Interpretive asymmetries between null and overt PRO in complement and adjunct infinitives in (Colombian) Spanish

Cross-linguistically, control complement clauses have been reported to allow overt pronominal subjects displaying the diagnostics of obligatory control (‘Overt PRO’; see Livitz 2011; Mensching 2000; Szabolcsi 2009). Building on Gómez (2017), we extend the empirical range of the overt PRO phenomena to para-finality adjunct clauses in (Colombian) Spanish. We show that the controlled subject of para-infinitives —be it null or overt PRO— has the same distribution and interpretive properties as that of complement infinitives. We bring to light unexpected asymmetries in the interpretive properties of overt vs. null PRO which we dub the Overt vs. covert PRO paradox: while they both only allow a bound variable reading under ellipsis, overt PRO, unlike null PRO, also allows a 2 Isogloss 2022, 8(2)/5 Kryzzya Gómez, Maia Duguine & Hamida Demirdache coreferential reading under association-with-focus. Here again, the data are identical in complement vs. adjunct control clauses. We account for this paradox with the Anaphor Generalizations, which state that (i) both overt and null anaphors must be syntactically bound, and (ii) while null anaphors must, overt anaphors can but need not be semantically bound. We further show how these generalizations can be extended to account for similar patterns of interpretation with English/French inherently vs. overtly reflexivized predicates.


Introduction *
The literature on obligatory control in complement clauses has observed the existence of overt controlled pronominal subjects that alternate with null PRO. This alternation is illustrated in (1)-(4): (  Barbosa (2009:103-104)) * We would like to thank to all survey respondents for their judgements and two anonymous reviewers for their useful comments. This work was partially funded by the following projects: AThEME (Advancing the European Multilingual Experience) funded by the European Seventh Framework Programme for research, technological development and demonstration grant agreement no 613465, Région des Pays de la Loire; UV2 ANR18-FRAL0006 (ANR-DFG); PGC2018-096870-B-I00 (MICINN & EAI); IT769-13 (Eusko Jaurlaritza); BIM ANR17-CE27-0011 (ANR). Szabolcsi (2009) and Livitz (2011Livitz ( , 2014 show that these pronominal subjects display properties of obligatory control (OC) PRO, and can therefore be characterized as the overt counterpart of null PRO, so called 'overt PRO' (see also Mensching (2000, pp. 61-62) Alonso-Ovalle & D'Introno (2000), Barbosa (2009Barbosa ( , 2018, and Herbeck (2015Herbeck ( , 2018). Barbosa (2009Barbosa ( , 2018 further points out that languages with overt PRO seem generally to be pro-drop languages. The existence of overt controlled pronominal subjects raises important issues for the theory of Control. 1 This paper sets out to contribute novel empirical and theoretical generalizations to these issues. First, we extend the empirical range of these phenomena to adjunct clauses in Colombian Spanish. We provide a detailed study of the properties of overt vs. null subjects in a subtype of infinitives introduced by the preposition para, namely para-finality clauses (or para-adjuncts for short) according to the terminology of Galán Rodriguez (1999), María Luisa Hernanz (1999, Pérez Vázquez (2007), and Schulte (2007).
Second, we carry out a detailed comparative analysis of the distributive and interpretive properties of overt vs. null PRO across both complement and para-finality infinitival clauses in final position.
In Pérez Vázquez's (2007) typology, para-finality clauses are adjuncts that serve to express an intention motivating the action described by the main clause. Crucially, they allow controlled overt pronominals -i.e., overt PROs-in subject position, in alternation with PRO. Gómez (2017) was to our knowledge the first to systematically test the availability and the interpretation of overt PRO in para-finality adjunct clauses in Colombian Spanish with an experimental study (see also Gómez (in progress)). 2 Obligatory control in para-adjuncts with both overt and null PRO is illustrated in (5) and (6) There is a rich literature on the properties of overt and null subjects of infinitives in Spanish. A broad comparative analysis of the different types of infinitives is beyond the scope of this paper. See however Gómez, Duguine, & Demirdache (to appear) (and references therein) for a systematic exploration of the properties of overt and null subjects of non-finite adjuncts headed by the prepositions para, sin and al, where we observe important variation in the properties of both overt and null subjects. 2 The experimental study carried out in Gómez (in progress) involved 34 native speakers from Colombia, 1 speaker from Bolivia and 1 from Mexico, all of them native speakers of Spanish. The goal was to test the interpretation of overt pronominal subjects in three different types of adjunct clauses, using the distribution of sloppy and strict readings to establish whether control is obligatory or not. The results confirmed, in particular, that the overt pronominal subject in parainfinitives is obligatorily controlled (e.g. it only allows sloppy construals), thus patterning like the null subject of para-infinitives, which likewise was only accepted on a sloppy reading (see section 2 below for further discussion).
