Troy Messick (U of Connecticut)

Pronouns and agreement in Telugu embedded contexts INTRODUCTION: Schlenker (1999, 2003) proposes that pronouns under the de se reading in embedded speech reports are semantically bound by an embedded author coordinate even though the morphological form of the pronoun can be 3rd person. So he argues that in (1) under the de se interpretation, even though the embedded subject appears to be morphologically 3rd person, it is semantically 1st person. (1) Johni said hei passed the exam. In this paper, I present novel data from Telugu which reveals a previously unobserved pattern that provides evidence for the idea that 3rd person pronouns in de se contexts are semantically 1st person. PRELIMINARY DATA: It has been noted by several authors that in embedded contexts, 3rd person pronouns can control first person agreement on the verb (Novelli 1985, von Roncador 1992, Curnow 2002). For example, in Karimojong, the 3rd person pronoun iŋèz can control 1st person agreement (2). (2) ɑ̀bʊ́ pɑpɑ̀ tolim εbè ɑ̀lózì iŋèz morotó AUX father say that 1SG.go.NONPAST 3SG Moroto ‘The father said that he is going to Moroto’ (Novelli 1985:531) The same facts are noted for the related language of Lotuko (von Roncador 1992). The same pattern exists in Telugu where the 3rd person t̪ anu can control either 3rd person (3a) or 1st person (3b) agreement. (3) raani [t̪ anu exam pass {a. ajj- ind-ani b. ajj-aa-nu-ani}] nammut̪ -und̪ i Rani [3SG exam pass {a. happen- F.SG-C b.happen-PAST-1SG-C}] believe-F.SG “Rani believes that she passed the exam” rd THE STATUS OF t̪ anu: I treat t̪ anu as a 3 person pronoun although it is cognate of the logophoric ta(a)n found in Dravidian languages (see Anand 2006 for Malayalam, Sundaresan 2013 for Tamil). It is noted in the descriptive grammar that although t̪ anu in Telugu was once logophoric, as well, speakers now use it as a third person pronoun (Krishnamurti & Gwynn 1985:73). OBLIGATORILY DE SE: There is a semantic effect of the first person agreement with t̪ anu. If t̪ anu controls first person agreement on the embedded verb then only the de se reading is available. So in the scenario in (4), which forces the de re reading, only (3a) could be uttered felicitously. (4) Situation: Rani took an exam, and later saw the top 10 scores with the respective ID numbers. She forgot her own ID number, so didn’t know who was who. Pointing to the top score, she thought, “This student definitely passed!” But it turned out that she was that student. a. (3a) OK b. (3b) # AN INSTANCE OF INDEXICAL SHIFT: The phenomena noted in Karimojong and Telugu is reminiscent of indexical shift (Schlenker 2003, Anand & Nevins 2004 a.o.). However, in languages that allow indexical shift, the indexicals themselves are the elements that shift (though see Sundaresan 2013 for cases where there is apparent 1st person agreement with a logophoric subject). For example, the first person pronoun εz in Zazaki can refer to the author of the reported speech act (Anand & Nevins 2004). In (2) and (3), however, it is not the indexical itself that shifts. It is only the agreement on the verb that appears to be 1st person. In Telugu, if the first person indexical neenu is used, it must refer to the current speaker (5). (5) raaǰu [neenu eemi tinn-aa-nu ani] čepp-ææ-Du Raaǰu [I what eat-PAST-1SG C] say-PAST-M.SG “#What did Raju say that he ate?” “What did Raju say that I ate?” ANALYSIS: To account for the fact that in de se readings in Telugu, first person agreement appears on the verb despite the subject being a 3rd person pronoun, I assume, following Schlenker (1999, 2003) that pronouns in de se readings are bound by an embedded author coordinate but morphologically appear to be 3rd person. I implement this using Kratzer (2009)’s minimal pronoun theory, and the interpretable/uninterpretable distinction proposed in Chomsky (2000, 2001). The minimal pronoun that begins the derivation has both an interpretable and uninterpretable person feature. This is similar to the idea that nominals have both CONCORD and INDEX features in Wechsler & Zlatić (2000, 2003). Both of these features begin the derivation unvalued. In Telugu and Karimojong, as well as English, the pronoun is bound by the author coordinate of the embedded context as in Schlenker (1999, 2003) and its interpretable person feature is valued as 1st. The uninterpretable person feature of the pronoun is valued by feature transmission with the 3rd person matrix subject.

