Mbyá resultatives and the structure of causation

In Mbyá, target state resultative predicates can only be derived from inchoative verbs. This is in contrast with target state adjectival passives in better studied languages such as English, German and Greek, which can be derived from transitive causatives. I argue that the limited distribution of Mbyá resultatives reveals a point of variation in the association between roots and external-argument introducing Voice heads: while roots of non-agentive transitive causatives need not be lexically associated to agent or causer Voice in English and similar languages, they do in Mbyá. Together with the well documented incompatibility of target state resultatives with agent/causer Voice, this difference explains the restricted distribution of Mbyá resultatives. The proposed analysis stresses the importance of the distinction between target states and resultant states in the structure of adjectival passives and related resultative predicates cross-linguistically.


Mbyá resultatives and valency alternations
This paper explores a point of cross-linguistic variation in the formation of resultative predicates, through a detailed study of resultatives in Mbyá Guaraní, a Tupí-Guaraní language spoken in Argentina, Brazil and Paraguay. I use the term resultatives to refer to stative predicates that entail the existence of a causing event, 1 such as adjectival passives in English (Wasow 1977;Levin & Rappaport Hovav 1986). Using the label adjectival passive would be misleading in Mbyá, since there is no participial morphology in the language and resultatives are formally distinct from passives.
My starting point is the observation that Mbyá resultatives have a more restricted distribution than English adjectival passives. In English, adjectival passives can be derived not only from inchoative verbs like break but also from transitive causatives that do not participate in labile alternations, like bury: (1) English a. The door broke. b. Chris broke the door. c. Chris will repair the broken door.
1 This use of 'resultative' is close to Nedjalkov and Jaxontov's (1988) use of the term. However, while I label resultant state predicate as resultatives, Nedjalkov and Jaxontov classify them as perfects, as observed by Kratzer (2000). Note that I do not discuss complex resultative predicates like kick the door open in this paper.
Glossa general linguistics a journal of Thomas, Guillaume. 2019. Mbyá resultatives and the structure of causation. Glossa: a journal of general linguistics 4(1): 136. 1-38. DOI: https://doi.org/10.5334/gjgl.688 (2) English a. *The body buried. b. Chris buried the body. c. (…) cadaver dogs found the buried body. 2 German and Greek are similar to English in this respect. In Mbyá on the other hand, resultatives can only be derived from inchoative verbs. As I will discuss in more detail in section 3, resultative predicates are derived by adding the suffix -kue or its voiced allomorph -gue to an inchoative verb stem. 3 Resultative derivation is ungrammatical not only with derived and underived transitive causatives, as illustrated in (4b) and (5b), but also with causative stems whose valency has been reduced with the reflexive/passive prefix je-/nhe-, as illustrated in (4c) and (5c), to be compared with the grammatical reflexives/passives in (4d)/(5d): c. *Jagua o-jou jaixa o-je-aty-kue. dog A3-find paca A3-PASS-bury-RES Intended: 'The dog found the buried paca.'

Theoretical assumptions
My analysis is grounded in constructivist theories of argument structure, which spring from the seminal work of Hale and Keyser (2002), and are represented by studies such as Ramchand (2008), Harley (2012) and Marantz (2013). In this tradition, the event structure and argument structure of verbs is built syntactically, through the combinations of roots with functional heads that enrich their event structure and introduce their arguments.
On the semantic side, I adopt a Neo-Davidsonian analysis of event descriptions, with full syntactic decomposition. In particular, I will assume that verb roots introduce predicates of events or states, and arguments are introduced by separate thematic heads. For a defence of thematic separation in syntax, see Borer (2003;. For a review of semantic arguments in favour of thematic decomposition, see Williams (2015: Chapter 9).
Finally, I will adopt a realizational analysis of morphology, according to which syntax acts as a generative engine that builds abstract representations that are then spelled out phonologically. In section 5, I will argue that restrictions on the distribution of Mbyá resultatives are elegantly captured in a spanning theory, which holds that sequences of non-terminal nodes in a linearized syntactic structure can be realized as single morphemes, see in particular Haugen and Siddiqi (2016).

Ethics and data sources
Mbyá examples, judgments of grammaticality and truth-value were elicited by the author with two native speakers of Mbyá from Paraguay and one speaker living in the state of Rio de Janeiro in Brazil in July 2017 and July 2018. The author sought the informed consent of the consultants prior to elicitation. The three consultants authorized the author to publish the contents of these elicitation sessions in scientific communications. This study also relies on Robert Dooley's (2015;2016)

Caveat lector
Mbyá is spoken on a large territory that includes the Misiones province in Argentina, parts of Paraguay and several states in Brazil. Therefore, it goes without saying that regional variation is expected, and I can only claim that the description of resultatives presented in this paper reflects the usage of the three consultants I worked with. However, it is worth noting that this description is consistent with Dooley's (2015;2016) own description of Mbyá resultatives, which is based mainly on data collected in the Rio das Cobras community, in the state of Paraná in Brazil.

Structure of the paper
Sections 2 and 3 present the core Mbyá data to be analyzed. Section 2 focuses on valency alternations, and section 3 on resultative predicates. I argue in section 3 that Mbyá resultatives denote properties of target states rather than of resultant states. In section 4, I compare the distributions of target state resultatives in Mbyá, English, German and Greek, and I argue that the restricted distribution observed in Mbyá is due to the more stringent constraints that Mbyá roots impose on their syntactic frame, in combination with the cross-linguistically attested incompatibility of target stativizers with agent or causer Voice heads. Section 5 presents an implementation of the analysis in a realizational theory of Thomas: Mbyá resultatives Art. 136,page 5 of 38 morphology with insertion at non-terminal nodes, and compares it to an implementation in Distributed Morphology. Section 6 concludes.

