Hybrid intransitives in Basque

This paper deals with a group of agentive verbs in Eastern dialects of Basque that show mixed unergative and unaccusative properties. Although they pattern with unergatives in certain aspects, they combine with an absolutive subject and the auxiliary 'be', contrary to what one would expect for Basque unergative verbs. Additionally, they behave like unaccusatives in a number of other tests, such as in their inability to take cognate objects and in allowing partitive subjects. The analysis put forward in this paper accounts for the hybrid nature of these verbs. In particular, we claim that their subject is introduced in the specifier of vP, and that it is co-indexed with a thematic but non-pronounced argument of Voice. As a consequence, the subject shows both external and internal properties. This paper thus challenges the mutually exclusive external/internal division of the subject in intransitive verbs and argues that intransitive verbs can be classified into more groups than just two, as also argued elsewhere. Additionally, it argues that the different types of intransitive verbs are grammatically encoded and shaped by different versions of Voice and v. Thus, this analysis assumes the typology of Voice proposed by Schafer (2008) and developed by Alexiadou, Anagnostopoulou and Schafer (2015), and extends it to the functional head v.


Introduction
Basque is a morphologically ergative language in which intransitive verbs show two different morphological alignments: broadly speaking, agentive verbs usually occur with an ergative subject and the auxiliary edun 'have', whereas patientive verbs combine with an absolutive subject and the auxiliary izan 'be' (Levin 1983). Nevertheless, there exists some dialectal variation regarding the use of certain agentive verbs, specifically those that convey (non-static) processes, where the subject is interpreted as both the force initiating the process and the entity undergoing it. We will henceforth refer to this group of verbs as agentive process verbs. In Eastern dialects, 1 some agentive process verbs combine an absolutive subject with the auxiliary izan 'be' (Oyharçabal 1992;Aldai 2006;Berro 2010;, thus resembling other ergative -non-active -languages like Warlpiri, Niuean and Samoan, where all intransitive verbs combine with an absolutive subject (Levin 1983;Massam 2009;Tollan 2018). Even if we do not believe that verb meaning by itself (in rough terms, agentive vs. patientive) is a totally reliable predictor of the syntactic behavior of verbs, it is important to pay attention to and account for those cases where semantic generalizations are clearly contravened, as occurs with the verbs under study. With the exception of Levin (1983), previous studies have assumed that the agentive process verbs that combine with an absolutive subject in Eastern dialects are unergative and have accounted independently for the absolutive marking. In fact, these verbs belong to semantic groups that are usually considered unergative (controlled and agentive processes), and include bazkaldu 'have lunch ', solastatu 'talk', borrokatu 'fight' or jolastu 'play'. Moreover, they behave syntactically like unergatives in their inability to undergo causativization (Oyharçabal 2003).
However, this study shows that agentive process verbs in Eastern dialects of Basque do not always pattern syntactically with unergative verbs, thus contravening the abovementioned generalizations or cross-linguistic tendencies: they are not compatible with cognate objects, they can combine with partitive subjects in interrogatives and negative clauses, and some of them can be embedded in adjectival participles. In order to account for this range of syntactic properties, we depart from the standard assumption that the subject of unergative verbs is introduced as a mere external argument (i.e., in the specifier position of Voice, Kratzer 1994;, proposing instead that their subject has both internal and external properties given that it is introduced in the specifier of Voice (1a) or, alternatively, in the specifier of v (1b), but is co-indexed with the non-pronounced specifier of the other functional head (Voice or v), thus fulfilling two different semantic roles, as shown in (1) In proposing this analysis, we will also argue against an account whereby Eastern dialects have an ergative case system based strictly on valence (non-active) that would contrast with the semantically aligned case system of Western and Central dialects (Aldai 2006;. In fact, not all intransitive verbs in Eastern dialects combine with an absolutive subject, and therefore, the case system of Eastern varieties cannot be fully equated to that of other ergative languages like Warlpiri, Niuean and Samoan, where most intransitives have absolutive subjects (Levin 1983;Massam 2009;Tollan 2018). We argue that the morphosyntactic distribution of these agentive process verbs is shaped by the combination of different versions of Voice and v, which happens to be different across Basque dialects. Thus, we argue that agentive process verbs have different argument structures across the Basque dialects and cannot be categorically classified as either unaccusative or unergative. Our proposal is that, apart from unaccusative and unergative verbs, there are what we call hybrid verbs, as depicted in (1). Therefore, this study makes a contribution to the view defended in several works (Kural 2002;Alexiadou & Anagnostopoulou 2004;Deal 2009;Alexiadou & Schäfer 2011;Irwin 2012, Alexiadou 2014 according to which monoargumental verbs can be classified into more structural classes than just two (cf. Perlmutter 1978;Burzio 1986). We assume the typology of Voice proposed by Schäfer (2008) and developed by Alexiadou, Anagnostopoulou and Schäfer (2015) whereby Voice can come in different flavors, 2 and extend it to the functional head v. With these different versions of Voice and v, we account for two different "intermediate" (hybrid) verb argument structures in the dialects. For Eastern varieties, the analysis put forward in this paper is that the subject of these verbs is introduced in the specifier of vP, but that it is co-indexed with a non-pronounced thematic argument of Voice (Kratzer 1996) (1b). These verbs thus have a hybrid structure with what we call a low subject. In this way, the subject is interpreted both as the argument of Voice and as the argument of v. In Western and Central dialects, by contrast, the subject of these verbs is introduced in the specifier of Voice, as an external subject, but it is also interpreted as an argument of v. That is, these verbs have a hybrid structure with a high subject (1a). Thus, this proposal challenges the mutually exclusive external/internal division of the subject in intransitive verbs and accounts for the dialectal differences observed across Basque varieties through an analysis that involves different versions of Voice and v, which yield different argument structures and morphosyntactic variation. We claim that these configurations are ruled by the LF-instructions associated with each Root (in the sense used by Harley 2014 andWood 2016), which restrict the interpretation of the Roots in the structure, and which differ across the dialects under study.
One main point argued for in this paper is that, although semantics usually gives rise to generalizations predicting the syntactic behavior of verbs, not all verbs with similar meanings will behave as unaccusative or unergative across languages, and even within languages, as noted early on by Rosen (1984). In order to establish whether the sole argument of a given intransitive verb is syntactically or semantically vP-internal or vP-external, we will use a series of diagnostics including, among other things, case marking and the availability of resultative adjectival passives. As we will see, rather than unaccusativity or unergativity diagnostics, what we have at our disposal are specific tests that are sensitive to different structural properties and allow us to distinguish not only unergative or unaccusative verbs, but also what we call hybrid verbs. Thus, by positing different structures for intransitive verbs, we are able to account for the mixed behavior of these verbs in the different syntactic tests. 3 The structure of the paper is as follows. Firstly, we briefly present the case system and argument structure of Basque in Section 2. Then, we describe the dialectal phenomenon under analysis in Section 3, noting that agentive process verbs take the auxiliary izan 'be' in Eastern dialects. In Section 4, we present our proposal for a syntactic analysis that reflects the hybrid nature of agentive process verbs and accounts for their mixed unaccusative/unergative behavior. Next, in Section 5, by means of the above-mentioned diagnostics, we offer several pieces of evidence that agentive process verbs display a hybrid behavior in Basque that can be accounted for in the terms of the analysis proposed. Additionally, in Section 6, we argue that izan 'be' auxiliary selection and absolutive case assignment of the subject in agentive process verbs cannot be explained by appealing to a different case system (that would be present in these dialects) (Aldai 2006;, or with an analysis where the complement NP is incorporated in the l-syntax (Fernández 1997). Finally, in Section 7 we conclude.

