2.1 Case

Telugu is a nominative-accusative language. The subject may be structurally case-marked nominative, in which case the verb agrees with it (e.g. (6a, b)). However, if the predicate is an experiential predicate (i.e. a predicate that expresses a physical or emotional state, such as hunger or anger, or possession), the subject is inherently case-marked dative (e.g. (6c)).

1. (6)

   1. (a) Kumaar naaTyam ceesaa-DuKumar.nom dance did-3.m.s‘Kumar danced.’

   2. (b) caalaa mandi naaTyam ceesaa-rumany people.nom dance did-3.m.p‘Many people danced.’

   3. (c) Kumaar-ki koopam waccin-diKumar-dat anger.nom came-3.n.s‘Kumar got angry.’

Evidence that the dative NP in (6c) is the subject comes from the fact that it may function as an antecedent to an anaphor or as the unpronounced argument of the subordinate clause in a control construction. In (7a), Kumaar functions as the antecedent to the anaphor tana-ku tana-miida ‘himself’. In (7b), the dative NP is the unpronounced controllee (presented as Δ), which can only be the subject.

1. (7)

   1. (a) Kumaar-ki tana-ku tana-miida koopam waccin-diKumar-dat him-dat him-upon anger.nom came-3.n.s‘Kumar got angry with himself.’

   2. (b) [Kumaar_i [Δ_i koopam raawaTam] iShTapaDa-Du]Footnote ⁴Kumar.nom Δ.dat anger.nom coming doesn't.like-3.m.s‘Kumar does not like to get angry.’

Unlike nominative NPs, however, dative NPs do not trigger agreement on the verb. In (7a) above, the verb agrees with the (neuter) nominative NP koopam ‘anger (nom)’ rather than with the (masculine) dative NP Kumaar-ki ‘Kumar (dat)’. For further illustration, the sentences in (8) also have dative subjects. The verb in (8a) agrees with the singular NP paniwaaDu ‘servant (nom)’, whereas in (8b) it agrees with the plural NP paniwaaLLu ‘servants (nom)’.

1. (8)

   1. (a) atani-ki paniwaaDu unnaa-Duhe-dat servant.nom is-3.m.s‘He has a servant.’

   2. (b) atani-ki paniwaaLLu unnaa-ruhe-dat servants.nom are-3.m.p‘He has servants.’

The following subsections highlight the characteristics of clauses in Telugu.

2.2 Finite clauses in Telugu

Finite declarative clauses in Telugu take verbs that are inflected for tense and agreement. The verb may belong to one of the following morphological paradigms: past, non-past (future or habitual), or negative (future or habitual). Concerning agreement, verbs inflect for person, gender (only with 3rd person), and number. For example, the verb ammu ‘to sell’ takes one of the forms in (9a–c) if used in a declarative finite clause with a 3rd person feminine subject; compare to (9d–f), in which the verb agrees with a 3rd person masculine subject. The variation in suffix forms is morphophonological (Krishnamurti Reference Krishnamurti and Sanford1997: 216–221).Footnote ⁵

1. (9)

   1. (a) amm-in-dusell-past-3.n.s‘she sold’

   2. (b) ammu-tun-disell-nonpast-3.n.s‘she sells/will sell’

   3. (c) amm-a-dusell-neg-3.n.s‘she won't/doesn't sell’

   4. (d) amm-aa-Dusell-past-3.m.s‘he sold’

   5. (e) ammu-taa-Dusell-nonpast-3.m.s‘he sells/will sell’

   6. (f) amm-a-Dusell-neg-3.m.s‘he won't/doesn't sell’

2.3 Non-finite clauses in Telugu

Telugu has two types of non-finite subordinate clauses that function as adjuncts. I will refer to the first type as infinitive (INF) clauses; the second type is known as adverbial or conjunctive participle (CNP) clauses (Linholm Reference Linholm1975 and Klaiman Reference Klaiman1981). The latter is central to the domain of investigation of this paper.

3.1 Forward Control

The sentences in (17) are instances of Forward Control in which the matrix subject is pronounced, determining the identity of the unpronounced subject in the adjunct clause.Footnote ⁷

1. (17)

   1. (a) [Kumaar_i [Δ_i koopam wacc-i] naa illu kuulcaa-Du]Kumar.nom Δ.dat anger get-part.cnp my house destroyed-3.m.s‘Having got angry, Kumar destroyed my house.’

   2. (b) [Kumaar_i [Δ_i laybrarii-ki weLL-i] pustakam cadiwaa-Du]Kumar.nom Δ.nom library-dat go-part.cnp book read-3.m.s‘Having gone to the library, Kumar read a book.’

In (17a), the case-marking of the pronounced subject shows that it is licensed by the matrix predicate and that the structure is an instance of Forward Control. In (17b), on the other hand, the matrix and CNP subjects would be case-marked the same. This means that the structure can be an instance of Forward Control, but it may also be analyzed as an instance of Backward Control. In the latter case, the CNP subject is overt and the matrix subject is implied, as illustrated in (18). Both types of control result in the same word order.

1. (18) [Δ_i [Kumaar_i laybrarii-ki weLL-i] pustakam cadiwaa-Du]Δ.nom Kumar.nom library-dat go-part.cnp book read-3.m.s‘Having gone to the library, Kumar read a book.’

In order to make sure that (17b) may be analyzed as an instance of Forward Control, I repeat it as (19a), only this time with the CNP clause realized sentence-initially. Given that the locus of the verb in Telugu is strictly clause-final, (19a) can only be analyzed as a Forward Control construction in which the pronounced subject is licensed by the matrix predicate; it may not be analyzed as a Backward Control construction (e.g. (19b)). Similarly, the grammatical sentence (17a) above may not be realized with a postverbal CNP subject (e.g. (20a)); compare to (20b), which has the same word order as (20a), but where the subject is now licensed by the matrix predicate.

1. (19)

   1. (a) [[Δ_i laybrarii-ki weLL-i] Kumaar_i pustakam cadiwaa-Du]Δ.nom library-dat go-part.cnp Kumar.nom book read-3.m.s‘Having gone to the library, Kumar read a book.’

   2. (b) *[[laybrarii-ki weLL-i Kumaar_i] Δ_i pustakam cadiwaa-Du]library-dat go-part.cnp Kumar.nom Δ.nom book read-3.m.s‘Having gone to the library, Kumar read a book.’

1. (20)

   1. (a) *[[koopam wacc-i Kumaar-ki_i] Δ_i naa illu kuulcaa-Du]anger.nom get-part.cnp Kumar-dat Δ.nom my house destroyed-3.m.s‘Having got angry, Kumar destroyed my house.’

