4.1 The basic effect: third person pronouns

The clear answer for simple cases is that honorific features are interpreted relative to the speaker, as we have seen so far. We have captured this by writing “than I am” into the informal definitions of our features in (3), and by saying that Hon heads bear a dedicated index bound by Sp in our official implementation in (16) and (17).

However, there are sometimes other possibilities which show that it is not as simple as this. Consider an example with a third person pronoun in an embedded clause, such as (20). Imagine a context in which Deepak is talking about his friend, and Santee is Deepak’s younger brother. The pronoun in the embedded clause referring to Deepak’s friend can be the NH form okraa, reflecting the speaker Deepak’s peer relationship to his friend, as in the other examples that we have seen. But it can also be the H form unkaa reflecting the matrix subject Santee’s relationship to Deepak’s friend.

This may give the impression that the honorific features on a pronoun can be interpreted pragmatically, with the standard of social status being any salient person in the discourse context.¹⁶

That conclusion would be hasty, however. It so happens that Magahi is a language that optionally allows indexical shift in complement clauses (cf. Schlenker 1999; Anand & Nevins 2004; Anand 2006; Deal 2020, among others). For example, the first person pronouns in the complement clauses in (21) can refer to the speaker of the sentence as a whole (which we call Sp*), as in English, but they can also refer to the subject of the matrix clause Santee. This indexical shift is possible both with dyadic cognition verbs like ‘think’ ((21a)) and with triadic verbs of communication that also (optionally) take a goal argument like ‘tell’ ((21b)). Unlike Mishar Tatar (Podobryaev 2014), indexical shift in Magahi is optional regardless of whether the first person pronoun is overt or undergoes pro-drop, although the preferred reading varies: the null pronoun prefers the shifted reading unless it is in a context that clearly supports the unshifted meaning, whereas the overt pronoun easily gets either reading in relative isolation. In this section, we concentrate on examples with overt first person pronouns, turning to some differences with null pronouns in section 4.3.

Indexical shift in Magahi has been studied in detail in Alok & Baker (2018) and Alok (2020). These works perform the standard tests to show that the “I=Santee” readings in sentences like (21) are not (necessarily) the result of the material following the complementizer ki being a direct quotation. This shifted reading is compatible with adjunct wh-phrases being extracted from the embedded clause, with negative polarity items in the embedded clause being licensed by negation in the matrix clause, and with the presence of material interpreted de re in the embedded clause; see Alok & Baker (2018: 27–29) and Alok (2020: 145–148). Indeed, one of the relevant tests (involving embedded addressee agreement) implies that having a direct quotation following the complementizer ki in Magahi, although not entirely impossible, does not come readily to the mind of a Magahi speaker, but probably requires a special intonational contour and a particular context. Therefore, there seems to be relatively little danger of mistaking a direct quotation for a case of indexical shift in this language (Alok & Baker 2018: 29). Given this result and strict space restrictions, we only apply the quotation tests explicitly to two particularly important examples here (see (24) and (25)).

What is significant about indexical shift for this work is that it interacts with the ambiguity in the interpretation of honorification features in (20) in an interesting way. To see this, consider the sentences in (22), imagined in the same scenario as before: Deepak is talking about his friend, who is older than Deepak’s younger brother Santee. Here the embedded subject ham ‘I’ can refer to the matrix subject Santee, or to the Sp* Deepak, as in (21). However, there is not a free choice as to the form of the third person pronoun in this case. If ‘I’ refers to Deepak, then the pronoun referring to Deepak’s friend must be okraa, the NH form reflecting Deepak’s relationship to his friend. In contrast, if ‘I’ refers to Santee, then the pronoun referring to Deepak’s friend must be unkaa, the H form reflecting Santee’s relationship to Deepak’s friend. This covariance is seen both when the matrix verb is ‘think’, as in (22a), and when it is ‘tell’, as in (22b).¹⁷

(23) gives other examples of this kind, showing the generality of the effect. These are imagined in the context of one professor talking to another professor about Deepak and a third professor. If ‘my’ in the embedded clause is given a shifted reading where it refers to Deepak, then the subject pronoun referring to the third professor must trigger HH agreement -thin on the embedded verb, as in (23a). This HH marking reflects Deepak’s relationship to the third professor. In contrast, if ‘my’ is understood as referring to the professor who utters the sentence, then the pronoun u referring to the third professor must trigger NH marking on the embedded verb, as in (23b). This NH marking reflects the relationship of the first professor to the third professor.

