In reflexive constructions, two arguments corefer. This makes it hard to decide which argument bears which θ-role, and consequently to assess whether unaccusativity is involved. A new test is proposed using focus alternatives, which overcomes this difficulty and can also be used to assess reflexivity strength.
Passive structures are analyzed as involving a predicate which, when appearing in a simple clause with a surface syntactic subject, takes this subject as an underlying object (and more generally does not take it underlyingly as its syntactically highest thematic argument). Famously, this mismatch between underlying and surface grammatical relations was extended in Perlmutter 1978 and Burzio 1986 to structures involving (superficially) morphologically simple predicates, henceforth called unaccusatives, that (superficially) take a single DP argument: for example, go, arrive, exist (see Levin and Rappaport Hovav 1994 for extensive discussion).
However, as passive structures demonstrate, structures containing predicates with two DP arguments could in principle be unaccusative too.
How do we know that a simple sentence like (1) is not unaccusative in the above sense?
(1) John fired Bill.
(2) Bill fire John → [[fire John]i Bill ti] → Johnj [[fire tj]i Bill ti]
One may be tempted to argue that such a sentence fails known unaccusativity tests. However, given the (current or possibly principled) absence of a general property identifying all unaccusative predicates, this is not sufficient, as it may merely diagnose that unaccusative predicates do not form a uniform class (known unaccusativity tests already distinguish between at least two classes (see, e.g., Levin and Rappaport Hovav 1994) and could be irrelevant for a third class including superficially simple transitive constructions) or that a general test applying to all unaccusatives has not yet been identified.
For a sentence like (1), however, an unaccusative analysis is implausible on other grounds. For example, John and Bill have different semantic functions ( the DPs John and Bill receive the Agent and Theme roles, respectively), and Agents are underlyingly syntactically higher than Themes (as shown by compound formation (OK employeeTheme-firing, *employerAgent-firing), possible idioms, lack of reconstruction for binding properties, etc.).
This type of consideration underdetermines what happens in reflexive constructions such as the following in English or in French:
(3) John fired himself.
Jean s’est renvoyé.1
Jean SE is fired
‘Jean fired himself.’
In these examples, the Theme argument and the Agent argument have the same referent. While a Theme can indeed be shown to be underlyingly lower than an Agent in a nonreflexive transitive case, it is not obvious which DP gets which θ-role on the surface: the DP Jean could equally well be the Theme or the Agent (or both under certain analyses). Thus, it may well be that an unaccusative derivation akin to the one illustrated in (2) for English can or must apply just in case the Agent is a reflexive form. This option is illustrated for example by Rooryck and Vanden Wyngaerd’s (somewhat more complex) 2011 proposal that attributes an unaccusative analysis to the seemingly simple Dutch object reflexive zich (to which I return in section 2.1).
This is in large part why the question of whether reflexive constructions are unaccusative (e.g., in French or in Dutch) has generated controversy and is not settled. Thus, Reinhart and Siloni (2005) argue that (in addition to Hebrew hitpa’el reflexives) French and more generally Romance reflexive se-si constructions (as in (4)) are not unaccusative ( joining Grimshaw (1982), Wehrli (1986), Chierchia (1989), Reinhart (1996), and Labelle (2008)), while Bouchard (1984), Marantz (1984), Kayne (1989), Grimshaw (1990), Sportiche (1990), Pesetsky (1995), and Rooryck and Vanden Wyngaerd (2011) argue that French reflexive se constructions are unaccusative (and see, e.g., Chief 1998 for Mandarin Chinese). The question has not been raised in English, possibly because a structural analogy is assumed between (1) and (3), though without argument.
In what follows, I will briefly discuss why an unaccusative analysis is a priori attractive for French and why the grounds on which Reinhart and Siloni (2005) reach their conclusion are unconvincing. Next, I will show that this conclusion is nevertheless correct, by introducing a new way to help settle this question, which clearly sides with Reinhart and Siloni’s analysis in showing that reflexive constructions in French with the clitic se are indeed in general not unaccusative. On the way, I will discuss how such a diagnostic could apply to English as well. Finally, I will discuss briefly (a) what this means for the properties that made unaccusativity appealing for French se-reflexives and (b) how this tool can further distinguish among various types of reflexive constructions.