   2. (b) [[Δ_i koopam wacc-i] Kumaar_i naa illu kuulcaa-Du]Δ.dat anger.nom get-part.cnp Kumar.nom my house destroyed-3.m.s‘Having got angry, Kumar destroyed my house.’

Adjunct control in Telugu qualifies as Obligatory Control in the sense that the CNP subject has to take the matrix subject as an antecedent. The CNP subject cannot be coreferential with any other NP in the sentence (e.g. the possessor of the matrix subject or a non-local NP); it cannot take a split antecedent (e.g. the matrix subject plus another NP in the sentence); and it cannot be coreferential with an NP selected from surrounding discourse (see Williams Reference Williams1980, Hornstein Reference Hornstein1999, Jackendoff & Culicover Reference Jackendoff and Culicover2003, and Polinsky & Potsdam Reference Polinsky and Potsdam2004, among others).

To illustrate, in (21a) the CNP subject takes the matrix subject as an antecedent; the sentence is grammatical. In (21b), the antecedent is the possessor of the matrix subject atani ‘his’ or the dative NP atani-ki ‘him (dat)’; the sentence is ungrammatical under the designated reading. Similarly, sentences (22a, b) illustrate that the CNP subject may only take the matrix subject as an antecedent, and that a reading with a split antecedent is not available. Note that in (22a) one of the antecedents is a non-local NP, while in (22b) both antecedents are local. Still, both sentences are ungrammatical on the split-antecedent reading.

1. (21)

   1. (a) [Kumaar_i [Δ_i aakali wees-i] saandwic tinnaa-Du]Kumar.nom Δ.dat hunger.nom fall-part.cnp sandwich ate. 3.m.s‘Having got hungry, Kumar ate a sandwich.’

   2. (b) *[[atani_i amma]_k [Δ_i aakali wees-i] atani-ki annam peTTin-di]his mother Δ.dat hunger.nom fall-part.cnp him-dat food put-3.n.sIntended meaning: ‘He got hungry, and his mother gave him food.’

1. (22)

   1. (a) [[Kumaar_i [Δ_i+k sinimaa cuus-tuu] paapkaarn tinnaa-Du aNi] Sarita_k ceppin-di]Kumar.nom Δ.nom movie watch-dur.cnp popcorn ate-3.m.s so/that Sarita.nom said-3.n.s‘Sarita said that Kumar ate popcorn while he was watching a movie.’*‘Sarita said that Kumar ate popcorn while they were watching a movie.’

   2. (b) [Kumaar_i [Δ_i+k sinimaa cuus-tuu] Sarita-ki_k paapkaarn peTTaa-Du]Kumar.nom Δ.nom movie watch-dur.cnp Sarita-dat popcorn put-3.m.s‘Kumar gave Sarita popcorn while he was watching a movie.’*‘Kumar gave Sarita popcorn while they were watching a movie.’

Sentence (23a) is similar to (21a) in that the reference of the CNP subject is determined by the reference of the matrix subject; the sentence is acceptable. In (23b), however, the antecedent is selected from surrounding discourse (speaker, hearer, or a person mentioned earlier in discourse). The sentence is ungrammatical under the designated reading.

1. (23)

   1. (a) [Kumaar_i [Δ_i jwaram wacc-i] mandulu waaDaa-Du]Kumar.nom Δ.dat fever.nom come-part.cnp medicines used-3.m.s‘Having had a fever, Kumaar took medication.’

   2. (b) *[Sarita [Δ_i/k/j jwaram wacc-i] (naa-ku_i/mii-ku_k/atani-ki_j)Footnote ⁸ mandulu iccin-di]Sarita.nom Δ.dat fever.nom come-part.cnp me-dat/you-dat/him-dat medicines gave-3.n.sIntended meaning: ‘I/You/He (mentioned earlier in discourse) had a fever, and Sarita gave me/you/him medication.’

In addition, Telugu adjunct control can be classified as Exhaustive Control, in that the identities of the CNP and matrix subjects must fully coincide. Stated differently, Telugu does not allow Partial Control into CNP clauses. Partial Control obtains when the reference of the unpronounced subject does not necessarily fully coincide with the reference of the overt subject (Landau Reference Landau2004: 833). To illustrate from English, in (24a) the unpronounced subject (symbolized by Δ) refers to the manager plus a particular group with whom the manager prefers to work. Compare to (24b), in which Δ can only refer to the manager; it cannot include other individuals. Sentence (24a) is an example of Partial Control, and sentence (24b) of Exhaustive Control.

1. (24)

   1. (a) The manager prefers [Δ to work on the project together].

   2. (b) The manager forgot [Δ to work on the project (*together)].

Now observe sentence (25). The matrix subject Kumaar functions as an antecedent to the unpronounced CNP subject. The sentence is considered grammatical under the reading that Kumar watched a movie and ate a sandwich. However, it is judged as unacceptable under the reading that Kumar and a designated person or group watched a movie, and Kumar ate a sandwich.

1. (25) [Kumaar_i [Δ_{i/*i+k(person or group)} sinimaa cuus-tuu] saandwic tinnaa-Du]Kumar.nom Δ.nom movie watch-dur.cnp sandwich ate-3.m.s

The above discussion shows that Forward Control into CNP clauses qualifies as Obligatory Control. The following section shows that Backward Control is similar in this respect.

3.2 Backward Control

In Backward Control, the subordinate subject is pronounced, determining the identity of the unpronounced subject in the matrix clause. The sentences in (26) are examples.

1. (26)

   1. (a) [Δ_i [Kumaar-ki_i Sarita-miida koopam wacc-i] akkadi-nunci weLLipoyinaa-Du]Δ.nom Kumar-dat Sarita-on anger.nom come-part.cnp there-from left-3.m.s‘Having got angry at Sarita, Kumar left.’

   2. (b) [Δ_i [Kumaar_i laybrarii-ki weLL-i] pustakam cadiwaa-Du]Δ.nom Kumar.nom library-dat go-part.cnp book read-3.m.s‘Having gone to the library, Kumar read a book.’

Sentence (26a) is clearly a Backward Control construction, as the case-marking of the pronounced subject indicates. Sentence (26b), on the other hand, could also be analyzed as an instance of Forward Control (e.g. (27)).

1. (27) [Kumaar_i [Δ_i laybrarii-ki weLL-i] pustakam cadiwaa-Du]Kumar.nom Δ.nom library-dat go-part.cnp book read-3.m.s‘Having gone to the library, Kumar read a book.’

Evidence that (26b) may also qualify as an example of Backward Control comes from scrambling. To elaborate, sentence (28a) is an example of Backward Control. It is minimally different from (26a) in that Sarita-miida ‘on Sarita’ is scrambled to a clause-initial position, past the CNP subject. The sentence is acceptable. Compare to the Forward Control equivalent in (28b). Sentence (28b) is ungrammatical because it involves scrambling beyond the boundaries of the adjunct/CNP clause.

