Reciprocal Verbs and Symmetry

56
Reciprocal Verbs and Symmetry Tal Siloni Abstract The paper shows that in addition to periphrastic reciprocal constructions and lexical reciprocal verbs, there is a third type – found in Romance and certain Slavic languages – whose reciprocity is not periphrastic but nonetheless composed only in the course of the syntactic derivation. Examining a sample of ten languages, the study reveals and derives the syntactic and semantic properties of these syntactic reciprocal verbs in comparison with their lexical counterparts. It further formulates the precise mechanisms forming the two types. Among other things, the paper devotes considerable attention to the notion “symmetric verb”, to the so-called “I” reading of embedded reciprocal clauses, and to a particular reciprocal construction that denotes reciprocity between two discontinuous phrases. Keywords: Lexical reciprocal verbs, Syntactic reciprocal verbs, Symmetric verb, "I" reading, Discontinuous construction, Lexicon-syntax parameter. 1. Introduction Reciprocal constructions have been the focus of much recent research, typological and theoretical (e.g., Beck (2001), Frajzyngier and Curl (2000), König and Gast (2008) and references therein). In the literature they are often argued to split into two types: (i) "periphrastic reciprocal constructions" such as They kissed each other, where reciprocity is expressed by means of a reciprocal anaphor (each other), and (ii) reciprocal constructions headed by a "lexical reciprocal verb" such as They kissed, which describes a reciprocal situation without the aid of a reciprocal anaphor (languages, unlike English, tend to mark such verbs morphologically). The paper presents robust empirical evidence for the existence of a third type found in Romance and certain Slavic languages, whose reciprocity is not periphrastic in the sense of (i), but nonetheless composed only in the course of the syntactic derivation, unlike that of lexical reciprocal verbs. Various arguments are provided to show that these "syntactic reciprocal verbs" do not involve a reciprocal anaphor. Further, it is known that periphrastic reciprocal constructions allow a particular additional reading in embedded contexts, the so-called "I" reading (Higginbotham 1980, Heim, Lasnik, May 1991a, 1991b, Williams 1991, Carlson 1998). Syntactic reciprocal verbs, it is shown, just like their lexical counterparts, disallow this construal, which, given the existing accounts of what makes it available, is rather puzzling. The paper offers a solution to this puzzle. The unavailability of the "embedded" reading follows straightforwardly from the fact that both types of reciprocal verbs are derived from their transitive alternate by an operation that eliminates a syntactic position in the complement domain, associating the corresponding θ-role with the subject argument. The empirical array of reciprocal verbs is sorted and assessed in a comparative perspective. The paper uncovers a voluminous cluster of semantic and syntactic distinctions between lexical and syntactic reciprocal verbs in a sample of ten languages. The distinctions all follow from the fact that the two types of verbs are formed in different components of the grammar: the lexicon and the syntax. The exact working of the operation of reciprocalization in the different components is discussed in detail, as well as their corresponding interpretation. It is important to understand that the distinctions between the two types of reciprocal verbs are directly related to their locus of formation. Thus, for instance, it is shown that the reciprocity of lexical reciprocal verbs, unlike their syntactic counterparts (and unlike periphrastic constructions), necessarily involves an atomic

Transcript of Reciprocal Verbs and Symmetry

Reciprocal Verbs and Symmetry Tal Siloni

Abstract The paper shows that in addition to periphrastic reciprocal constructions and lexical reciprocal verbs, there is a third type – found in Romance and certain Slavic languages – whose reciprocity is not periphrastic but nonetheless composed only in the course of the syntactic derivation. Examining a sample of ten languages, the study reveals and derives the syntactic and semantic properties of these syntactic reciprocal verbs in comparison with their lexical counterparts. It further formulates the precise mechanisms forming the two types. Among other things, the paper devotes considerable attention to the notion “symmetric verb”, to the so-called “I” reading of embedded reciprocal clauses, and to a particular reciprocal construction that denotes reciprocity between two discontinuous phrases. Keywords: Lexical reciprocal verbs, Syntactic reciprocal verbs, Symmetric verb, "I" reading, Discontinuous construction, Lexicon-syntax parameter. 1. Introduction Reciprocal constructions have been the focus of much recent research, typological and theoretical (e.g., Beck (2001), Frajzyngier and Curl (2000), König and Gast (2008) and references therein). In the literature they are often argued to split into two types: (i) "periphrastic reciprocal constructions" such as They kissed each other, where reciprocity is expressed by means of a reciprocal anaphor (each other), and (ii) reciprocal constructions headed by a "lexical reciprocal verb" such as They kissed, which describes a reciprocal situation without the aid of a reciprocal anaphor (languages, unlike English, tend to mark such verbs morphologically). The paper presents robust empirical evidence for the existence of a third type found in Romance and certain Slavic languages, whose reciprocity is not periphrastic in the sense of (i), but nonetheless composed only in the course of the syntactic derivation, unlike that of lexical reciprocal verbs. Various arguments are provided to show that these "syntactic reciprocal verbs" do not involve a reciprocal anaphor. Further, it is known that periphrastic reciprocal constructions allow a particular additional reading in embedded contexts, the so-called "I" reading (Higginbotham 1980, Heim, Lasnik, May 1991a, 1991b, Williams 1991, Carlson 1998). Syntactic reciprocal verbs, it is shown, just like their lexical counterparts, disallow this construal, which, given the existing accounts of what makes it available, is rather puzzling. The paper offers a solution to this puzzle. The unavailability of the "embedded" reading follows straightforwardly from the fact that both types of reciprocal verbs are derived from their transitive alternate by an operation that eliminates a syntactic position in the complement domain, associating the corresponding θ-role with the subject argument. The empirical array of reciprocal verbs is sorted and assessed in a comparative perspective. The paper uncovers a voluminous cluster of semantic and syntactic distinctions between lexical and syntactic reciprocal verbs in a sample of ten languages. The distinctions all follow from the fact that the two types of verbs are formed in different components of the grammar: the lexicon and the syntax. The exact working of the operation of reciprocalization in the different components is discussed in detail, as well as their corresponding interpretation. It is important to understand that the distinctions between the two types of reciprocal verbs are directly related to their locus of formation. Thus, for instance, it is shown that the reciprocity of lexical reciprocal verbs, unlike their syntactic counterparts (and unlike periphrastic constructions), necessarily involves an atomic

2

event. Their reciprocity cannot be composed via the accumulation of sub-events, as is the case with reciprocity that is composed in the syntax. The reason for that, I will show, is the fact that plural events are not part of the lexicon's inventory. Particular attention is given to the so-called “discontinuous reciprocal construction”, which expresses reciprocity between two discontinuous constituents: the subject and an oblique phrase, e.g., John corresponded with Mary, unlike the prototypical construction, which involves reciprocity between the members of the subject set (John and Mary corresponded). The construction is studied in detail; after showing that it is possible with lexical reciprocal verbs only, I explain what makes it possible and how it is interpreted. The organization of the paper is as follows. Section 2 discusses the so-called “I” reading of embedded reciprocal sentences. This reading has been argued to be contingent upon the availability of a plurality of sub-events. Indeed, it is available in periphrastic constructions, which, as will be shown, denote reciprocity by the accumulation of sub-events, and impossible with lexical reciprocal verbs whose reciprocity involves a singular event. Count adverbials allow us to easily set these two types of reciprocity apart. It is then argued that verbs whose reciprocity involves a singular event are necessarily “symmetric verbs”. However, certain reciprocal verbs found in Romance and certain Slavic languages do not allow the “I” construal although their reciprocity involves a plurality of sub-events. Section 3 explores the status of these reciprocals, concluding that they are not periphrastic to the extent that they certainly do not involve a reciprocal anaphor. The singular event reading versus the sub-event reading are then shown to follow directly from the component of grammar where the reciprocal verb is formed, lexicon versus syntax, respectively. Section 4 is devoted to formalizing the operation forming reciprocal verbs in the lexicon and in the syntax. Further, it shows how the different reciprocal interpretation that each type of reciprocal verb is associated with is obtained. Once the operations have been defined, the “I” reading puzzle is solved. A cluster of distinctions that follow from the different locus of formation of reciprocal verbs is discussed in section 5. The last two sections deal with properties that do not seem to neatly follow from the different loci of formation. On closer inspection, these differences, too, match the split to lexical and syntactic reciprocal verbs. Section 6 focuses on reciprocal event nominals, and section 7 scrutinizes the so-called “discontinuous reciprocal construction”, and its intriguing properties. The appendix shows that reciprocal verbs whether lexical or syntactic have an external argument subject. 2. A puzzle This section discusses two interpretative distinctions between periphrastic reciprocal constructions and lexical reciprocal verbs, one distinction concerns distributivity and the other, the so-called “I” vs. “we” readings (to be explained directly). These distinctions single out a third type of reciprocal construction found in Romance and certain Slavic languages. As will be shown in detail, this third type, which is formed by a clitic, is reminiscent of periphrastic reciprocal constructions as far as distributivity is concerned. But with regard to the so-call “I” reading, it behaves on a par with lexical reciprocal verbs. Let us start with the “I” vs. “we” readings. 2.1 The “we” and “I” readings A well-known ambiguity first brought to light by Higginbotham (1980) arises in

3

embedded reciprocal sentences of the type in (1). Such sentences have the so-called “we” and “I” readings. Under the “we” reading (i), John and Paul said: “we saw each other”. Under the “I” reading (ii): John said: “I saw Paul” and Paul said: “I saw John” (Heim, Lasnik, May 1991a, 1991b, Williams 1991). (1) John and Paul said that they saw each other. i. John and Paul said that they (i.e., John and Paul) saw each other. ii. John said that he saw Paul and Paul said that he saw John. The two readings are especially conspicuous in the context of (2). (2a) is a contradiction because it can only mean that John and Paul defeated each other in the final. But under embedding as in (2b), the sentence has, in addition to the contradictory reading (the “we” reading), a noncontradictory reading (the “I” reading), according to which John said that he defeated Paul and Paul said that he defeated John in the final. (2) a. # John and Paul defeated each other in the final. b. John and Paul said that they defeated each other in the final. Thus, on the "we" reading (1i) John and Paul report that a situation identical to that described by the simple reciprocal clause: “we saw each other” took place. The pronominal subject they is coreferent with the plural matrix subject John and Paul. Under the “I” reading (1ii), in contrast, the plural matrix subject involves distribution over its members, and the embedded, pronominal subject is bound by the distributive operator ranging over the members of the matrix subject set. Thus, (1ii) means that each of the pair John and Paul said that he saw the other. Roughly speaking, then, the source of ambiguity is the plural matrix subject. Given that, it is very intriguing to discover that the Romance reciprocal constructions formed by the clitic (se/si) are not equally ambiguous. The contradictory (3a) remains a contradiction also under embedding (3b). It does not avail itself of the “I” reading.1 (3) a. #Pierre et Jean se sont vaincus à la finale. (French) Pierre and Jean SE are defeated in the final ‘Pierre and Jean defeated each other in the final’ b. #Pierre et Jean ont dit qu’ils se sont vaincus à la finale. Pierre and Jean have said that they SE are defeated in the final Noting that reciprocal verbs such as collide do not allow the “I” reading either (4b), Williams (1991) suggests that these reciprocals are morphological units that require a

1 In the same vein, (i) is a contradiction. This is so because its sole reading is that Jean and Marie admitted being in love; hence, the addition 'but Jean doesn't know that Marie loves him' is contradictory. Under the "I" reading, it could have meant that Jean admitted that he loved Marie without knowing that Marie loved him too, although Marie herself admitted (separately) that she loved Jean. (i) #Jean et Marie m'ont avoué qu'ils s'aiment, mais Jean ne sait pas que Marie l'aime. Jean and Marie to+me have confessed that they love(rec), but Jean knows NEG that Marie him

loves 'Jean and Marie confessed to me that they are in love, but Jean does not know that Marie loves

him.' Certain verbs taking an infinitival complement seem at first glance to contradict the generalization that se reciprocals disallow the "I" reading. These cases are discussed in section 4.3 where the "I" reading puzzle is solved.

4

semantically plural subject, unlike periphrastic reciprocal constructions. Under the “I” reading, the embedded subject is bound by the distributive operator ranging over the plural matrix subject, and therefore does not qualify as semantically plural.2 (4) a. The truck and the car collided. b. Mary and John said they collided. But why do these verbs require a semantically plural subject? A response is offered by Carlson (1998), who argues that verbs in themselves denote a set of singular events. The ability to express a plurality of events arises from the presence of a plural operator in the sentence distinct from the verb. Reciprocal verbs such as the intransitive kiss (5a), argues Carlson, are group level predicates, on a par with collective predicates such as gather, and therefore require a semantically plural subject. But being group level predicates, they cannot distribute over their plural subject, and the sentence necessarily denotes a singular event of kissing, just like gather. This, according to Carlson, explains why they have only the “we” reading: The “I” reading entails that the embedded subject is bound by the distributive operator ranging over the matrix subject, and accordingly more than one event, e.g., of kissing in (5bii): “the event of John kissing Mary” and “the event of Mary kissing John”. The same is true for collide.3 (5) a. John and Mary said they kissed. ‘John and Mary said that they were involved in an event a mutual kissing.’ b. John and Mary said they kissed each other. i. ‘John and Mary said that they were involved in an event a mutual kissing.’ ii. ‘John said that he kissed Mary and Mary said that she kissed John.’ Let us examine Carlson’s proposal in greater detail, first introducing a technique to count events.

2 In their (1991a) paper, Heim, Lasnik, and May assume that each (and its equivalent in other languages) is a distributive operator contained within the reciprocal expression. They then derive the ambiguity in (1) from the distinct scope each can gain by raising. Attachment of the raised each to the embedded subject generates the “we” reading, while its attachment to the matrix subject gives rise to the “I” reading. The movement approach, as noted by Heim, Lasnik and May (1991a), suggests an explanation to the Romance facts. Heim, Lasnik and May (1991a) assume that reciprocals formed by se/si do not include a detachable distributive operator of the each type, which can take scope over the matrix subject. Hence the “I” reading is not available. But the movement analysis is not tenable, as Heim, Lasnik and May (1991b) themselves acknowledge. Williams (1991) shows that the “we”/”I” ambiguity is not unique to sentences involving reciprocals, but to plurals in general. (i) is a simple illustration of that (for more examples and discussion, see cited references). Discussing the "I" reading of periphrastic reciprocal constructions is beyond the scope of this paper; for discussion, see cited references as well as Dimitriadis (2000). (i) John and Paul think they are sick. (a) John and Paul think that they (John and Paul) are sick. (b) John thinks that John is sick and Paul thinks that Paul is sick. 3 Carlson (1998:48) does not treat collide on a par with the intransitive kiss, arguing that cases such as A and B colliding denote two distinct events. I believe that this is unjustified for the following reasons. First, just like (5a), (4) disallows the "I" reading. Moreover, as will be shown directly (section 2.2), the test of count adverbials confirms that both the intransitive kiss in (5a) and the verb collide in (4) cannot distribute over their pair subject and have only the singular event reading. Section 4.1 further discusses this issue and makes the comparison between reciprocal verbs and group level predicates more precise.

5

2.2 Count adverbials: counting events Let us start with a note on plurals. It is commonly agreed that sentences with a plural subject, such as (6), have two readings: distributive and collective. On the collective reading, John and Bill got together and managed to carry a piano upstairs; there is just one event under discussion, whose Agent is the group John and Bill (an unordered pair). The distributive reading, in contrast, has it that each of the pair John and Bill carried a piano upstairs individually, thus entailing two carrying events, one by John and one by Bill (given the weight of the piano this is the less likely scenario). (For more discussion, see Carlson (1998:42), Landman (2000:167), among others). (6) a. John and Bill carried a piano upstairs. b. John and Bill carried a piano upstairs five times. Siloni (2002) and Dimitriadis (2004) observe that count adverbials, such as x times, can count the number of events the sentence denotes. Hence, they help discerning the two readings. Consider (6b). On the collective reading it involves five carrying events (each of which has the group John and Bill as its Agent). This is so because the collective reading involves one carrying event, which, modified by the adverbial, is multiplied by five. On the distributive reading, modification by five times involves two additional readings. (i) There was a total of five carrying events (each of which had either John or Bill as an Agent). On this reading, the adverb denotes the sum of carrying events performed individually by the members of the subject set. (ii) There was a total of ten carrying events: five by John and five by Bill. On this reading, the adverb modifies (multiplies) the number of carrying events performed individually by each member of the plural subject. Crucially, when the verb can distribute over its subject, the ten event reading is available. When it is unavailable, it means that the verb imposes a collective reading, a singular event reading. Indeed, the reciprocal (intransitive) kiss and collide disallow the ten event reading (7). Modification by the count adverbial five times gives rise to exactly five events of mutual kissing in (7a), and five events of collision involving the car bumping the truck and vice versa in (7b). Although their subject is a pair, these verbs cannot distribute over its members, and only the singular event reading is available. This also explains why the addition of the PP on the forehead to modify the event of kissing between John and Mary would result in an anomalous sentence (7c). A single event of mutual kissing between John and Mary cannot be an event of kissing on the forehead, as observed by an anonymous reviewer. (7) a. John and Mary kissed (five times). b. The car and the truck collided (five times). c. John and Mary kissed (#on the forehead). We can now turn to the Romance clitic reciprocals. The puzzle we started with was why they do not have the “I” reading under embedding. Carlson’s account predicts that just like the intransitive kiss (or collide), they should disallow distribution over the members of their pair subject, imposing a singular event reading. The “I” reading, recall, requires distribution and accordingly a plurality of sub-events (e.g., of kissing in (5b): "John said that he kissed Mary and Mary said that she kissed John"). The count adverbial test immediately shows that this prediction is not borne out: the Romance reciprocals under examination do not impose a singular event reading. (8) can

6

mean either that there were all in all five events of examination (i), or that there were ten events of examination: five where the laryngologist examined the dentist, and five where the dentist examined the laryngologist (ii). Let us call the latter reading the (Reciprocal) Sub-event Reading as it is based on a plurality of sub-events. The fact that (8) allows this reading shows that clitic reciprocals allow distribution over their pair subject. Note that sentences involving se are by and large ambiguous between a reciprocal and a reflexive reading (as will be briefly discussed in section 3.5); unless otherwise specified, we focus in this paper on the reciprocal interpretation. (8) Le laryngologue et le dentiste se sont examinés cinq fois. (French) The laryngologist and the dentist SE are examined five times i. There were five events of examination ii There were ten events of examination: five by the laryngologist, and five by

the dentist. In the same vein, the sentence in (9), which naturally describes five mutual kissing events, can denote ten sequential events of kissing (some by Jean and some by Marie) in the appropriate context, such as the following scenario: Jean and Marie are playing a game; the loser at each turn has to kiss the winner. At the end of the game, we can announce the final score using (9). (9) Jean et Marie se sont embrassés cinq fois. (French) Marie and Jean SE are kissed five times i. There were five events of kissing between Jean and Marie. ii There were ten events of kissing: Marie kissed Jean five times, and Jean

kissed Marie five times. Likewise, modification by sur le front 'on the forehead' triggers the reciprocal sub-event reading (10), as sub-events of kissing can naturally be events of kissing on the forehead, unlike a single event of mutual kissing (see (7c) above). (10) Jean et Marie se sont embrassés sur le front. Jean and Marie SE are kissed on the forehead 'Jean and Marie kissed each other on the forehead' Finally, compare (11) to (3a). While (3a) is a contradiction, (11) has the same noncontradictory status as its English paraphrase; namely, it allows the reciprocal sub-event reading because in the course of the tournament or a month, Jean can defeat Pierre several times, and vice versa. (11) Jean et Pierre se sont vaincus plusieurs fois durant le tournoi/le mois passé. J. and P. SE are defeated several times during the tournament/last month ‘J. and P. defeated each other several times during the tournament/last month’ (French) In sum, despite the fact that the Romance clitic reciprocals allow the reciprocal sub-event reading, they disallow the "I" reading. If so, then we have not solved the puzzle as to why the Romance clitic reciprocals fail to allow the “I” reading. More generally, although the reciprocal sub-event reading is a necessary condition for the “I” reading, as suggested by Carlson, it turns out that it is not a sufficient condition.