We see, then, that (20) does not reflect pragmatic vagueness about how honorific features are interpreted; rather there is a grammatical ambiguity here, and it interacts with the grammar of indexical shift.

Here we do need to make sure that we are dealing with true indexical shift, rather than the difference between indirect speech and direct quotation. If the complement clauses in (22) and (23) are quotations of what the subject of the matrix verb said or thought, then it makes sense that that person refers to themself with ‘I’ and uses pronouns that reflect their own relationship to the referent of the pronoun. As is well known, one cannot extract a question word out of a quoted sentence, whereas one can extract out of a clause with true indexical shift. Thus in (24) when ‘when’ is interpreted as questioning the time of the dying event, the pronoun ‘I’ in the embedded clause can still have the shifted reading in which it refers to the matrix subject Ram, showing this to be true indexical shift.

Then (25) shows that including a third person pronoun in the embedded clause does not change this. This sentence is imagined in a context in which one professor is talking to another professor about a student Chotu and a third professor. Here it is possible for ‘when’ to question the time of the proposal’s acceptance, with ‘my’ referring to Chotu, and ‘he’ triggering HH agreement, expressing Chotu’s relationship to the third professor (not Sp*’s).

The negative polarity item test gives similar results in that shifted readings of ‘my’ and ‘he’ in a sentence like (25) are compatible with there being a negative polarity item in the embedded clause licensed by negation in the matrix clause.

In fact, the principal groundwork for deriving these facts is already in place: it is our assumption that Hon heads bear the designated index {1} which they share with Sp, Speas & Tenny’s (2003) syntactic representation of the speaker of a sentence. This assumption was already partially motivated by the basic meanings of simple sentences with honorific-marked elements and by our derivation of the fact that first person pronouns are always nonhonorific in Magahi. However, now it truly comes into its own. Given this assumption, the fact that the point of reference for evaluating the honorific status of the pronoun shifts if and only if first person pronouns in the same clause shift qualifies as a Shift Together effect, familiar from Anand & Nevins (2004) and subsequent work. Magahi obeys Shift Together quite strictly, in that two first person pronouns in the same clause must have the same reference. For example, in (26a) the two pronouns must both refer to Sp* or they must both refer to the matrix subject Santee; it is not possible for one of them to refer to Sp* and the other one to refer to Santee. This is reinforced by the fact that an example like (26b) is impossible: readings in which the two pronouns have different referents are ruled out by Shift Together, whereas readings where they have the same referent are ruled out by Condition B of the Binding theory. (Magahi is different from some other languages including Amharic in this respect; see Schlenker 1999; 2003, Anand 2006, Deal 2020.)¹⁸

We assume that an ingredient in this kind of Shift Together is the assumption that first person pronouns also share with Sp the designated index {1}. As such, two of them in the same clause will both be bound by the closest instance of Sp, and they thus have the same reference as each other, both referring to whatever Sp does in that structure. Our hypothesis that Hon heads also bear the index {1} then implies that their interpretation covaries with the factors that determine the reference of first person pronouns.

We flesh out the details of this analysis within the control-based theory of indexical shift of Alok & Baker (2018) and Alok (2020), developing a view sketched in Baker (2008)—while acknowledging that the more standard shifty operator theory of Anand & Nevins (2004), Anand (2006) and Deal (2020) could likely work as well (see below for some discussion). The assumptions of the Baker/Alok view are outlined in (27), some of which have already been anticipated.

Assumptions (27a) and (27b) are taken from Speas & Tenny’s (2003) neoperformative theory, which is intended to account for a wide range of phenomena at the interface of syntax and pragmatics. (Recall that in section 2 we used it to account for addressee agreement in Magahi.) Assumption (27c) descends directly from Baker’s (2008) effort to derive why first and second person features often behave differently from third person features, being subject to stricter locality conditions. (27d) is the option that results in indexical shift in languages that permit it: Amharic, Zazaki, Nez Perce, and Magahi allow this type of control; English does not. Finally (27e) fills in what happens when control does not take place: then Sp and Ad default to the values of the next highest Sp and Ad. In cases with only one level of embedding, uncontrolled Sp and Ad are bound by the matrix Sp and Ad, and hence refer to Sp* and Ad* by (27b). This gives the unshifted reading.