1. (28)

   1. (a) [Δ_i [Sarita-miida Kumaar-ki_i koopam wacc-i] akkadi-nunci weLLipoyinaa-Du]Δ.nom Sarita-on Kumar-dat anger.nom come-part.cnp there-from left-3.m.s‘Having got angry at Sarita, Kumar left.’

   2. (b) *[Sarita-miida Kumaar_i [Δ_i koopam wacc-i] akkadi-nunci weLLipoyinaa-Du]Sarita-on Kumar.nom Δ.dat anger.nom come-part.cnp there-from left-3.m.s‘Having got angry at Sarita, Kumar left.’

Now consider the grammatical structure in (29). Based on the discussion of (28), we can conclude that (29a) can be acceptable only if it is taken as (29b), where the scrambling of laybrarii-ki ‘to the library’ takes place within the boundaries of the CNP clause. Sentence (29c) is similar to (28b) in that it involves scrambling outside the boundaries of the CNP clause. Thus, it should be considered ungrammatical. This indicates that (29a) qualifies as an example of Backward Control. By the same token, (26b) above can qualify as an instance of Backward Control.

1. (29)

   1. (a) laybrarii-ki Kumaar weLL-i pustakam cadiwaa-Dulibrary-dat Kumar.nom go-part.cnp book read-3.m.s‘Having gone to the library, Kumar read a book.’

   2. (b) [Δ_i [laybrarii-ki Kumaar_i weLL-i] pustakam cadiwaa-Du]Δ.nom library-dat Kumar.nom go-part.cnp book read-3.m.s‘Having gone to the library, Kumar read a book.’

   3. (c) *[laybrarii-ki Kumaar_i [Δ_i weLL-i] pustakam cadiwaa-Du]library-dat Kumar.nom Δ.nom go-part.cnp book read-3.m.s‘Having gone to the library, Kumar read a book.’

Like their Forward Control counterparts, Backward Control structures are instances of Obligatory Control. If the CNP subject fails to fully determine the identity of the matrix subject, the result is ungrammaticality. Observe the sentences in (30). The two subjects in each sentence are interpreted as coreferential; both sentences are grammatical. Compare to the sentences in (31). In (31a), the CNP subject fails to determine the identity of the matrix subject. Rather, the reference of the matrix subject coincides with the reference of the non-local NP ‘Sarita’; the sentence is unacceptable. In (31b, c), the CNP subject takes the matrix subject plus another NP as a split antecedent. In (31b), the other NP is non-local, while in (31c) it is local. In both sentences, the reference for the CNP subject does not fully coincide with the reference of the matrix subject; as a result, the sentences are unacceptable. In (31d, e), the unpronounced matrix subject is coreferential with an NP selected from surrounding discourse. This means that no interpretation dependency could be established between the CNP and the matrix subject. The result is ungrammaticality.

1. (30)

   1. (a) [Δ_i [Kumaar-ki_i aakali wees-i] saandwic tinnaa-Du]Δ.nom Kumar-dat hunger.nom fall-part.cnp sandwich ate. 3.m.s‘Having got hungry, Kumar ate a sandwich.’

   2. (b) [Δ_i [Kumaar-ki_i jwaram wacc-i] mandulu waaDaa-Du]Δ.nom Kumar-dat fever.nom come-part.cnp medicines used-3.m.s‘Having had a fever, Kumar took medication.’

1. (31)

   1. (a) *[[Δ_i [Kumaar_k sinimaa cuus-tuu] paapkaarn tinna-di aNi] Sarita_i ceppin-di]Δ.nom Kumar.nom movie watch-dur.cnp popcorn ate-3.n.s so/that] Sarita.nom said-3.n.sIntended meaning: ‘Sarita said that while Kumar was watching a movie, she ate popcorn.’

   2. (b) *[[Δ_i [Kumaar_iunnuSarita_k sinimaa cuus-tuu] paapkaarn tinnaa-Du aNi] Sarita_k ceppin-di]Δ.nom Kumar.nom and Sarita.nom movie watch-dur.cnp popcorn ate-3.m.s so/that Sarita.nom said-3.n.sIntended meaning: ‘Sarita said that while Kumar and Sarita were watching a movie, Kumar ate popcorn.’

   3. (c) *[Δ_i [Kumaar_iunnuSarita_k sinimaa cuus-tuu] Sarita-ki_k paapkaarn peTTaa-Du]Δ.nom Kumar.nom and Saita.nom movie watch-dur.cnp Sarita-dat popcorn put-3.m.sIntended meaning: ‘Kumar gave Sarita popcorn while Kumar and Sarita were watching a movie.’

   4. (d) *Δ_k [Kumaar-ki_i aakali wees-i] atani-ki annam peTTaa-nu/peTTaa-wu/peTTin-di Δ.nom Kumar-dat hunger.nom fall-part.cnp him-dat food put-1.s/put-2.s/put-3.n.sIntended meaning: ‘Kumar got hungry, and I/you/she [e.g. his mother] gave him food.’

   5. (e) *[Δ_i [Kumaar-ki_k jwaram wacc-i] (atani-ki_k) mandulu iccaa-nu/iccaa-wu/iccin-di]Δ.nom Kumar-dat fever.nom come-part.cnp him-dat medicines gave-1.s/gave-2.s/gave-3.n.sIntended meaning: ‘Kumar had a fever, and I/you/she [mentioned earlier in discourse] gave him medication.’

Based on the above discussion, we can conclude that Backward Control into CNP clauses in Telugu is Obligatory Control.

A point is in order before we proceed. Given that Telugu is a pro-drop language (see section 2), one might try to argue that in each of the grammatical instances of Forward and Backward Control presented in this section the unpronounced subject is a base-generated pro. As mentioned in section 2, although Telugu is referred to in the literature as a pro-drop language, pro and lexicalized NPs are actually in free variation. This means that if the unpronounced subject in Forward and Backward Control were truly a base-generated pro, one would expect the sentences to be grammatical with lexicalized NPs. This is not the case, however, as sentences (32a, b) show. Note that these sentences are ungrammatical regardless of how the matrix subject or the CNP subject is realized – whether as an overt pronoun or as an r-expression. The sentences would be grammatical if either the matrix subject or the CNP subject were unpronounced.

1. (32)

   1. (a) *[Kumaar/atanu [Kumaar-ki/atani-ki aakali wees-i] Saandwic tinnaa-Du]Kumar/he.nom Kumar/he-dat hunger.nom fall-part.cnp Sandwich ate. 3.m.s‘Having got hungry, Kumar ate a sandwich.’

