1 Introduction

Lexical causative verbs such as the verb in (1a) and their anticausative counterparts such as the verb in (1b) are standardly assumed to have semantics along the lines of (2a, b) (e.g. Parsons 1990). Under this view (henceforth: ‘the standard theory’), a causative verb denotes a causal relation between an event and a change of state, the latter being denoted by the corresponding anticausative verb. Consequently, the truth of a clause headed by a causative verb entails the truth of the clause headed by its anticausative counterpart, that is if (1a) is true then (1b) is necessarily true, too:¹

Most English causative-anticausative verb pairs are morphologically identical, constituting a so-called labile alternation (with very few exceptions such as lay vs. lie or raise vs. rise, which show stem allomorphy). Other languages, however, often differentiate one of the alternates morphologically (e.g. Haspelmath 1993). Many Indo-European languages also have, besides a set of labile or unmarked anticausative verbs, a set of morphologically marked anticausative verbs. The Spanish examples in (3) and (4) illustrate this. While the anticausative verb in (3b) is morphologically identical to its causative counterpart in (3a), the anticausative verb in (4b) differs from its causative counterpart in (4a) in that it obligatorily co-occurs with a SE-reflexive clitic (other Indo-European languages use a free SE-reflexive pronoun or a reflexive verbal affix to mark (a subset of) their anticausatives.)² Without going into any theoretical details, many accounts build on the pre-theoretical intuition depicted in (5a, b) that the presence of the SE-morpheme in marked anticausatives reflects the absence of the external causer argument and the eventuality introducing this external causer argument (the CAUSE predicate and its argument y) in the semantic representation of anticausatives (e.g. Grimshaw 1981; Reinhart 2000; 2002; Doron 2003; Reinhart & Siloni 2005; Schäfer 2008; Alexiadou et al. 2015, though the theoretical assumptions and semantic details differ a lot). Consequently, marked anticausatives denote simple inchoative events, too, as shown in (5b).

Under the standard account then, SE-marked and unmarked anticausatives do not differ semantically (cf. (5b) and (2b)); both denote simple inchoative events undergone by their sole DP argument. Consequently, both should be entailed by their causative counterparts because the latter denotes the causation of the event denoted by the former.

This standard view on anticausatives in general, and on reflexively marked anticausatives in particular, leaves open, however, an important puzzle. Why are (across languages and far beyond the Indo-European language family; cf. Haspelmath 1993) markers of anticausatives often syncretic with markers of reflexivity, that is, why do we find so often the same morphological marker with anticausative verbs that normally produces canonically reflexive verbs? The Spanish canonically reflexive verb in (6b) illustrates the point again; (6a) shows what we call the ‘transitive-disjoint use’ of the very same verb.³

In (6b), the SE-morpheme arguably acts as a reflexivizer with the meaning in (7) (or as a locally bound variable, i.e. an anaphor)⁴ in that it takes a transitive relation ℜ such as the transitive verb lavar (wash) in (6a) with the semantics in (8a) as its argument and identifies the two arguments of the relation, thereby producing the reflexive verbal meaning in (8b). The sentence in (6b) then means that the boy was the agent as well as the patient of a washing event:

A number of scholars, in particular Chierchia (2004) and Koontz-Garboden (2009), suggested that the puzzling syncretism between SE-marked anticausatives and SE-marked canonically reflexive verbs follows from an identical semantic content or function of the SE-morpheme in the two verb classes; SE acts as a reflexivizer in anticausatives just as it does in canonical reflexive verbs. We call this proposal the ‘reflexivization analysis of anticausatives’ (henceforth RAoAC). Under this proposal, a SE-marked anticausative verb is derived from a transitive lexical entry simply by adding the SE-reflexivizer in (7). The only difference between verbs undergoing canonical reflexivization and verbs forming marked anticausatives lies in the nature of the external argument role. As discussed in more detail in the next section, the external argument role of a causative verb with an anticausative alternate is not a human agent but it is an underspecified effector (see 9a), which can also be realized as a non-human causer entity. Adding the reflexivizer in (7) to the transitive causative verb in (9a) sets identical the external argument effector with the internal argument theme as in (9b). Then, according to (9b) the anticausative clause in (4b) says that ‘the glass’ was the effector as well as the theme of a breaking event. Since break is a causative verb, this means that ‘the glass caused its own breaking’.