Ergativity and argument structure in Basque
Basque is a morphologically ergative language, which means that subjects of intransitive verbs (S) and direct objects of transitive verbs (O) constitute a class for case-marking and agreement, in contrast with subjects of transitive verbs (A), as schematically illustrated in (2) (Dixon 1994, Ortiz de Urbina 1989. (2) a. Transitive: Ergative case is assigned to the subject of transitive predicates (3), whereas absolutive case (zero case) is assigned to the direct object of transitive predicates, as in (3), and also to the subject of intransitive predicates (4). 4 (3) Jon-ek sagar bat jan du. John-erg apple a.abs eat have.(3sgabs.)3sgerg 'John has eaten an apple.' Mahai-a apurtu da. However, not all intransitive verbs behave alike. Especially in Central and Western dialects (Fernández 1997;Aldai 2006;Berro 2010;Berro & Etxepare 2017), a syntactic distinction between unaccusative (4) and unergative (5) verbs is systematically at play, the latter occurring with an ergative subject and the auxiliary edun 'have', just like transitive verbs (3) (Levin 1983;Etxepare 2003 In general terms, this syntactic behavior is found with verbs that belong to semantic classes that are usually considered unergative (Perlmutter 1978). 6 This group includes intransitive predicates expressing volitional acts (like (5)), verbs describing non-animate 4 Case assignment in Basque is phrasal, so that case marking follows the determiner in DPs where the determiner is overt (2)-(3). In such cases, it is therefore not marked on the noun. 5 In this paper, we are only considering simplex verbs, that is, verbs that consist of a single phonological word. Note, however, that many unergative verbs in Basque are morphologically complex, involving the light verb egin 'do' and a non-verbal element, usually a noun (see Acedo-Matellán & Pineda 2019 for a recent overview). The dialectal variation that concerns us here is only attested in simplex verbs. 6 Perlmutter (1978) also defined unergative and unaccusative verbs syntactically, but here we are only referring to the semantic classification.
activities, verbs of emission, verbs describing involuntary bodily processes, some verbs describing the existence of a state and some aspectual verbs. On the other hand, verbs of other semantic classes that are classified as unaccusative (Perlmutter 1978) show a distinct syntactic behavior, as they combine with an absolutive subject and occur with the auxiliary izan 'be' (Levin 1983;Etxepare 2003). This group comprises predicates expressing change of state (whether telic or atelic) (like (4)) or location, predicates of appearance or occurrence, verbs of existence, and (some) aspectual predicates.
As ergative case is assigned to the subject of transitive as well as to the subject of unergative verbs -especially in Western and Central dialects -we consider ergative case in Basque to be inherent and assigned by Voice (Kratzer 1994; to the argument introduced in its specifier position. 7 In fact, assuming that transitive and unergative subjects are not introduced directly by the predicate (Marantz 1984) and that they are projected externally to the verb phrase (Kratzer 1994;, ergative can be considered the case of external arguments. Given that this pattern is also observed in other ergative languages like Georgian and Hindi, many authors have suggested that ergative case is an inherent case related to theta-marking (Nash 1996;Legate 2002;Aldridge 2004;Woolford 2006;Torrego 2012;see also Oyharçabal 1992;Holmer 1999;Laka 2006 for Basque). More specifically, Massam (2002), Woolford (2006), Legate (2002;, Aldridge (2004) and Coon (2010) have claimed that it is assigned by the head introducing the external argument (Voice or little v), which may also be responsible for case-marking the internal argument. In this paper, we will follow this line and assume that the subjects bearing ergative case are introduced in [spec, Voice].
Summing up, Basque is a morphologically ergative language where both transitive and unergative verbs generally tend to occur with an ergative subject and auxiliary 'have'. Nevertheless, as we will show in the next section, there is a group of agentive intransitive verbs in Eastern dialects that, even though they pattern with unergative verbs in certain aspects, combine with an absolutive subject and the auxiliary 'be'.

