Why Is There No Raising to Object in Spanish? A Study of Code-Switching

: Raising to Object (RtoObj), like other types of Raising conﬁgurations, features a determiner phrase (DP) in a dual-clausal relationship with both the matrix and the embedded clauses. RtoO is possible in English and a few other languages, most famously, Icelandic. However, it is not possible in many other languages, such as Spanish. As far as we can tell, insight into what licenses RtoObj is largely speculative. The main goal of this paper is to limit the range of possible hypotheses by pinpointing the source of the cross-linguistic difference using code-switching data. A priori, we could hypothesize two possible sources for the licensor of RtoObj: it could be a feature in the matrix clause or a feature in the inﬁnitival complement. In this chapter, we present code-switching data that support the second option: English T def P is linked to the licensing of RtoObj. We ﬁnd that early Spanish/English bilinguals overwhelmingly prefer code-switched RtoObj samples when the inﬁnitival complement is in English and they reject RtoObj when the complement is a Spanish inﬁnitival. This suggests that Spanish T def is either different or altogether missing.


Introduction
Raising to Object (henceforth RtoObj), like other types of Raising configurations, features a determiner phrase (DP) in a dual-clausal relationship with both the matrix and the embedded clauses. Consider Example (1).
1. John believes Mary to be intelligent.
In (1), Mary is thematically linked to the infinitival complement-'Mary' is an argument of 'being intelligent'. At the same time, it has the grammatical function of object in the matrix clause, as is revealed using common tests, such as passivization or case morphology: as we can tell, insight into what licenses RtoObj is largely speculative. Additionally, RtoObj has limited cross-linguistic distribution, which creates additional difficulty to further investigate the question of licensure. The main goal of this paper is to limit the range of possible hypotheses by pinpointing the source of the cross-linguistic difference.
The licensing of a RtoObj structure requires the presence of two features in the syntactic structure: a feature in the matrix clause and a feature in the infinitival complement. Let us use the abstract tree in (4) to illustrate the discussion: Languages 2021, 6, x FOR PEER REVIEW 2 of 15 The contrast between (1) and (3) raises the question of what feature or features differentiate Spanish from English and give(s) rise to the distinct acceptability judgments. As far as we can tell, insight into what licenses RtoObj is largely speculative. Additionally, RtoObj has limited cross-linguistic distribution, which creates additional difficulty to further investigate the question of licensure. The main goal of this paper is to limit the range of possible hypotheses by pinpointing the source of the cross-linguistic difference.
The licensing of a RtoObj structure requires the presence of two features in the syntactic structure: a feature in the matrix clause and a feature in the infinitival complement. Let us use the abstract tree in (4)  RtoObj involves a functional feature in the matrix clause that establishes a dependency with an argument in the lower clause-hence, the accusative case and the object-like property of the raised DP. This functional feature must be able to probe into a subordinate clause. In our structure in (4), and following a tradition that begins with Chomsky (1995), we take it that the head that assigns accusative case to the argument of the lower clause is v.
Additionally, RtoObj requires a feature in the infinitival complement that makes it transparent for a probe in the matrix clause. Following a line of thinking that originates in Chomsky (1995), we assume that English epistemic verbs can select a deficient T phrase (TdefP) that is unable to license an overt or covert DP, with the consequence that the thematic subject of the infinitival complement must establish a dependency in the matrix clause.
Since RtoObj requires two features in the structure, the absence of RtoObj in Spanish could come about due to the absence of one of these features in the Spanish inventory. One possibility is that the Spanish v does not have the ability to probe lower than a TP barrier. Alternatively, the absence of RtoObj in Spanish would suggest that epistemic verbs cannot select for Tdef or that Spanish lacks this category altogether.
Thus, the question that this article addresses is: What makes English and Spanish distinct-is it the matrix v or the subordinate Tdef? In order to extricate the feature or features that yield RtoObj, we propose using code-switching data. As we shall show, codeswitching by deep bilinguals-those that acquired both languages from a very early age and continued to develop both languages into adulthood (see López 2020 for discussion of the concept of 'deep bilingual')-helps us set the laboratory conditions to investigate alternative hypotheses.