7

Our discussion, then, has revealed an intriguing three-way split among reciprocal sentences. (i) Reciprocal sentences involving each other have a reciprocal sub-event reading and allow the “I” reading under embedding. (ii) Reciprocal sentences involving verbs such as collide or kiss disallow the sub-event reading as well as the “I” reading. (iii) The Romance clitic reciprocals allow the sub-event reading, but disallow the “I” reading. In what follows, we will be mainly concerned with classes (ii) and (iii), their properties and formation. This will also allow us to solve the “I’ reading puzzle. Section 2.4 widens our cross-linguistic empirical basis. Prior to this, however, let me throw some more light on class (ii), clarifying what it means for a reciprocal verb to describe a singular event. The notion of symmetry will play an important role here. 2.3 Symmetric verbs Reciprocal sentences involving two participants will usually express a symmetric relation between them. (For ease of exposition, the discussion across the paper concentrates on reciprocal situations involving two participants, putting aside the readings a bigger set brings about, as they are orthogonal to the point at stake.4). Thus, for instance, the reciprocal reading of sentence (12), which involves a reciprocal anaphoric object, is composed of sub-events of writing preformed separately by John and by Mary. As a result, John and Mary are in a symmetric relation, as John wrote to Mary and Mary wrote to John. (12) John and Mary wrote to each other. In light of that, consider the reciprocal kiss and collide again. (13) a. John and Mary kissed. b. The car and the truck collided. As shown by the count adverbial test (section 2.2.), (13a) denotes a singular, atomic event that in itself involves John kissing Mary and vice versa. Thus, the participants in the singular event that the verb describes are also in a symmetric relation just like in (12).5 But although in both (12) and (13a) a symmetric relation holds between John and Mary, there is still an important distinction between the two reciprocal sentences. (12) involves a plurality of sub-events, some of which are events of John writing to Mary and some of Mary writing to John. The sentence is true if for each participant there is some event of writing and some event of being written to. This reading is

4 Nonetheless, for the sake of completeness, let me mention that when the subject set is bigger than two, periphrastic reciprocal sentences (class (i)) are usually associated with weak reciprocity (∀x ∈ D ∃y,z ∈ D (x≠y & x≠z & Rxy & Rzx)) (Langendoen 1978). For extensive discussion, see Kim and Peters (1998), Langendoen (1978, 1992), Langendoen and Magloire (2003) and references therein. Clitic reciprocals (class (iii)) behave on a par. As mentioned, these readings are irrelevant for our purposes. 5 When the subject set is bigger than two, sentences with reciprocal verbs such as kiss (class (ii)) involve symmetrical reciprocity (∀x ∈ D ∃y ∈ D (x≠y & Rxy & Ryx) (Langendoen 1978), not the symmetric relation proper (Rxy is a symmetric relation if and only if for all x and y, if Rxy then Ryx), as noted by an anonymous Reviewer. Thus, in John, Nary, Paul and Sara kissed each member of the subject set has to be engaged in mutual kissing with some other member of the set. Note that The boys kissed may be used by speakers (in some context) even if one individual of the subject set was not engaged in kissing at all. I believe this is so, because in most situations one can speak a little bit loosely (see Lesersohn's (1999) discussion of "pragmatic halos" or "pragmatic slack"), not because the truth-conditions of the sentence allow the exception.

8

impossible in (13a). The reciprocal (intransitive) kiss cannot denote an event of John kissing Mary without that same event also being an event of Mary kissing John (and vice versa). The same is true for collide in (13b). It denotes an event of collision that is inevitably an event of the car bumping the truck and vice versa. Thus, in (12) the symmetric relation is a result of the accumulation of sub-events of writing each of which has either John or Mary as an Agent; the symmetric relation is not associated with the atomic (sub-) events of writing that compose the correspondence. (13a), in contrast, entails a symmetric relation between John and Mary although it has a singular (atomic) event in its denotation. This amounts to saying that in (13a) John and Mary have identical participation in the singular event that the verb denotes. That is so because the reciprocal relation the verb expresses is not based on the accumulation of sub-events but on a singular (atomic) event. I will refer to such verbs as “symmetric verbs”. For now let us content with this informal explanation of what their nature is. In section 4.1, they will be defined by a meaning postulate relating them to their transitive alternates. Importantly, the claim here is not that any instance of kiss and collide constitutes a symmetric verb. The transitive (two-place) verb kiss is, of course, not a symmetric verb, unlike its one-place counterpart. It appears in nonreciprocal sentences, and in the presence of a reciprocal anaphoric object (14), it can (unlike symmetric verbs) have the reciprocal sub-event reading ("John kissed Mary (say, on the hand) and Mary kissed John (on the hand)" in (14)).6 (14) John and Mary kissed each other. The intransitive kiss and its transitive alternate are two (related) lexical entries, as will be shown in section 4. This is not an ad hoc claim. Setting aside English, whose morphology is not rich, all the other languages discussed in the paper mark the two instances by different morphology; in Hebrew, for example, the transitive kiss is nišek and its reciprocal counterpart is hitnašek (more examples are given in section 2.4). The verb collide also has a nonreciprocal occurrence, as in The truck collided with the lamppost (Gleitman 1965, Dowty 1991). This point is discussed in section 7.1; let us put it aside until then. What is crucial for our purposes is the fact that kiss and collide in sentences such as (13) are symmetric verbs, as their reciprocity is necessarily based on a singular event. 2.4 Interim Conclusion and Cross-linguistic Widening Recall that our discussion has revealed a three-way split regarding reciprocal sentences. (i) Sentences involving each other have a reciprocal sub-event reading and allow the “I” reading under embedding. (ii) Sentences involving symmetric verbs such as the reciprocal collide or kiss disallow the sub-event reading as well as the “I” reading. (iii) The Romance clitic reciprocals have the sub-event reading, but disallow the “I” reading; for ease of reference, let us label their behavior a “mixed” behavior. Across languages, complex reciprocal expressions of the each other type consistently fall in class (i). Once we examine the two other classes, the following picture emerges. In the sample of languages I have looked at, reciprocal verbs in Hebrew, Hungarian and Russian belong to class (ii). They are symmetric verbs: they disallow the reciprocal sub- 6 As expected, upon modification by the adverb five times, (14) can denote ten events of kissing: five by John, and five by Mary (unlike 7a), but it can also denote five events of mutual kissing. The latter reading, I believe, is derivable from the former, along lines suggested in section 4.2 for se reciprocals.

9

event reading, as shown by the fact that modification by the adverbial ‘five times’ gives rise to exactly five events (15a-c).7 Further, as expected, modification of the reciprocal 'kiss' by 'on the forehead' results in an anomalous sentence, as illustrated in (15d) with a Hebrew example. This is so because a single event of mutual kissing cannot be an event of kissing one the forehead. Finally, reciprocal verbs in these languages disallow the "I" reading under embedding, as illustrated in (15e) with regard to Hebrew. Unlike their English counterparts, reciprocals in Hebrew, Hungarian, and Russian are easily distinguished from their nonreciprocal counterparts as they are morphologically marked, by a verbal template in Hebrew, the so-called hitpa'el template in (15a, e), and by a verbal suffix in Russian (15b) and Hungarian (15c). (15) a. Dan ve-Rina hitnašku xameš pe’amim. (Hebrew) Dan and-Rina kissed(rec) five times 'There were five symmetric kissing events between Dan and Rina'. b. Masha i Dima pocelovali-s' pjat' raz. (Russian)

Masha and Dima kissed-rec five times 'There were five symmetric kissing events between Masha and Dima.' c. János és Mari öt-ször csókol-óz-t-ak. (Hungarian) János and Mari five-times kissed-rec-past-3Pl 'There were five symmetric kissing events between János and Mari.' d. Dan ve-Rina hitnašku (#al ha-mecax). (Hebrew) Dan and-Rina kissed(rec) (on the-forehead) e. Dan ve-Ron amru še-hem hitnagšu. (Hebrew) Dan and-Ron said that-they collided ‘Dan and Ron said: we collided’ The “mixed” behavior exhibited by Romance clitic reciprocals (class (iii)) is manifested also by Czech, Bulgarian and Serbo-Croatian reciprocals formed by a clitic (I will discuss this morphological correlation: clitic-“mixed” behavior in section 7.3). Thus, for instance, (16a) can mean either that there were all in all five events of examination, or that there were ten events of examination, five where Petr examined Pavel, and five where Pavel examined Petr. Likewise, the reciprocal 'kiss' in Czech, just like its French equivalent, most dominantly describes an event of mutual kissing (five kissing events in (16b)), but can also denote ten events of kissing, five by Petr and five by Pavel if the context enforces it (as the scenario suggested for (9) does). Since the sub-event reading is available, modification by na čelo 'on forehead' is possible (16c); separate sub-events of kissing can be events of kissing one the forehead. The lack of “I” reading is illustrated in (16d), where embedding of a contradictory reciprocal clause does not resolve the contradiction as the “I” reading is not available. (16) a. Petr a Pavel se pětkrát prověřili. (Czech) Petr and Pavel SE five times examined ‘Petr and Pavel examined each other five times’

7 Such sentences have an additional, marginal reading. (15a), for instance, can marginally mean that Dan and Rina, each, were engaged in a series of five mutual kissing events with some implicit individual; hence, ten kissing events are involved. This reading (which is not available with any symmetric verb) is not directly relevant here. It is a manifestation of the discontinuous construction (see section 7 for extensive discussion), with omission of the discontinuous (oblique) argument, which is implicit.

10

b. Petr a Pavel se pětkrát políbili. Petr a Pavel SE five times kissed ‘Petr and Pavel kissed each other five times’ c. Petr a Pavel se políbili na čelo. Petr and Pavel SE kissed on forehead 'Petr and Pavel kissed each other on forehead' d. #Petr a Pavel říkali, že se porazili ve včerejší šachové partii. Petr and Pavel said that SE defeated in yesterday(adj) Chess game Note now that a complex reciprocal expression of the each other type (class (i)) is undoubtedly an anaphoric argument of a transitive verb. In contrast, there is absolutely no reason to believe that class (ii) reciprocal verbs involve a reciprocal anaphoric object. The question arises how the clitic reciprocals should be classified. In order to solve the puzzle they pose regarding the “I” reading, we must first clarify their status. It is not trivial to decide whether they instantiate the reciprocal verb strategy or the reciprocal anaphor one. A priori, the clitic (which I will refer to as se) can be analyzed either as the morphological marking of a reciprocal verb or as a reciprocal object clitic, given that its location and morphology are similar to that of pronominal object clitics (e.g., Jean le voit ‘Jean himcl sees’). It is worth noting immediately that the clitic, just like the relevant verbal morphology in Hebrew, Russian and Hungarian, is also found with other types of predicates: reflexives, unaccusatives, subject-Experiencer verbs, middles, and passives. This morphology has been argued to be a marking typical of valence reducing operations (see Grimshaw 1982, Wherli 1986, Rappaport-Hovav and Levin 1998, Reinhart and Siloni 2005 among others). Clearly, with unaccusatives, subject-Experiencer verbs, passives, and middles, se cannot be analyzed as an object clitic. The next section offers robust evidence that se indeed is not an object clitic. Our focus will be on the reciprocal reading of the se sentences, but the diagnostics used below all give identical results also under the reflexive reading. 3. Against an object clitic analysis of se 3.1 The case of depictive adjectives In languages with morphological case, such as Czech, a depictive adjective modifying the object naturally bears accusative case, in agreement with the case of the modified object (17). (17) a. Petr ho vyfotografoval neučesaného. (Czech)

Petr him(ACC) photographed unkempt(ACC) ‘Petr photographed him unkempt’ b. Petr a Pavel vyfotografovali jeden druhého neučesaného. Petr and Pavel photographed each other unkempt(ACC) Important for our purposes is the fact that a depictive adjective in sentences denoting reciprocity via se must bear nominative case, even when it modifies the Theme role, as in (18).

11

(18) Petr a Pavel se vyfotografovali neučesaní. (Czech) Petr a Pavel SE photographed unkempt(NOM) ‘Petr and Pavel photographed each other unkempt’ Why is that so? A straightforward account is that se is not the object argument of a transitive verb. Rather, the subject is the sole argument of the verb, and hence associated with both θ-roles of the verbal predicate, Agent and Theme; Petr and Pavel, each, are the Agent and Theme of the events. Consequently the depictive adjective necessarily modifies the subject argument and therefore agrees with it in case, even when it refers to the Theme. I’ll resume discussion of this unusual thematic state of affairs in section 4. We now turn to comparative ellipsis constructions, which allow us to diagnose argumenthood. 3.2 Comparative ellipsis constructions The sentences in (19) include a comparative ellipsis construction. The comparative remnant (in brackets) in (19a) can refer to the subject (i) or object (ii), because the matrix verb is transitive and the remnant is not marked for case.8 When the matrix verb is a reciprocal verb, only the subject reading is available (19b), as there is no syntactic object in the sentence. (19) a. Dan ve-Ron nišku exad et ha-šeni [yoter me’ašer yeladim axerim]. Dan and-Ron kissed each other more than children other i. ‘D. and R. kissed each other more than other children kissed each other.’ ii. ‘D. and R. kissed each other more than they kissed the other children.’ b. Dan ve-Ron hitnašku yoter me’ašer yeladim axerim. Dan and-Ron kissed(rec) more than children other ‘Dan and Ron kissed more than other children kissed.’ (Hebrew) According to the test of depictive adjectives, se is not an argument. If this is indeed so, then, we expect languages with accusative case marking to disallow an accusative comparative remnant, as there is no accusative argument in the sentence, which the remnant can refer to. Indeed, Zec (1985) and Dimitriadis (2004) have observed that in Serbo-Croatian the remnant of the comparative ellipsis construction cannot bear accusative case (20a), suggesting that se is not an argument. The same holds for Czech, as illustrated in (20b). The comparative remnant can only refer to the subject and must therefore bear nominative case, as shown in (21). (20) a. * Medicinske sestre su se branili uspešnije nego Lekare. (Serbo-Croatian) nurses(NOM) are SE defended better than doctors(ACC) b. *Petr a Pavel se obviňovali víc než Annu a Janu. (Czech) Peter and Pavel SE accused more than Anna(ACC) and Jana(ACC)

8 In (19a), when the comparative remnant refers to the subject, we get, in addition to the so-called sloppy reading, as in (19ai), also the “strict” reading, as in (i) below. The strict reading requires reference to the object (as is clear from (i)), and hence is never allowed when the object reading is disallowed. As the use of the comparative ellipsis construction here aims to test whether or not the matrix sentence includes an object, it suffices to check whether the object reading is available, abstracting away from the strict reading, which is sometimes blocked for independent reasons. (i) ‘Dan and Ron kissed [each other]i more than the other children kissed themi.’

12

(21) a. Medicinske sestre su se branile uspesnije nego lekari. (Serbo-Croatian) Nurses(NOM) are SE defended better than doctors(NOM) 'The Nurses defended each other better than the doctors' b. Petr a Pavel se obviňovali víc než Anna a Jana. (Czech) Peter and Pavel(NOM) SE accused more than Anna(NOM) and Jana(NOM) ‘P. and P. accused each other more than A. and J. accused each other.’ In languages where nouns are not marked for case (Romance, Bulgarian), only a subject reading is available for the comparative remnant (22), just like in Hebrew (19b). (22) a. Pierre et Nathalie se dessinent mieux que Jean et Marie. (French) Pierre and Nathalie SE draw better than Jean and Marie ‘Pierre and Nathalie draw each other better than Jean and Marie (draw…)’ b. Ivan i Maria se risuvat po-dobre ot Peter i Maya (Bulgarian)

Ivan and Maria SE draw-3Pl better than Peter and Maya ‘Ivan and Maria draw each other better than Peter and Maya (draw…)’ When the remnant is not comparable with the subject owing to its inanimacy, the sentence loses its reciprocal reading as the subject reading is anomalous (the sentence has a middle reading, as is clear from the English translation). (23) Pierre et Jean se dessinent mieux que toute autre chose. (French) Pierre and Jean SE draw better that any other thing 'Pierre and Jean can be drawn better that anything else.' As expected, when the construction involves a reciprocal object, the remnant allows the object reading (24a) and can bear accusative case (24b). (24) a. Pierre et Nathalie dessinent l’un l’autre mieux que Jean et Marie (French) Pierre and Nathalie draw each other better than Jean and Marie i.. ‘P. and N. draw each other better than J. and M. draw each other’ ii. ‘Pierre and Nathalie draw each other better than they draw Jean and Marie’ b. Petr a Pavel namalovali jeden druhého lépe než Evu a Janu. (Czech) Petr and Pavel draw each other better than Eva(ACC) and Jana(ACC) 'Petr and Pavel draw each other better than they draw Eva and Jana' In sum, comparative ellipsis constructions provide strong evidence that se is not an argument of a transitive verb.9 This conclusion can also account for the syntactic

9 As will be discussed in detail in section 4.2, se can also reciprocalize Exceptional Case Marking verbs. In this context too, the remnant can neither bear accusative case (i) nor be associated with an object reading (ii-iii). (i) * Peter a Pavel se viděli tančit lépe než Annu a Janu. (Czech) Peter and Pavel SE saw to+dance better than Anna(ACC) and Jana(ACC) (ii) Jean et Marie se voient danser plus souvent que Paul et Nathalie. (French) Jean and Marie SE see dance more often than Paul and Nathalie. ‘Jean and Marie see each other dance more often than Paul and Nathalie see each other

dance’ (iii) Gianni e Maria si guardano ballare più spesso che Paolo e Natalia. (Italian) Gianni and Maria SE watch dance more often than Paolo and Natalia

13

behavior of ‘se verbs’ in French causative constructions, as explained below. 3.3 French causative constructions As observed by Kayne (1975), French causative constructions treat transitive and intransitive verbs differently. When the verb embedded under the causative verb faire ‘make’ is a transitive verb, its subject must be introduced by the preposition à ('to') (25a). When the lower verb is intransitive, its subject cannot be introduced by à (25b).10 As is clear from (25c), when the direct object of the embedded verb is a pronominal clitic, the verb patterns with transitive verbs as expected. Interestingly, when se is used, the subject surfaces without a preposition (25d), just like the subject of intransitive verbs. This is expected under the present analysis that takes se to be the verbal encoding of reciprocity (reflexivity etc.), and not an object of a transitive verb. Note incidentally that the different positioning of the pronominal clitic and se in the causatives of (25) suggests in itself that they deserve a different syntactic treatment. (25) a. Pierre a fait embrasser Jean à Marie. (French) Pierre has made kiss Jean to Marie 'Pierre made Marie kiss Jean' b. Pierre a fait courir Marie. Pierre has made run Marie 'Pierre made Marie run' c. Pierre l’a fait embrasser à Marie. Pierre him/her has made kiss to Marie 'Pierre made Marie kiss him/her' d. Pierre a fait s’embrasser Jean et Marie. Pierre has made SE kiss Jean and Mary 'Pierre made Jean and Marie kiss' Below I discuss the incompatibility of se with derived subjects. In this respect too, it does not pattern with reciprocal anaphors; this, of course, is not surprising if se is not an anaphor but rather a morphological marking. 3.4 Derived Subjects Se reciprocals are subject oriented, and so are certain anaphors. However, se rejects derived subjects: it can neither appear with passives (26a) nor with raising verbs (26b),

‘Gianni and Maria watch each other dance more often than Paolo and Natalia watch each other dance’

The same is true for reciprocal causative verbs: (iv) Les dames se font danser plus souvent que les filles. (French) The women SE make dance more often than the girls ‘The women make each other dance more often than the girls make each other dance’ Again, if a reciprocal anaphor is used, the object reading is available. (v) Les dames font danser l'une l'autre plus souvent que les filles. (French) The women make each other dance more often than the girls i. ‘The women make each other dance more often than the girls make each other dance’ ii. ‘The women make each other dance more often than they make the girls dance’ 10 The subject of intransitives is an accusative argument. This is clear when it is cliticized, as the accusative clitic is used. (i) Je le ferai courir. I himcl will+make run

14

as already noted by Kayne (1975), and Burzio (1986). This is untypical of subject oriented anaphors, such as the Japanese jibun (27), for instance. For completeness, note that passive and raising verbs are not incompatible with pronominal clitics (28). (26) a. * Gianni e Maria si sono stati affidati. (Italian) Gianni and Maria SI are been entrusted b. * Jean et Marie se semblent intelligents. Jean and Maria SE seem intelligent (27) Taroo-ga jibun-no sensei-ni sikar-are-ta. Taro-NOM self-GEN teacher-by scold-PASS- PAST ‘Taroi was scolded by hisi teacher’ (Japanese: Miyagawa and Babyonyshev 2004) (28) a. Gianni e Maria gli sono stati affidati (Italian) Gianni and Maria to+them(CL) are been entrusted b. Jean et Marie leur semblent intelligents. Jean and Mraia to+them(CL) seem intelligent If se were an anaphoric object, this would be unexpected. As will become clear in section 4.2, this characteristic follows directly from the nature of the operation forming reciprocal verbs. 3.5 Interim conclusion: Lexicon versus syntax Various types of evidence converge to show that se is not an anaphoric object clitic of a transitive verb. Rather, the sentences under examination involve reciprocal verbs. Indeed, on a par with reciprocal verbs, ‘se verbs’ disallow the “I” reading under embedding. Unlike other reciprocal verbs, however, they are not symmetric verbs: they have the sub-event reading, as shown by the count adverbial test (sections 2.2 and 2.3). What could this distinction follow from? What could the reason be that reciprocal verbs in certain languages (Hebrew, Russian etc.) but not in others are symmetric verbs, whose reciprocity is associated with a singular event (and not based on the accumulation of sub-events)? I embrace Carlson's claim that verbs in themselves denote singular events. Assuming a modular approach to grammar, I believe that reciprocal verbs in Hebrew, Russian etc. become reciprocal at a level where verbs have no access to a plural operator. As a plurality of sub-events arises from the effect of a plural operator, plural events do not exist at that level. At which level (component of grammar) do verbs have no access to other constituents such as a plural subject (by virtue of which the denotation can be plural)? Consider our mental lexicon. It includes lists of entries (some of which, as will be discussed, are derivationally related), but it does not involve any syntactic structure; this would be a superfluous duplication of the syntactic component. This is why in the lexicon there is no relation whatsoever between verbs and their actual arguments or modifiers; only syntactic structure puts these together. As plural events arise from the effect of a plural argument (or modifier), they are not part of the lexicon's inventory and cannot underlie lexical operations. Only upon syntactic merging can verbs be associated with a plural element, which, owing to its plurality, can increase the number of sub-events that the sentence denotes. (29) states this generalization. (29) Plural events are not part of the lexicon's inventory.