1 Assessing Unaccusativity of Reflexives
Consider the following examples:
Jean se regarde.
Jean SE looks.at
‘Jean looks at himself.’
Jean se croit idiot.
Jean SE believes stupid
‘Jean finds himself stupid.’
In French, and Romance more generally, the presence of se cannot be analyzed as turning a predicate into a reflexive relation by removing a syntactic argument of the verb and reassigning two θ-roles to a single argument (as many, including Reinhart and Siloni (2005), note; they argue that se removes a Case property but crucially does not trigger object-to-subject movement), since the two arguments understood as coreferential need not be coarguments of the same predicate. This is seen in (5b), for example, where one argument is an argument of idiot ‘stupid’ and the other an argument of croire ‘believe’. Following many earlier works (e.g., Kayne 1975) and, more recently, Kayne 2000 and Sportiche 2010, I take it that se can be analyzed as a clitic DP argument of the verb.
If se-reflexives are unaccusative, the surface subject Jean in (5a) is not the highest syntactic argument of regarder ‘look at’. Rather, Jean must be bearing a lower (internal) θ-role.2 It must be the case, then, that se is the highest argument of the verb. In other words, the derivation of this sentence must proceed as follows, where se binds e (as a trace), and Jean has moved from the position of t (see Sportiche 1990, for further justifications):
(6) Jeanj sek [ek regarde tj] (where j=k)
There is prima facie excellent independent evidence for such a derivation, having to do with regularities found in the French (and Romance) systems that I briefly mention below; however, this evidence ultimately proves insufficient.
1.1 Evidence for the Unaccusativity of Reflexive Structures
Extending the empirical basis beyond reflexives to include other French constructions in which se is involved shows why an unaccusative analysis is plausible. Se is also found in middles (7a) and inchoatives/anticausatives (7b).
On a vendu ces livres. → Ces livres se sont vendus.
one has sold these books.MASC.PL these books.MASC.PLSE are sold.MASC.PL
‘They sold these books.’ → ‘These books sold.’
Le vent a renversé la palissade. → La palissade s’est
the wind has tipped.over the fence.FEM.SG the fence.FEM.SGSE is renversée.
‘The wind tipped over the fence.’ → ‘The fence tipped over.’
These constructions illustrate the following properties of se-constructions:
The addition of se removes the expression of what would otherwise be the highest argument of the thematic complex signaled by the verb, and never any other argument.
The perfect aspectual auxiliary used is always ‘be’, never ‘have’.
The participle obligatorily agrees with the superficial subject of the clause.
Reflexives aside, these se-constructions are uncontroversially unaccusative constructions, as the superficial subject is an internal argument.3 And properties (8b–c) are explained by the fact that these constructions are unaccusative. Indeed, (a) there is no non-unaccusative VP in French using ‘be’ as aspectual auxiliary (apart possibly from reflexives), and (b) in transitive structures, participle agreement in French only occurs with direct objects, and such agreement is obligatory only if the direct object moves to subject position (see Sportiche 1990 for detailed discussion).
If reflexive constructions are unaccusative, we can immediately make sense of the fact that reflexive constructions also
obligatorily use ‘be’ as the perfect auxiliary (like other se-constructions) and
require participle agreement when the direct object is ‘‘reflexivized.’’
1.2 An Insufficient Argument against an Unaccusative Analysis of Reflexive se-Constructions
The difficulty of determining the correct analysis for se-reflexives is illustrated by the unconvincing character of Reinhart and Siloni’s (2005) argument against an unaccusative analysis of se reflexives.
Unaccusative predicates typically display two properties, exemplified in (9), taken as diagnostic of (i.e., as sufficient conditions for) unaccusativity: the possibility of impersonal constructions with inverted subject (9b), and en-placement from this inverted subject (9c)—the latter the main original motivation for postulating an unaccusative analysis.
Like unaccusatives, and as expected, middles and inchoatives/anticausatives also allow impersonal constructions with inverted subject, and en-placement from this inverted subject.
Plusieurs livres se sont vendus.
several books SE are sold.MASC.PL
‘Several books sold.’
Les palissades se sont renversées.
the fences SE are tipped.over.FEM.PL
‘The fences tipped over.’
there SE is sold several books
‘Several books sold.’