Note that the reflexivization analysis of anticausatives in (9a, b) crucially differs from the standard analysis in (2a, b/4a, b) in that it predicts that a causative clause does not entail its corresponding anticausative clause, i.e. (4a) should not entail (4b). This is the same as with the canonically reflexive sentence (6b), which is, of course, not entailed by its transitive-disjoint counterpart in (6a).

The reflexivization analysis of anticausatives (RAoAC) was proposed most prominently in the work by Chierchia (2004) for Italian and Koontz-Garboden (2009) for Spanish (see also Beavers & Koontz-Garboden 2013a; b) and we will concentrate on their proposals because they provide explicit semantics that allow for falsification (in particular (9a, b) which is taken from Koontz-Garboden 2009). However, other authors have made (sometimes only in passing) similar suggestions for different languages, for example Manzini & Savoia (2001; 2011) for Italian, Medová (2012) for Czech, Beavers & Zubair (2013) for Sinhala, Fehrmann et al. (2014) for Russian or Lundquist et al. (ms.) for Norwegian.

We will discuss a number of conceptual and empirical arguments that suggest that the RAoAC cannot be upheld.⁵ In fact, a number of authors have argued against the RAoAC before, in particular Horvath & Siloni (2011; 2013) (with rejoinders in Beavers & Koontz-Garboden 2013a; b) and Alexiadou et al. (2015). In section 2, we will take up from these preceding publications some basically conceptual aspects in the discussion of the RAoAC in order to set some background. In section 3, we will add new semantic arguments that are complementary to those given in the above-mentioned earlier work. We, thereby, hope to add further support to the conclusion reached by these authors and to further clarify aspects in the argumentation that seemed to remain undecided. In particular, we will apply in section 3 tests that show that marked and unmarked anticausatives are entailed by their causative counterparts, while the same does, of course, not hold for canonically reflexive verbs and their transitive-disjoint counterparts. As we will also discuss, the arguments presented in the literature in favor of the RAoAC turn out to be conceptually and/or empirically misguided; under closer scrutiny they do not point to a reflexive relationship between causatives and anticausatives but to a scalar relationship. Such a scalar relationship follows from the standard semantics in (2a, b)/(4a, b) which state that a causative clause entails its anticausative counterpart. As is well known, scalar predicates can trigger implicatures; and in fact, the use of an anticausative often triggers the implicature that its causative counterpart does not hold true, an effect that derives a number of observations in the context of the causative alternation in general, and in the discussion of the RAoAC in particular (see Rappaport Hovav 2014 for related observations about contextual restrictions on the use of (anti-)causatives).

More far-reaching, our results support the conclusion that SE-marked anticausatives form a semantically uniform class with unmarked anticausatives (e.g. Schäfer 2008; Horvath & Siloni 2011; 2013; Alexiadou et al. 2015) as well as with other clearly inchoative structures such as non-alternating unaccusative verbs and eventive copula+adjective constructions (Schäfer 2008; Alexiadou et al. 2015). This means that the SE-morpheme does not always act as a reflexivizer (or a bound anaphor; cf. fn. 4) and that the identical morphological marking found in anticausatives and canonically reflexive verbs across languages is a real syncretism (same form, different function). However, as we will discuss, nothing (except world knowledge) blocks semantic reflexivization of causative verbs. Therefore, a surface string such as (4b) can, in principle, receive the reading in (8b), but only under very specific circumstances. The RAoAC, on the other hand, wrongly predicts meaning (2b) to be generally unavailable for SE-marked anticausatives.

For practical reasons, we restrict our discussion to Indo-European languages with a SE-morpheme/SE-reflexive element. In fact, we will mainly use examples from German and Spanish, but in order to reach some cross-linguistic significance, we will set out at least some of our empirical tests in two Romance languages (Spanish, Italian), two Germanic languages (German, Norwegian) and two Slavic languages (Russian and Czech). The semantic and pragmatic tests presented below should, however, be applicable to languages from other language families using different morphological devices, too.⁶

Before we turn to a discussion of the semantic predictions of the RAoAC, we will first look in the next section at some conceptual arguments that might speak in favor of the RAoAC as this allows us to characterize a bit more this proposal as well as some linguistic issues against which this proposal should be evaluated.

2 Conceptual evaluation of the RAoAC

In this section, we survey three conceptual arguments brought forward in favor of the RAoAC.