15

Thus, reciprocal verbs in Hebrew, Russian etc. are symmetric verbs as they become reciprocal in the lexicon, prior to the emergence of sub-events. Given the existence of a morphologically related transitive alternate for lots of them, the simplest hypothesis is that they are derived from the corresponding transitive entry by a reciprocalization operation, and not stored as items independent of their transitive alternates.11 Taking our reasoning a step further, it also follows that those reciprocal verbs that do have the reciprocal sub-event reading ought to be formed in the syntax, where a plural subject can give rise to a plurality of sub-events, which the reciprocal relation can be built of. More explicitly, I argue that reciprocal verbs split into two types. In our sample of languages, we find the symmetric type, which is formed in the lexicon, in Hebrew, Russian, Hungarian, and English, and the other type, the one formed in the syntax, in the Romance family, Serbo-Croatian, Czech, and Bulgarian.12 Both types of languages can, in addition, express reciprocity using a periphrastic construction involving a reciprocal anaphor. This option does not concern the above split and is orthogonal to the discussion to follow. It is immediately noticeable that languages forming reciprocal verbs in the lexicon have a restricted, relatively small set of reciprocal verbs, rather coherent across the various languages. Roughly speaking, the set includes verbs denoting situations of interaction between humans, as illustrated in (30). In languages forming reciprocal verbs in the syntax, in contrast, the operation of reciprocalization is productive. Thus, for example, while French has a reciprocal verb se dessiner 'draw(rec)' (31), Hebrew, Russian, Hungarian, and English do not have such a verb. A reciprocal anaphor must be inserted in the object position to obtain the relevant interpretation, as in the gloss of (31). (30) a. Dan ve-Dina hitxabku. (Hebrew) Dan and-Dina hugged(rec) 'Dan and Dina hugged'

11 For example, in Hebrew we find the following pairs: xibek ‘hugged’, hitxabek ‘hugged(rec)’; laxaš ‘whispered’, hitlaxeš ‘whispered(rec)'; nišek ‘kissed’, hitnašek ‘kissed(rec)’; litef ‘caressed’, hitlatef ‘caressed(rec)’; katav ‘wrote’, hitkatev ‘corresponded’; ra’a ‘saw’, hitra’a ‘met’; mizmez ‘fondled’, hitmazmez ‘fondled(rec)’; gipef 'cuddle', hitgapef 'cuddle(rec)'; simes ‘smsed’, histames ‘smsed(rec). (The citation form of Hebrew verbs is conventionally the past tense form.) 12 More generally, East Slavic languages (Russian, Ukrainian, and Belarusian) form reciprocal verbs in the lexicon, while West Slavic languages (Czech, Slovak, and Polish) and South Slavic languages (Bulgarian, Serbo-Croatian, Slovenian, and Macedonian) form them in the syntax (see Hron 2009 for a survey of Slavic languages). German reciprocals formed with sich seem to constitute another instance of syntactic reciprocal verbs. They pattern with reciprocal verbs on the “I” reading diagnostic. Thus, the reciprocal reading in (i) is contradictory, unlike the one in (ii), where the reciprocal anaphor einander is used. Unlike (ii), (i) does not have the meaning 'Hans said that he defeated Paul in the final and Paul said that he defeated Hans in the final' (Volker Gast, personal communication). This suggests that (i) contains a reciprocal verb, and not a reciprocal anaphor. (i) #Hans und Paul sagten, dass sie sich im Finale besiegt hatten. Hans and Paul said that they sich in the final defeated have (ii) Hans und Paul sagten, dass sie einander im Finale besiegt hatten. The same is argued by Gast and Haas (2008) on different grounds. Reinhart and Siloni (2005) provide evidence that the reflexive sich in local contexts (not the long distance anaphor sich) forms a reflexive verb, and does not have the syntactic status of an argument. I will not discuss German any further here. For more discussion of the status of sich, see cited references as well as Steinbach (1998) and Dimitriadis (2004).

16

b. Dan ve-Dina hitkatvu. Dan and-Dina wrote(rec) 'Dan and Dina corresponded'

(31) Jean et Marie se sont dessinés. (French) Jean and Marie SE are drawn 'Jean and Marie drew each other' In languages with a limited set of reciprocal verbs, the set of reflexive verbs is also restricted. The two sets are distinct; the set of reflexive verbs includes grooming verbs and a few others. In languages where the formation of reciprocal verbs is productive, the formation of reflexive verbs is also productive. Hence, it is not surprising that in the latter languages verbal forms can be ambiguous between the reflexive and reciprocal readings; recall that in our sample of languages reflexive and reciprocal verbs share the same morphology. In addition to the reciprocal reading, (31) also has a reflexive reading ('Jean and Marie drew themselves'). Similarly, (32) has a reflexive (32i) and a reciprocal reading (32ii). Of course, the reciprocal reading of (32) requires an appropriate context (e.g., (32b)), otherwise speakers strongly prefer the reflexive interpretation, on the basis of world knowledge (i.e., people tend to dress themselves rather than each other). The parallel sentence in Hebrew (33) is unambiguous. Its sole reading is reflexive, as the verb hitlabeš 'dressed' belongs to the set of reflexive verbs and not to the set of reciprocal verbs.13 (32) a. Pierre et Jean se sont habillés. (French) Pierre and Jean SE are dressed i Pierre and Jean dressed (themselves). ii Pierre and Jean dressed each other. b. …pour montrer aux enfants comment l’un habille l’autre. …in order to show the children how one dresses the other. (33) Dan ve-Ron hitlabšu. (Hebrew) Dan and-Ron dressed(refl) Why is the set of lexical reciprocal verbs restricted the way it is (or the set of reflexive verbs for that matter)? The exact definition of the set is not yet understood. It may turn out that specific properties of the lexicon determine the set. If the set is to some extent language specific, this ought to be acquired (at the acquisition stage). As traditionally noted, it seems theoretically desirable that such idiosyncrasies fall in the domain of the lexicon, which contains (finite) lists that have to be acquired anyway, whereas the syntactic component is a productive engine that the theory can preserve "uncontaminated" with idiosyncrasies. Crucially, as will be discussed in sections 5-7, alongside the level of productivity, and the type of reciprocal reading that they denote (a sub-event reading or a singular event reading), reciprocal verbs in the two types of languages differ in many other important respects. The cluster of distinctions setting the two types apart follows straightforwardly from the assumption that they are formed in different components of the language faculty. Before discussing the cluster of properties distinguishing between lexical and syntactic reciprocal verbs, let us turn to the way reciprocalization applies. As will 13 It seems that for some speakers, the verb hitlatef 'caressed(rec)' in Hebrew is ambiguous between a reflexive and a reciprocal interpretation; that is, it belongs to both sets.

17

become clear in section 4.3, a basic trait of the reciprocalization operation makes the “I” reading, which we started the discussion with, altogether impossible with reciprocal verbs. 4. The Reciprocalization Operation As discussed above, the subject of reciprocal verbs is understood to be associated with two θ-roles. The subject in (34a) is interpreted as both the Agent and Theme (which is not mapped to the object position) of the corresponding transitive alternate (34b). Just like the subject of a reflexive verb (34c) is associated with both roles of the transitive entry (34d), so is the subject of a reciprocal verb. But while the reflexive verb describes a reflexive relation, the reciprocal verb describes reciprocity. (34) a. John and Mary kissed. b. John kissed Mary. c. John dressed. d. John dressed her. Below I discuss the operation that brings about this thematic state of affairs. I start with lexical reciprocalization. Prior to that, a note on mapping is in order. Given the thematically double nature of the subject, the question arises whether it is mapped to syntax as an internal or external argument. The Appendix offers strong evidence that the subject of reciprocal verbs is an external argument; it systematically fails the tests diagnosing internality in a variety of languages. 4.1 Reciprocalization in the lexicon If the subject of reciprocal verbs is understood to be associated with two roles, the question is: how does the operation produce this effect? As mentioned in section 3.5, in the lexicon there is no relation whatsoever between a predicate and its actual (DP) argument(s), as it is only the syntax that puts them together. So, a lexical operation cannot actually assign roles to arguments. A way to ensure in the lexicon that two roles will end up associated with a single argument is by forming a role composed of the two roles. Such complex role formation is labeled “bundling” by Reinhart and Siloni (2005) with regard to the formation of reflexive verbs. The bundle [θi - θj] retains the thematic properties of [ θi ] and [ θj ]. Lexical reciprocalization can thus be defined as in (35).14, 15 14 Alsina (1996) advances a similar claim regarding Catalan reciprocal verbs, in the LFG framework. Thus, under his analysis, both the Agent and Theme roles are mapped to the subject position at f-structure (the representation of grammatical functions). Alternatively, it has been argued that the internal role is eliminated altogether in the lexicon in order to form a one-place (reciprocal) predicate in the syntax (see Mchombo (1991) for Chichewa reciprocal verbs, and Chierchia (2004) for reflexive verbs). However, since this role exists at the level of interpretation, some meaning postulate that ensures that the remaining role is interpreted not as a simple Agent, but as an Agent and a Theme must be added. But what does it actually mean to say that the output of lexical reciprocalization has to be interpreted this way? In terms of event semantics, if the same argument is both the Agent and the Theme of the event, then the Theme role was not eliminated, although the predicate is monadic. The crucial observation here is that the Theme role, which is not realized syntactically, is nevertheless present in the semantics, so we cannot view lexical reciprocalization as a simple elimination of the Theme. Bundling provides an explicit way to capture this observation. 15 Bundling should not be confused with the mechanism of θ-identification suggested by Higginbotham (1985), which is responsible for adjectival modification. The mechanism of θ-identification identifies

18

(35) Reciprocalization in the lexicon (to be revisited in (36)) Bundling: V(ACC)[ θi ] [ θj ] → V [ θi - θj ] Note that a verb undergoing reciprocalization loses its ability to assign accusative case. Following Reinhart and Siloni (2005), I assume first, that the accusative feature is associated with the base transitive verb in the lexicon, and second, that all valence reducing operations applying in the lexicon reduce accusative case. I resume discussion of this point in section 5.1.

Recall that the set of lexical reflexive verbs and the set of lexical reciprocal verbs are fixed and distinct. Consequently, it must somehow be marked whether the entry formed by bundling is reflexive or reciprocal. How can a lexical operation shape a verb as reciprocal? As discussed in sections 2.2 and 3.5 above, a reciprocal reading is either the result of an accumulation of sub-events or associated with the singular/atomic event itself. As is clear from (29), a lexical operation cannot involve sub-events, as these arise only in the syntax owing to a plural operator. As verbs in themselves range over singular events, if they turn reciprocal prior to syntactic structure, their reciprocity is associated with the atomic event. This is what makes them necessarily symmetric verbs, as explained in section 3.5. Reciprocal bundling, then, should mark the verb as symmetric (SYM is not a feature; it is shorthand for the meaning postulate relating VSYM to V, as will become clear shortly). (36) Reciprocal bundling V(ACC) [ θi ] [ θj ] → VSYM [ θi - θj ] Applying (36) to the verb entry kiss in (37a), we obtain the verb entry in (37b). The new entry has only one complex θ-role, bundle, to assign, and this bundle is assigned to the external argument as in (37c). (37) a. Verb entry: kiss [Ag] [Th] b. Reciprocalization output: kissSYM [Ag-Th] c. Syntactic representation: John and MaryAg-Th kissedSYM. The next question concerns the interpretation of (37c). Its event semantic representation is given in (38). (38) ∃e [kissSYM (e) & [Ag-Th](e, John and Mary) ] The bundled role is assigned to an unordered pair, John and Mary, a group of two individuals. Let us now throw more light on this point. Recall first that sentences such as (39) have two readings: distributive and collective. On the collective reading, the boys got together and managed to carry a piano upstairs; there is just one event under discussion. The distributive reading, in contrast, has it that each boy carried a piano upstairs individually, thus entailing a plurality of carrying events. (39) The boys carried a piano upstairs. the open slot of N(oun) with that of A(djective) under government. It does not form a new complex θ-role and does not apply in the lexicon.

19

Following Carlson (1998:42), Landman (2000:167), among others, I assume that on the distributive reading there is a different Agent for each of the distinct carrying events. But on the group reading, the group itself is the sole Agent of the singular carrying event denoted by the verb. Landman (2000:182) accordingly argues that θ-roles are defined per atomic events. Turning back to (37c), the sentence denotes a singular event whose complex role Agent-Theme is assigned to a group of two individuals. In this respect, reciprocal verbs are reminiscent of group level predicates, but their group argument is an unordered pair.16 The formula then involves a verb denoting a singular event and taking John and Mary as its sole Agent-Theme argument. But there is a further entailment about what the division of labor between John and Mary is with regard to the event. So this is what we need to define next. Notice that the problem we face is two-fold. First, how can we define more accurately the role that each member fulfills in the event if the event is atomic, and its sole role is assigned to the pair? Second, it is well known that a singular event cannot involve two identical roles (Bresnan 1982, Carlson 1998, Parsons 1990, Pesetsky 1995, Williams 1981, among others). Suppose then that the specific role each member fulfills is successfully specified, would we be able to avoid the problem of assigning two identical roles (say Agent) per event? After all, both members of the pair do fulfill the same role. So first, how can an atomic event specify the division of labor between the members of its sole argument? Link (1998: 255-256) in his discussion of event structures raises a somewhat similar problem. Link examines the event of Mozart’s death as referred to in the sentence Mozart died on the 5th of December 1791. The verb die is a verb denoting a durationless change (it is an achievement verb), and therefore treated as an atomic event. There is however a level of description on which Mozart’s death is composed of a more or less complex series of events, e.g., the physical processes involved in dying. The formal problem here is how those intuitive events are to be located in the already atomic event of dying. Link suggests that a certain event can be atomic in a coarser domain and still be the image of a more complex sum of events on a more fine-grained perspective. Putting aside the example of die and the type of "underlying events" it may involve, the idea relevant for us is that atomic events, which linguistic diagnostics identify as indecomposable, can still involve underlying events that are part of the speaker's knowledge, but inaccessible in the syntactic and semantic representations of the relevant sentence. Indeed, Dimitriadis (2008b) suggests utilizing this idea in the realm of reciprocal verbs. Following Dimitriadis' insight, I assume that the singular events denoted by symmetric verbs involve underlying events inacessible to linguistic diagnostics. This information should be part of their lexical meaning. I opt to define this information via a meaning postulate constraining the interpretation of VSYM (the output of lexical reciprocalization) by the interpretation of its corresponding transitive (the input for 16 For the sake of simplicity I limit the discussion to pairs, as mentioned in section 2.2. When the subject set is bigger than two, we have distribution to pairs. Thus, (i) denotes a plurality of events of mutual kissing, each of which involves an unordered pair. (i) The roommates kissed. This is reminiscent of distribution to collections, which can be triggered with collective predicates (ii) (see Landman 1989). (ii) The boys and the girls meet (but not in same room). Usually the pair has to involve individuals and not groups. Thus, The boys and the girls kissed implies kissing events between individuals and not a singular event of kissing between the two groups. The boys and the girls smsed may involve an sms exchange between the group of boys and the group of girls. I put aside here investigation of these issues and related questions.

20

reciprocalization). VSYM will be defined as entailing two underlying events of the same type. As observed by Dimitriadis, this would allow us to define the symmetric distribution of roles between the participants in the event. Let us see how this could work. First, note that meaning postulates place constraints on the relation between lexical items without assuming actual decomposition, that is, without assuming further internal structure (see Chierchia and McConnell-Ginet's (1990) discussion). This can explain why the underlying events are inaccessible. Specifically, this can account for why the sentence ‘John and Mary kissed five times’, for instance, necessarily denotes five kissing events. This is so because the underlying events are invisible to the count adverbial. Moreover, even a lexical reciprocal verb such as hitkatev 'corresponded (write(rec))' (or histames ‘smsed(rec)’), which we know must involve distinct writing events, necessarily refers to exchanges of letters (or SMS messages), and not to the separate writing events. Thus, to the extent that it is possible to modify hitkatev by the adverb ‘five times’ (40), the only available reading is that there were five events of letter exchanging between Dan and Ron (five ‘units’ of corresponding). Although we know that the corresponding eventuality involves separate events of writing performed by the participants in the event separately, as far as the reciprocal verb hitkatev 'write(rec)' is concerned, these are underlying events and therefore invisible for modification. (40) Dan ve-Ron hitkatvu xameš pe'amim. (Hebrew) Dan and-Ron corresponded five times Next, the Agent-Theme role can now be symmetrically distributed with respect to the underlying events between the participants in the atomic event (the members of the unordered pair) (for formalization see note 17). Thus, in our example (37c), the two underlying events represent John kissing Mary and Mary kissing John. They are components of the singular event which they both underlie. The simplex roles Agent and Theme composing the complex role Agent-Theme will be defined at the level of the underlying events. The two Agent roles and the two Theme roles will be defined for different underlying events, and the problem of two identical roles per event will not arise. (38) repeated in (41a) then entails the existence of two "inaccessible" underlying events. This can be formulated utilizing the apparatus proposed by Landman (2000) to map atoms "up" and "down" in the domain of individuals. Specifically, his "down" operation, represented by ↓, maps atoms down, that is, allows group atoms to specify the individual members of the group. If such an operation is applicable in the domain of individuals, it should also be valid in the domain of events (see also Artstein 1997, Rubinstein 2008). Following Lasersohn (1992), Krifka (1992, 1998), Landman (2000), Rothstein (2000), among others, I assume that the domain of events forms a Boolean semilattice, with the sum operation, represented by |_|, which forms a plural event from singular events; each singular event in the sum is part of the plural event. On the distributive reading of say The boys jumped once, the plural event of jumping is the sum of the jumping sub-events that the boys each performed. The down operation

21

allows mapping atomic events to the sum of the events underlying them. (41a) then entails (41b), e1 and e2 being the underlying events.17 (41) a. ∃e [kissSYM (e) & [Ag-Th](e, John and Mary)] b. ∃e∃e1∃e2 [↓(e) = (e1 |_| e2) & [kiss(e1) & Ag(e1, John) & Th(e1, Mary) &

kiss(e2) & Ag(e2, Mary) & Th(e2, John)] Finally, note that reflexive bundling obviously does not mark the verb as symmetric, and its complex role is not assigned to an unordered pair. Lexical reflexive verbs such as dress, shave etc. impose distributivity; they do not allow a collective reading. Consider the examples in (42). Langendoen and Magloire (2003) observe that if (42c) and (42d) are true, then together they imply (42b), where a reflexive anaphor is used, but not (42a). This is so because lexical reflexive verbs impose distributiviy to individual atoms, and therefore disallow the collective reflexive reading necessary to allow the situation described by (42c) and (42d) together. (42) a. Peter and John are shaving. b. Peter and John are shaving themselves. c. Peter and John are shaving John. d. Peter and John are shaving Peter. Bundling, then, in the case of reflexive verbs requires distributivity (to individual atoms). The bundled role is associated with one individual that is both the Agent and Theme of the atomic event. 4.2. Reciprocalization in the syntax What is to become a reciprocal verb in the syntax leaves the lexicon with the same number of θ-roles to be assigned as the basic verbal entry. I assume with others that the clitic se reduces case (e.g., Cinque 1988). Consequently, there is a surplus role, which cannot be mapped in the canonical fashion and must be differently handled by the syntactic engine. Let us examine how syntactic reciprocalization does that.18 I believe that there are good reasons to assume that unlike the lexical operation, syntactic reciprocalization does not involve the formation of a complex, bundled θ-role. For one thing, it is not at all evident that the syntactic component should allow such a manipulation of thematic information. Moreover, if indeed θ-roles are defined 17 The lexical information associated with kissSYM can be defined as in (i). The sum operation maps the sum of singular individuals to a plural individual. Landman's "up" operation represented by ↑ further maps the plural individual to a group atom. (i) KissSYM is a set of singular events for which the role Ag-Th is defined such that for some d1 and d2 (d1≠d2): Ag-Th (e) = ↑ (1 |_| d2). Meaning postulate for kissSYM If e ∈ kissSYM and Ag-Th (e) = ↑ (d1 |_| d2) then for some e1, e2 ∈ kiss: ↓(e) = (e1 |_|e2) and Ag(e1) = d1 and Th(e1) = d2 and Ag(e2) = d2 and Th(e2) = d1 18 The fact that what is to become a reciprocal verb in syntax leaves the lexicon with the same number of θ-roles as the basic entry does not at all contradict the conclusion that reciprocal verbs derived from a two-place entry are syntactically intransitive (section 3). Although they keep the same number of roles as the basic entry, they do not map the internal role onto the object position and are thus syntactically intransitive.