   2. (b) *[Kumaar/atanu [Kumaar-ki/atani-ki jwaram wacc-i] mandulu waaDaa-Du]Kumar/he.nom Kumar/he-dat fever.nom come-part.cnp medicines used-3.m.s‘Having had a fever, Kumar took medication.’

Nevertheless, under the right conditions, adjunct control structures in Telugu may be realized with two pronounced subjects, which brings us to the central topic of this paper: Copy Control. Section 4 presents the empirical picture. In section 5, I provide an analysis using the movement theory of control. I suggest that decisions concerning the pronunciation/deletion of copies in the different types of adjunct control (Forward, Backward, Copy) are prepared for in the syntax, but are actually made at PF.

4.1 Copy Control: conditions and restrictions

A closer look at the grammatical Copy Control structures presented above shows that the CNP clause is always realized sentence-initially, and that the CNP subject is never a pronominal. In fact, Copy Control obtains only if these two conditions are met:

• • Condition 1: The CNP clause is sentence-initial.

• • Condition 2: The CNP subject is an r-expression (non-pronominal).

Concerning the first condition, if the CNP clause is realized sentence-internally, a Copy Control construction becomes ungrammatical, as illustrated in (39a). A JL reviewer commented that the restriction might be on adjacent subjects, which is why longer sentences (e.g. (38) above) are judged as less redundant. Sentence (39b) shows that even if material intervenes between the two subjects, a Copy Control structure with a sentence-internal CNP clause is still unacceptable.

1. (39)

   1. (a) *[atanu/Kumaar [atanu-ki/Kumaar-ki aakali wees-i] bhoojanamu tinnaa-Du]he/Kumar.nom he/Kumar-dat hunger.nom fall-part.cnp dinner ate-3.m.s]‘Having got hungry, Kumar ate dinner.’

   2. (b) *[atanu/Kumaar [enimidiki Sarita-too atanu/Kumaar naTyamu cees-tuu] aami-ki kata ceppaa-Du]he/Kumar.nom at 8:00 Sarita-with he/Kumar.nom dance do-dur.cnp her-dat story told-3.m.s‘While dancing with Sarita at 8:00, Kumar told her a story.’

The second condition states that the CNP subject has to be an r-expression. By r-expression, I mean any NP that is not pronominal. The data introduced thus far involve proper nouns. This is not a requirement, however, as (40) illustrates.

1. (40) [[naa baas-ki pooyina-waaram jwaram wacc-i] atanu/naa baas muuDu roojulu mandulu waaDaa-Du]my boss-dat last-week fever.nom come-part.cnp he/my boss.nom three days medicines used ‘My boss had a fever last week, and he/my boss was on medication for three days.’

If a pronominal is used as the CNP subject, a Copy Control construction becomes ungrammatical (e.g. (41)). Note that the CNP subject has to be an r-expression – that is, it cannot be a pronoun – even if the matrix subject is a pronoun. I consider this as a language-specific restriction. Other South Asian languages (e.g. Dakkhini) allow Copy Control structures in which both subjects are pronouns (see section 6.3 below).

1. (41) *[[atani-ki aakali wees-i] atanu/Kumaar bhoojanamu tinnaa-Du]he-dat hunger.nom fall-part.cnp he/Kumar.nom dinner ate-3.m.s‘He got hungry, and Kumar ate dinner.’

Before proceeding to the following section, a word about the two coreferential subjects in Copy Control is appropriate. It can be observed that the matrix subject in a Copy Control structure may be realized as an exact copy of the CNP subject. This applies only if the CNP subject does not exceed one or two words. For example, the CNP subject in (42) is a conjunct. In this case, only a pronoun or an epithet may be used as a subject in the matrix clause. I do not analyze this restriction here, but simply present the facts for descriptive completeness.

1. (42) [[KumaarunnuSarita sinimaa cuus-tuu] waLLu/aa pichooLLu/*Kumaar unnu Sarita paapkaarn tinnaa-ru]Kumar.nom and Sarita.nom movie watch-dur.cnp they/those idiots/*Kumar.nom and Sarita.nom popcorn ate-3.m.p‘While Kumar and Sarita were watching a movie, they/the idiots ate popcorn.’

4.2 Copy Control as Obligatory Control

Like their Forward and Backward Control counterparts, Copy Control constructions are instances of Obligatory Control. The referential properties of the subordinate CNP subject and those of the matrix subject have to fully coincide. A referential dependency between the CNP subject and any other NP in the sentence results in ungrammaticality. To illustrate, in (43) the CNP subject refers to the possessor of the matrix subject, and the sentence is ungrammatical. Similarly, (44) and (45) are unacceptable because the CNP subject refers to a split antecedent. In addition, the reference of the CNP subject may not be determined by surrounding discourse, including speaker/hearer (e.g. (46)).

1. (43) *[[Kumaar-ki aakali wees-i] [atani amma] atani-ki annam peTTin-di]Kumar-dat hunger.nom fall-part.cnp his mother him-dat food put-3.n.sIntended meaning: ‘Kumar having got hungry, his mother gave him food.’

2. (44) *[[[SaritaunnuKumaar sinimaa cuus-tuu] Kumaar paapkaarn tinna-Du aNi] Sarita ceppin-di]Sarita.nom and Kumar.nom movie watch-dur.cnp Kumar.nom popcorn ate-3.m.s so/that Sarita.nom said-3.n.sIntended meaning: ‘Sarita said that Kumar ate popcorn while they were watching a movie.’

3. (45) *[[SaritaunnuKumaar sinimaa cuus-tuu] KumaarSarita-ki paapkaarn peTTaa-Du]Sarita.nom and Kumar.nom movie watch-dur.cnp] Kumar.nom Sarita-dat popcorn put-3.m.sIntended meaning: ‘Kumar gave Sarita popcorn while they were watching a movie.’

4. (46) *[[naa-ku/mii-ku/Kumaar-ki jwaram wacc-i] aame/Sarita (naa-ku/mii-ku atani-ki) mandulu iccin-di]I/you/Kumar-dat fever.nom come-part.cnp she/Sarita.nom I/you/him-dat medicines gave-3.n.sIntended meaning: ‘I/You/Kumar had a fever, and she/Sarita gave me/you/him medication.’

In the following section, I present a syntactic analysis of Telugu adjunct control. Following Nunes (Reference Nunes1995, Reference Nunes2004) and Hornstein (Reference Hornstein1999, Reference Hornstein and Randall2003), I analyze adjunct control as derived via sideward movement. I highlight the conditions that drive and constrain each type of control, focusing especially on Copy Control. Section 6 brings to light a theory-internal problem in the derivation. Contrary to the facts, the theory predicts that the derivation presented in section 5 will not converge. Section 6 proceeds to show that the problem can be solved by adopting Uriagereka's (Reference Uriagereka, Samuel and Norbert1999) Multiple Spell-Out.