One main argument concerns the puzzle about the multifunctionality of reflexive morphology. If SE acts as a reflexivizer both in canonical reflexive verbs and in marked anticausatives, then the puzzle why SE shows up with anticausatives is solved. It applies to causative verbs to derive, in a well-defined and fully compositional way, what received the descriptive term ‘marked anticausative’ in the literature. Under the standard analysis, the semantic impact of the SE-morpheme in anticausatives remains unclear; in any case, it cannot be reduced to the canonical use of the SE-morpheme.

As is well known, SE-reflexive morphemes can be used for a wider range of constructions than just canonically reflexive verbs and marked anticausatives; in many languages, they are also used in generic middles, passives or even in impersonal constructions (cf. Spanish (10a-c)) (see also Koontz-Garboden 2009: 92 fn. 11; Horvath & Siloni 2013: 218). We agree with Horvath & Siloni (2013) that it is impossible to account for the semantics of these constructions in terms of reflexivization. If true, morphological identity does not necessarily reflect semantic identity and this particular conceptual advantage of the RAoAC vanishes. However, the data in (10a-c) do by no means invalidate the RAoAC. In any case, the question remains significant: how do we correctly divide the space marked with SE-morphemes into a semantically reflexive part and one with different semantics.⁷

Second, Koontz-Garboden (2007; 2009) points out that the RAoAC avoids the violation of the monotonicity hypothesis in (11):

Many accounts that assume the standard semantics of (marked) anticausatives in (5b), on the other hand, violate monotonicity. This is so if they assume that verbs forming marked anticausatives are basically stored with their transitive lexical entry (i.e. 5a) and the SE-marked anticausative use of the verb with the semantics in (5b) is derived from the transitive entry. Under such a view, adding the SE-morpheme to the entry in (5a) leads to a deletion of the CAUSE predicate (cf. (5b)), and this clearly violates monotonicity. However, the problem arises only if SE-marked anticausatives are literally derived from their transitive counterpart. There are, however, theories of word formation that technically do not derive marked anticausatives from their causative variants (e.g. Piñón 2001; Doron 2003; Schäfer 2008; Alexiadou et al. 2015); these do not violate monotonicity (as already pointed out by Horvath & Siloni 2013: 218). The latter authors argue that the lexical derivation of anticausatives from causatives as it is executed in the Theta System (Reinhart 2000; 2002; Reinhart & Siloni 2005 et seq.) should not be seen as a violation of monotonicity. In this framework, decausativization applies at a lexical level that is assumed to lack event decomposition; consequently, it does not delete a causative event but only the thematic information about a verb’s external argument/causer (see Horvath & Siloni 2013: 218 for details).

We conclude that the above conceptual arguments in favor of the RAoAC are not decisive. Whether SE-marked anticausatives should be subsumed under the semantics of canonically reflexive verbs seems to remain then a purely empirical question. When we turn to a third conceptual argument, however, brought forward in favor of the RAoAC, we will see that under closer scrutiny, this argument not only turns against the RAoAC under conceptual reasoning, but it also brings about a first set of empirical problems for the RAoAC.

Koontz-Garboden (2009) argues that only the RAoAC can derive the so-called ‘underspecified external argument generalization’ on anticausative formation given in (12), which was first discussed in the work by Levin & Rappaport Hovav (1995) and Reinhart (2000). Other theories, he argues, in particular those building on detransitivization, must stipulate the existence of (12).⁸

The Spanish examples in (13) and (14) illustrate the generalization in (12). A verb like desmontar (‘dismantle’) selects a human agent as its external argument; non-human causers and instruments are not allowed (13a). (12) correctly predicts that this verb does not form an anticausative (13b). The verb romper (‘break’), on the other hand, selects an underspecified effector as its external argument since agents, non-human causers as well as instruments can appear (14a), and in accordance with (12) the verb allows an anticausative use (14b).

Recall that the thematic underspecification of the external argument of verbs like romper is at the core of the RAoAC. Since romper selects an underspecified effector lacking any agent entailments, the non-human theme the window can also be assigned this external effector role in (14b) under reflexivization. The ungrammaticality of (13b) on the other hand follows because the verb involved selects for an agent external argument and since the theme is non-human and lacks agentive properties, this selectional restriction is not fulfilled if (13b) involves reflexivization.