Il s’est renversé plein de palissades.
there SE is tipped.over lots of fences
‘Many fences tipped over.’
there SE of.them is sold several
‘Several of them sold.’
Il s’en est renversé plein.
there SE of.them is tipped.over lots
‘Many of them tipped over.’
But unlike unaccusatives, reflexive constructions mildly disallow impersonal constructions with a postverbal subject, and they strongly disallow en-placement from such postverbal positions—a property Reinhart and Siloni (2005) take to be a necessary property of unaccusatives.
The following comparison illustrates this:
Why is there a difference between unaccusatives, middles, and inchoatives/anticausatives, on the one hand, and reflexives, on the other? This seems unexpected on an unaccusative analysis for reflexives. However, a careful assessment of what is expected may make this outcome unsurprising. First, the moved en in (11c) must bind its trace inside the postverbal DP: this means this DP must remain syntactically lower than (i.e., c-commanded by) en. The fact that these constructions are reflexive also needs to be taken into account: se can be interpreted as a variable bound by its antecedent.4 This requires semantic binding of se by the superficial subject, which in turn requires c-command, suggesting that se must be syntactically bound by its antecedent. These two requirements could plausibly yield contradictory demands, and result in ill-formedness, if se is higher than en5 and if the object must be both lower than en and higher than se. The deviance of (11b) shows that the postverbal DP is normally too low to bind se. The fact that it is not as deviant as (11c) suggests that a level of attachment may be marginally available for this binding to take place (as is also the case with inverted subjects of unergatives; see Sportiche 1990)6 but is excluded in (11c) as the postverbal DP must remain low enough so that en can bind its own trace. To discount this argument, it would have to be shown either that the c-command relations do not have to be as described or that se cannot be analyzed as a clitic DP, neither of which is known to be true.
Furthermore, the assumption that inversion and en-placement are necessary properties of unaccusative verbs is too strong. Thus, the verb aller ‘go’ or aller à ‘go to’ has the archetypal semantics of unaccusative verbs; it selects aspectual ‘be’,7 whose participle obligatorily agrees with a preverbal subject and can modify a nominal (12), yet it fails both tests.
tous ceux déjà allés en Italie
all those.MASC.PL already gone.MASC.PL to Italy
*Il est allé plusieurs enfants au musée.
there is gone several children to.the museum
*Il en est allé plusieurs au musée.
there of.them is gone several to.the museum
This indicates the presence of an additional variable that could be responsible for the failure of en-placement (or the relative failure of subject inversion) with reflexives, if it is not due to the contradictory binding requirements invoked above.8
1.3 A New Argument That Reflexives Are Not Unaccusative
I will now present a new diagnostic based on the computation of focus alternatives and show why it argues against the straight unaccusative analysis.9 This means that Reinhart and Siloni (2005) and their predecessors (see, e.g., Grimshaw 1982, Wehrli 1986, Chierchia 1989, Reinhart 1996) reached a correct conclusion. Consider the following sentences:
Seul Pierre se trouve intelligent.
only Pierre SE finds intelligent
Only Pierre finds himself smart.
The focus particle only or seul associates with the superficial subject Pierre. The meaning contribution of the focus particle seul is that no contextually salient alternative to Pierre satisfies the property denoted by the VP. But what is this property? In cases such as (14a), which contain a (syntactically) bound reflexive, this VP can be interpreted in two ways, illustrated by the two distinct ways in which the sentence could be denied.
Non, moi aussi je me trouve intelligent.
no me too I me find smart
‘No, I find myself smart too.’
Non, moi aussi je le trouve intelligent.
no me too I him find smart
‘No, I find him smart too.’
The ordinary meaning of (14a) is unique and always reflexive: Pierre finds himself smart. The focus meaning, however, is ambiguous. There are two ways of computing the alternatives to this reflexive reading:
Pierre is the only one to have a reflexive property.
Pierre is the only one to find Pierre smart.
λx (x finds x smart)
λx (x finds Pierre smart)
These observations about the focus meanings of reflexive constructions are not new. For French, they are reported by, for example, Schlenker (2005:73). The facts about English are reported by Büring (2005:141), for example, as follows:
If reflexives needed to be semantically bound [i.e., interpreted as bound variables], this latter construal [(15b)] should be out for the reflexive case, given that the reflexive in [(15b)] is syntactically, but not semantically, bound. In other words, [(14b)] . . . is predicted to be unambiguous. This, however, does not accord with speakers’ intuitions. While the sloppy reading for [(14b)] is generally preferred, the strict one is clearly judged possible.