22

per atomic events, then the formation of one complex role would lead to the wrong interpretation: If the bundled role is assigned (distributed) to individuals, then both the Theme and Agent will be associated with the same atomic event, which would necessarily result in a reflexive interpretation. And if the role is assigned to pairs, the pair will be associated with a singular event, and, contrary to facts, the only reciprocal reading the verb could have would be the one associated with a singular event, namely, that of symmetric verbs.19 So what does take care of the surplus role? θ-information carried by the verb must be assigned, as stated by the θ-criterion. As the status of the criterion has been debated, it is important to stress that I believe that this requirement of the criterion is indispensable. However, the criterion also includes a biuniqueness condition requiring that (i) each argument receive only one θ-role and (ii) each θ-role be assigned to only one argument. The latter part (ii) of the biuniqueness condition is obvious: an assigned θ-role is not available for the verb to reassign. The former part is empirically problematic and was primarily meant to rule out movement to θ-positions (Chomsky 1981). As movement to θ-positions can be excluded on different grounds (Chomsky 1995, Bošković 1994 among others), this requirement should be discarded. If so, then given gradual building of structure along minimalist lines (Chomsky 1995), the syntactic mechanism can proceed as follows. The choice of morphology (se) reduces accusative case (or another case as will be discussed in section 5.1). I assume the clitic originates on V and then moves with the verb to T, but nothing hinges on that. An internal θ-role is not mapped onto its canonical position (the case it usually checks was reduced). The unassigned role is retained on the verbal projection until the merger of another argument. It is then assigned to that same argument, resulting in the association of two roles with the same argument. The assignment of the retained role is parasitic on the assignment of another θ-role. It is obviously not the standard procedure of θ-assignment; I suggest it is a last resort mechanism, and consequently applicable only upon assignment of an external role. I resume discussion of the workings of parasitic assignment in sections 5.1 and 5.2. The way reciprocalization applies in the syntax is summarized in (43).20 (43) Reciprocalization in the syntax a. Case reduction (enabled by the appropriate morphology (se)). b. Parasitic assignment: a retained θ-role is assigned upon merger of another

argument (a last resort operation).

19 In Reinhart and Siloni (2005) it is assumed that syntactic reflexivization does involve bundling. Given that, the lexical and syntactic operations could be viewed as identical, and subject to the same constraint, namely, that bundling must apply upon the assignment of an external role. Note that the problem of interpretation in case of syntactic bundling, which was just explained in the text, does not arise regarding reflexive verbs. However, as will become clear in the course of the paper, there is no need to impose “the external role constraint” on bundling. In the lexicon, the set of reciprocal verbs (or reflexive verbs) is anyhow much more limited. In the syntax, the constraint follows straightforwardly, as will be explained in sections 5.1-5.2. Given that and the problem of interpretation that bundling encounters regarding reciprocal verbs, I believe that the syntactic operation does not form a complex θ-role. 20 As will become clear shortly, I believe reflexive verbs are formed by the same operation. Further, if the internal role does merge in the canonical object position, it ends up associated with nominative Case, as the verb lost its ability to check accusative. In that case, I assume, with Chierchia (2004) and Reinhart and Siloni (2005), that the external role cannot merge and ought to undergo existential binding at the level of interpretation, which is what happens in passives, for instance.

23

The operation of syntactic reciprocalization is illustrated in (44) with the French reciprocal se voir ‘see(rec)’. The derivation includes a two-place verb voir ‘see’ with Agent and Theme roles, the clitic se, and the DP Jean et Marie. The selection of se reduces the verb's ability to check accusative case. The Theme role is not mapped onto the object position, but is retained on the verbal projection. Thus, at the VP level, the verb still has two unassigned roles (44b). Upon merger of Jean et Marie, namely, upon the assignment of the Agent role, the Theme role is assigned together with the Agent, as roughly schematized in (44c) (for our purposes, it is immaterial where exactly the external argument is merged). (44) a. Jean et Marie se voient. Jean and Marie SE see 'Jean and Marie see(rec).' b. VP: [se voientAg, Th]] c. TP: [Jean and MarieAg, Th [ se voientj [VP tj]]]

Since the operation is enabled when a θ-role is retained on the verbal projection, the question arises whether there is a maximal domain where the role must be discharged. Or more generally, how is the operation constrained? Section 5.2 discusses these questions in detail. Recall now that verbs whose subject is a derived subject cannot undergo reciprocalization (section 3.4). This ban is a straightforward result of (43b). With both raising and passive verbs ((26) repeated in (45)), there is no merger of a θ-role upon which the retained θ-role (say Target in (45a) and Experiencer in (45b)) can be parasitically assigned. The matrix subject is formed by an internal merger (movement), which is not a merger of a θ-role. (45)a. * Gianni e Maria si sono stati affidati. (Italian) Gianni and Maria SI are been entrusted b. * Jean et Marie se semblent intelligents. (French) Jean and Maria SE seem intelligent I now turn to the interpretation of the syntactic outputs. Recall that the reciprocity denoted by (46a) for instance involves either a singular event reading or a sub-event reading. Let us start with the latter. The event denoted by the verb is the sum (|_|) of two sub-events, e1, e2. Each individual in the subject set is associated with two distinct roles each of a distinct sub-event. Thus, there is a symmetric relation between Jean and Marie owing to the two sub-events of kissing: the kissing of Jean by Marie and the kissing of Marie by Jean. More precisely, there is a plural event of kissing that is the sum (|_|) of the event of Jean kissing Marie (e1) and the event of Marie kissing Jean (e2), as formalized in (46b). (46) a. Jean et Marie s'embrassent. (French) Jean and Marie SE kiss b. ∃e∃e1∃e2 [e= (e1 |_| e2) & kiss (e1) & Ag(e1, Jean) & Th(e1, Marie) & kiss(e2)

& Ag(e2, Marie) & Th(e2, Jean)] Since each individual is assigned two distinct roles, the roles can either be associated with two distinct sub-events, as in (46b) or with the same sub-event, thus resulting in a reflexive interpretation. Indeed, as already mentioned in section 3.5, se sentences are a priori ambiguous between a reflexive and a reciprocal reading.

24

Turning now to the singular event reading, let us represent it using Landman's (2000) apparatus to map atoms "up". Specifically, in the domain of events the "up" operation, which is represented by ↑, allows to map a sum of events to an atomic group event. I suggest the "up" operation maps the denotation of the two event expressions in (47) to yield a singular event. (47) represents the singular event reading, which is made available in the semantics by the "up" operation. (47) is true if the sum of the event of Jean kissing Marie and the event of Marie kissing Jean forms a singular event (an atomic group event). (47) ∃e∃e1∃e2 [e= ↑(e1 |_| e2) & kiss (e1) & Ag(e1, Jean) & Th(e1, Marie) &

kiss(e2) & Ag(e2, Marie) & Th(e2, Jean)] Although the verb s'embrasser is not a symmetric verb, the interpretaion in (47) is identical to that of kissSYM (41b), as desired.21 4.3.Back to the "I" reading We are now in a position to solve the “I” reading puzzle, which was the starting point of our discussion: why do syntactic reciprocal verbs disallow the “I” reading of the embedded clause ((3b) repeated as (48a)) on a par with their lexical counterparts (15e) repeated by (48b), although they allow the sub-event reading required by the "I" reading (see section 2.1 and 2.2). (48) a. #Pierre et Jean ont dit qu’ils se sont vaincus à la finale. (French) Pierre and Jean have said that they SE are defeated in the final b. Dan ve-Ron amru še-hem hitnagšu. (Hebrew) Dan and-Ron said that-they collided ‘Dan and Ron said: we collided’ Consider (49), which does allow the “I” reading (ii) as it does not involve a reciprocal verb but a reciprocal anaphoric object. It could be that John thought that he photographed Paul, and Paul thought that he photographed John, but each was unaware of having been photographed by the other. On the “I” reading, the reciprocal nature of the interaction is not in the awareness of John and Paul. (49) John and Paul thought they photographed each other. i. ‘John and Paul each thought: we photographed each other.’ ii. ‘John thought that he photographed Paul and Paul thought that he photographed John.’ Common to lexical (48b) and syntactic reciprocal verbs (48a) is the fact that their subject is assigned both the Agent and Theme role via bundling or parasitic assignment respectively. This thematic state of affairs is incompatible with the “I” reading. Under the “I” reading, the interaction denoted by the embedded verb is not conceived by the members of the matrix subject as reciprocal. Rather, each of them considers himself to be associated with one role of the embedded predicate, the Agent role exclusively in (49). Reciprocal verbs do not allow this reading, as their subject necessarily involves two roles, Agent and Theme in (48). Sentences denoting reciprocity via a reciprocal 21 Further discussion of the precise working and applicability of the "up" operation in the domain of events is beyond the scope of this paper. Note

25

anaphoric object do not involve this thematic state of affairs. They are transitive verbs whose subject is associated with one role only (Agent in (49)). Hence, the “I” reading is possible. Thus, the “I” reading requires both that the embedded verb allows the sub-event reading, and that its subject be associated with one thematic role. Thus, (49), on its “I” reading denotes two sub-events of photographing for which the members of the subject set are each an Agent exclusively. Both lexical and syntactic reciprocal verbs fail to meet the "sole role requirement", as their subject is associated with two thematic roles; hence, they do not allow the "I" reading. Lexical reciprocal verbs, in addition, do not even have the sub-event reading, as extensively discussed in section 2.2, 2.3 and 4.1. Notice that reciprocal verbs in infinitival complements also disallow the “I” reading, as already observed by Heim, Lasnik and May (1991a), who attribute the observation to Luigi Rizzi (personal communication). This is illustrated in (50). (50) a. #Jean et Paul ont décidé de se vaincre à la finale. (French) Jean and Paul have decided to SE defeat in the final b. #Gianni e Paolo pensano di esser-si sconfitti nella finale. (Italian) Gianni and Paolo think to be- SE defeated in+the final Interestingly, however, it seems that with certain verbs, the “I” reading becomes available with infinitivals. Thus, for instance, the examples in (51) allow the noncontradictory reading according to which each of the pair Jean and Paul (Gianni and Paolo) wants to defeat the other.22 Why is it so? (51) a. Jean et Paul veulent se vaincre à la finale. (French) Jean and Paul want SE defeat in the final b. Gianni e Paolo vogliono sconfigger-si nella finale. (Italian) Gianni and Paolo want to defeat- SE in+the final The class of verbs allowing the “I” reading of the embedded clause seems to be the class of verbs exhibiting transparency effects, namely, the class of the so-called Restructuring verbs, including modals, aspectuals and motion verbs. These verbs allow phenomena that are otherwise clause-bound, such as clitic placement (52), to span over two clauses (see Rizzi 1978, 1982, Cinque 2002). One finds transparency phenomena comparable to clitic climbing (52) in a variety of languages, and with the same set of verbs (or subsets thereof). (52) Gianni e Paolo si vogliono sconfiggere nella finale. (Italian) Gianni and Paolo SE want to defeat in-the final Under Rizzi’s analysis, modal, aspectual and motion verbs can trigger a process of restructuring that turns an original biclausal configuration into a monoclausal one, forming a complex verb out of the matrix and embedded infinitive. On Cinque’s analysis, restructuring verbs are functional verbs in a monoclausal configuration, even in the variant which shows no overt transparency effects. 22 Similarly, although the “we” reading in (i) is dominant, an appropriate scenario can trigger the “I” reading. For instance: The children play chess. At the end of each turn, the loser has to choose either to kiss the other or to give him/her five cents. In this context, (i) would have the “I” reading. (i) Jean et Marie veulent toujours s’embrasser (et non pas se payer). Jean and Marie want always SE to+kiss (and NEG SE pay)

26

If so, then the “I” reading in cases such as (51-52) is just another transparency phenomenon restructuring verbs show. Let me explain why. If the structure of the sentences in (51-52) is monoclausal, then the sentences have the interpretation: "Jean and Paul (or Gianni and Paolo) ‘want-defeat’ each other" and more explicitly, "each of the pair Jean and Paul 'wants-defeat' the other". This interpretation is equivalent to the "I" reading ("Jean wants to defeat Paul and Paul wants to defeat Jean"), but it does not involve embedding; rather 'want-defeat' is a reciprocal complex verbal predicate assigning its Agent and Theme roles to the subject by (43b). The "I" reading, in contrast, is obtained under embedding and accordingly requires that the embedded subject be associated with one role only, as explained above ("John thought that heAgent photographed Paul" and vice versa, in (49ii)) The fact that (51b), where the clitic does not surface attached to the restructuring verb nonetheless has this reading shows that the structure that these verbs project may be monoclausal, even in the absence of overt symptoms, as suggested by Cinque (2002). The next section discusses the cluster of properties that characterize syntactic versus lexical reciprocal verbs. 5. Lexical versus syntactic reciprocalization A coherent cluster of distinctions sets the two types of reciprocal verbs apart, as first observed by Siloni (2001). The distinctions can all be accounted for under the assumption that reciprocal verbs can be formed in different components of the mental grammar: in the lexicon – in Hebrew, Russian, Hungarian, and English, and in the syntax – in Romance, Serbo-Croatian, and Czech.23 The “lexicon-syntax” split is also found among reflexive verbs (Reinhart and Siloni 2005) and middles (Marelj 2004). Reinhart and Siloni capture the split by means of the lex(icon)-syn(tax) Parameter in (53). (53) The Lex-Syn Parameter UG allows arity (valence changing) operations to apply in the lexicon or in

the syntax. Across the languages studied here, the parameter seems to be set once for the various valence reducing operations. Is that a necessity? In section 7.3, once the cluster of distinctions divorcing lexical and syntactic reciprocals have been spelled out, I discuss the parameter, its nature and scope together with the morphological correlation between a syntactic setting of the parameter and the use of a clitic, which has been noted in section 2.4.24 In section 3.5, we have already discussed two important respects distinguishing lexical and syntactic reciprocal verbs. First, lexical but not syntactic reciprocal verbs are symmetric verbs. Second, while the set of lexical reciprocal verbs is closed and limited, 23 Discussion of additional languages is beyond the scope of this paper. Thus, reciprocal verbs in languages such as Chichewa (Mchombo 1991) or Malagasy (Keenan and Razafimamonjy 2001 and Hurst 2006) are not dealt with here, although they constitute a rich empirical array for testing the typology proposed in this paper. I leave this for future research. 24 The correlation holds across Slavic languages, too. A recent study shows that East Slavic languages (Russian, Ukrainian, and Belarusian), which use verbal morphology to form reciprocal, reflexive and middle verbs, set the parameter to "lexicon", while West Slavic languages (Czech ,Slovak, and Polish) and South Slavic languages (Bulgarian, Serbo-Croatian, Slovenian, and Macedonian), which use a clitic, set the parameter to "syntax" (Hron 2009).

27

reciprocalization in the syntax is a productive operation. We now turn to discuss the other distinctions. In the course of the discussion, I freely alternate between the languages in my sample, when illustrating the distinctions. 5.1 Case The operation of reciprocalization, whether lexical or syntactic, can suppress the syntactic realization of an accusative (54) or a dative (55) argument. However, in case a dative argument is suppressed, syntactic reciprocal verbs can realize an accusative argument (55a-b), while their lexical counterparts cannot (55c-d). (54) a. Pierre et Marie se sont embrassés. (French)

Pierre and Marie SE are kissed 'Pierre and Marie kissed'

b. Juan y María se han besado. (Spanish) Juan and María SE have kissed 'Juan and María kissed' c. dan ve-dina hitnašku. (Hebrew) Dan and-Dina kissed(rec) d. János és Mari csókol-óz-t-ak. (Hungarian) János and Mari kissed-rec-past-3pl ‘János and Mari kissed ’ (55) a. Pierre et Marie se sont chuchoté des mots d'amour. (French) Pierre and Marie SE are whispered words of love ‘Pierre and Marie whispered words of love to each other' b. Juan y María se dicen palabras de amor. (Spanish) Juan and María SE say words of love Juan and María said words of love to each other' c. Dan ve-Dina hitlaxšu (*milot ahava / et ha-sodot). (Hebrew) Dan and-dina whispered(rec) (words of love / acc the-secrets ) d. János és Mari (*hízelgő szavak-at) sugdol-ódz-t-ak. (Hungarian) János and Mari (flattering words-acc) whisper-rec-past-3pl The direct object cannot be realized in (55c-d), although reciprocalization has eliminated the dative argument. This reinforces the claim that valence reducing operations applying in the lexicon must reduce accusative case (section 4.1). This is so even when the case of the suppressed argument is dative. When no operation applies in the lexicon, there is no case reduction in the lexicon, and the verb is inserted in the syntax with its case abilities. As mentioned in section 4.2, I assume the clitic (se) reduces case, and is indiscriminant as to whether that case is accusative or dative.25 Note that whether the reduced argument is accusative or dative, syntactic reciprocalization is applicable only upon the merger of the external argument, as predicted (see (45) and the related discussion), even when the verb is a three-place predicate. Thus, while it is possible for a reciprocal anaphoric object to be bound by

25 Syntactic reciprocal verbs can also realize a dative argument when the accusative one is suppressed. This cannot be tested with regard to their lexical counterparts as it seems that there is no instance where reciprocalization suppresses the accusative argument of an input that takes a dative in addition. Se can also reduce nominative case; this happens in certain impersonal constructions (see Cinque 1988, Reinhart and Siloni 2005, Rivero and Milojević Sheppard 2003).

