Control theory has been a topic of discussion and heated debate for the last few decades, and increasingly so since the publication of Hornstein (Reference Hornstein1999). In the wake of the Minimalist Program, linguists working within the framework of Principles and Parameters were divided into a number of camps. Two such camps, the PRO theorists and the Movement theorists, have been especially active, each group rigorously defending their approach and identifying setbacks in the other group's approach. The volume under review is a collection of eleven chapters brought together by Norbert Hornstein and Maria Polinsky. Its main purpose is to provide further support for the Movement theory of control (MTC) by exploring control phenomena in languages and/or structures that have not been (adequately) studied.

The editors open the volume with their chapter ‘Control as movement: Across languages and constructions’. They argue that the MTC is superior to a theory that makes use of PRO on both theoretical and empirical grounds. Theoretically, PRO is a formative that was postulated for theory-internal reasons in the pre-Minimalism Principles and Parameters framework. The advent of Minimalism made it possible to account for the distribution and interpretation of control structures as movement and thus to dispense of PRO.

Empirically, Hornstein & Polinsky add, the MTC is superior because it is able to account not only for Forward Control, in which the matrix copy is phonologically realized, but also for Backward Control, where it is the subordinate copy that is pronounced. One should add that the MTC, but not PRO theory, is also able to account for Copy Control, as attested in a number of languages, including San Lucas Quiaviní Zapotec (Boeckx, Hornstein & Nunes Reference Boeckx, Hornstein and Nunes2007) and Telugu (Haddad Reference Haddad2009).

More than half of Hornstein & Polinsky's chapter is devoted to the discussion of data that are arguably problematic for the MTC. These include instances of control shift and of control into the complement of promise-type verbs. The authors succeed in showing that these data do not pose a problem for the MTC but in fact constitute evidence in its favor. At the same time, the authors are not shy to admit that there are other data that pose challenges which the MTC is not yet able to resolve. However, as they demonstrate, these data are also problematic for other theories of control and should therefore be considered a challenge more generally.

In my opinion, the chapter that best serves the purpose of the book is Edith Aldridge's ‘Clitic climbing in archaic Chinese: Evidence for the movement analysis of control’. It is an excellent example of how empirical evidence may provide solid support for a theoretical framework. Aldridge uses data from Classical Chinese to bolster the superiority of the MTC over the PRO theory. Classical Chinese licenses clitic climbing from embedded clauses when the subject of the matrix clause is the negative quantifier mo ‘none’, see (1).

1. (1) hu fu yu, mo+zhi gan [yingzhi]tiger back crevice none+3.ojb dare approach 3.ojb ‘The tiger backed into a crevice and no one dared to approach it.’(163, ex. (25b), modified)

The clitic climbing in (1) is mysterious because cliticization is strictly clause-bound in Classical Chinese. Aldridge shows that (1) is not an exception to this restriction. That is, in her analysis, (1) has the underlying structure in (2): the subject mo undergoes first merge in the subordinate clause, and the cliticization of zhi takes place clause-internally; the subject then moves to the matrix clause, pied-piping the pronoun with it. This analysis is possible only under the MTC. The PRO theory cannot account for this type of data without considering (1) to be an exception to the clause-boundedness of cliticization.

1. (2) hu fu yu, mo+zhi gan [mo+zhi yingzhi] tiger back crevice none+3.ojb dare none+3.ojb approach 3.ojb(165, ex. (28a), modified)

Aldridge further addresses two apparent counterexamples to her analysis and manages to subsume these examples under her theory, thereby providing thorough empirical evidence in support of the MTC.

In ‘Movement theory of control and CP-infinitives in Polish’, Jacek Witkoś analyzes control into two types of complements in Polish, viz. bare infinitives or Tense Phrases (TPs), and C(omplementizer)P(hrase)-infinitives. The latter are further divided into two classes: (i) CPs introduced by żeby ‘so that’, which can be strong or weak phases, and (ii) interrogative CPs, which are categorically treated as strong phases.

Witkoś argues that control in Polish is derived by both Move and Agree. The former involves movement of the subject from the embedded clause to the specifier of the matrix vP, as indicated by the arrows placed above in the examples (3)–(6). The latter mediates transmission of case from matrix I⁰ to the embedded adjective, as schematically represented by the arrows placed below in these examples. Crucially, Move and Agree apply independently, and each may operate over a maximum of one strong phase. Thus, in (5) and (6), case transmission is not possible because the matrix I⁰ cannot probe across two strong phases into the complement.

