Object Sharing in Mbya Guarani: A Case of Asymmetrical Verbal Serialization?

In this paper, we intend to describe and discuss the grammatical status of the V1-V2 (Cy/vy) constructions found in Mbya Guarani which can express simultaneous events, among other meanings, and which involve a single clause. We suggest here that this verbal complex can be treated as a case of asymmetrical verbal serialization because it contains verbs from a major lexical class, occupying the V1 slot, followed by a more restricted intransitive verbal class, such as movement, postural, or stative verbs, which stands in the V2 position. The curious property of these constructions is that V2 can be transitivized through the attachment of applicative or causative morphemes and “share” its object with transitive V1. “Object sharing” is another property attributed to serialization, as suggested by Baker and Baker and Stewart, which may be seen as a strong argument in favor of the present hypothesis. We will also provide evidence to distinguish Mbya Guarani V1-V2 (Cy/vy) complex from other constructions, such as temporal and purpose subordinate clauses, involving the particle vy.


Introduction
The aim of this paper is to describe and discuss the grammatical status of the widespread V1-V2 (Cy/vy) construction, employed by Mbya Guarani native speakers to express one single event. This and similar verbal complexes have been regarded in the Tupi-Guarani literature as cases of verbal serialization (Jensen 1998a;Velázquez-Castillo 2004), supplementary verbs (Dooley 2013), auxiliary verbs (Harrison 1986), or, more traditionally, as gerunds (de Anchieta 1595;Martins 2004;Cabral and Rodrigues 2006;Seki 2014).
Our present proposal is to try to show that, in Mbya Guarani, these constructions constitute an instance of what has been cross-linguistically described as asymmetrical serial verb constructions (Aikhenvald 2006(Aikhenvald , 2011, namely, when a small group of intransitive verbs-movement, postural or stative-occupy the V2 slot, adding specific semantics to the clause, such as direction, manner, and aspect. Postural verbs, for instance, can provide a progressive reading (cfr. 1a and 1b): (1) a. a-ke 1SG-sleep 'I sleep'/'I slept' b. a-ke a-iko -vy 1SG-sleep 1SG-be -SER 'I am sleeping ' We have observed, however, that, in Mbya Guarani, V2 can be transitivized through the attachment of applicative or causative morphemes when V1 is also transitive. In this case, in addition to the subject, both verbs have to share the same object. Compare (2a) with (2b) where the applicative version of ainy occurs and where "basket" is interpreted as the object of V1 and V2: 1 (2) a. a-japo ajaka a-i -ny 1SG-make basket 1SG-sit -SER 'I was making a basket (seated)' b. a-japo ajaka a-re-i -ny 1SG-make basket 1SG-APPL-sit -SER 'I am making a basket, having it (seated) with me' Based on Aikhenvald's typology for verbal serialization (Aikhenvald 2006), we, following Vieira (2017), reinforce the idea that Mbya Guarani has asymmetrical serial verb constructions in which object sharing can occur between V1 and V2. This latter property is, according to Baker (1989) and Baker and Stewart's diagnosis (Baker andStewart 1999, 2002), an essential characteristic of Serial Verb Constructions (hereafter SVCs). It is the occurrence of internal argument sharing between V1 and V2 that made us suggest that the verbal complex under investigation here can be regarded as one type of serialization.
Nonetheless, additionally, this paper includes a review of what has been traditionally labelled as SVCs in Tupi-Guarani languages. The field literature treats simultaneity and purpose constructions involving the particle vy in Mbya Guarani as instances of the same basic SVC configuration. However, we intend to provide morphological, syntactic, and semantic evidence that show that they do not reflect the same phenomenon, and that the morpheme under discussion is a suffix in SVCs and a subordinative conjuction in dependent purpose clauses (Section 6.2). We will also distinguish serials from other structures, such as coordination and other types of subordination, based on the data collected among native speakers during fieldwork carried out in Argentina (Misiones) and Brazil (São Paulo and Paraná).
This paper is organized as follows: Section 2 provides a brief summary of the main features which identify SVCs in the world languages; Sections 3 and 4 present some phonological, morphological, and syntactic aspects of Mbya Guarani, relevant for the understanding of the data under discussion; Section 5 reviews some of the Tupi-Guarani literature on verbal complexes; and Section 6 discusses the analysis assumed here.

On the Definition of SVCs
In the field literature, SVCs appear to function as a pre-theoretical umbrella term (Haspelmath 2016), historically described by different sets of properties that often combine both syntactic/morphological and semantic criteria, as the items in (3) can show and example (4) can illustrate: (3) SVC properties (Aikhenvald 2006): (a) they involve more than one verb; (b) there is no linking morpheme between the verbs involved; (c) they are mono-clausal; (d) they share identical TAM and negation values; (e) they share one S argument; (f) they may share, eventually, an O argument as well; (g) they describe one single event; (h) they present one single intonation contour.

