Anticausatives compete but do not differ in meaning : a French case study

entity is not “presented as undergoing a physical process of [...] thickening up; they are rather presented as being in a state which [...] can be described as [...] more thick”. Apart from the fact that it is not completely clear why metaphorical uses should involve a focus on the result state rather than on the process, we do not think that there is a true generalization here. First, we observe that with other verbs, the metaphorical use enters the unmarked construction, cf. (50), again taken from corpora. Second, in some cases, the zero-AC is marked on the literal use, rather than on the metaphorical one, cf. (51). (50) a. Le progrès ralentit. The progress is getting slower. b. Lorsqu'ils sont entourés de béni oui-oui, leur esprit ramollit. When they are surrounded by yes-men, their spirit gets softer c. A la fin l'argent a tari et le commerce a fini. At the end money dried up and business stopped. (51) a. ?Mon livre épaissit. My book is thickening. b. ?Mes fichiers épaississent. My files are thickening.


Introduction
In French as in many other Romance and Germanic languages, verbs undergoing the causative/anticausative alternation divide into two morphological and three distributional classes. With verbs of class A, the anticausative (AC) is morphologically unmarked (∅-ACs), cf. (1). With verbs of class B, the AC is marked with the reflexive clitic se (se-ACs), cf.
(2). ACs of class C allow both markings (∅/se-ACs) allow both markings, cf. (3). layer on top of vP, a middle or expletive Voice (Doron 2003, Alexiadou et al. 2006, Schäfer 2008. The presence of this expletive Voice projection triggers (morpho-)syntactic differences (e.g. auxiliary selection) but does not add any semantics. Semantically reflexive verbs and reflexively marked anticausatives, on the other hand, differ syntactically and semantically, but this difference is not visible at the surface. The relevant (underlying)  D P NOM se-verb The paper is divided as follows. In section 2, we discuss the arguments presented in favour of Claim 1, show the empirical difficulties raised by these arguments and offer an alternative competition-driven account of the remaining data. In section 3, we present the arguments offered in favour of Claim 2, and show that they do not support the hypothesis of a systematic meaning difference between marked and unmarked anticausatives either.

Alleged meaning difference 1: external vs. internal causation
The distinction between internal and external causation was established by Levin & Rappaport Hovav (1995) (who build on Smith 1970) to answer the question of when an intransitive verb has a transitive, causative counterpart. Externally caused VPs are defined as describing eventualities that are under control of some external cause that brings such an eventuality about (break, open). Internally caused VPs describe eventualities where some property inherent to the theme argument is 'responsible' for bringing about the eventuality (run, glow, blossom). The only test offered for internal vs. external causation is the (non-)existence of a causative counterpart, see. (4)-(5).
The door opened. (externally caused) b.
John opened the door.
a. The flower blossomed. (internally caused) b.
*The gardener/*The sun blossomed the flower.
A number of authors have argued that when a French verb is attested in both constructions (class C), the change of state (COS) is presented as internally caused when expressed by ∅-ACs and as externally caused when expressed by se-ACs (Claim 1, cf. Rothemberg 1974, Bernard 1971, Burston 1979 a. Il vit le mouchoir *(se) rougir.
(internally caused) Jeanne (SE) reddened/blushed. This claim raises the following problems. First, if the presence/absence of morphology has semantic effects in class C, the relevant semantic properties should also hold in class A and class B, respectively. This prediction holds at least in frameworks where the different morphology is directly correlated with different syntactic structures, which, in turn, are correlated with different event semantics, e.g. Labelle 1992, Labelle & Doron 2010, Doron & Labelle 2011 (it does not hold in the system proposed by Legendre & Smolensky 2009). However, as acknowledged by Rothemberg 1974  A third problem for Claim 1 is that when the COS is conceived as externally caused according to the transitivization test, the ∅-AC of class C verbs is not systematically ungrammatical. For us, (8a) repeated below is perfect without se, and we find examples such as in (16) in corpora (again also accepted by our informants). a.
[Son] cou a rougi au soleil. Her neck ∅ reddened from the sun. b.
La peinture a noirci sous l'effet de la fumée des bougies. The painting ∅ blackened under the effect of the candles' smoke.
A fourth problem for Claim 1 is that it contradicts the intuition conveyed by traditional grammars. Indeed, se-ACs are often said there to underline some responsibility of the subject for the COS to evolve (authors cited by Zribi-Hertz 1987:24). For instance, Vendryes 1948 claims that the reflexive clitic marks "la participation du sujet au procès" [the participation of the subject to the process], Grevisse 1969 suggests that se "met en relief l'activité personnelle du sujet", marks "un intérêt particulier de ce sujet dans l'action" [it focuses on the personal activity of the subject, marks a particular interest of this subject in this action], and Gougenheim assumes that se indicates that the subject "a contribué pour une part si minime soit-elle à l'action subie" [contributed even for a very minimal part to the endured action]. We backed up this intuition with a small experiment: 8 native speakers were asked to pick among the se-AC and ∅-AC of two verbs of class C the one which attributes more responsibility/agentivity to the (inanimate) subject. They all chose the se-AC in (17) and (18): The metal SE rusted.
Note that these verbs can, in principle, be used transitively with a causer subject (so they are really anticausatives, not unergatives or pure unaccusatives), cf. (19).
Le gel a flétri les rameaux. b. L'humidité a rouillé le métal. The frost faded the branches.
*Après l'extraction du nerf, les dents se noircissent. (good for us) After the extraction of the nerve, the teeth blacken. c.
(good for us)