Let us say a few words about intra-sentential code-switching. For starters, let us introduce an example that appeared in the Facebook feed of one of the authors of this article: 5. Antes de que se vaya, thank President Obama for everything he's achieved. He's worked hard to protect and defend nuestros terrenos, nuestro aire, nuestras aguas, nuestras comunidades, y nuestra madre tierra. Add your name to our thank you letter today! ("antes de que se vaya"= "before he leaves") RtoObj involves a functional feature in the matrix clause that establishes a dependency with an argument in the lower clause-hence, the accusative case and the object-like property of the raised DP. This functional feature must be able to probe into a subordinate clause. In our structure in (4), and following a tradition that begins with Chomsky (1995), we take it that the head that assigns accusative case to the argument of the lower clause is v.
Additionally, RtoObj requires a feature in the infinitival complement that makes it transparent for a probe in the matrix clause. Following a line of thinking that originates in Chomsky (1995), we assume that English epistemic verbs can select a deficient T phrase (T def P) that is unable to license an overt or covert DP, with the consequence that the thematic subject of the infinitival complement must establish a dependency in the matrix clause.
Since RtoObj requires two features in the structure, the absence of RtoObj in Spanish could come about due to the absence of one of these features in the Spanish inventory. One possibility is that the Spanish v does not have the ability to probe lower than a TP barrier. Alternatively, the absence of RtoObj in Spanish would suggest that epistemic verbs cannot select for T def or that Spanish lacks this category altogether.
Thus, the question that this article addresses is: What makes English and Spanish distinct-is it the matrix v or the subordinate T def ? In order to extricate the feature or features that yield RtoObj, we propose using code-switching data. As we shall show, codeswitching by deep bilinguals-those that acquired both languages from a very early age and continued to develop both languages into adulthood (see López 2020 for discussion of the concept of 'deep bilingual')-helps us set the laboratory conditions to investigate alternative hypotheses.
Let us say a few words about intra-sentential code-switching. For starters, let us introduce an example that appeared in the Facebook feed of one of the authors of this article:

5.
Antes de que se vaya, thank President Obama for everything he's achieved. He's worked hard to protect and defend nuestros terrenos, nuestro aire, nuestras aguas, nuestras comunidades, y nuestra madre tierra. Add your name to our thank you letter today! ("antes de que se vaya" = "before he leaves") ("nuestros terrenos, nuestro aire, nuestras aguas, nuestras comunidades, y nuestra madre tierra" = our land, our air, our waters, our communities and our mother earth") As you can see, constituents from both English and Spanish find their way into the structure of the clause. For deep bilingual speakers, code-switching should be regarded as an integral component of their linguistic competence. Consequently, there are rulegoverned instances of code-switching and unacceptable instances and deep bilingual speakers can provide acceptability judgments on code-switched sentences just like they do with monolingual sentences.
Many linguists who focus on code-switching assume the No Third Grammar Approach (MacSwan 1999). Under the No Third Grammar approach, any unacceptability that arises in code-switching is due to restrictions inherent to the two languages themselves rather than a separate, code-switching-specific rule system. We fully endorse this assumption, which is foundational in our code-switching work.
In light of the previous discussion, consider the following fabricated code-switching sentences: In the first sentence, the matrix predicate is in English while the subordinate clause is in Spanish. In the second sentence, it is the other way around. Will these sentences be acceptable to Spanish/English bilingual code-switchers? The No Third Grammar Approach informs our understanding of RtoObj and, therefore, we expect that certain combinations will be acceptable to code-switchers, whereas others will not be only on the basis of the features that appear in the structure (4). If a property of the matrix predicate licenses RtoObj, (6) should be acceptable because the matrix clause is in English and, therefore, so is the matrix v. On the other hand, if a property of the subordinate clause licenses RtoObj, then (7) should be acceptable, because the subordinate clause is in English.
In this article, we present data that support the second option: English/Spanish bilinguals accept (7) and reject (6). We find that early Spanish/English bilinguals overwhelmingly prefer code-switched RtoObj samples when the infinitival complement is in English and they reject RtoObj when the complement is a Spanish infinitival. Consequently, English T def P is linked to the licensing of RtoObj. This suggests that Spanish T def is either different or altogether missing as a grammatical ingredient.
