Involvee causatives in Turkish

. In this study, we investigate a morphologically but not semantically causative construction in Turkish, which we label ‘involvee causative’ (InvC). In contrast to regular causative, the external argument in InvCs is not interpreted as an agent or causer but as merely being involved in or experiencing the event. Additionally, InvCs also differ from regular causatives in not licensing agent-oriented adverbs, instrument phrases and passivization. In previous research, failure of these diagnostics to apply has been taken as evidence for an unaccusative structure. However, we argue against such an analysis for InvCs and show that the latter contain a thematic Voice head. The above diagnostics, we conclude, are sensitive not only to the syntactic status of the external argument but also to its semantic properties, and are licensed only if the argument receives an agentive interpretation.

We will refer to InvCs such as (1b) which pattern with lexical causatives as simple InvCs, those like (2b) which pattern with indirect causatives as complex InvCs.
While InvCs are formed productively, their use is subject to pragmatic restrictions, in that the event in which the subject is involved must qualify as significant or newsworthy.For instance, in (3), the transitive only allows for an Involvee reading if the sinking ship is famous (e.g., the Titanic) or if the sinking is staged in front of a large audience: (3) a. Gemi ship bat-tı.sink-PST.3SG'The ship sank.' b.(In an amusement park, we enjoy a lot of fun activities one by one, and next is the sinking of the famous Titanic ship): pro gemi-yi ship-ACC de de bat-ır-dı-k.sink-CAUS-PST-1PL Besides the non-causative interpretation of the subject argument, InvCs have three other properties which distinguish them from regular causatives.First, the Involvee cannot be associated with agent-oriented adverbs, (4b), which are normally licensed by the subject of causatives, as illustrated in (4a): (4) a. pro gemi-yi ship-ACC (s ¸evkle) enthusiastically bat-ır-dı-k.sink-CAUS-PST-1PL 'We sank the ship enthusiastically.' b. pro günes ¸-i sun-ACC (#s ¸evkle) enthusiastically bat-ır-dı-k.set-CAUS-PST-1PL #'We were involved enthusiastically in the event of the sun setting.'Secondly, in contrast to regular causatives (5a), neither simple (5b) nor complex (5c) InvCs permit instrumental phrases.In the former, the PP cannot receive an instrumental reading; in the latter, only the regular, indirect causative interpretation is available.
In this paper, we argue that InvCs, despite their non-standard properties, do not have an unaccusative structure but feature a thematic Voice head.Unlike in regular causatives, however, their external argument is not interpreted as an agent but as an involvee. 2The main contribution of this paper, besides the concrete analysis proposed for InvCs, is thus methodological: we establish that agent-oriented adverbs, instruments and passivization are sensitive not only to the syntactic presence of an external argument but also to its semantic interpretation.Moreover, InvCs demonstrate that causative morphology is independent of causative semantics.Complex InvCs in particular offer evidence that a single event can take two external arguments as long as their interpretation is sufficiently distinct, as we will show in Section 3.2.
We will proceed as follows.We first establish that InvCs are not unaccusatives (Section 2), then develop our own proposal based on the claim that InvCs contain thematic Voice (Section 3) and end by briefly considering the morphology of InvCs (Section 4).Section 5 concludes.
2. Against an unaccusative structure.In this section, we offer three pieces of evidence for the claim that InvCs are not unaccusatives, that is, do contain a thematic Voice head:3 control into purpose clauses, adverbial gerundives in -ArAk and minimum size of the embedded constituent.

CONTROL INTO PURPOSE CLAUSES.
In Turkish, unaccusatives do not allow control into purpose clauses (8). 4 This also holds for transitive anticausatives, analyzed by Schäfer (2022Schäfer ( , 2023) ) (Akkus ¸ 2021(Akkus ¸ , 2023)).Embedding of smaller structures such as vP is ruled out.InvCs, both simple ( 14) and complex (15), can be embedded in an indirect causative structure, again corroborating the claim that they contain a thematic VoiceP.3. Proposal.To give a brief intermediate summary, we have shown that InvCs contain a thematic Voice head, as evidenced by their ability to control into purpose clauses, combine with unergative/transitive -ArAk gerundives and be embedded in indirect causatives.We now propose an analysis based on this finding, looking first at simple and then at complex InvCs.
