Basque impersonals in comparison

1 Introduction

The aim of this article is to provide an analysis of Basque impersonals (Albizu 1997, 2001; Ortiz de Urbina 1989, 1991, 2003, ).^[1] In using the term “impersonal” we refer to a detransitivized or valence decreased construction such as (1b), which alternates with a transitive counterpart as illustrated in (1a).

Several properties of the voice alternation in (1) deserve special attention. First of all, the transitive sentence in (1a) involves two arguments, agent and theme, marked by ergative and absolutive case respectively, whereas the impersonal in (1b) involves a sole argument marked by absolutive case. This argument is not the agent but the theme, as in (1a), the transitive counterpart of (1b). Second, Basque exhibiting a systematic auxiliary alternation, the auxiliary selected in (1a) is *edun ‘have’,^[2] i.e., the auxiliary attested in transitive clauses, whereas the auxiliary selected in (1b) is izan ‘be’, the auxiliary attested in intransitive clauses. Third, the inflected auxiliary form in (1a) agrees with both the ergative and the absolutive arguments, whereas the auxiliary form in (1b) shows only absolutive agreement.

Constructions similar to the one in (1b) have been described and analyzed in Basque under the label of “mediopassives” (Brettschneider 1979; de Rijk 2008; Ortiz de Urbina 2006; Rezac 2009) and “middles” (Berro and Fernández 2019; Urrestarazu 2019). Actually, (1b) can be taken as an example of a Basque middle, as discussed by Berro and Fernández (2019). Nevertheless, this example is due not to a particular middle-voice mechanism, as in English, but to a more general mechanism that is not as restrictive as in English and English-like middles, i.e., ‘type I middles’ in Ackema and Schoorlemmer’s (2005) typology. In this typology, two types of middles are distinguished crosslinguistically, ‘type I’ and ‘type II’ middle, with the former being much more restrictive than the latter, both syntactically and semantically. As will be shown, Basque impersonals share some properties with type I middles, as the thematic object is the grammatical subject and the implied agent cannot be formally represented as an adjunct. However, many other properties exhibited by Basque impersonals are also attested in canonical passives, showing that Basque goes beyond the behavior of ‘type I middles’ in the abovementioned typology. Crucially, in Basque impersonals, as generally in passives, there is an implicit agent that is syntactically active. Hence, this syntactically active implicit argument can license i) the PRO subject of purpose clauses; ii) agent-oriented adverbial modifiers; iii) agent-oriented instrumental and comitative phrases; and iv) secondary predicates.

Still, Basque impersonals show a property that distinguishes them from canonical passives. Firstly, Basque impersonals lack the counterpart of an English by-phrase, i.e., a demoted agent expressed by a postposition (Zuñiga and Kittilä 2019). This being the case, Basque impersonals would be akin to agentless passives. Secondly, Basque impersonals show a person restriction on the subject, as 1st and 2nd person subjects are not allowed. Finally, the demoted agent must be interpreted as an indefinite human, another restriction that passives are not subject to. If canonical passives are subject to these restrictions, then contemporary Basque impersonals cannot be regarded as passives. Nevertheless, with respect to some other properties, Basque impersonals behave similarly to passives. Particularly, they are strikingly similar to Romance reflexive-marked passives (see Mendikoetxea 1999, 2012 for Spanish and Cabredo-Hofherr 2017 for an overview of other Romance languages). In any case, the same construction used to be more similar to canonical passives in previous stages of the language, given that demoted agents marked by postpositions were allowed and there was no person restriction on the theme, as shown by Irigoyen (1992), Zulaika (1998) and Ortiz de Urbina (2006). This construction was termed “mediopassive” in Ortiz de Urbina’s (2006).

