1. Introduction
The tough-construction in (1) has generated an enormous amount of healthy theoretical debate.Footnote 2
The alternation in (1) is not confined to just adjectives (cf. Lasnik & Fiengo Reference Lasnik and Fiengo1974; Williams Reference Williams1983), though this fact has not generally played a significant role in the analysis of (1). Nevertheless, it has been recognized that other kinds of predicates can be tough-predicates, including nouns (Lasnik & Fiengo Reference Lasnik and Fiengo1974) and psych-verbs (Pesetsky Reference Pesetsky1987). The focus of this study is on the take-time construction (TTC), which has also been observed to allow the tough-alternation (Chomsky Reference Chomsky1981: 319, credited to Tim Stowell; Jones Reference Jones1991: 227; Klingvall Reference Klingvall2018; Gluckman Reference Gluckman, Stockwell, O’Leary, Xu and Zhou2019).Footnote 3
As I illustrate in Section 2, the alternation in (2) is identical to that in (1), and so should be given the same theoretical explanation. The contribution of this article is to investigate the syntactic properties of the tough-construction through the lens of the TTC. As an instantiation of the general phenomenon that comprises the tough-construction, a close look at the alternation in (2) sheds light on what is, and is not, a viable analysis of (1).
The TTC provides clarity on two core issues with respect to the tough-construction. Foremost, we find that the subject this test does not get to its surface position in (2) via movement out of the lower clause; however, we find evidence that it has moved from somewhere lower in the main clause. This finding is compatible with predication-based approaches to the tough-construction (as in, e.g., Williams Reference Williams1983), rather than movement-based approaches (as in, e.g. Postal Reference Postal1971). Moreover, it discriminates among various kinds of predication-based accounts as well in that it is not consistent with licensing the non-expletive subject this test in its surface position (as argued in, e.g. Rezac Reference Rezac and Boeckx2006), rather, the ‘tough-subject’ is a (nonthematic) argument of the tough-predicate (Jones Reference Jones1991).
Second, we find that the relationship between the nonfinite clause and the main clause is a modificational, rather than a selectional relationship. In (2), the nonfinite clause is a VP modifier. This again differentiates among analyses of the tough-construction between those that treat the nonfinite clause as an argument of the tough-predicate (e.g. Keine & Poole Reference Keine and Poole2017) and those that treat it as adjoined to the tough-predicate (e.g. Mulder & Den Dikken Reference Mulder and den Dikken1992; Hornstein Reference Hornstein2001).
In addition to these two core observations, I make an ancillary observation about English syntax. We find clear evidence for a high applicative position in English – a language that is otherwise argued to lack high applicatives (Pylkkänen Reference Pylkkänen2008). The data are consistent with what is argued in Kim (Reference Kim2012) and lexical decomposition approaches to light verbs (Ritter & Rosen Reference Ritter and Rosen1997; Hale & Keyser Reference Hale and Keyser2002).
The final point of this article is more general. The TTC is representative of the class of predicates that participate in the tough-construction, including cost and set X back (Jones Reference Jones1991), and possibly psych-verbs (Pesetsky Reference Pesetsky1987). Thus, the findings below are not simply a ‘quirk’ of the TTC, rather the properties that I investigate here are broadly applicable in English syntax. In the specific investigation of the TTC, I therefore address both the homogeneity and heterogeneity of the general class of predicates that permit the tough-alternation. The finding is that all predicate types, adjectives, nouns, and verbs, are potential tough-predicates (see Williams Reference Williams1983). In this way, we start to build a profile of the range of core properties of the tough-construction, where each predicate type differs, and why.
This article is structured in the following way. I will first confirm in Section 2 the parallels between (1) and (2), showing that both constructions involve the same somewhat idiosyncratic properties. I will also note there how the two constructions diverge in both form and meaning. I investigate the TTC’s specific properties in Section 3, using standard tests for constituency, movement, and c-command. I then turn back to the tough-construction in Section 4, showing how the findings shed light on the numerous previous proposals of the alternation in (1). In the conclusion, Section 5, I briefly expand the investigation to comment on other predicates that could possibly provide further insight into the tough-construction, as well as the general argument structure of English.
2. Shared properties of the TTC and tough-construction
The purpose of this section is to establish the (well-known) defining properties of the tough-construction, and illustrate that the TTC represents an instance of the same idea. The first and central observation is that, in both cases, we find an alternation between an expletive/pleonastic subject and non-expletive subject binding a (non-subject) gap in a lower nonfinite clause (represented throughout with ‘
$ e $
’).
