Turkish Causatives are Recursive: A Response to Key 2013 Open Access

In Turkish, it is possible to form a “double” causative, involving two causees and two instances of the productive causative affix on the same verb stem, as shown in (1).¹

Turkish double causatives are usually assumed to involve two causing events in addition to the caused event (e.g., Göksel 1993, Kural 1996, Çetinoğlu, Butt, and Oflazer 2009), much like make-causatives in English. However, Key’s (2013) dissertation on the morphosyntax of Turkish causatives claims that apparent double causatives in Turkish do not in fact involve the syntactic recursion of causing events; instead, they merely involve morphological reduplication of the causative morpheme. According to Key, then, Turkish does not have true double causatives. Key extends his argument to several other languages with productive affixal causatives, suggesting that double causatives in these languages are illusory and never encode more than two events in the syntax (one causing event and one caused event). If his claim is correct, then it would have important implications for the syntax of causatives and other complex predicates; we would have to explain why productive affixal causatives never recurse in these languages and why they differ in this respect from periphrastic causatives in other languages.

In this squib, I demonstrate using evidence from several diagnostic tests for eventhood that Turkish double causatives indeed have two distinct, syntactically represented causing events in addition to the caused event. Thus, productive affixal causatives in at least some languages involve true causative recursion. This finding supports approaches to affixal causatives that allow embedding of the same category, such as flavors of little v (e.g., Harley 1995, Pylkkänen 2008, Legate 2014), over approaches that employ fixed functional hierarchies to capture causative structure (e.g., Ramchand 2008, Key 2013).

Productive affixal causatives in Turkish have been shown to encode two events that stand in a causal relation: one event (the “causing event”) brings about another event (the “caused event”). Some evidence for this in the literature has come from scope of negation (e.g., Bainbridge 1987, Göksel 1993, Kural 1996). The negated simple causative in (2) has two interpretations: one where negation scopes over the causing event (2a), and another where negation scopes over just the caused event (2b). While Kural (1996) points out a confound in the scope-of-negation test with respect to the type of causation conveyed (I return to this point in section 3.4), it is by all accounts agreed that there are two syntactically represented events in simple productive causatives in Turkish.

Double causative constructions, then, might be expected to encode three events in Turkish: a higher causing event, an intermediate causing event, and the caused event. However, Turkish does not always exhibit a one-to-one relationship between causative marking and interpretation. As shown in (3), the same situation involving two causees can be expressed using either one or two instances of the causative marker on the verb.

Turkish also allows the opposite pattern: a causative with one causee can cooccur with either single, double, or even quintuple causative marking, as in (4). As Kural (1996:125) writes, “Turkish places no upper limit as to how many times the causative morpheme may be iterated” (see also Göksel 1993).

Thus, there is not always a one-to-one mapping between the number of causees and the number of causative affixes in Turkish. Key (2013) reports similar evidence to this effect from Hungarian (Hetzron 1976), Kashmiri (Manetta 2014), and Tsez (Kulikov 1993).

From this evidence, Key argues that double causatives in Turkish maximally encode two events in the syntax (one causing event and one caused event).² To account for the availability of multiple causees as in (3), Key suggests that “the semantics of indirect causation allow for an indefinite number of causal links”; however, these links crucially “are not syntactically represented” (2013:225). That is, intermediate causees might be associated with intermediate causing events in the semantics, but not in the syntax. Key does not address the issue of how exactly intermediate causees can be overtly pronounced without an associated causing event in the syntax. He nonetheless proposes that such constructions encode just one causative head/event, and that double causatives arise when a causative head marked with a [+focus] feature undergoes morphological reduplication (Key 2013:220). Despite providing no evidence from eventhood diagnostics to support his claim, Key concludes that double causative constructions in Turkish and other affixal languages are illusory (see also Svenonius 2005, Harley 2017).

Key points out that theories of affixal causatives that appeal to some form of recursion, defined here as the embedding of one category within another category of the same type, predict that true double causatives should be attested. For instance, the flavors-of-little-v approach (Harley 1995, 2008, Cuervo 2003, Folli and Harley 2005, Pylkkänen 2008, Legate 2014) allows one vP to embed a second vP, which can be of one of various flavors, such that a causative vP should be able to embed another causative vP. In order to avoid predicting such causative recursion, Key (2013:216) proposes instead that affixal causatives obey the fixed universal functional hierarchy in (5) where a dedicated Caus head, associated with the causing event and the causee, selects for a vP associated with the caused event; Voice (Kratzer 1996) introduces the causer argument.

