Portal de Periódicos da CAPES

Symmetric Reciprocal Semantics as a Predictor of Partial Control

2018; The MIT Press; Volume: 49; Issue: 2 Linguagem: Inglês

10.1162/ling_a_00276

1530-9150

J.‐Marc Authier, Lisa A. Reed,

Linguistics and Discourse Analysis

Wilkinson (1971) and Lawler (1972) originally observed the phenomenon of partial control (PC). Descriptively, PC refers to situations in which the reference of PRO must include that of an overt argument in the matrix clause, but is not exhaustively determined by that argument. The effects of PC are best observed in sentences like (1a), which contain an infinitival whose predicate is unambiguously collective (i.e., one that requires, rather than just allows, its subject to denote a plural entity; cf. (1b)).Most of the research on PC has focused on the (semantic) properties of those matrix predicates that license the phenomenon (see, e.g., Landau 2000, White and Grano 2013 for a survey and experimental data, as well as Pearson 2016). One notable exception to this trend is provided by Sheehan (2012, 2014), who observes that PC in European Portuguese, Italian, Spanish, and French displays a selective availability based on what kind of collective predicate appears in the embedded infinitival containing PRO.1 Specifically, she points out that the descriptive generalization in (2) seems to hold.Thus, French se réunir 'meet/gather' can take an overt comitative argument but s'embrasser 'kiss/hug' cannot, and, as a result, only the former can occur in a PC infinitival.2Since the generalization in (2) does not seem to apply to English (see the English glosses in (4)), Sheehan calls examples like (3b) instances of "fake PC" and argues that this distinct phenomenon arises indirectly from a silent comitative phrase present in the infinitival complement (as Hornstein (2003) and others have proposed is the case in all languages). Landau (2016a) convincingly shows, however, that Sheehan's analysis of PC in Romance is untenable by pointing out that elements syntactically and/or semantically associated with overt comitatives are systematically unavailable with PC complements. For example, while an adverb like séparément 'separately' can modify an overt comitative (5a), it fails to occur in those PC complements alleged to have a null comitative structure (5b).We add to Landau's arguments one of our own, one that is based on the observation, due to Dimitriadis (2004), that the semantics of simple reciprocals enriched with a comitative phrase (also called discontinuous reciprocals) is more specific (or expressive) than that of their corresponding simple reciprocals. Consider in this respect the paradigm in (6).As Dimitriadis points out, a sentence like (6a) describes a quarreling event involving Eric, Nadine, and someone else with no specification as to who was in conflict with whom. The interpretation of the discontinuous reciprocal construction in (6b), on the other hand, is more specific in that it expresses a reciprocal relation between pairs consisting of one participant (possibly plural) taken from the denotation of the subject and another participant taken from the denotation of the comitative phrase. Thus, (6b) is either about a disagreement involving Eric and Nadine versus someone else, or about two separate conflicts, one involving Eric versus someone else and another involving Nadine versus that someone else. Consider next the sentence in (7) on the PC reading symbolized by the indices.On the assumption that (7) contains a null comitative phrase, we expect the interpretation of that sentence to be akin to that of (6b) in that it should involve pairs whose first member corresponds to the denotation of the controller (i.e., Eric and Nadine) and whose second member corresponds to the denotation of the null comitative. This expectation is not fulfilled, however. Indeed, the PC reading associated with (7) can only be paraphrased as follows: Eric and Nadine remember that there was a quarreling event involving Eric, Nadine, and some other unspecified individual(s) with no specification as to who was in conflict with whom. The fact that (7) cannot have the more specific reading tied to the presence of an overt comitative phrase in (6b) thus provides an additional argument against the presence of a null comitative in PC contexts.Landau's (2016a,b) analysis of PC, which does not posit a null comitative phrase but takes PRO in PC to be a group-denoting, syntactically singular but semantically plural noun, is not (directly) challenged by the facts in (5) and (7). However, given the untenable character of Sheehan's analysis of Romance "fake PC" demonstrated above, neither Landau's account of PC nor its alternatives can explain why the selective availability of PC exhibited by Romance embedded predicates is not also present in English (see, e.g., (4b) vs. (1a)).In this squib, we will claim that (a) there is no such thing as "fake PC"; that is, there is only one kind of PC, and this phenomenon is subject to the same conditions in English as it is in Romance; (b) one such condition is that the embedded collective predicate have (irreducibly) symmetric reciprocal semantics in the sense of Siloni (2002, 2012) and