Given these assumptions, an ordinary Shift Together example like (26a) has the representation in (28a) or (28b). One possibility is that the subject of ‘tell’ controls Sp, as shown in (28a). In this case, the same DP (Sp) is both a binder (of ‘my’ and ‘me’) and a bindee (of ‘Santee’, as a result of control). To represent this clearly, we adopt the semantic convention of distinguishing between the variable/bindee index that a DP might bear and the abstractor/binder index that it might be associated with. We indicate the variable/bindee index as a subscript at the end of the DP itself, whereas the abstractor/binder index associated with the DP is subscripted at the beginning of the sister of the binder—i.e., on the phrase that is abstracted over. The two indices can be distinct. In particular, in (28a) ‘Santee’ binds Sp by the index (say) 3 as a result of control, and Sp binds both ‘my’ and ‘me’ by the designated index 1. The result is that both ‘my’ and ‘me’ end up referring to Santee—the Shift Together reading. In contrast, if the subject of ‘tell’ does not control Sp, as in (28b), then Sp is bound by Sp* via the designated index 1. Meanwhile, both ‘my’ and ‘me’ are bound by Sp by the designated index 1; therefore, both ‘me’ and ‘my’ refer to the speaker of the whole sentence—the unshifted reading.

Now compare this with (29), the corresponding representations of a sentence like (22a), which contains both a first person pronoun and a third person pronoun with honorific features.²⁰

Here if ‘Santee’ controls Sp in the complement clause, as in (29a), then ‘I’ refers to Santee, as in (28a). At the same time, the honorification status of ‘him’ is fixed relative to Santee, because Hon bears the same designated index as ‘I’ does, by hypothesis. In contrast, if ‘Santee’ does not control Sp, as in (29b), then Sp* binds Sp. Then ‘I’ refers to Sp* and the honorification status of ‘him’ is fixed relative to Sp*—again because Hon bears the same designated index 1 as ‘I’ does. In this way, we capture the covariance we observe in (22) and (23) by the same mechanisms used to capture Shift Together.

Before moving on, a brief comparison of our approach to indexical shift in Magahi with the more familiar “shifty operators” theory of Anand (2006) and Deal (2020) is in order. The context-shifting operator account says that an embedded clause in some languages can be interpreted relative to a context different from that of the sentence as a whole. Specifically, the embedded clause can be interpreted in a context associated with the event denoted by the matrix verb. First person pronouns inside the embedded clause then refer to the author of the context associated with that event. Lexical meanings for the honorification features could presumably also be stated such that they refer to the author of the local context; for example, Hon:HH could be assigned the meaning in (30), where the Hon refers to the author of the current context, rather than to a variable bound by Sp.²¹

Although we do not doubt that the gist of our account can be captured in this way, we prefer (27) because of concerns about the well-definedness of “author” in this shifty-operator analysis. How precisely this author relates to the syntactic arguments of the matrix verb is not specified explicitly in this account; rather, it is taken to follow in particular cases from what “author” means together with the lexical semantics of the matrix verb. With this in mind, consider the Magahi examples in (31).

In purely semantic terms, it seems reasonable to say that Santee counts as the author in the context associated with the matrix event of saying in all these examples; certainly he is the source of the content expressible as “I will come to visit soon” in all three. Therefore, the shifty operator account might well predict that I=Santee will be possible in all three examples. Indeed, some constructions that are “logophoric” in a broad sense do show this grammatical freedom; for example NOC PRO in English is possible in (32b,c), where the antecedent of PRO is not an argument of the matrix clause, as well as in (32a), where the antecedent of PRO is an argument of the matrix clause.