Explaining the generalization in (12) via the RAoAC raises, however, a problem. To see this, note that Koontz-Garboden (2009) assumes that only SE-marked anticausatives are semantically reflexive (cf. (9b)). Unmarked anticausatives such as (3b) on the other hand are assumed to have the standard semantics in (2b/4b) expressing simple one-place inchoative events. Unmarked anticausatives must then be lexically stored and the causative version must be derived by causativization.⁹ However, the generalization in (12) holds for marked as well as unmarked anticausatives. While the RAoAC seems to account for the fact that marked anticausatives are subject to (12), it has nothing to say as to why all verbs forming unmarked anticausatives also allow non-agentive external arguments in their transitive use, i.e. why they are subject to (12), too. Assuming that SE-marked and unmarked anticausatives have fundamentally different semantics forces to come up with two different explanations of what seems to be a unitary phenomenon.

It should be noted that this problem does not exist under a version of the RAoAC as Chierchia (2004) suggests it (cf. also Lundquist et al. to appear for Norwegian). He proposes that both marked and unmarked anticausatives are derived by reflexivization from a transitive lexical entry; reflexivization is just not always reflected by surface morphology. The idea is that unmarked anticausatives such as (3b) are similar to English John washed which is arguably semantically reflexive (John washed himself) even though no reflexivizing morphology is present. This idea is not as innocent as suggested, however. While English John washed receives a reflexive interpretation, this option seems to crucially depend on the fact that English lacks a SE-reflexive morpheme. All other Indo-European languages that have a SE-reflexive morpheme lack zero-derived reflexive verbs of the English type with human subjects (see Alexiadou et al. 2014). That is, in all these languages, strings similar to English John washed, John dressed or John and Mary kissed can only get an interpretation involving object drop (John washed something) but not a reflexive or reciprocal interpretation.¹⁰

Turning back to (12), note that this generalization is strong but not perfect. Basically all languages have individual transitive, causative verbs with an underspecified external argument that do actually not form an anticausative. This is illustrated in (15) with a German example (see Rappaport Hovav 2014 for English verbs of destruction and killing showing the same behavior, see Alexiadou et al. 2015 for further discussion and further German examples).

In the worst case, every theory has to stipulate something (e.g. in a lexical entry) to capture the ungrammaticality of (15b) and similar cases. The problem for the RAoAC is that it would have to stipulate that reflexivization is blocked. However, this move seems too strong as reflexivization is arguably not blocked with such a verb whether the external argument is a human or a non-human entity (cf. 16a, b):

Since reflexivization of transitive predicates cannot be blocked in any reasonable way, the proposal makes the even worse prediction that all unmarked anticausatives, e.g. (17a), should optionally allow SE-marking as in (17d). This is so as these verbs allow non-human subjects (17b) and they allow reflexivization (17c). In order to avoid (17d), the RAoAC would have to stipulate that such verbs reflexivize only if the subject is a human agent.¹¹

Finally, it remains unclear why we find reflexively marked anticausatives only in languages with a SE-reflexive element. English is a case in point. As the example in (18a) shows, the English SELF-reflexive pronoun is not restricted to human (or anthropomorphic) antecedents (pace Beavers & Koontz-Garboden 2013a: 214; see also many examples in Stevens 2006); but then, why can’t English use its SELF-reflexive pronoun to derive anticausatives (18b-e)?

Note that (18b-e) are not formally ungrammatical. Arguably, speakers reject these examples because their semantics are incompatible with the way they conceive the world. For example, (18c) without the reflexive pronoun expresses that a gap came into existence, but with reflexive pronoun, it expresses that a gap caused itself to come into existence. But gaps cannot open themselves for conceptual reasons (cf. fn. 11) because this would amount to a world where a causer causes its own existence (see Doron 2003; Alexiadou et al. 2015 for similar examples). Note next that the RAoAC proposes exactly such fallacious semantics for similar examples in languages with SE-reflexives. In German, for example, the anticausative version of open is obligatorily expressed with the SE-reflexive pronoun (Ein Spalt öffnete sich (‘a gap opened SE’)). To conclude, while it is well known that only SE-reflexives but never SELF-reflexives are used across languages to form marked anticausatives (e.g. Faltz 1985; Kemmer 1993), the RAoAC has no way to integrate this fact, because the semantic outcome of the two reflexivization strategies is identical.¹²

3 Entailment in the causative alternation

Recall that the semantic relation between a SE-marked anticausative and its causative alternate is fundamentally different under the RAoAC and under the standard account. As discussed in detail in Koontz-Garboden (2009), the standard account in (4a, b) predicts a clause headed by a causative verb to entail a clause headed by its anticausative counterpart, while under the RAoAC in (9a, b) no such entailment relation should hold. Koontz-Garboden (2009: 103) sees the prediction of the RAoAC confirmed by examples such as the Spanish one in (19), where a clause headed by a SE-marked anticausative is negated while its transitive counterpart is asserted. Examples corresponding to (19) are acceptable in many, if not all languages.