Last but not least, Basque impersonals seem to be similar to subject-suppressing impersonals (Blevins 2003), attested in Balto-Finnic and Balto-Slavic languages, among others. These impersonals have usually been analyzed under the “(impersonal) passive” label. However, as argued by Blevins, passives and impersonals must be distinguished from one another. First of all, impersonals can be built not only on transitive verbs but also on unergatives and unaccusatives. Secondly, the suppressed subject is interpreted as indefinite and human. Thirdly, the subject cannot be expressed by an oblique phrase. As argued by Blevins (2003), Polish -no/-to constructions (Kibort 2004) are impersonals of this type (similar constructions are also attested in other Slavic languages (Siewierska 1988)). As shown by Kibort (2004), these apparently subjectless constructions show that the implicit subject is, paradoxically, syntactically active as it can: i) license a reflexive anaphor; ii) control on purpose clauses; and iii) license agent-oriented adverbial modifiers, among other things. As we will see, all these properties are met in Basque impersonals (although licensing of anaphors needs some discussion). However, there is a crucial distinction between subject-suppressing impersonals of the Polish type and Basque impersonals: the former are valence-preserving constructions (for instance, impersonals built on transitive verbs retain the direct object) whereas the latter are detransitivized constructions with a theme subject (apart from the implicit but syntactically active subject). In this regard, Basque is closer to English passives than to Polish impersonals.

The article is structured as follows. In Section 1.1 we present some background information on Basque, and in Section 1.2 we provide some definitions of the main constructions discussed in the study. In Section 2 Basque impersonals are compared to middles, particularly with regard to their aspectual interpretation and modality (Section 2.1), auxiliary selection (Section 2.2), compatible verb classes (Section 2.3) and the anti-double object condition (Section 2.4). In Section 2.5 we provide an interim summary. In Section 3 we present the properties of Basque impersonals that are shared by passives (Section 3.1) and those that are not (Section 3.2), and we provide a second interim summary (Section 3.3). In Section 4, we explore a non-standard strategy to circumvent a non-third person subject restriction. In Section 5 we present the properties of subject-suppressing impersonals and compare them to Basque impersonals. We start with the main distinction (Section 5.1) and continue with the shared properties (Section 5.2). Finally, a third interim summary is provided (Section 5.3). Section 6 discusses the main results of the study, and in Section 7 we present our main conclusions.

2 Basque impersonals in comparison to middles

In this section, we will explore the similarities and differences between Basque impersonals and middles of the type illustrated in (7). As we saw in Section 2.1, in Basque a given labile verb like hil ‘kill/die’ can occur in three different morphosyntactic contexts: a causative construction (12a), an inchoative construction (12b) or an intransitive construction that is nonetheless interpreted agentively (12c), a construction we have termed “impersonal”.

As we mentioned in Section 1.2, the inchoative and impersonal variants are morphologically similar. In fact, they would be ambiguous if the modifiers berez ‘by themselves’ and zapata honekin ‘with this shoe’ were not included. In both constructions, the grammatical subject – which corresponds to the logical object – has absolutive case and is cross-referenced by absolutive agreement markers in the auxiliary, in this case izan ‘be’ (cf. 12a). Additionally, no logical subject bearing ergative case is present, nor are there ergative agreement markers on the auxiliary (cf. 12a). The thematic interpretations of (12b) and (12c) are different, however. In (12b) no agent is implied, i.e., the spiders have died or die without the volitional implication of an agent.^[6] In contrast, in (12c) it is understood that someone deliberately has killed or kills the spiders.^[7]

The potential ambiguity of the examples in (12b) and (12c) only arises if the predicate in question is labile, i.e., if it allows both the causative and the inchoative variants. In other verbs that only accept the transitive variant, the intransitive variant can only be interpreted as an impersonal, that is, in such a way that an implicit arbitrary agent is understood to initiate the process undergone by the logical object.

In modern Basque syntax this construction has been classified as impersonal (Albizu 1997, 2001; Berro et al. 2021; Ortiz de Urbina 1989, 1991, 2003, ), and more recently as middle (Urrestarazu 2019; see also Berro and Fernández 2019). It would be easy to argue that sentences like (12c) or (13b) are really middles. Middles are often regarded as intransitive predicates with an implicit agent which – unlike passives – show no special morphology in English (e.g., Keyser and Roeper 1984) – although some other languages such as Greek show special middle morphology. (14b) is a pertinent example. Regarding the implicit agent, middles would be similar to passives, differing from them in not exhibiting special morphology (see Section 2.1).