Yoruba
(4) Ó mu iwé wá he brought book came 'He brought back the book' Therefore, Alexandra Aikhenvald (2006) suggests a continuum-type approach to this phenomenon where constructions can be described by resemblance to a prototype characterized by the numerous features listed above. The author defines serial verb constructions as "a sequence of verbs which act together as a single predicate, without any overt marker of coordination, subordination, or syntactic dependency of any other sort" (Aikhenvald 2006, p. 1). Haspelmath (2016, p. 292), on the other hand, advocates for a narrow definition of SVCs, more suitable for comparative purposes: "a mono-clausal construction consisting of multiple independent verbs with no element linking them and with no predicateargument relation between the verbs". Hence, the author excludes from his definition constructions that have been historically regarded as SVCs, such as: (i) non-compositional verb combinations, as they do not follow regular schematic patterns and, hence, do not fall within the construction definition; (ii) auxiliaries and adpositions, as they cannot occur without another verb; and (iii) causative and complement-clause constructions, as one of the verbs involved is necessarily (part of) an argument of the other verb.
Following Bohnemeyer et al. (2007, p. 501), Haspelmath (2016 points to the relevance of negation when defining clausehood. For the author, single negatability is the only appropriate criterion to recognize a single clause from a cross-linguistic perspective. This is an important feature for identifying verbal serialization, as serial verbs occur in the same clause and negation has scope over all of them. One very good example of a SVC is, in fact, provided by Aikhenvald's description of Tariana, a northwest Amazonian language. In the sentence below, (5), we can see how Tariana speakers make use of five different verbs in order to describe a set of sub-events that conform to one single event, while the translation consists of only one verb due to the lack of SVCs in languages like Portuguese and English. This difference is, in fact, emphasized by native speakers of Tariana: "during early fieldwork sessions, one of the consultants remarked: "It is not like Portuguese, we just cannot say it with one verb'" (Aikhenvald 2003, p. 428): 3SGNF-go.out 3SGNF-come 3SGNF-ararrive 33SGNF-S sit REM.P.REP 'The bacurau-bird arrived here (from there)'[lit: he flew-he went-he came-he arrived-he sat] (Aikhenvald 2003, p. 425) Aikhenvald also makes a distinction between symmetrical and asymmetrical SVCs, based on the types of verbs involved. In symmetrical SVCs, all verbs come from an open class, such as "cook" and "eat" in (6). In asymmetrical SVCs, the first verb (V1) belongs to an open class, while the second verb, V2, belongs to a restricted class, being often a motion or posture verb, providing aspectual, manner, or direction values to the construction as a whole, as the verb lai 4 "come" in (7) can illustrate: FUT.cook thing eat 'Áma is going to cook and eat' (Aikhenvald 2006, p. 28 lei 5 lo 2 di 1 saam 1 lai 4 you take PL clothing come 'Bring some clothes' (Aikhenvald 2006, p. 21) There is also another important property attributed to verb serialization: "the sharing of objects" between the verbs involved, as suggested by Baker (1989) and Baker andStewart (1999, 2002) to account for cases like (8), from Edo, a Niger-Congo language of West Africa. In reality, each of the transitive verbs involved in the serialization selects its own internal argument. Notice that in (8), the objects of both verbs share the same referent, but only one of them can be lexically realized: the internal argument of V1. The internal argument of V2 is represented by a null pronoun (pro): Ozo is going to kill bode i (and) sell (pro i )' (Baker and Stewart 2002, p. 2) According to the authors, "the sharing of objects" in a SVC construction like (8) is a property of languages in which vP can have two heads. Baker and Stewart identify three types of SVCs, each one derived from a configuration in which one of the following functional projections can be double-headed: VoiceP, vP, or VP. The type represented in (8), called Consequential Serial Verb Construction, is derived from the possibility of adjunction of vP2 to vP1. This is the parameter responsible for the occurrence of "object sharing" in SVCs.
On the one hand, Aikhenvald's distinction comes especially handy for our analysis, as the asymmetrical type seems to describe the type of verbs involved in Mbya Guarani V1-V2 (Cy/vy) complex. Also, Baker and Stewart's hypothesis seems to explain the transitivity restrictions observed between the verbs in these structures. On the other hand, Haspelmath's proposal highlights the relevance of negation when defining and identifying clausehood (Haspelmath 2016, p. 299). This last point is also relevant for our analysis since negation proves to be one of the most useful tools to distinguish SVCs in Mbya Guarani, which are monoclausal, from other similar constructions which involve more than one clause. In the former structures, there is only one way to form the negation, marked on V1 with scope over all the verbs within the same clause.
We have seen, then, that SVCs are of different types, classified according to the nature of the verbs involved and to the syntactic configuration from which they are derived.
In the next two sections, we provide some pieces of information about Mbya Guarani language which are relevant for the understanding of the data to be presented and of the phenomena under discussion.