Split intransitivity in Mbyá
In order to discuss valency alternations in Mbyá, it is important to give an overview of agreement in the language. Mbyá verbs are uninflected for tense and aspect (Dooley 2015: §12.3). Bare verbs are interpreted with a non-future reference time, and are underspecified for viewpoint aspect. 4 By contrast, subjects and/or objects are cross-referenced on the verb, following a split-S system. Two paradigms of cross-reference markers are used, which I refer to as class A and class B markers (see Table 1). Some intransitive verbs cross-reference their subject with class A markers, others with class B markers. Intransitive verbs that use class A markers are known as active, while those that use class B markers are known as inactive. Transitive verbs crossreference either their subject or their object, with the exception of combinations of first person subjects and second person objects, which are cross-referenced with the portmanteau prefix ro-. 5 With all other combinations of persons, a person hierarchy is used to determine which argument is cross-referenced (where "n > m" stands for "n wins over m"): (6) Mbyá person hierarchy: 1 > 2 > 3 Transitive subjects are cross-referenced with class A markers, while objects are crossreferenced with class B markers. With two third person arguments (e.g. third person subject and third person object), the subject is cross-referenced. 6 In other configurations, the highest argument on the hierarchy is cross-referenced. The following constructed examples illustrate: (7) a. A-exa. A1.SG-see 'I saw her/him/them/it.' 4 Unless required by the context of an example, I will translate bare achievements and accomplishments with past perfective English verbs, and bare activities and states with present (progressive) English verbs or adjectives. 5 Note that, contrary to Paraguayan Guarani, Mbyá does not have different portmanteau prefixes for singular and plural second person objects. 6 A subset of transitive verbs cross-reference both their subject and object, when the latter is third person.
Following Dooley (2015), I analyze the i-and j-prefixes in word forms such as o-j-aty ('A3-B3-bury') and o-i-kuaa ('A3-B3-know') as third person cross-reference markers. The nature of the cross-referencing system of Guaraní languages has been the subject of some theoretical debates, see in particular Velázquez-Castillo (2002a) and Zubizarreta & Pancheva (2017a) on Paraguayan Guaraní. One fact that bears mentioning in this section is that the distinction between intransitive verbs that select class A markers and those that select class B markers is orthogonal to the distinction between intransitive verbs that support resultative derivations and those that don't.
The following examples show that resultative predicates can be derived from either class of verbs: Consequently, the opposition between class A and class B markers will not play a direct role in my analysis of Mbyá resultatives.

Stative/inchoative alternations
In Mbyá, stative verbs may alternate between their stative interpretation and a dynamic, inchoative interpretation without morphological marking. This is illustrated in (9)

Causative alternations and causative verbs
Alternations between intransitive and causative uses of a verb are marked by the causative prefix mbo-and its allomorph mo-: Finally, it should be mentioned that mo-/mbo-is not the only causative marker attested in Mbyá. Transitive verbs are causativized by the suffix -uka, and sociative causatives (cf. Shibatani & Pardeshi 2002;Guillaume & Rose 2010) are built with the prefix (gu)ero-, see Dooley (2015: §13.2). Since causatives of transitives and sociative causatives are not directly relevant to the argument that will be developed in the rest of the paper, I will not discuss them.
door A3-CAUS-break wood with 'The door was broken with a stick.' Not only do (20c) and (21c) imply the existence of an external argument, it is also understood that this argument is an agent, rather than the door or the road itself: clearly, the door cannot be the participant manipulating the stick in (20c), and if it were the understood external argument in (19b), negating the existence of an agent with avave ('nobody') should be coherent. In sum, it appears that the reduced causatives of Mbyá cannot be analyzed as reflexive verbs with a semantically impoverished effector external argument, as proposed by Koontz-Garboden (2009) for marked anti-causatives. In these respects, non-reflexive causatives whose valency was reduced with je-/nhe-are more similar to English passives, which also imply the existence of an external argument, than to marked anti-causatives. French SE-anticausatives, for instance, are consistent with negating the existence of an external argument, and license neither instrument modifiers nor agentive prepositional phrases: French a. La porte s'est cassée, mais personne ne l'a cassée. the door SE=is broken but nobody NEG it=has broken 'The door broke, but nobody broke it.' b. La porte s'est cassée (*avec un bâton).
the door SE=is broken with a stick c. La porte s'est cassée (*par Jean). the door SE=is broken by Jean I will not offer a detailed analysis of the structure and interpretation of valency reduction with je-/nhe-, nor will I explain the preference for its reflexive interpretation, since these questions are orthogonal to our main argument. What matters to the analysis of resultative formation is that transitive verbs prefixed with je-/nhe-introduce an external argument, which is either existentially quantified or bound through reflexivization. In both cases, these verb forms contrast with inchoatives, which lack an external argument altogether.

There are no anticausative or labile alternations in Mbyá
We have seen that causativization in Mbyá is marked by the prefix mo-/mbo-, the suffix -uka or the sociative causative prefix (gue)ro-. The prefix je-/nhe-is used as a reflexive or passive marker. Two other valency increasing or decreasing markers that we will not discuss here are the reciprocal prefix jo-/nho-, and the impersonal voice marker -a, which binds the external argument of a transitive or intransitive verb without promoting its internal argument (Dooley 2015: §13.2). The demotion of a direct object to an oblique argument is not marked by derivational morphology, but is marked syntactically by the use of postpositions: Given this rich inventory of valency changing markers, we may ask ourselves whether there are any unmarked valency alternations in the language. With respect to external argument alternations, the answer is clearly negative. As Dooley observes in his grammatical sketch of Mbyá, alternations that affect the external argument are marked morphologically (Dooley 2015: §13.2).
The situation is more complex with alternations involving internal arguments, which are optional with some verbs:  I conclude that, with the possible exception of optional internal arguments, every operation that increases or decreases the valency of a verb is marked either by derivational morphology, or by the use of adpositions. In particular, all operations that introduce or bind an external argument are marked by derivational morphology.
An important consequence of this state of affairs is that Mbyá lacks any form of anticausative or labile alternation. We saw that valency reduction with je-/nhe-should be analyzed as a form of passivization or reflexivization. So-called labile alternations, covert alternations between inchoative and transitive causative uses of a verb, are also unattested, as is middle voice. In these respects, Mbyá differs from English, in which labile alternations are attested, as well as German and Greek, in which both labile alternations and marked anticausatives are attested (Alexiadou et al. 2006). In section 4, I will argue that this property of Mbyá is closely linked to the restricted distribution of resultatives in the language.

