Egophoricity, Involvement, and Semantic Roles in Tibeto-Burman Languages

Abstract In this study, we explore typological aspects of egophoricity marking based on selected Tibeto- Burman languages. Conceptualizing egophoricity as an autonomous grammatical category that marks access to knowledge, we first discuss how egophoricity marking interacts with evidentiality in the Tibeto-Burman languages Shigatse Tibetan and Bunan. We then go on to explore the differences between the egophoricity systems of Shigatse Tibetan and Bunan, arguing that the variability of egophoricity within and across languages can be captured if we distinguish (i) constructions in which egophoricity markers express privileged access to knowledge due to actional involvement in the role of an event participant from (ii) constructions in which egophoricity markers express privileged access to knowledge due to epistemic involvement in the role of a “knower” whose precise relation to the event is not specified. We additionally introduce a set of five semantic roles to offer a more detailed description of the egophoricity systems of Shigatse Tibetan and Bunan (and also, albeit marginally, Kathmandu Newar and Galo). This study thus offers a new perspective on the variability of egophoricity systems in Tibeto-Burman and propagates an analytical approach that may also be helpful for analyzing egophoricity systems in other language families of the world.


Introduction
Egophoricity is a comparatively rare phenomenon in natural languages and has attracted the attention of descriptivists and typologists only recently. It has often been studied resorting to a different terminology -most prominently, the opposition between "conjunct" and "disjunct" verb forms (Hale & Watters 1973); more recently, the marking of the "assertor's involvement" (Creissels 2008). Based on an approach developed by Hargreaves (1991Hargreaves ( , 2005, we define egophoricity as a grammatical category that expresses access to knowledge or, more precisely, to the particular information conveyed in a given utterance. It typically appears marked on the predicate, showing a binary opposition between an egophoric form that denotes privileged access and an allophoric (or non-egophoric) form that denotes general, or non-privileged, access. 1 This study explores, against the background of a number of selected Tibeto-Burman languages, some important parameters of variation of egophoricity. An in-depth exploration of interactions with other grammatical categories such as tense, mood, as well as with clause type, would merit a separate study; we limit ourselves here to outline the relevant interactions with evidentiality (Section 2). We then outline two parameters of language-internal and cross-linguistic variation as found within Tibeto-Burman, viz. involvement and semantic roles, and discuss aspectual correlates of these parameters (Section 3). Lastly, Section 4 closes the paper with some concluding remarks and some suggestions for further research.

Defining egophoricity
Since Hale and Watter's (1973) seminal description of egophoricity in Kathmandu Newar, our understanding of the phenomenon has been enhanced by an ever-growing body of descriptive work.
Nevertheless, in spite of all these studies, there is still no generally accepted definition of egophoricity. In this article, we basically follow Hargreaves (1991Hargreaves ( , 2005, who defines egophoricity as a binary grammatical category that marks one's access to mental states as privileged or nonprivileged. We extend Hargreaves's original conception and define egophoricity as a binary grammatical category that specifies whether one has privileged or non-privileged access to the knowledge that is conveyed in a proposition. The notion of privileged access as we understand it describes a privileged epistemic relationship that holds between a speech-act participant (SAP) and the knowledge conveyed in a proposition. Egophoric markers thus indicate that a given SAP has an inside perspective on an event and possesses the "epistemic authority" (see Hargreaves 1991) to assert the relevant facts. Allophoric markers, in contrary, indicate that a given SAP merely has an outside perspective on an event and does not possess any special epistemic authority to assert the relevant facts.
It is well known that egophoricity markers do not always relate to the perspective of the speaker but may relate to the perspective of other SAPs depending on the type of speech act. Most notably, there is a striking contrast between assertions, in which egophoricity markers commonly occur in combination with 1 st person pronouns, and questions, in which egophoricity markers commonly occur in combination with 2 nd person pronouns. This characteristic pattern is illustrated in Table 1 below. A number of scholars have tried to account for these perspective shifts by postulating a category that comprises the different SAPs to which egophoricity markers may relate, e.g. "locutor" (Aikhenvald 2004), "informant" (Bickel 2008), "epistemic source" (Hargreaves 1991(Hargreaves , 2005, "assertor" (Creissels 2008), and "source of information" (Tournadre 2008). For the following discussion, we adopt Creissels' term "assertor", as it captures the functional motivation of the category rather well and -unlike some other proposed terms -is not already associated with a different meaning. We understand the assertor as an EPISTEMIC ROLE that refers to the SAP from whose perspective the epistemic status of a given proposition is evaluated. Our definition of the assertor and egophoricity is closely related to Kamio's (1997) notion of TERRITORY OF INFORMATION, defined as one's sphere of personal knowledge. The assertor can thus be thought of as the SAP against whose territory of information a proposition is profiled. Egophoricity, in turn, can be thought of as a grammatical category that indicates whether or not the knowledge in a proposition is conceptualized as being part of the assertor's territory of information (see Dickinson 2016).
As noted above, the assertor role typically falls to the speaker in assertions and to the addressee in questions. This is illustrated by the following examples from Kathmandu Newar, 2 in which the egophoric form -ā relates to the speaker in (1a) and (1b) and to the addressee in (1c) and (1d).
2 Kathmandu Newar is a Tibeto-Burman language that is spoken by approximately 600,000 people in Central Nepal. All data have been adopted from Hale & Shrestha (2006).
(1) Kathmandu Newar (Hale & Shrestha 2006)  Things are slightly more complex in the domain of reported speech, where two different speech events have to be distinguished: (i) a primary speech event, which minimally includes a primary speaker and a primary addressee, and (ii) a reported speech event, which minimally includes a reported speaker and a reported addressee (see Evans 2012). In reported speech clauses, the assertor role commonly falls to the reported SAPs, i.e., the reported speaker of a reported assertion and the reported addressee of a reported question. This is illustrated by the following Kathmandu Newar examples, in which the egophoric form -a relates to the perspective of the reported speaker in (2a) and (2b) and to the perspective of the reported addressee in (2c) and (2d).
(2) Kathmandu Newar There is thus a strong correlation between the distribution of the assertor role and the parameter of clause type. In declarative clauses, the role commonly falls to the speakers, whereas in interrogative clauses, the role falls to the addressee. Nevertheless, the correlation is not perfect, as there are certain exceptions, e.g. rhetorical interrogative clauses. Such clauses differ from true interrogative clauses in the sense that they do not ask for information but rather assert facts (see Heritage 2012).
In other words, a rhetorical interrogative clause profiles a proposition against the speaker's perspective rather than the addressee's. As a consequence, one often encounters a non-canonical distribution of egophoricity markers in such contexts. This is illustrated by the following we espouse here, grammatical person is defined in the same way across languages, based on speech-act roles (with languages differing as to which values they instantiate, and how); some languages additionally feature a grammaticalized system of elements that denote epistemic role.