1. (3)

   

2. (4)

   

3. (5)

   

4. (6)

   

Witkoś's account is elegant and clever. If correct, it may resolve the problem of the discrepant behavior of raising and control structures concerning case, which is always brought up as an argument against the MTC. Meanwhile, it remains to be determined whether his account works for languages other than Polish. San Martin (Reference San Martin2004) presents data from several languages showing that the complements of both raising and (exhaustive) control structures are TPs rather than CPs, as the presence of a complementizer rules out both raising and control.

Norbert Hornstein & Paul Pietroski's contribution, ‘Obligatory control and local reflexives: Copies as vehicles for de se readings’, presents semantic evidence to show that locally bound reflexives (LBR) and obligatory control PRO (OC PRO) are the outcome of movement rather than co-indexing. The focus is on verbs like expect and believe; see examples (7)–(8), here used in reference to an amnesiac war hero who accidently came across information about himself.

1. (7)

   1. (a) The unfortunate expected to get a medal.

   2. (b) The unfortunate believed himself to be brave.

1. (8)

   1. (a) The unfortunate_i expected that he_i would get a medal.

   2. (b) The unfortunate_i believed that he_i was brave.

      (68, exx. (2)–(3), modified)

Hornstein & Pietroski argue that the sentences in (7) may not be used in reference to the amnesiac war hero even after his recovery because they are strictly de se reports. That is, to use Langacker's (Reference Langacker, Radden, Köpcke, Berg and Siemund2007) terms, the subject as a conceptualizer is aware that he is the object of conceptualization – or the conceptualized. The sentences in (8), however, are appropriate in this context; although they normally allow a de se interpretation, a de re reading is also available, whereby the conceptualizer is not aware that he is the conceptualized. For Hornstein & Pietroski, the reason behind this semantic distinction is syntactic. In (7), OC PRO and the LBR are construed as the unfortunate as the result of movement/replication. In (8), on the other hand, the pronouns are related to the matrix subjects through co-indexing.

Hornstein & Pietroski's analysis successfully accounts for this relation between syntax and semantics. According to the authors, movement results in semantic identicalness in a way that co-indexing cannot. See, however, Anand (Reference Anand2006) and works cited within for ECM-structures of the believe-type that allow a de re interpretation. Moreover, Hornstein & Pietroski's proposed analysis gives rise to an important question. As observed by Reuland (Reference Reuland2001), in Dutch, a ‘self’-anaphor, unlike its se counterpart, may be distinguishable from its antecedent. For example, imagine that John walks into a museum where there is a statue of John, and he looks into a mirror. The Dutch equivalent of John saw himself, with a ‘self’-anaphor, can then have two interpretations: John saw his own reflection or John saw (the reflection of) his statue. This is not expected under Hornstein & Pietroski's account, which predicts that the reflexive and its antecedent should be identical since they are derivationally related as copies of the same element.

Following the introduction of the MTC as a viable theory of control, the first full-fledged research study that presented empirical challenges to the PRO theory was provided by Polinsky & Potsdam (Reference Polinsky and Potsdam2002), which focused on Backward Control (BC) in the Northeast Caucasian language Tsez. However, according to Landau (Reference Landau, Davis and Dubinksy2007), BC in Tsez is problematic on two counts: (i) it is a rare phenomenon since it is only licensed by two verbs; and (ii) the matrix and subordinate subjects in a BC structure can check separate cases. On the one hand, this means that disjoint subjects should be possible; on the other hand, this is an indication that control cannot be raising, given that the two subjects in raising structures bear the same case. In their chapter, ‘No objections to Backward Control’, Artemis Alexiadou, Elena Anagnostopoulou, Gianina Iordăchioaia & Mihaela Marchis discuss evidence from Romanian and Greek in order to show that the two challenges identified with respect to Tsez do not present counterarguments against the MTC or BC. The authors also try to provide an explanation as to why BC is available in some languages but not in others.

Considering first the limited distribution of BC in Tsez, Alexiadou et al. show that unlike in Tsez, all OC verbs in Romanian and Greek license BC. They run multiple tests to prove that the BC structures under analysis are real instances of BC. This is a warranted move, given that some instances of BC may turn out to be ‘impostors’, as Nayoung Kwon, Philip J. Monahan & Maria Polinsky show convincingly and elegantly in their chapter ‘Object control in Korean: A Backward Control impostor’.