Dialectal variation: Agentive process verbs with unaccusative distribution
The dialectal alternation we are concerned with affects only simplex verbs that belong to semantic classes that are usually considered unergative, and is attested only in a subset of them. In fact, in Eastern dialects, certain simplex agentive verbs align with unaccusatives under a number of syntactic diagnostics, such as the fact that they occur with absolutive subjects and the auxiliary izan 'be'. We have referred to this group of verbs as agentive process verbs, as they convey (non-static) processes, where the subject is interpreted as both the force initiating the process and the entity undergoing it, as illustrated in (6).
(6) a. Antton bazkaldu da. Antton.abs have.lunch be.3sgabs 'Antton has had lunch.' b. Maialen borrokatu da. Maialen.abs fight be.3sgabs 'Maialen has fought.' 7 It must be noted that our proposal is also compatible with a structural approach to ergative case in Basque (e.g., Laka 1993;Fernández 1997;Rezac et al. 2014) as long as the DP in [spec, Voice] is always assigned ergative case and the DP within vP always gets absolutive. As a reviewer notes, this can be the case if T assigns ergative to the highest caseless DP in its c-command domain, and Voice assigns absolutive to the DP within vP (see Tollan 2013 for a similar approach). Within this analysis, absolutive would remain unassigned when there is no DP within vP, and ergative case would only be available for arguments sitting above Voice. In any case, given that ergative case is only assigned to the argument in [spec, Voice], in this paper we will assume that ergative is inherently assigned by Voice to the argument sitting in its specifier.
These agentive process verbs in Eastern dialects belong to the volitional verb classes listed and exemplified in (7)  With these agentive verbs that are morphosyntactically aligned as unaccusatives according to a number of diagnostics, the situation in Eastern dialects resembles the distribution of intransitive predicates in other ergative languages like Warlpiri, Niuean and Samoan, where most intransitive verbs -whether patient verbs or agent verbs -combine with an absolutive subject (Levin 1983;Massam 2009;Tollan 2018). On the basis of the pattern found in these kinds of ergative languages, Massam (2009), Tollan (2018 and Tollan and Oxford (2018) have proposed that external subjects can be introduced in two different positions -specifier of Voice and specifier of v -and that the subject of unergative verbs is introduced in the specifier of vP, that is, lower than transitive subjects, which are merged in the specifier of Voice. As unergative subjects are not introduced by Voice, they are not assigned ergative case by it, but rather bear absolutive case. (i) Horiek entzu-te-arekin jauzi zen. those.abs hear-nmlz-with jump be.3sgabs.pst 'He/She jumped when he/she heard that.' 9 With regard to speech verbs, it is worth noting that they do not have a simplex counterpart in Western and Central dialects, as illustrated in (i), but rather involve morphologically complex unergatives, with an ergative-marked subject and the auxiliary 'have', as expected.

(i) Western and Central dialects
Antton-ek hitz egin / berba egin du. Antton-erg word do talk do have.3sgerg 'Antton has spoken/talked. ' We argue that this analysis, although explanatory for ergative languages like Niuean and Samoan, cannot be extended to Eastern dialects of Basque. In fact, the variation presented in this section only affects a small group of predicates, particularly those denoting animate and agentive processes, where the subject both initiates and undergoes the event. Other verb classes like inanimate manner of motion verbs, verbs expressing duration and light emission verbs take an ergative subject and the auxiliary edun 'have', also in Eastern dialects ((12), see also Section 6). Thus, it cannot be claimed that all intransitive verbs have the same syntactic configuration in Eastern dialects.
Izarr-a-k distira-tzen du. star-det-erg shine-ipfv have.3sgerg 'The star shines/is shining.' We therefore cannot assume analyses, such as Aldai's (2006;, that argue that Eastern dialects have a specific case system. According to Aldai, the dialectal divide between Western and Central varieties on the one hand and Eastern varieties on the other can be explained if there are two different case-marking systems in Basque. In his account, Eastern varieties would resemble the above-mentioned ergative languages, with an ergative case system based strictly on valence and thus having all intransitive subjects marked with absolutive case, whereas in Western and Central the case system would be semantically aligned (i.e., based on a semantic property). As we just suggested -and as we shall see in Sections 4 and 6 -this proposal is not satisfactory with regards to the distribution of intransitives in Eastern dialects, since not all intransitive predicates combine with absolutive subjects and the auxiliary izan 'be'.