Dimitriadis (2004, 2008); and (c) the difference between English and Romance boils down to the facts that only reciprocals formed in the lexicon introduce symmetric semantics and that the set of reciprocals formed in the lexicon in English and that formed in the lexicon in Romance are not identical. The assumption that reciprocal verbs can be formed in different components of the grammar—namely, the lexicon and the syntax—is based on a coherent cluster of distinctive properties (such as degree of productivity) first uncovered by Siloni (2001) and further substantiated by Reinhart and Siloni (2005), who capture the split via a Lex(icon)-Syn(tax) parameter. This parameter states that Universal Grammar allows valence-changing operations to apply either in the lexicon or in the syntax. Following Siloni (2002), we will argue that only lexical reciprocals are symmetric and that French, being a language with a syntactic setting of the Lex-Syn parameter, only has isolated instances of lexical reciprocal verbs. This, as we will demonstrate, is the reason why French has a much more restricted set of PC-inducing embedded predicates than English.English has a restricted set of covert reciprocals such as kiss, meet, hug, embrace, nuzzle (but not, e.g., love, elect, beat, follow) that bear no special morphology and are therefore idiosyncratically marked as reciprocals in the lexicon. Covert reciprocals like (8a) have periphrastic counterparts, which involve pairing the verb with an object reciprocal anaphor (8b).3It has been known for some time, however, that English lexical and periphrastic reciprocal constructions are not semantically equivalent (see in particular Leonard and Goodman 1940, Langendoen 1978, and references cited there). That is, in English, only covert lexical reciprocals are symmetric in the sense made explicit in (9), adapted from Siloni 2002, 2008, 2012 and Dimitriadis 2004, 2008.While (8b) is vague in being able to refer to an atomic kissing event (with simultaneous participation of Ron and Sally) or to the accumulation of separate kissing events (e.g., Ron kissed Sally on the forehead and then she kissed him on the cheek), (8a) can only refer to a single event with symmetric participation, that is, simultaneous kissing on the lips. This can be shown clearly by means of a test originally developed by Siloni (2002). This test uses count adverbials of the x times type that quantify over the number of events a sentence may denote. Thus, in an English periphrastic reciprocal construction like kiss each other, which is nonsymmetric (i.e., vague with respect to symmetry), a count adverbial can count either the total number of atomic events or the number of subevents attributable to each participant. This is shown in (10a), a sentence that can be interpreted in the two ways expressed by the glosses. However, with the symmetric lexical reciprocal kiss, we can only count events once; hence, (10b) does not display the same counting ambiguity (10a) does.Interestingly, as (11) illustrates, only covert reciprocals can participate in PC in English.This suggests that (a) only those reciprocal verbs that have symmetric semantics are compatible with PC PRO and (b) as originally argued by Siloni (2002) and Dimitriadis (2004), only verbs that have acquired a reciprocal meaning through a lexical (not syntactic) operation have symmetric semantics. If these generalizations are on the right track, we then expect them to extend to (e.g.) French and possibly explain Sheehan's observation in (2).Before we turn to the French data, however, we must briefly discuss Landau's (2000) alternative account of the ungrammaticality of the periphrastic reciprocal option (11). According to Landau, this option is ruled out because (a) the binder of a plural anaphor like each other must be syntactically plural and (b) PRO in PC contexts, being akin to a group-denoting noun like committee, is syntactically singular (though semantically plural) and cannot therefore serve as a binder for the plural anaphor. This approach appears to make a problematic prediction, however. As is well-known, group-denoting nouns can, in those dialects of English spoken in the United Kingdom, serve as antecedents for plural anaphors. This is illustrated in (12).The prediction is therefore that in such dialects, the fact that group-denoting nouns can be syntactically plural should correlate with PC PRO being able to antecede plural anaphors. The data collected from our informants (Nigel Duffield and Ian Roberts) do not, however, support the existence of such a correlation. Indeed, both speakers accept examples like (12) but reject examples like (11) (with the reciprocal anaphor) and (13).4On the other hand, as we will see shortly, there is at least one good reason to adopt the hypothesis that only those reciprocal verbs that have symmetric semantics are compatible with PC PRO: namely, that it can account for the crosslinguistic differences in PC licensing between English and Romance uncovered by Sheehan (2012, 2014).Turning now to French, the only natural way to express (8) is to combine the reciprocal morpheme se with transitive embrasser 'kiss' as in (14).5We first note that a se morpheme with reciprocal