Morphological makeup
Resultative predicates are formed by attaching the suffix -kue/-gue to the verb root. Both active and inactive verbs are attested in resultatives. The following examples illustrate the predicative use of resultatives: Mbyá resultatives also have attributive uses: cable tear-RES 'I am going to use this torn cable.' We now come to the observation at the centre of this paper, namely that the resultative suffix -kue/-gue cannot attach to causative verbs, whether they are derived causatives or underived transitive causatives. This is illustrated in examples (28) and (29): (28) a. A-j-aty jaixa. A3-B3-bury paca 'I buried the paca. b. *Jagua o-jou jaixa (j)-aty-kue. dog A3-find paca 3-bury-RES Intended: 'The dog found the buried paca.' (29) *A-mo-ĩ porã ta okẽ o-mo-pẽ-gue. A1.SG-CAUS-be good PROSP door A3-CAUS-break-RES Intended: 'I am going to fix the broken door.' (cf. (27a)) Resultative formation is also unacceptable with causative verbs whose valency has been reduced with the prefix je-/nhe-: Note that resultative formation is not acceptable with all intransitive verbs. Resultatives formed from activity verbs are unacceptable, as illustrated in (31): Furthermore, resultatives are not attested with all inchoative verbs: (32) a. A-mo-ĩ axuka ka'ay py, ha'e he'ẽ reve'i. A1.SG-CAUS-be sugar mate in and sweet quickly 'I put sugar in the mate, and it got sweet quickly.' b. ?A-ka'ay-'u-xe ka'ay he'ẽ-gue. A1.SG-mate-drink-DES mate sweet-RES Intended: 'I want to drink sweetened mate.' The following resultatives were identified in Dooley's (2016) lexicon. Note that all base verbs in the list are intransitive in Mbyá, with the apparent exception of monda, to which we come back below: (33) a. aguino > aguino-gue rot > rotten b. jy > jy-kue cook > cooked c. kanhy > kanhy-gue disappear > disappeared d. ke > ke-kue turn (of milk) > turned e. monda > monda-kue steal > stolen f. u'ũ > u'ũ-gue rot > rotten g. xo > xo-gue break off > broken off h. xoro > xoro-kue tear > torn i. xyryry > xyryry-kue fry > fried j. ypi > ypi-kue dry (of a tree) > dried k. vo > vo-kue crack > cracked My consultants judged that these resultatives are well formed. They also accepted the following ones. Again, all base verbs in the list are intransitive: All verbs in these lists but monda have inchoative uses. Indeed, in his lexicon, Dooley (2016: 92) observes that the formation of resultatives with -kue/-gue is restricted to intransitive verbs, with the exception of the form mondakue ('stolen'), which my consultants also accepted: (35) A-jo-gua ao monda-kue. A1.SG-TR-buy clothes steal-RES 'I bought stolen clothes.' It should be noted that monda is not a direct transitive verb, but rather an intransitive verb with an oblique theme argument: Still, monda differs from the other predicates listed in (33) and (34) insofar as its subject is not a theme but an agent, and constitutes an apparent exception to the generalization that Mbyá resultatives may only be derived from inchoative verbs. In section 5, I will argue that the compatibility of monda with resultative derivation is due to the irregular character of this verb, and that this exception actually brings further support to the proposed analysis of the restricted distribution of Mbyá resultatives. Before I close this section, I would like to point out that the Guaraní suffix -kue/-gue is also attested as nominal temporal marker (Tonhauser 2006;Thomas 2014): However, uses of -kue/-gue as a nominal tense and as a resultative marker have different distributions and interpretations. As a past temporal marker, -kue/-gue cannot be suffixed to verbs that have not been nominalized. If the verb is nominalized, -kue/-gue can attach to it regardless of its event structure. In (38a), the bare verb o-o is compatible with a past tense interpretation. Attaching the temporal marker -kue to the verb in (38b) is ungram-matical. In (38c), the temporal marker attaches to the nominalizer va'e, which rescues the sentence: By contrast, the resultative use of -kue/-gue is restricted to inchoative verbs, and does not require nominalization. Since the present paper is only concerned with the resultative use of -kue/-gue, and its use as a temporal marker has a different distribution, the question of the unification of resultative and past uses will be left to future research.

Semantic properties of Mbyá resultatives
Embick (2004)  Since adjectival passives describe states that result from a previous causing event, asserting that an entity was in a result state at its time of creation is incoherent. Applying the same diagnostics in Mbyá confirms that inchoative verbs suffixed with -kue/-gue denote properties of result states, rather than properties of pure states: The door can only be broken after it's been made." One may further ask what kind of result state is described by Mbyá resultatives. In a seminal study, Kratzer (2000) observed that German and English adjectival passives do not form a homogeneous class. Kratzer identified two kinds of adjectival passives in these languages, depending on the nature of the state and its relation to the causing event. Resultant state adjectival passives denote mere states of events having culminated, which Kratzer models as properties of times that follow the culmination of an event. Target state adjectival passives, on the other hand, describe semantically richer states that result from the culmination of an event, but may not persist indefinitely. In English and German, target state adjectival passives that do not describe permanent states are compatible with modification by still (in German: immer noch), unlike resultant state adjectival passives, as illustrated by (42a)  b. Das Theorem ist (*immer noch) bewiesen. the theorem is (*always still) proven '*The theorem is still proven.' The distinction between resultant states and target states was initially formulated by Parsons (1990), who offers an elegant description of the difference between these two kinds of states: (43) Parsons (1990: 234-235) "For every event e that culminates, there is a corresponding state that holds forever after. This is "the state of e's having culminated," which I call the "Resultant state of e" (…) It is important not to identify the Resultant-state of an event with its "target" state. If I throw a ball onto the roof, the target state of this event is the ball's being on the roof, a state that may or may not last for a long time. What I am calling the Resultant state is different; it is the state of my having thrown the ball onto the roof, and it is a state that cannot cease holding at some later time." Following this intuition, Kratzer (2000) proposes that target state adjectival passives are derived by applying a stativizer to a relation between events and states, while resultant state adjectival passives are derived by using a stativizer that is essentially a perfect operator, which maps a property of events to a property of times: Kratzer's denotations for the two stativizers are given in (45). I have renamed the target and resultant stativizers RES TARGET and RES RESULT , respectively: Coming back to Mbyá, all the resultatives that my consultants accepted and that do not describe permanent states are compatible with modification by teri ('still'): We must of course make allowance for resultative predicates that are incompatible with teri because they denote permanent states: 11 Finally, the incompatibility of -kue/-gue resultatives with activity predicates (cf. example (31)) is also explained if -kue/-gue denotes a target stativizer, which takes relations between events and states as arguments. Since activity predicates denote properties of events without an associated target state, they will not be able to combine with -kue/-gue. I conclude that -kue/-gue resultatives denote target states rather than resultant states.