Sentences like (12c) and (13b) would naturally fit into the class of middles given that, as we have mentioned, they show no special morphology – thus apparently resembling inchoatives – but are interpreted agentively, like passives. In what follows, we will compare the Basque construction (12c) with middles, considering the tests and features often attributed to middles in the literature.

Middles have been defined as constructions that include a particular set of syntactic and semantic features. Schäfer (2006: 183) states that in middles (i) the grammatical subject corresponds to the thematic or internal object; (ii) the agent is demoted and has an arbitrary interpretation; and (iii) the aspectual interpretation of the sentence is non-episodic, that is, it has a generic or stative interpretation. Additionally, other properties of middles that have been reported are (iv) their necessary co-occurrence with a certain kind of modal adverb; (v) their restriction to particular verb classes; (vi) the impossibility of their occurring in the imperative or progressive; and in some languages (vii) their selection of the have auxiliary (Ackema and Schoorlemmer 2005; Keyser and Roeper 1984). As can be seen, middles have sometimes been argued to be quite restrictive in terms of the syntactic and semantic environments licensing them. Nevertheless, the cross-linguistic picture seems to be rather complex, and in several languages the constructions giving rise to middles have been shown to be more permissive (Lekakou 2005). For instance, Ackema and Schoorlemmer (2005) draw a distinction between what they call “type I middles”, which fulfill all the conditions mentioned above, and “type II middles”, which fulfill some but not all of those conditions. English has type I middles, whereas Romance languages like Spanish and French have type II middles. As will be shown in the following subsections, Basque patterns with Romance languages in having what Ackema and Schoorlemmer term “type II middles”. On the basis of our scrutiny of those conditions in Basque, we will conclude that the middle is best regarded as a particular semantic notion – not a particular morphosyntactic construction – and that the impersonal is a construction that is more broadly used in the language.

3 Basque impersonals in comparison to passives

Thus far we have shown that Basque impersonals cover a wider range of instances than English-type middles, i.e., type I middles in Ackema and Schoorlemmer’s (2005) typology. In this section, we will explore some properties of Basque impersonals that seem to match up with those of passives, thus departing from the behavior of type I middles. In particular, the implicit agent of Basque impersonals is syntactically active and not inert, as in type I middles. In this respect, Basque impersonals behave like passives. Still, this behavior in itself does not allow us to consider them prototypical passives (Kulikov 2010; Zuñiga and Kittilä 2019). In fact, other properties exhibited by Basque impersonals depart from those of passives in languages like English, such as the impossibility for the agent to be overtly marked by an instrumental phrase (similar to a by-phrase). As explained in Section 1.2, in this respect Basque impersonals could be considered instances of agentless passives. However, there are two other properties that separate the constructions under consideration from passives of any kind, to wit, the person restrictions of the grammatical subject, as it can only be 3rd person, and the obligatory human interpretation of the implicit logical subject. In the latter respect, Basque impersonals are closer to subject-suppressing impersonals (Blevins 2003) (see Section 5).

We will start this section by discussing three properties of English-type passives that are not attested in Basque impersonals: (i) the potential overt realization of the agent in a by-phrase; (ii) the absence of a person restriction on the grammatical subject; and (iii) the absence of an obligatory human interpretation of the implicit logical subject. We will then explore the properties shared by English-type passives and Basque impersonals, mainly those related to the syntactic behavior of the implicit argument (Ackema and Schoorlemmer 2005; Bhatt and Pancheva 2005 and references therein), namely (i) control of the subject of a rationale/purpose clause; (ii) licensing of the agent-oriented adverbs; (iii) licensing of agent-oriented secondary predicates; (iv) control of the subject of adjunct clauses; and (v) licensing of instrumental/comitative phrases.