We apply obligatory control diagnostics to both overt and null PRO and in both complement and para-adjunct clauses to probe their distribution and interpretation. We will show that null and overt PRO subjects have identical syntactic properties across both complement and para-adjunct clauses. In a nutshell, they are obligatorily c-commanded by a local controller. We uncover, however, a surprising asymmetry in the interpretive properties of null vs. overt PRO which, importantly, holds alike across both complement and para-adjunct infinitival clauses.
Obligatory controlled PRO is expected to allow only bound variable (BV) construals under the two canonical tests for pronominal interpretation: ellipsis and associationwith-focus (where the controller for PRO is associated with a focus particle such as only or even) -see Hornstein (1999) and Landau (2000Landau ( , 2013 and references therein. Here we show that both null and overt PRO behave as obligatory controlled under the ellipsis test. However, overt PRO contrasts with null PRO in not behaving as obligatory controlled under the association-with-focus test, since it allows for coreferential readings (in addition to the expected BV reading). We dub these asymmetrical patterns of interpretation the overt vs. covert PRO paradox: (7) Overt vs. Covert PRO paradox (i) Both null and overt PRO only allow BV interpretations under the ellipsis test.
(ii) Overt PRO, unlike null PRO, also allows coreferential interpretations under the association-with-focus test.
Why do null PRO and overt PRO pattern differently (yield conflicting results) with respect to the two standard tests for pronominal interpretation? In particular, why does overt PRO, unlike null PRO, also allows coreference interpretations, and why so only under the association-with-focus test, but not under the ellipsis test?
Importantly, we further show that the conflicting patterns of interpretations that arise across the two types of control clauses is not an exclusive property of overt referentially dependent pronouns surfacing in control constructions, but more generally of overt SELF-anaphors such as himself in English and se in French. As pointed out by Büring (2005) and Sportiche (2014, pp. 6-7), SELF-anaphors only allow sloppy identity (BV) readings in contexts of VP-ellipsis (VPE). However, in contexts of association-with-focus, they allow both strict (coreferential) and sloppy readings, as illustrated by the paradigm in (8) No one other than Pierre shaves Pierre.
We show that these facts follow naturally from the Anaphor Generalizations in (10), which appeal to a distinction between syntactic vs. semantic binding, as advocated by Büring (2005)  (10) The Anaphor Generalizations: (i) Both null and overt anaphors need to be syntactically bound.
(ii) Overt anaphors can be semantically bound, null anaphors must be semantically bound.
The paper is structured as follows. Section 2 establishes the properties of overt and null subjects in both complements and para-adjunct clauses, on the basis of the diagnostic tests for obligatory control identified in the literature, converging on the paradox put forth in (7). Section 3 develops an analysis in terms of the Anaphor Generalizations in (10), showing that the latter can account not only for the contrast between null PRO vs. overt PRO, but can also be extended to overt SELF-anaphors. Section 4 concludes the paper, drawing a parallel between the asymmetries found with null vs. overt PRO and those found with inherently vs. overtly reflexivized predicates in English and French.

A double paradox in complement and adjunct control infinitivals
The goal of the present section is to explore the distributive and interpretive properties of null vs. overt pronominal subjects in nonfinite complement clauses, as well as in nonfinite para-adjuncts in (Colombian) Spanish. 4 4 It has sometimes been claimed in the literature that the overt pronominals illustrated in (1)-(6) are not the genuine subject of the infinitival clause, but are rather doubling elements modifying the null PRO subject (see for example Piera 1987;Suñer 1986;Torrego 1996 on Spanish). More recently however, scholars have provided evidence showing that such pronominals are the actual subjects of the clause (see Alonso-Ovalle & D'Introno 2000; Barbosa 2009Barbosa , 2018Duguine 2013;Mensching 2000;Szabolcsi 2009). We offer below an argument from Spanish (inspired by Barbosa (2009;): overt pronominals in Spanish cannot double any DP, including in particular preverbal derived (unaccusative/passive) subjects (i), unlike English adjunct anaphors which can modify any type of argument DP (ii). The contrast between ungrammatical (i) and grammatical (iv) follows on the assumption that the overt pronominal in (iv) is not a doubling element, but a postverbal subject, on a par with (iii). To this effect, we apply to the null and overt pronominal subjects of both types of infinitival clauses a set of criteria developed and widely used in the literature for identifying obligatory control (OC) vs. non obligatory control (NOC) (cf. Baltin, Déchaine, & Wiltschko 2015;Hornstein 1999;Landau 2000Landau , 2013Williams 1980 No one else claimed that Peter is the winner. Obligatorily controlled PRO must be c-commanded by an antecedent (11a), in a local dependency (11b). It also allows only sloppy interpretations under the two standard tests for bound variable vs. coreferential interpretations -that is, ellipsis (Lebeaux, 1985, p. 351) and association-with-focus (Jerry, 1975)-, as shown in (11c) and (11d) respectively.