3.1.SIMPLE INVCS.We argue that simple InvCs such as (1b), repeated here as (16a), have the simple transitive structure given in (16c).The only difference to regular lexical causatives, with the structure in (16b), concerns the external argument, which in InvCs is interpreted as an involvee rather than as an agent.To formalize this observation, we propose that lexical causatives and InvCs contain different flavors of Voice which assign an agent and an involvee θ-role, respectively, to their specifier positions.This involvee θ-role should not be understood as a fully novel and distinct role but simply realizes a subset of proto-agent properties as proposed by Dowty (1991) (see also Baker 1997 for the same view): like prototypical agents, involvees are sentient, but they lack intentionality and causal power.This analysis leaves the question open why agent-oriented adverbs, instruments and passivization are not licensed with InvCs.We argue that this falls out from the non-agentive interpretation of the external argument.This claim is supported by the fact that the diagnostics in question equally fail in Turkish for other non-agent arguments such as instruments (17) and inanimate causers (18) (see also Kural 2000).Note that for obvious reasons, we do not give an example showing that instrument subjects cannot combine with instrumental modifiers.Taken together, these data confirm that agent-oriented adverbs, instruments and passivization are not only sensitive to the syntactic presence of an external argument but also to its semantic interpretation, being licensed only with agentive Spec,VoiceP arguments.The broader question this raises is, naturally, why this generalization should hold.On the one hand side, the fact that agentoriented adverbs and instruments are only licensed in the presence of an argument interpreted with intentionality and volition arguably does not come as a surprise but simply follows from the semantic and/or pragmatic requirements of this diagnostic.On the other hand, the fact that passivization requires an agentive argument in Turkish is less straightforward, and also does not seem to replicate in other languages such as English.If lexical causatives and InvCs are assumed to contain different flavors of Voice which have some syntactic r eality,5 passivization could be confined to agentive arguments by positing that only agentive Voice can have an empty specifier.We leave the further exploration of this question to future work.
So far, we have argued that InvCs are syntactically regular transitives but contain a non-agent external argument, similar to instruments and inanimate causers.However, this analogy might be taken to obscure a more fundamental difference: unlike any other kind of causative, regardless of the semantics of their external argument, InvCs are not interpreted as involving any causation.This raises the question whether pinpointing the difference between regular causatives and InvCs on the θ-role of the external argument is really sufficient or whether we should not rather develop a distinct eventive semantics for the latter.
What makes this question difficult to address is the fact that causative semantics has been analyzed in a variety of ways in the literature, and even more often, assumptions are not usu-ally made explicit.We propose here one answer based on what is arguably currently the standard approach to causative semantics, without regarding it as the last word in this matter.It has been argued that lexical causatives and anticausatives have the same event structure, containing a low stative projection and an event-introducing v head (19b-20b).Making use of the "Process" semantics (e.g., Pietroski 2004; Williams 2015, see also Biggs & Embick 2022), our analysis adopts the relation END featuring in both (19b) and (20b), rather than e.g., introducing a predicate Cause.The Process semantics is typically applied to the analysis of change-of-state verbs; the main intuition being an event e ENDs in a state s.6 Against this background, InvC causatives can be interpreted along the same lines (20c), with the sole difference being that the subject argument is not interpreted as the agent but as an involvee of the event: In short, under this view, InvCs do not require a distinct event semantics: the fact that they are not interpreted as causative follows exclusively from the non-agentive interpretation of the external argument.
As an alternative view on causative semantics, it has been suggested that lexical causatives, unlike anticausatives, might contain a functional head which explicitly encodes causative semantics (e.g., Cuervo's (2003) and Folli & Harley's (2007) v DO or v CAUS(E) , contrasting with v BECOME ).In this case, InvCs would require a semantic variant of this head.While we believe that the first approach to the event semantics of InvCs outlined above holds more promise, we leave it to future work to provide a more definite answer.

COMPLEX INVCS.
Having provided an analysis for simple InvCs, we now turn to complex InvCs such as (2b), repeated here as (21a).For the reader's reference, the structure of the regular indirect causative interpretation in the tradition of Pylkkänen ( 2008) is given in (21b).