The paper is organized as follows. In Section 2, we discuss RtoObj more formally, and we introduce two related phenomena: Raising to Subject and Object Control. Section 3 discusses code-switching as a means of analyzing the nature of RtoObj and introduces our research questions. Section 4 presents the study, including methods, and results. The discussion and conclusions appear in Sections 5 and 6, respectively.

Raising
Raising to Subject (RtoSubj) (8) and RtoObj (9) are characterized by having a nonfinite complement and a DP that is simultaneously in a thematic relationship with a predicate in the subordinate clause and in a grammatical dependency with a predicate in the matrix clause.

Raising to Subject
Ludwig seems to be talented. 9. Raising to Object Wolfgang believes Ludwig to be talented.
In both the RtoSubj (8) and RtoObj (9) examples above, the DP in a dual-clausal relationship is Ludwig, which receives its θ-role from the adjective in the small clause that belongs to the non-finite complement. In this way, the proposition of the complement in both (8) and (9) is that Ludwig is talented.
What differs for the DP between (8) and (9) is its relationship with the matrix clause. To make our discussion more explicit, we adopt a fairly standard view on clause structure, the one in Chomsky (2000) and represented in (10) and (11): In both the RtoSubj (8) and RtoObj (9) examples above, the DP in a dual-clausal relationship is Ludwig, which receives its θ-role from the adjective in the small clause that belongs to the non-finite complement. In this way, the proposition of the complement in both (8) and (9) is that Ludwig is talented.
What differs for the DP between (8) and (9) is its relationship with the matrix clause. To make our discussion more explicit, we adopt a fairly standard view on clause structure, the one in Chomsky (2000) and represented in (10) and (11) That is, we assume two relevant functional categories in the clause, T and v. Both of them can establish dependencies with a DP argument. In Case Theory terms, we say that T assigns nominative case and v accusative case. Additionally, we adopt the broad out-tionship is Ludwig, which receives its θ-role from the adjective in the small clause that belongs to the non-finite complement. In this way, the proposition of the complement in both (8) and (9) is that Ludwig is talented.
What differs for the DP between (8) and (9) is its relationship with the matrix clause. To make our discussion more explicit, we adopt a fairly standard view on clause structure, the one in Chomsky (2000) and represented in (10) and (11) That is, we assume two relevant functional categories in the clause, T and v. Both of them can establish dependencies with a DP argument. In Case Theory terms, we say that T assigns nominative case and v accusative case. Additionally, we adopt the broad out-That is, we assume two relevant functional categories in the clause, T and v. Both of them can establish dependencies with a DP argument. In Case Theory terms, we say that T assigns nominative case and v accusative case. Additionally, we adopt the broad outlines of the Agree (p,g) framework of Chomsky (2000). The idea is that syntactic dependencies are established when a functional category with a bundle of unvalued features (the probe) finds in its c-command domain a constituent with matching valued features (the goal). If the probe bears an EPP feature, it can attract the goal and form a spec position.
Both examples in (8) and (9) have in common that the non-finite T of the subordinate clause does not have any φ-features that would establish a dependency with the DP argument in the subordinate clause. This is what we called T def above. This lack of φ-features on T def makes the DP available to a higher probe. Examples (10) and (12) represent a RtoSubj structure. The v in the matrix predicate is an intransitive v without φ-features. The DP eventually establishes a dependency with the φ-features of the matrix T. If Case Theory is assumed, the DP receives the nominative case. Examples (9) and (13) represent RtoObj. Here, the v of the matrix clause is a transitive v in full possession of φ-features, which are valued against the φ-features of the DP: it is said that the DP receives accusative case.
English clearly has an EPP feature in T acting in conjunction with Agree. As a result, the DP of the subordinate clause in a RtoSubj structure raises and merges with T, forming a spec. This is shown in (10) and again in (12). As for RtoObj, we are not certain that v triggers movement of the DP (despite some arguments in Bowers 1993) and, therefore, we provide two choices, (13a) and (13b). In (13a), Ludwig has raised out of the subordinate clause; in (13b), it stays in situ. The assumption that the argument in RtoObj constructions stays, in fact, in the subordinate clause was predominant in the 1980s and led to the alternative moniker, Exceptional Case Marking (ECM). For our purposes, the decision between (13a) and (13b)  As mentioned, RtoObj is not possible in Spanish (15). However, RtoSubj is fine (14).