Why these two readings are possible is discussed by, among others, Büring (2005). He proposes that although the reflexive must be syntactically bound, it can be treated semantically either as a bound variable or not (much as a pronoun would be treated in Only Pierre shaves his head).10
The strict reading crucially shows that to compute the interpretation of (14a), we can take the reference of the reflexive argument as fixed by its antecedent and constant, and compute alternatives on the antecedent alone. In other words, it is possible to treat the meaning of a sentence like (14a) as if it were expressed by ‘Only Pierre finds Pierre smart’, before computing alternatives on the focused subject. So doing allows denials with the meaning expressed by ‘Someone other than Pierre finds Pierre smart’.
Note that a felicitous denial of some assertion need not have the same form as this assertion (it suffices that they have the same meaning except for the part that is denied). This can be seen with middle se-constructions. Recall that middle se-constructions are unaccusative: the superficial subject of a middle can be the argument of another predicate, and it passes the en-placement test for unaccusativity. Consider the middle se-sentence in (17a).
En Inde, seul le riz se mange avec les doigts.
in India only the rice SE eats with the fingers
‘In India, only rice is eaten with the fingers.’
*Non, en Chine aussi, le riz se mange avec les doigts.
no in China too the rice SE eats with the fingers
‘No, in China too, rice is eaten with the fingers.’
*Non, les Chinois aussi mangent le riz avec les doigts.
no the Chinese too eat the rice with the fingers
‘No, Chinese people too eat rice with the fingers.’
Non, en Inde le pain aussi se mange avec les doigts.
no in India the bread too SE eats with the fingers
‘No, in India, bread too is eaten with the fingers.’
c′. Non, les Indiens mangent aussi le pain avec les doigts.
no the Indians eat too the bread with the fingers
‘No, Indian people too eat bread with the fingers.’
French se-middles are interpreted as having an implicit agent that typically can be paraphrased in French by the pronoun on with a generic reading. In (17a), the context makes it clear that the agent is a ‘‘typical Indian person.’’ As expected, the truth of (17a) cannot be denied using (17b) or (17b′) (alternatives on the implicit agent but not on the theme), but it can be denied using (17c) or (17c′), in which the alternatives are computed on the theme alone exactly as expected, given that we are dealing with an unaccusative structure. In particular, we can treat what is asserted as expressed by ‘A typical Indian person eats only rice with the fingers’, which can be denied by asserting either that other things are eaten with the fingers (using another unaccusative middle construction as in (17c)) or that typical Indian people eat other things with the fingers (using a transitive structure as in (17c′)). (And it could also be denied with a passive structure . . . ) That is, there is no need to deny an ‘‘unaccusative assertion’’ with an ‘‘unaccusative denial’’: as stated above, felicitously contradicting an assertion is primarily a matter of content, which underdetermines what form can deliver it.
Superficially, the only difference between such se-middles and reflexive se-middles is the reference of se (generic with these middles, coreferential with the non-se-argument in reflexives). Consequently, just as (17c′) is a felicitous denial of (17a), so we would fully expect (18) to be a denial of (14a) or (14b), were those sentences unaccusative.
What would we expect if (14a) were unaccusative? If it were, it should be possible to treat its meaning as if it were expressed by ‘Pierre finds only Pierre smart’, where, just as in the case of the sloppy reading, the reference of se is fixed by its coreferential argument and constant, and where, just as in middles, alternatives are computed on the (superficial) focused subject (which is the experiencer, ‘‘logical’’ subject of ‘find’). This would predict that (18) is a possible denial.
Non, Pierre me trouve intelligent moi aussi.
no Pierre me finds smart me too
‘No, Pierre finds me smart too.’
Strikingly, and this is the crucial new observation, this denial of (14a) (or (14b)) is not an option. This shows that (14a) (and similar reflexive constructions more generally) cannot be unaccusative.