We follow below these criteria to determine whether null and overt nonfinite subjects display obligatory control in Spanish complement and para-adjunct control clauses. 5 Note moreover that Barbosa shows with focus tests that infinitival postverbal pronominal subjects such as the one in (iv) must be interpreted inside the infinitival clause (as opposed to the matrix). 5 Perez Vazquez (2007) divides adjunct clauses headed by the preposition para in three classes: concessives, consecutives and finality/finals. Concessives allow referentially free postverbal subjects, consecutives do not accept overt subjects, and finals only allow referentially dependent overt pronominal subjects (see also María L. Hernanz (1982, p. 413)). As pointed out by a reviewer, Torrego (1998) gives the following piece of data with an overt non-pronominal subject, in what we would classify as a finality/final para-adjunct under Perez Vazquez' (2007)  Results reveal a striking parallelism between these two types of subordinate clauses. We go on to highlight, however, a surprising set of interpretive asymmetries -between overt vs. null dependent subjects on the one hand, and between contexts of ellipsis vs. contexts of association-with-focus on the other hand-which we formulate as the "Overt vs. covert PRO paradox".

Obligatory control criteria with overt and null subjects
We saw in section 1 that both complement and para-adjunct nonfinite clauses allow null, as well as overt, referentially dependent pronominal subjects. We apply below the criteria for distinguishing OC vs. NOC to these two types of pronominal subjects. It appears that both types satisfy the criteria for OC (12)-(17), except in one unexpected instance (18). First, c-command by the matrix controller is mandatory for both null PRO subjects ((a) examples) and overt PRO subjects ((b) examples), be it in complement (12) or paraadjunct clauses (13) The possibility of having a referentially free overt DP in a para-finality clause is unexpected under Perez Vazquez' typology. Moreover, there appears to be some variation since none of the native speakers we consulted accept this sentence as grammatical. Interestingly, however, the subjects of the para-adjunct and the matrix clause are anaphorically linked in (i), just like in control constructions. We leave the exploration of these issues for future research. 6 Spanish does not admit VP-ellipsis, but allows ellipsis of larger structures (Dagnac, 2010;Saab, 2010 María left in order for Juan to be happy. Recapitulating: overt and null subjects pattern exactly alike in the above contexts. They satisfy the first three criteria for obligatory control (as established in (11)) in both types of subordinate clauses. These results confirm the hypothesis that the null subject in the (a) examples is OC PRO, and the pronominal subject in the (b) examples is its overtly realized counterpart 'overt PRO' (Mensching 2000;Ovalle and D'Introno (2001), Livitz 2011Livitz , 2014Herbeck 2015Herbeck , 2018.
Importantly, however, null PRO and overt PRO subjects do not pattern alike under association-with-focus: (18)  No, Daniel also promised to prepare himself the dinner. Daniel ( y (y also promised that y prepares the dinner)). (ii) Strict reading (coreference): No, Daniel also promised that Eduard would prepare the dinner Daniel ( y (y also promised that he would prepare the dinner)). No, Daniel also promised to prepare himself the dinner. Daniel ( y (y also promised that y prepares the dinner)). (ii) Strict reading (coreference): No, Daniel also promised that Eduard would prepare the dinner Daniel ( y (y also promised that he would prepare the dinner)).

(he= Eduard)
The statement in (18a) with a null (by hypothesis, PRO) subject can only be denied in one way -that is, on its BVA construal-, as expected. This result signals yet again obligatory control. Crucially, if we substitute null PRO in (18a) with overt PRO as in (18b), the resulting statement can now be denied in either of two ways: on its BV construal (18ai) or its coreferential construal (18aii). This asymmetry between null and overt PRO under the association-with-focus test holds across both complement (18) and para-adjunct clauses (19) No, Daniela also cheated in order for herself to win. Daniela ( y (y also cheated in order for y to win)). (ii) Strict reading (coreference): No, Daniela also cheated in order for María to win. Daniela ( y (y also cheated in order for her to win)). No, Daniela also cheated in order for herself to win. Daniela ( y (y also cheated in order for y to win)). (ii) Strict reading (coreference): No, Daniela also cheated in order for María to win. Daniela ( y (y also cheated in order for her to win)).