28

an internal coargument, as illustrated for French in (56a), the reciprocalization operation cannot involve two internal coarguments, as in (56b). (56) a. Sur cette photo je montrerai [les deux jeunes]i [l’un à l’autre]i. (French) On this picture I will+show the young man and woman to each other b.**Sur cette photo je se montrerai les deux jeunes. On this picture I SE will+show the young man and woman 5.2 ECM Further, languages forming their reciprocals in the syntax allow reciprocalization of Exceptional Case Marking (ECM) verbs, unlike languages forming them in the lexicon. Consider the ECM verb in (57a) and its reciprocal equivalent in (57b). The matrix predicate voit ‘see’ does not take a DP as its internal argument, but rather a clause. Marie in (57a), to which voit assigns accusative case is not an argument of voit, but the subject of the clause, and receives its θ-role from the embedded verb danser. Nonetheless, voit can undergo reciprocalization involving in addition to its own external role, the external role of the verb danser, as is clear from (57b). (57) a. Pierre voit Marie danser. (French) Pierre sees Marie dance b. Pierre et Jean se voient danser. Pierre and Jean SE see dance ‘Pierre and Jean see each other dance’ Unlike languages with a syntactic setting of the lex-syn parameter, languages where it is set to “lexicon” do not allow ECM reciprocal verbs (58a). They must use a reciprocal anaphoric object to express this meaning (58b). It is worth noting that the verb ‘see’ in Hebrew, for instance, has a reciprocal counterpart (58c), thereby showing that the form is possible, although it has undergone semantic drift (see the English translation) and lost its original meaning. (58) a. * Dan ve-Ron hitra'u racim. (Hebrew) Dan and-Ron see(rec) run b. Dan ve-Ron ra'u [exad et ha-šeni] racim. Dan and-Ron saw each other run

c. Dan ve-Ron hitra'u. Dan and Ron see(rec) 'Dan and Ron met'

This linguistic variation follows directly from the proposal that reciprocalization applies in different components in the two types of languages. Let us see how. As mentioned in section 3.5, the lexicon is composed of lists of entries, and does not involve any syntactic structure. Therefore in the lexicon there is no relation whatsoever between distinct (derivationally unrelated) predicates; they are items on a list that nothing ties together. Only the syntax merges them into structure, thereby establishing structural relations between them. Since reciprocalization of ECM verbs involves roles of two distinct predicates, it cannot take place in the lexicon. When the operation is syntactic it applies after the formation of syntactic structure, which establishes structural relations between distinct lexical items. It is thus not

29

surprising that syntactic reciprocalization can affect θ-roles of two distinct predicates that occur in a local syntactic domain. What defines this domain? As will become clear shortly, it is precisely the domain where A-movement is applicable. This is expected as in both cases the dependency is thematic. Let us see what this means. Consider the Reciprocal ECM verb in (59). The external role of the embedded infinitival is retained and assigned outside its clause upon the merger of the external role of the matrix verb. We need to define the maximal domain in which a θ-role can be retained. In other words, we need to define the maximal domain of application for the θ-criterion. Reinhart and Siloni (2005) suggest that the domain can be defined by means of the EPP, as is shown directly. (59) Pierre et Jean s’entendent [ chanter] . (French) Pierre and Jean SE hear sing Note first that the (CP-less) infinitival complement of the ECM verb in (59) is a TP (and not a smaller clause), as it can host the sentential negation ne pas (60), which is known to be impossible in reduced clauses. (60) Pierre et Jean s’entendent ne pas dire ce qu’il fallait (dire). (French) Pierre and Jean SE hear not say(INF) what should have been (said) 'Pierre and Jean hear themselves not say what they should have said' Since the external role of the embedded infinitival in these configurations is not merged but retained, the embedded TP ought to be specifierless, in other words, EPP-deficient. This claim is not unprecedented; it has already been suggested that the EPP is a requirement of the Cycle (the CP phase) (see Bošković (2002) and Chomsky (2001), among others). That is, C cannot merge with an EPP-deficient TP; hence, the merger of Spec,TP is necessary to complete the CP phase. Other, intermediate Spec,TPs are optional. As the sentential complement of ECM verbs does not project a CP, its Spec,TP can be deficient. (61) states the EPP as a constraint imposed on the CP phase. The maximal domain of application for the θ-criterion can now be defined as the domain of the smallest full (non EPP-deficient) TP. This is precisely the domain defined by the Specified Subject Condition. (61) EPP Merging the outermost Spec,TP of the CP is obligatory. (62) θ-criterion Every θ-role must be assigned in the smallest full (non EPP-deficient) TP. Let us now follow the derivation of the ECM reciprocal in (59). Through the derivation (63a-c), I mark θ-role assignment by labeling constituents with the role assigned to them; unassigned roles appear in angle brackets. In (63a), the Agent role of the embedded verb is not assigned (θi); hence, it stays on the verb and can be carried along. The embedded TP is EPP-deficient; CP cannot be projected. The derivation proceeds to the next VP as in (63b). The higher verb is associated with se; hence, it has no accusative case to check. The matrix verb assigns its Theme role to the TP clause (marked as θf in (63b)). Its external role θk still waits to be assigned. At that stage, in (63c), two θ-roles need to be assigned: the unassigned Agent role (θi) of

30

chanter, which has been retained, and θk of entendent. Upon merger of θk, θi is also assigned, so that both end up associated with the same argument. (63)a. Embedded TP: [TP [chanter <θi>]] b. Matrix VP: [VP se entendent <θk> [TP [chanter <θi>]]θf] c. Matrix TP: [TP Jeanθk+θi [VP se entendent [TP [chanter]]θf]] Thus, the mechanism of parasitic assignment (43) allows assignment of two θ-roles to a single argument in the domain of the smallest non EPP-deficient TP. I would like to raise the possibility that parasitic assignment requires morphological marking, because it is not the regular (canonical) procedure of θ-role merging. An obvious candidate to fulfill this morphological requirement is the clitic se, as it always accompanies verbs undergoing syntactic reciprocalization (or reflexivization). Examining the validity and scope of this requirement is beyond the scope of this paper. Let me nonetheless note that as far as I can see, it rules out the sole potential case of overgeneration that parasitic assignment may encounter. For the sake of simplicity, let me illustrate the case with a reflexive sentence (a parallel sentence can, of course, be construed to allow a reciprocal reading). Consider (64). The question is: Why is parasitic assignment of the role of intelligent, which is retained, blocked upon the merger of the indirect object lui? It is not obvious that the fact that parasitic assignment is a last resort operation suffices to ban this inconceivable sentence, as sembler ‘seem’ is a raising verb with no external role. (64) *Il(EXPL) lui semble intelligent It seems to him intelligent (Impossible intended meaning: It seems to himi that hei is intelligent) Importantly, if morphological marking is required, then the sentence is immediately ruled out as semble is not marked as required.26 Lexical reciprocal (or reflexive) verbs do not involve parasitic assignment and consequently can but do not have to be marked. I now turn to discuss a cluster of three properties, which characterize lexical outputs in general. 5.3 Frozen Input, semantic drift and Idioms There are instances of lexical reciprocal verbs whose transitive alternate does not exist in the vocabulary. For example, the verbs hitgošeš 'wrestled' (65a) and borot’sja 'fight'

26 This requirement may either be specific to verbs assigning two of their roles to the same argument (namely, syntactic reciprocal and reflexive verbs) or broader in scope, applying whenever a θ-role is not assigned by the standard merger (e.g., in passives, where it is suggested that the role undergoes existential binding (saturation) at the level of interpretation (Chierchea 2004, Reinhart and Siloni 2005); indeed passives are morphologically marked across languages (see Haspelmath 1990 and Kazenin 2001). In the former case the requirement may follow from condition B of Reflexivity (Reinhart and Reuland 1993), which requires that a predicate that has two of its θ-roles assigned to coindexed arguments must be morphologically marked. In the latter case, the other way round, it may be that condition B of reflexivity is a particular case of the more general requirement for morphological marking of nonstandard θ-role assignment. I leave further evaluation of this proposal for future research. Note that parasitic assignment in sentences such Jean dessine 'Jean draws' can arguably be banned not only by lack of morphological marking but also due to the unchecked (unassigned) accusative of the verb.

31

(65b) do not have a transitive counterpart. There are no instances of syntactic reciprocal verbs lacking a transitive alternate. (65) a. Dan ve-Dina hitgošešu. (Hebrew) Dan and-Dina wrestled b. Masha i Dima borolis'. (Russian) Masha and Dima fought. Why should that be so? It has often been suggested that the lexicon includes entries that are frozen in the sense that they exist in the lexicon but cannot be inserted into the syntactic derivation, and hence are not part of the actual vocabulary of the language; by vocabulary I mean the sum of words speakers can utter, not a component of the grammar. If frozen entries are available in the lexicon, they can feed lexical operations, in particular, the formation of reciprocal verbs. But they cannot feed syntactic operations because they are not accessible to the syntax. Hence, syntactic reciprocal verbs always have a transitive alternate in the vocabulary of the language (for more on frozen entries, see Horvath and Siloni 2008; for experimental evidence in favor of the existence of frozen entries, see Fadlon 2008). Further, lexical reciprocal verbs can undergo a semantic drift (shift), thereby acquiring a new meaning, alongside the original meaning or replacing it. Semantically drifted reciprocal verbs are found in Hebrew, Hungarian and Russian. For example, the verb vstrechat’sja ‘meet’ in Russian also has the meaning ‘to go out on a date’, which is not shared by its transitive counterpart. Horvath and Siloni (2008) argue that only items that are lexical entries can easily acquire an innovative, drifted meaning, as otherwise this meaning cannot be straightforwardly listed. It automatically follows that lexical reciprocal verbs can drift, while syntactic reciprocal verbs normally keep the meaning of their transitive alternate, as they are not listed in the lexicon.27 Finally, reciprocal verbs formed in the lexicon can appear in phrasal idioms that are not available for their transitive counterparts. The transitive alternate of (66a) and (67a), for example, has only a literal meaning (66b, 67b). Preliminary searches suggest that syntactic reciprocal verbs cannot form idioms not shared by the corresponding transitive verbs. (66) a. Nipageš ba-sivuv. (Hebrew) will+meet(REC.WE) in+the-turn 'Just you wait and see' b. Nifgoš otxa ba-sivuv. Will+meet(we) you in+the-turn 'We will meet you at the turn' (67) a. Nipageš be-emca ha-derex. Will+meet(REC.WE) in+the-middle the-way ‘We will reach a compromise’ b. Nifgoš otxa be-emca ha-derex. Will+meet(we) you in+the-middle the-way 'We will meet you in the middle of the way' 27 Note that an output that has undergone semantic drift may turn into a lexical entry synchronically lacking a transitive alternate. The border line between the two may not always be obvious; this is not important for our purposes.

32

Syntactic reciprocal verbs are completely unavailable in the lexicon. They are reciprocalized in the syntax. Horvath and Siloni (2008) argue that phrasal idioms, such as (66a) and (67a), are listed in the lexicon as subentries of their matrix predicate, that is, their lexical head, the reciprocal verb. Diachronically, idioms start out literal, and acquire a special meaning in some specialized contexts (by ad hoc inferences). After consistent use of the expression with that contextually adapted interpretation, the innovative interpretation enters the lexicon, that is, gets lexicalized. A special meaning of a phrasal expression cannot be readily listed in the lexicon if its matrix predicate is not an entry in the lexicon, as is the case with syntactic reciprocal verbs. A syntactic reciprocal verb can head an idiom only if it shares the idiomatic meaning with its transitive alternate, with which the idiom can be stored because the transitive is a lexical entry.28 There are two additional distinctions between reciprocal verbs across languages, which, at first glance, do not seem to neatly follow from the lex-syn parameter. In section 6 and 7, I will show that on closer inspection, these differences too match the partition to lexical and syntactic outputs. 6. Event nominals Languages setting the lex-syn parameter to “lexicon” have reciprocal event nominals. In Hebrew (68a) and Hungarian (68b), reciprocal nominals are marked for reciprocity by the same morphology used by their verbal counterparts. We do not find anything of the sort in Romance languages, Serbo-Croatian and Bulgarian.29 (68) a. hitnaškut bney ha-esre (Hebrew) kissing(rec) the teenagers 'the teenagers' mutual kissing' b. a gyerekek csókol-óz-ás-a (Hungarian) the-children kiss-rec-nominal-Agr 'the children's mutual kissing' It may be suggested that the explanation for this fact is just morphological, namely that se is incompatible with nominal morphology. While se, indeed, is incompatible with nominal morphology, French (Romance) unaccusative verbs formed with se (69b), nonetheless have nominal counterparts without se (69a). So the question is: why are there no reciprocal event nominals of the sort? (69) a. Le développement de ce garçon est remarquable. (French) the development of this boy is remarkable

b. Le garçon s'est développé de façon remarquable. the boy SE is developed in fashion remarkable 'The boy developed remarkably'

28 As is well known clausal idioms may involve syntactic phenomena such as wh-fronting or passivization. These idioms are often argued to be of a distinct nature (Marantz (1984), which most probably justifies a different listing method for them. 29 The discussion is limited to event nominals, as the interpretation of result nominals is vaguer. The fact that a noun such as kiss could be used in the context of a mutual kissing does not make it a reciprocal nominal.

33

Similarly, in Russian the reciprocal suffix (-sja) is incompatible with nominals. Reciprocal event nominals can appear without it. Thus, (70), for instance, is ambiguous between a transitive and a reciprocal interpretation. (70) obnimanie detej (Russian) hugging children(GEN) 'the children's (mutual) hugging' Moreover, Hron (2006) observes that Czech, unlike Romance languages, Bulgarian and Serbo-Croatian, systematically allows reciprocal event nominals (71), although its reciprocal verbs pattern with syntactic reciprocal verbs. Why does Czech differ in this respect from the other languages forming reciprocal verbs in the syntax? (71) Nepřetržité hádání se jejich dětí jim zkazilo celou dovolenou. Constant quarreling SE their children(GEN) them ruined whole vacation 'Constant quarreling of their children ruined them the whole vacation' (Czech) As observed by Hron (2006), a priori there are two possible derivational paths for reciprocal event nominals: either reciprocalization is followed by nominalization or vice versa. Let us put aside Czech momentarily, and examine how we can account for the split between the other languages setting the parameter to “syntax” and those setting it to “lexicon”. I assume that nominals are formed in the lexicon, as argued in Siloni (1997). If nominalization is a lexical operation, then Hebrew type languages can in principle both nominalize reciprocal entries or reciprocalize nominals, as both operations are lexical and can feed each other. Some evidence in favor of the former derivational path comes from agglutinative languages, such as Hungarian, where the reciprocalization suffix is closer to the root than the nominalizing one, as is clear from the gloss of example (68b). It may still be that this is not the sole derivational path used to form reciprocal nominals in the lexicon. Languages setting the parameter to "syntax", in contrast, cannot nominalize reciprocal entries because the latter are syntactic outputs and therefore not present in the lexicon, where nominals are formed. So this derivational path is not available for them. Why is nominalization of unaccusatives nonetheless possible (69a) in these languages? It has often been argued that unaccusative verbs are present in the lexicon across languages.30 If that is correct, then they can undoubtedly feed nominalization. Can languages with a syntactic setting reciprocalize nominals? Showing that nominalization in Czech is a lexical operation, Hron (2006) suggests that Czech reciprocal nominals are derived by reciprocalization of the corresponding transitive nominal. He further shows that, as is expected from the setting of the lex-syn parameter, reciprocalization of nominals in Czech is a syntactic operation, just like reciprocalization of verbs: it is productive and possible with ECM nominals etc.31 The question then arises why reciprocalization of nominals cannot take place in Romance, Bulgarian and Serbo-Croatian the way it does in Czech. Recall first that the clitic in these languages is a verbal clitic and can never be attached to nouns, unlike its Czech equivalent. Hence, reciprocal nominals with reciprocal morphology (of the 30 See Chierchia (2004), Levin and Rappaport-Hovav (1995), and Reinhart and Siloni (2005), who argue that unaccusatives are universally derived by a lexical operation. 31 See Fassi Fehri (1993), Siloni (1997) for the claim that nominals in certain configurations (languages) license Exceptional Case Marking.

34

Czech type) are blocked. Why is it impossible to derive reciprocal nominals with no reciprocal morphology from the corresponding transitive nouns? The reason for that, I believe, lies in the role of the clitic. In section 5.2, I entertained the idea that syntactic but not lexical reciprocal verbs must be morphologically marked (as their θ-assignment mechanism is not standard). If this is indeed so, then syntactic reciprocalization of nouns is expected to be possible in Czech because it can use the clitic to mark the predicate as required, but not in Romance where the clitic is incompatible with nouns. Lexical reciprocal verbs, as discussed in section 5.2, do not involve nonstandard θ-assignment and therefore do not have to be morphologically marked. 32 7. The discontinuous construction 7.1 Setting the stage Alongside the reciprocal construction discussed so far (e.g., (72a)), languages with lexical setting also manifest the so-called discontinuous construction (or briefly, discontinuity) (72b):33 (72) a. Ha-yeladim ve-ha-yeladot hitnašku. (Hebrew) The-boys and-the-girls kissed(rec) b. Ha-yeladim hitnašku im ha-yeladot. The boys kissed(rec) with the-girls As pointed out by Frajzyngier (1999) and Dimitriadis (2008a), in the discontinuous construction (72b), reciprocity holds between the subject set and the oblique set introduced by the preposition with, and not between the members of the subject set as in (72a). Thus, in (72b), there were mutual kissing events between boys and girls; but no kissing events within the set of boys or girls. Reciprocity in (72a) is not limited this way: kissing events in (72a) are possible between all members of the subject set. As the discontinuous construction denotes reciprocity between the subject set and the oblique set, the subject set can be a singleton set, unlike in regular reciprocal constructions, which require that the subject set be equal to or bigger than two for reciprocity to be possible. The preposition introducing the discontinuous phrase is with or a parallel case morphology. The discontinuous construction entails that both the subject and the oblique constituent play the same role in the event (this is, of course, intriguing and will be discussed in detail in sections 7.4 and 7.5). The construction thus involves a symmetric relation between the subject and oblique constituent. (73a) and (73b), where the syntactic positions of Dan and Dina are reversed, are equivalent. (73c) is a 32 As will become clear in section 7.2, there are instances of lexical reciprocal verbs in “syntax languages". These instances can have nominal counterparts without se in Romance, Serbo-Croatian and Bulgarian, as expected. Further, there are reciprocal nominals such as the French entraide (‘mutual aid’) etc., which involve a derivational affix (entre- ‘mutual’). The existence of such nominals is orthogonal to the present discussion, as they involve a reciprocal affix that is compatible with verbs and nouns, and their formation is idiosyncratic and most probably lexical (not productive, etc.). 33 In English reciprocal verbs such as kiss and hug disallow the discontinuous construction although the language is a language that set the lex-syn parameter to "lexicon" (it has a limited set of reciprocals, no ECM reciprocals, etc.). This may be owing to the morphological identity between the reciprocal verb and its transitive alternate. I will not discuss this issue any further here.

35

contradiction as its second conjunct negates the equal participation of both constituents.34 (73) a. Dan hitnašek im Dina. (Hebrew) Dan kissed(rec) with Dina 'Dan and Dina kissed' b. Dina hitnaška im Dan. Dina kissed(rec) with Dan 'Dan and Dina kissed' c. # Dan hitnašek im Dina, aval Dina lo hitnaška im Dan. Dan kissed(rec) with Dina, but Dina NEG kiss(rec) with Dan It has often been argued that the discontinuous construction is not always reciprocal (does not always involve a symmetric relation between the subject and the oblique constituent), because there are examples where only the subject is volitional (e.g., Gleitman 1965, Dowty 1991, Rákosi 2003, 2008). A famous example is The truck collided with the lamppost. Interestingly, in Hebrew, for instance, the nonreciprocal hitnageš 'collide' introduces its oblique constituent by means of the preposition be- 'in' and not im 'with'. Thus, for instance, in (74a-b), the subject and the discontinuous (with) phrase play the same role in the event, and therefore the discontinuous phrase rejects stationary entities such as ‘the fence’ (74a). In (75a), in contrast, the oblique phrase introduced by the preposition be- can host the stationary object ‘fence’. This is so because the construction is not reciprocal. The oblique phrase does not play the same role in the event as the subject, as shown by the fact that they cannot switch positions (75b). (74) a. Ha-mexonit hitnagša im ha-masa’it/*ha-gader. (Hebrew) the-car collided with the-truck/the-fence b. Ha-masa’it hitnagša im ha-mexonit. the- truck collided with the- car (75) a. Ha-mexonit hitnagša ba-gader

the-car collided with in+the-fence b. *Ha-gader hitnagša ba-mexonit. the-fence collided in+the-car This suggests that there are in fact two distinct lexical entries associated with the form hitnageš 'collide': reciprocal and nonreciprocal. The suggestion receives further support from Hungarian, where two morphologically distinct 'verb particle' complexes realize the two meanings: összeütközik 'collide (with)' (76) vs. bele-ütközik 'bump/run into' (77). 34 To be more explicit (although this is not relevant for our purposes), when the subject set and the oblique set are bigger than two, the sentence involves symmetrical reciprocity (∀x ∈ D ∃y ∈ D' (Rxy & Ryx) (not the symmetric relation proper: Rxy is a symmetric relation if and only if for all x and y, if Rxy then Ryx)). Note that The boys kissed with the girls may be used by speakers (in some context) even if one individual of the subject set was not engaged in kissing at all. As mentioned in note 5, I believe this is so, because in most situations one can speak a little bit loosely (see Lesersohn's (1999) discussion of "pragmatic halos" or "pragmatic slack"), not because the truth-conditions of the sentence allow the exception.