The characteristic property of this alternation is that the non-expletive subject in the examples in (3b) and (4b) are syntactically arguments of the main clause, but thematically arguments of the lower clause. The latter point is illustrated by the fact that without the nonfinite clause the tough-subject is not possible, demonstrated by the lack of entailment in (5).Footnote 4
To the extent that we can understand the second sentences in (5), it must be with respect to an elided or implicit event. Thus, we appear to have a case of non-local selection. The natural response is to treat this as a case of movement (as in, e.g. Chomsky Reference Chomsky1981). But this raises more questions, because there is very good evidence that the movement step in the lower clause comprises a step of A
$ ^{\prime } $
-movement. This would make the antecedent-gap chain an instance of improper movement, i.e. an A
$ ^{\prime } $
-chain headed by something in an A-position. Evidence for the A
$ ^{\prime } $
-step comes from standard diagnostics like islandhood, extraction of goals in double object constructions, and licensing of parasitic gaps (Chomsky Reference Chomsky, Culicover, Wasow and Akmajian1977).
Even more unusual, the A
$ ^{\prime } $
-movement is restricted in ways that other A
$ ^{\prime } $
-movements are not. For instance, it does not appear to generally cross finite clausal boundaries.Footnote
6
It has also been widely noted that the tough-construction resists connectivity effects. Tough-subjects cannot be interpreted for scope (10) or for variable binding (11) inside at the gap position (Postal Reference Postal1974; Epstein Reference Epstein1989; Fleisher Reference Fleisher2013).Footnote 7 Tough-subjects also permit Condition C obviation (Munn Reference Munn1994) in (12).
Similarly, as Wilder (Reference Wilder1991: 123) points out (see also Mulder & Den Dikken Reference Mulder and den Dikken1992: 316), examples such as (13a) demonstrate that the tough-subject is not necessarily a selected argument in the nonfinite clause. Because the verb believe can only appear with finite or purely nominal complements, the infinitival clause tough-subject could not have originated as the complement to the infinitive. The same fact holds for refute in (13b).
It is worth noting one argument that, at least superficially, suggests that connectivity effects are possible: idiom chunks. It is widely reported that certain idioms survive in the tough-construction (14a). However, it is also widely reported that not all idioms are possible (14b).
I will not provide a solution to this puzzle here. I adopt Hicks’s (Reference Hicks2009: 554) stance that ‘the behavior of each type of idiom chunk under
$ \Big[ $
tough-movement
$ \Big] $
at least mirrors its behavior under passivization’. Whatever the explanation for this, I simply note that the tough-construction proper and the TTC share the same judgments.
I finally briefly note that the tough-construction and TTC share many similarities in where the gap in the lower clause is allowed to appear. An illustrative example is the fact that raising-to-object/ECM’d arguments are not permitted as tough-subjects, though object control is perfectly fine as a target for the gap (Postal Reference Postal1974: 193).
The examples illustrate that the exact same particular (and somewhat peculiar) properties of the tough-construction are also found in the TTC.
This is not to say that the constructions are entirely identical. There are some notable and important differences between the tough-construction and the TTC. First, obviously, they mean different things. This is important to point out because it affects which nonfinite verbs are permitted. The TTC imposes a telicity restriction on the nonfinite verb and therefore it is incompatible with stative verbs (Mourelatos Reference Mourelatos1978; Mittwoch Reference Mittwoch1991), a property not shared by the tough-construction.Footnote 8
The TTC also differs syntactically in a striking way: it has a richer argument structure, licensing what I identify as an applied argument.Footnote 9
The syntactic status of Mary in (19b) will play an important role in the analysis of the TTC. I return to it in Subsection 3.2.
The TTC also has a more ‘flexible’ argument structure in that it permits the non-expletive subject to bind a non-object gap, something which is disallowed in canonical tough-predicates.Footnote 10
I note that there is an alternative reading of (21a) that treats take as a ‘lexical’ verb (i.e. not a light verb). On this reading, Mary set aside an hour to complete the test. This version of take is diagnosable in two ways. First, lexical-take may appear with verbal particles. Verbal particles are barred in the presence of an expletive subject and a gap in the nonfinite clause (i.e. cases where take is a light verb).
Second, constructions with lexical-take lose the telicity restriction. Thus it is possible to follow sentences like (21a) with a statement that asserts the nonculmination of the nonfinite event (23a), and it is possible to use a stative predicate (23b). When take is a light verb, the event must culminate, and stative predicates (or any non-culminating event) are not possible (24).