Concerning the second challenge presented by Tsez, Alexiadou et al. show that the matrix and embedded subjects in Romanian and Greek pattern with Tsez in bearing different case markers. The authors attribute this to movement in combination with multiple case checking, and dismiss the distinction between control and raising with regard to case as irrelevant. I disagree with Alexiadou et al. on this point, as it seems to me that the distinction between control and raising with regard to case is one of the issues that the MTC must resolve in order to establish itself as the uniquely viable approach to control within generative syntax. A possible answer to this problem is provided in Witkoś's chapter (see also Haddad Reference Haddad2011: 101–110).

Two chapters in this volume focus on control structures in which the controllee is a possessor inside a Determiner Phrase (DP). These are Ciline Rodrigues's ‘Possessor raising through thematic positions’ and Stanley Dubinsky & Shoko Hamano's ‘Framing the syntax of control in Japanese (and English)’.

Rodrigues analyzes a case of control into the DP domain in Brazilian Portuguese. She looks at structures like (9), in which the subject a Maria is coreferential with the unpronounced possessor of as unhas ‘the nails’. Rodrigues offers ample and thorough evidence to show that (9) is an instance of OC.

1. (9) A Maria_i cortou as unhase _i/*k. the Maria cut-3sg the nails ‘Maria cut her nails.’(125, ex. (12a), modified)

According to Rodrigues, the coreferentiality in (9) is the outcome of movement. A Maria ‘the Maria’ undergoes first merge as the possessor in the lower DP, where it may or may not check genitive case. If it does not check genitive case, it moves to the specifier of IP (SpecIP) in order to check nominative case. On its way to SpecIP, the possessor ‘touches down’ in SpecvP, where it takes on another theta role. If, on the other hand, the possessor checks genitive case DP-internally, the outcome is (10). Note that the possessor dela ‘her’ may or may not take a Maria as its referent.

1. (10) A Maria_i cortou as unhas dela_i/k.the Maria cut-3sg the nails of.she ‘Maria cut her nails.’(124, ex. (11a), modified)

I believe that a slightly modified analysis of these Brazilian Portuguese data is possible. Given the availability of (10) as an alternative of (9) and the fact that the possessor may check case DP-internally, I suggest that (9) is an instance of circumstantial control (Haddad Reference Haddad2010), whereby the possessor moves to SpecvP only when the numeration is exhausted and there is no token to satisfy the thematic requirement of the predicate.

Dubinsky & Hamano are concerned with Japanese structures like (11) and (12). (11) is a control structure, in which the implied possessor of yoko ‘side’ in the adjunct introduced by ni ‘at’ is obligatorily coreferential with the matrix subject. (12) is minimally different from (11) in that the adverbial clause is a ni site ‘at do-te’ adjunct; as a result, it is a non-control structure in the sense that the possessor does not have to be coreferential with the matrix subject.

1. (11) Mari-wa [tue-o yoko ni] tatiagatta. Mari-top cane-acc side at stood.up ‘Mari stood up, with the cane at her side.’

1. (12) Mari-wa [tue-o yoko ni site] tatiagatta. Mari-top cane-acc side at do.TE stood.up ‘Mari stood up, with the cane at her side.’ or: ‘Mari stood up, having laid the cane flat on its side.’(187, ex. (8), modified)

Like Rodrigues, Dubinsky & Hamano analyze (11) as movement. Their focus, however, is on the contrast between (11) and (12). Why does the addition of site – or, more specifically, the addition of the te form – convert the sentence into a non-control structure? The authors hold that te reflects the projection of an event phrase that assigns the possessor an event index and blocks its movement to a thematic position in the matrix clause. Stated differently, only a noun phrase with no event index can move to a thematic position, which is the case in (11). Dubinsky & Hamano extend their analysis to English, showing that what Landau (Reference Landau2004) attributes to tense in the dichotomy of control as exhaustive vs. partial control should be attributed to the absence vs. presence of an event projection.

The approach adopted in this chapter seems to work for Japanese and English. It would be interesting to test its cross-linguistic validity. As I pointed out in a review of an earlier version of this chapter (Haddad Reference Haddad2011: 179–182), Telugu and Assamese also license exhaustive control into adjuncts; however, the adjunct and the matrix clauses may depict two separate events.