Does this conclusion apply to English? It should be noted that most speakers (in English, or in French for that matter) find the sloppy reading to be more easily available than the strict one. As a reviewer notes, some English speakers do not seem to allow the strict reading (contrary to what Büring (2005) states).11 Why is unknown.12 In a worst-case scenario, the conclusion that reflexive constructions cannot be unaccusative would hold only for speakers who allow the strict reading. It should be noted, however, that the strict reading made prominent by (15b) is deemed less deviant than the strict reading made prominent by (18). Since the contrast seems to hold for all speakers, this conclusion may still be generally warranted.
Finally, let it be noted that there is no principled bar to having such a reading generated because, say, the reflexive ends up higher than the antecedent to the reflexive. Thus, consider the following case where the superficial subject has raised from the embedded clause:
(19) Only John seemed to himself to be sick.
It is possible for some speakers (although not for all—a necessary condition being that a speaker must accept strict readings with reflexives) to construct a strict reading meaning, as in (20), and to deny it with (21a) (which is, from the present point of view, equivalent to (21b)).
(20) Only John seemed to John to be sick.
No, Mary also seems to John to be sick.
No, it seems to John that Mary is sick too.
1.4 Further Predictions and New Puzzles
This conclusion now raises questions for other se-constructions—in particular, inchoatives/anticausatives (and other se-constructions not mentioned here). Exploring these questions would require first selecting among various analytical options for inchoatives/anticausatives—for example, the widely held analysis of se as an argument suppressor versus Homer and Sportiche’s (2011) alternatives taking these constructions to involve nonagentive causal reflexive or middle causal constructions. Justifying such a choice goes well beyond the scope of these short remarks.
In addition, the conclusion that se-reflexives are not unaccusative raises the question, What do some of the properties discussed earlier, which are classically taken to be sufficient for unaccusativity, actually diagnose? In particular:
Why do reflexive constructions consistently select the aspectual auxiliary ‘be’?13 More generally, what governs the ‘have’/‘be’ alternation in French? This is particularly interesting in view of the (reasonable) claim that in Italian and Dutch, for example, selecting ‘be’ normally entails being unaccusative. The behavior of reflexive constructions probably provides an important clue bearing on auxiliary selection.14
What do French (and Romance) se-constructions share?
Why must reflexive se link to an internal argument?
Why can’t nonreflexive se (e.g., middles and inchoatives/anticausatives) link to (or suppress) an internal argument?
Why is participle agreement obligatory in direct object reflexives?
Some of these questions are addressed in Sportiche 2010.
2 Assessing Reflexivity and Unaccusativity
2.1 Dutch zich and zichzelf and French se
Further illustrating the current relevance of the debate regarding the proper analysis of reflexives, and the usefulness of the focus test, consider Rooryck and Vanden Wyngaerd’s (2011) approach. They also suggest ( pp. 76–77) that Reinhart and Siloni’s (2005) argument is not demonstrative (on different grounds than those discussed here), and they do propose an unaccusative analysis for French se-reflexives, as well as for Dutch zich-reflexives. For sentences (22a–b), they put forth the analysis and derivations in (23a–b) (identical for the two languages except for the fact that se moves), in effect an unaccusative-cum-possessor-raising analysis.
Jean se rase.
Jean SE shaves
‘Jean shaves himself.’
Jan scheert zich.
Jan shaves himself
[rase [se [Z [P Jean]]]] → Jean se [rase + Z + P [
se[ Z[ P Jean]]]]
[scheer [zich [Z [P Jan]]]] → Jan [scheer + Z + P [zich [
Z[ P Jan]]]]
Here, Jean/Jan inalienably possesses se/zich (meaning something like ‘self’), a relation mediated by a ( possession) head Z, and P is a preposition (with both heads incorporating into the verb).
What does the focus test tell us about an analysis such as (23a) for French given that Seul Jean s’est rasé ‘Only Jean shaved himself ’ does allow a strict reading? Notice first that strict readings are not allowed in cases of inalienable possession parallel to (23a). Consider the following, where inalienable possession is signaled by the lack of a visible possessor in the object DP:
Seul Jean a levé le bras.
only Jean has raised the arm
‘Only Jean raised his arm.’
*Non, Pierre a levé le bras de Jean aussi.
no Pierre has raised the arm of Jean too
‘No, Pierre raised Jean’s arm too.’