The walls near the chimney are becoming black.
Other examples where Labelle 1992 identifies a meaning difference involve a contrast between bare verbs and a prefixed counterpart, cf. (21)-(22). We think that the slight meaning differences in these examples result from meaning differences between grandir/agrandir (the latter restricted to spatial/physical contexts), not from the class A vs. class B membership of the two verbs.
The companies are getting bigger.
La plage d'ombre s'est agrandie. The shadow spot got bigger.
Our interim conclusions are as follows. First, the claim that se-ACs are externally caused while ∅-ACs are internally caused cannot be upheld as it meets numerous counterexamples, cf. also Zribi-Hertz 1987. Second, anticausatives of class C (optional verbs) have to be realized as ∅-ACs if their DP is +human. However, this holds even if the verb expresses an externally caused event. Third, for verbs of Class C, speakers ascribe more responsibility for the COS to the DP in the se-variant than in the ∅-variant.

An alternative explanation: A pragmatic account
As just emphasized, there remain two robust differences between se-ACs and ∅-ACs, but they emerge only with ACs of class C: (i) Human DPs are out in se-ACs (difference A); (ii) If forced to choose the structure that ascribes more responsibility to the DP, speakers prefer the se-variant over the ∅-variant (difference B). We account for these both differences through a competition-driven pragmatic explanation.