meaning can be added to just about any transitive verb, and the productivity of this process immediately suggests that it is syntactic in nature. Furthermore, semantically, the sentence in (14) is akin to (8b) rather than (8a); that is, it is unmarked with respect to symmetry and can therefore describe a nonsymmetric situation, as the possible rejoinder in (15) makes clear.This is further confirmed by the fact that the reciprocal verb s'embrasser gives rise to the sort of counting ambiguities that arise with the English periphrastic reciprocal kiss each other, as (16) shows.These observations are, in fact, part of a much larger pattern first uncovered by Siloni (2001) and further substantiated by Reinhart and Siloni (2005) and Siloni (2008, 2012). That is, setting periphrastic reciprocals aside, reciprocal verbs exhibit a number of linguistic properties that cluster in such a way that they can be split into two groups crosslinguistically. In languages like French, Spanish, Serbian, and so on, reciprocalization is a highly productive operation that can target exceptional-case-marking (ECM) predicates (17a) and allows direct objects (17b).On the other hand, in languages like English, Russian, Hebrew, and so on, reciprocalization is restricted to a relatively small set of verbs, does not cooccur with direct objects, and never targets ECM predicates. Reinhart and Siloni's (2005) Lex-Syn parameter ascribes these divergent patterns to the fact that (nonperiphrastic) reciprocalization is a valence-reducing operation whose locus is either the syntax (in, e.g., French, Spanish, Serbian) or the lexicon (in, e.g., English, Russian, Hebrew).Further, as Dimitriadis (2004) and Siloni (2008, 2012) argue, when reciprocalization applies in the lexicon, it does not have access to subevents resulting from the interaction of a verbal predicate and its arguments, as this type of interaction is syntactically determined. Thus, lexical reciprocals, being denotationally restricted to atomic events, can only convey reciprocal meaning by being symmetric. Syntactic reciprocalization, on the other hand, is an operation that results from merging two distinct elements taken from the lexicon: a verb and the appropriate morphology (se in French). The verb in question must, of course, be a two-place predicate, given the valence-reducing character of the operation, which ties in with Siloni's (2012) observation that there are no instances of syntactic reciprocal verbs lacking a transitive alternate.Of course, when reciprocalization applies in the lexicon, it too originally affects a two-place predicate. However, the reciprocal lexical entry thus formed can, like any lexical item, undergo semantic drift over time, acquiring a meaning distinct from the transitive alternate it was originally based on. For example, the English lexical reciprocal kiss can only mean 'kiss on the lips' and therefore has a meaning that is more restricted than that of its transitive counterpart (and that of its periphrastic reciprocal counterpart). Similar examples are found in French. Reinhart and Siloni (2005:417n24) point out that in syntactic reciprocal languages like French, there can be instances of lexical reciprocals that constitute a closed set of verbs. Two such cases are discussed by Siloni (2002). These are the semantically ambiguous reciprocal verbs se battre 'beat each other or quarrel' and s'entendre 'hear each other or agree/get along'. The ambiguity, Siloni argues, comes from the fact that such verbs can instantiate syntactic reciprocals with a compositional meaning that is nonsymmetric (i.e., beat each other, hear each other) as well as lexical reciprocals with an idiomatic meaning that is symmetric (i.e., quarrel, agree/get along with each other). As expected, the transitive counterparts of such verbs only match the meaning of the corresponding syntactic reciprocals. This is illustrated in (18) and (19).Assuming that PC requires the embedded collective predicate to be symmetric and that only reciprocals formed in the syntax can be nonsymmetric, we then expect English kiss to mean only 'kiss on the lips' when it appears in a PC infinitival because only the lexical reciprocal version of kiss is symmetric. This prediction is borne out: a sentence like (20) can only mean that Eric wanted for himself and some other party to be involved in an atomic event of kissing on the lips; it cannot mean that Eric wanted for himself and some other party to exchange kisses on different parts of their bodies.Similarly, if the ambiguous reciprocal French verbs in (18a) and (19a) are used in PC infinitives, only the symmetric idiomatic interpretation of these verbs is available, as shown in (21) and (22).Finally, as Siloni (2008) observes, the main factor that determines the availability of a discontinuous reciprocal construction (i.e., a reciprocal with a comitative argument) is whether the reciprocal predicate it is based on is a lexical entry rather than the output of a syntactic operation.6 We illustrate this observation with the ambiguous s'entendre 'hear each other or agree/get along', which, as (23b) shows, can only have the symmetric idiomatic interpretation when used in the discontinuous