Just like sentence (22a), sentence (24a) is, or can be, interpreted agentively. If the strict reading were possible, we should be able to take it to mean ‘Only Jean raised an arm Jean inalienably possesses’ and deny this with sentence (24b), meaning ‘No, Pierre too raised an arm Jean inalienably possesses’ (much as Seul Jean se rase ‘Only Jean shaves the self that Jean inalienably possesses’ can be denied by Non, Pierre le rase aussi ‘No, Pierre also shaves the self that Jean inalienably possesses’ (lit., no Pierre him shaves too)). But this is not possible. At a minimum, then, we cannot take se in (22a) to be inalienably possessed.15 But a broader point can be made. Could we maintain an unaccusative analysis with alienable possession? The answer seems negative for the same reason. In a denial such as Non, Pierre aussi rase Jean ‘No, Pierre too shaves Jean’, there is no reason to assume any possession relation (e.g., between Jean and Pierre). How, then, could this be a denial of a proposition asserting (among other things) a possessive relation? This shows that (22a) cannot only be analyzed as unaccusative cum possessor raising.
This is in part compatible with what Rooryck and Vanden Wyngaerd (2011) conclude: namely, that with some verbs, some simple transitive-looking reflexive cases of the (22a) type are structurally and semantically ambiguous.16 For example, the French form in (25a) is ambiguous between the agentive reading in (25b) and the nonagentive one in (25c).
Thus, (25a) could receive two distinct analyses: a simple transitive analysis corresponding to the reading (25b), and, if Rooryck and Vanden Wyngaerd (2011) are right, an unaccusative-cum-possessor-raising analysis corresponding to the reading (25c). Given the argument based on focus alternatives, we expect that the strict reading should be available only for the agentive reading.
And indeed, as expected, the French Seul Jean s’est blessé ‘Only Jean got hurt’ (lit., only Jean SE is hurt) allows a strict reading only under an agentive reading (which, as we have seen, cannot be unaccusative).
Turning back now to (22a) (Jean se rase): this sentence is unambiguous, as it does not display a nonagentive reading. Given the above discussion, it cannot in fact be analyzed as in (23a).17 In other words, French reflexive constructions are in general not unaccusative.
Further, it is not even clear that the unaccusative-cum-possessor-raising analysis is right for the nonagentive case of (25a) with reading (25c). Indeed, the nonagentive meaning of (25a) requires that Jean somehow be ‘‘responsible’’ for what happened to him, as noted by Homer and Sportiche (2011). So the following French examples are infelicitous:
(Context: Jean is dead.)
Jean s’est coupé.
Jean SE is cut
‘Jean got (himself ) cut.’
(Context: Jean was unconscious in the trunk of a car.)
Jean s’est tué en voiture.
Jean SE is killed in car
‘Jean got himself killed in a car.’
In each case, Jean can’t be responsible for what happened to him (as either he is dead, or he is unconscious and cannot influence his fate). To explain this, the structure of (25a) must at the very least be a bit richer than what (23a) provides, and is thus unlikely to be unaccusative.
Turning now to Dutch, this crucial argument against an analysis of (22b) as (23b) does not carry over, as the focus version of (22b) in Dutch given in (27a) does not allow a strict reading.18 Its truth cannot be denied with (27b).
Alleen Jan scheert zich.
only Jan shaves himself
Nee, ook Piet scheert Jan.
no too Piet shaves Jan
‘No, Piet too shaves Jan.’
That zich is deemed to be inalienably possessed makes the analogy with (24) plausible. But note the following sentences with inalienable possession (signaled again by the lack of visible possessor in the object DP):
Seul Jean s’est frappé le bras.
only Jean SE is hit the arm
‘Only Jean hit his arm.’
Non, Pierre lui a frappé le bras aussi.
no Pierre him has hit the arm too
‘No, Pierre hit his ( Jean’s) arm too.’
Here, the denial in the second sentence is fine (at least under an agentive reading), showing (as expected when se is involved) that the strict reading is available. This casts doubt on inalienable possession per se being responsible for the unavailability of the strict reading. Still open, however, is the difference between zich and se.
A clue may come from contrasting zich and zichzelf. Indeed, an equivalent of the above difference between the behavior of zich and the behavior of se is reproduced with pairs involving verbs allowing either a full reflexive (zichzelf) or a nonreflexive pronoun (zich) as direct object, which get the readings indicated: zichzelf allows a strict reading, unlike zich.