Explaining difference A
Our account of difference A rests on three observations. The first one is that reflexively marked strings are ambiguous between a number of semantic argument structures. They can correspond to (i) a semantically reflexive verb, (ii) an anticausative and (iii) a reflexive passive or generic middle. Concentrating on the first two readings, a reflexively marked string could origin either from structure III or structure IV below: In sum, the structures III and IV differ syntactically and semantically, but do not differ differ in their morpho-phonology (i.e. surface string). The second relevant observation is that for some interpretations, the morphological form is fixed by the grammar (i.e. lexicon or syntax), whereas for others there is optionality: Reflexive semantics have to be expressed with the syntax in IV. b) Anticausative verbs of class A are lexically restricted to enter structure II. c) Anticausative verbs of class B are lexically restricted to enter structure III. d) Anticausatives of class C are compatible with both structure II and structure III.
With verbs of class A, an ambiguity between a reflexive and an anticausative interpretation will never arise. On the other hand, anticausative verbs of class B are at the surface ambiguous as their morphophonology could also convey the reflexive structure in IV. So here, grammar does not make disambiguation available. Finally, for anticausative verbs of class C, Structure II does not lead to ambiguity (only anticausative), while Structure III leads to an ambiguity of the surface string (anticausative or reflexive). In that latter case, grammar makes disambiguation available.
The third relevant observation is that semantically reflexive construals are overwhelmingly found with human agents acting on themselves but not with non-human causers acting on themselves. 1 Anticausatives, on the other hand, typically express a change of state and are overwhelmingly found with non-human DPs. As a consequence, Structure III leads to a salient ambiguity or alternative parse only if the DP is +human. Alltogether, these three observations allow us to account for difference A with the help of the Gricean Maxim Avoid ambiguity (if possible)! (Grice 1989). The reasoning goes as follows: if the full DP has agent properties (+human), the [∅-ACs] is preferred over the [se-ACs] to avoid that the hearer arrives wrongly at a reflexive interpretation. If we do not find a similar effect with verbs of class B, it is because those lexically enforce the presence of se: the speaker has therefore no choice between several structures. Note that related competition effects have been observed to hold between semantically reflexive construals and se-passives. For instance, Zribi-Hertz 1982:361 observes that the se-passive is acceptable with an animate DP only if the verb involved is not a so-called 'naturally reflexive verb'. Obviously, the logic behind these facts is of the same kind here, namely to avoid that the hearer arrives at a pragmatically favored but wrong interpretation.

Explaining difference B
Remember difference B between marked and unmarked anticausatives: if forced to choose the structure that ascribes more responsibility to the DP, speakers prefer the se-variant over the ∅-variant. We can easily explain it as follows: (i) Responsible arguments are agents or causers. (ii) Agents/Causers map to the external argument position. (iii) se-ACs are formally ambiguous between an anticausative and a reflexive structure. (iv) ∅-ACs are not ambiguous (v) only reflexive structures involve an external argument position which could host a responsible DP (agent/causers) That is, only the se-ACs string is compatible with an alternative derivation where the DP could be an external argument, i.e. responsible.

Further arguments for claim 1
In section 2.3, we exposed a first set of arguments in favour of Claim 1, showed the empirical problems they raise and provided an alternative pragmatic account of the remaining facts. The following subsections deal with additional arguments in favour of Claim 1 (external vs. internal causation) provided by Labelle 1992 and DL, and show that they are not decisive either.

Mettre à
As a further argument for Claim 1, Labelle 1992 and DL present the fact that se-ACs (both from class B and C) are excluded from the construction mettre x à P, cf. (24), an observation originally due to Zribi-Hertz 1987.
He has put the sugar to (SE) caramelize.
According to DL, this construction describes "the fact of creating the appropriate conditions for an autonomous [i.e. internally caused] process to take place". Given Claim 1, se-ACs are predicted to be excluded from the structure, because they denote externally caused (i.e. non-autonomous) events. However, this argumentation misses several empirical facts, which make the above semantic characterization of mettre à unfeasible. First, transitive verbs are sometimes embeddable under mettre à, cf. (25a) and (26). The COS expressed by these transitive verbs is not conceivable as autonomous: laver does not form an anticausative, cf. (25b); raccourcir does, but the COS in (26a) is clearly conceived as external caused given the context. The acceptability of (25a) and (26a) therefore raise a first difficulty for the hypothesis that mettre à requires the embedded verb to express an autonomous COS. Le fichier *(se) convertit.

The file (SE) is converting.
In conclusion, while DL are right to claim that there are clearly lexical-semantic restrictions on the verbs that can be embedded under 'mettre à', these are neither captured by the distinction between internal vs. external causation nor by the morphology an AC takes.