construction.Thus, only lexical reciprocals, which entail a symmetric event, may partake in the discontinuous reciprocal construction and, since lexical reciprocals are the only reciprocals that can be predicated of PC PRO, we end up with Sheehan's (2012, 2014) descriptive generalization in (2), repeated here as (24).The overall picture that emerges for French is that partial control is only possible with symmetric embedded reciprocals, those being a rather small subset of reciprocal predicates that license the discontinuous construction and do not have a transitive counterpart (at least not one with the same denotation). This small subset includes the following verbs:7As stated earlier, the majority of the present approaches to PC focus on predicting the class of control (matrix) verbs that create the proper environment for PRO to take on a PC reading. It is therefore unclear how such approaches could handle the symmetry constraint on those reciprocal predicates that take PC PRO as their subject discussed in this squib.8 One potential exception is the theory of PC proposed by Landau (2016a,b). Landau's theory takes PRO in PC to be a group-denoting, syntactically singular, but semantically plural noun, and it therefore seems worthwhile to explore the question of whether this characterization of PC PRO suffices to explain the fact that it can only be the subject of symmetric reciprocal predicates. In what follows, we discuss evidence suggesting that this is not the case.An important distinction between group nouns and PC PRO was already pointed out in section 2: in (at least some) English dialects spoken in the United Kingdom, group-denoting nouns can bind plural anaphors in the so-called periphrastic reciprocal construction (see (26a)) and are therefore compatible with nonsymmetric reciprocals. This, however, does not correlate with PC PRO being able to antecede plural anaphors (see (26b)).Thus, within the same English dialect, a group noun, but not PC PRO, can be the subject of a nonsymmetric reciprocal.Group nouns and PC PRO differ in a similar way when we examine their compatibility with those nonsymmetric reciprocal predicates that are not periphrastic reciprocals. Recall that PC PRO can only be predicated of symmetric reciprocal predicates—that is, predicates that always yield an atomic event interpretation. Group nouns, on the other hand, are not subject to this constraint: they are compatible with nonsymmetric reciprocal predicates on both of their interpretations, as the examples in (27), found on the Internet, make clear. The predicate s'envoyer des messages 'send messages to each other' in (27b) is a nonsymmetric reciprocal predicate that takes the group noun couple as its subject and is incompatible with an atomic event interpretation (i.e., it can only be understood as the accumulation of message-sending subevents). In (27a), on the other hand, the nonsymmetric reciprocal s'embrasser 'kiss', predicated of the same group noun, is compatible with both an accumulation-of-subevents and an atomic-event interpretation.It thus appears that PC PRO in French does not allow nonsymmetric predicates, while group nouns do. A reviewer suggests the following possible explanation. The semantic plurality of group nouns is clear from their intrinsic meaning and therefore this type of semantic content can, to some extent, license syntactic plurality, a necessary ingredient for distributivity.9 PC PRO, on the other hand, is an empty category whose semantic plurality is determined by a covert operator.10 It therefore has no intrinsic semantic content that can license syntactic plurality and consequently distributivity. This approach, which is based on the observation that PC PRO is not distributable (see, e.g., Landau 2016a) does not, however, predict the fact that a sentence like (4b), repeated here as (28), remains ungrammatical under any interpretation.To explain, the French syntactic reciprocal s'embrasser is nonsymmetric; that is, it is compatible with both an accumulation-of-subevents reading, which arguably involves distributivity, and an atomic-event reading, which does not. It therefore ought to be possible to interpret (28) in exactly the same manner as its English translation. This possibility is, however, ruled out: only those reciprocal predicates that are symmetric—that is, those whose reciprocality is determined in the lexicon—license PC PRO. We therefore conclude that the constraint uncovered in this squib is distinct from Landau's distributivity constraint and will have to follow from other considerations. We leave this as an open issue for future research.We gratefully acknowledge the helpful comments and suggestions made by Benjamin Bruening, Idan Landau, Tal Siloni, Dominique Sportiche, the LI reviewers, the Squibs and Discussion editors, and the audiences at the 41st Penn Linguistics Colloquium and the 47th Linguistic Symposium on Romance Languages hosted by the University of Delaware. We also wish to thank our informants: Nigel Duffield and Ian Roberts for (British) English and Simon Cottart, Morgane Haesen, Johann LeGuelte, Marie Paillard, and Timothée Valentin for French. All remaining errors are, naturally enough, our sole responsibility.