I will put aside the lexical version of take in this article. I assume that, in this case, the subject is introduced as an agent in its normal position. To control for this issue, when applicable, I will use inanimate subjects, which cannot be agentive. For example, (21b) is not ambiguous in the same way as (21a).Footnote 11
Finally, there is the obvious observation that there are simply more parts to the TTC. It minimally consists of the light verb take plus a ‘measure phrase’. Note that the measure phrase need not be a temporal unit as long as it describes some bounded interval.Footnote 12
Thus, while the tough-construction proper involves a syntactic relation between an adjective and nonfinite clause, the TTC is a more complex syntactic creature. I view this as a benefit, because I believe that the relative simplicity of the tough-construction hides many of the complex factors that go into the relation between the two clauses. The TTC’s relative ‘complexity’ actually makes the issues somewhat more transparent.
In sum, despite some non-trivial differences, I take the preceding correlations to validate treating the TTC as a proxy for the tough-construction, along with the authors cited above. I turn in the next sections to a thorough investigation of the TTC, putting the tough-construction aside until Section 4.
3. Properties and analysis of the TTC
3.1 Constituency
I will start with a discussion of constituency. In principle, there is nothing wrong with a measure phrase like an hour and a nonfinite clause like to complete the test forming a constituent. However, they do not form a constituent in the TTC. Though this may seem counterintuitive at first glance, this fact is demonstrated through basic constituency tests which force the measure phrase and the nonfinite clause to form a constituent, for instance, all- and pseudo-clefting. The (b) examples simply demonstrate that clefting is possible in the TTC.
A measure phrase and a nonfinite clause also cannot be a fragment answer to the question What did it take? (The responses are marked infelicitous because they are grammatical utterances, just not in the given context.)
It is worth comparing examples where the measure phrase and nonfinite clause do form a constituent. An illustration of such a context is the have-time construction (HTC).Footnote 13
I also point out that the semantic role of the nonfinite clause is different when the measure phrase and nonfinite clause form a constituent. In the HTC, it is possible to paraphrase the relationship as a relative clause, whose head is the measure phrase (30). This is not possible with the TTC (31).
Instead, the nonfinite clause in the TTC is more accurately parsed as a purpose/rationale clause.Footnote 14 (This reading is also available with the HTC, see footnote 13.)
The lack of constituency with the measure phrase and the parse as a purpose/rationale clause lead to the conclusion that the nonfinite clause is merged as a modifier (that is, an unselected argument) of the verb phrase (Faraci Reference Faraci1974), here assumed to be the complex
$ v $
+V. (I discuss control of PRO in the next section.)Footnote
15
The tree in (33) predicts that the measure phrase does not c-command the nonfinite clause. The binding data in (34) confirm this prediction.Footnote 16
I understand modification in terms of the semantic process of predicate modification (Heim & Kratzer Reference Heim and Kratzer1998) or meaning conjunction. In a Neo-Davidsonian event semantics the combined meaning of the two clauses is given in (35a). The predicate take-an-hour is a function that ‘measures’ or ‘bounds’ an event, such that take-an-hour
$ (e) $
means, ‘
$ e $
measures/is bounded at one hour’.Footnote
17
I assume existential closure over events, so that the sentence It took an hour to complete the test has the truth conditions in (35b).
Notice that we neatly explain the telicity restriction found in the TTC. Given that the two events are conjoined via predicate modification, the event described by the nonfinite predicate must measure an hour – that is, it is bounded – as they are in fact the same event.Footnote 18
3.2 High applicatives
The TTC permits an additional argument, sitting between the light verb take and the measure phrase.
This ‘intermediate’ argument is in complementary distribution with an overt subject in the nonfinite clause.
I identify the relationship between this argument and the empty subject position in the nonfinite clause as (obligatory) control, rather than raising.Footnote 19 This is diagnosed by the fact that this position does not tolerate expletive subjects (38), nor does it permit idiom chunk interpretations (39), nor meaning-preservation under passivization (40).
I further observe that the arguments sitting in this position have available scope readings which are not possible for arguments inside of the nonfinite clause. Such lack of connectivity I again take as evidence for a control relation. The comparison between the sentence-medial position and the subject position inside of the nonfinite clause concisely makes the point.
The example in (41a) is true in a context in which each woman separately needed an hour to stand up. Assuming that the group of women is Janice and Mary, (41a) is true in a context in which, when Janice stood up at 8 a.m. and Mary at 11 a.m., both women needed a full hour to rise. The example in (41b) is not true in this context, meaning only that both women rose in the same hour. Similarly, in (42a), when three students appears in front of a year, it is true in the context that each of the students needed different years (e.g. 2016, 2017, 2018) to learn French. When three students appears after a year, the only available reading is that all three students learned French in the same year.