Structure of Mbyá resultatives
My analysis of the structure of Mbyá resultatives builds on the constructivist analysis of causativization of Schäfer (2007) and Alexiadou et al. (2006;. Verbs are decomposed into an uncategorized root together with a number of functional projections: Following Marantz (1997), roots are category neutral. A little v head categorizes them as verbal. Following earlier proposals by Kratzer (1996), Hale & Keiser (2002) and Pylkkännen (2002;, external arguments are introduced by a Voice head. In addition, Schäfer (2007) and Alexiadou et al. (2015) analyze causative/inchoative alternations as Voice alternations. Inchoative verbs (i.e. unmarked anticausatives) consist of a Voice-less vP. In causative verbs, an agent or causer argument is introduced by an additional Voice head: This analysis follows a tradition of decomposing causative verbs into a state denoting root, and an operator that introduces a causing event (see among many others Lakoff 1965;McCawley 1968;Dowty 1979;von Stechow 1996;Harley 2008;2012). In addition, Pylkkänen (2002; argued based on the analysis of Finnish desiderative causatives and Japanese adversative causatives that causing events may be introduced independently from causer arguments. She concludes that causative relations between eventualities and agents of causing events are introduced by different heads, which in the present framework are identified with little v and Voice, respectively. Further extending this line of reasoning, Kratzer (2005) observes that Pylkkänen's dissociation of causing events from causer arguments makes it possible to analyze causative alternations as Voice alternations. Kratzer argues that causation should be analyzed as a relation between a causing event and an event or state that the culmination of this event brings about. In this perspective, one may assume that inchoative predicates are causative too: (50) Kratzer (2005) "the commonly posited BECOME operator becomes superfluous in the decomposition of inchoatives, causatives, and anticausatives. Those three types of verbs are all plain causatives. They differ with respect to voice." Finally, Schäfer (2007) and Alexiadou et al. (2015) provide syntactic arguments for the presence of a causative little v head in inchoatives, in the absence of external argument introducing head.
Building on this analysis of causative structures, Alexiadou et al. (2015) argue that resultative operators differ with respect to their height relative to the Voice head. In resultant state adjectival passives, the resultative head (which I will refer to as RES) is introduced above Voice, while in target state adjectival passives, it is introduced below the Voice head. In resultant state adjectival passives, Voice may introduce an agent or a causer argument. On the other hand, when Voice is present in target state adjectival passives, it introduces as an external argument a participant 12 in the target state, which accounts for the incompatibility of target state adjectival passives with modifiers that relate to agents or causers, with the exception of agents and causers that participate in the result state (see McIntyre 2013 and section 4.1). The class of target state adjectival passives is not itself syntactically homogeneous, since target state participles in English and German as well as target state interpretations of Greek participles formed with the suffix -menos are argued to embed a full vP, while Greek target state participles formed with the suffix -tos only embed a root or √P:  (41), where it was shown that they are unacceptable in complements of verbs of creation.
In sum, resultative predicates formed with -kue/-gue denote a target state, which entails the existence of a causing event. The resultative head is introduced above little v, and is incompatible with an agent/causer Voice head. This suggests the following structure for Mbyá resultatives: In this structure, the stative root √DRY is adjoined to the v CAUS head, which introduces a causing event. The thematic head θ THEME introduces the holder of the state, in the complement domain of little v. The little vP denotes a relation between states of the theme being dry and their causing event. The resultative head RES TARGET maps these relations to a property of target states of such events.
In the syntactic representation in (54), it is assumed that the root denotes a state. This is appropriate for verbs like piru, which can be used as pure stative predicates. However, some inchoative verbs only have a dynamic use. This is the case for instance with pẽ ('break'). How can we capture the dynamic nature of such verbs? Following Embick (2009), I propose that the roots of predicates of caused states denote a property of events and must co-occur with a "proxy" predicate of states. In inchoative structures, the root √BREAK denotes a property of events that is adjoined to a causative little v head, and specifies a manner of causing. The caused state is denoted by an abstract predicate ST that is c-commanded by little v: 13 Given these assumptions, we may represent the structure of resultatives derived from predicates of caused states as follows: The compositional interpretation of these structures is discussed in the appendix.