An analysis for Basque hybrid intransitives
Traditionally, unergative and unaccusative verbs have been defined on the basis of the nature of their single argument (Perlmutter 1978;1989;Perlmutter & Postal 1984;Burzio 1981;1986), with the subjects of unergative verbs having more subject-like properties and the subjects of unaccusative verbs having more object-like properties. Assuming that subjects are not introduced by the verb (Marantz 1984;Kratzer 1994;, in the generative tradition unaccusative verbs have been considered to involve an internal argument (vP or VP-internal) and no external argument, whereas unergatives have been analyzed as involving an external argument but no internal argument (e.g., Embick 2004;Irwin 2012). For instance, Irwin (2012) defines unaccusativity on the basis of the presence or absence of two properties: unaccusative sentences have no external argument but have a VP-internal argument requiring structural case, whereas by contrast unergatives have an external argument and no VP-internal argument requiring structural case. Assuming that the external argument is introduced by Voice (Kratzer 1994; and that verbs are combinations of a functional v head and a Root, the structure of unaccusative and unergative verbs can be illustrated in the following way 10,11 : Our proposal does not refute the existence of these structures, but rather argues that a given DP argument can be interpreted as an argument of both Voice and v. Therefore, this analysis makes a contribution to the line of research that challenges a strict two-way classification of intransitive verbs (e.g., Kural 2002;Alexiadou & Anagnostopoulou 2004;Deal 2009;Alexiadou & Schäfer 2011), by showing in particular that the discrepancies of these intransitive verbs in the syntactic tests can be explained by positing hybrid configurations. We argue that apart from the argument structures illustrated in (13ab), there are two more intermediate configurations where the subject gets interpreted in both [spec, Voice] and [spec, v]. As we will explain, this is the case of the agentive process verbs analyzed in this paper.
Since we are working within the framework of Distributed Morphology (Halle & Marantz 1993;Marantz 1997;Harley & Noyer 1999), we consider that verbs are combinations of Roots and a verbalizing functional projection, namely v. Additionally, we assume that Voice, the head introducing the external argument, is not part of the verb root (Kratzer 1996); v introduces the event and an associated undergoer argument, and Voice introduces an agent or initiator of the same event, combining with the vP by means of Event Identification (Kratzer 1996). Regarding Voice, we adopt Schäfer's (2008) original analysis (and later developed in Alexiadou, Anagnostopoulou & Schäfer 2015) of the typology of Voice. According to this view, Voice is the locus of both syntactic and semantic transitivity, but these two notions need not coincide. For instance, Voice can be thematic (i.e., semantically transitive), and assign a theta role to the DP in its specifier position (e.g., agent), or it can be expletive (i.e., semantically intransitive), in which case the specifier will remain without any semantic interpretation. On the other hand, Voice can be syntactically transitive if it has a D feature, and require a DP to be merged in its specifier. Or, alternatively, it can be intransitive if it has no D feature (Ø). As a logical possibility within this framework, we propose to extend this typology to v. In this way, we argue that the verbalizer v can also come in different versions. Particularly, it can be thematic or non-thematic and have a D feature or not. 12 The combination of the differ-ent versions of Voice and v renders possible a variety of intransitive verbs with different argument structures. Apart from the standard verbs with a mere external subject or a mere internal subject -like those depicted in (13) -we expect to find what we call hybrid verbs, that is, verbs that have their single argument introduced in the specifier of Voice (14a) or, alternatively, in the specifier of v (14b), but with their subject also interpreted in the other specifier position, thus fulfilling two different semantic roles (14ab). (14) a. Hybrid verb with high subject Agentive process verbs have one of the two argument structures seen in (14). Let us compare now the structures in (14) with those in (13) and the definition of unaccusative and unergative verbs mentioned at the beginning of this section. As we can see, the structures in (14) do not overstep the definitions but rather add some specifications that can account for the mixed behavior of these verbs. We have said that unergative verbs are those that have an external argument and no vP-internal argument requiring structural case (13a). The high subject structure (14a) has both these properties, with the peculiarity that its argument -merged in [spec, Voice] -is also interpreted as the argument of v, that is, as an internal argument. On the other hand, unaccusatives are verbs that have no external argument projected but have a vP-internal argument requiring case (13b). The low subject structure (14b) also has both these properties, since it does not have an external argument externally merged in [spec, Voice], but it has the specification that Voice is semantically active and that the same discourse referent of the internal argument is also assigned to the semantic argument of Voice. As in these structures the subjects are not interpreted as the specifiers of a single functional head, we can account for the mixed behavior of these verbs in the syntactic tests. Our proposal is that these are what we call hybrid verbs, intermediate argument structures whose configuration is shaped by Voice and v.
Recall that in Eastern varieties, agentive process verbs combine with an absolutive subject and the auxiliary izan 'be', thus patterning syntactically with unaccusatives. In order to account for this semantic and syntactic mismatch, we claim that, in this case, the verbs have a low subject structure, as illustrated in (14b). In this configuration, v is thematic -it introduces an undergoer argument -and requires a DP argument (v {λx, D} ), so that a DP is introduced in its specifier. Voice is thematic but syntactically intransitive (Voice {λx, Ø} ). It does not introduce a DP in its specifier, but has an agent argument, assigned to the discourse referent introduced within vP. Thus, the semantic interpretation of (14b) would be as follows: firstly, v introduces a free variable that gets bound by Functional Application when the DP is introduced in the specifier of v and, secondly, Voice introduces another free variable that is assigned the same discourse referent as that of the argument of vP. The co-reference value of Voice must be specified in Voice, assigning the variable y the same index as that assigned to the x introduced by v. Given that the low subject in (14b) is an argument of both v and Voice, it shows mixed semantic and syntactic behavior. Indeed, the absolutive subject of hybrid intransitive verbs in Eastern dialects has properties of both external and internal subjects, as we will show in Section 5. Regarding semantics, it is interpreted both as an agent and as an undergoer. As for the syntax, on the one hand, the argument of Voice is saturated and co-referent with the argument of v, so the verb cannot undergo direct causativization. On the other, as the low subject is overtly introduced within the vP, it gets absolutive case and the predicate shows unaccusative properties in other syntactic diagnostics: it is incompatible with overt direct objects, the subject can be marked partitive, and in certain cases it can also occur in adjectival participles (see Section 5).
As we showed above, the low subject structure (14b) is not the only way in which a DP argument can be interpreted as both an agent and an undergoer. In fact, the system also allows a high subject configuration where the single argument is introduced in [spec, Voice], but it also binds the free variable introduced by v (14a) (repeated in (15)). Therefore, it is interpreted as both the agent argument of Voice and the undergoer argument of v. The high subject configuration is the result of having the version of Voice {λx, D} and the version of v {λx, Ø} . We propose that this is the configuration of the agentive verbs that we are analyzing in Central and Western varieties.
(15) Configuration of verbs in Western varieties: high subject In this case, Voice is both thematic and syntactically transitive and v, in contrast, is thematic but introduces no argument in its specifier. Since the DP is introduced in the specifier of Voice, it gets inherent ergative case from it, and given that it is an external argument, the predicate behaves syntactically like an unergative verb in certain tests. On the one hand, as in Eastern varieties, in this configuration too the argument of Voice is saturated and is co-referent with the argument of v, so that the verb cannot undergo direct causativization. On the other hand, since the subject is assigned inherent case from Voice, an additional argument (a direct object) can be introduced within vP and be case-assigned by Voice.
Thus, we argue that there are two different argument structures for these verbs in Basque dialects and that they are basically the consequence of the different make-up of the two functional heads, Voice and v. As a consequence, the existing morphosyntactic variation in Basque is not related to the case system (cf. Aldai 2006;. All dialects have the same case system -where we argue the DP in [spec, Voice] gets inherent ergative case from Voice -but in agentive process verbs the Roots combine with different versions of Voice and v, and these combinations yield the morphosyntactic differences observed. The choice for the particular structure comes from different LF-instructions of the Roots in each dialect (in the sense of Harley 2014 andWood 2016). 13 We assume that Roots are introduced in syntax with no phonological or syntactic information (e.g., Halle & Marantz 1993;Harley & Noyer 1999;Embick & Noyer 2007) and that in principle they can combine with any syntactic structure (Borer 2005). Phonological and semantic information is provided after the syntactic derivation, and for that to be possible Roots have a series of PF and LF instructions (Harley 2014;Wood 2016). The interpretation of that Root will be subject to the surrounding syntactic structure. If a Root is in a particular configuration for which its LF-instructions provide no interpretation, the syntactic object will not be interpreted (Wood 2016). Therefore, it is necessary for the Root to occur in a structure that is listed in its LF-instructions, because otherwise there will not be any interpretation and the derivation will fail. As for the dialectal variation existing in Basque, we propose that the Roots have different LF-instructions in Western and Central dialects relative to Eastern dialects. A given Root like √bazkal in Eastern dialects has the interpretation of ' (15), with the overt subject introduced in [spec, Voice] but interpreted as both the agent and the undergoer of the event.