Wolfgang believes ACC Ludwig to be talented The unacceptability of (15) poses an interesting puzzle for syntactic theory. What is the property or properties that leads to the difference between (9) and (15)? Now we have the tools to pose this question a little more formally than in the introduction. One possibility is that matrix v has different properties in English and Spanish: the English v can establish a dependency long distance, while Spanish v cannot. The other possibility is that the subordinate T has different properties. The complement of epistemic verbs in Spanish does not select a T def : the non-finite T projects a minimality barrier that prevents an outside probe to reach inside the TP. Notice that this second solution leads to another question: why is (14) grammatical? Is the absence of a T def a property of epistemic verbs only or is it a general property of Spanish? If the second, should the Spanish lack of T def also not prevent RtoSubj? There is in fact a proposal along these lines in Ausín (2001). He argues that RtoSubj in Spanish involves, in fact, raising out of a vP and not out of a TP def . If so, then T def simply does not exist in Spanish and verbs such as creer 'believe' select a CP, like regular attitude verbs. We leave the question open at this point and go back to it in Section 5.

Code-Switching as a Tool
One way to learn about languages is to study speakers' I-languages via elicited judgments of acceptability. The intuitions used in the study of I-languages are typically monolingual intuitions on the consultants' native language, but deep bilinguals can also provide consistent acceptability judgments about code-switched stimuli (see González-Vilbazo et al. 2013 for further discussion). We assume that code-switching judgments reflect the I-language of bilinguals in the same way that monolingual intuitions reflect the I-language of monolinguals.
Additionally, we assume a No Third Grammar approach (González-Vilbazo and López 2012;MacSwan 1999;Woolford 1983), which states that there is no code-switching-specific rules and restrictions. Instead, code-switching restrictions emerge as a result of the interaction of the properties of the participating languages as well as common universal properties.
This article is meant as a contribution to the larger research project of using codeswitching to create laboratory conditions to test theoretical hypotheses. For example: González-Vilbazo and Koronkiewicz (2016) and Koronkiewicz (2014) used code-switching to test competing theories of pronouns. Ebert (2014)  In this article, we expand the use of code-switching to better understand RtoObj as well. Recall the fundamental question that we posed above: What property or set of properties allows RtoObj in English, and how is it disallowed in Spanish? Recall also that we proposed two possible accounts: either a property of the matrix v or a property of the T in the subordinate clause teases the two languages apart. In code-switching contexts, the two options lead to distinct predictions. Consider the following two sentences: 16. I believe John ser inteligente.
Eng/Span 17. Creo a Juan to be intelligent. Eng/Span In sentence (16), the matrix v is English while the subordinate T is Spanish. In sentence (17), the reverse is the case: v is Spanish and non-finite T is English. These yield the following two predictions, which we now state formally:  (17) should be judged as better than (16).
Notice that the predictions in (18) and (19) arise due to the impossibility of RtoObj in Spanish.
In order to tighten up our argument, we included Object Control (ObjC) sentences in our study. ObjC sentences are superficially similar or identical to RtoObj sentences, but their underlying syntax is very different. ObjC structures are available in Spanish as well as English. Example (20) is an ObjC in English, (21) in Spanish, and (22)  As indicated in (20), the object of an ObjC verb is in fact a member of the θ-structure of the matrix predicate; this is the major difference with RtoObj, where the DP that plays the role of the object receives no θ-role from the matrix predicate. By hypothesis, the non-finite T of ObjC sentences includes a silent subject whose reference is dependent on the controlling object. This realization is what led to the analysis of ObjC as in (22), where the subordinate predicate has a silent argument referred to as PRO. 2 We decided to include ObjC in our study as a necessary contrast with RtoObj. Since ObjC is possible in both English and Spanish, no code-switching configuration is predicted to result in unacceptability-mutatis mutandis. Thus, switches with English matrix clauses and English infinitival complements should provide equivalent acceptability judgments. Both English matrix (23) and English complement (24) are expected to be equally acceptable.