Indexical shift in Magahi, however, does not have this latitude: it is possible in (31a) but not in (31b) or (31c). Thus, there seems to be an additional constraint on indexical shift: not only does the understood antecedent of ‘I’ in the embedded clause need to count as an author semantically, but it needs to be a grammatical argument of the matrix verb. Our account differs from the shifty operator account in using the well-studied operation of obligatory control to capture this additional constraint (see (27d); compare with Landau’s 2013: 29 “OC signature”). In other words, at a high theoretical level we take (31) in Magahi to be akin to the OC paradigm in (33) rather than to the purely semantic/pragmatic NOC/logophoric paradigm in (32).

This is one of our primary motivations for adopting a control-based theory of indexical shift as in (27), even if the main insights about the nature of honorification features could also be expressed in a shifty operator approach.

4.2 Extension to second person pronouns

Armed with the explicit theory of indexical shift in (27), we can go on to generalize our account to honorific marking on second person pronouns in embedded clauses. Previous work has shown that second person pronouns also shift their reference in Magahi when the matrix verb is a dyadic verb such as ‘tell’ or ‘ask’ (Alok & Baker 2018; Alok 2020). For example, ‘you’ in (34) can refer to John, the goal of the telling event, or it can refer to Ad*, the addressee of the sentence as a whole. This is parallel to the optionality of first person indexical shift in (21).

The assumptions in (27) provide for this by positing Ad as well as Sp ((27a)), Ad bearing its own designated index {2} ((27c)), and by optionally allowing the goal of the CP-selecting verb to control the Ad of its CP complement, parallel to the agent of the verb controlling the Sp of its CP complement ((27d)). Indeed, these two control options are linked, as seen in an example like (35) which has both a first person pronoun and a second person pronoun in the complement clause. This sentence has two readings rather than four: it is possible with ‘I’=Santee and ‘you’=Bantee, or with ‘I’=Sp* and ‘you’=Ad*. However, it cannot have the meaning ‘I’=Santee and ‘you’=Ad* or ‘I’=Sp* and ‘you’=Bantee. (This restriction, known also from Zazaki and Nez Perce among other languages, is also called Shift Together by Anand & Nevins (2004), but it is conceptually distinct and admits some variation across languages and constructions; see Deal (2020: 125–132) for clarifying discussion. Here we call the fact that two pronouns with the same person feature have to shift together as in (26) Shift Together 1 and the fact that two pronouns with different person features have to shift together as in (35) Shift Together 2.)

This Shift Together 2 effect is asserted in (27d) in the statement that the arguments of the matrix verb either control the null DPs at the edge of the CP together as a package, or not at all. Alok (2020) suggests a deeper derivation of this in terms of saying that this type of control is fundamentally a relationship between heads: the matrix verb controls the head of its CP complement, and an implication of this control relation is that the arguments of the verb control the corresponding arguments of the C. We do not develop that further idea here, however.

So far this discussion has not taken into account honorific marking on second person pronouns. The ambiguity in (34) and (35) is partly due to the fact that the people imagined as being involved—the referent of the subject, the referent of the object, Sp*, and Ad*—are all peers, so second person NH marking is expected on both interpretations of the sentences. However, other scenarios show that honorific marking is affected by the control of Sp and Ad as much as person marking is. This can be seen in examples like (36). The context for these sentences is that Deepak is speaking to a teacher—note the HH addressee marking on the matrix verb—and Santee and Bantee are friends.

Here if ‘I’ shifts to refer to the matrix subject Santee, then ‘you’ also shifts to refer to Bantee and the form of the pronoun ‘you’ is NH, the form consistent with Santee’s relationship to Bantee ((36b)). In contrast, if ‘I’ is unshifted and refers to Sp* (Deepak), then ‘you’ is unshifted, referring to Ad* (the professor), and its form is HH expressing the relationship of Deepak to the professor ((36a)). In broad terms, this is like what we saw with third person pronouns: in both cases, honorification on a pronoun in the embedded clause is determined relative to the matrix subject if ‘I’ shifts and relative to Sp* if ‘I’ does not shift. The explanation is also the same: Hon with ‘you’ as its complement is like Hon with ‘he/she’ as its complement in bearing the designated index {1} which is bound by the closest Sp. As a result, the status of the DP complement of Hon is evaluated relative to the matrix subject if the matrix subject controls the embedded Sp, and it is evaluated relative to Sp* if the matrix subject does not control the embedded Sp. Indeed, even if there is no first person pronoun in the embedded clause, the same possibilities arise for Hon, since Sp is still optionally controlled. A significant difference is that the reference of ‘you’ in the embedded clause changes depending on whether indexical shift happens or not, whereas the reference of ‘he’ need not change; nevertheless, the same theory of the honorific features applies to both.