Negation is a downward entailing operator licensing inferences from supersets to subsets. Informally speaking, if an expression A is entailed by B, while the reverse does not hold, then A is semantically weaker than B. And if a weaker semantic expression is not satisfied (or false), any corresponding stronger expression is thereby not satisfied (or false), too. Since the standard theory of the causative alternation repeated in (20) characterizes the meaning of a causative verb and its anticausative alternate as strong and weak expressions, it predicts that the negation of an anticausative event should necessarily make the corresponding causative event false, too. And the example in (19) seems to prove this to be empirically wrong.

The argument goes through, however, only if the negation in (19) is a truth-functional statement about the proposition expressed by the first clause, i.e. if it is a ‘logical’ (or ‘descriptive’) use of negation. Alternatively, (19) might involve the so-called ‘metalinguistic’ use of negation, which does not negate the truth-value of a proposition but objects to a pragmatic aspect of the utterance (Horn 1985, and many others before and since). The different uses of negation are illustrated in (21a-c) and (22a-c) respectively:

The propositions in the first clauses of (21a-c) are logically negated, and the following clauses express that, in fact, something different holds true. In (22a-c), however, the negation does not apply to the propositions expressed by the first clauses. If John has five children, as asserted in the second clause of (22a), it is logically true that he has four children, and if the soup is hot, as asserted in the second clause of (22b), it is also warm. Instead, here the negation objects to a scalar implicature that puts an upper bound to the scalar expressions used in the first clauses, the numeral four with its scalar implicature ‘not more than four’, the adjective warm with its scalar implicature ‘not more than warm, i.e. not hot’ and the verb hate with its scalar implicature ‘not more than hate, i.e. not loathe’. In general, scalar implicatures may arise if a speaker uses a weak scalar expression instead of a stronger alternative. In such a context, the hearer may compute that the stronger alternative does not hold true, because otherwise the speaker would have used this stronger alternative (Maxim of quantity: “make your contribution as informative as required (for the current purposes of exchange)”; Horn 1985; Grice 1989: 26).¹³/¹⁴ Since the standard analysis in (20a, b) assigns a stronger meaning to the causative verb than to the anticausative counterpart, the causative-anticausative pair makes up a scale under such an account. The negation in (19) might then simply express that the use of the weaker scale member as in the first clause is not appropriate and should, instead be replaced by the stronger scale member, as in the second clause.

We chose the term ‘appropriate’ instead of ‘true’ for the following reason (see e.g. Higginbotham 1997). Arguably, every inchoative change-of-state event has some cause in the actual world. However, a speaker that utters a clause headed by an anticausative either does not want to or cannot identify the cause of this event. Under such a speaker’s perspective, the use of an anticausative is appropriate, while the use of the corresponding causative verb (be it in the active or in the passive Voice) would not be (cf. also Rappaport Hovav 2014). Whether it is appropriate or not is, however, a question of perspective and context; if one disagrees with the speaker in considering the use of the causative verb appropriate, ‘metalinguistic’ negation can be applied to the clause headed by the anticausative verb in order to object to and remove the upper bounding scalar implicature associated with this verb.

4 Is a reflexive reading impossible for SE-marked anticausatives?

We have argued against the reflexivization analysis of SE-marked anticausatives, proposing that they have inchoative semantics along the lines in (5b). But can’t strings such as (4b), at least optionally, have a reflexive construal derived from (5a) via reflexivization/a bound anaphor in object position?

This option cannot be blocked by formal grammar because reflexivization is a productive process formally available for all transitive verbs (cf. already Horvath & Siloni 2011: 218, who point out that reflexive verb formation is productive, at least in Romance languages). Instead, we think that it is blocked by conceptual considerations: such a construal is as odd as the English The glass broke itself (see the discussion in section 2). But in specific contexts, if the nonsensical construal is negated (cf. fn. 11) and/or enforced by an intensifier (e.g. Spanish sí mismo ‘itself’), it becomes available as the licensing of but1 and but2 in (80) shows. Note that (80) has nothing to do with anticausativization, since unmarked anticausatives as in (3b) enter the reflexive construal under such conditions, too, cf. (81).