4 Circumventing person restrictions: impersonals with a dative subject

In Section 3.1.2, we showed that Basque impersonals can only be formed if the grammatical subject (thematic object) is in the 3rd person. Leaving aside examples from ealier language stages like those shown in (47) and (48) and some exceptional and controversial contemporary examples as shown in (50) and (51), the impersonal construction is assumed to be ungrammatical with 1st or 2nd person grammatical subjects. As discussed, this is a feature that Basque impersonal shares with English middles and that sets them apart from English-type passives. In this connection, there is one strategy that allows the person restriction to be circumvented: the shift to a dative subject (Ortiz de Urbina 1988, 2003, ). In the examples below we can see a transitive personal sentence (69a), its impersonal ungrammatical counterpart (69b) and an alternative impersonal with the dative theme (69c).

In the ungrammatical impersonal construction in (69b), the thematic object becomes the grammatical subject, bearing both absolutive case and showing absolutive agreement. As shown in Section 3.1.2, this impersonal is ungrammatical because the grammatical subject is 2nd person. Nevertheless, if the 2nd person theme is marked dative and a corresponding dative agreement marker is included in the auxiliary, as in (69c), the result becomes grammatical for many speakers.

Much research is still needed to fully understand the nature of impersonals with dative theme. Nonetheless, we can share some preliminary thoughts. First of all, the strategy seems to be restricted to the spoken language, given that no written examples can be found in the Contemporary Reference Prose or Contemporary Dynamic Prose corpora of Basque. Second, this is unquestionably not a standard strategy and it does not seem to belong to a particular dialect of Basque. Specifically, it does not seem to be restricted to Western Basque, although the speakers we informally consulted are all from this dialectal area.^[14] Third, the strategy is reminiscent of differential object marking (DOM) (for Basque, see Fernández and Rezac 2010, 2016; Odria 2012, 2014, 2017), although, as pointed out by Ortiz de Urbina (2011–2019), the speakers that accept and produce these impersonal constructions do not necessarily accept DOM in the transitive structure, such as zuri ikusi dizut ‘I have seen you’, with a 2nd person dative object instead of the canonical absolutive. Dative themes of impersonals are also available with 3rd person subjects, though only if they are [+human].

Apart from dative themes, other dative arguments can be present in impersonal constructions. In fact, Example (69c), repeated here for convenience, can have another interpretation involving an indirect object.

This example can be interpreted not only as an impersonal including a 2nd person dative theme (69c) but also as an impersonal based on a ditransitive verb such as eraman ‘take’, like the example (73) (see Section 2.4 for the anti-double-object condition).

In this ditransitive construction, the 2nd person dative agreement marker corresponds to the covert indirect object (zuri ‘to you’). Similarly, in its impersonal counterpart (72), the dative argument is an indirect object and not a theme. In this second reading, there is a covert absolutive theme that could also appear overtly, such as gutuna ‘the letter’. Note that the auxiliary form is apparently the same, namely zaizu be.2sgDat[3abs], in both the impersonal involving the dative theme in (69c) and the impersonal involving the dative indirect object in (73). However, the former, unlike the latter, shows 3rd person dummy agreement that cross-references no absolutive argument.

The dative arguments occurring in both constructions are not syntactically similar. The dative theme of the impersonal licenses secondary predicates (74a), just like absolutive themes (74b).

In contrast, secondary predicates in impersonals involving a dative indirect object cannot be controlled by the dative. In (75), the predicates pozik ‘happy’ cannot be predicated of the dative argument zuri ‘you’– it can only be predicated of the implicit agent (indexed as i).

Thus, the dative theme of impersonals seems to be syntactically similar to the absolutive theme, at least with regard to the licensing of secondary predicates. This is reminiscent of the contrast found with DOM objects and indirect objects. In fact, DOM objects license secondary predicates even though they are marked dative, whereas indirect objects do not (Fernández and Rezac 2010, 2016; Odria 2012, 2014, 2017; Ortiz de Urbina and Fernández 2016). For a formal analysis of impersonals including dative themes, see Berro et al. (2021).