Target stativizers and Voice
Following previous studies by Rapp (1996), Schlüker (2005, Gehrke (2011), Meltzer-Asscher (2011) and McIntyre (2013), Alexiadou et al. (2015) note that English, Greek and German target state adjectival passives fail to license agent/causer related modifiers, unless the agent they refer to is present in the target state, or the adverb modifies that state directly. The following contrasts illustrate this fact: The target state adjectival passives in (57a) and (58a) are incompatible with agentive byphrases. In (57b) and (58b), agentive modification is acceptable. McIntyre argues that the relevant contrast lies in the relation of the agent to the target state. While the agents of (57a) and (58a) only participate in the causing event, those of (57b) and (58b) are interpreted as participants in the target state. As McIntyre puts it, they are "responsible for continuing the state expressed by the participle" or are somehow essential to the characterization of the state. It appears then that the arguments introduced in by-phrases in (57b) and (58b) are arguments of the target state, and that these examples are consistent with the generalization that target state adjectival passives are incompatible with agent oriented modifiers that relate to the causing event rather than its result state. This generalization is further supported by the incompatibility of target state adjectival passives with instrument modifiers and agent oriented adverbs. The following examples from Greek illustrate: Greek (Alexiadou et al. 2015: 158) To thisavrofilakio itan (*akoma) prosektika anigmeno. the safe was still cautiously opened 'The safe was (*still) cautiously opened.' Based on these observations and additional modification tests, Alexiadou et al. (2015) conclude that target stativizers are introduced below Voice, and block the introduction of agent/causer Voice.
In order to explain the incompatibility of target stativizers with agent/causer Voice, Anagnostopoulou (2017; proposes that target state but not resultant state participles represent outcomes of scalar changes. She further argues that verbs that lexicalize scalar changes must express change along a single dimension. The presence of Voice is then argued to be blocked in target state adjectival passives because it would lead to interactions between the agent and the theme that would render the change too complex.
In the next subsection, I will argue that the incompatibility of target stativizers with agent/causer Voice explains the restricted distribution of resultatives in Mbyá.

Explaining the restricted distribution of Mbyá resultatives
Mbyá target state resultatives are unacceptable with all transitive verbs. This contrasts with target state adjectival passives in English and German, which can be derived from a subset of non-alternating transitive causatives, as illustrated by the following examples: The fact that these adjectival passives are compatible with modification by still suggests that they are interpreted as target state resultatives. Importantly, the transitive verb stems from which they are derived lack inchoative alternants: In German, the native speakers I consulted accept sich-forms of these verbs only to the extent that internal causation is conceivable. To illustrate, one speaker commented that (64b) is acceptable "if I imagine a building that has a function of ejecting all residents in the case of a threat". Another speaker observed that (64a) suggests that "the tires have a built-in inflation mechanism (much like life jackets in an airplane)." One speaker judged that, as a consequence of these implications, (64b) and (64d) are odd. Another one judged that (64c) and (64d)  This discussion shows that target state resultatives are compatible with some transitive verbs that do not participate in causative alternations in English and German. Why then are resultatives incompatible with transitive causatives in Mbyá? We saw in section 3 that Mbyá lacks any form of anticausative or labile alternations. Furthermore, the only operations that reduce the valency of transitive verbs preserve external argument entailments: both reflexive and passive verbs entail the existence of an external argument, and there appears to be no middle voice in the language. In the present analytical framework, this suggests that the roots of lexically causative verbs in Mbyá are only ever attested in syntactic frames that include an agent or causer Voice head, which introduces an external argument in the semantic representation of the verb. I conclude that these roots are lexically required to occur in such syntactic frames. This entails that these verbs will be incompatible with resultative derivation with the suffix -kue/-gue, since this suffix is a target stativizer and as such is incompatible with agent or causer Voice. The same reasoning also explains the ungrammaticality of resultatives derived from morphological causatives, under the assumption that the causative prefix mo-/mbo-introduces a thematic Voice head.
By contrast, given the compatibility of the English and German transitive causatives with target state derivation in (61), we must conclude that the roots of these verbs may occur in syntactic frames that lack agent or causer Voice, despite the fact that these verbs do not have inchoative alternants.
Note that a similar analysis of restrictions on the distribution of target state resultatives was already proposed in Greek by Anagnostopoulou (2017), who observed that verbs of manner of killing and verbs of creation, as well as a subset of verbs with manner entailments, cannot form target state adjectival passives: Greek (Anagnostopoulou 2017: 115) Aftos o anthropos ine (*akomi) dolofonimenos/straggalismenos this the man is (*still) murdered/strangled dhilitiriasmenos/pirovolimenos/mexeromenos. poisoned/shot/knived (66) Greek ( Anagnostopoulou observes that the causative verbs that cannot form target state adjectival passives require an agent as an external argument. By contrast, at least some verbs of external causation that are not strictly agentive appear to form acceptable target state adjectival passives in Greek, as illustrated by the following examples (the judgments in (69) were provided by a native speaker of Greek): Greek (Alexiadou et al. 2015: 157)  In the present framework, verbs that require an agent as external argument must be realized in a frame that includes an agentive Voice head, which explains their incompatibility with target stativizers. By contrast, the roots of non-agentive causative verbs that are compatible with target state adjectival passives must be able to occur in syntactic frames that lack an agent or causer Voice head, even if they fail to participate in labile alternations.
In sum, the crucial difference between Mbyá and languages like English, German or Greek with respect to the distribution of target state resultatives appears to be range of roots that are required to co-occur with an agent or causer Voice head. In Mbyá, all roots of transitive causatives are subject to this requirement. In English, German and Greek, at least some roots of non-alternating transitive causative verbs must be able to occur in frames that do not include agent or causer Voice.

Coda
I have argued that all roots of transitive lexical causatives in Mbyá require the presence of an agent or causer Voice head in their syntactic frame. This, in combination with the incompatibility of target state resultatives with such heads, explains the restricted distribution of resultatives in this language. What is still unclear at this point is the nature of the association between Voice and the roots of causative verbs in Mbyá. In the next section, I sketch two alternative analyses of this dependency. The first one builds on the assumption that phonological exponents may spell out linearized sequences of terminal nodes, and proposes that exponents of roots of transitive lexical causatives in Mbyá spell out sequences of heads that include the root itself and its associated Voice head. The second adopts a more classical approach to Vocabulary Insertion, and posits that these roots bear a feature that must be checked by a Voice head.