5.1 Adjunct control as movement

The Copy Control structures under examination involve control into adjuncts. It is well-known that adjuncts are islands to movement. An exception to this observation is parasitic gap constructions which, according to Nunes (Reference Nunes1995, Reference Nunes2004), are derived via sideward movement out of an adjunct. Sideward movement is an operation that allows movement between two unconnected syntactic objects.

Under the Copy Theory of Movement as proposed by Chomsky (Reference Chomsky1995), all movement takes place between two positions that are in a c-command relationship. A constituent undergoes copy-plus-merge into a c-commanding position and the two copies form a chain. In Nunes's system, the Copy Theory of Movement is reformulated as the Copy-plus-Merge Theory of Movement. According to this theory, movement comprises four independent operations: Copy, Merge, Form Chain, and Chain Reduction. The two operations Copy and Merge are in principle similar to Chomsky's, except that chain formation does not follow naturally from them. Form Chain is an independent operation, the essence of which is captured in (49). Note that (49) is a simplified version of Nunes's (Reference Nunes2004: 91) original formulation.

1. (49) Form Chain: Two constituents X and Y form a chain iff

   1. (a) X and Y are non-distinct copies of the same token; that is, they are identical copies related through movement.

   2. (b) X c-commands Y.

With Form Chain as an independent step, movement does not have to target a c-commanding position. In other words, movement between two unconnected syntactic objects is now possible. For example, α in (50) may copy out of the syntactic object L and merge in the unconnected syntactic object M, as (55a) illustrates. Subsequently, L and M undergo merge in (55b). This type of movement is called sideward movement (Nunes Reference Nunes2004). Note that if L is an adjunct, it becomes an island after – not before – merging with M.

1. (50)

   1. (a)

      

   2. (b) [_m [_l α …] [_m α […]]]

While the three operations Copy, Merge, and Form Chain take place in the syntax, Chain Reduction takes place in the phonological component. According to Nunes, if two elements form a chain, one of them has to be deleted. Both operations, Form Chain and Chain Reduction, take place for the purpose of linearization. They satisfy the Linear Correspondence Axiom in (51), which dictates that an element cannot both asymmetrically c-command and be asymmetrically c-commanded by the same element in a structure. By the same token, an element cannot follow and precede itself, as this induces a violation of irreflexivity. Asymmetry and irreflexivity are defined in (52)–(53), respectively (Nunes Reference Nunes2004: 24). To satisfy the Linear Correspondence Axiom, Chain Reduction applies at PF. This PF operation reads as (54) (Nunes Reference Nunes2004: 27 ex. (44)).

1. (51) Linear Correspondence Axiom (LCA)

   Let X, Y be nonterminals and x, y terminals such that X dominates x and Y dominates y. Then if X asymmetrically c-commands Y, x precedes y.

   (Kayne Reference Kayne1994: 33)

1. (52) Asymmetry

   If x precedes y, y necessarily does not precede x.

1. (53) Irreflexivity

   If x precedes y, then x and y are distinct copies.

1. (54) Chain Reduction

   Delete the minimal number of constituents of a nontrivial chain CH that suffices for CH to be mapped into a linear order in accordance with the LCA.

Hornstein (Reference Hornstein1999, Reference Hornstein and Randall2003) follows Nunes and considers adjunct control structures as instances of sideward movement. I adopt the same approach in order to account for adjunct control in Telugu. To illustrate, observe the Forward Control structure (55). Following Hornstein and Nunes, I propose that (55) has the derivation in (56). In (56a), the CNP clause (CNPP) and the matrix clause form independently, and Kumaar copies out of the CNP clause. In (56b), Kumaar merges in the matrix clause. The copy-plus-merge operation between the two unconnected syntactic objects, the CNP and the matrix clauses, is an instance of sideward movement.Footnote ⁹ Subsequently, the CNP clause adjoins to matrix vP, as shown in (56c). Upon adjunction, the CNP clause becomes an island. In (56d), the matrix subject Kumaar moves from Spec,vP to Spec,IP to check the EPP feature. As the dotted arrows in (56e) show, the copy of Kumaar in Spec,IP c-commands both the copy in the CNP clause and the copy in Spec,vP, forming a chain with each – thus, Form Chain. The pronunciation of all the non-distinct copies of Kumaar at PF would induce a violation of the irreflexivity condition in (53). The reason is that Kumaar ends up preceding and following itself. This is why the PF operation Chain Reduction applies in (56f). This is when the lower copy in each chain is deleted in order for the structure to be linearized in accordance with the Linear Correspondence Axiom in (51).

1. (55) [Kumaar [Kumaar-ki jwaram wacc-i] haaspaTal weLLaa-Du] Kumar.nomKumar-dat fever.nom come-part.cnp hospital went-3.m.s‘Having had a fever, Kumar went to the hospital.’

1. (56)
   1. (a)

      1. (i)

         

      2. (ii)

         

   2. (b) [_{Matrix vP} [_NPKumaar] haaspaTal weLLaa-Du]

   3. (c) [_{Matrix IP} [_vP [_CNP [_NPKumaar-ki] jwaram wacc-i] [_vP [_NPKumaar] haaspaTal weLLaa-Du]]]

   4. (d) [_{Matrix IP} [_NPKumaar] [_vP [_CNP [_NPKumaar-ki] jwaram wacc-i]

      [_vP [_NPKumaar] haaspaTal weLLaa-Du]]]

   5. (e)

      

   6. (f) At PF:

      [_{Matrix IP} [_NPKumaar] [_vP [_CNP[_NPKumaar-ki] jwaram wacc-i]

      [_vP[_NPKumaar] haaspaTal weLLaa-Du]]]

The outcome of (56f) can be slightly different. As (56f′) below shows, in the chain {[_NP Kumaar]^{Matrix IP}, [_NP Kumaar-ki]^CNP}, the higher copy may get deleted; the outcome will be Backward Control.

1. (56f′) At PF:

   [_{Matrix IP}[_NPKumaar] [_vP [_CNP [_NPKumaar-ki] jwaram wacc-i] [_vP[_NPKumaar] haaspaTal weLLaa-Du]]]

Note that this free variation between Forward and Backward Control is not unique to Telugu. For example, it is also attested in Malagasy Object Control, (e.g. (57a, b); Potsdam Reference Potsdam2006: 1 exx. (1a, b)).

1. (57) Malagasy

   1. (a) naneren'i Mery ny zaza [hofafana ny zaza ny trano …forced Mary the child sweep the child the house …‘Mary forced the child to sweep the house.’