I identify the controller of PRO as the specifier of an applicative phrase, and it is in an obligatory control relationship with the adjoined nonfinite clause. Furthermore, because of the structural position of the adjoined clause, this must be an applicative head that relates an argument to an event – a high applicative in the terminology of Pylkkänen (Reference Pylkkänen2008).Footnote 20
I assume, following Landau (Reference Landau2015), that obligatory control requires a strict c-command requirement between the controller and PRO, which in turn forces the applicative to be a high, rather than low applicative head. That is, if Appl were merged above an hour as a low applicative, the applied argument would not c-command the nonfinite clause. (It would also fail to interact correctly with a tough-subject, as I discuss in the next section.)
An applicative analysis is confirmed by considering languages with overt applicative morphology. Consider the Bantu language Logoori (Luhya, Bantu). In Logoori, the TTC patterns identically to English on all relevant diagnostics, and licenses an intermediate object with the applicative morpheme.Footnote 21
Finally, I point out that treating this argument as a high applicative again matches intuitions. The applied argument seems to be ‘involved’ in the event in some way that the subject of the nonfinite clause is not. For instance, in (45), the difference between the two sentences seems to be in whether Mary is ‘measuring out’ the event of taking an hour. In (45a) we get the sense that Mary has attempted to stand for an hour. Example (45b) also has this reading, but it additionally has a reading in which the speaker in some way is measuring out this event, like they are waiting for Mary to stand up.
I interpret this to be a result of the Appl head mapping its specifier directly to the taking an hour event in (45a). I will refer to this thematic relation as an Affected thematic relation. This affected reading is a result of Mary being in a relationship with the event that measures an hour, as described by the higher
$ v $
P.Footnote
22
Controlled PRO subsumes the thematic role of the lower clause. In contrast, merged inside of the nonfinite clause, Mary is not directly related to this event in an affected thematic relation. She is simply the agent of the event of standing up. In such cases, I assume that the measurement of time is ‘speaker-oriented’ – a notion that I will not attempt to formalize. Thus, the truth conditions of (46)/(47) minimally differ in that (46a) has an extra thematic relation that (47a) lacks.
If this is correct, the data from the TTC point to a potential problem: English is generally thought to lack an applicative head that relates its specifier to an event. Indeed, in the typology of applied arguments, English is considered to be a canonical example of a language that only has a low applicative, i.e. a head that relates an individual to another individual (Pylkkänen Reference Pylkkänen2008).
I will, however, adopt the view of Kim (Reference Kim2012), who argues that English does have structurally higher applicatives which can be observed in the following data.Footnote 23
The idea explored by Kim is that have in general is merely the realization of the complex of Appl and the higher verbal head
$ v $
– taking a cue from Freeze’s (Reference Freeze1992) analysis of have as P-incorporation. In the examples in (48a) and (48b), Appl is merged above the verbal domain. In these cases, Appl ‘denotes a relation between the causee, Mary, and the event’ described by the verb phrase (Kim Reference Kim2012: 77).Footnote
24
To the extent that light verbs like have are related to take, an idea that I consider highly plausible from a lexical decomposition point of view (Hale & Keyser Reference Hale and Keyser2002), then I believe that postulating a high applicative head with take is motivated in English. However, I will remain agnostic as to the name of this projection. It is not the goal of this article to provide a decompositional analysis of the light verb take in English, nor to derive the distribution of high applicatives in English. The point made here is simply that there is sufficient evidence for treating this clause medial argument as a high applied object.Footnote 25
3.3 The interaction of subjects
In this section I will consider the interaction of the tough-subject (the test) and the applied argument (Mary) in (49).
Though the applied argument cannot be interpreted for scope and variable binding inside of the nonfinite clause (as shown in Subsection 3.2), nor can the tough-subject (as shown in Section 2), the two positions do scopally interact with each other, as we might expect if they are in the same clause. Surprisingly, however, the tough-subject can be interpreted for scope and variable binding below the applied argument. Again, it is worth comparing the applied argument versions with the nonfinite clause versions. In the versions with an applied argument, inverse scope of the tough-subject relative to the applied argument is permitted; the tough-subject can be interpreted below the applicative.
The example in (50a) is true in a context in which student Mary completed her French and Spanish tests in less than an hour, student Sally completed her Spanish and German test in an hour, and student Greta completed her German and Swahili test in an hour. In this context, two scopes under no. The example in (50b) is not possible in this same context. Likewise, the crucial reading in (51a) is the one where there are six different languages in total, two for each student, and the students each spent a year learning their two languages. This reading is available in (51a) but unavailable in (51b).
We find the same thing with variable binding.