About Mbya Guarani Dialects
According to Rodrigues (1958), Mbya is included in the Guaranian languages subgroup of the Tupi-Guarani branch of the Tupi stock, along with other genetically related languages, such as Paraguayan Guarani, Nhandeva, Kaiowa, Chiriguano, Aché, Xetá, and Izoceño. Being the most widely spoken language of the Guarani group, Mbya has roughly 27,000 native speakers distributed in Argentina, Brazil, and Paraguay (Ladeira 2020).
The data discussed in this paper were collected by the authors during several field trips in different communities in Brazil (Paraná and São Paulo) and Argentina (Misiones).  (El Alcázar, Misiones). During each fieldwork trip, semi-spontaneous texts were collected and, afterwards, several elicitation sessions were conducted in order to test specific features of the constructions discussed in this paper. Mbya Guarani transcriptions and Portuguese and Spanish translations were made in collaboration with bilingual native speakers both in Brazil and Argentina 2 . The orthography employed for the transcription of the primary data cited in this paper follows the conventions used by native speakers. Mbya has six oral vowels: a (written a), ε (e), i (i), o (o), u (u), 1 ɨ ã, ẽ, ĩ, õ, ũ, ỹ (y), and their nasal counterparts, written: ã, ẽ, ĩ, õ, ũ, ỹ. The consonants are: p (written p), t (t), k (k), m (m or mb before oral vowels), n (n or nd before oral vowels), . Along with elicited data, secondary sources were also consulted for comparative purposes, in order to illustrate specific phenomena both in Mbya and other languages of the Guarani group, in which case the reference follows the translation.

Grammatical Aspects of Mbya Guarani
Mbya Guarani has a rich verbal morphology where the arguments are cross-referenced and some grammatical categories, such as mood and negation, can be expressed. It is a null subject and a null object language which can exhibit both SOV and SVO word orders when the core arguments are lexically realized:

Verbal Morphology
Tupi-Guarani languages have an active/non-active split system and their verbal morphology expresses the person and number of the subject and/or the object. Mbya presents two sets of verbal prefixes employed in the indexation of arguments (Table 1). On a general basis, prefixes from Set 1 index the A argument of transitive verbs and the S A argument of intransitive active verbs, while Set 2 prefixes index the P argument of transitive verbs and the S P arguments of intransitive stative verbs.
In transitive constructions, the choice of the affix to be expressed on the verb follows the referential hierarchy 1 > 2 > 3, meaning that the P arguments are cross-referenced by Set 2 prefixes when a third person subject acts upon a 1st or a 2nd person object (cfr. 10a and 10b) 3 . The language retained a portmanteau prefix, originally taken from the co-referential indexation paradigm (Set 3) reconstructed for the proto-language (Jensen 1998b), that is employed in main clauses when the referential hierarchy is inverted, meaning when a 1st person acts on a 2nd person (11): Besides the presence of personal affixes, the verb can also be marked with future 4 and desiderative morphemes, as in (12). Other tense markers, such as past and present, are not represented in the verb morphology. Negation is expressed either by the suffix -e'ỹhe'ỹ~-he'ỹre or by a discontinuous morpheme. The former is found in nominalized and subordinate clauses (13a) and the latter, in main clauses (13b): Intransitive verbs can change their valence through the adding of causative or applicative 5 morphemes. The causative prefix licenses an external argument, while the applicative introduces an internal argument interpreted as a comitative. Transitive verbs can become intransitive by the adding of the reflexive/reciprocal affixes:

Coordinate Clauses
In other Tupi-Guarani languages, when the subject of the main clause and the subject of the following clause are coreferential, the second verb assumes a specific form referred to as "gerund" in the literature. The gerund changes the verb's form through the adding of special suffixes and employs the absolute cross-referencing system. If the second verb is intransitive, it is marked with coreferential personal prefixes referring to the subject. If it is transitive, only the person markers associated to the object are employed, as the following data from Tocantins Asurini illustrate. The gerund can be translated into Portuguese as a gerund, temporal and purpose clauses, or even as coordinate clauses: Similar to many other genetically closely related languages, Guarani has lost its gerund forms (Jensen 1998b). Due to this fact, coordinate and subordinate verbs are expressed as finite verb forms, identical to those of independent clauses. Coordinate clauses can employ the conjunction (ha'e/ha) which is optional. Note that in (18), each verb can exhibit its own morphology and select its own arguments independently: We will kill the jaguar and bring its skin'

Subordinated Clauses
Subordinate clauses have special markers in Mbya, depending on their semantic type. Purpose dependent clauses can be headed either by aguã or vy. The latter is only used when the subjects of both clauses have the same referent: (19) a-ata [mbojape a-Ø-jogua aguã/vy 1SG-walk bread 1SG-3-buy CONJ 'I walked in order to buy bread' In temporal subordinate contexts, vy is also employed when there is co-referentiality between the subjects, as illustrated in (20), but when the subjects have different referents, ramo is used instead: 3-see When the man was going, he saw a snake' (Dooley 1988, p 3-see 'When the man was going, the snake saw him' (Dooley 1988, p. 97) According to the literature, Mbya Guarani is the only language in the Guarani group that exhibits a switch-reference system. Robert Dooley (1988) describes the subordinating conjuction vy as a same-subject marker (SS), which occurs in dependent clauses whose subject is co-referential with the subject of the main clause, as opposed to ramo, the differentsubject marker (DS), which occurs when the subject of the subordinate and main clauses are not co-referential. The pair of examples in (20) above clearly shows how the switchreference system works in Mbya Guarani.
In the section below, we will present a brief review of the analyses found in the literature about the existence of verbal serialization in Tupi-Guarani languages.