   2. (b) naneren'i Mery ny zaza [hofafan' ny zaza ny trano …forced Mary the child sweep the child the house …‘Mary forced the child to sweep the house.’

To summarize, according to Nunes (Reference Nunes2004), movement comprises four independent operations: Copy, Merge, Form Chain, and Chain Reduction. When a syntactic object α moves, it copies out of a subordinated domain and merges into a subordinating domain. Subsequently, the two copies of α form a chain in accordance with Form Chain as formulated in (49) above. At PF, the structure is linearized in order to satisfy the Linear Correspondence Axiom in (51). The main idea behind (51) is that at PF a syntactic object cannot follow and precede itself at the same time. This is when Chain Reduction, as stated in (54), applies. As a result only one of the copies of α is phonologically realized.

Before proceeding to Copy Control, recall from section 3.1 that adjunct control structures may be realized with a sentence-initial CNP clause and one pronounced subject (e.g. (58a)). Since Telugu is a pro-drop language, it might be assumed that the subject in the sentence-initial CNP clause is a silent pronoun pro, in which case (58a) would be analyzed as an instance of Copy Control. This conclusion is not viable, however. As I mention in section 4.1, the CNP subject in Copy Control structures has to be an r-expression; it cannot be a pronominal (e.g. (58b)). Therefore, pro is ruled out.

1. (58)

   1. (a) [[Δ_i koopam wacc-i] Kumaar_i naa illu kuulcaa-Du]Δ.dat anger get-part.cnp Kumar.nom my house destroyed-3.m.s]‘Having got angry, Kumar destroyed my house.’

   2. (b) *[[atani-ki_i koopam wacc-i] Kumaar_i naa illu kuulcaa-Du]he-dat anger get-part.cnp Kumar.nom my house destroyed-3.m.s‘Having got angry, Kumar destroyed my house.’

I suggest that the CNP clause in (58a) is base-generated at vP of the matrix clause before it moves to the sentence-initial position where it is pronounced. In other words, (58a) has the structure in (59). The copy of the subject in Spec,IP of the matrix clause c-commands both the copy in Spec,vP and the copy in the lower CNP clause (CNPP₁). It forms a chain with each of them. At PF, Chain Reduction applies, and the lower copy in each chain is deleted. Further, the two copies of the CNP clause, CNPP₂ and CNPP₁, also form a chain. At PF, the lower copy is deleted.

1. (59)

   

Now the question is: how does the subject in CNPP₂ get deleted? The answer resides in the particular type of movement that the CNP clause undergoes. The movement of the CNP clause in (59) is commonly referred to as remnant movement. This involves movement of a constituent out of which extraction has taken place (Müller Reference Müller2000). In the case of (59), this means that the CNP clause moves to matrix CP after the CNP subject has already moved to the matrix clause.

In order to account for a similar case of remnant movement and deletion of copies, Nunes (Reference Nunes2004: 50–55) adopts a more elaborate definition of chains and chain links. Following Chomsky (Reference Chomsky1995: 300), he holds that ‘the individual links of a chain must … be identified not only in terms of their content, but also in terms of their local structural configuration’.

To illustrate, consider the chain {[Kumaar], [Kumaar-ki]} in (59) above. It is made of the copy of the subject in Spec,IP of the matrix clause and the copy of the subject in the CNP clause (CNPP₁). Nunes holds that the two copies must be identified, not only in terms of their content as Kumaar, but also in terms of their local structural configuration. That is, the chain {[Kumaar], [Kumaar-ki]} must be presented as (60a) in which one link is identified as the sister of matrix I′ and the other link as the sister of CNP′ of the CNP clause. At PF, Chain Reduction instructs the phonological component to delete the occurrence of Kumaar-ki that has the structural configuration ([Kumaar-ki], [_CNP′ koopam wacc-i]). Two such copies exist in (59), one in CNPP₁ and one in CNPP₂. As Nunes (Reference Nunes2004: 54) maintains, ‘assuming that the phonological component blindly scans the structure to carry out the deletion instructed by Chain Reduction’, it ends up deleting the two copies of ([Kumaar-ki], [_CNP′ koopam wacc-i]) in CNPP₁ and CNPP₂. Given that CNPP₁ and CNPP₂ also form a chain, one of them has to be deleted. In this case, CNPP₁ undergoes deletion, as (60b) shows. Note that the instance of Kumaar in the lower CNP clause in (60b) has double strikethrough, one for being deleted itself, and one because the whole CNP clause is deleted.

1. (60)

   1. (a) {([Kumaar], [_I′ …]]), ([Kumaar-ki], [_CNP′ koopam wacc-i])}

   2. (b) [[Kumaar-ki koopam wacc-i] Kumaar[Kumaar-ki koopam wacc-i] naa illu kuulcaa-Du] Kumar-dat anger get-part.cnp] Kumar.nomKumar-dat anger get-part.cnp my house destroyed-3.m.s‘Having got angry, Kumar destroyed my house.’

Note that the outcome can be slightly different. At PF, Chain Reduction may instruct the phonological component to delete the occurrence of Kumaar that has the structural configuration ([Kumaar], [_I′ …]). In this case, it is the copy of Kumaar in the matrix clause that undergoes deletion, and the result is Backward Control, as (61) illustrates.

1. (61) [[Kumaar-ki koopam wacc-i] Kumaar[Kumaar-kikoopamwacc-i] naa illu kuulcaa-Du][[Kumar-dat anger get-part.cnp] Kumar.nomKumar-datanger get-part.cnp my house destroyed-3.m.s‘Having got angry, Kumar destroyed my house.’

In the following section, I adopt Nunes's system to account for Copy Control in Telugu.

5.2 Copy Control: the derivational history

The derivation of Forward/Backward Control presented in the previous section applies, with minor adjustments, to Copy Control. The Copy Control structure in (62) has the derivational history outlined in (63).

1. (62) [[Kumaar sinimaa cuus-tuu] Kumaar paapkaarn tinnaa-Du]Kumar.nom movie watch-dur.cnp Kumar.nom popcorn ate-3.m.s‘While watching a movie, Kumar ate popocorn.’