The distribution of de-PPs
The second additional argument that Labelle 1992 a. Le ballon gonfle (*de gaz carbonique). The ball ∅ is inflating (with carbon dioxide). b.
Labelle sees the paradigm illustrated in (29)-(30) as a further confirmation of Claim 1. Her reasoning can roughly be reconstructed as follows: (i) ∅-ACs denote autonomous processes (Claim 1) (ii) Given Claim 1, the subject of ∅-ACs must be conceived as the main Actor (the Responsible entity) of the event e. (iii) Distinction between Causing entity and Responsible entity: the de-PP is acceptable in ∅-ACs as soon as the entity it denotes is a mere Cause of e, but does not have the responsibility in e. (iv) the de-PP expresses a mere Cause in (30), but the responsible entity in (29a), hence the contrast.
This line of argumentation raises two difficulties. First, Labelle offers no clear independent ground for the distinction between causing and responsible entities. Second, the ∅-form of gonfler intransitive is acceptable with avec-PPs, whose argument arguably plays the very same role in the event as the problematic de-PP in (29a), cf. (31)-(33), all taken from the internet. So whatever is the constraint explaining the ban on de-PPs with some ∅-ACs like gonfler in (29a), it cannot be due to the fact that the denoted entity is conceived as responsible of the event, since the licensed avec-PP has arguably the same thematic role. Note that enfler, another translation of inflate, does not take a de-PP, even with the reflexive morphology, cf. (34). This suggests that the constraint on the de-PP is not purely of a conceptual nature.

The balloon (SE) inflated with carbon dioxide.
We are unsure whether the (non-)availability of de-phrases is totally idiosyncratic (verb specific) or whether there are generalizations to be made. In any case, the explanation building on the contrast 'responsible entity' vs. 'causer' does not seem to accurately account for the empirical picture.

Alleged meaning difference 2: Aspectual differences
As mentioned in the introduction, ∅-ACs and se-ACs have also been claimed to show aspectual differences. According to Labelle 1992 and DL, ∅-ACs focus on the process, whereas se-ACs focus on the final state of the COS (what we called claim 2 above). LD explain these alleged aspectual contrasts through a difference in the syntactic structure of each form. Very briefly, in the ∅-AC, the two verbal projections v and V are supposed to be present; v introduces an activity subevent, and V a change leading to a state. The verbal root merges with v, hence the focus on the process. On the other hand, in the se-AC, only V is present and thus no process subevent is introduced. The verbal root therefore has to merge with V, hence the focus on the result state. One immediate problem for an analysis along this line is that there is no indication/empirical argument that the two ACs differ in event complexity (e.g. the number of readings with adverbs like 'again' seems to be identical). In fact, traditional event decomposition tests, like the interpretation of negation or the French construals corresponding to almost and again, force to conclude that the two constructions have the same event decomposition (we illustrate this for class C, but the point holds for class A and class B, too): (35) a. La branche a failli casser.
The branch almost broke.
(i) counterfactual reading (nothing happened to the branch) (ii) scalar reading (something happened to the branch but it was not broken at the end) b.
La branche a failli se casser. The branch almost SE broke.
(i) counterfactual reading (ii) scalar reading (36) a. La soupe a re-refroidi. The soup has again cooled down.
(i) repetitive reading (two cooling down events) (ii) restitutive reading (two states of being cool) b.
La soupe s'est re-refroidie. (i) repetitive reading (ii) restitutive reading Moreover, as we show in detail through sections 3.1 to 3.4 below, the empirical arguments in favour of Claim 2 are not convincing (pace Legendre & Smolensky 2009, who also argue for an aspectual difference between marked and unmarked AC, but only within class C verbs).