23. I persuade John ser honesto. 24. Persuado a Juan to be honest. (23) and (24). Thus, testing the acceptability of ObjC in code-switching grounds our analysis and provides additional evidence that Languages 2021, 6, 172 7 of 14 the methodology employed here is on the right track. In sum, we propose the following research question (25) and hypotheses (26) and (27) with regard to the whether the matrix clause or the complement is in English.

Research Question
Do deep Spanish/English bilinguals rate code-switched sentences differently by whether the English clause is matrix (CP1) or embedded (CP2) for RtoObj or ObjC? 26. Hypothesis 1-Raising to Object There will be a difference in rating between English CP1 and English CP2 because RtoObj exists in only one of the languages, resulting in lacking some property or properties in one or more combinations.

Hypothesis 2-Object Control
There will be no difference in rating between English CP1 and English CP2 because Object Control exists in both languages, allowing its necessary properties to be available in all combinations.

Methods
For the experiment, we followed the methodological considerations in González-Vilbazo et al. (2013), including the design of a background questionnaire to identify deep bilinguals by age of acquisition and daily usage. 3 A group of 15 deep Spanish/English bilinguals were recruited at a large Midwestern public institution. All bilinguals had learned Spanish in the home and English either upon entering school or before, resulting in an age of acquisition of 6 or younger for both languages. The bilinguals used both languages every day and had at least some college education due to being recruited from an undergraduate population at a large Midwestern university.
Stimuli in both the RtoObj and ObjC conditions were varied with regard to English CP1/2 (English CP1, English CP2). We also made sure that we included the language of the raising or controlling DP as an independent variable and, consequently, the examples included an equal number of DPs in Spanish and English. Why should we do this? The possibility that the language of the DP could prevent RtoObj is not regarded as a viable hypothesis because DPs in both languages enter all kinds of dependencies. However, we could really not be sure that the language of the DP was not going to play a role for extraneous reasons. Thus, we wanted to make sure that the language of the DP did not intrude as an unwelcome confounding variable. For the same reason, we also included Spanish DPs with and without accusative a. Participants saw two of each remaining combination of factors (28) All participants completed the background questionnaire followed by the six blocks of code-switched ratings. In total, participants saw 28 items related to the current study in addition to 239 other distractor items as part of a six-block Latin Square design. Participants were asked to rate each item on a 1-5 Likert scale. Appendix A includes a full list of items.
Due to stigma issues inherent to studying code-switching (González-Vilbazo et al. 2013), the analysis of Likert judgments presents a unique statistical problem. Badiola et al. (2018) found that participants' perception of code-switching had a relationship with the maximum rating given to any item. Whereas participants with a positive perception of code-switching used an entire 1-7 Likert scale, those with a negative perception used only the lower part of the scale. However, the authors found that all participants rated some types of switches higher than others regardless of whether they used all or part of the rating scale. As a result, they concluded that bilinguals distinguish between "good" and "bad" switches regardless of the amplitude of difference between the ratings of the two categories.
Upon initial analysis of the current study's dataset, the problem of compression of judgments was also found. Figure 1 shows the average and maximum ratings for each participant across all subtypes. Whereas most participants used the entire scale, three (1, 4, 16) did not. Participants' average ratings also differed for more than two points.
To statistically analyze existing Likert data with clear differences in use of the scale, there are at least three options. The first is to normalize the distributions. It is unclear, however, whether it is truly the case that one participant's maximum rating is equivalent to another's. Another option is to use repeated methods to compare bilinguals directly to themselves. Although participants' categorical acceptance can be inferred in this way, it still assumes that there are equal numbers of good and bad items. In a code-switching study where ratings are impossible to anticipate, a correct/violation paradigm is untenable, and thus, an assumption of equal numbers of good and bad items is unwarranted.
We adopt a third strategy, which is to recode the 1-5 Likert ratings as a binary rating. This normalizes the data and affords us the possibility that more than half of the items are Whereas most participants used the entire scale, three (1, 4, 16) did not. Participants' average ratings also differed for more than two points.