In (63a), the CNP clause (i) and the matrix clause (ii) form independently, and the CNP subject copies out of the CNP clause. In (63b), Kumaar merges with the matrix vP. In (63c), the matrix subject moves from Spec,vP to Spec,IP to check the EPP feature. The CNP clause then merges with the matrix clause at CP, as (63d) demonstrates. As the dotted arrow shows, the two matrix copies of Kumaar {[_NP Kumaar]^{Matrix IP}, [_NP Kumaar]^{Matrix vP}} enter into a c-command relationship and form a chain. The CNP copy of ‘Kumaar’, on the other hand, does not enter into a c-command relation with either of the matrix copies. At PF, Chain Reduction applies for the purpose of Linearization. As (63e) illustrates, two copies of Kumaar, namely [_NP Kumaar]^CNP and [_NP Kumaar]^{Matrix IP}, survive deletion, resulting in Copy Control.Footnote ¹¹

1. (1)
   1. (a)

      1. (i)

         

      2. (ii) [_{Matrix vP} paapkaarn tinnaa-Du]

   2. (b) [_{Matrix vP} [_NPKumaar] paapkaarn tinnaa-Du]

   3. (c) [_CP [_{Matrix IP} [_NPKumaar] [_{Matrix vP} [_NPKumaar] paapkaarn tinnaa-Du]]]

   4. (d)

      

   5. (e) [_CP [_CNP [_NPKumaar] sinimaa cuus-tuu] [_CP [_{Matrix IP} [_NPKumaar] [_{Matrix vP}[_NPKumaar] paapkaarn tinnaa-Du]]]]

The main difference between (63) and the derivational history of Forward/Backward Control structures is the merging site of the CNP clause. In Forward/Backward Control, the CNP clause merges clause-internally at vP. In Copy Control, however, the CNP clause merges clause-initially at CP, as (63d) shows. Evidence that this is the case comes from the fact that a bound-variable interpretation is not available in Copy Control constructions. Because the CNP clause undergoes first merge (i.e. merge as a result of base-generation and not as a result of movement) at CP of the matrix clause, the CNP and matrix subjects do not enter into a c-command relation at any point in the derivation. As a result, a bound-variable relation between the two subjects is not possible.

To illustrate, the quantifier prati ‘every’ in Telugu allows two types of readings: (i) a bound-variable interpretation (e.g. (64a)) and (ii) a group interpretation (e.g. (64b)). In the latter case, the pronoun takes as its antecedent ‘the group that satisfies the quantified expression’ (Chao & Sells Reference Chao, Sells, Peter and Charles1983: 50f.). Note that irrespective of the interpretation (bound-variable vs. group), prati baas ‘every boss’ triggers singular agreement on the verb.

1. (64)

   1. (a) Bound-variable reading

      prati baas atani implayis-ki bonus iccaa-Duevery boss.nom his employees-dat bonus gave-3.m.s‘Every boss gave his employees a bonus.’

   2. (b) Group reading

      prati baas waaLLa implayis-ki bonus iccaa-Duevery boss.nom their employees-dat bonus gave-3.m.s‘All bosses gave their employees a bonus.’

When Copy Control with a quantificational phrase is involved, the bound-variable reading is unavailable, as (65a) shows. The only available interpretation is the group reading, whereby the quantificational expression is considered as a specified group of individuals, as illustrated in (65b). This indicates that the matrix subject is not bound, which is expected due to the lack of c-command. Further, we know from section 4.1 that the CNP subject may not be a pronominal. In other words, it is not possible to have a pronominal CNP subject resuming a quantificational subject in the matrix clause (e.g. (65c)). If the CNP clause underwent first merge at vP before moving to CP, restructuring would be possible, and a bound-variable relation of this sort would be available.

1. (65)

   1. (a) *[[_CNP prati baas-ki pooyina-waaram manciga anipinc-i] atanu/aa pichooDu implayis-ki bonus iccaa-Du] every boss-dat last-week good feel-part.cnp he.nom/the idiot.nom employees-dat bonus gave-3.m.s‘Every boss felt good last week, and he/the idiot gave his employees a bonus.’

   2. (b) [[_CNP prati baas-ki pooyina-waaram manciga anipinc-i] waaLLu-andaru/aa pichooLLu implayis-ki bonus iccaa-ru] every boss-dat last-week good feel-part.CNP they all.nom/the idiots.nom employees-dat bonus gave-3.m.p‘All the bosses felt good last week, and they/the idiots gave their employees a bonus.’

   3. (c) *[[_CNP atanu/aa pichooDu pooyina-waaram manciga anipinc-i] prati baas-ki implayis-ki bonus iccaa-Du] he.nom/the idiot.nom last-week good feel-part.CNP every boss-dat employees-dat bonus gave-3.m.s‘Every boss felt good last week, and he/the idiot gave his employees a bonus.’

One question that comes to mind is the following: how can a control interpretation be established without c-command? In other words, how does the matrix subject determine the identity of the CNP subject – or vice versa – if the two do not enter into a c-command relationship? The simple answer, and probably the only one, is movement. The two copies are coreferential because they are copies of the same token derived via movement.

6.1 Multiple copy spell-out

Multiple copy spell-out is attested in several languages. For example, the Kru language Vata allows multiple copies of a verb chain to be pronounced, (e.g. (68); Koopman Reference Koopman1984: 38).

1. (68) Vata

   Li à li-da zué saká.eat 1.p eat-past yesterday rice ‘We did eat rice yesterday.’

The multiple copies realized in (68) are non-distinct copies in a c-command relationship. Based on Nunes's system, the two copies form a chain and, accordingly, they must undergo Chain Reduction. But they do not. How do two non-distinct copies of the same token (i.e. two copies of the same syntactic object related through movement) escape Chain Reduction? According to Nunes (Reference Nunes2004: 40), this is possible only if one of the copies ‘hides’ inside another word, thus becoming invisible to the Linear Correspondence Axiom. More specifically, if a copy in a given chain adjoins to another head, both the copy and the head may be ‘morphologically reanalyzed as a single terminal element’ or a single ‘phonological word’. In the theory of Distributed Morphology (Halle & Marantz Reference Halle, Marantz, Kenneth and Samuel1993), this process is called fusion. The Linear Correspondence Axiom cannot see into fused links and, consequently, the lower copy escapes deletion (see Kandybowicz Reference Kandybowicz2006 for a slightly different approach).

To illustrate, building on Koopman's analysis, Nunes (Reference Nunes2004: 46–48) discusses the example in (68) and holds that the higher copy of the verb li ‘eat’ moves to a focus position preceding IP. The verb and the head of the focus projection are fused into a single terminal and are reanalyzed as a new phonological word. In this way, the Linear Correspondence Axiom does not detect the two occurrences of li ‘eat’ as copies of the same token. This means that the structure can be linearized without the deletion of either copy. Consequently, neither copy undergoes Chain Reduction, given that Chain Reduction is a costly operation that applies only in minimal fashion for the purpose of linearization and convergence, as (69) explicitly states.