This suggests that there is a position below the applied object, but outside of the nonfinite clause, in which the surface subject starts. The natural choice is spec-
$ v $
P, which I assume to be the first-merge position of external arguments in general.
The argument in spec-
$ v $
P promotes to the subject position, past the applied object. I assume that the applicative is licensed in situ and is therefore inactive for further syntactic processes (Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001). (See further discussion concerning intervention effects in Subsection 3.4.)Footnote
27
The interaction of the subjects illustrated here provides strong evidence against treating the applied object as a low applicative, merged below V, c-commanding the measure phrase. In such a position, it would be impossible to reconstruct the tough-subject below the applied object. The applicative must be structurally higher than the tough-subject in its base position, which requires that it be merged somewhere higher in the verbal domain.
The structure is consistent with approaches to the tough-construction in which the nonfinite clause merges an operator and then is later predicated of the subject which is generated (athematically) in the main clause. The subject then gets a thematic role through some mechanism of ‘thematic transmission’, whereby the head of the chain is assigned a thematic role through the operator (Williams Reference Williams1983; Browning Reference Browning1987; Heycock Reference Heycock1994).Footnote 28 Though it is possible to capture this idea in a variety of different formalisms, I believe that the notion of ‘thematic transmission’ (as it is intended for the tough-construction) follows from independent syntactico-semantic principles.
Keeping the assumption from earlier that VP/
$ v $
P describes a property of events, and that
$ v $
is responsible for introducing external arguments, I additionally assume that
$ v $
in the TTC introduces a nonthematic argument, i.e. an argument which is not mapped to the event via a thematic relation. This is functionally equivalent to saying that
$ v $
introduces vacuous quantification over individuals, as shown in (55b).Footnote
29
The problem of course is that without a thematic role the added argument is vacuous, and so is plausibly excluded on semantic/pragmatic grounds – though of course it is a violation of more well-known syntactic constraints like the theta criterion of the government and binding framework (see Bruening Reference Bruening2013 for recent discussion), as well as the ban on vacuous quantification.Footnote 30
To fix the issue, the nonfinite clause is predicated of the tough-subject, providing what the main clause cannot, a thematic role, thereby making the quantification over individuals non-vacuous. I assume that operator insertion is permitted to apply freely to form predicates out of nonfinite clauses (Nissenbaum Reference Nissenbaum2000; Landau Reference Landau2011), and that the nonfinite clause is merged again via predicate modification, yielding the structure and meaning in (56). This analysis employs a standard notion of chain-formation via predication (Williams Reference Williams1983; Heycock Reference Heycock1994).Footnote 31
In effect, the nonfinite clause ‘rescues’ the nonthematic subject by providing a complex event description which includes a thematic relation for the subject.
Most importantly, given that the high applied object gets a thematic role in the main clause, it is not eligible for the same ‘rescuing’ via predication that applies to the nonthematic arguments in spec-
$ v $
P. This explains why the applied object cannot bind a non-subject gap.
The ungrammaticality of (57) follows if syntactic objects can have one and only one thematic role, i.e. the theta criterion (or however this is captured in minimalism). The problem with (57) is that the predication relation attributes to the applied object the thematic role made available in the lower clause, but the applied object already has a thematic role in the main clause in (58).
Because we still wish to exclude attributing multiple thematic roles to a syntactic chain (modulo Hornstein Reference Hornstein2001), the sentences in (57) require the applied object to bear two thematic roles, and therefore are out on independent grounds. A PRO argument, however, independently bears the thematic role assigned in the nonfinite clause. A full sketch of the proposal is provided below.
Thus, the fact that the light verb take does not assign a thematic role ensures that a subject generated in its specifier has to get one from somewhere else, like predication. In contrast, because the high applied object does get a theta role in the main clause, it is precluded from getting one via predication.
A related question is why the applied argument, when present, must bind something in the nonfinite clause. That is, it is not immediately clear from the analysis why (60) is ungrammatical.Footnote 32
This is, of course, a well-known issue in the study of control, i.e. why some predicates require an obligatory control relationship (Grano Reference Grano2012; Landau Reference Landau2015). I will not settle the question here.Footnote 33 , Footnote 34
3.4 Subject control and intervention
To account for sentences like The bus took an hour to arrive, I propose that the applied argument in spec-ApplP is able to raise to spec-TP. In this, I am adopting the idea that Appl can come in two ‘flavors’, one which licenses its specifier in situ (i.e. renders its specifier inactive) and another which leaves its specifier fully active for agree.