To statistically analyze existing Likert data with clear differences in use of the scale, there are at least three options. The first is to normalize the distributions. It is unclear, however, whether it is truly the case that one participant's maximum rating is equivalent to another's. Another option is to use repeated methods to compare bilinguals directly to themselves. Although participants' categorical acceptance can be inferred in this way, it still assumes that there are equal numbers of good and bad items. In a code-switching study where ratings are impossible to anticipate, a correct/violation paradigm is untenable, and thus, an assumption of equal numbers of good and bad items is unwarranted.
We adopt a third strategy, which is to recode the 1-5 Likert ratings as a binary rating. This normalizes the data and affords us the possibility that more than half of the items are good or bad. Binary ratings reflect acceptability as a scalar proportion of acceptance out of 1, which takes the place of average ratings. A binary coding also allows us to use a binary logistic regression model, which indicates the strength of each input factor in predicting the outcome rating. We chose to remove ratings of 3 because it is unclear whether 3 indicates acceptance or non-acceptance among participants. Ratings of 1-2 were then coded as 0 (not accepted) and 4-5 as 1 (accepted).
For the analysis, we ran a Binary Logistic Regressions with input variables Type (Raising to Object, Object Control), English CP1/2 5 (English CP1, English CP2), and Language of DP (English, Spanish). In step one of the model, we analyzed only the predictors, and we added the interaction CP1*DP in step two. The step with the better fit as measured by the -2 Log Likelihood is reported in the results. Figure 2 shows that potential asymmetries in the bilinguals' acceptance rate were found both in the Raising to Object and the Object Control conditions. The Raising to Object switches differed by English CP1/2 (English CP2 > English CP1), whereas the baseline Object Control condition did not. The results of the Binary Logistic Regression appear in Table 1. The only categorical predictor with a main effect was English CP1/2. Inspection of the data reveals that English CP2 was preferred to English CP1 overall. However, type of sentence (Raising to Object, Object Control) significantly interacted with English CP1/2, driven by the difference between the acceptance of English CP2 and English CP1 in RtoObj (0.707 and 0.329, respectively) rather than in Object Control (0.629 and 0.700). Given these results, we return to the research question regarding an effect for English CP1/2. We accept both hypotheses: there is a difference between Spanish and English matrix clauses for RtoObj but not Object Control.  The results of the Binary Logistic Regression appear in Table 1. The only categorical predictor with a main effect was English CP1/2. Inspection of the data reveals that English CP2 was preferred to English CP1 overall. However, type of sentence (Raising to Object, Object Control) significantly interacted with English CP1/2, driven by the difference between the acceptance of English CP2 and English CP1 in RtoObj (0.707 and 0.329, respectively) rather than in Object Control (0.629 and 0.700). Given these results, we return to the research question regarding an effect for English CP1/2. We accept both hypotheses: there is a difference between Spanish and English matrix clauses for RtoObj but not Object Control. Interestingly, there was an unexpected interaction between type of sentence and DP language, revealing a preference for Spanish DPs in Object Control but not RtoObj. Importantly, there was no interaction between English CP1/2 and language of the DP and no interaction between type, English CP1/2, and language of the DP. The relationship between language of the DP and Object Control is a notable datum that is beyond the scope of the present analysis. 6

Discussion
It is not surprising that an effect for English CP1/2 was only found for Raising to Object. We put forth two predictions, repeated as (32) and (33)  Prediction 2 (33) was corroborated: structures with a Spanish matrix and an English non-finite complement were accepted more than twice as often (M = 0.707) as the structures with an English matrix complement (M = 0.329) and a Spanish subordinate clause. The same is not true of ObjC, with similar acceptance rates for English complement (M = 0.629) and English matrix clause (M = 0.700). The OC data confirm that the difference in acceptability between an English subordinate clause and a Spanish subordinate clause in RtoObj is indeed linked to a property of T that is specific to raising constructions and not of control constructions.