Leyla
Continuing the strategy to treat InvCs largely on a par with regular causatives, we argue that complex InvC differ from (21b) in two ways.First, as already seen for simple InvCs, their external argument is interpreted as an involvee as opposed to an agent.Secondly, whereas regular indirect causatives are commonly analyzed as containing two separate events, a causing and a caused event, each encoded on a distinct v, this does not seem to hold for complex InvCs.Example (22a) shows that the involvee reading does not license two adverbs to each be associated with a distinct event, which is possible for the regular indirect causative reading.Equally, as shown in (22b), targeting two separate events by count phrases is only licensed with regular indirect causatives, not with complex InvCs.In short, the evidence indicates that complex InvCs do not contain two distinct v heads each encoding an event.
This observation lends itself to two analyses, the choice between which makes no difference as far as we can tell.Complex InvCs might either feature a semantically vacuous v or no embedding v at all, which would give rise to the Voice-over-Voice structure in (23) (see also Nie 2020).The fact that in this structure, agent is not the highest θ-role in the verbal domain, while perhaps unexpected, is in line with previous work by Bosse et al. (2012).Note also that while it is not possible to passivize on the involvee, as previously mentioned, it is possible to causativize on the embedded agent (in (21a), 'the thief') in complex InvCs, as correctly predicted by the structure in (23).
Overall, regardless of whether the embedding v is vacuous or absent altogether, both external arguments in complex InvCs are associated with the same event.This is a non-trivial finding.It is well-known that at least in the vast majority of languages, including Turkish, transitives cannot be causativized without the addition of a separate causing event,8 suggesting that in those cases, the two external arguments -causer and causee -cannot be part of the same event (but see Nie 2020).Hence, it is not immediately clear why such a constellation should be licensed in the case of complex InvCs.
We argue that nothing per se rules out two Spec,VoiceP arguments as part of a single verbal domain but that their co-occurrence is restricted by thematic uniqueness (Carlson 1984(Carlson , 1998;;Parsons 1990), in that their θ-role must be distinct.Concretely, two agent arguments, as in indirect causatives, cannot be part of the same event whereas an agent and an involvee, as in complex InvCs, can.Under this view, InvCs demonstrate that at least in Turkish, the co-occurrence of arguments within the same event is not, or not only, restricted syntactically but semantically.Two arguments occupying the same syntactic position can be part of the same event as long as their interpretation is sufficiently distinct.
The question this raises is, naturally, what precisely counts as sufficiently distinct.For instance, while there are substantial differences between the interpretation of agents, inanimate versity causatives.Adversity causatives are generally only possible on transitive structures for which there also exists an intransitive unaccusative counterpart.As such, they are morphophonologically identical to lexical causatives, built directly on a verbal root, and not to syntactic causatives in the many cases where these are morphophonologically distinguishable (e.g., when the syntactic causative is built on top of a lexical causative) (Wood & Marantz 2017:274).causers and instruments, respectively, these arguments do not seem to be compatible with each other.In response to this, we hypothesize that two external arguments might be blocked from cooccurring if they are both interpreted with causal force.Multicausality arguably poses a cognitive challenge for humans, suggesting that linguistic representations featuring two independent loci of causal power could be ruled out on such grounds.This would correctly account for the fact that involvees, lacking causal force, can be part of the same event as causally efficacious arguments such as agents, inanimate causers and instruments, while the latter are incompatible with each other.
4. The morphology of InvCs.Finally, we briefly turn to the question of how the causative morphology on InvCs, both simple and complex, should be analyzed.Crucially, since InvCs lack causative semantics, they provide evidence that the latter is independent from causative morphology (see Wood & Marantz 2017;Schäfer 2022).This contradicts the claim that causative morphemes spell out a dedicated head denoting a causal relation, such as v CAUS (Ramchand 2008;Key 2013;Harley 2017).Causative semantics and causative morphology, InvCs suggest, are correlated but not inextricably linked.