Note that an argument introduced in spec-
$ v $
P does not have the option of remaining in situ (*Mary took the test an hour to complete), and so when present, must raise to spec-TP, potentially crossing in inactive applied argument. This is because Appl has the specific property of being able to deactivate its specifier, while
$ v $
does not. Anything introduced in spec-
$ v $
P must interact with some non-local head to check its features.
Our analysis raises an important distinction between the TTC and the tough-construction proper: it is generally thought that the latter explicitly and uniformly bans gaps in the subject position; (62a) is repeated from earlier.Footnote 35
In fact, the ban on subject control with tough-predicates appears to be over-stated. There are tough-predicates that permit both object gaps and subject control. Consider Stowell’s (Reference Stowell and Rothstein1991) adjectives describing mental properties, such as mean, kind, nice, petty, …. etc. They permit the controller of PRO to be introduced in a prepositional phrase or as the subject in (63a) and (63b). Mental property adjectives also permit non-subject operator-gap chains in (64).Footnote 36
Thus, the fact that the TCC permits both subject control and an object gap is not unprecedented. Though I cannot offer a full account for why canonical tough-predicates do not allow this, I will speculate briefly. What permits this kind of three-way alternation appears to be related to the semantic function of the controller of PRO. With the canonical tough-predicates, PRO is always controlled by the judge of the tough-predicate – whether the judge is explicit or implicit (Landau Reference Landau2013: 41). Judges of tough-predicates are introduced in a prepositional phrase (Keine & Poole Reference Keine and Poole2017).Footnote 37
There emerges a basic generalization about judge arguments of adjectives: they never participate in argument structure alternations. Judges are ‘fixed’ relative to a given adjective.Footnote 38 In contrast, non-judge arguments are not fixed, and may participate in argument structure alternations. The applied argument in the TTC and the ‘agent’ with kind-predicates are not judges, and freely alternate.
To be clear, I have no explanation for why non-judge arguments are freer in this respect. But this fact is almost certainly related to another issue that arises with the TTC: defective intervention. As Hartman (Reference Hartman2011) notes, evaluator arguments (including judges) in a number of constructions across languages (tough-constructions and various raising contexts) act as interveners for movement, despite the fact that the intervening arguments themselves cannot enter into the movement relationship.Footnote 39
The TTC appears to be an exception to these intervention facts, because the applied argument clearly sits between the subject and the gap – and it moreover can (at least sometimes) undergo raising. Recently, the defective intervention facts for the tough-construction have been accounted for in Keine & Poole (Reference Keine and Poole2017) by attributing the issue to a type mismatch stemming from the introduction of the judge argument. Alternatively, Gluckman (Reference Gluckman2018) suggests that defective intervention reduces to a general constraint on chains that cross attitude holders. Both accounts ultimately attribute the intervention effects to the fact that defective interveners in the tough-construction are judges. Whatever its eventual explanation, if this descriptive generalization holds, then the applied argument in the TTC will not be a defective intervener because it is not an attitude holder: applied arguments in the TTC do not hold a belief about the lower clause. Indeed, applied arguments do not even need to be animate: It took the tree an hour to fall. This is of course consistent as well with the data in (64), which demonstrates another case of non-intervention by a non-judge argument.
Indeed, the parallel between the TTC and mental predicates goes further. Stowell (Reference Stowell and Rothstein1991) argues that Mary is introduced in the specifier of AP in (63b), making an explicit comparison to Larson’s (Reference Larson1988) analysis of the double object construction, where the goal is in spec-VP. In modern frameworks, of course, this has been reanalyzed in terms of an applicative phrase. Stowell’s analysis therefore suggests that the analysis above in which the applied argument raises to spec-TP in the TTC has precedent. Alternatively, Bennis (Reference Bennis, Alexiadou, Anagnostopoulou and Everaert2004) argues that the alternation in (63) is reminiscent of an active/passive alternation, where the prepositional phrase is equivalent to the passive by-phrase. Though I cannot undertake a close examination of these parallel behaviors here, I believe that the similarities are at least strongly suggestive of a shared analysis.Footnote 40
4. Implications for the tough-construction
Given the extensive theoretical landscape concerning the tough-construction, in this section I wish to illustrate how the TTC sheds light on which proposed analyses of the tough-construction are plausible. I focus on two factors which have been debated in the literature: (a) predication vs. movement of the tough-subject and (b) selection vs. modification of the nonfinite clause. I illustrate how the predication-based analysis of the TTC is supported in the tough-construction proper, and that the nonfinite clause is not a selected argument of the tough-predicate. I consequently review the noted evidence against this position, i.e. that the tough-subject is not an argument of the tough-predicate and that the nonfinite clause is selected, and point out the faults in these arguments.