The judgments in (80) and (81) hold in other languages, too (e.g. in German and Italian).³² Note further that all other diagnostics discussed above such as the licensing of NPIs, the licensing of concessive vs. corrective conjunctions and adverbs or conditionals also lead to the opposite result if the SE-reflexive is combinbed with an intensifier. We exemplify this with one further Spanish sentence pair involving the corrective vs. concessive adverbs ‘rather’ vs. ‘instead’.

Note, finally, that we do not claim that adding an intensifier is the only way how a semantically reflexive construal can emerge with a SE-marked anticausative. Intensifiers are just one way to enforce a semantically reflexive construal. Other options are devices which, indeed, can act on themselves and, thereby, cause a change in one of their properties; with an automatic door, designed to cause itself to go open, or a computer program that can shut down itself, the intensifier is not necessary to receive a semantically reflexive construal.³³ The same holds for human DPs as in (83b).³⁴

5 The licensing of by-itself

In this last section, we address a final prominent argument brought up in favor of the RAoAC and originally due to Chierchia (2004), namely the licensing of by itself and its counterparts in other languages (e.g. Italian da sé, Spanish sí mismo, German von selbst)

Chierchia’s argument runs as follows (see also Koontz-Garboden 2009). In (84), da sé is successfully bound by an agent subject. In (85), the subject is not an agent, and modification with da sé is claimed to be bad. Finally, in passives such as (86) the derived theme subject cannot license da sé.

Chierchia concludes from this set of data that da sé must be bound by a structural subject, and that this subject must have the theta role agent or causer. Furthermore, he proposes that this phrase expresses that its antecedent is the sole causer/only agent of the event expressed by the lexical verb. Crucially, anticausatives are compatible with da sé, cf. (87). Chierchia takes this as a confirmation for the RAoAC: since (87) is good, la porta, the antecedent of da sé, must be a causer (in addition of being a theme).

Why should the antecedent of da sé be the sole causer? This would follow from the observation that a simple anticausative clause can be followed by a clause introducing a (further) causer for the inchoative event, while this is impossible if the anticausative is modified with da sé, as illustrated for English in (88a, b).

A number of authors have argued against the idea that the distribution of da sé and its counterparts in other languages should be taken as an argument in favor of the RAoAC, in particular that it can be used to diagnose the presence of an agent or causer (see Reinhart 2000: 28; Schäfer 2007; Horvath & Siloni 2011; 2013). Recently, Alexiadou et al. (2015) explicitly argued that neither the distribution nor the interpretation of these phrases fit the above characterization given by Chierchia (2004); neither is the antecedent of da sé necessarily an agent or causer, nor is it necessarily the sole causer. Before we turn to some of their examples showing this, let us first consider how the standard theory of the causative alternation could account for the above data.

Let us start with the contrast in (88a, b). As discussed above, under the standard analysis of the causative alternation the two verbal phrases are ordered on a scale. Furthermore, the anticausative verb, being the weaker scalar item, triggers the implicature that the stronger item is not true. Scalar implicatures (and more generally all conversational implicatures) can be computed on top of an utterance, but they are not asserted and, therefore, not explicitly endorsed by the speaker. We have already seen in section 3.5 that sentences beginning with in fact are used to cancel an implicature associated with a scalar expression used in the preceding clause. We suggest then that in fact in (88a) cancels the scalar implicature associated with the anticausative verb in the first clause, while by itself in (88b) reinforces the effect of this implicature as an entailment. And an entailment, in contrast to an implicature, cannot be cancelled (cf. e.g. Sadock 1978; Grice 1989; Winterstein 2013 about canceling and reinforcing implicatures).

Next, how can a standard theory account for the (non-)licensing of by itself in transitive, passive and anticausative clauses, repeated below for English?