Spelling out syntactic frames with post-linearization spanning
An elegant way of capturing the requirement that roots of transitive causatives impose on the presence of a Voice head is to assume that the exponent of these verbs spell out a chunk of structure that includes both heads. Theories of Vocabulary Insertion at nonterminal nodes are well suited to implement this idea. In this subsection, I will sketch an analysis of the morphological realization of Mbyá verbs using the theory of post-linearization spanning of Haugen & Siddiqi (2016).
An important motivation of realizational models of morphology with non-terminal insertion is the avoidance of the proliferation of null exponents and post-syntactic readjustment operations like fusion and fission. Null exponents and readjustment operations are invoked in realizational models of morphology when a piece of syntactic structure that contains multiple terminal nodes is spelled out by a single exponent. This arises for instance in the analysis of irregular past forms of English verb tenses like hit or sang (Halle & Marantz 1993). The former may be analyzed by positing a null allomorph of the past tense, whose use is triggered by the presence of the root √HIT: An analysis of sang will involve not only positing a null exponent for the past tense in the context of √SING, but also invoking a phonological readjustment rule that will change the nucleus of the exponent of the root in the context of the past tense. c. Phonological realization: sang The realization of sequences of heads by a single Vocabulary Item is subject to the two following principles: (73) Post-linearization Contiguous Morpheme Insertion Principle (Haugen & Siddiqi 2016: 369) Insertion may realize multiple adjacent X 0 s (features) provided that the features realized by the inserted Vocabulary Item are as large a subset of the string of adjacent X 0 s (features) as that which could otherwise be expressed by separate overt Vocabulary Items at the contained X 0 s (features).
(74) Minimize Exponence (Siddiqi 2009: 4) The most economical derivation will be the one that maximally realizes all the formal features of the derivation with the fewest morphemes.
Note that in post-linearization spanning, a Vocabulary Item may provide an exponent for a series of heads, without matching all the heads in the series. This is similar to the insertion of underspecified Vocabulary Items under bundles of features in Distributed Morphology. In view of this fact, the discussion that follows will distinguish exponence (the phonological realization of a series of heads) from feature matching.
In the graphical representation of Vocabulary Insertion that are used in this section, e.g. (72b)-(72c), horizontal curly brackets represent exponence, rather than feature matching. Keeping this in mind, let us now illustrate how the two principles in (73) and (74) interact in the realization of the past form of sing. This process involves a competition between the irregular form sang and the regular form *singed, which exploits the following set of Vocabulary Items: Two options compete for the phonological realization (i.e. exponence) of the linearized sequence of heads. For ease of reading, feature matching between Vocabulary Items and sequences of heads are indicated through boldface, while exponence is indicated with curly brackets:  (75), the little v head does not qualify as realized in either option, in the matching sense of realization. Nevertheless, this head is still spelled out by sang in (76c) and by sing in (77c). It is worthwhile to discuss the process that leads to the realization of the past form of hit, since this will illustrate an important feature of post-linearization spanning, namely the use of homophony in the analysis of morphological irregularities. Both the present nonthird person form of this verb and its past tense form are realized as hit. The Vocabulary Item in (78b) may be used to spell out the former. However, spelling out the past form of this verb with this Vocabulary Item is less optimal than using the regular form *hitted, since (78b) does not match the past tense feature. Consequently, we must introduce a homophone of hit in (78c), in order to account for the irregular realization of the verb in the past tense. While the appeal to root homophony in post-linearization spanning may be criticized by some, it is embraced by Haugen & Siddiqui (2016), who argue that this approach is no more stipulative than positing a contextually conditioned zero-allomorph of the past tense, and reflects some degree of arbitrariness in the history of these irregular forms.
Let us now consider how post-linearization spanning may be used to provide a model of the morphological realization of Mbyá resultatives and valency alternations. In a realizational morphological framework, the most delicate aspect of Mbyá verb morphology to model is the lack of labile alternations. Indeed, we must make sure that (i) exponents of inchoative verbs cannot be used to spell out transitive structures, and (ii) exponents of transitive causatives cannot be used to spell out inchoative structures. In Distributed Morphology, both problems would arise if null exponents were used indiscriminately to spell out little v heads and Voice. Remembering that the verb piru ('dry') is restricted to intransitive inchoative or stative uses, while aty ('bury') is restricted to transitive uses, consider the following inventory of Vocabulary Items: Unrestricted use of null exponents for v CAUS and Voice predicts, incorrectly, that the exponent piru may realize a causative verb, and that the exponent aty may realize an inchoative verb: These cases of over-generation are elegantly avoided using post-linearization spanning. One may restrict the use of aty to causative structures by assuming that this exponent spells out a sequence of heads that includes both the root and Voice: Because the Vocabulary Item in (85a) is specified for Voice, it cannot spell out a sequence that does not include this head. The impossibility to spell out transitive structures with piru is also easily accounted for with Minimize Exponence, given the availability of the alternative realization mombiru. To illustrate, the exponent piru defined in (86a) spells out all three features present in (87), but it matches only the root, while the combination of mo-and piru (mombiru) in (88) matches all three features:  (87), since the causative morpheme mo-matches the Voice and v CAUS heads, which are left unmatched in (87). Note that this analysis relies on an auxiliary assumption, which is that there is no Vocabulary Item that spells out Voice with a null exponent. This however is a consequence of a design feature of post-linearization spanning, which eschews the use of null exponents altogether. The analysis also captures covert event structure alternations between stative and inchoative uses of piru and other stative roots, under the assumption that little v heads lack independent exponents in the language (which is supported by the lack of overt marking of alternations between stative and dynamic interpretations of verbs, as well as by the lack of overt marking of the opposition between inchoative and process verbs, all these oppositions being encoded in flavours of little v in the proposed analysis): In sum, this analysis captures dependencies between the roots of lexical causatives and Voice lexically, by encoding in Vocabulary Items the information that the exponents of these roots also spell out Voice. In this perspective, the incompatibility of roots of underived transitive causatives with resultative derivation receives the same explanation as the impossibility of these roots to occur in inchoative verb frames. Namely, since both structures lack agent or causer Voice, the only Vocabulary Item that may spell out the root is overspecified. Compare (92) to the grammatical resultative in (91)