4.1 Predication, not movement
Exporting the analysis of the TTC to the tough-construction, the tough-subject is generated in the specifier of the adjectival projection
$ a $
P, and the nonfinite clause is adjoined to
$ a $
P. Semantically, I take tough-predicates proper to describe properties of events (Gluckman Reference Gluckman, Stockwell, O’Leary, Xu and Zhou2019), which compose with the nonfinite clause as indicated earlier. Like
$ v $
introduced above,
$ a $
may also vacuously introduce a nonthematic subject.
This analysis explicitly denies that the various movement-based analyses, most recently in Hicks (Reference Hicks2009), Hartman (Reference Hartman2012), and Longenbaugh (Reference Longenbaugh2016), are correct.Footnote 41
Moreover, with regard to the position of the subject, the data differentiates between some models of predication analyses. The data are consistent with proposals such as Williams (Reference Williams1983), Jones (Reference Jones1991), Wilder (Reference Wilder1991), Mulder & Den Dikken (Reference Mulder and den Dikken1992), Nissenbaum (Reference Nissenbaum2000), and Keine & Poole (Reference Keine and Poole2017), which treat the tough-subject as a selected argument of the predicate. In contrast, the predication analyses offered in Browning (Reference Browning1987), Heycock (Reference Heycock1994), and Rezac (Reference Rezac and Boeckx2006) in which the subject does not have a selectional relationship with the main-clause predicate are not consistent with the TTC data.Footnote 42 The interaction of the high applied object and tough-subject demonstrate that there must be a position lower in the clause into which the subject can reconstruct.
The main argument against a selectional relationship between the tough-subject and the main clause predicate comes from nominalizations. It is noted that the tough-construction does not survive nominalization of the tough-predicate as in (68) (Chomsky Reference Chomsky, Culicover, Wasow and Akmajian1977: 109; Pesetsky Reference Pesetsky1991: 101). Thus, the reasoning goes, the tough-subject cannot be selected, unlike e.g. John’s eagerness to please.
On the present analysis, (68) are understood by the fact that such nominalizations are root nominalizations, and so have the structure [
$ n $
[
$ \sqrt{\mathrm{ROOT}} $
] ]. They therefore lack an adjectival projection which selects for the subject and which is an appropriate position for the nonfinite clause to adjoin to. This idea is supported by Pesetsky’s (Reference Pesetsky1991) observation that nominalizations in -ness do permit the antecedent-gap chain, though he notes speaker variation.
These facts follow if -ness nominals (for some people) are derived from adjectival predicates, [
$ n $
[
$ a $
[
$ \sqrt{\mathrm{ROOT}} $
] ] ], and therefore include a projection in which the nonthematic subject can be generated and that the nonfinite clause can adjoin to. The TTC makes the same point more explicitly because it lacks a root-derived nominal, but has a gerundive nominalization. Because this nominalization includes
$ v $
P (and apparently the applicative phrase given the possibility of an applied object), there is a position (spec-
$ v $
P) that selects for a subject before nominalization.Footnote
43
What is crucial then for a tough-construction to survive nominalization is that the category that selects for the tough-subject be included in the nominalization.Footnote 44 The nominalization facts are unexplained in accounts that generate the tough-subject higher in the clause.Footnote 45
Independent evidence for selection is also observed in Fleisher’s (Reference Fleisher2015) rare-class predicates (although Fleisher does not interpret it as such, adopting an analysis based on Rezac Reference Rezac and Boeckx2006).
Fleisher’s core observation is that rare-predicates only permit kind-denoting subjects, rather than type-denoting subjects, even when used as tough-predicates. At the very least, such data indicate that the availability of a tough-subject is in part dependent on the lexical semantics of the main clause predicate, which in turn argues against an analysis that completely severs this link.
4.2 Modification, not selection
The analysis above also has implications for the relationship between the tough-predicate and the nonfinite clause. In particular, we have found evidence that this is not a selectional relationship, rather, it is one of modification, as in Williams (Reference Williams1983), Wilder (Reference Wilder1991), Mulder & Den Dikken (Reference Mulder and den Dikken1992), Contreras (Reference Contreras1993), and Hornstein (Reference Hornstein2001). This rules out analyses of the tough-construction which treat the nonfinite clause as an argument of the tough-predicate (Longenbaugh Reference Longenbaugh2015; Keine & Poole Reference Keine and Poole2017; Salzmann Reference Salzmann2017).
However, there are two fairly strong arguments supporting a selectional relationship between the tough-predicate and the nonfinite clause. First, it is noted that there are idiosyncratic restrictions concerning which adjectives can and cannot be tough-predicates. For instance, Landau (Reference Landau2011) offers the following evidence to suggest that the nonfinite clauses (‘Op-derived clauses’) are selected in the tough-construction.