A distinctive feature of our control-based approach to indexical shift is that it fits well with a simple analysis of addressee agreement in Magahi, as seen again in (37).

As mentioned in section 2, our analysis of this taken from earlier work (Alok & Baker 2018; Alok 2020; Alok 2021) is that addressee agreement is the result of a C like head (Fin) agreeing with Ad—the same Ad that is posited in (27a). In root clauses, this Ad refers to the addressee of the sentence ((27b)), resulting in addressee agreement. This is essentially Oyharçabal’s (1993) hypothesis about allocutivity in Basque put into the framework of Speas & Tenny (2003), as in Miyagawa (2012; 2017) among others. Now our hypothesis about indexical shift assumes that Sp and Ad can be found in embedded clauses in Magahi too. Therefore, we can find addressee agreement with an Ad on Fin in embedded clauses as well.²² Moreover, under a verb like ‘tell’, there are two possibilities. The first is that Ad can be controlled by the goal of the matrix verb; then agreement with Ad on Fin expresses the social relationship that holds between the referent of the matrix subject and the referent of the matrix goal, as shown in (38a). The second option is that Ad can avoid being controlled by the goal of the matrix verb; then it is bound by the next highest Ad (here Ad*), and the features copied from it onto Fin in the embedded clause are those that express the social relationship between Sp* and Ad*, as in (38b). (See Alok 2020 for more examples and discussion.)

A formal implication of this is that Ad itself is embedded in an HonP with an index bound by the closest Sp—the Sp that inhabits the same CP periphery as Ad. (Note that this implies that Sp c-commands Ad, as Speas & Tenny 2003 assume.) This is represented in the bolded phrases in (39); (39a) shows the shifted Ad of (38a) and (39b) shows the unshifted Ad of (38b).

This elaborated structure for Ad may seem a bit surprising at first glance, but it is simply the unpacking of our claim that Ad bears the same person and honorification features as overt ‘you’ does, as shown by the fact that addressee agreement on Fin manifests the same three-way honorification distinctions as second person subject agreement does (compare (4) and (7) above). Since the Hon features are on the Hon head and Ad has Hon features, Ad must be embedded inside an HonP. However, in most cases we continue to omit the Hon head associated with Ad to make the structures more readable.

So far the examples in this section all involve a triadic matrix verb like ‘tell’. The patterns are somewhat different if the matrix verb is a dyadic verb of cognition like ‘think’. Unlike (36), in this situation there is only one choice for the honorification features on ‘you’ in the embedded clause: they must be determined by Sp*’s relationship to the referent of ‘you’, not by the matrix subject’s relationship to the referent of ‘you’. This is shown by (40) spoken by Deepak to his friend about his younger brother Santee, who admires Deepak’s friend. ‘You’ here must be NH tu reflecting Sp*’s relationship to the friend; it cannot be HH apne reflecting Santee’s relationship to the friend.²³

This is superficially different from a second or third person pronoun in the complement of ‘tell’ and from a third person pronoun in the complement of ‘think’, any of which can have their honorificity level calculated relative to the embedded subject.

This difference can be derived from the close relationship between first person indexical shift and second person indexical shift—the Shift Together 2 phenomenon. In our theory, this is expressed in (27d), the claim that the arguments of the verb control as a package.²⁴ What happens, then, when the matrix verb does not supply a controller for Ad? The answer in Magahi as in Uyghur and Nez Perce is that when the subject of a goalless verb controls Sp, Ad is defective, receiving the value 0 (Deal 2020:84–86). Ad₀ does not give a pronoun that it binds a referent. The result is that a second person pronoun cannot appear along with shifted first person pronouns in the embedded clause of a verb like ‘think’, as shown in (41b). In contrast, the presence of a third person pronoun puts no restrictions on the shifting of a first person pronoun in the same clause (see (41a)), because third person pronouns do not have to be bound by anything the way that second person pronouns are bound by Ad.