Instead, we propose two simple ways in which the causative marking on InvCs could be analyzed.In line with much previous research, the causative morpheme can either be regarded as the spell-out of v in the context of thematic Voice (Legate 2014;Wood 2015), or as realizing thematic Voice itself (Key to appear).In both cases, causative morphology is correctly predicted to be absent from the anticausative variants which lack a thematic Voice head.Note that both theories struggle with the fact that unergatives do not surface with causative morphology (Neu 2024); however, this is a general problem independent of InvCs.
Which of these two morphological analyses to adopt also depends on the question whether complex InvCs should be considered as containing a semantically vacuous v or no embedding v at all.In the latter case, the causative morpheme would have to spell out Voice whereas in the former case, both options remain on the table.In short, InvCs provide clear evidence against a strict coupling of causative semantics and causative morphology (in support of Wood & Marantz 2017;Schäfer 2022), but they are fully compatible with other standard approaches to the latter.

Conclusion.
To summarize, InvCs exhibit several non-standard properties compared to regular causatives, namely, failure to license agent-oriented adverbs, instrument phrases and passivization.In contrast to previous research which has taken these properties as indicative of an unaccusative syntax, we have provided several diagnostics demonstrating that InvCs must in fact contain a thematic Voice head.On a methodological level, we thus conclude that agent-oriented adverbs, instrument phrases and passivization do not constitute valid unaccusativity diagnostics cross-linguistically. Rather, at least in some languages they are sensitive to the interpretation of the external argument.Our findings thus have a direct bearing on future work on non-canonical transitives, as well as on argument structure in general.
Three main questions are left for further research to address.First, while the fact that agentoriented adverbs and instrument phrases require an agentive argument does not require an extensive explanation, we would like to have a better account for why the same holds for passivization in Turkish.To this end, it would also be useful to gain a clearer understanding of the typological facts, determining whether other languages place similar restrictions on passivization and how those restrictions compare with each other.
Secondly, we have provided only a tentative answer to the question of how to derive the non-causative interpretation of InvCs.We believe that InvCs can provide a valuable testing ground for both past and future theories of causative event semantics.Such theories will need to account for the fact that what we intuitively perceive as the causative component of the semantics of transitives can be systematically absent without any associated morphosyntactic change.
Finally, complex InvCs have raised intriguing questions about the uniqueness conditions holding on the arguments of a single event.We have argued that a single event can be associated with two Spec,VoiceP arguments as long as their interpretation or θ-role is sufficiently distinct, such as agent and involvee arguments, and we have suggested that the relevant factor distinguishing the two might be the fact that involvees lack causal force.Finding support or counterevidence for this claim cross-linguistically and exploring its broader cognitive implications would be a promising avenue for future research.
We were involved/part of the event when the daylight went away.' YES: #'We caused the daylight to go away.ıl-dı.darken-CAUS-PASS-PSTNO: 'The daylight went away, and we were somehow involved/part of it.'YES: #'The daylight was caused (by us) to go away.' -PST YES: 'Leyla was somehow involved in the thief stealing the purse (e.g., by carelessly leaving the purse on the ground)' YES: 'Leyla caused the thief to steal the purse.' bdır-ıl-mıs ¸. steal-CAUS-PASS-PST NO: The purse was stolen by the thief, and Leyla was involved in/affected by this.' YES: 'The purse was made by Leyla [for the thief to steal ].' tir-di.do]-CAUS-PST 'Final exams made [us see through the morning].'(15) Dikkatsizlik carelessness [Leyla-ya Leyla-DAT c ¸anta-yı purse-ACC hırsız-a thief-DAT c ¸al]-dır-mıs ¸. steal-CAUS-PST 'Carelessness caused [Leyla to have the purse stolen by the thief (on her)].' 5 'The prisoner ended the day (while) being beaten.' (16) a. pro günes ¸-i sun-ACC bat-ır-dı-k.set-CAUS-PST-1PL YES: 'We set the sun.' YES: 'The sun set, and we were involved/around when it happened.'b.
calmly made [the thief steal the purse in a flash].'NO: 'Leyla calmly had the thief steal the purse in a flash (on her).ontwo separate occasions made [the thief steal the purse 3 times].'NO: 'Leyla on two separate occasions had the thief steal the purse 3 times (on her).' 7