36

(76) a. Az autó össze-ütközött a teherautó-val/*a kerítés-sel. (Hungarian) the car together-collided the truck-with/the fence-with b. A teherautó össze-ütközött az autó-val. the truck together-collided the car-with

(77) a. Az autó bele-ütközött a kerítés-be. the car into-collided the fence-into b. * A kerítés bele-ütközött az autó-ba. the fence into-collided the car-into The crucial point for our purposes is that every discontinuous construction has a reciprocal (symmetric) interpretation. It may additionally have a nonreciprocal reading, which I believe, given the above arguments, is based on a homophonuous nonreciprocal verb.35 Thus, (78) allows both readings: either Dina argued with Dan as well (i), or she did not (ii). (78) a. Dan hitvake’ax im Dina. (Hebrew) Dan argued with Dina

i. Dina hitvakxa im Dan. Dina argued with Dan ii. Dan hitvake’ax im Dina, aval Dina lo hitvakxa ito.

Dan argued with Dina, but Dina NEG argued with Dan. Our discussion hereafter concerns reciprocal verbs only; using the term discontinuous construction (or discontinuity) I refer exclusively to reciprocal constructions. The discontinuous construction seems to be impossible with syntactic reciprocal verbs, as illustrated below. (79) a. *Jean s’est embrassé avec Marie. (French) Jean SE is kissed with Marie b. *Giovanni si è abbracciato con Maria. (Italian) Giovanni SI is hugged with Maria c. *Juan se acusó con María. (Spanish) Juan SE accuse with Maria d. * Ana s-a curăţat cu Ion. (Romanian) Ana SE-has cleaned with Ion e. * Dan se obviňoval s Petrem. (Czech)

Dan SE accused with Petr However, a closer inspection reveals that certain reciprocal verbs in languages where the parameter is set to “syntax” do allow the discontinuous construction. In the next subsection, I will nonetheless argue that the discontinuous construction is possible only with lexical reciprocal verbs. I will show that there are good reasons to believe that there are instances of lexical reciprocal verbs in languages with a syntactic

35 This is not surprising. Reasonably, the nonreciprocal entry has developed from the reciprocal one by semantic drift. As will become clear in section 7.4, the subject of the discontinuous construction is not only structurally but also thematically more prominent than the discontinuous phrase. This seems to facilitate the drift into a nonreciprocal verb. Note incidentally that at least in Hebrew most (if not all) verbs that have acquired a nonreciprocal reading are verbs that lack a transitive alternate in the vocabulary.

37

setting, and only these can appear in the discontinuous construction and show other properties typical of lexical outputs, such as semantic drift. Syntactic reciprocal verbs disallow the discontinuous construction. 7.2 Lexical reciprocal verbs in “syntax languages” In certain languages with a syntactic setting of the lex-syn parameter (e.g., French and Italian), we find isolated reciprocal verbs allowing the discontinuous construction, while in others, i.e., Serbo-Croatian, Bulgarian and Czech a wider set of reciprocal verbs allows discontinuity. Below I will show that there are good reasons to believe that these are instances of lexical reciprocal verbs. The French verb se battre allows discontinuity. The basic entry, battre, means ‘beat’ (and ‘defeat’). Se battre usually means ‘fight’ (80i) and can also mean ‘beat’ (80ii); the two meanings are distinct, as shown by the discourse context suggested for (80ii). (The meaning ‘defeat’ is logically impossible for se battre, as it necessarily results in a contradiction.) Thus, a semantic drift has applied to se battre, associating it with the meaning ‘fight’. As mentioned in section 5.3, only items that are lexical entries can acquire an innovative, drifted meaning, as otherwise this meaning cannot be listed. If that is correct, it follows that the reciprocal verb se battre 'fight' must be listed in the lexicon, because it is associated with a special meaning. (80) Jean et Pierre se sont battus. (French) Jean and Pierre SE are beat i. Jean and Pierre fought. ii. Jean and Pierre beat each other. Discourse context for (ii): Jean et Pierre étaient sensés battre les noyers pour

faire tomber les noix, mais par mégarde ils se sont battus occasionnellement. (‘Jean and Pierre were supposed to beat the walnuts to make the nuts fall, but by mistake they beat each other occasionally’)

Moreover, it turns out that when se battre occurs in the discontinuous construction, it can only mean 'fight' (81). (It has an additional drifted nonreciprocal meaning ‘struggle with something’ when it appears with a with phrase; this reading is not reciprocal and hence irrelevant here, as mentioned above.) As ‘fight’ is a drifted meaning, it means that only the lexical reciprocal se battre licenses the discontinuous construction. Note that this means that the lexical reciprocal se battre has lost the original meaning of battre, otherwise it would also be able to mean 'beat' in the discontinuous construction. (This is not always the case: sometimes items acquire a drifted meaning in addition to the original one.) Note that in (80) se battre can also mean ‘beat’ (ii). This is so because in (80) it is ambiguous between a syntactic reciprocal verb (ii) and a lexical one (i). The fact that the lexical se battre has lost the original meaning allows us to determine that only the lexical alternate can feed the discontinuous construction, thus supporting my claim that the discontinuous construction cannot be fed by outputs of a syntactic operation. If indeed se battre in the discontinuous construction is a lexical reciprocal verb, as I argue, it ought to show other properties of lexical reciprocal verbs. This prediction is borne out. While in general sentences with se battre allow both a singular event reading and a sub-event reading (e.g. (80)), the discontinuous construction can only describe a singular event of fighting (81). This is why

38

modification by the adverbial cinq fois 'five times' gives rise to exactly five events of fighting.36 (81) Jean s’est battu avec Pierre cinq fois. (French) Jean SE is beat with Pierre five times There were five events of fighting between Jean and Pierre. In the same vein, the reciprocal verb ‘kiss’ in Czech ((16b) repeated as (82a)) and Romanian (82b) can denote ten events if the context enforces it, just like their French counterpart (see (9) and the relevant scenario). (82) a. Petr a Pavel se pětkrát políbili. (Czech) Petr and Pavel SE five times kissed b. Ana şi Ion s-au sărutat de cinci ori. (Romanian) Ana and Ion SE-have kissed five times In both languages, the reciprocal 'kiss' licenses the discontinuous construction (83), in contrast with its French equivalent (79a). (83) a. Pavel se políbil s Petrem. (Czech)

Pavel SE kissed with Petr. 'Pavel and Petr kissed' b. Ana s-a sărutat cu Ion. (Romanian) Ana SE-has kissed with Ion 'Ana and Ion kissed' Crucially, upon modification by the adverbial 'five times', the discontinuous construction (84a-b) allows the five event reading only, in sharp contrast with (82). The reciprocal 'kiss' in Romanian and Czech, when appearing in the discontinuous construction, necessarily denotes a singular event. This is why modification by 'on the forehead' results in an anomalous sentence, as illustrated in (84c) with a Czech example; a singular event of mutual kissing cannot be an event of kissing on the forehead (compare (84c) with its nondiscontinuous counterpart (16c)). This is exactly what we predict if the construction is limited to lexical reciprocal verbs. (84) a. Pavel se pětkrát políbil s Petrem. (Czech)

Pavel SE five times kissed with Petr 'Pavel and Petr kissed five times' b. Ana s-a sărutat cu Ion de cinci ori. (Romanian) Ana SE-has kissed with Ion five times 'Ana and Ion kissed five times' c. Petr se s Pavlem políbil (#na čelo). Petr SE with Pavel kissed (on forehead) Finally, one does not find ECM verbs in the discontinuous construction. Thus, while (82a) is grammatical in Romanian, (85b), which involves an ECM reciprocal, is completely impossible. 36 In general, we predict that the reciprocity of drifted reciprocals in both languages setting the parameter to "lexicon" and "syntax" to be associated with a singular event, as they are lexical entries. Being lexical entries, this is the only type of reciprocity they can denote.

39

(85) a. Ana s-a sărutat cu Ion. (Romanian) Ana SE-has kissed with Ion 'Ana and Ion kissed' b. *Ana s-au auzit cu Ion cantand Marsilieza. Ana SE-has heard with Ion singing Marseillaise This is expected if only lexical (not syntactic) reciprocal verbs can feed the discontinuous construction. ECM reciprocals involve two distinct predicates and therefore can only be formed in the syntax (see section 5.2). As they are syntactic outputs, they cannot appear in the discontinuous construction. In Czech we find a minimal pair of the same sort. Among the reciprocal verbs that Czech allows to appear in the discontinuous construction, we find the verb vidět se (86). The verb vidět means ‘see’. The reciprocal vidět se means ‘see each other’ or ‘meet’. When vidět se appears in the discontinuous construction, it predominantly means ‘meet’ (the drifted meaning), and can also mean ‘see each other’ at least for some speakers. (86) Pavel se viděl s Petrem. (Czech) Pavel SE saw with Petr 'Pavel and Petr saw/met each other' Crucially, when vidět se functions as an ECM predicate, it has its original meaning only, and reciprocal discontinuity is ruled out (87b). Again, this is expected if only lexical reciprocal verbs can give rise to the discontinuous construction. (87) a. Pavel a Petr se viděli tančit. (Czech) Pavel and Petr SE saw dance b. *Pavel se viděl s Petrem tančit. Pavel SE saw with Peter dance Intended meaning: ' Pavel and Petr saw each other dance' Given the above data, we must conclude that there are instances of lexical reciprocal verbs in languages with a syntactic setting of the lex-syn parameter, and only these reciprocal verbs allow the discontinuous construction (see also note 32).37

37 There is an additional related distinction between lexical and syntactic reciprocal verbs that reinforces this claim. Note first that reciprocal anaphors are not always in complementary distribution with reciprocal verbs. In Romance languages, in case an accusative argument is suppressed by reciprocalization, a bare reciprocal anaphor can be added (i), and in case a dative is suppressed, it is introduced by the dative preposition (ii). In Hebrew (iii) and Hungarian, reciprocal anaphors can only be added introduced by with. Russian does not allow them readily but to the extent that it does, they must be introduced via with. The generalization seems to be that with lexical reciprocal verbs, reciprocal anaphors can only be added when introduced by with. Syntactic reciprocal verbs in Serbo-Croatian and Czech do not avail themselves of the Romance option. This may be related to the different ways they realize emphatic pronouns (Siloni 2001). Important for our purposes is the fact that lexical reciprocal verbs in languages with syntactic setting, to the extent that they allow the addition of a reciprocal anaphor, pattern with Hebrew and Hungarian, adding the anaphor via with. Compare the minimal pair in (iv-v). When the reciprocal anaphor is bare, it is the syntactic se battre 'beat' (iv), when the reciprocal anaphor is introduced via with, it is the lexical reciprocal 'fight' (v). This provides further support to the claim that there are instances of lexical reciprocal verbs in "syntax languages", and that

40

7.3 Interim conclusion: On the nature of the lex-syn parameter We have seen that there are instances of lexical reciprocal verbs in languages where the lex-syn parameter is set to “syntax”. These instances seem to belong to the core set typical of languages with a lexical setting. But there are considerable variations: the number of lexical reciprocal verbs varies from one “syntax language” to another, and the exact choice of verbs seems to be idiosyncratic. In languages forming their reciprocal verbs in the lexicon, in contrast, we never find instances of syntactic reciprocals, e.g., reciprocal ECM verbs. This is expected, because when the operation is syntactic it is necessarily productive. It is important to emphasize that the classification to lexical and syntactic settings dictated by the lex-syn parameter is well-founded and insightful, despite the existence of instances of lexical reciprocal verbs in “syntax languages”. These instances do not substitute their syntactic equivalents, as discussed in detail in section 7.2, and therefore they do not undermine the generalizations discussed in section 5. Rather, in addition to a wide set of syntactic reciprocal verbs, which pattern as expected by the lex-syn parameter, these languages can also have additional instances that systematically behave as lexical outputs. As already mentioned in the beginning of section 5, a set of valence changing operations is subject to the lex-syn parameter, and there is evidence that the value of the parameter is identical across the various operations, at least those that use the same morphological form, as is the case with reciprocalization, reflexivization, and middle formation in my sample. The parameter, thus, derives important cross-linguistic generalizations. Languages setting the parameter to “lexicon” are languages that can only carry out these operations in the lexicon. That is, they ban the application of reciprocalization in the syntax. Hence, there are no instances of syntactic reciprocal verbs in “lexicon languages”. Languages setting the parameter to “syntax”, in contrast, do not exclude the possibility of having instances of lexical reciprocal verbs. The question is whether these instances are outputs of lexical reciprocalization or reciprocal verbs that have gotten lexicalized as such, and have to be acquired on an individual basis. Taking parameter setting at the acquisition stage into consideration, I tend to opt for the latter alternative. Note first that the cluster of distinctions between the two types of languages provide the necessary triggers for acquisition. As the value of the lex-syn parameter turns out to be identical across various operations (reciprocalization, reflexivization, and middle formation), parameter setting is facilitated because evidence from various sources (operations) converges to set the choice. More they systematically pattern with lexical reciprocal verbs. I will not investigate these constructions any further here. (i) Jean et Pierre se dessinent l’un l’autre. Jean and Marie SE draw one the other (ii) Jean et Pierre s’écrivent l’un à l’autre.

Jean and Marie SE write one to the other (iii) Dan ve-Dina hitnašku exad im ha-šeni. Dan and-Dina kissed(rec) one with the-other (iv) Jean et Pierre se sont battus l'un l'autre. Jean and Pierre SE are beat one the other (v) Jean et Pierre se sont battus l'un avec l'autre. Jean and Pierre SE fought one with the other

41

specifically, consider reciprocalization. Setting the lex-syn parameter to “syntax” will be triggered by encountering reciprocal ECM verbs and reciprocal verbs that do not belong to the universal lexical set. By contrast, the existence of reciprocal event nominals and discontinuous reciprocal constructions will trigger a lexical setting. But given the existence of instances of lexical reciprocal verbs in “syntax languages” as well as reciprocal event nominals in Czech, the child may be exposed to both types of triggers. How are such data processed by the acquirer? Concluding that the acquired language allows both syntactic and lexical application of reciprocalization, the child risks overgeneralizing, as certain languages with a syntactic setting (e.g., French) have only isolated instances of lexical reciprocal verbs, and not the whole lexical set. It thus seems more plausible that upon exposure to both types of evidence, the child has to acquire the lexical reciprocal verbs on an individual basis. If this is correct, then also in these languages the operation of reciprocalization is limited to apply in one component only, namely, the syntax, just like in “lexicon languages” the operation applies exclusively in the lexicon. The fact that “syntax languages” can have instances of lexical reciprocal verbs but not vice versa naturally follows from the different nature of these two components. As already mentioned in section 5.3, the lexicon allows listing of irregularities, while the syntax is a computational system, not an inventory of items. There is additional evidence that indeed these instances of lexical reciprocal verbs in "syntax languages" are not the outputs of lexical reciprocalization, but rather listed on an individual basis. Most properties of lexical reciprocal verbs discussed in sections 5-6 follow from their lexical status. But one characteristic is a direct result of the operation they undergo, namely the lack of accusative case. As discussed in section 5.1, the operation of lexical reciprocalization has the effect of reducing accusative case, even if it targets the dative argument ((55c-d) repeated in (88a-b)). (88) a. Dan ve-Dina hitlaxšu (*milot ahava / et ha-sodot). (Hebrew) Dan and-dina whispered(rec) (words of love / acc the-secrets) b. János és Mari (*hízelgő szavak-at) sugdol-ódz-t-ak. (Hungarian) János and Mari (flattering words-acc) whisper-rec-past-3pl Obviously, also in the discontinuous construction lexical reciprocals cannot realize an accusative complement. (89) a. Dan hitlaxeš im Dina (*milot ahava / et ha-sodot). (Hebrew) Dan whispered(rec) with Dina (words of love / acc the-secrets) b. János (*hízelgő szavak-at) sugdol-ódz-ott Mari-val. (Hungarian) János (flattering words-acc) whisper-rec-past.3sg Mari-with If in "syntax languages", the reciprocal verb occurring in the discontinuous construction is not an output of the operation of lexical reciprocalization, but rather listed as is, then the verb is expected not to lose its accusative case ability as it is the operation of reciprocalization that reduces it. This prediction turns out to be correct. Unlike lexical reciprocal verbs in languages with a lexical setting of the parameter, lexical reciprocal verbs in "syntax languages" can take an accusative complement, as shown by the Czech examples in (90).38

38 Dimitriadis (2004) entertains the idea that the lex-syn parameter should determine whether or not the operation can apply in the syntax, whereas lexical application is always possible. Given our conclusion

42

(90) a. Petr si psal s Marií dopisy. (Czech) Petr SE wrote with Marií letters 'Petr and Marií wrote letters to each other' b. Petr si šeptal s marií tajnosti. Petr SE whispered with Marií secrets 'Petr and Marií whispered secrets to each other' Turning back to the setting of the lex-syn parameter, the next question is whether the setting is necessarily coherent across the various valence changing operations of a given language. This is an empirical question. It may be that the parameter is indeed set once for all valence changing operations per language. Judging by the sample of languages studied here, this seems to be the situation. But it could also be that the setting of the parameter is not automatically coherent per language, but more directly contingent upon the morphological inventory of the language, along lines suggested by Borer (1984), for instance. Under this option, one could imagine that verbs with different morphology could be associated with different values of the parameter. This leads us to the question whether the correlation observed in the present sample of languages between syntactic application of the operation and the use of a clitic (se) is a universal generalization. Again, this is an empirical issue. A priori, there is no obvious reason why this must be so. It is well documented that there is a diachronic pattern of development from reflexive anaphors to verbal markers of intransitivity (Ariel 2006, Gast and Haas 2008, Haspelmath 1990, Kemmer 1993). Reflexive anaphors become phonologically reduced, develop into nonargumental elements, and then into verbal morphology. Both markers of lexical and syntactic reciprocalization in the sample of languages studied here are markers of intransitivity to the extent that they mark syntactic valence reduction. The empirical question is whether there is some strict correlation between the morphological stage of development (clitic, verbal morphology) and the locus of application of the operation of reciprocalization (reflexivization, or middle formation).39 The next section examines the status of the discontinuous phrase with regard to argumenthood. 7.4 Argument or adjunct? The discontinuous phrase is reminiscent of the so-called comitative with phrase, which can be added rather freely to sentences, as in (91). A conceivable analysis entertained by Siloni (2001) is that the discontinuous construction simply utilizes a

that the instances of lexical reciprocal verbs in "syntax languages" are not outputs of lexical reciprocalization, but rather listed on an individual basis, this proposal cannot be maintained. As noted in section 3.5, the set of reciprocal verbs across "lexicon languages" is rather coherent but not identical (roughly, verbs denoting interaction between people). The definition of the set is not yet understood. It may turn out that specific properties of the lexicon determine the set. Any language specific properties of the set would also have to be acquired on an individual basis (like the instances of lexical reciprocal verbs in "syntax languages"). 39 A lexical setting of the lex-syn parameter is marked by verbal morphology in Hebrew, Russian and Hungarian. It can also be marked by a nonargumental element, such as the Dutch zich, which marks lexical reflexivization (Reinhart and Siloni 2005). A syntactic setting can be marked by se, and by the phonologically weak occurrence of the German sich (Gast and Haas 2008, and Reinahrt and Siloni 2005). See Papangeli (2004) for the claim that a syntactic setting can also be marked by verbal morphology. I will not pursue this issue any further here.