This difference correlates with another difference, discussed by Lidz (2001). Lidz notes that zichzelf can be used to refer to a statue of Jan in (29a). Just as we can refer to a statue of Jan by using the name Jan, so the reflexive zichzelf can refer to a statue of its antecedent’s referent, the ‘‘proxy reading.’’ But zich can’t be so used in (29b). This illustrates the fact that the referent of zich must be identical to the referent of its antecedent, not merely able to be designated by the same expression. A ‘‘stronger reflexivity’’ is involved. This stronger identity requirement could be exactly what blocks the strict reading with zich, precisely because it is not possible to paraphrase the meaning of (29b) as ‘Only Jan washed Jan’.19
If this correlation is correct, one prediction that it makes is indeed right: French se, which behaves like Dutch zichzelf, does allow proxy usages: Jean se lave can refer to a situation in which Jean is washing a statue of himself.20
2.2 Beyond French se
Deciding against an unaccusative analysis of French se-reflexives (or possibly English x-self reflexive constructions) is possible precisely because they allow strict readings under the focus alternative diagnostic test. But not all semantically reflexive constructions do so, as seen with zich. This is also true in English. Compare the following two examples (for the speakers who allow strict readings in the first one):
(30b)—call it a silent-object reflexive typically allowed with ‘‘grooming’’ verbs—only allows a sloppy reading. This means the reflexivity of (30b) cannot be ignored to get a strict reading under focus alternatives. It behaves like Dutch zich-constructions: it may well be unaccusative or allow an unaccusative derivation. Now (30b) (or (29b)) does not allow a denial of the sort No, John washed me too, which may prima facie argue against the possibility of an unaccusative derivation. However, this could be because strict readings are simply not available for this type of reflexive construction, owing to the stronger kind of identity required with the antecedent.
As a last note, Reinhart and Reuland (1993) lexically mark the verb in (29b) as [+ reflexive]. Reinhart and Siloni (2005) take Hebrew hitpa’el reflexives to also be lexically reflexive, and they indeed behave like the corresponding Dutch example (29b).21
Thus, while DPs such as himself or zichzelf allow weaker reflexivity, Dutch zich, hitpa’el morphology, or silent objects as in (30b) do not. It is to be hoped that the nature of the morphology signaling reflexivity and its syntactic position is relevant, but nothing simple emerges from a quick crosslinguistic survey. Thus, French se allows weaker reflexive readings (even though it is underspecified; see Kayne 2000). Mandarin Chinese allows reflexive readings with a bare reflexive DP (ziji) or with a preverbal particle zi reminiscent of French se; neither of these allows a strict reading.22
And neither does ta-ziji, as in (33), even though it comprises both a ‘self ’ particle and a pronoun, on the himself/zichzelf model, and minimally contrasts with bare ziji from a morphological standpoint.
Zhangsan fen le ta-ziji.
Zhangsan burn ASP him-self
‘Zhangsan burned himself.’
The focus alternative diagnosis provides a tool for assessing unaccusativity of reflexive constructions, as well as strength of reflexivity. To explain away the initial puzzle about the properties of the French se-reflexive constructions, one analysis implied that they had to be unaccusative constructions. We now see that they cannot be. This in turn leaves unresolved questions concerning auxiliary choice and participle agreement in French and concerning linguistic variation in the properties of reflexive constructions.
For their comments and questions, many thanks to Byron Ahn, Adriana Belletti, Isabelle Charnavel, Guglielmo Cinque, Vincent Homer, Hilda Koopman, Luigi Rizzi, Ian Roberts, Viola Schmidt, Benjamin Spector, Tim Stowell, and the audiences at the University of Cambridge, the Ca’Foscari University in Venice, the University of Siena, and the participants at the 2008 Austrian Linguistics Conference meeting, but most of all to Martin Prinzhorn, whose insightful comment made me think about the main idea developed here.
This research has been supported in part by a UCLA Academic Senate Grant and by a Face Foundation PUF grant.
1 The presence of the French morpheme se yields different effects (middle, reflexive, anticausative, etc.), which I think arise from its unique lexical entry interacting with the structure in which it is inserted. I thus gloss it simply as SE.