It Runs in the Family: SVCs in Other Tupi-Guarani Languages and in Mbya Guarani
Instances of SVCs in languages of the Tupi-Guarani (TG) group have already been postulated to occur by Cheryl Jensen (1998a) and in Paraguayan Guarani, by Maura Velázquez-Castillo (2004). Lucy Seki (2014) describes a similar phenomenon in Kamaiurá, the gerund construction. However, the latter author is reluctant to label it as SVC, given that they present specific syntactic constraints that would be disregarded under such a broad label.

The Gerund Constructions
According to Jensen (1998a, p. 17), in Tupi-Guarani languages, verbal serialization can be found in constructions where either V1 or V2 belongs to the closed class of movement, postural, or state verbs. V2 in these cases occurs in the gerund form 6 . For the author (Jensen 1998a, p. 17): "in Tupi-Guarani languages, an action or a series of actions having the same subject may be perceived as part of a single event and expressed as a series of verbs in a single clause". The following sentences are examples from Tupinambá and Tocantins Asuriní, provided by Jensen, where she analyses the gerund suffix -a as a serializer attached to V2: He came to see the boy' (Jensen 1998a, p. 17) Tocantins Asuriní (22) ere-ha e-seegat -a 2SG-go 2SG.COR-sing -SER 'You (SG) went singing' (Jensen 1998a, p. 24) Jensen describes serial verbs in TG languages as "a verb which appears together with an independent verb to express simultaneous action, purpose or sequential action, when the subject of both verbs is identical" (Jensen 1998a, p. 24).
The author, however, seems to agree with Nicholson's analysis of Tocantins Asurini complex verbal constructions in which V2 belongs to a closed class of intransitive verbs, such as movement, postural, and stative, and do not exhibit the gerund suffix. For Nicholson, V2 in these cases should be treated as an auxiliary verb which conveys direction, position or progressive meanings 8 : (2014), on the other hand, argues that Jensen's definition is too broad, as it may include different constructions in Kamaiurá that clearly do not involve serialization (i.e., same subject subordinate clauses). The author distinguishes four different types of gerund constructions in Kamaiurá, depending on the types of the verbs involved. Especially relevant for our analysis are Seki's types (i) and (ii), presented below:  (Seki 2014, p. 694) Seki claims that type (i) constructions can receive either a purpose or a simultaneous reading. However, they appear to be subordinate clauses in Kamaiurá rather than SVCs, due to the presence of the linking morpheme which, in these cases, is the gerund marker, and to the fact that both verbs can be negated independently (Seki 2014, p. 698). Regarding type (ii), the author states that V2 resembles an auxiliary rather than a serial verb, given the lack of contiguity among the verbs, as they admit the presence of other elements between them. Aside from that, V2 in the (ii) type can convey different types of aspectual semantics: durative, progressive, and even inceptive. This latter analysis is the same suggested by Nicholson (1975) for Tocantins Asurini, where V2 is taken from a closed verbal class, as expressed in (26): The problem with the auxiliary analysis is the fact that V2 can be transitivized in Tocantins Asurini, as pointed out by Vieira (2002Vieira ( , 2017, and as illustrated by (27), where intransitive aka in (26) receives the applicative morphology, turning itself into a transitive verb: By definition, auxiliary verbs do not select arguments and so, there is no reason for aka in (27) to get transitivized. Due to this valency-changing possibility, we do not agree on labeling V2 in these structures as auxiliaries.
Velázquez-Castillo (2004) analyzes as SVC the -vo construction in Paraguayan Guarani, which, according to her "presents two distinct patterns: one involving a sequential interpretation with purposive overtones and another with a simultaneous interpretation" (Velázquez-Castillo 2004, p. 193), as (28) and (29) illustrate, respectively: He slept as he went (the whole way) to his hometown' (Zarratea 1981, as cited in Velázquez-Castillo 2004 The author's proposal is that, given the structural similarities between the simultaneous and purpose -vo, they should be treated as different occurrences of the same serial construction. Along the same lines, we encounter Cheryl Jensen's analysis of SVCs in different languages of the Tupi-Guarani group.

Finitization of Gerunds and SVCs in Emerillon
Rose (2009,2013) distinguishes between SVCs and gerundives in Emerillon. The former is seen as an innovation resulting from a shift in the language's indexation system and the loss of dependent/subordinate markings. According to the investigator, gerundive forms underwent a finitization process where they gained finite features and became independent predicates. Emerillon SVCs semantics include sequential actions, motion, and direction. Their re-ranked nature (Rose 2009) resides in the fact that they have lost the serial suffix and the verbs present an active-inactive indexation system (30a), as opposed to Emerillon gerundives that retain the prefixes from Set 2 for argument indexation (30b). As Rose points out, this finitization process has not affected the transitive gerundives yet. Observe the data below in which intransitive V2 takes the prefix from Set 1 (active), as independent verbs do, while transitive V2 keeps the marking of its object through the person series extracted form Set 2 (non-active): 3.I-make -COMPL 3.II-make.go metal He thus finished to make his weapon (by streaching out a piece of metal) (Rose 2009, p. 48) Dooley (1991Dooley ( , 2013 analyzes the same Mbya Guarani verbal constructions under investigation here, such as (31) below. V2s are compared to gerunds (Cabral and Rodrigues 2006 and to dependent serial verbs (Jensen 1998b), but the author chooses to call them supplementary verbs (Dooley 2013, p. 69):