43

comitative phrase. (91) John rode to the store (with Mary). There are however important distinctions between the comitative phrase and the discontinuous one. For one thing, the comitative phrase can be freely dropped (91), while the discontinuous one does not readily allow omission (92). Further, the interpretation of the comitative phrase is vaguer than that of the discontinuous phrase. Dimitriadis (2004) observes that in (91), for instance, Mary may have been given a ride rather than riding a bike herself. Dimitriadis (2004), Komlósy (1994), and Rákosi (2003, 2008) convincingly argue that while the comitative phrase is an adjunct, the discontinuous phrase is an argument. If so, then discontinuous reciprocal verbs must be two-place predicates, unlike regular (i.e., non-discontinuous) reciprocal verbs. This explains why the discontinuous phrase is not readily dropped and why its interpretation is stricter. Moreover, to the extent that the discontinuous phrase can be omitted, it is implicit, as expected if it is an argument (92). But if we drop the comitative phrase in (91), the sentence, obviously, does not imply the participation of any additional entity in the event (Komlósy 1994; Rákosi 2003). (92) ??Dan hitnašek. (Hebrew) Dan kissed(rec) 'Dan and someone kissed' An additional test that the comitative phrase passes but the discontinuous phrase fails is the addition of the modifier 'together' (Komlósy 1994, Rákosi 2003). (93a) cannot mean that there was a singular event of kissing whose participants are Dan and Dina, because the discontinuous phrase disallows the modifier 'together'. It is nonetheless marginally acceptable with the meaning: 'Dan and someone kissed in the presence of Dina, or while Dina was also kissing someone'. This meaning results from analyzing the sentence as involving a comitative phrase, 'together with Dina', and omission of the discontinuous phrase (the omission is only marginally possible, as already illustrated in (92)). By contrast, (93b) is possible, as the with phrase is a comitative phrase. (93) a. ??Dan hitnašek yaxad im Dina. (Hebrew) Dan kissed(rec) together with Dina b. Dan halax habayta yaxad im Dina. Dan went home together with Dina One may suggest that 'together' cannot be added to the discontinuous phrase, as it is redundant, because it is clear that if Dan and Dina kissed they were together. But it is equally clear in (93b) that they were together, and nonetheless modification by 'together' is possible. Moreover, writing to each other does not entail being together, and still (94) is impossible (I find omission of the discontinuous phrase impossible here; hence, the sentence is ungrammatical). The modifier 'together', thus, serves as an additional test distinguishing between the comitative and discontinuous phrases. (94) *Dan hitkatev yaxad im Dina. (Hebrew) Dan wrote(rec) together with Dina

44

In sum, the discontinuous phrase is an argument, whose participation in the event seems equal to that of the subject. Given that, we would expect it to pass tests diagnosing Agenthood. However, Rákosi (2003) observes that when an infinitival adjunct and an Agent oriented adverb, which both diagnose Agenthood, are added to the discontinuous construction, they can refer to the subject only, and not the discontinuous phrase. This is illustrated in (95a-b). If both arguments are interpreted as Agents, why do these Agent tests diagnose the subject only? . (95) a. Dan hitnašek im Dina [bli le-hit'ayef]. (Hebrew) Dan kissed with Dina without to-be+tired b. Dan hitnašek im Dina be-xavana. Dan kissed with Dina in-intention It can be argued that this follows from the fact that the subject is structurally higher than the discontinuous phrase. Consider, for instance, the French causative construction in (96). It is a biclausal structure containing two Agents, that of the higher verb, Jean, and that of the lower verb, Paul. Only the higher Agent can control the subject of avoir peur 'being afraid' in (96a) and be modified by exprès 'on purpose' in (96b). (96) a. Jean a fait courir Paul [sans avoir peur]. (French) Jean has made run Paul without have fear 'Jean, without being afraid, made Paul run'

b. Jean a fait courir Paul exprès. Jean has made run Paul deliberately 'Jean, deliberately, made Paul run' But the Instrument diagnostic of Agents is not equally sensitive to structural hierarchy. An Instrument requires the explicit or implicit presence of an Agent. In (97), the Instrument 'with a cane' can modify the Agent of either faire 'make' or marcher 'walk', despite the structural 'superiority' of the former. (97) Jean a fait marcher Paul avec une canne. (French) Jean has made walk Paul with a cane 'Jean made Paul walk with a cane' An Instrument added to the discontinuous construction also diagnoses as Agent the subject only and not the discontinuous phrase. Sentence (98a) states that Dan used a fountain pen to write to Dina. As to Dina, (98a) does not supply any information regarding the instrument she used. Hence, the addition of 'and she wrote to him with a pencil' does not give rise to a contradiction (98b). (98) a. Dan hitkatev im Dina be-et nove’a. (Hebrew) Dan wrote(rec) with Dina in-pen fountain 'Dan corresponded with Dina using a fountain pen' b. Dan hitkatev im Dina be-et nove’a ve-hi katva lo be-iparon. Dan wrote(rec) with Dina in-pen fountain and-she wrote to him with pencil 'Dan corresponded with Dina using a fountain pen and she wrote to him

with a pencil'

45

Despite the fact that the discontinuous phrase is an argument whose participation in the event seems equal to that of the subject, it fails to be diagnosed as Agent. This leads me to conclude that lexical reciprocal verbs assign their Agent-Theme role exclusively to the subject.40 The discontinuous phrase is not assigned this role. Hence, the Instrument diagnostic does not detect it. But this is very puzzling as the phrase is nonetheless an argument. How is it licensed? 7.5 Dyadic reciprocal verbs As shown in the previous section, the discontinuous phrase is an argument. This means that lexical reciprocal verbs, in fact, have two possible realizations: as monadic predicates and as dyadic predicates. As will become clear at the end of the section, this behavior is not their peculiarity; it turns out to be typical of any verb whose lexical meaning codes symmetry. In other words, the discontinuous construction is a property of symmetric verbs. Hence, the construction is possible with lexical reciprocal verbs, but not with syntactic reciprocal verbs, whose reciprocity is formed by the accumulation of sub-events after syntactic merging. Both the monadic reciprocal verb (99a) and its dyadic counterpart (99b) involve an [Agent-Theme] role. In both it is assigned to the subject argument. However, the dyadic reciprocal verb differs from the monadic one in two important respects. (i) It has an additional argument that is not assigned the Agent-Theme role but is nonetheless interpreted as being an Agent and a Theme (see (73) and the relevant discussion). (ii) It involves a symmetric relation between its two arguments, and not between the individuals of the subject set, as is the case with monadic reciprocal verbs (see note 34). (99) a. Dan ve-Dina hitnašku. (Hebrew) Dan and-Dina kissed b. Dan hitnašek im Dina. Dan kissed(rec) with Dina ‘Dan and Dina kissed’ These two characteristics are related. The symmetric relation holding between the two arguments of (lexical) dyadic reciprocal verbs explains why the oblique argument is interpreted as an Agent (and Theme), although it is not assigned this role. Being in a symmetric relation with the subject entails being interpreted as associated with the same θ-role assigned to the subject. This will be formulated shortly. Before doing that, however, it must be spelled out what it means to be an argument whose role does not specify any thematic content in the lexicon. If θ-roles are grammatical primitives, it is not clear how an argument can lack thematic content. But if θ-roles are conventionalized labels for feature clusters, then an unspecified role is possible. Following Reinhart (2002), I assume that θ-roles are composed of two atomic features: c, which determines whether or not the argument in question is necessarily responsible for causing the denoted event, and m, which determines whether or not the mental state of the argument in question is relevant to the denoted event. Each of these features can be valued for [+], [−], or left unvalued/unspecified. Thus, for example, the Agent role is [+c +m], as it brings about the relevant event or change and must be animate (its mental state is relevant). The Cause role, in contrast, is 40 I am not aware of any diagnostic specifically detecting Themes. Hence, it is impossible to apply parallel diagnostics regarding the Theme.

46

[+c] as it is unspecified with regard to mental state, and can either be realized by an animate argument or not. The Theme role is [-c -m], as it does not trigger the change in question nor is its mental state relevant to the event.41 Crucial for our purposes is the fact that when both features are unspecified, the role lacks thematic content. This is an empty cluster [Ø], an argument whose thematic content must reasonably be determined in the course of the derivation, or else the derivation would crash owing to the requirement of Full Interpretation (Chomsky 1986, 1995).42 The lexical representation of reciprocal verbs, then, includes two entries, monadic (V) and dyadic (V'), as schematized in (100a) and (100b) respectively.43 The empty role [Ø] is introduced into the structure by the preposition with (or a parallel case morphology). The two together form one semantically interpreted role [Ø-WITH]. (100) Reciprocal verb a. Monadic: VSYM [Ag-Th] b. Dyadic: V'SYM [Ag-Th], [Ø-WITH] The event semantic representation of (99b) is given in (101). The next question is how the empty role gets interpreted. (101) ∃e[kiss'SYM (e) & [Ag-Th](e, Dan) & [Ø-WITH] (e, Dina)] I assume a meaning postulate relates V'SYM to V, the input of lexical reciprocalization. Following the guidelines suggested in section 4.1 with regard to the lexical semantics of VSYM, I assume the meaning postulate defines V'SYM as entailing two underlying events of the same type (e1 and e2), by means of Landman's down (↓) operation (which maps atomic events to the sum of the events underlying them). The simplex roles making up the complex role [Agent-Theme] are defined at the level of the underlying events. Crucially, each simplex role is defined for a different underlying event (as in the case of VSYM). Thus, the subject argument is associated with the Agent role of the underlying event e1 and the Theme role of e2. Finally, the thematically empty argument [Ø-WITH] is interpreted as being in a symmetric relation with the subject argument; more precisely, it is interpreted as the Agent role of the underlying event e2 and the Theme role of e1. (101) then entails (102).44 41 The mental state of an argument specified -m, say [-c-m], the door in (i), remains irrelevant at all stages of the derivation. In contrast, an argument bearing a [+c] cluster, which is unspecified with regard to m, is interpreted at the semantics either as an Agent (Dan in (i)) or as a non-Agent (say, an Instrument, this key in (i)) (see Marelj 2002 for more discussion). (i) Dan /This key opened the door. 42 The idea that the oblique argument in the discontinuous construction is somewhat underspecified has already been suggested by Rákosi (2008). However, according to Rákosi, the role, which he labels Partner, is underspecified for the Proto-Agent properties: volitional involvement, sentience and being a causer (see Dowty 1991). Moreover, the role is underspecified to the extent that it may, but does not have to, be associated with these properties. That is, it is not a radically empty role whose content has to be determined in the course of the derivation. This view is incongruous with our finding that the oblique role is interpreted as being an Agent and a Theme. 43 The claim that one concept can give rise to two distinct entries is not new. See, for instance, Higginbotham's (1989) claim that root-related homonyms such as spray paint on the wall and spray the wall with paint are distinct entries derived from the same "theoretical predicate" (concept). 44 The lexical information associated with kiss'SYM can be defined as in (i). (i) Kiss'SYM is a set of singular events for which the roles [Ag-Th] and [Ø-WITH] are defined such that

for some d1: Ag-Th (e) = d1

47

(102) ∃e∃e1∃e2 [↓(e) = (e1 |_| e2) & [kiss (e1) & Ag(e1, Dan) & Th(e1, Dina) &

kiss(e2) & Ag(e2, Dina) & Th(e2, Dan)] Recall now that the oblique argument fails Agenthood tests. Why is that so? After all, the phrase ends up interpreted as an Agent (and a Theme) on a par with the subject (102). In fact, this is exactly what our approach predicts. (101) entails (102) by a meaning postulate (see note 44), but it does not involve a parallel decomposition, that is, it is not equivalent to (102).The Agent role is not assigned to the discontinuous phrase. Moreover, the discontinuous phrase is not the Agent of e but rather the Agent of one of its underlying events, which are inaccessible to the syntactic and semantic representations as they are entailed by a meaning postulate. Recall that a count adverbial, too, does not "see" unerlying events owing to their "invisible status", as discussed in section 4.1. Thus, modification of both the monadic VSYM (2.2) and the dyadic V'SYM (section 7.2) by the count adverb five times results in exactly five events. Similarly, the discontinuous phrase, which is an Agent of an underlying event, is invisible to Agent diagnostics (modifers). The subject, Dan, is also an Agent (and a Theme) of an underlying event. However, crucially, it also receives the [Agent-Theme] role of e, which is visible to the semantic and syntactic representations, and it is precisely this which makes it detectable by Agenthood diagnostics, unlike its discontinuous counterpart. In sum, the discontinuous phrase gets interpreted as an Agent and a Theme, although it is not directly assigned these roles. This is possible as the discontinuous phrase is an argument whose thematic content is initially unvalued. The phrase gets associated with semantic content (gets interpreted) in the course of the derivation owing to the symmetric relation it bears to the subject. Across languages, the discontinuous phrase is introduced by the preposition 'with' (or its equivalent case marking), which is also the preposition that introduces comitative phrases (see section 7.4). This is not arbitrary. In both cases the preposition introduces phrases whose semantic content is specified in the course of the derivation, either owing to the symmetric relation that the verb imposes (in the case of the discontinuous phrase) or by the context (in the comitative case; see example (91) above and the related discussion). The availability of a thematically unspecified role is a property of symmetric verbs, which in addition to having a monadic instantiation can also have a dyadic one (100). Their dyadic realization involves a thematically empty argument. When they are monadic, they enforce a symmetric relation between the individuals of their sole argument, the subject. As dyadic predicates, they enforce a symmetric relation between their two arguments, the subject, and the oblique phrase. As discontinuity is a property of symmetric verbs, syntactic reciprocal verbs do not qualify; they are not symmetric predicates, as their reciprocity is built from sub-events after syntactic merging. They are not entries in the lexicon, and cannot license a thematically empty argument, which is a trait of the discontinuous construction. If discontinuity is a property of symmetric verbs, then it is expected that any verb whose lexical meaning codes symmetry should avail itself of the discontinuous for some d2: [Ø-WITH] (e) = d2

Meaning postulate for kiss'SYM If e ∈ kiss'SYM and Ag-Th (e) = d1 and [Ø-with](e)=d2 then for some e1, e2 ∈ kiss: ↓(e) = (e1|_| e2) and Ag(e1) = d1 and Th(e1) = d2 and Ag(e2) = d2 and Th(e2) = d1

48

construction, not only lexically derived reciprocal verbs. Consider verbs such as shake hands and play. While shake hands must denote reciprocity, play can, but does not have to, describe a reciprocal situation. Let us limit our discussion here to the reciprocal play. Both shake hands and the (reciprocal) play are not the outputs of reciprocalization; their reciprocity is intrinsic. The criteria for deciding that are the following. First, they do not have a two-place alternate from which they could be derived. But, given the existence of reciprocal verbs whose transitive alternate is frozen (section 5.3), the question arises how it can be decided whether they (or any other instance of the sort) are underived or derived from a frozen input. Importantly, not only do these verbs lack a transitive alternate, but in addition it is even hard to imagine what the corresponding transitive concept could be. Moreover, across languages both verbs do not bear morphology typical of valence reducing operations; this, too, suggests that they are underived. Finally, at least play can assign accusative case (shake hands is composed of the verb and its accusative complement). A verb assigning accusative cannot be formed by the operation of lexical reciprocalization, as the operation always eliminates accusative case (see discussion in sections 5.1 and 7.3). Since the lexical meaning of these underived entries encompasses reciprocity, we predict that their reciprocity ought to be associated with a singular event, as plural events are not part of the lexicon's inventory (generalization (29)). Indeed, as expected, both (103a) and (103b) do not have the reading where there were more than five events of shaking hands and playing chess respectively between John and Mary. (103) a. John and Mary shook hands five times. b. John and Mary played chess five times. Furthermore, as expected if discontinuity is a trait of verbs whose lexical meaning codes symmetry, they allow the discontinuous construction. Both shake hands and the (reciprocal) play clearly take a discontinuous argument, not a comitative adjunct. To the extent that with Mary can be omitted in (104a), the participation of an additional entity in the event is implicit. Likewise, as arguments are not readily dropped, the preferred reading of (104b) without the oblique phrase is not reciprocal. (104) a. John shook hands ??(with Mary). b. John played (with Mary). Now, although (105a) states that both Dan and Dina played, the Instrument 'red pen' modifies Dan's playing only, as is the case with (lexically) derived reciprocal verbs. Thus, we have no information as to which instrument Dina used in the game: it may be the same pen but it may also be another pen, a pencil, etc. Hence, the addition of 'and she used a blue pen' does not give rise to a contradiction (105b).

(105) a. Dan sixek iks-miks-driks im Dina be-et adom. (Hebrew) Dan played x-mix-drix with Dina in-pen red 'Dan played x-mix-drix with Dina with a red pen' b. Dan sixek iks-miks-driks im Dina be-et adom ve-hi hištamša be-et kaxol. Dan played x-mix-drix with Dina in-pen red and-she used in-pen blue 'Dan played x-mix-drix with Dina with a red pen and she used a blue pen' Finally, Dimitriadis (2004) has suggested that the dyadic reciprocal verb is formed by

49

a specific arity operation. The fact that the discontinuous construction is possible with underived symmetric verbs undermines this proposal. Further, Dimitriadis (2004) and Rákosi (2008) derive the monadic reciprocal verb form its dyadic counterpart by an operation identifying (unifying) the two arguments of the dyadic input. However, as the reader may recall, languages such as Hebrew, Russian, and Hungarian morphologically code arity operations. But they systematically do not employ any morphological marking distinguishing between the monadic and dyadic alternates. This casts doubts on the plausibility of this suggestion. Conclusion In addition to periphrastic reciprocal constructions, languages exhibit verbs that do not govern a reciprocal anaphoric object, but nonetheless express reciprocity. These verbs fall into two types: those that are formed in the lexicon and those whose formation is entirely syntactic. This split is expected by the lex(icon)-syn(tax) parameter (Reinhart and Siloni 2005), which states the arity operations can apply in the lexicon or in the syntax. Unlike periphrastic reciprocal constructions, both types of reciprocal verbs do not avail themselves of the so-called "I" reading, namely, the (noncontradictory) reading of John and Paul said that they defeated each other in the final. This is shown to follow directly from the fact that both types of reciprocal verbs are derived from their transitive alternate by an operation that eliminates a syntactic position in the complement domain, associating the corresponding θ-role with the subject argument. A coherent cluster of cross-linguistic distinctions follows straightforwardly from the different locus of formation of reciprocal verbs. The distinctions concern: (i) whether the reciprocal reading is associated with a singular event (namely, the verb is symmetric) or based on the accumulation of sub-events, (ii) the level of productivity of the operation, (iii) the availability vs. unavailability of reciprocal ECM verbs, (iv) whether or not the verb can be formed from a frozen input, (v) whether or not it allows semantic drift, (vi) whether or not it appears in idioms unavailable for its transitive alternate, (vii) whether or not the verb licenses an accusative argument in case reciprocalization does not target the Theme, (viii) the derivation of reciprocal event nominal, (ix) and whether or not the verb allows the so-called discontinuous construction. Beyond its contribution to the study of the syntax and semantics of reciprocal verbs, the paper has more general theoretical consequences: It shows that the lexicon is an active (operative) component of the grammar, where arity operations can apply. Further, it provides evidence that argument structure (including the external argument) is part of the lexical representation of lexical entries (and not syntactically inserted). Appendix: On the Externality of the subject Since the subject of reciprocal verbs is understood to be associated with two thematic roles, the question arises whether the subject is an internal or external argument. Below I show that the subject of reciprocal verbs fails common tests diagnosing internality in a variety of languages. 45 45 Crosslinguistically, there is a set of symmetric verbs whose subject is an internal argument. These verbs (e.g., the Hebrew hit’arbev ‘mixed(rec)’(i)) show morphology typical of valence reducing

50

As noted by Shlonsky (1997) among others, Hebrew allows two types of "inversion": triggered inversion, which is licensed by some XP immediately preceding the verb [XP V S], and simple inversion, that is, [V S] order. The latter is possible only with subjects that are internal arguments, e.g., subjects of passives (106a) and unaccusatives (106b). Reciprocal verbs do not allow [V S] order (106c) just like other verbs whose subject is an external argument (106d).46 (106) a. Butlu štey harca’ot. (Hebrew) were+cancelled two lectures b. Higi’u šney studentim. arrived two students c. *Hitnašku šney studentim. kissed(rec) two students d. *Avdu šney studentim. worked two students Similarly, possessive datives cannot modify the external argument. As noted by Borer and Grodzinsky (1986), possessive datives, which denote possession in the loose sense (here le-Dan (‘to-Dan’ roughly ‘Dan's’) diagnose internality. They can modify the subject of passive (107a) and unaccusative verbs (107b), but not the subject of unergatives (107c). The subject of reciprocal verbs (107d) patterns with the subject of unergatives.47 (107) a. Ha-pgiša butla le-Dan. (Hebrew) the-meeting cancelled(passive) to-Dan 'Dan’s meeting was canelled.' a. Kol ha-kir hitkalef le-Dan. all the-wall peeled to-Dan operations, and have a transitive alternate that denotes a symmetric relation between the members of its Theme argument (ii). The verbs in this set denote symmetry inherently; their symmetry is not acquired by reciprocalization, as shown by the fact that both the intransitive and transitive alternates denote a symmetric relation. As their lexical meaning is symmetric, they allow the discontinuous construction (iii), as expected. For reasons of space, I do not discuss these verbs any further here (for discussion, see Siloni 2008). (i) Ha-cva’im hit’arbevu. (Hebrew) the-colors mixed ‘The colors mixed’ (ii) Dan irbev et ha-cva’im. Dan mixed the colors (iii) Ha-adom hit'arbev im ha-šaxor. the-red mixed with the-black 46 [V S] order is always ruled out with proper names and pronouns. It diagnoses internality provided that there is no material intervening between the verb and its subject as such an intervention can in certain cases license "inversion" also with unergatives. Further, it is important to note that the sole counterexample to the generalization that strict VS order is impossible with unergatives is the verb tilpen (alias tilfen), which to some extent licenses VS (see Shlonsky 1987). This, however, seems to be a special use of the verb, as suggested by the fact that in this environment it does not allow complements; hence the marginality of tilfen avixa le-dan ('called your father to Dan'). 47 Modification by possessive datives is limited to verbs whose subject is an internal argument provided that the subject is an alienable noun and the possessive dative a lexical noun phrase, not a personal pronoun. Inalienable subjects license possessive datives with unergatives, too. Personal pronouns can be ethical datives, which are also possible with unergatives. The possessee can be neither a proper name nor a kinship noun.