Diagnostics of the hybrid nature
In this section, we will provide evidence for the hybrid nature of agentive process verbs. In order to do so, we will examine a series of diagnostics, all of which are summarized in Table 1. As we will see, according to these diagnostics, as they are defined here in (13b) 13 Crucially, we do not believe that root meaning determines syntactic structure, at least not before syntactic derivation. Roots are freely introduced in syntax (without semantic information) but the LF-instructions restrict, post-syntactically, the contexts in which roots can be interpreted. agentive process verbs cannot be classified entirely as either unergative or unaccusative verbs. In fact, unaccusative verbs (13b), but not unergatives (13a), are generally able to causativize and can have their subject marked with the partitive. On the other hand, unergative verbs, but not unaccusatives, have an ergative marked subject, are able to take cognate objects and cannot form a resultative adjectival participle. The verbs that we are examining show a mixed behavior in these tests, and in certain cases the results differ depending on the dialect. Our proposal is that these tests do not categorically separate unergative and unaccusative verbs but rather target different structural properties. This is why hybrid verbs behave in certain cases like unergatives and in other cases like unaccusatives. Additionally, as agentive process verbs have different configurations in Basque dialects, they show different results depending on the dialect. Table 1 describes the requirements targeted by each test, and the behavior of unergative and unaccusative verbs as traditionally defined (13a vs. 13b). As can be seen, high subject verbs and low subject verbs (14a and 14b) do not always pattern like good unergative (13a) or unaccusative (13b) verbs.
First, we observe that high subject verbs -as they are used in Western and Central varieties -present several syntactic similarities with unergative verbs: they trigger ergative marking of the subject, which is merged in [spec, Voice] (recall Section 2); they are not able to causativize, as they cannot have two non-co-referent arguments in Voice and v (see Section 5.1); they are compatible with cognate objects, since they lack a syntactic internal argument (see Section 5.2); and their subject cannot take partitive marking in interrogative or negative environments (see Section 5.3). There is one last diagnostic, the compatibility with resultative adjectival participles, which generally identifies unaccusative verbs by assessing the presence of a semantic internal argument: it turns out that Western and Central varieties allow for such constructions with agentive process verbs, thus offering proof of the hybrid nature of such verbs that we are arguing for.
On the other hand, we see that low subject verbs -the same semantic class but as they are used in Eastern dialects -seem to align syntactically with unaccusative verbs under some diagnostics; for example, they present absolutive case marking of the subject, disallow cognate objects (Section 5.2), are compatible with partitive-marked subjects (Section 5.3) and allow for resultative adjectival participles (Section 5.4). These results suggest the presence of a syntactic, as well as semantic, internal argument. However, in these dialects the verbs under study are not able to causativize, thus patterning syntactically with unergatives (Section 5.1), and again reinforcing the proposal of a hybrid nature for such verbs in Basque.
In the following sub-sections, we will examine these diagnostics in more detail.

Ability to causativize
In all Basque dialects, the class of verbs denoting an agentive process shares with unergative verbs the inability to causativize. In Basque, direct causativization occurs freely in intransitive verbs whose argument is marked absolutive and which denote a change (Oyharçabal 2003;Ormazabal 2008;Berro, Oihartzabal & Fernández 2018): change of state verbs, change of position verbs and change of location verbs. In the causative variant, the undergoer argument keeps absolutive case, the causer is introduced with ergative case, and the auxiliary is edun 'have' (16).
(16) a. Katu-a hil da. cat-det.abs die be.3sgabs 'The cat died.' b. Haurr-a-k katu-a hil du. child-det-erg cat-det.abs die have.(3sgabs.)3sgerg 'The child killed the cat.' Regarding change of state verbs that are interpreted as internally caused (such as urtu 'melt' or gorritu 'blush/turn red') (Levin & Rappaport Hovav 1995), Basque seems to allow direct causativization (17), although it is more difficult when the undergoer of the change is animate (18)-(19) (Oyharçabal 2003: 240-241 On the other hand, some verbs describing inherently directed motion, such as ailegatu 'arrive', erori 'fall', etorri 'come' or irten 'go out', cannot undergo direct causativization. As pointed out by Levin and Rappaport Hovav (1995), these verbs do not causativize because they are internally caused, which in our analysis means that they have a saturated argument of Voice. We assume that the causative alternation is a Voice alternation (Alexiadou 2010;Alexiadou et al. 2006;Wood 2012;2016;among others), and that the causative variant emerges when Voice is semantically and syntactically transitive and the discourse referent of the argument of Voice is different from that of the argument of v -so that we have two different arguments, namely a causer and a theme. In internally caused verbs like verbs of inherently directed motion, the subject DP is introduced in [spec, v] but it is interpreted as the argument of both v and Voice. As the argument structure of VoiceP is saturated and Voice has no syntactic argument different from that of v, the causative variant cannot be built. We argue that this analysis can be extended to the agentive process verbs, to which we turn now.
As reported by Oyharçabal (2003), agentive process verbs, which according to him are semantically "clearly unergative", cannot undergo direct causativization, even though their single argument is marked absolutive in Eastern dialects (like in unaccusative verbs) (Oyharçabal 2003: 235). 14 Therefore, in terms of allowing direct causativization, agentive process verbs pattern with unergative verbs and unaccusative verbs of internal causation.
14 Some of these verbs can occur in transitive configurations, as in (i), but in these cases, the direct object denotes the target of the event, rather than the entity that undergoes the change (see Oyharçabal 2003 We claim this is due to the fact that in agentive process verbs (20)-(21), the DP is an argument of both Voice and v; Voice cannot introduce another independent argument and thus the conditions to build the causative variant are not met.
In the following sub-sections, we will show that agentive process verbs in Eastern dialects also share some distributional features with unaccusatives, evidence that supports the claim that they must not be considered unergative. In other words, they must not be regarded as verbs whose single argument is simply introduced as the specifier of Voice.