(40) can now be seen as a result of the same forces at work, with the two imaginable representations in (42), where the Hon head associated with ‘you’ is added in.

‘You’ is present in the complement, so its binder Ad in the embedded CP must not get the value 0 ((42a) is ruled out). Therefore, the arguments of ‘think’ must not be controlling the Sp/Ad arguments of the embedded C. Therefore, Sp in the CP complement of (42) must be bound by Sp* by (27e), as shown in (42b). A consequence of this is that the Hon head that takes ‘you’ as its complement has its designated index bound by the local Sp, which in turn is bound by Sp*, just as ‘I’ is in (41b). Therefore, the honorific features of ‘you’ must show the social relationship between Sp* and the referent of ‘you’. The matrix subject cannot come into the honorification calculations in this case.

Overall, we see that honorific features have fundamentally the same formal possibilities and interpretations on second person pronouns in embedded clauses as they do on third person pronouns in embedded clauses—a desirable result. However, second person pronouns are subject to a condition which does not apply to third person pronouns: they must be bound by the Ad argument of C. The control of Sp is entangled with the control of Ad (Shift Together 2), and this breaks the symmetry between second person and third person pronouns in a few well-defined respects.

4.3 Advanced indexical shift: pro versus overt pronouns

So far in this section we have concentrated on examples in which the pronouns in the embedded clause are all overt. However, we mentioned that Magahi has subject agreement, and therefore ordinary pro-drop is possible for subject pronouns.²⁵ It turns out that when the embedded subject is a first person null pronoun, some different patterns can emerge—a plot twist. Although our account of this phenomenon does not go beyond a descriptive level, our discussion of this enriched data set makes clear one important result: here too the patterns of honorific interpretation are predictable from the patterns of indexical shift.

It is already known that some languages draw a distinction between overt pronouns and null pronouns licensed by rich agreement when it comes to indexical shift. For example, in Mishar Tatar, first person pro can undergo indexical shift, whereas overt first person pronouns cannot (Podobryaev 2014). This fundamental difference is not found in Magahi: indexical shift is generally optional with both pro and overt pronouns. But a difference with respect to Shift Together 2 arises in sentences like (41b). When the embedded subject is ham ‘I’, the presence of ‘you’ in the complement clause makes indexical shift of ‘I’ impossible, but when first person pro is used instead of ham, as in (43), it can undergo indexical shift. Here ‘you’ is licit and interpreted as Ad*. This can result in a so-called mixed context reading in which pro_1st gets its referent from the thinking situation and ‘you’ gets its referent from the speech situation.

We describe this pattern in theoretical terms as in (44).²⁶

We acknowledge that we do not fully understand this addendum to the theory of indexical shift and refrain from speculating about it here (see Alok 2020 for a proposal).²⁷ Whatever its ultimate nature turns out to be, the idea is that (44) makes possible the representation in (45a) for (43), alongside (45b) carried forward from the previous section.

Our main interest here is not to derive (44) but to understand its implications for the interpretation of honorific features on second and third person pronouns. The upshot is that the implications are positive. Consider first the possibility of third person pronouns in the complement of ‘think’. We saw in section 4.1 that when a first person pronoun is overt, honorific features on the third person pronoun are interpreted relative to the matrix subject if and only if the overt first person pronoun also refers to the matrix subject. This is not necessarily the case with a pro_1st subject. Consider (46), again in the context where Deepak is talking about his younger brother Santee and referring to Deepak’s friend. In this case, the third person pronoun can either be honorific unkaa, showing the relationship of the matrix subject Santee to its referent (Deepak’s friend), or it can be nonhonorific okraa, showing the relationship of Sp* Deepak to its referent (his friend).

This optionality is what we expect given (44). Since there is no second person pronoun in (46), there are two ways that the matrix subject Santee can control: it can do so directly, using (44), or it can do so indirectly by using (27d), the same way that overt ham is shifted. This gives the two representations in (47).²⁸

The index 1 associated with Hon is bound by the closest Sp in both cases, but in (47b) this Sp is controlled by Santee whereas in (47a) it is not involved in control and is therefore bound by Sp*. The result is that Hon can express the social relationship of its DP complement relative to either Santee or Sp* in this case.