Feature Transmission (6)

SUBJECT[iPerson: 3rd] MATRIX VERB λ[. … x[iPerson: 1st uPerson: 3rd] …]

Binding So the person features for a de se pronoun in Telugu, Karimojong and English are [iPerson: 1st uPerson: 3rd]. The only difference between English and Telugu and Karimojong verbal agreement with these pronouns is which feature of the pronoun the verbal inflection targets. English verbal inflection agrees with the uninterpretable 3rd person feature, while Telugu verbal inflection targets the interpretable 1st person feature. The Telugu and Karimojong agreement is simply an instance of semantic agreement in the sense of Corbett (1979). It is well known that semantic features can control agreement. E.g., semantically plural collective noun phrases can trigger plural agreement on the verb in some dialects of English even though they appear to be morphologically singular as in (7). See Wechsler & Zlatić (2000, 2003) and Corbett (2006) for more examples. (7) The committee {a. has b. %have} decided. (Corbett 1979:203) The analysis of Telugu agreement is then part of a larger paradigm found for φ-agreement more generally. FURTHER IMPLICATIONS AND CONCLUSION: This analysis also accounts for cases where 1st person pronouns control 2nd or 3rd person agreement on the verb in embedded contexts in a number of languages. For instance, in the Uyghur example in (8), the 1st person pronoun meni controls 2nd person agreement on the verb (Shklovsky & Sudo 2014:399). Similar data has been noted for Donno Sɔ (Curly 1994:123) and Golin (Loughane 2005). (8) Ahmet [meni nan ye-isen] di-di. Ahmet [1SG.ACC bread eat-IMPF.2] say-PAST.3 `Ahmet said to me, “You eat bread.”’ I argue that this is also a case of semantic agreement. Since the current speaker is the addressee of the reported speech act, the pronoun would be bound by the addressee coordinate of the embedded context, but since it is the author of the current utterance, it will be bound be bound by the author coordinate of the matrix contexts, resulting in the person features [iPerson: 2nd uPerson: 1st]. The verb can then target the interpretable 2nd person features to gets its value. The final section of this paper discusses the distribution of first person agreement with t̪ anu. In Telugu, this agreement appears to occur embedded under attitude predicates as in (3), but it can also occur embedded under factive verbs and other subordinate clauses. (9) raani [t̪ anu exam pass ajj-aa-nu-ani] t̪ elusu-kon-und̪ i Rani [3SG exam pass happen-PAST-1SG-C] know-REFL-F.SG “Rani found out she passed the exam” (10) raani [t̪ anu exam pass ajj-aa-nu-ani] santošanga und̪ i Rani [3SG exam pass happen-PAST-1SG-C] happy BE “Rani is happy because she passed the exam” The distribution is not completely free, however. First person agreement with t̪ anu can only occur under the complementizer ani .Ani historically comes from the verb an, which means ‘to say’ (Krishnamurti & Gwynn 1985:363). This same restriction is noted for logophors (Sells 1987) and also for indexical shift in Uyghur (Sudo 2012:202 fn. 143). To conclude, this paper has introduced and analyzed a new agreement paradigm in Telugu. The paradigm provides evidence that morphologically third person pronouns are semantically 1st person in de se contexts. REFERENCES: Anand & Nevins 2004. Shifty Operators in changing contexts. SALT. Corbett 2006. Agreement. Curly 1994. Aspects of logophoric marking. Linguistics. Curnow 2002. Three types of verbal logophoricity in African languages. Studies in African linguistics. Kratzer. 2009. Making a pronoun. LI. Krishnamurti & Gwynn. 1985. A grammar of modern Telugu. Loughane 2005. Reported speech constructions in Golin. Novelli 1985. A grammar of the Karimojong language.. .Shklovsky & Sudo 2014. The syntax of monsters. LI. Sudo. 2012. On the semantics of phi features on pronouns. Sundaresan 2013. Context and (co)reference. Wechsler & Zlatić 2003. The many faces of agreement.