Cognate objects or measuring direct objects
Unergative verbs are usually compatible with cognate objects or direct objects (called rhemes in Ramchand 2008) that measure out the event denoted by the verb (22) (see Tenny 1987; also Berro 2015 for an analysis applied to Basque). Syntactically, the subject in unergatives is introduced outside vP (13), in the specifier of Voice (Kratzer 1994; (see Section 4). As it does not get case from the verb, it can co-occur with another case-marked argument within the vP. The subject of unaccusative verbs, in contrast, is introduced within the verb phrase (13), and as a consequence, unaccusatives are not generally compatible with any overt object (23) (see Levin & Rappaport Hovav 1995: 40, who discuss the cognate object construction in English). Transposing this generalization to Basque, it is worth noting that in Eastern dialects, agentive process verbs do not combine with cognate objects or measuring direct objects. This restriction is particularly telling in the case of meal verbs like bazkaldu 'have lunch' and animate activity verbs like borrokatu 'fight', since the restriction seems to have a dialectal distribution. In Central and Western varieties of Basque, meal verbs are compatible with hyponymous direct objects that denote the substance that is eaten (Berro 2010: 14), as illustrated in (24) b. Bazkari-a-n / Bazkal egi-te-ko / Bazkal-tze -ko zopa/fruitu-a lunch-det-ines lunch do-nmlz-dest lunch-nmlz-dest soup/fruit-det jan dut. eat have.(3sgabs.)1sgerg 'I have eaten soup/fruit in the lunch/for lunch.' The dialectal difference regarding the compatibility with direct objects correlates with the subject marking and auxiliary selection of the verbs. As explained in Section 2, meal verbs combine with an ergative subject and the auxiliary edun 'have' in Western and Central varieties, whereas they take an absolutive subject and the auxiliary izan 'be' in Eastern ones. We argue that this contrast is related to the fact that, in Eastern varieties, the subject is syntactically introduced within vP, as in unaccusatives, whereas in Western and Central dialects it is introduced in the specifier of Voice.
The dialectal difference regarding the compatibility with measuring direct objects does not necessarily hold for other agentive process verbs. For instance, animate activity verbs like jolastu/jostatu 'play' (28) and speech verbs like mintzatu 'talk' (29) do not accept cognate or measuring objects in either dialect. In order to express overtly the object in play verbs, speakers have to introduce a postpositonal phrase (30), or use a transitive verb like esan 'say' in the case of speech verbs (31).

Compatibility with partitive subjects
Another distributional feature that agentive process verbs in Eastern dialects and unaccusative verbs share is their ability to have partitive subjects in interrogatives and negative clauses. In a study of Basque ergativity, Levin (1983) showed that internal arguments differ from external ones in their compatibility with the partitive (called the ZERIK case) -(r) ik. In Basque, the partitive shows up in a number of contexts, and provides a non-definite interpretation to the argument it marks. According to Levin (1983) (see also Ortiz de Urbina 1989;Salaburu 1992), the arguments occurring with the partitive bear the same semantic role as the arguments occurring with absolutive case. Thus, in transitive predicates, the direct object may occur with partitive under certain semantic conditions, like in (32), whereas the subject cannot (32).
(32) a. Emakume-ek sagarr-a-k erosi dituzte. woman-pl.erg apple-det-pl.abs buy have.3plabs.3plerg '(The) women have bought (the) apples.' b. Emakume-ek ez dute sagarr-ik erosi. women-pl.erg no have.3plerg apple-part buy '(The) women have not bought apples.' c. *Emakume-rik ez ditu sagarr-a-k erosi. women-part no have.3sgerg.3plabs apple-det-pl.abs buy Intended: 'No woman has bought (the) apples.' Coming to the divide between unergative and unaccusative verbs, Levin (1983) shows that only subjects of patientive verbs -in her terms -can occur with the partitive case (33). Subjects of agentive verbs are not acceptable in this context (34) (Salaburu 1992 Thus, partitive marking can be considered a diagnostic that distinguishes unaccusative verbs (verbs that have their subject introduced in [spec, v]) from unergative verbs (verbs with the subject in [spec, Voice]). 17 Nevertheless, some verbs that could be semantically characterized as unergatives also accept partitive subjects in Eastern dialects of Basque. As pointed out by Levin (1983), the subject of mintzatu 'talk', an agentive process verb original from Eastern varieties, can bear the partitive case despite being agentive. Levin (1983: 318) gives an example of this use in a negative clause (35). In (36), we provide a literary example where the subject of mintzatu has the partitive case in a conditional. The same holds for all other agentive process verbs that have an absolutive subject in Eastern dialects, as confirmed by the Eastern speakers consulted and exemplified in (37) and (38).

(37)
Gaur ume-rik ez da bazkaldu. today child-part no be.3sgabs have.dinner 'Today, no child has had lunch.' (38) Gaur gizon-ik ez da promenatu. today man-part no be.3sgabs stroll 'Today, no man has strolled.' Thus, as can be seen, the subject of agentive process verbs in Eastern dialects behaves exactly like the subject of an unaccusative verb in terms of accepting partitive marking.

Adjectival participles
Finally, there is a further property that calls into question the merely external status of the subject of agentive process verbs. Some of these verbs can form adjectival participles with a resultative interpretation, both in Eastern dialects and in Central and Western varieties. For instance, adjectival resultative participles can build on meal verbs like bazkaldu 'have lunch', speech verbs like mintzatu 'talk' and animate activity verbs such as borrokatu 'fight'. 17 As an anonymous reviewer notes, the availability of partitive marking could be considered to be related to case -only in those absolutive arguments that allow it -and not related to the structural position of the argument. Nevertheless, not all absolutive arguments are compatible with the partitive, given that the transitive subjects of the ari progressive do not seem to accept it (even though they are marked absolutive). Therefore, partitive marking is not systematically related to absolutive case and could be considered to be dependent on the structural position of the argument.
(39) Bazkaldu-a/ Afaldu-a naiz. have.lunch-ptcp/ have.dinner-ptcp be.1sgabs 'I have eaten lunch/dinner (lit. I am eaten lunch/dinner).' (40) Jonekin mintzatu-a / solastatu-a naiz. John-soc talk-ptcp be.1sgabs 'I have talked to John (lit. I am spoken with John).' Borrokatu-a / Gudukatu-a naiz. fight-ptcp be.1sgabs 'I have fought (lit. I am fighted).' Adjectival resultative participles denote a consequent state reached after a prior event (among many others, Kratzer 1994;2000;Anagnostopoulou 2003;Embick 2004;Alexiadou & Anagnostopoulou 2008;Gehrke 2011) and are mainly compatible with telic verbs that involve an internal argument, in other words, transitive verbs and unaccusatives. Additionally, in Basque, the configuration illustrated in (39)-(41), with the adjectival participle ending in -a, can have, apart from the resultative meaning, an experiential one (Hualde et al. 1994;Krajewska 2012;2013a;b). The experiential interpretation arises when the participle is built on verbs that do not yield a good result state, 18 such as unergatives, and when it is built on transitive verbs and the theme of predication is co-indexed with the external argument of the verb (rather than the internal one) (Berro 2019). According to the speakers consulted, the examples above have a resultative flavor rather than an experiential one: the theme of predication is asserted to be in a consequent state triggered by the event underlying the participle. According to this judgment, the predicates above pattern more with transitive and unaccusative verbs, that is, verbs with an internal argument, rather than with unergatives, which only have an external argument, since they can give rise to adjectival participles with a resultative interpretation. We propose that this is because in agentive process verbs, v introduces an undergoer argument.