An important claim here is that honorific marking in a sentence like (46) is not syntactically unconstrained; rather the sentence is structurally ambiguous, with its two versions corresponding to two distinct syntactic representations. An empirical consequence of this arises if one puts two coreferential third person pronouns in the embedded clause with a first person pro subject. Imagine (48) in the context where Deepak is talking about his younger brother Santee and the two third person pronouns refer to Deepak’s peer friend. Suppose that pro refers to Santee, the shifted reading. Here the two pronouns can both have the honorific form unkaa, showing Santee’s relationship to Deepak’s friend, or they can both have the nonhonorific form okraa, showing Deepak’s relationship to his friend. One cannot, however, mix and match, with one okraa and one unkaa. This is a novel kind of Shift Together effect.²⁹

This is predicted by our analysis. The two {1} indices associated with the HonPs containing the third person pronouns must both be bound by Sp, so they must have the same value. In addition, the two third person pronouns have the same referent, by hypothesis. Therefore, the same honorific features must be on both Hon heads, continuing to assume that two people can only stand in one social relationship at a time. There are two possibilities in (48) because Sp can be controlled by Santee or bound by Sp* (Santee controlling the subject pro_1st directly via (44)). However, these are the only possibilities allowed within this formal system.

Next, we consider the consequences of (44) for sentences that contain a second person pronoun. We already discussed (43) with matrix verb ‘think’, pro_1st with a shifted interpretation as the embedded subject, and ‘you’ in the complement CP, showing that this is possible with ‘you’ referring to Ad*. Now we consider the honorific marking on ‘you’ in this situation. Imagine (49) being said by one teacher to a fellow teacher. Here the second person pronoun in the CP complement must be nonhonorific toraa, reflecting the relationship of Sp* to the addressee, not the honorific apne-ke, reflecting the relationship of Santee (a student) to the addressee.

The representations already given in (45) explain why. Here Santee must control pro_1st directly, not using Sp/Ad, or Ad would be defective and then ‘you’ in the complement CP would not be possible. In other words, the representation must be (45a), not (45b). Since Ad is uncontrolled in (45a), Sp is uncontrolled (Shift Together 2); therefore, Sp and Ad must be bound by Sp* and Ad*. The index {1} associated with the HonP containing ‘you’ is bound by Sp, which is in turn bound by Sp*. Therefore, ‘you.NH’ in (49) indicates that the addressee is socially equal to Sp*, not Santee.

Finally, consider what happens when both ‘you’ and ‘him’ are in the complement of ‘think’ with a pro_1st subject. In this situation, ‘you’ can only refer to Ad* and must have honorific marking relative to Sp*, just as in (43). The interesting thing is that in this environment ‘him’ loses the flexibility that it has in (46): now it too can only bear honorific marking relative to Sp*, as shown in (50). Here the speaker Deepak is talking to his friend Bantee. He says of his younger brother Santee that he thought that he Santee gave Bantee’s phone to a third friend of Deepak’s and Bantee’s, Chotu. Here the nonhonorific forms of both ‘you’ and ‘him’ must be used, showing Deepak’s relationship to Bantee and Chotu, not the honorific forms showing Santee’s relationship to Bantee and Chotu.

This result is also predicted. The presence of ‘you’ implies that the embedded Sp/Ad must not be controlled by the arguments of ‘think’; rather ‘Santee’ must control the subject pro_1st directly. This implies that Sp and Ad are bound by Sp* and Ad*. But the {1} index of Hon in [3^rd Hon] is bound by the closest Sp, just as the {1} index of Hon in [2^nd Hon] is. Therefore, the third person pronoun is interpreted relative to Sp* in this case. As in sections 4.1–4.2, we see that the honorific features on second and third person pronouns behave grammatically in the same way, but some of the surface patterns are different because of the requirement that the second person pronoun is bound by the closest Ad.

In this subsection, we presented a puzzle for the theory of indexical shift: how best to fit (44) into our understanding of this phenomenon. But we have also presented rich additional evidence that the grammar of honorificity marking is deeply related to the grammar of participant pronouns, such that quirks in the distribution of indexical shift correspond closely to quirks in the distribution of honorific marking in Magahi. And that is something worth knowing.