Since the vP in (5) is crucially selected by Caus and not another vP, there is no category recursion, thus precluding the possibility of causative recursion.

While at least some double causatives in Turkish may indeed involve reduplication of a single causative morpheme, I contend in this squib that this is not the case for all double causative constructions. In the next section, I show using diagnostic tests for eventhood that true double causatives do exist in Turkish.

Previous work has revealed the presence of two events in Turkish simple causatives. In this section, I use four diagnostic tests for eventhood to demonstrate the presence of three events (two causing events and a caused event) in Turkish double causatives: scope of ‘again’, manner adverbials, temporal adverbials, and permission readings.

The data reported in this section involve double causatives of unergative predicates with animate causees. The data represent truth value judgments provided by at least three native-speaker consultants, based on the contexts given (or similar ones). Some speakers found double causatives of transitive predicates to be degraded, independently of the eventhood diagnostics. This could be due to a number of reasons: a general dispreference in the language for multiple dative arguments in the same clause (e.g., Kural 1996), a difference in the syntax of unergative and transitive causees,³ and/or parsing limitations. I leave this issue for future research, focusing on the grammatical double causatives of unergatives and whether their causing events can recurse.

‘Again’ attachment has been widely used to diagnose event decomposition in the syntax (e.g., McCawley 1968, Dowty 1979, von Stechow 1996); ‘again’ introduces a presupposition that an event has previously occurred at least once. Simple causatives with the adverb yine ‘again’ in Turkish are ambiguous between two interpretations, one where it scopes over the causing event (6a) and another where it scopes over just the caused event (6b).

Double causatives with yine are ambiguous between three interpretations, where yine scopes over the higher causing event (7a), the intermediate causing event (7b), or just the caused event (7c). Thus, ‘again’ diagnoses the presence of three syntactically represented events in double causatives in Turkish.

Distinct events should also be able to receive independent manner modification (e.g., Davidson 1967, Jackendoff 1972). Examples of a double causative with the manner adverb sessizce ‘quietly’ are given in (8) and (9). Different readings emerge depending on the placement of the adverb. When the adverb immediately precedes the direct causee (i.e., the argument that participates in the caused event), the adverb can modify either the higher causing event or the intermediate causing event (8). When the adverb follows the direct causee, it can modify either the higher causing event or the caused event (9). The unavailable readings for each word order are indicated using the symbol #.

Example (7) from section 3.1 showed that ‘again’ can attach at three different points in double causatives in Turkish. It is not clear to me why manner adverbs appear to attach at only two, with the interpretations modulated by word order; information structure may play a role. Setting this issue aside, the interpretations available with manner adverbials nonetheless indicate the presence of three events in Turkish double causatives.

Like manner adverbials, temporal adverbials should also be able to independently modify each event in a double causative (e.g., Fodor 1970, Jackendoff 1972, Wierzbicka 1975). A double causative with two temporal adverbials is given in (10). While pazartesi ‘Monday’ modifies the higher causing event, gelecek hafta ‘next week’ can modify either the intermediate causing event (10a) or the caused event (10b). Thus, temporal adverbials also indicate the presence of three events in Turkish double causatives.⁴

In (10), gelecek hafta linearly follows the direct causee. The same interpretations are available if the temporal adverb immediately precedes the direct causee.

Finally, Turkish verbs marked with causative morphology often also have a permission reading (e.g., Bainbridge 1987, Göksel 1993, Kural 1996). Example (11), for instance, is ambiguous between a causative and a permission interpretation.

Recall from section 2 that scope of negation has been used to diagnose the presence of two events in Turkish simple causatives, where negation appears to scope over either the causing event or the caused event, as in (12). However, the negative morpheme always appears outside of the causative morpheme in Turkish; allowing negation to scope below the causing event would therefore violate the Mirror Principle (Baker 1985). Kural (1996, 1997) argues that this puzzle can be solved, using evidence from the permission reading of Turkish causatives. He shows that the scope order CAUS > NEG is truth-conditionally equivalent to NEG > LET, as in (12b). Thus, “it can be maintained that negation uniformly takes scope over [the causative morpheme] in Turkish” (Kural 1996:89), and differences in interpretation derive from the causative vs. permission reading of the causative morpheme.