Interim conclusion
So far we have presented a variety of diagnostics. Table 2 below provides a summary of how the different verb classes under consideration behave with respect to each of the tests discussed. As can be seen, the verbs under study consistently show a mixed behavior in the two varieties: in Western and Central dialects, they align with unergative verbs in most respects, but they show an unaccusative behavior regarding the availability of resultative adjectival participles; on the other hand, in Eastern varieties, agentive process verbs generally show a syntactic unaccusative behavior, although they disallow direct causativization, just like unergatives.
To sum up, the agentive process verbs in Eastern dialects share distributional properties with both unergative and unaccusative predicates; they resemble unergatives in their ability to undergo direct causativization, but behave like unaccusatives in that they cannot combine with cognate or hyponymous objects, in that they allow partitive subjects in interrogative and negative clauses and in that some of them can occur embedded in adjectival resultative participles. As we will argue in Section 6, izan 'be' auxiliary selection and absolutive case assignment of the subject in agentive process verbs in Eastern dialects cannot be explained by resorting to a different case system (that would be present in these dialects) (Aldai 2006;, or with an analysis where the complement NP is incorporated in the l-syntax (Fernández 1997). Instead, we argue that this particular morphosyntactic make-up is evidence of something that goes beyond that: the hybrid status of these verbs, which boils down to different combinations of Voice and v.

Explaining the variation
In Section 3, we presented a number of dialectal differences across Basque varieties concerning certain agentive process verbs that show unergative morphology in Western and Central dialects but unaccusative morphology in Eastern dialects. We have shown that the verbs in question pattern with unergative verbs in their meaning and also in not allowing direct causativization. However, in Eastern dialects they show unaccusative-like distributional features in a number of other syntactic diagnostics, such as their incompatibility with overt direct objects and acceptance of partitive subjects, unlike what occurs in Western and Central varieties.
In Section 4 we argued that the dialectal differences under study rest on the typology of Voice and v, which can in fact share the same argument. Because, in this analysis, a single argument can be interpreted as both an agent and an undergoer -theta roles assigned by two different functional heads, namely Voice and v -this analysis is somewhat reminiscent of Ramchand (2008), where it is proposed that unergative subjects are both initiators and undergoers at the same time. In Ramchand's system, each role is assigned by a different subevent, which are all syntactically represented. The initiator and undergoer arguments are introduced by the initiation and process subevents respectively and, in the case of unergative verbs, both the initiation head and the process head are coindexed and share the same subject. As a consequence, the arguments introduced in their specifier positions have the same referential index. Although this system shares with our analysis the property that a single argument can get two different theta roles assigned by two different heads, this argument is not sufficient because it cannot explain the morphosyntactic variation existing in Basque dialects. Ramchand's system would predict a single argument structure for all dialects -with the subject introduced in both the specifier of initiation and the specifier of process -and would not be able to derive the case variation attested as well as other syntactic phenomena. For instance, it cannot explain why these verbs combine with an ergative subject in Western and Central varieties yet with an absolutive subject in Eastern ones. In contrast, as our analysis argues that the subjects are introduced in different positions -in [spec, Voice] in Western and Central dialects and in [spec, v] in Eastern ones -it can account for the different morphological realizations as well as the different syntactic distribution of these verbs in the dialects. This dialectal difference results from by the different LF-instructions (Harley 2014;Wood 2016) of the Roots in question, which vary across dialects.
We will now further argue that the variation cannot be explained on the basis of different case systems existing in Basque dialects (as proposed in Aldai 2006; or to different incorporation patterns of a hypothetical direct object of the verb (Fernández 1997).
As shown in Section 3, the pattern arising in Eastern dialects resembles that of intransitive verbs of some ergative languages like Warlpiri, where most intransitive verbs combine with an absolutive subject. On the basis on this similarity, Aldai (2006; claims that the case system in Eastern dialects of Basque is strictly valence-dependent. If a verb has a single argument, this argument will be assigned absolutive case. Nevertheless, this characterization of the Eastern case-alignment situation is called into question by the behavior of some verbs, such as inanimate manner of motion verbs (42), inanimate activity verbs (42b), verbs directly expressing duration (42) and emission verbs (42), which take a single argument marked ergative and the auxiliary edun 'have' also in Eastern varieties (see Berro 2012;Berro & Etxepare 2017 d. Izarr-a-k distira-tzen du. star-det-erg shine-ipfv have.3sgerg 'The star shines/is shining.' As can be seen, case marking in Eastern dialects is not exclusively valence-based, given that not all intransitive verbs are absolutive cased in these varieties. Thus, the distinction between Eastern and Western/Central dialects in terms of an ergative alignment versus a semantically based alignment is not as clear as suggested. We suggest that the variation is not related to the case system, but, as explained in Section 4, is instead grammatically encoded in the LF-instructions of the verbs. All dialects have the same case system, but in the verbs analyzed the Roots combine with different versions of Voice and v, and these combinations yield the morphosyntactic differences observed. The verbs in (42) can show unergative morphology -ergative subject and edun 'have' -in two scenarios. In one scenario, the subject of the verb can be introduced in the specifier of Voice and thus get interpreted as an initiator, but not as an argument of v. We argue this to be the case of (42abc). In the other scenario, the only subject can be interpreted both as an argument of Voice and as an argument of v, but it is syntactically introduced in the specifier of Voice. This would be the case of (42d). In both scenarios, the subject is syntactically external, rather than internal, and thus gets assigned ergative case.
This analysis is also superior to the NP-incorporation account provided in Fernández (1997) and mentioned in Oyharçabal (2003). In these works, following Hale and Keyser (1993), some of these agentive process verbs are considered to involve a complement NP in l(exical)-syntax. This complement is incorporated at this level, and the subject is therefore regarded as the only argument when the predicate enters the syntactic component (see also Bobaljik 1993). In this account, Basque would have a valence-based case system and, thus, the only argument of the intransitive predicate would be assigned absolutive case in the derivation. Regarding the dialectal variation, Fernández (1997) argues that the complement NP does not incorporate at the l-syntax in Western and Central dialects, and she explains in this fashion the ergative case marking of the subject in those varieties. Nevertheless, this analysis falls short in several aspects: under an approach like this that involves different components or levels for incorporation, one cannot in fact predict whether a given unergative verb will surface with unergative morphology (with the auxiliary 'have' and an ergative subject) or an unaccusative one (with the auxiliary 'be' and an absolutive subject); nor can one account for dialectal differences in a systematic way. In fact, we could not claim that noun-incorporation always takes place at the l-syntax in Eastern dialects, given that absolutive case assignment and 'be' auxiliary selection is not systematic in intransitive verbs in such varieties. Moreover, in more recent analyses of unergatives (e.g., Hale & Keyser 2005;Harley 2005), it is assumed that morphologically simple (synthetic) unergative verbs consist of a verbal head and a Root, not a verb and an NP. This is also the position that we have adopted in this paper.
By claiming that agentive process verbs are hybrid, with their subject syntactically introduced in [spec, vP] but semantically interpreted as the argument of both v and Voice, we can account for their mixed syntactic distribution, which, as we have shown, in some cases patterns with unergatives and in others patterns with unaccusatives. On the other hand, our analysis also explains why not all intransitive verbs have the same morphosyntax: in some of them, like in light emission verbs (42d), the only subject is syntactically introduced in the specifier of Voice. As a result, the subject of these verbs is assigned ergative case in all varieties. In contrast, in the intransitive verbs that denote an externally caused change, the subject is syntactically and semantically only introduced in the specifier of vP, and as a consequence it is assigned absolutive case in all varieties. As for hybrid verbs, that is, verbs that are internally caused and where the argument undergoes a change, the subject can be syntactically high (introduced in [spec, Voice]) or low (introduced in [spec, v]) and semantically interpreted as the argument of both Voice and v. This analysis is schematically spelled out in Table 3.
Depending on the position where the subject is syntactically introduced, agentive process verbs will have an ergative or an absolutive subject, but syntactically they will show mixed distributional properties.