Egophoricity and evidentiality
The grammatical status of egophoricity and its relationship to other grammatical categories associated with knowledge such as evidentiality and epistemic modality (all of which we will subsume under the label EPISTEMIC MARKERS here) is still a matter of dispute.
We distinguish egophoricity from these two notions in this study. First, when asserting a proposition p, the speaker may comment on how certain (s)he is that p is true. Given the definition of egophoricity we provided in the introduction (i.e. privileged access to knowledge about a state of affairs), it is evident that there might be interesting correlations between egophoricity and individual values of these two related categories. It is equally evident, however, that neither category can be mechanically reduced to any other one. On the other hand, just as dynamic and deontic modals may resort to similar means of expression as epistemic modals, as well as the latter and evidentials, epistemic modals and evidentials can become conflated with egophoricity. (By "conflated" we mean not only that similar or the same formal means may develop in such a way that they mark egophoricity and evidentiality, for example, but also that one particular morpheme may express both categories in an instance of cumulative exponence.) The matter is an intricate one, both at the conceptual and the empirical level, and it is not our intention to treat the legitimate concerns voiced by scholars as to the adequacy of the approach espoused here (e.g. Tournadre & LaPolla 2014) in a cavalier fashion. Rather, we chose not to focus on these intricacies and to use a comparatively simple model where categories are distinct in principle in order to more easily explore some specific formal and content-related interactions.

Egophoricity in Shigatse Tibetan and Bunan
In the following subsections, we describe the interaction of egophoricity and evidentiality in two Tibeto-Burman languages of the Himalayas: Shigatse Tibetan and Bunan. 3 We chose these two languages for two reasons. First, we had access to comprehensive grammatical descriptions as well as native speakers for both languages. Second, the egophoricity systems of Shigatse Tibetan and Bunan show a considerable deal of variation both within and between the two languages.
Accordingly, Shigatse Tibetan and Bunan are ideal candidates for exploring the typological parameters along which egophoricity systems may vary. We limit ourselves to the discussion of present/imperfective as well as past/perfective paradigms and exclude future and perfect forms. 4