Remarks on thieving
It was observed in section 3.1 that the verb monda ('steal') is compatible with resultative -kue/-gue, which appears to contradict the generalization that resultative -kue/-gue only combines with inchoative predicates. I believe that the source of this exception lies in the irregular character of the predicate monda itself. This is a dynamic predicate which requires the presence of an agent argument and a theme that undergoes a transfer of possession. As such, monda falls under Tsunoda's (1981) category of predicates of effective action. The fact that the theme of this verb is realized as an oblique argument is somewhat surprising, since effective action predicates tend to be realized as transitive verbs cross-linguistically (Tsunoda 1985;Malchukov 2005;Grimm 2011). Even more surprising is the fact that the agent of monda is cross-referenced with a class B inactive marker. Velázquez-Castillo (2012a) observed that arguments that initiate a process of change (in this case a transfer of possession) are marked as actors in Paraguayan Guaraní. Mbyá is similar to Paraguayan Guaraní in this respect (see Dooley 2015: §10.1). We would therefore expect the agent of monda to be crossreferenced with a class A marker. This is indeed how the agent of dynamic uses of monda is cross-referenced in Paraguayan Guaraní (see Jover Peralta & Osuna 1950;Guasch & Ortiz 2008). Given this state of affairs, I would like to suggest that monda has an argument structure that is not characteristic of agentive verbs, and that this property is reflected both in its unexpected cross-referencing pattern and in its compatibility with resultative derivation. More precisely, I will argue that because of its peculiar argument structure, monda differs from other agentive verbs in Mbyá in not requiring the presence of a Voice head in its syntactic frame. My argument will rely on Zubizarreta & Pancheva's (2017a;b) theory of cross-referencing in Paraguayan Guaraní, which I apply to Mbyá. Zubizarreta & Pancheva (2017a;b) analyze the cross-referencing system of Paraguayan Guaraní transitive verbs as a direct/inverse system, class A markers being used to crossreference subjects in the direct order, while class B markers are used to cross-reference objects in the indirect order. In addition, Zubizarreta and Pancheva argue that while class A markers are agreement prefixes, class B 1 st and 2 nd person markers are pronouns. In the simplest case of indirect order, they are object pronouns that are promoted to a position of adjunction to the inflection head I, via the edge of vP, as illustrated by the following example from In (94b), the 1 st person cross-reference marker is argued to be base generated as a possessor in the specifier of the incorporated nP che=rova, before undergoing cyclic movement to I. Zubizarreta and Pancheva extend this analysis to class B 1 st and 2 nd person cross-reference markers of so-called "triform-verbs", which they argue originate as possessors of an nP incorporated into little v. Consider for instance the verb che=rasẽ in (95). This verb consists of the nP asẽ ('cry') incorporated into a light v. Its class B 1 st person cross-reference marker originates as a possessor of the incorporated nP, and is cyclically moved to a position of adjunction to the I head: (95) Zubizarreta & Pancheva (2017b: 97) (Che) che=rasẽ. I 1POSS=cry 'I cry.' I propose that the Mbyá verb monda ('steal') has a structure similar to that of intransitive tri-form verbs, where the incorporated phrase is a possessive DP, rather than an nP: In this example, the stem xemonda is derived by incorporating an event denoting DP interpreted as 'my theft' into a light v. Its 1 st person cross-reference marker originates as a possessor (nP [1SG] ) in the specifier of DP, and is cyclically moved until it ends up as a D head adjoined to I. The theme is introduced as a post-positional argument of the verb. The introduction of the agent as a possessor explains its unexpected cross-referencing with a class B marker. Moving on to the interpretation of (96b), we can assume that the possessor nP [1SG] is base-generated in a thematic position. More precisely, the root √THEFT denotes a property of events of stealing, and the possessive head D POS introduces introduces nP [1SG] as the agent of these events (cf. John's theft of the $1,000).
Crucially, while we argued that agent-introducing Voice is always spelled out together with the root of transitive causative verbs, this generalization need not apply to the relation between nominal roots and D POS . If √THEFT can be spelled out independently of D POS , it should be possible for the target stativizer -kue to combine with the stem monda, as illustrated in (97), repeated from (35), and in (98) In sum, analyzing monda as an incorporated DP explains its irregular inflection class and argument structure. In addition, the hypothesis that its agent is not introduced by Voice but rather by a possessive D POS head explains why monda is not subject to the requirement that all roots of lexical transitive causatives in Mbyá should be spelled out together with Voice.