Supplementary verbs in Mbya Guarani
Mbya Guarani has lost its gerund forms. According to Dooley (2013, p. 69), vy can occur attached to what he calls supplementary verbs in the V2 slot, similar to the gerund forms of the other Tupi-Guarani languages. -Vy has four variants (-py, -my, -ny) due to the nasalization or to the loss of a final consonant of the root. As pointed out by the author, there is also a vy form which marks subordinate clauses of the adverbial and the purpose types when main and subordinate subjects have the same referent.
In relation to the personal morphology, supplementary and subordinate verbs follow the same rules of independent clause verbs, not the absolutive pattern found in the original gerundive forms, employed in other languages of the group. Dooley also observes that all of these supplementary verbs can undergo a process of transitivization, as example (32)  Although the author finds similarities between these supplementary structures and SVCs, he refuses to treat them as instances of serialization (Dooley 1991, p. 33): "The Mbya [V1-V2] constructions turn out to behave like SVCs in its semantics and in some syntactic aspects. However, in four respects it is grammatically tighter than stock SVC. Dooley prefers to treat V2 as a special type of subordinate verb due to the following facts: (i) The V2 has an identifying suffix. (ii) The V2 has a distinct, reduced agreement marker. (iii) The V2 is required to have the same subject and, if transitive, the same object as V1. (iv) The construction is virtually impervious to the occurrence of arguments between V1 and V2.
According to Dooley (2013, p. 73): "supplementary verbs have the suffix -vy which marks its syntactic dependency." There is a much tighter dependency link between V1 and V2, rather than in regular serial verb constructions, maybe because of the "fairly high index of synthesis" of the language.
For the investigator, supplementary and subordinate verbs differ both syntactically and morphologically, due to following facts: (i) the highly grammaticalized supplementary verbs are included in the main predicate, while the subordinate clauses function as a peripheral adjunct; (ii) while the supplementary verbs show a fairly fixed position, the order of a subordinate clause and the main clause may vary for information structure purposes; (iii) in the first case, the morpheme -vy constitutes a verbal suffix, attached to the radical, while in the latter case, it takes the form of a conjunction encliticized to the subordinate clause; and (iv) supplementary verbs cannot introduce new arguments, while subordinate verbs lack this constraint (Dooley 2013, p. 71).
Summarizing thus far, on one hand, from the broad approach adopted by Jensen and Velázquez-Castillo, the purpose and simultaneous constructions in Tupi-Guarani languages, expressed by the so-called gerund forms, should be both analyzed as SVCs. On the other hand, in order to avoid this approach, Seki ends up with a definition so narrow that it is not applicable to describe most of the gerund constructions she encounters in Kamaiurá. Our proposal, then, lies between these two approaches, as we argue in favour of the existence of SVCs in Mbya Guarani, but we suggest two different analysis for the simultaneous and purpose constructions involving the morpheme Cy/vy in the verbal complex, regarding only the former as a case of SVC and the latter as a case of clausal subordination.

Asymmetrical SVCs: The Case of Mbya Guarani
As we have already mentioned above, the definition of serial verb constructions is too broad and many different types of SVCs are identified in the literature. For the analysis of Mbya Guarani V1-V2 (Cy/vy) complex assumed here as verbal serialization, we have chosen as the main defining characteristics the existence of asymmetrical serialization (Aikhenvald 2006) and of object sharing (Baker 1989and Baker and Stewart 1999. In order to prove that both verbs are projected in the very same clause, we rely on negation tests, among others, as suggested by Haspelmath (2016).
In the following sections, we will provide a description of Mbya Guarani SVCs, including evidence for the object sharing property and for the mono-clausal status of this verbal complex. We will also focus on the distinction between serial and subordinate constructions involving the vy morpheme.

Asymmetrical Serialization: Basic Characterization and Object Sharing Restrictions
In the following sub-sections, we will describe Mbya SVCs according to Aikhenvald's parameters presented in Section 2.

One Single Event
SVCs are employed by Mbya Guarani speakers to express one single event and involve a verb from an open class (V1) followed by a closed-class verb (V2), which can express movement, position, direction, and continuity of actions. Both verbs share the same subject. Regarding the indexation of arguments, Mbya Guarani SVCs generally present the same indexation system as that of independent verbs, as can be observed in (33-34) below, where both V1 and V2 receive the personal prefixes from the same paradigm (Set 1): (33) Ha As can be inferred from the translations of the sentences above, Mbya Guarani SVCs ultimately describe one single event. The acts of singing and of planting, for instance, took place at the same time as the act of going away and of coming.