51

'The whole of Dan’s wall peeled.' c. *Ha-studentim kafcu le-Dan. the-students jumped to-Dan d. *Šney studentim hitxabku le-Dan. Two students hugged(rec) to-Dan As already mentioned, the so-called hitpa’el template appears not only with reciprocal verbs, but also with other types of predicates, including unaccusatives, reflexives, and (a few) passives. Alongside the biblical form hitpa’el, literary Hebrew has a post-biblical form nitpa’el (in past tense). Interestingly, for speakers whose grammar includes the form, its distribution is limited. While unaccusatives and passives can use the nitpa’el form, reciprocal verbs just like reflexive verbs cannot. The possibility to use nitpa’el seems to correlate with the type of argument functioning as a subject. Unaccusatives (108a-b) and passives (108c-d), whose subject is an internal argument, allow it, but not reciprocal verbs (108e-f) and reflexive verbs (108g-h), whose subject is an external argument (for more on reflexive verbs, see Reinhart and Siloni 2004). (108) a. Ha-mixnasayim nitkavcu. (Hebrew) the-pants shrank(nitpa'el) 'The pants shrank' b. Ha-sukar nitmoses. the-sugar dissolved(nitpa'el) 'The sugar dissolved' c. Ha-uvdot nitgalu al yedey xoker svecari. the-facts were+discovered(nitpa'el) by researcher Swiss The facts were discovered by a Swiss researcher' d. Ha-yecira nitxabra al yedey malxin carfati. the-work was+composed(nitpa'el) by composer French 'The work was composed by a French composer' e. * Hem nitnašku. they kissed(nitpa'el) f. *Hem nitxabku. they hugged(nitpa'el) g. * Hu nitlabeš. he dressed(nitpa'el) h. * Hu nitgale’ax. he shaved(nitpa'el) In French, verbs whose subject is an internal argument can appear in expletive constructions (109a,c) and typically allow en cliticization out of their subject (109b,d), as en can cliticize only out of internal DP arguments. French reciprocal verbs on a par with their Hebrew equivalents do not pattern with verbs whose subject is an internal argument. They are marginal in expletive constructions (109e) and disallow en cliticization (109f). (109) a. Il est arrivé trois filles hier soir. (French) there is arrived three girls yesterday evening 'There arrived three girls yesterday evening' b. Il en est arrivé trois hier soir . there of+them is arrived three yesterday evening

52

'There arrived three of them yesterday evening' c. Il s’est cassé beaucoup de verres dans ce lave-vaisselle. there SE is broken many glasses in this dishwasher 'Many glasses broke in this dishwasher' d. Il s’en est cassé beaucoup dans ce lave-vaisselle. there SE of+them is broken many in this dishwasher 'Many of them broke in this dishwasher' e. ??Il s’est embrassé beaucoup de filles à cette fête. there SE is kissed many girls in this party f. *Il s’en est embrassé beaucoup à cette fête. there SE is kissed many in this party The subject of reduced relatives whose matrix predicate is the perfect participle must be an internal argument. Hence the predicate can be passive (110a) unaccusative (110b) but not unergative (110c) (see Siloni 1995, 1997). Reciprocal verbs (110d) pattern with unergatives.48 (110) a. L'uomo arrestato dalla polizia è una spia. (Italian) the man arrested by the police is a spy 'The man arrested by the police is a spy' b. Il bicchiere rottosi ieri apparteneva a mio nonno. the glass broken- SI yesterday belonged to my grandfather 'The glass that broke yesterday belonged to my grandfather' c. * L'uomo telefonato a suo nonno è una spia. the man telephoned to his grandfather is a spy d. * I ragazzi baciati-si ieri sono miei alunni. the children kissed-SI are my pupils In Russian, the so-called genitive of negation can be used as a diagnostic for internality. Internal arguments can bear genitive case when the predicate is negated (Pesetsky 1982). Unlike the subject of unaccusatives (111a) and passives (111b), the subject of reciprocal verbs cannot bear genitive case (111d), just like the subject of unergatives (111c).49 (111) a. Ne ob''javilos' studentov. (Russian) NEG showed up students(GEN) 'Students did not show up' b. ?ne moetsja okon. NEG washed(PASS) the windows(GEN) c. *Ne tancevalo studentov. NEG danced students(GEN) d. *Ne obnimalos’ detej. NEG embraced(rec) children(GEN) In sum, crosslinguistic evidence shows that the subject of reciprocal verbs is an external argument. 48 In French the test is not applicable, as participial relatives of this kind disallow clitics altogether. 49 Judgments vary as genitive of negation is not equally productive among speakers.

53

References Alsina, Alex. (1996). The role of argument structure in grammar. CSLI lecture notes

62. Stanford, California: CSLI Publications. Ariel, Mira. (2006). The Making of a Construction: From Reflexive Marking to

Lower Transitivity. Ms. Tel Aviv University. Artstein, Ron. (1997). Group events as means for representing collectivity. In

Benjamin Bruening (Ed.) MIT Working Papers in Linguistics 31: Proceedings of the Eighth Student Conference in Linguistics (pp 41–51). Cambridge, MA.

Borer, Hagit. (1984). Parametric Syntax. Foris, Dordrecht. Borer, Hagit & Grodzinsky, Yosef. (1986). Syntactic vs. Lexical Cliticization: The Case

of Hebrew Dative Clitics. In Hagit Borer (Ed.) The Syntax of Pronominal Clitics (pp.175-217). San Francisco: Academic Press.

Bošković, Željko. (1994). D-structure, θ-criterion and Movement into θ-positions. Linguistic Analysis, 24, 264-286.

Bošković, Željko. (2002). A-movement and the EPP. Syntax, 5, 167-218. Bresnan, Joan. (1982). .The Passive in Lexical Theory. In Joan Bresnan (Ed.) The

Mental Representation of Grammatical Relations (pp. 3-86). Cambridge, MA: MIT Press.

Burzio, Luigi. (1986). Italian Syntax: A Government-Binding Approach. Dordrecht: Reidel.

Carlson, Greg. (1998). Thematic Roles and the Individuation of Events. In Susan Rothstein (Ed.) Events and Grammar (pp. 35-51). Dordrecht: Kluwer Academic Publishers.

Chierchia, Genaro. (2004). A Semantic for Unaccusatives and its Syntactic Consequences. In The Unaccusativity Puzzle, ed. by Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert. 288-331.Oxford: Oxford University Press.

Chierchia, Gennaro & Sally McConnell-Ginet. (1990). Meaning and Grammar. Cambridge, MA: MIT Press.

Chomsky, Noam. (1981). Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam. (1986). Knowledge of Language: Its Nature, Origin and Use. New

York: Praeger. Chomsky, Noam. (1995). The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam. (2001). Derivation by Phase. In Michael Kenstowicz & Ken Hale

(Eds.) A Life in Language (pp. 1-52). Cambridge, MA: MIT Press. Cinque, Guglielmo. (1988). On si Constructions and the theory of arb. Linguistic

Inquiry 19: 521-581. Cinque, Guglielmo. (2002). "Restructuring" and Functional Structure. Ms. University

of Venice. Dimitriadis, Alexis. (2000) Beyond Identity: Topics in Pronominal and Reciprocal

anaphora. Doctoral dissertation, University of Pennsylvania. Dimitriadis, Alexis. (2004). Discontinuous Reciprocals. Ms. Utrecht Institute of

Linguistics OTS. Dimitriadis, Alexis. (2008a). Irreducible Symmetry in Reciprocal Constructions. In.

Ekkehard König and Volker Gast (Eds.) Reciprocals and Reflexives: Cross-linguistic and theoretical explorations. Mouton de Gruyter.

Dimitriadis, Alexis. (2008b). The Event Structure of Irreducibly Symmetric Reciprocals. In Johannes Dolling & Tatjana Heyde-Zybatow (Eds.) Event Structures in Linguistic Form and Interpretation. Berlin: De Gruyter.

Dowty, David. (1991). Thematic Proto-roles and Argument Selection. Language, 67,

54

547–619. Fadlon, Julie. MA thesis (2008) “The Psychological Reality of Hidden Entries”. Tel

Aviv University. Fassi Fehri, Abdelkader. (1993). Issues in the Structure of Arabic Clauses and Words.

Dordrecht: Kluwer Academic Publishers. Frajzyngier, Zygmunt. (1999). Coding the Reciprocal Function: Two Solutions. In

Zygmunt Frajzyngier & Traci S. Curl (Eds.). (pp. 179-194). Amsterdam: John Benjamins.

Frajzyngier, Zygmunt, and Traci S. Curl (eds.). (1999). Reciprocals: Forms and Function. Typological Studies in Language 41. Amsterdam: John Benjamins.

Gast, Volker & Florian Haas (2008). Reflexive and Reciprocal Readings of Anaphors in German and other European Languages. Ms. Free University of Berlin.

Gleitman, Grimshaw, Jane (1982). On the Lexical Representation of Romance Reflexive clitics. In

Joan Bresnan (Ed.) The Mental Representation of Grammatical Relations. Cambridge, MA: MIT Press.

Haspelmath, Martin. (1990). The Grammaticalization of Passive Morphology. Studies in Language, 14:1, 25-72.

Heim, Irene, Howard Lasnik & Robert May. (1991a). Reciprocity and Plurality. Linguistic Inquiry 22, 63-101.

Heim, Irene, Howard Lasnik & Robert May. (1991b). On Reciprocal Scope. Linguistic Inquiry, 22, 173-192.

Higginbotham, James. (1980). Reciprocal Interpretation. Journal of Linguistic Research, 1:3, 97-117.

Higginbotham, James. (1985). On Semantics. Linguisitc Inquiry, 16, 547-593 Horvath, Julia & Tal Siloni. 2008. Active Lexicon: Adjectival and Verbal Passives. In

Sharon Armon-Lotem, Gabi Danon & Susan Rothstein (Eds.) Generative Approaches to Hebrew Linguistics. Amsterdam/Philadelphia: John Benjamins.

Hron, David. (2005). On the Derivation of Czech Reflexive and Reciprocal Nouns. MA thesis. Tel Aviv University.

Hron, David. (2006). On the Derivation of Czech reflexive and Reciprocal Nouns. Proceedings of IATL 22.

Hron, David. (2009). Slavic languages in the Light of Lex-Syn Parameter. Ms. Tel Aviv University.

Hurst, Peter. (2006). The syntax of the Malagasy reciprocal construction: an LFG account. Proceedings of the LFG06 Conference. Available online at http://www.csli-publications.stanford.edu/LFG/11/lfg06hurst.pdf

Kayne, Richard. (1975). French Syntax. The Transformational Cycle. MIT Press, Cambridge, Mass.

Kazenin, Konstantin. (2001). The Passive Voice. In Martin Haspelmath, Ekkehard Köning, Wulf Oesterreicher and Wolfgang Raible (Eds.) Language Typology and Langugae UniversalsI. Berlin: de Gruyter, vol 2, 899-916.

Keenan, Edward and Jean-Paulin Razafimamonjy. (2001). Reciprocals in Malagasy. In Torrence, Harold (ed.) 2001.Papers in African Linguistics 1, UCLA Department of Linguistics, 6. UCLA working papers in linguistics.

Kemmer, Suzanne. (1993). The Middle Voice. Typological Studies in Language, 23. Amsterdam: John Benjamins.

Kim, Yookyung & Stanley Peters. (1998). Semantic and Pragmatic Context-Dependence: The Case of Reciprocals. In Pilar Barbosa, Danny Fox, Paul

55

Hagstrom, Martha McGinnis & David Pesetsky (Eds.) Is the Best Good Enough (pp. 221–248). Cambridge, MA: MIT Press.

Komlósy, András. (1994). Complements and Adjuncts. In Ferenc Kiefer & Katalin É. Kiss (Eds.) The Syntactic Structure of Hungarian. Syntax and Semantics, 27, 91-178. San Diego: Academic Press.

Krifka, Manfred. (1992). A Compositional Semantics for Multiple Focus Constructions. In Joachim Jacobs (Ed.) Informationsstruktur und Grammatik (pp. 17-53). Opladen: Westdeutscher Verlag.

Krifka, Manfred. (1998). The Origins of Telicity. In Susan Rothstein (Ed.) Events and Grammar (pp. 197–235). Kluwer Academic

Landman, Fred. (1989). Groups I. Linguistics and Philosophy, 12:5, 559–605. Landman, Fred. (2000). Events and Plurality. Dordrecht: Kluwer. Langendoen, D. Terence. (1978). The logic of Reciprocity. Linguistic Inquiry, 9, 177-

197. Langendoen, D. Terence. (1992). Symmetric Relations. In Diane Brentari, Gary N.

Larson and Lynn A. MacLeod (Eds) The Joy of Grammar. Amsterdam: John Benjamins.

Langendoen, D. Terence and Joël Magloire. (2003). The Logic of Reflexivity and Reciprocity. In Andrew Barss (Ed.) Anaphora: a Reference Guide (pp, 237-263). Oxford: Blackwell Publishers.

Lasersohn, Peter. (1992). Generalized Conjunction and Temporal Modification. Linguistics and Philosophy, 15, 381-410.

Lasersohn, Peter. (1999). Pragmatic Halos. Language 75:3, 522-551. Levin, Beth and Malka Rappaport- Hovav. (1995). Unaccusativity. Cambridge, Mass.:

MIT Press. Link, Godehard. (1998). Algebraic Semantics in Language and Philosophy. Stanford:

CSLI Publications. Marelj, Marijana. (2002). Rules that Govern the Occurrences of Theta-clusters in the

Theta-System. Theoretical Linguistics 28, 357-373. Marelj, Marijana. (2004). Middles and Argument Structure across Languages.

Doctoral dissertation. OTS, University of Utrecht. Mchombo, Sam A. (1991). Reciprocalization in Chichewa: A lexical account. Linguistic Analysis, 21, 3-22. Miyagawa, Shigeru & Maria Babyonyshev. (2004). The EPP, Unaccusativity, and the

Resultative Constructions in Japanese. Scientific Approaches to Language No. 3. Center for Language Sciences. Kanda University of International Studies, Kanda, Japan.

Papangeli, Dimitra. (2004). The Morphosyntax of Argument Realization. Doctoral dissertation. OTS, University of Utrecht.

Parsons, Terence. (1990). Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge, MA: MIT Press.

Pesetsky, David. (1982). Path and Categories. Doctoral dissertation. Cambridge, MA: MIT Press.

Pesetsky, David. (1995). Zero Syntax: Experiencers and Cascades. Cambridge MA: MIT Press.

Rákosi, György. (2003). Comitative arguments in Hungarian. In Uil-OTS yearbook (pp. 47-57). Utrecht Institute of Linguistics.

Rákosi, György. (2008). The Inherent Reflexive and the Inherent Reciprocal Predicate in Hungarian: Each to Their Own Argument Structure. In. Ekkehard König and

56

Volker Gast (Eds.) Reciprocals and Reflexives: Cross-linguistic and theoretical explorations. Mouton de Gruyter.

Rappaport-Hovav, Malka, and Beth Levin. (1998). Morphology and Lexical Semantics. In Andrew Spencer and Arnold M. Zwicky (Eds.) The Handbook of Morphology. Oxford: Blackwell Publishers.

Reinhart, Tanya. (2002). The Theta System: an Overview. Theoretical Linguistics, 28, 229-290.

Reinhart, Tanya & Eric Reuland. (1993). Reflexivity. Linguistic Inquiry, 24:4, 657-720.

Reinhart, Tanya and Tal Siloni. (2004). Against the Unaccusative Analysis of Reflexives. In Artemis Alexiadou, Elena Anagnostopoulou & Martin Everaert (Eds.) The Unaccusativity Puzzle (pp. 288-331). Oxford: Oxford University Press.

Reinhart, Tanya and Tal Siloni. (2005). The Lexicon-Syntax Parameter: Reflexivization and other Arity Operations. Linguistic Inquiry 36:3, 389-436.

Rivero, Maria Luisa and Milena Milojević Sheppard (2003). Indefinite Reflexive Clitics in Slavic: Polish and Slovenian.In: Natural Language and Linguistic Theory, 21:1, 89-155

Rizzi, Luigi. (1978). A Restructuring Rule in Italian Syntax. In S. Jay Keyser (Ed.) Recent Transformational Studies in European Languages (pp. 113-158). Cambridge, MA: MIT Press. (also in Rizzi 1982)

Rizzi, Luigi .(1982). Issues in Italian Syntax, Dordrecht: Foris. Rothstein, Susan. (2000). Secondary Predication and Aspectual Structure. In Ewald

Lang, David Holsinger, Kerstin Schwabe & Oliver Teuber (Eds.) ZAS Papers in Linguistics (pp. 241-264). Zentrum für Allgemeine Sprachwissenschaft, Sprachtzpologies und Universalienforschung.

Rubinstein, Aynat. (2007). Groups in the Semantics of Reciprocal Verbs. Talk presented at "Reciprocals cross-linguistically". Freie Univerität Berlin.

Shlonsky, Ur. (1997). Clause Structure and Word Order in Hebrew: An Essay in Comparative Semitic Syntax. Oxford: Oxford University Press.

Siloni, Tal. (1995). On Participial Relatives and Complementizer D0: a Case Study in Hebrew and French. Natural Language and Linguistic Theory, 13:3, 445-487.

Siloni, Tal. (1997). Noun Phrases and Nominalizations: The Syntax of DPs. Dordrecht: Kluwer Academic Publishers.

Siloni, Tal (2001). Reciprocal Verbs. In Yehuda N Falk (Ed.), Proceedings of the Israel Association of Theoretical Linguistics. Online publication at http://atar.mscc.huji.ac.il/~english/IATL/17/.

Siloni, Tal. (2002). Active Lexicon. Theoretical Linguistics, 28, 383-400. Steinbach, Markus. (1998). Middles in German. Doctoral Dissertation, Humboldt

University, Berlin. Siloni, Tal. (2008). The syntax of reciprocal verbs: An overview. In Reciprocals and

reflexives: Cross-linguistic and theoretical explorations, eds. Ekkehard Konig and Volker Gast. Berlin: Mouton de Gruyter. 451-498.

Wehrli, Eric. (1986) On some properties of French Clitic se. In Hagit Borer (Ed.) The Syntax of Pronominal Clitics, 175-217, San Francisco: Academic Press.

Williams, Edwin. (1981). Argument Structure and Morphology. The Linguistic Review 1, 81-114.

Williams. Edwin. (1991). Reciprocal Scope. Linguistic Inquiry 22, 159–73. Zec, Draga. (1985). Objects in Serbo-Croatian. In Mary Niepokuj, Mary Van Clay,

Vassiliki Nikiforidou, and Deborah Feder (Eds.) Proceedings of the 11th Annual Meeting of the Berkeley Linguistics Society (pp. 358–371).