Cross-linguistic Variation
The analysis sketched in the previous subsection accounts for the lack of target resultatives derived from transitive causatives in Mbyá. However, we haven't yet explained why such alternations are attested in English and similar languages. For alternating verbs like dry, the answer is simple. The Vocabulary Item that is used to spell out the root of such verbs is underspecified for Voice. Since causative alternations are unmarked in English, the same Vocabulary Item can be used to spell out the root in an inchoative or a transitive frame, the former being compatible with resultative formation: The situation is more complicated with transitive causatives that do not alternate, like bury. Because these verbs can form target state adjectival passives, the Vocabulary Items that realize their roots must not be specified for agent or causer Voice: However, this cannot be the whole story, since the Vocabulary Item in (101) could then be used to spell out the root √BURY in an inchoative frame, which we know is unattested.
In order to solve this puzzle, I propose that the constraints that prevent the realization of these roots in inchoative frames are not lexical, but pragmatic. I will now sketch an analysis along these lines, by building on an earlier proposal by Rappaport Hovav (2014).
Following Levin & Rappaport Hovav (1995), Rappaport Hovav (2014)  Rappaport Hovav argues that the availability of the inchoative form with clear type verbs is not due to lexical restrictions, but to pragmatic constraints which favour the more informative expression of the cause argument: (104) Rappaport Hovav (2014: 23) "In the description of a change of state, the cause of the change of state is relevant; therefore, since an utterance which specifies the cause of the change of state is more informative than one which expresses just the change of state, it is to be preferred, all things being equal." If the cause of a change of state is always relevant, speakers should use the more informative transitive verb form, unless they have additional reasons not to do so. One such reason would be that the speaker assumes there is no external cause. In Gricean terms, using the transitive form would then violate the maxim of quality (Grice 1975). In this scenario, the inchoative form is preferred: (105) Rappaport Hovav (2014: 25) a. My watch broke after the warranty ran out. b. #I broke my watch after the warranty ran out.
(does not have the same interpretation as (a)).
Note that Rappaport Hovav (2014) does not intend her pragmatic analysis to apply to verbs like kill and destroy. Since these verbs never alternate, she argues that they are lexically associated with an external argument. However, given the availability of target state adjectival passives with non-agentive transitive causatives like bury, I believe that a lexical analysis of the lack of alternation of bury-type verbs is not tenable. Instead, I suggest that Rappaport Hovav's pragmatic analysis of causative alternations with clear-type verbs can be extended to non-alternating non-agentive transitive causatives. While the Vocabulary Items that spell out the roots of these verbs are by themselves compatible with inchoative frames, transitive frames are preferred due to their greater informativity. The challenge that we face is to explain why this preference rules out the inchoative use of bury type verbs, while it can be overridden by contextual factors for break and clear type verbs. I suggest that the unacceptability of inchoative alternants with bury type verbs can be attributed to an obligatory implicature. Let us assume that the inchoative form of lexical causatives competes with its passive alternative: (106) English a. *The body buried. b. The body was buried.
(106b) is more informative than (106a), since it entails the existence of an external cause that lead to the body being buried. Everything else being equal, an utterance of (106a) will therefore lead to an implicature that the speaker is not in a position to assert (106b). This implicature, however, conflicts with the common knowledge that events of burying have external causes. Indeed, bury is similar to kill and destroy insofar as these verbs denote non-spontaneous events, i.e. events that are unlikely to occur without the intervention of a external cause (Haspelmath 1993). In this respect, bury differs from verbs like break and clear, which are compatible with internal as well as external causes (Harley & Noyer 2000), and will therefore trigger non-contradictory implicatures. Note that this analysis requires the computation of the implicature of (106a) to be mandatory and blind to common knowledge. Indeed, if the implicature could be cancelled, it would not lead to a contradiction and unacceptability. In addition, the computation of the implicature must be blind to common knowledge. Otherwise, given the piece of information that events of burying are always associated with an external cause, (106a) and (106b) would be truth-conditionally equivalent, and an utterance of (106a) would not trigger the implicature that the speaker is not in a position to assert (106b). A theory of blind and mandatory implicatures was developed by Magri (2009), in order to account for lifetime effects of individual-level predicates. For reasons of space, I will not discuss how the present analysis could be implemented in Magri's framework. There are certainly loose ends to tie up, but I believe that the broad lines of this analysis are clear enough for our purposes.
Importantly, this analysis can account for the felicity of target state resultatives derived from non-alternating lexical causatives, given adequate assumptions about the set of syntactic alternatives to adjectival passives. Consider for instance example (107):

(107) English
The body is still buried.
The implicature that is responsible for the unacceptability of inchoative uses of buried depends on a competition between the verbal passive and inchoative forms of the verb.
Under the assumption that target state adjectival passives do not compete with verbal passives, they will not trigger the faulty implicature. If adjectival and verbal passives differ in syntactic category in English, this assumption could be reduced to the more general requirement that two constituents cannot be syntactic alternatives of one another unless they have the same category, a requirement that is already part of Katzir's (2007) theory of structural alternatives.

Conspiracies
The analysis that was presented in this section relies on post-linearization spanning. However, it is also possible to model Mbyá verb morphology even if Vocabulary Insertion is limited to terminal nodes. To do so, one may wish to posit null exponents for Voice and little v heads. In order to account for the lack of covert alternations targeting external arguments, the former should be restricted to contexts that include the root of a lexically transitive verb: (108) a. Voice ↔ ∅ /{√BURY, √BLOW, …} b. Voice ↔ moc. v CAUSE ↔ ∅ Likewise, in order to block the insertion of roots of transitive verbs in inchoative verb frames, the exponents of these roots should be specified for insertion in the context of a Voice head: ] Voice ] c. … In such a system, the absence of external argument alternations with roots of causative verbs arises as a conspiracy, as in the case of English irregular past tense inflection. 19 In the latter case, the blocking of a regular past tense verb form in the context of √SING conspires with the blocking of the regular exponent of √SING in the context of T [+PAST] . In the former, the realization of agent/causer Voice by the causative prefix mo-is blocked in the context of a transitive root, while the phonological realization of the root is restricted to the context of agent/causer Voice.
While such an analysis is tenable, I believe that an analysis based on insertion at nonterminal nodes is more elegant. Indeed, in the conspiracy analysis, the restriction of the root of a causative verb to a transitive frame is encoded in the Vocabulary Item that specifies the exponent of this root. The fact that Voice is not realized independently from the root is arbitrarily encoded in a second Vocabulary Item. While this arbitrariness may be justified in the analysis of irregular processes, the morphology of Mbyá verbs appears to be a paragon of regularity, which is elegantly captured by the spanning analysis.
Note that I am not claiming that analyses of irregularities in post-linearization spanning are generally more principled than analyses that rely on null exponents and readjustment rules. As Haugen & Siddiqi (2016) recognize, the use of homophony in the former is no less stipulative than that of lexically conditioned zero allomorphs. In the specific case that concerns us here, however, a spanning analysis appears to more elegant.

Conclusion
I have argued that what appeared to be an arbitrary selectional restriction governing the distribution of the Mbyá resultative suffix -kue/-gue is in fact an epiphenomenon, which reveals the interplay between lexical constraints on the frames in which roots may be inserted, and the syntactic properties of target stativizers.
The proposed analysis brings to light a point of variation in the lexical encoding of the dependencies between roots and functional heads: the distribution of resultatives in Mbyá suggests that all roots of underived transitive causatives are lexically associated with an agent/causer Voice head in this language, in contrast to English, German and Greek, where this lexical association only holds of roots of strictly agentive verbs.
Finally, this study of Mbyá resultatives offers new empirical support for the analysis of adjectival passives of Alexiadou et al. (2015), and stresses the importance of the distinction between target states and resultant states in the structure of adjectival passives and related resultative predicates.