As we see above, RtoObj in code-switching contexts is very much preferred when the non-finite T is English. We take it then that the property that makes RtoObj grammatical in English and ungrammatical in Spanish resides in the complement clause and not in the matrix v. This result is consistent with Chomsky's (1981) proposal that RtoObj should be analyzed as resulting from transparency of the non-finite T to external government, what he called Exceptional Case Marking, which became reanalyzed as the T def property of Chomsky (1995). However, this result leads to another puzzle. As shown in Example (14), Spanish allows what appear to be RtoSubj sentences. It is commonly assumed that RtoSubj sentences should require a T def in the subordinate clause as well. If T def is part of the repertoire of Spanish grammar, we need to explain why T def is not available with epistemic predicates to form RtoObj sentences.
Here, are the options. Option 1 would be to stipulate this property of verbs such as creer 'believe', considerar 'consider', esperar 'expect': they simply cannot select for T def . Option 2 is the more intriguing one: despite appearances, there is no T def in Spanish at all. What appear to be instances of RtoSubj in Spanish actually do not involve a TP at all but a Small Clause structure consisting only of a vP, as in (34). Epistemic verbs select a regular complement clause (TP or CP).

María parece [ vP t ser lista]
'Maria seems to be clever.' To our knowledge, the only proposal that assumes no T def in Spanish is Ausín (2001), and Ausín's proposal is controversial (see, e.g., Gallego 2007). The evidence against T def in Ausín is due to his analysis of RtoSubj verb parecer. In particular, he argues that parecer + infinitive is a modal construction.
'What Juan {hopes (for), desires}, is to know the news.' (Ausín (2001): (98) Based on the evidence in (35)-(38), Ausín concludes that parecer + infinitive is a modal verb, and he proposes that its complement is a VP/vP in examples such as (34). 7 We can adopt Ausín's insights to account for the results found in this investigation: The reason why there is no RtoObj in Spanish is because there is no Tdef in this language. Epistemic verbs select a regular clause structure.

Conclusions
This study has shown that code-switching can be used to provide evidence for or against existing theoretical proposals. This particular study investigated the possible grammatical factors that give rise to RtoObj. We pointed out that the crux could be found either in a feature of the matrix clause-by hypothesis, associated with little v-or with a feature of the subordinate clause-a feature in T that makes it transparent for external probes. By using code-switching, we were able to limit the scope of our search for the necessary properties that give rise to RtoObj, a search that now is restricted to the infinitival complement. At this point, two options were presented: one that requires a stipulation that epistemic verbs such as 'think' and 'consider' do not select for Tdef and one that proposes an absence of T def in Spanish altogether, following Ausín's (2001) analysis of the RtoSubj verb parecer. We tentatively adopt the second option because it seems to provide a more parsimonious understanding of Spanish syntax. Further study is needed to corroborate this analysis, including a potential avenue via a code-switching study of Raising to Subject.
The subjects that participated in the study were described as "deep bilinguals", that is, people who acquired both languages since birth or early childhood and who have been able to develop both languages into adulthood. A reviewer for Languages wonders about the generalizability of our results, given that the participants are heritage speakers. The grammars of heritage speakers indeed diverge from those of monolingual speakers in all kinds of interesting ways (see Polinsky and Scontras 2020 for an overview), which can indeed pose challenges for generalizability. However, we think that our result can be generalized beyond this particular group of subjects on the grounds of existing asymmetries between English and Spanish with respect to Raising. As mentioned above, we have argued that the code-switching experiment shows that the feature that is responsible for the absence of RtoObj in the Spanish of our bilingual subjects must be found in the subordinate clause-be it a TdefP or a vP. It could be the case that the rejection of RtoObj among monolingual speakers is due to something else-such as the matrix v. Or it could also be that both the matrix v and the subordinate TdefP or vP contribute to the rejection of RtoObj among monolingual Spanish speakers but not among the bilingual ones. However, a sensible application of the Ockam's Razor heuristic leads us to think that these scenarios are less plausible than the one presented in these pages. Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author Acknowledgments: Thanks to Laura Bartlett for her work in design and administration on previous stages of this project. Thanks is also due to the undergrad research assistants in the University of Illinois at Chicago's Bilingualism Research Laboratory for their help in administering the study.  Luján (1980) proposes that clitic climbing is not possible if the matrix verb can take an indicative complement. This claim predicts correctly that parecer as well as lamentar (regret) and deplorar (deplore) do not allow clitic climbing.