Conclusions
In this paper, we have proposed a novel analysis to explain the dialectal differences found in some intransitive verbs in Basque. More specifically, we have shown that agentive process verbs like bazkaldu 'have lunch', borrokatu 'fight' and jolastu/jostatu 'play, have fun' have some unaccusative and some unergative properties in terms of, for example, subject case-marking and auxiliary selection. We account for this behavior by means of an analysis in which these verbs have a hybrid nature and their subject is introduced either high or low. This proposal adds to previous studies such as Kural (2002), Alexiadou and Anagnostopoulou (2004), Alexiadou and Schäfer (2011) and Irwin (2012), among others, which challenged the binary classification of intransitive verbs and argued that further divisions were necessary. Specifically, under the analysis proposed in this paper, agentive process verbs are "intermediate" verbs whose subject can be syntactically high or low (introduced either in the specifier of Voice or in the specifier of v), but which is interpreted as the argument of both Voice and v. Our analysis is supported by the mixed syntactic behavior of these predicates in Basque. In our analysis, we have claimed that the dialectal differences observed across Basque varieties are a consequence of the various combinations of different versions of Voice and v, and that these combinations are encoded in the LF-instructions (Harley 2014;Wood 2016) of the Roots in question, which can vary from one variety to another. In Eastern varieties of Basque the subject is introduced low, within the verbal phrase, as the specifier of vP, and it is co-indexed with a non-pronounced thematic argument introduced by Voice. Being an argument of both v and Voice, the DP argument shows mixed syntactic behavior. On the one hand, the structure does not allow causativization, aligning with unergative verbs. On the other, the subject gets absolutive case and the verb shows other unaccusative properties in a number of contexts: the subject can also be marked with the partitive, the verb is not compatible with cognate objects and it can form resultative adjectival participles.
However, this is not the only type of hybrid verb that the system allows. In fact, in agentive process verbs in Western and Central varieties of Basque, the overt DP is introduced high, in the specifier of Voice -rather than vP-internally -but this DP also binds the free variable introduced by v. Therefore, it is interpreted both as the agent argument of Voice and as the undergoer argument of v. This being the case, these verbs pattern with unergative verbs in that their subject is assigned ergative case, they do not allow causativization and they are compatible with cognate objects. However, as the subject is also semantically an argument of v, the predicates can form resultative adjectival participles. Thus, as can be seen, the diagnostics we use do not systematically separate unaccusative verbs from unergative ones, but merely target different structural and semantic properties. The results obtained in these diagnostics reveal the existence of hybrid verbs in Basque that cannot be categorically classified as unergative or unaccusative. If our analysis is right, we expect to find hybrid intransitive verbs in other languages as well. In fact, the hybrid status of these verbs is not restricted to Basque dialects. Significantly, the behavior of the verbs in each dialect of Basque is closely related to the use of semantically (and sometimes also phonologically) similar verbs in the Romance languages that are or have been historically in contact with these dialects, namely French and Occitan on the one hand and Spanish on the other (see Berro & Pineda forthcoming). In particular, agentive process verbs also present mixed distributional features in Romance, under a number of syntactic diagnostics: in both French and Occitan, these verbs tend to take the se clitic (usually found in unaccusative change-of-state verbs in both French and Occitan), they are not compatible with cognate objects, and they are sometimes acceptable in the adjectival participial form or the causative configuration. This resembles to a certain extent what is going on in the Eastern Basque varieties, which are spoken in the same area, and contrasts with the behavior of the verbs in Western and Central varieties of Basque and the Romance language in contact with them, namely Spanish. Therefore, it seems that the LF-instructions of the Roots in a given language of a bilingual speaker can influence the LF-instructions of semantically (and phonologically) similar verbs in the other language.
In sum, in this paper we have proposed an analysis for intransitive Basque verbs that are considered hybrid. The study of hybrid verbs in other languages, such as Romance, remains an important avenue for future research.