On the Finite and Independent Nature of V2
In Mbya Guarani, V2 belongs to a restricted verbal class and adds specific semantics to the construction, such as direction, manner, and aspect. We regard Mbya V2s in these SVCs as independent verbal forms, given the fact that most of them also occur as main predicates. 9 The following (Table 2) lists the forms of V2 roots found in Mbya SVCs:

Independent Verb Forms V2s in SVCs
'ã (to stand up) 'a-my (positional) ĩ (to sit, to be located, to exist) i-ny (positional, progressive aspect) (i)ko (to be, to walk, to roam) (i)ko-vy (continuous action, movement) kua (to be all together) kua-py (be all together, to act uniformly) o (to go) o-vy (directional, progressive aspect) (j)u (to come) (j)u-vy (directional, progressive aspect) (j)upy (to lie down) (j)upy (positional, perfective aspect) The data below in (35) provide evidence for the independent status of Mbya V2s as they can occur as the only predicate in an independent clause: ne-akã ngyta rejupy 2SG-head pillow 2SG-lie.down 'You lie with your head on the pillow' (Dooley 2013, p. 78) As the data above show, most SVC verbs can also occur as main predicates without the serializing suffix, except (j)upy 'to lie down' which, although can occur as an independent verb, still retains the serializing suffix.
In addition to that, when functioning as V2s, they are marked by the same rules employed in main clauses verbs. That is, these verbs receive the same set of personal affixes as any independent verb 10 , with a very few exceptions. For instance, when a 1st person subject acts upon a 2nd person object, V2 in SVC also gets the portmanteau prefix (1>2), as any finite, autonomous verb in the language. Observe in (36) that both verbs are affixed with the same morpheme. The portmanteau prefix is used only in finite and independent verbs: (36) Oro-jou oro-guer-u -vy 1>2-find 1>2-APPL-come -SER 'I found you and brought you with me' (Dooley 1991, p. 47)