The distribution of in and for adverbials
According to Labelle 1992:398, for-phrases are acceptable only with ∅-ACs, because they denote a process and no state, cf. (37a). Since se-ACs focus on the result state, they are more natural with inadverbials, cf. (37b). The cement ∅/SE hardened for 3 hours. b.
According to our judgements however, the empirical claim illustrated in (37a) does not hold. For us (37a) is acceptable with both ACs, and further examples where the reflexive variant of class C verbs combines with for-adverbials can be found in corpora and are accepted by our informants, cf. e.g. (38a-e).  , which SE blackened for twenty years in the window of a second-hand bookseller, encloses in its thin sides three or four pages having escaped the hunting of the most tireless book lovers.
Besides this first difficulty, we note that the alleged contrast in (37a,b) does also not manifest itself in class A or class B. Indeed, we find ∅-ACs of class A licensing in-adverbials as well as se-ACs of class B licensing for-adverbials, cf. (39). According to our intuition however, cuire is not shifted to class C in (40) ---with the reflexive, it strongly tends to be interpreted as a passive. One piece of evidence for this is provided by the interpretation of tout seul 'by itself'. Under the most natural interpretation of (41b), tout seul gets with se cuire the reading it gets with se-passives: For us, (40b) and (40c) are in fact acceptable once the passive interpretation is made salient in the context. We think that if this passive interpretation is more accessible out of the blue with a en-adverbial (cf. (40a) than with a for-adverbial (cf. (40c), it is because the telic interpretation triggered by enadverbials is very easily associated with an intentional interpretation. This in turns makes the interpretation of the implicit argument that goes with the passive reading more salient. As a further argument for our reanalysis of Labelle's data, we observe that causer-PPs are acceptable with the ∅-version but are quite degraded under the se-marking; this follows if se-passives have an agentivity restriction on the implicit argument.
(42) a. L'oeuf a cuit sous l'effet de la chaleur. The egg cooked under the effect of the warmth. b.
??L'oeuf s'est cuit sous l'effet de la chaleur. (cf. also Kupferman 2009) The egg cooked under the effect of the warmth.

The argument of muer
The second set of arguments in favour of Claim 2 offered by Labelle 1992 and DL bears on the verb muer, which, so the claim, forms a se-AC only in presence of a resultative en-PP (43a), while it forms a ∅-AC only without it, cf. (43b). DL argue that since the se-AC focuses on a result state, an explicit description of this state is needed in (43a), and since the ∅-AC focuses on the process, the state is deemphasized. This is supposed to make an explicit state description impossible.
The bird SE changed (in a monster with three heads). b.
L'oiseau a mué (*en un monstre à trois têtes). The bird ∅ moulted (in a monster with three heads). We propose the following alternative analysis for the remaining accurate data (43a). On one hand, muer with en-PP (muer en) is an AC of class C (or alternatively of class A for speakers rejecting (43b) with the en-PP and examples in (44)). On the other hand, muer without en-PP is an unergative verb (cf. also Zribi-Hertz 1987). We see three arguments in favour of the hypothesis that the intransitive muer is lexically ambiguous (between an anticausative and an unergative construal

The argument of the restriction on metaphorical uses
A third kind of arguments in favour of Claim 2 rests on the restriction on the metaphorical uses of anticausatives. Lagane 1967 andRuwet 1972 observed that sometimes, with verbs of class C, the se-AC is blocked when the verb is used in a metaphorical way, cf. (47) Apart from the fact that it is not completely clear why metaphorical uses should involve a focus on the result state rather than on the process, we do not think that there is a true generalization here. First, we observe that with other verbs, the metaphorical use enters the unmarked construction, cf. (50), again taken from corpora. Second, in some cases, the zero-AC is marked on the literal use, rather than on the metaphorical one, cf. (51).
Le progrès ralentit. The progress is getting slower. b.
Lorsqu'ils sont entourés de béni oui-oui, leur esprit ramollit. When they are surrounded by yes-men, their spirit gets softer c.
A la fin l'argent a tari et le commerce a fini. At the end money dried up and business stopped. a.

Conclusions
In this paper, we contested previous analyses according to which ∅-ACs and se-ACs systematically differ in meaning, on the basis of new data, mostly found in corpora. We argued that French ∅-ACs and se-ACs do not show systematic meaning differences wrt focusing on subevents and internal vs. external causation. This supports our claim that the presence vs. absence of se should not be associated with fundamentally different syntactic structures predicting such semantic differences. Finally, we showed that the true remaining meaning differences accurately noticed by Labelle 1992 and DL are either restricted to class C verbs and follow from pragmatic considerations, or are idiosyncratic to individual verbs.