(her= María) The availability of a strict coreferential reading for overt PRO alongside its BV reading is surprising, since under all the previous tests, overt PRO has shown the expected OC pattern of interpretation. Association with-focus is thus the only test distinguishing overt PRO from null PRO and calling into question the former's OC status. 7, 8 7 Judgments for the association-with-focus test and, more generally all the Spanish data presented here, have been confirmed by 13 native speakers of Spanish: 7 Colombian, 1 Bolivian, 1 Mexican, and 4 speaking Peninsular Spanish. 8 Herbeck (2018) also observes interpretive asymmetries between null and overt PRO in control configurations. He reports, in particular, that PRO can yield distributive (bound variable) as well as collective (coreferential) readings with a weak/cardinal QP controller, whereas with overt PRO, the collective reading is preferred to the distributive reading. He gives the following paradigm to illustrate the contrast (from Herbeck 2018: 183-184). The context for the distributive reading is one in which every neighbor promises to prepare a dinner on a different day of the week, and the context for the collective reading is one where there are four neighbors in the  i /? BV: Each of the four boys promised that he would prepare dinner ii Coreference:The four boys promised that they would prepare dinner together Note, however, that if we change the direct object la cena in (2), to piñatas as in (3), both the collective reading where the three neighbors meet to buy and bring all the piñatas together to the party, and the distributive reading where each neighbor buys and brings a different piñata to the party are readily available with overt PRO. (4) further illustrates the availability of a distributive reading for overt PRO with an inherently distributive infinitival predicate. Interpretive asymmetries between null and overt PRO Isogloss 2022, 8(2)/5 11 commanded by its antecedent, and only yields BV readings under both the ellipsis and the association-with-focus test, overt PRO exhibits most, but not all the properties of OC. It must be locally c-commanded by its antecedent, and yields only BV readings under ellipsis, just as is the case with null PRO. Crucially, however, under the association-withfocus test, overt PRO shows more interpretive possibilities than null PRO, since it yields a coreferential reading alongside the BV reading. In sum, both null and overt PRO pattern alike with respect to the syntactic criteria for OC (c-command and locality) across these two types of clauses. Regarding their interpretive properties, however, an unexpected contrast arises, which we dub the overt vs. covert PRO paradox: (20) Overt vs. Covert PRO paradox (i) Both null and overt PRO only allow BV interpretations under the ellipsis test.
(ii) Overt PRO, unlike null PRO, also allows coreferential interpretations under the association-with-focus test.
Why do null PRO and overt PRO pattern differently (yield conflicting results) with respect to the two standard tests for pronominal interpretation? In particular, why does overt PRO, unlike null PRO, also allows coreference interpretations, and why so only under the association-with-focus test, but not under the ellipsis test?
As we shall now see, the answer to these paradoxes is to be found in the interaction of the binding requirements holding over null vs. overt anaphors.

The Anaphor Generalizations
We argue that the solution to the Overt vs. Covert PRO paradox lies in the Anaphor Generalizations put forth in (21) (repeated from (10) above), which appeal to the distinction between syntactic binding and semantic binding, as advocated by Büring (2005) and Heim & Kratzer (1998) for instance. Syntactic binding requires a bindee to be c-commanded by and coindexed with its binder (cf. Chomsky (1981)), while semantic binding requires the bindee to be interpreted at LF as a variable bound by a predicate abstractor/ -operator (i) Both null and overt anaphors need to be syntactically bound.
(ii) Overt anaphors can be semantically bound, null anaphors must be semantically bound.
(21) requires null anaphoric expressions such as PRO to be both syntactically and semantically bound, while only enforcing syntactic binding for overt anaphors.
Consider first the ellipsis context in (18)-(19) above, repeated in (22) and (23). Recall that both overt and null PRO allow sloppy, but not strict interpretations under ellipsis: María left in order (for herself) to be happy.
(ii) Strict reading (coreference): María left in order for Juan to be happy.
The syntactic binding requirement in (21i) straightforwardly accounts for this contrast. As shown in (24), the sloppy reading of the elided infinitival clause satisfies syntactic binding since, be it null or overt, PRO in the second conjunct is bound by the matrix subject.
(24) Sloppy reading (ellipsis test estar feliz]. 'Juan left in order for himself to be happy and María also left in order for herself to be happy.' In contrast, the strict (coreferential) reading of the elided infinitival clause would necessarily involve a configuration where the overt/null PRO in the second conjunct is bound by its antecedent, the matrix subject in the first conjunct, in contravention of the syntactic binding requirement in (21i), as shown in (25)