidirectional and, more generally, semantics should not be the decisive criterion.

Let us now move to the second issue, namely, morphologization. In Section 3 we discussed converb suffixes that are fully decomposable. The question is how the proposed analysis deals with cases of partial segmentability. In light of the structure proposed for morphologically decomposable converbs (cf. (18)), potentially, two types of partial segmentability should be distinguished: (i) the non-finite suffix is no longer segmentable or productive, (ii) the case suffix is no longer used.

The former type of partial segmentability is attested in Udmurt, for example. The -toź-clause in (64a) contains the fully productive terminative case -oź but the segment -t is not transparent: it is argued to be a (fossilized) deverbal nominalizer (see Bartens 2000, 250, 254; Fokos-Fuchs 1958, 287, 299) or a caritive suffix (Georgieva 2018, 108–109). Given the parallelism with terminative-marked phrasal PPs (64b), though, it is safe to assume that the ‘converb’ is outwardly a PP.

A similar situation is observed with the Finnish non-finites -ma: they combine only with a few semantic cases (7), but not with the structural cases (genitive, partitive), unlike the nominalizations formed with the suffix -minen, for example. Ylikoski (2003, 203) mentions this fact as an argument against analyzing the ma-forms as case-marked nominalizations. In my view, however, there might be independent reasons why -ma-clauses never combine with other cases. For example, this could be related to the structure and the size of these non-finite clauses: as discussed in Vainikka (see 1989, Ch. 5), they lack certain functional projections. Thus, the fact that -ma-clauses do not occur with all cases in Finnish does not necessarily present a counterargument against the PP analysis.

The reverse scenario of partial morphological segmentation arises when the case suffix is no longer (productively) used with phrasal PPs. Let us consider a few examples. The ‘until’/‘before’-clauses in the Turkic languages, illustrated for Mishar Tatar in (65), pose such a problem. This non-finite form clearly contains the participle -GAn, but is considered to be a converb (Tatevosov, Pazelskaya & Suleymanov 2017; Zakiev 1993,  223–230; see also Washington, Tyers & Salimzianov 2022). Although the combination of the P head and the participle is less transparent, based on comparative evidence (cf. Johanson 2021, 757, 910–911) we can argue that -GAnčA-clauses are still clausal PPs.

A further example of this kind of partial segmentability is the converb -me-ke in Mari, which was briefly mentioned in Section 3. This suffix contains the non-finite form -mE, also found in -me-ške, combined with the lative case (Isanbaev 1961, 60). This gives rise to an ‘after’-clause, as illustrated in (66).¹⁹

The descriptive literature treats -me-ke as a converb alongside with -me-ške and -šə̑-la (Isanbaev 1961; Pengitov 1961; Alhoniemi 2010; Riese et al. 2022). The only exception is the diachronic study by Bereczki (2002, 123–124, 126, 136–137), who lists only -me-ke as a converb (határozói igenév ‘adverbial participle’ in his terms), and treats the other two as morphologically transparent forms – the decision is probably related to diachrony, no explicit arguments are given in his work. The question whether -meke-clauses should be classified as converbs seems to be rooted in the fact that the lative suffix is no longer used. It is still attested in some adverbs, though, e.g., ümba-ke ‘on top’ (< ümbal ‘surface’), jə̑ma-ke ‘down, underneath’ (< jə̑mal ‘bottom, lower part’). Given that the lative case is attested in adverbs, it can be proposed that -meke-clauses are still PPs. As such, they can be coordinated with PPs (67).

This type of partial segmentability may seem more worrisome for the present proposal, but I argue that the cases discussed above do not constitute a serious problem and shall be analyzed as PPs. First, it is often noted that diachronic processes like phonological attrition and morpholigization may obscure morphological segmentability. I mentioned three pieces of support in favour of analyzing the partially segmentable converbs as PPs: the comparative evidence within the language family (cf. Johanson 2021, Ch. 46 on Turkic), the occurrence of the relevant P head in some lexicalized forms and the coordination test. More general arguments can be based on the structure of PPs, including their history. For example, it is well-known that P heads may grammaticalize not only from relational nouns but also from adverbs, the latter being exemplified by English outside which originates from the adverb outen (Waters 2009, 296). The diachronic changes in the structure of this preposition suggest that the P layer may undergo a substantial reanalysis with multiple steps involved. This is a further piece of (indirect) support in favour of treating the partially segmentable converbs as PPs: in light of the observations regarding the diachrony of P heads, it is not surprising to find a ‘frozen’ P head that occurs in some clausal PPs but is not attested with phrasal PPs or vice versa.

In sum, the aim of this section was to reflect on certain points of the PP analysis that I put forward for morphologically segmentable converbs and to emphasize the gains of the present proposal for the typological studies. I also addressed some issues that could appear to challenge the PP analysis, such as the morphologization of converb suffixes and the form-meaning mismatches, and argued that these issues do not falsify the proposed analysis.