5 Basque impersonals and subject-suppressing impersonals

In the preceding section, we compared Basque impersonals to English-like passives. As we have seen, there are some striking similarities between the two languages regarding the syntactic behavior of the implicit subject. However, unlike English, Basque cannot express the logical subject by means of an oblique of the by-phrase type. This can be interpreted as meaning that Basque impersonals are agentless passives or, alternatively, that they are subject-suppressing impersonals (Blevins 2003). This particular instance of impersonal constructions has been observed and analyzed by Blevins in Balto-Finnic languages such as Estonian, Balto-Slavic languages such as Polish and Celtic languages such as Welsh. Blevins identifies three main features that distinguish these impersonals from passives. First of all, not only transitive verbs but also unaccusative and unergative verbs may participate in impersonal constructions. Secondly, the suppressed subject is interpreted as human and non-human subjects are not allowed. Third and lastly, agentive obliques of the by-phrase type analyzed above (Section 3.1.1) are either less acceptable or unacceptable altogether. As we will see, these three characteristics are shared by Basque impersonals. However, unlike the subject-suppressing impersonals analyzed by Blevins, Basque impersonals are detransitivized constructions. In this section, we will focus on one of the many constructions analyzed by Blevins under the label “impersonal”. In particular, we will compare Basque impersonals to Polish no/to-constructions (Blevins 2003; Kibort 2004; Siewierska 1988). As we will see, Basque impersonals share many characteristics with Polish no/to-constructions, such as the three general properties listed above, as well as other properties explored by Kibort (2004), e.g. concerning syntactic control and the binding of reflexive anaphora.

6 Discussion

In this section, we will discuss some of the outcomes of the present study, considering the status of Basque impersonals within the typological framework of middles, passives and subject-suppressing impersonals. On the one hand, middles and passives are similar in that in both cases the logic object is realized as the grammatical subject, and the agent or logical subject, although not formally present, is semantically implied. The latter is also true for subject-suppressing impersonals. On the other hand, these three constructions are different with regard to a number of other morphosyntactic properties, some of them discussed in this study. As shown in the previous sections, the Basque impersonal construction shares some properties with middles (of type I as classified in Ackema and Schoorlemmer 2005) and many properties with passives and subject-suppressing impersonals. Specifically, the Basque impersonal resembles middles and subject-suppressing impersonals in that the implicit external argument cannot be overtly expressed in an adpositional phrase and, particularly, it patterns with middles in that the grammatical (absolutive) subject must be 3rd person (Urrestarazu 2019).^[17] Nevertheless, Basque impersonals differ from the middle in languages such as English in that the event is actually instantiated, it allows more aspectual and modal interpretations, and it can be formed with more classes of verbs. In terms of all those properties, the Basque impersonal seems to be closer to the passive and the subject-suppressing impersonals than to middles. In fact, even though the implicit agent argument of impersonals is not reflected in auxiliary choice or verbal agreement, it is syntactically active, as can be concluded from the tests presented in Section 3.2: it can control purpose/rationale clauses and it can license agent-oriented adverbs, secondary predicates and instrumental phrases. Thus, in Basque impersonals, the implicit agent argument is both semantically and syntactically present, like in passives and subject-suppressing impersonals, and unlike in middles, which have semantically, but not syntactically, active agents. Moreover, Basque impersonals pattern with subject-suppressing impersonals in having a logical subject that is necessarily interpreted as human, but resemble passives in showing detransitivized verbal morphology.

Thus, attending to the syntactic contribution of the implicit agent in Basque impersonals and the detransitivized verbal morphology, one could consider the Basque impersonal construction to be in reality a sort of agentless passive. In fact, Ackema and Schoorlemmer (2005) suggest that type II middles – a group in which Basque impersonals could be classified – should be analyzed as passives. In a similar fashion, Lekakou (2005) argues that Greek and French middles, which according to her have a syntactically active implicit agent, are syntactically indistinguishable from passives. Nevertheless, the Basque impersonal exhibits a person restriction on the grammatical subject, a feature that significantly limits the use of the impersonal in comparison to the passive in other languages such as English or Spanish. Moreover, it also requires a human implicit agent, a property not attested in passives of any type. Therefore, Basque impersonals can be thought of as an intermediate construction which does not completely coincide with middles, passives or subject-suppressing impersonals, thus lying somewhere in the middle.