2 By internal argument, I simply mean an argument that is not the highest of its predicate in the ordinary sense.
3 For exactly the same (well-known) reason that reflexive se cannot be seen as removing a syntactic argument in (5b), middle se cannot be either. Thus, in the middle construction in (i), the superficial subject is an argument of idiot, not of juger.
Ce genre de choses, ça se juge idiot.
this kind of things that SE judges stupid
‘These kinds of things are judged stupid.’
4 As can be seen in ellipsis constructions, for example; see footnote 10.
5 It is actually difficult to show that this is true. If the current treatments related to the Person Case Constraint prohibiting combinations such as *me lui, *te lui, *se lui are on the right track (see, e.g., Anagnostopoulou 2005), me/te (and thus se) must be higher than lui, which in turn is higher than en, as shown in periphrastic V-V causative structures, for example, in which en can appear on the lower V in contexts excluding lui.
Il lui a fait en livrer trois à Pierre.
he him has made of.them deliver three to Pierre
‘He made him deliver three of them to Pierre.’
??Il lui a fait lui en livrer trois.
he him has made to.him of.them deliver three
‘He made him deliver three of them to him.’
6 A reviewer asks why the following sentence is perfect even though the reflexive is bound by a postverbal subject too:
Marie fait se raser Jean.
Marie makes SE shave Jean
‘Marie makes Jean shave himself.’
In causatives like (i), subject VP inversion is due to preposing of a verbal projection (see, e.g., Kayne 1975 and most subsequent literature) and not to low attachment of the subject.
7 Thanks to Nora Boneh for pointing this out to me.
9 Thanks to Martin Prinzhorn, whose remark about German led me to this new diagnosis.
10 That the second option is available may seem surprising at first, as most speakers necessarily treat such reflexives as semantically bound—that is, as bound variables in some simple cases of VP-ellipsis in English, or in French as Schlenker (2005:73) notes. For example:
(i) John shaved himself and Bill did too. (*shave him = John)
However, as documented for example by Hestvik (1995) and Kehler (2002) (see also Büring 2005:138), strict readings are available in such contexts for some speakers, and other cases of strict readings for reflexives are readily available.
(ii) Bill defended himself before John did. (defend him = Bill)
12 A reviewer wonders whether the reflexive in the strict reading cases (for speakers who allow them) is emphatic or necessarily focused. The answer is negative: no focal stress on the reflexive is necessary, and the type of reading available is simply not a reading associated with emphatic or focused reflexives (see Ahn 2010, 2012). Thus (for speakers allowing the strict reading), (14b) can be an answer with only focus on only Pierre to the question Who finds Pierresmart?
13Charnavel’s (2008) work on the surcomposé, showing that se-constructions and unaccusatives actually function differently regarding auxiliary choice, is particularly significant in this respect.
14 One property that reflexives share with unaccusatives is obligatory participle agreement, suggesting that the object moves by A-movement (to subject or, in reflexives, to the se position).
15 See below for further discussion of this point.
16 Thanks to Johan Rooryck and Guido Vanden Wyngaerd for discussion on this point.
17 This fact calls into question Rooryck and Vanden Wyngaerd’s attribution of the deviance of (11c) (repeated here together with their analysis of it) to en’s being extracted from a PP (independently of the fact that it is not clear why this is excluded given that the P is incorporated, thus allowing Jean to move in (23a)).
(i) Il s’en rase plusieurs = se + en [rase + Z + P [se[Z[P [plusieurs en]]]]]
This fact also shows that raser ‘shave’ is different from blesser ‘hurt’, which could be related to their different internal structure: blesser is a causative, raser is not (cf. have a shave).
18 Thanks to Riny Huijbregts, Hilda Koopman, and Jos Telling for help with the Dutch examples.
19 The behavior of zich is sometimes described as being obligatorily de se, which I refrain from doing here because the standard notion of de se readings of a pronoun ( perhaps in need of revision) requires it to be in an attitude context, which is not the case here in any simple sense (but see Reuland and Winter 2009).
20 The variation observed among English speakers regarding the availability of the strict reading may parallel these data if it correlates with the option of using the reflexive x-self in proxy readings, as not all speakers allow such readings equally easily.
21 Thanks to Hagit Borer for help with the Hebrew examples.
22 Thanks to C.-T. James Huang and Y.-H. Audrey Li for help with the Chinese examples.