Permission readings are also found in double causatives. A canonical causing event can embed another causing event (13a) or a permission event (13b). A permission event can similarly embed another permission event (13c) or a causing event (13d). Permission readings therefore also diagnose the presence of three events in double causatives, including two events associated with the causative morpheme.

Summing up, the evidence from scope of ‘again’, manner adverbials, temporal adverbials, and permission readings all indicates that Turkish double causatives encode three events in the syntax: two causing events and a caused event.⁵

The evidence from eventhood diagnostics demonstrates that productive affixal causatives of unergatives in Turkish are indeed recursive; they allow the embedding of multiple events. This result accords with previous work reporting the availability of true double causatives in some varieties of Japanese (Kuroda 1993, Nie 2020a). Like Turkish, Japanese allows double causatives to be expressed with one or two causative markers on the verb. Kuroda (1993) suggests that a process of morphological haplology deletes one of the causative morphemes in Japanese. Turkish similarly exhibits morphological haplology (deletion) and phonological haplology (dissimilation) in a number of domains (Kornfilt 1986, Haig 2004). Tat and Kornfilt (2018) show that Turkish causatives undergo dissimilation; I suggest that causatives may also be subject to morpheme deletion.

These results furthermore support the predictions of recursive-embedding approaches to causatives, such as flavors of v (Harley 1995, 2008, Cuervo 2003, Folli and Harley 2005, Pylkkänen 2008, Legate 2014). The flavors-of-v analysis is sketched in (14), in which a vP headed by any event-introducing v can be embedded by another vP, whose head introduces a causing event. I assume here that the causee in Turkish is introduced by a lower Voice head, as has been suggested for other languages (Harley 2013, Legate 2014, Nash 2020, Nie 2020b, Akkuş 2021b, Sigurðsson and Wood 2021).⁶

Importantly, the causative vP in (14) can be further embedded by another causative vP, thus deriving double causatives and causative recursion. By contrast, approaches to causatives that employ fixed functional hierarchies (e.g., Ramchand 2008, Key 2013) are unable to account for the availability of causative recursion.

The finding that recursion is possible in languages with affixal causatives also provides support for a syntactic approach to argument structure and word building (e.g., Marantz 1984, Baker 1988, Halle and Marantz 1993, Pesetsky 1995, Borer 2005, Koopman 2005). Productive affixal causatives in Turkish and Japanese are recursive, just like periphrastic causatives in other languages.⁷ This suggests that the same recursive rules that operate on words and phrases also apply below the word level, to morphemes. Word structure and phrase structure are thereby built the same way.

Finally, there has been recent work arguing that productive causatives in some languages involve stacking of just the Voice head, without an additional little v (Nash 2020, Nie 2020a,b, Sigurðsson and Wood 2021). Such Voice-type causatives do not involve a syntactically identifiable causing event; the VoiceP that introduces the causer thus directly embeds the VoiceP that introduces the causee. In Nie 2020a, I show that, despite its being in principle possible, Voice-type causatives do not recurse without limit; they cannot embed a third VoiceP. The question thus arises why vP-type causatives can recurse but Voice-type causatives cannot. I suggest that this contrast lies not in whether Voice and v heads themselves can recurse, but in independent restrictions on thematic role assignment: namely, that the causee role (just like any other role) can only be assigned once per thematic domain. I propose that thematic domains are defined by events. Thus, vP-type causatives, which encode multiple events, can introduce multiple causees, whereas Voice-type causatives, which encode a single event, do not allow multiple causees.

To conclude, this squib has demonstrated using evidence from syntactic eventhood diagnostics that Turkish has true double causatives. I argued, contra Key 2013, that productive affixal causatives in Turkish do allow the recursion of events, which is best captured in recursive-embedding approaches to causatives.

I am grateful to Furkan Dikmen, Rümeysa Dilje, and Özlem Ergelen for providing and collecting Turkish judgments. Thanks also to Faruk Akkuş, Laura Kalin, Alec Marantz, and audiences at NYU, Leiden, HU Berlin, and NELS 51 for their feedback at various stages of this work. Comments from two anonymous reviewers greatly helped to improve the squib. This research was supported by funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (LeibnizDream, grant agreement 856421; PIs: Artemis Alexiadou, Maria Teresa Guasti, Uli Sauerland). Any errors are mine.