Whether Basque has a real passive construction or not is a topic that has been much debated in the literature. As already pointed out in the introduction to this article, Basque grammarians have attributed the term “passive” to some structures that do not fit the traditional typological definition. In fact, Saltarelli et al. (1988: 219) consider the construction involving the adjectival participle (ending in -a, see Berro 2019; Eguzkitza 1981; Ortiz de Urbina and Uribe-Etxebarria 1991) to be a passive (see also Alberdi and Sarasola 2001: 44–45, 160–164; Bollebacher 1977; Euskaltzaindia 1991: 294–296, 1993: 16–19; Rebuschi 1984; and others).

As can be seen in (97a) and (97b), in this construction the perfective participle is headed by -a and agrees in number with the theme gutuna/gutunak ‘letter/letters’, i.e., the direct object of the lexical verb bidali ‘send’.^[18] On the other hand, the implied ergative-marked agent is missing, although it could be introduced in a fixed position preceding the participle, as can be seen in (98).

Even in this case, the ergative agent does not trigger agreement, just like in (97). Thus, the inflected element is an intransitive copula (izan ‘be’ or the locative copula egon ‘be’), even though the verb bidali ‘send’ is (di)transitive and is interpreted agentively.

Given that in this construction the thematic object behaves like a grammatical subject and the agent is usually not overtly realized, it has been considered a passive by some authors (Alberdi and Sarasola 2001: 44–45, 160–164; Euskaltzaindia 1991: 294–296, 1993: 16–19; Rebuschi 1984; Saltarelli 1988; and others).^[19] However, as argued by Ortiz de Urbina and Uribe-Etxebarria (1991), there are many reasons not to regard it as a passive. In fact, the following evidence supports a biclausal analysis of the construction, where the inflected element, the copula, takes the participial clause – with subject/object gap – as a complement. First of all, the agent of the event denoted by the verb embedded in the participle is not demoted, as we have already shown in (98). This is particularly relevant when it comes to considering (97) a passive, i.e., a detransitivized construction in which the direct object becomes the subject (or in other words, the direct object is promoted) and the subject becomes an adjunct (i.e., it is demoted) and thus can be omitted. Second, there are some word-order constraints that are surprising in a free word order language such as Basque. For instance, the agent marked by ergative cannot precede the theme, a consequence of the fact that the agent is inside the participial clause and therefore not in the sphere of verbal inflection.

If this construction were monoclausal, the agent would be expected to appear either before the theme or after the inflected element, without causing ungrammaticality. Compare (99) with (100), a monoclausal transitive sentence where the agent is in the same clause as the inflected verb.

Third, in the presence of an indirect object marked by dative, dative agreement is not obligatory in the copula, a surprising fact taking into account that dative arguments must generally agree with the auxiliary.

These facts and other evidence suggest, as proposed by Ortiz de Urbina and Uribe-Etxebarria (1991), that the construction in (97) is really biclausal, and not an instance of a monoclausal passive (see Berro 2019 for a syntactic and semantic analysis).

On the other hand, in analyzing classical texts, some authors such as Altuna (1987: 232, 234) and Zulaika (1998: 779–784) use the label ‘passive’ not only for constructions like those in (97) but also for those instances of impersonals that can be found in previous stages of the language, where 1st and 2nd person subjects were allowed (Examples (47), (48) and (50)). We repeat Example (50c) as (102) here for convenience.

With regard to these two constructions, that is, the one including the -a headed participle in (97) and the one in (102), only the latter can be considered a true passive. Unlike the -a participial construction, the old impersonal construction is monoclausal, and has no person restrictions. In relation to this, it must be noted that the person restriction is one of the main criteria distinguishing Basque impersonals from passives cross-linguistically. Thus, the old impersonal fulfills almost all the conditions to be considered a real passive.

Last but not least, Urrestarazu (2019) refers to Basque impersonals as middles. This designation can be also misleading, as Basque impersonals behave in a way that is much more similar to passives and impersonals than to middles, at least to type I middles in Ackema and Schoorlemmer’s (2005) typology. As shown in Section 2, the Basque impersonal construction seems to be broader than this kind of middle. Given this, we claim that in Basque the middle is best regarded as a semantic interpretation, parasitic on an impersonal construction that can be used in more contexts than the middle (see Lekakou 2005 for a similar approach but defending a passive analysis). As we have shown, the impersonal has some properties that are usually attributed to passives, and in all these features, the Basque impersonal seems to be very close to the reflexive-marked passives of Romance languages, particularly those of Spanish. However, this line of analysis merits more detailed research.