Tomohiro Fujii's contribution, ‘Split control and the Principle of Minimal Distance’, focuses on split control structures like (13), in which the subordinate controllee is coreferential with two noun phrases in the matrix clause, each occupying a different argument position.

1. (13) John _i proposed to/asked Mary_k [ _____i+k to meet each other at 6].

   (214, ex. (6b), modified)

Building on Portner & Zanuttini (Reference Portner and Zanuttini2005), Fujii maintains that verbs like propose select for exhortative subordinate clauses, such that the property expressed in the verb phrase may be added to the to-do list of both the speaker and addressee, which results in split control. Propose contrasts with, for example, verbs like recommend and order, which select for imperative embedded clauses. These verbs add the property expressed in the verb phrase to the to-do list of the addressee, and the result is non-split object control.

Fujii presents data from Japanese to show that the subordinate clause of a split control structure is different from its counterpart in a non-split control structure in that it is headed by the exhortative particle yoo, see (14).

1. (14) Taro-wa Hanako-ni bokuno beeguru-o tabe-yoo-to teiansita. Taro-top Hanako-dat my bagel-acceat-YOO-c proposed ‘Taro proposed to Hanako to eat (together) my bagel.’(226, ex. (40b), modified)

The author analyzes split control structures like (13) and (14) as movement plus pied-piping. As (15) illustrates, both noun phrases start out as conjoined elements in the embedded clause. The second noun phrase (NP2) then moves to the indirect object position in the matrix clause, pied-piping NP1 along with it. In this position, only NP2 checks the theta-role feature of V, while NP1 moves to SpecvP, where it checks the theta-role feature of v.

1. (15) [_vP NP1 [_VP [NP1+NP2] [_CP [_MoodP [NP1+NP2] yoo]]]]

Fujii's elegant proposal satisfies minimality but nevertheless raises a number of important questions. To begin with, Fujii does not explain what would prevent [NP1+NP2] from being assigned a theta role by V as a conjoined NP. This should be possible given that the NP1 and NP2 are assigned a theta role collectively in the embedded clause. And this is precisely what happens in (16).

1. (16) I asked John_i and Mary_k [ _____i+k to love each other].

Alternatively, would it be possible for only one of the conjoined elements to move out of the embedded clauses, stranding the other element? In other words, is the Japanese equivalent of (17), where the embedded subject is understood as John and Mary, a possible structure, and if not, why not? After all, the copy of NP1 in the matrix VP in (15) moves to SpecvP without pied-piping NP2.

1. (17) I asked Mary_k [ _____i+k to love each other].

Of course split control structures require that the controllee be coreferential with two separate NPs in the matrix clause, but my point is that it is not clear how this restriction can follow in a principled way under the movement analysis proposed here.

Ivy Sichel's contribution, ‘Towards a typology of control in DP’, and Laura Kertz's ‘The argument structure of evaluative adjectives: A case of pseudo-raising’ are two well-written chapters that deal with two very thorny topics. Sichel provides a typology of control in nominals. And although it is not conclusive, the chapter constitutes much needed work that is likely to trigger more research on control in DPs and the OC–NOC dichotomy.

Kertz analyzes control structures that involve evaluative adjectives, such as smart and silly. These are divided into two types: subject control structures like (18) and sentential subject structures like (19). Kertz convincingly shows that only (18) is the result of syntactic control. The control relation in (19), on the other hand, warrants a semantic rather than a syntactic analysis.

1. (18) Arthur is smart/stupid/silly/rude to press the issue.(272, ex. (13))

2. (19) It is smart/stupid/silly/rude of Arthur to press the issue.(272, ex. (14))

Despite the pressing need for research on these phenomena, Sichel's and Kertz's analyses need not be framed within the MTC. This means that their papers do not straightforwardly contribute to the overall aim of this volume as they neither provide an unambiguous argument in favor of the movement approach nor offer means to constrain it.

The inclusion of Sichel's and Kertz's chapters, as well as the lack of cross-reference among the different chapters, may be considered to be the only drawback in this otherwise very good and much welcome volume. My comments and questions are mostly suggestions for further research; only quality work of the caliber of the various chapters of this volume may give rise to such questions.