One Single Clause
Another property of SVCs is that the verbs involved must occur in the same clause. This explains the fact that there must be only one TAM and negative morphemes.
In Mbya Guarani, there is just one temporal marker to express future meaning. This morpheme can be attached only to V1, as the ungrammaticality of the (37b) example can illustrate: all -SER 'They will be all dancing' b.
* o-jeroky o-kua-py-ta 3-dance 3-be.all-SER-FUT 'They will be all dancing' Sentential negation in Mbya Guarani is constructed by means of the cirfumfix n(d)-...-i, which gets attached to independent verb forms. In SVCs, the negation circumfix is placed on the main verb, V1, and its scope semantically affects the whole construction, only V1 can be negated. When V2 is negated, the construction is not accepted, as the ungrammatical status of (50b) and (50c) show. The whole verbal complex cannot receive the negation circumfix, (50d), and V2 cannot take the standard negation used for dependent clauses, as in (50e) which our consultants consistently rejected: Another piece of evidence that led us to the conclusion that V1 and V2 are in the same clause comes from the fact that the arguments selected by V1 can show up to the right of V2, as (39)  We would like to notice here that V2 cannot be viewed as an auxiliary verb in these constructions due to the fact that it can be transitivized, as (40) indicates, when compared to (39). Auxiliary verbs do not select arguments, and so there is no reason for V2 to turn itself into a transitive verb. In addition to that, if V2 were auxiliary, it could also carry some verbal morphology, such as tense and aspect markers, but it does not: (40) Xee a-ra-a a-re-ko -vy xe-r-a'y teko'a -py I 1SG-APPL-go 1SG-APPL-be -SER 1SG-REL-son village -LOC 'I am taking my son to the village, having him with me' 6.1.4. The "Sharing" of the O Argument with Transitivized V2 In her previous work, based on Aikhenvald's (2006) hypothesis regarding the existence of asymmetrical SVCs in natural languages, Vieira (2017) suggests that even though V2s in Mbya Guarani's SVCs do belong to a semantically and grammatically restricted class, they do not match the criteria to be labeled as light or auxiliary verbs, as they can undergo valency-changing operations. Additionally, the author states that the "object-sharing" constraint (Baker (1989) and Baker andStewart (1999, 2002)) is an important and salient feature to classify these constructions as SVCs.
In Mbya Guarani, V2 may become a transitive verb when it receives causative or applicative morphology. This valence change is licensed only when V1 is also transitive. When these transitivity features match, both verbs must select two different objects, but with the same referent. While the object selected by V1 is lexically realized, the object of V2 must be null-an empty category (pro). The following sentences illustrate well the "object-sharing" constraint. In (41a), V1 and V2 are both intransitives and the sentence is grammatical. In (41b), only V2 gets transitivized and the sentence is ruled out. In (41c), where the causative prefix mbogets attached to V1 and V2, turning them into transitive predicates, the sentence is well-formed again: The same scenario is seen in the data in (41c), (42b), (43) and (44); where we can notice that there is also a matching in the kind of transitivizer each verb can take. If V2 gets the causative prefix, V1 also gets transitivized by the same morpheme. If V2 gets the applicative prefix, so does V1. Also observe the transitivity restriction in (42a) which is ungrammatical because V1 is intransitive and so, there is no context for transitive V2 to "share" its object with the same referent: Transitivity restrictions have been, then, proven to be the key in defining the syntactic status of these constructions, and also to distinguish them from adverbial subordinate clauses, as stated by Dooley (2013, p. 71). The problem with Dooley's analysis is that he does not treat such constructions under investigation here as cases of verbal serialization. For the author, the so-called "supplementary verbs" are a type of subordinated constituents, very tight to V1. One piece of evidence that he gives for his hypothesis comes from the fact that no element can stand between V1 and V2 due their syntactical dependent nature (Dooley 1991). We have found examples, however, which show that many kinds of constituents can interfere between them, as the object in (45) and (46) In reality, the object in these SVCs can show up in any position because all the constituents are in the very same clause, as exemplified by (47): a-Ø-'u pira a-a -vy 1SG-3-eat fish 1SG-go -SER 'I ate fish going' [V1 O V2] Another feature that the author calls attention to in relation to Mbya Guarani supplementary verbs comes from the fact that when V2 is transitive and the object it selects is 1st or 2nd person, the personal prefix chosen refers to a 3rd person. This is another reason Dooley (1991Dooley ( , 2013 gives not to consider them as serial verbs, but as a tigher type of subordination. Observe in (48) that the verbs "to feed" and "to make sit" choose a 1st person object which is only lexically expressed on V1. In V2 morphology, a third person object prefix is placed instead: The man fed me by making me sit' (Dooley 1991, p. 52) We would like to suggest here that the reason why V2 does not get the 1st person object marker comes from the fact that the object xe is a clitic pronoun, not an instance of agreement. Being so, it cannot be present twice in serializing contexts where there is object sharing. Remember that the object of V2 must be null. This way, as the verb in Mbya Guarani cannot occur in its plain form, a third person prefix is used instead, as a default form. In the case of the portmanteau prefix illustrated in example (36) above, where the features of both subject and object are expressed on both V1 and V2, we suggest that it is possible because this morpheme is an instance of agreement. It is not a pronominal argument, as the 1st person clitic is. Summarizing, the possible transitivity matches found in the V1 V2 Cy/vy verbal complex are: (49) Transitivity matches a. V1 intransitive V2 intransitive b. V1 transitive V2 intransitive c. V1 transitive V2 transitive 6.1.5. Absence of a Dependency Marker As both Aikhenvald (2006) and Haspelmath (2016) state, the absence of a linking element is a key feature to identify SVCs In the following section we will deal with the status of the serializing suffix in Mbya, noting that serializing suffixes have been already documented in other languages of the world (Aikhenvald 2011). 6.1.5.1. On the Distinction between Serial and Subordinate Constructions Involving vy Cabral and Rodrigues (2006) track down the origin of vy morpheme to earlier nominalizations involving the suffix *-áp~*-táp and the locative *-βo, and stress the fact that the outcome of this grammaticalization process may trigger a "sequence", "purpose", or "simultaneous" reading. Nonetheless, the -vy suffix has extended its domains nowadays to expressing different types of relations between predicates, as can be observed in the Mbyá Guarani data below: All sentences presented above involve two different verbs with a shared subject. However, while in (50) the events expressed by the verbs involved develop in the same time and space coordinates, overlapping completely and, hence, describing one single event, in (51-53) we find ourselves dealing with a very different state of affairs. The latter examples involve two easily distinguishable events, henceforth, two distinguishable clauses, where vy serves as a subordinate conjunction that expresses different kinds of semantic nexus between the dependent and the matrix clauses.
This morpheme can trigger a temporal reading in (51), where the dependent clause expresses an event that took place at the same time as the event presented in the main clause; a purpose reading in (52), where the dependent clause expresses the purpose of the event in the main clause; or even a cause reading in (53), where the event depicted in the dependent clause is understood as the cause or reason of the state of affairs presented in the main clause. From a cross-linguistic perspective, it is not uncommon to encounter temporal, purpose, and cause semantics being expressed by the same grammatical means (for further details see Cristófaro 2005, p. 161, on Ancient Greek adposition hös). However, it has become evident that constructions like (50) present different morphological and syntactic restrictions from the constructions presented in (51-53). 6.1.5.2. On the Status of Cy/vy: Suffix and Conjunction Having described the different phenomena involving the vy morpheme in Mbya Guarani, our proposal is that there are two syntactically distinguishable constructions involving the aforementioned form in this language: (i) SVCs-vy, which conveys a simultaneous reading or two aspects of the same event, as in (50); and (ii) the subordinating vy clauses, which can trigger either a temporal (51), a purposive (52), or a causal reading (53).
Aikhenvald's parameter (3b) presented in Section 2, that is, the absence of a linking morpheme, could be regarded as problematic for our proposal here, given that the vy morpheme could be regarded as a subordinating element between V1 and V2. However, if we take a closer look at the forms involving the Cy/vy morpheme in Mbya Guarani SVCs which express different aspects of the same event from an historical perspective, we can see that the suffix has experienced a process of fossilization over time, since it only remains attached to the members of this very small closed set of verbal forms, most of which fell into disuse as independent roots, resembling more a dummy marker (Aikhenvald 2011, pp. 22-23) than a subordinate linking element.
The -vy suffix is included in a much larger group of serializers found in Mbya Guarani that exhibit the -Cy form, its allomorphs originally being -ngy (54a), -py (54b), -my (53c), and -ny (53d), depending on the phonological environment in which it occurred (Dooley 2013, p. 69). These suffixes today are highly grammaticalized and have lost their productivity, meaning that they only attach to this closed set of verbal forms in these particular contexts: (54) a.
I-mba'eaxy oupy 3-sick 3-lay.down.SER 'He is sick in bed' (Dooley 2013, p. 70 In the data provided in (54), we can observe how these verbs contribute to the constructions as a whole, adding positional semantics (54a, 54c) manner (54b), and progressive aspect (54d). With this in mind, we can establish a key distinction between the serial -vy and the subordinative vy; while the former constitutes a verbal suffix, with different allomorphs depending on the phonological context in which it appears, the latter is an invariable free form. Evidence supporting our argument is that, in subordinate clauses headed by vy, other elements, like poã, the object, in (55), and voi, an adverb, in (56), can intervene between the dependent verb and the conjunction, while this has been proven to be ungrammatical in SVCs, as (57) (58) and (59) where V2 is used to provide a continuous aspect inside a subordinate clause headed by the conjunction ramo (DS) and aguã inside a main clause: In addition to data like these, we have also found examples in which serial Cy/vy co-occurs with the conjunction -vy (60), indicating that they have different functions. The first is an empty morpheme, a residue of an old gerundive marker, while the latter is a productive conjunction which also marks co-referentiality between subjects: 1SG-wash 1SG-be.sit -SER CONJ 1SG-fall 'When I was washing the bench, I fell'