All in all, the comparison of Basque impersonals with middles, passives and impersonals cross-linguistically, as well as with reflexive-marked passives in Spanish, reveals that it is very difficult to propose a single syntactic definition for passives, middles or impersonals that will hold cross-linguistically. In fact, all three constructions seem to be subject to variation across languages, presumably due to the morphosyntactic requirements of each language. For instance, regarding passives, Alexiadou (2014) and Alexiadou et al. (2015) have proposed two syntactic ways of forming passives: one is present in languages such as English or German, and the other is present in languages like Greek. Without going into the detail, the most important consequences of having two configurations is that the English- or German-type passive is much more productive and systematic than the Greek one, which, for instance, only accepts a subset of verbs (some agentive verbs) and exhibits some restrictions regarding the introduction of apo ‘by’ phrases. On the other hand, middles have also been shown to vary across languages (as we have also demonstrated in this article). In fact, Ackema and Schoorlemmer (2005) distinguish between type I and type II middles, with both of them having quite different morphosyntactic properties. Moreover, Lekakou (2005) argues that middles ascribe a dispositional property to the understood object, where dispositional ascriptions are subject-oriented generic sentences. According to her, languages differ in terms of the choice of the structure that gives rise to such an interpretation. Along the same lines, Fábregas and Putnam (2014) and Alexiadou (2014) propose that the middle is better defined as a semantic interpretation that can be obtained with different syntactic constructions across languages. In a different but similar direction, Kemmer (1993) suggested that the middle must be regarded as an expressive strategy with the particular function of signaling “a departure from the canonical transitive event type in the direction of an intransitive event type” (Kemmer 1993: 209), a definition that served better to account for the various realizations of the middle across languages. In conclusion, the morphosyntactic shape of middles varies across languages, and to a certain extent, the same is true of passives and impersonals. In fact, as we have seen, even though Basque impersonals share many properties with subject-suppressing impersonals as described by Blevins (2003), they are also different in the arrangement of verbal morphology and the type of grammatical subject permitted. Thus, as far as English-type middles, passives and subject-suppressing impersonals are concerned, Basque impersonals seem to share properties with all of these constructions, but do not fully coincide with any of them.

7 Conclusions

In this article, we have presented and analyzed Basque impersonals, comparing them in particular with English-type middles, English-type passives, and Polish subject-suppressing impersonals. We have concluded that, though Basque impersonals seem to be closer to the passive and the subject-suppressing impersonal, they cannot be fully categorized as either. First, Basque impersonals share with middles the inability to have the implicit agent realized as an adjunct, and the person restriction on the grammatical subject. Second, Basque impersonals are similar to the passive in that the subject is syntactically active: it allows agent-oriented modifiers, instrumental/comitative phrases, control of purpose clauses and secondary predicates co-indexed with the implicit agent. Finally, we have shown that Basque impersonals resemble subject-suppressing impersonals to a considerable degree, but differ from them in their detransitivized verbal morphology, and the person restriction on the grammatical subject. Thus, Basque impersonals have a syntactically active implicit agent, unlike English-type middles, but they cannot be considered either passives or subject-suppressing impersonals, as the person restriction on the grammatical subject considerably limits their distribution.

Abbreviations

abs

absolutive

abl

ablative

attr

attributivizer

comit

comitative

comp

comparative

dat

dative

det

determiner

erg

ergative

impers

impersonal

iness

inessive

ins

instrumental

ipfv

imperfective

neg

negation

n

neuter

nmlz

nominalizer

nvir

non-virile

pfv

perfective

pl

plural

pres

present

prf

prefix

pros

prospective

psr

possessor

ptcp

participle

ptcl

particle

ptv

partitive

pst

past

rel

relational

refl

reflexive

sg

singular

soc

sociative

subr

subordinator

sup

superlative

vir

virile