1. (69) Delete the minimal number of constituents of a nontrivial chain CH that suffices for CH to be mapped into a linear order in accordance with the LCA. (Nunes Reference Nunes2004: 101 ex. (31)) [my emphasis]

It is tempting to try to analyze Copy Control in Telugu along the same lines. In other words, it is tempting to assume that one of the copies in a Copy Control structure undergoes fusion with a null head of some sort. As a result, the copy becomes part of a new phonological word. Because linearization cannot see into words, the two identical copies are not detected as non-distinct, and they both escape deletion. Such an analysis faces a major problem, however. It predicts that the two copies can escape deletion irrespective of the merging site of the CNP clause. This prediction is not borne out. In sections 3 and 5.1, we saw that when the CNP clause merges at vP of the matrix clause, one of the copies has to be deleted, resulting in either Forward or Backward Control.

Still, an analysis similar to Nunes's is viable. The following section suggests that one of the copies in a Copy Control structure escapes linearization by becoming part of a giant phonological word, a spelled-out domain. The keyword is Multiple Spell-Out (Uriagereka Reference Uriagereka, Samuel and Norbert1999).

6.2 Multiple Spell-Out and Copy Control

6.3 Summary and predictions

I mentioned at the beginning of section 6 that there is a way for the theory to account for Copy Control in Telugu without compromising the essence of Kayne's Linear Correspondence Axiom and without tampering with Nunes's theory of movement. The rest of the section was devoted to showing that Copy Control is derivationally only slightly different from other instances of multiple copy spell-out. According to Nunes, occurrences of multiple copy spell-out involve two non-distinct copies, one of which has become invisible to linearization due to fusion. At PF, linearization cannot see into the fused element. Accordingly, no precedence relationship is detected, and no deletion/Chain Reduction takes place. The same mechanism applies in the case of Copy Control. The steps are summarized as follows:

• • Step 1: Two subject NPs are related through movement and are thus non-distinct copies of the same token.

• • Step 2: Due to Multiple Spell-Out, one copy ‘hides’ within a spelled-out domain and becomes part of a giant phonological word. This outcome is, in essence, identical to fusion.

• • Step 3: As a result, two copies escape deletion without inducing a violation of the Linear Correspondence Axiom simply because no precedence is detected.

The main difference between Nunes's multiple copy spell-out and Copy Control is that the former involves movement minus Chain Reduction, while the latter involves movement minus Form Chain and Chain Reduction. Both types of movement are allowed only if no violation of the Linear Correspondence Axiom is induced.

The analysis makes an important prediction. In order for Copy Control to obtain in a language, two conditions must be satisfied:

• • Condition 1: The two arguments entering into a control relation must check case, which may be inherent or structural case (or even default case).

• • Condition 2: At least one of the two arguments entering into a control relation must become part of a bigger word by PF.

The prediction is not entirely new. Condition 1 is based on the standard assumption that an NP has to check case in order to be visible, or phonologically realized (Chomsky Reference Chomsky1981: 49, Vergnaud Reference Vergnaud1982). Condition 2 is a central assumption in Nunes's system. It is employed in Boeckx et al. (Reference Boeckx, Hornstein, Nunes, Anastasia, Chunyuan, Chizuru and Eri2007) in order to account for Copy Control in San Lucas Quiaviní Zapotec. To elaborate, Boeckx et al. analyze the Copy Control construction in (75) (from Lee Reference Lee2003: 102 ex. (83)) as movement. The lower subject copies out of the subordinate clause and merges in the matrix clause. By PF, the lower copy adjoins to a covert head, a reflexive self-affix.Footnote ¹⁴ The copy and the head form a new word. Since linearization cannot see into words, the copy escapes deletion.

1. (75) San Lucas Quiaviní Zapotec

   R-cààa'z Gye'eihlly g-auh (Gye'eihlly) bxaadyhabitual-want Gye'eihlly irrealis-eat Gye'eihlly grasshopper ‘Mike wants to eat grasshopper.’

In this paper, I suggest that Condition 2 may also be satisfied if one of the copies in a control structure hides in a spelled-out domain, a phase that is transferred to the phonological component and transformed into a giant compound à la Uriagereka (Reference Uriagereka, Samuel and Norbert1999). In the case of adjunct control, satisfying Condition 2 seems to depend strictly on the first merging site of the adjunct. If the adjunct merges at the matrix CP – that is, after the matrix clause is spelled out and transformed into a giant compound – then the matrix subject gets the chance to hide in a bigger word.Footnote ¹⁵ If Condition 1 is also satisfied – that is, if the two control-related arguments check case – the result is Copy Control.

Conditions 1 and 2 are able to account for Copy Control in Telugu. They may also be able to generalize over Copy Control in other South Asian languages, such as Dakkhini (e.g. (76)) and Karnataka Konkani (e.g. (77); from Arora & Subbarao Reference Arora, Subbarao, Peri and Karumuri2004: 40 exx. (80)–(81)). (Note that in these languages, the CNP subject may be realized as a pronominal, which makes the restriction that the CNP subject must be an r-expression specific to Telugu.)

1. (76) Dakkhini

   [[us-ku bukhaar aa-ke] uno mar.gayaa]he-dat fever.nom come-cnp he.nom died ‘Having had a fever, he died.’

1. (77) Karnataka Konkani

   [[tak-ka taap yewa-nu] tO gellO]he-dat fever.nom come-cnp he died ‘Having had a fever, he died.’

On the other hand, if the adjunct undergoes first merge at vP of the matrix clause, neither the subordinate subject nor the matrix subject has the opportunity to become part of a spelled-out domain by PF. This is why both copies are detected as non-distinct at PF, and thus only one survives deletion. Note that this is true even if both control-related arguments check case. It is also true even if the adjunct subsequently undergoes remnant movement to a sentence-initial position, adjoining to CP of the matrix clause (see section 5.1 above).

If this explanation is on the right track, it may be able explain why not so many languages license Copy Control, including languages that are similar to Telugu. For example, Hindi-Urdu also licenses adjunct control into CNP clauses (e.g. (78a, b), adapted from Kachru Reference Kachru1981). Nevertheless, the language categorically disallows Copy Control (e.g. (78c)). Notice that, based on the movement approach adopted in this paper, (78b) would be an instance of remnant movement.

1. (78)

   1. (a) [Raaj-ne_i [Δ_i nahaa-kar] khaanaa khaayaa]Raj-erg bathe-cnp meal ate ‘Having bathed, Raj ate a meal.’

   2. (b) [[Δ_i nahaa-kar] Raaj-ne_i khaanaa khaayaa] bathe-cnp Raj-erg meal ate ‘Having bathed, Raj ate a meal.’

   3. (c) *[[Raaj_i nahaa-kar] usne khaanaa khaayaa]Raj.nom bathe-cnp he.erg meal ate ‘Raj bathed, and he ate a meal.’