As an important remark, note that by itself in English (as well as in Italian or Spanish, but not in German, for example) has two interpretations: one is roughly ‘alone’ (or ‘unaccompanied’ as in Yesterday I went to the cinema by myself). As pointed out by Levin and Rappaport Hovav (1995: 88), this reading is not relevant for the discussion of anticausatives. We follow Alexiadou et al. (2015) who label the second reading of by itself (the one relevant for the discussion of anticausatives) as ‘no particular cause’. As these authors propose, by using anaphoric by itself under this second reading, a speaker asserts that nobody and nothing can be identified that (directly or indirectly) caused the formal antecedent of by itself to participate in the event expressed by the predicate. (89c) expresses then that no one and nothing can be identified by the speaker to have forced or caused John to break the door. (89b) expresses that nothing can be identified as responsible for the breaking of the door. A speaker uttering (89b) indicates that she rejects the possibility that there is any particular causer or agent that brought about the opening of the door. In other words, modifying the anticausative in (89b) with by itself stresses that the use of the anticausative is the strongest statement the speaker is willing to make and that a causative use (passive or active, i.e. involving an external argument) of the same predicate is, in the opinion of the speaker, not justified. Under this view, by itself does not identify a causer but rejects the participation of a causer. Furthermore, by itself asserts what the plain use of an anticausative verb implicates. In (89a), finally, by itself has the same semantic contribution as in (89b), but now a contradiction arises because of the presence of the implicit external argument that denotes exactly what by itself denies, i.e. the passive semantics contain an implicit agent or causer, while by itself denies the presence of such an entity.³⁵

If the use of by itself asserts that “no causer can be identified”, then, arguably, modification of a predicate with by itself is possible only if a causative use of this predicate involving such a causer is a conceptually or contextually motivated option. With anticausatives this precondition is conceptually fulfilled because, by definition, these verbs are “lexically” associated with a stronger scalar item, i.e. the lexical causative alternate. This is so, whether the anticausative predicate is marked or unmarked, and in fact, across languages, both types of causatives license by itself smoothly (cf. Alexiadou et al. 2006; Schäfer 2008; Horvath & Siloni 2013).

With predicates that lack a causative transitive counterpart, by itself is, as expected, more difficult to use. This is the case with verbs that are already transitive such as the verb in (85) above, which is repeated with such a licensing context in (90) (and note that (84) is actually marked in out-of-the-blue contexts). This is also the case with inchoative structures that lack a lexical causative counterpart, such as so-called internally caused (or pure) unaccusative verbs such as blossom or eventive copula constructions, the latter exemplified in (91a, b). However, once it is contextually established that the events expressed by such predicates could, in principle, be caused, then the exclusion of such causation via the use of by itself becomes an option (see for this point also Schäfer 2007; Horvath & Siloni 2013: 220). Crucially, these example (and further examples in Alexiadou et al. 2015) make it clear that the antecedent of by itself can have other thematic roles than agent or causer.

Finally, Alexiadou et al. (2015) show that even in cases where by itself takes an agent or causer as its antecedent, this agent/causer does not have to be the sole agent/causer of the event. This is exemplified with the Italian example below which expresses that no one or nothing can be identified that caused Gianni to participate in the carrying event; he did it of his free will.³⁶

6 Conclusions

We argued on the bases of conceptual and empirical arguments that the reflexivization analysis of anticausatives (whether they are reflexively marked or plain, i.e. unmarked) cannot be upheld, thereby reconfirming a conclusion already defended in Horvath & Siloni (2011; 2013). All these arguments are, on the other hand, compatible with the standard view on the semantics of the causative alternation according to which anticausatives express one-place inchoative events and their lexical causative counterparts introduce a further argument, which causes the anticausative event. In particular, our empirical tests showed that a clause headed by a causative verb entails the clause headed by the corresponding anticausative verb, i.e. that causatives and anticausatives form a scale. This scalar relationship can explain many arguments brought forward in favor of the RAoAC; a clause headed by an anticausative verb can be negated, and a clause headed by the corresponding causative verb can be asserted at the same time, because the negation involved in the former clause necessarily has a ‘metalinguistic’ use, where the upper scalar implicature associated with anticausative verb is negated. Similarly, by itself appears happily with anticausatives, where this phrase expresses as an entailment what is otherwise just implied by an anticausative verb. However, we also stressed that the morphological string of a SE-marked anticausative verb can be forced into a semantically reflexive construal (either by conceptual or by grammatical clues). Our results reinforce the standard view that SE-morphemes used in many Indo-European languages to mark verbal derivations do not always act as an anaphor or reflexivizer.

Competing Interests

Schäfer’s work was partly supported by a DFG grant to the project B6 ‘Underspecification in Voice systems and the syntax-morphology interface’, as part of the Collaborative Research Center 732 Incremental Specification in Context at the University of Stuttgart. Vivanco’s work was supported by a FPU grant from the Ministerio de Educación, Cultura y Deporte of Spain. Both authors declare that they have no competing interests.