On Expressing Purpose Semantics with vy in Mbya Guarani
The grammaticalization of movement verbs into purpose constructions is a frequent process from a cross-linguistic perspective (Heine and Kuteva 2002, p. 163). In Mbya Guarani, when a movement verb stands in V1 position, followed by a regular lexical verb in V2, it triggers a purpose reading, as can be observed in (61) In constructions like (62a) and (62b), none of the verbs are from a restricted class nor do they exhibit any constraints regarding transitivity; each transitive verb can select its own object, as can be observed in (62a), where V1 'to plant' selects the argument avachi ('corn'), while V2 chooses cho'o ('meat') as its internal argument. Furthermore, we can observe that both V2s in (62a) and (62b) can be negated independently from V1, making it clear that they do not conform to a single syntactic unit. They belong to two different clauses. The same argument regarding negatability applies to motion-cum-purpose vy constructions, as can be observed in the following data in (63) Even if the motion verb receives the negation circumfix (64), the status of the event encoded by the purpose clause is not affected by its scope, proving the events involved constitute two separate predications. In the following examples, negation is placed on the movement verb, but it does not imply that the purpose clause is negated as well: Given the evidence presented above, we propose to analyze vy purpose constructions as subordinate clauses as they do not exhibit the same syntactic and morphological constraints serial verbs do, such as lack of subordinate connectors and identical values for negation and transitivity.

Conclusions
According to our data, Mbya Guarani SVCs fit Aikhenvald's definition of asymmetrical serialization in that they involve more than one verb, where V1 belongs to an open class, while V2 belong to a closed restricted set, they are mono-clausal, and they share just one TAM and negative marker. In addition to that, we also show that the transitivity restriction found between V1 and V2 reinforces Baker (1989)'s and Baker andStewart's ( 1999, 2002) hypothesis that object sharing is a defining property of SVCs. On this last matter, the possibility of transitivizing Mbya Guarani V2 in this complex should also prevent us from labeling it as an auxiliary, as those forms are not susceptible to valency-changing operations. Parameters notions account for these differences; as already noticed, vP in Mbya Guarani can contain two heads which allows the property of object sharing (Vieira 2017) as proposed by Baker (1989) and Baker andStewart (1999, 2002). Moreover, the transitivity restriction is not observed in any type of subordinate clauses in Mbya.
Regarding the status of the -Cy suffix as a dependency marker, we propose that this form resembles more a residual dummy marker, given that it has lost its productivity in Mbya. This last observation raises the question of dependency and the alternation between Set 1 and Set 2 indexation prefixes in intransitive and transitive V2s in Mbya SVCs discussed in this paper. The explanation for this current scenario can only come from a diachronic perspective. Both Mbya's serial suffix and its homophonous samesubject (SS) subordination conjunction have probably developed from the reconstructed PTG serial suffix *-abo (Jensen 1998b;Cabral and Rodrigues 2006). Regarding the vy purpose constructions, our analysis intended to show that this constitutes a case of clausal subordination, with different syntactic restrictions from that of the vy SVCs in Mbya. By no means do we propose that this analysis holds true for analogous constructions found in all languages in the Tupi-Guarani group, but rather we aim to make a contribution in order to show how one construction in the proto-language can evolve in different constructions exhibiting distinct syntactic and semantic features in different daughter languages, or even in the same language, as in the case of Mbya Guarani.
In order to close this paper, we would like to show, following Vieira (2002Vieira ( , 2017, that Tocantins Asurini, another Tupi-Guarani language, also exhibits serial verbs of the asymmetrical type and presents the object sharing property, as (26) and (27) repeated below illustrate. In (65b), V2 gets transitivized by the applicative morpheme and this way, it can " share" its object with transitive V1:  Informed Consent Statement: Informed consent to make this data available was obtained from all subjects involved in the study, who have participated in the data collection sessions whish took place in their villages or in their educational institutions. The data presented in this paper were collected in different times and places; 1995 and 1999 (Ubatuba, São Paulo); 2008 and 2009 (Faxinal do Céu, Paraná); 2017 (Misiones, Argentina), and 2019 (Misiones and Rio de Janeiro). Individually, the subjects were submitted to experiments which comprised of translation exercises and grammaticality judgement tests.

Data Availability Statement:
Enquires regarding the data presented in this study should be addressed directly to the authors as is not publicly available yet.