However, it seems that this distinction is not as robust as Landau claims. Many examples of forbidden as a tough-predicate can be found in a Google search.Footnote 46
In fact, this is probably evidence that certain tough-predicates select for subjects, rather than nonfinite clauses, just like what is illustrated in Fleisher (Reference Fleisher2015). That is, forbidden imposes selectional restrictions on what can be a tough-subject, not whether it can combine with a nonfinite clause. Indeed, the same sentences are perfectly grammatical without a gap, e.g. It is forbidden for us to eat two parts of any Kosher animal, showing that nonfinite clauses are compatible with these predicates.
The second argument against a modification relationship appeals to the semantic relationship between the tough-predicate and the embedded clause. Rezac (Reference Rezac and Boeckx2006: 291–292) argues that the lack of an entailment relationship illustrates that the nonfinite clause cannot be an adjunct, because entailment is a general property of (intersective) modification. (Judgments are cited as given.)
Of course, this argument only goes through if the tough-subject is in fact thematically licensed in the main clause, which it is not.
Arguments in favor of a modification analysis include the following ellipsis data from Contreras (Reference Contreras1993: 5–10). Contreras first notes that VP ellipsis is not possible when the VP is an adjunct (see Zagona Reference Zagona1988; Lobeck Reference Lobeck1986).
If the nonfinite clause is an argument of the tough-predicate, we would expect to be able to elide its VP, contrary to fact.Footnote 47
Empirical evidence for modification also comes from comparison with true complements to adjectives, which are not acceptable in attributive position, though (some) nonfinite clauses are.Footnote 48
Finally, Wilder (Reference Wilder1991: 125) notes an additional theory-internal argument for treating the nonfinite clause as an adjunct. He observes that, ‘TM infinitives now form a class with infinitival relatives and purpose clauses; they never occur as arguments to lexical heads, but only as adjuncts’. That is, nonfinite clauses with operator gaps are never selected for (pace Landau Reference Landau2011).
I note though, that unlike the TTC, the tough-construction does not have a paraphrase as a purpose/rationale clause (Wilder Reference Wilder1991: 129).Footnote 49
This is a natural consequence of the fact that the nonfinite clause is a
$ v $
P modifier in one case and an adjectival modifier in another. Given that purpose/rationale clauses are naturally VP/
$ v $
P oriented, then the lack of such a reading can be attributed to the fact that a nonfinite clause with the tough-construction proper modifies a different category.
The categorial difference between
$ a $
P and
$ v $
P also explains why the tough-construction does not license an applied object: High Appl heads select for verbal projections, not adjectival projections.
5. Conclusion
A close examination of the TTC reveals syntactic variation in tough-constructions. Though they share many core properties, the TTC and the ‘canonical’ tough-construction diverge in important syntactic dimensions. I have capitalized on these differences to explore what is, and is not, a viable analysis for this particular (heterogenous) class of predicates. I conclude that the non-expletive subject in the TTC/tough-subject is a selected argument of the tough-predicate. And I further conclude that the nonfinite clause is a modifier of the main clause. Both conclusions point to a particular kind of analysis of the tough-construction in general. The study expands the range of inquiry for tough-structures in general, as well as the various aspects of argument structure in English.
More broadly, the general point here is that a close look at the TTC reveals something deeper about the core alternation of the tough-construction. The study developed here should be expanded to look at additional predicates which undergo similar alternations and which have complex argument structures, like cost (This book cost me $20 to buy) or set X back (This book set me back $20 to buy). (See in particular Jones Reference Jones1991: 227, as well as discussion of Spanish light verbs in Fernández-Soriano & Rigau Reference Fernández-Soriano and Rigau2009.) Though there is variation among the class of tough-predicates, there are constant elements as well (Gluckman Reference Gluckman, Stockwell, O’Leary, Xu and Zhou2019). There is always an alternation between an expletive subject and non-expletive subject binding a non-subject gap; there is always a ‘weak’ A
$ ^{\prime } $
-step; there is always a nonfinite clause.
These facts in turn point to a particular brand of analysis of the tough-construction. The most accurate analyses are those which treat the tough-subject as an argument of the main clause and the nonfinite clause as a modifier. Thus, the ideas set forth in Mulder & Den Dikken (Reference Mulder and den Dikken1992) (who treat the tough-construction as a kind of parasitic gap) come closest to a correct response (but see also Jones Reference Jones1991; Nissenbaum Reference Nissenbaum2000).