Present tense / imperfective aspect
The imperfective paradigm of Shigatse Tibetan comprises three forms: (i) an egophoric form -kī=joe, (ii) an allophoric direct evidential form -kì, and (iii) an allophoric indirect evidential form -kī=joapie. This is schematically presented in Table 2 below. second most important Central Tibetan variety after Lhasa Tibetan, to which it is closely related (see Haller & Haller 2007). The Shigatse Tibetan data have been adopted from Haller & Haller (2007). Bunan (a.k.a. Gahri) is a West Himalayish language that is spoken by approximately 4,000 speakers in Himachal Pradesh, Northern India. The Bunan data have been adopted from Widmer (forthcoming). 4 It is in order to add that both languages show egophoricity in the future subparadigms as well. By contrast, only Shigatse Tibetan shows egophoricity with perfect forms; in Bunan, the egophoricity distinction, perhaps due to its relatively young age (see Widmer 2015), does not appear in perfect forms.
In addition, speakers use egophoric forms to describe other event types that occur habitually (5).
This is use of -kī=joe presupposes, however, that the relevant event can be construed as the result of a cause-effect relationship. Otherwise, the use of the marker -kī=joe is not possible. 5 The present tense paradigm of Bunan differs from the Shigatse imperfective paradigm in two ways.
First, the paradigm is only based on a simple egophoricity opposition, that is to say, there is no distinction between direct evidentiality and indirect evidentiality with allophoric forms. Second, the egophoricity distinction is paired with a singular-plural number distinction. The relevant forms are given in Table 3 below. There are two further differences between egophoricity marking in Shigatse Tibetan and Bunan that are noteworthy. First, the Bunan egophoric forms -ek and -h ek cannot be used to describe any habitual or generic situation that is part of the assertor's intimate and personal sphere of knowledge if the assertor (e.g. If the child becomes afraid, it cries; cf. (5c) above). Rather, the marker is bound to contexts in which speakers assume the role of a volitional agent or an experiencer of an internal stimulus (cf. (7) above). Second, the Bunan allophoric forms -are and -h ak cannot be used to express nonvolitionality in combination with verbs that denote prototypically controllable events (e.g. I am inadvertently giving the money to you; cf. (6c) above). Table 4 below compares the present/imperfective egophoricity systems of Shigatse Tibetan and Bunan. Table 4. Egophoricity and evidentiality in the two languages (present / imperfective)

Past tense / perfective aspect
The perfective paradigm of Shigatse Tibetan comprises four forms and is thus more complex than the imperfective paradigm. The relevant forms are (i) an egophoric form -pa=jĩ for actors, (ii) an egophoric form =tɕu for undergoers, (iii) an allophoric direct evidential form -so, and (iv) an allophoric indirect evidential form -papie; see Table 5 below.
The actor-egophoric form -pa=jĩ is used to describe one-time past events that speakers instigated volitionally (9). Note that -pa=jĩ cannot be used to describe recurring events in the past. Only the imperfective ending -kī=joe can serve this function as exemplified in (4b) above. The undergoer-egophoric form =tɕu, in turn, is used to describe events in which speakers experienced an internal state (10a), perceived an external stimulus (10b), performed a prototypically noncontrollable event (10c), or partook in a process or in which they can be construed as the goal of a (di)transitive event (10d). In addition, speakers use =tɕu to refer to past events with which they are well-acquainted, in which case the marker often implies that the event took place repeatedly in the past (11). Note, however, that the marker =tɕu cannot be used to express acquaintance with prototypically controllable events that occurred in the past. Only the marker -kī=joe can serve this function (cf. (4b) above). The allophoric marker -so is used for events that speakers observed directly, but which they do not consider to be part of their personal knowledge (12a). In combination with prototypically controllable events, the marker -so expresses unconscious acting on behalf of the speaker (12b). 1SG.ERG probably cup=DEF=ADD break-PFV.INDIR.ALLO 'Probably I also broke the cup (in the state of drunkenness).' (185) Bunan has four kinds of markers in the past tense: (i) an actor-egophoric marker, (ii) a composite undergoer-egophoric marker, (iii) allophoric direct evidential markers, and (iv) allophoric inferential evidential markers. In addition to the number-related allomorphy shown by some of the markers, some of the elements display different allomorphs based on the (transitivity-related) conjugation class of the verb to which they attach. 9 Table 6 below gives an overview of the different forms.
Bunan speakers use actor-egophoric forms to describe prototypically controllable events that they performed volitionally (14a) and mental states or body processes that involved an internal stimulus that was only directly accessible to themselves (14b).
1SG=ERG.SG food make-TR-PST.EGO.ACT In addition, actor-egophoric forms can be used to describe events with which the speaker is wellacquainted. Note that there are differences between speakers of different age groups in terms of the use of actor-egophoric forms. Members of the younger generations only use actor-egophoric forms to describe events in which they assumed the role of the most agent-like participant (15a)-(15b).
Members of the oldest generation also use such forms when the speaker assumes a less agentive role or does not occur as a core argument at all (15c)-(15d). with the allophoric direct evidential marker -dza / -ts h a. Also, the marker is only commonly encountered in the speech of the oldest generation, and even there it is not obligatory. Accordingly, (16a) and (16b) would also be grammatical without the undergoer-egophoric form. Allophoric direct evidential forms are used to describe events which were directly witnessed by speakers but to which they do not possess privileged access by virtue of assuming the role of agent or an experiencer of a mental state or body process (17a)-(17b).

Modeling egophoricity in a typological perspective
In this section, we model the egophoricity systems of Shigatse Tibetan and Bunan in a typological perspective in order to gain a better understanding of the language-internal and cross-linguistic variability of egophoricity. Before going into this matter, however, we first introduce and discuss two major parameters of variation: involvement and semantic roles.

Involvement
Bickel (2008) postulates a parameter "scope" to capture some aspects of the cross-linguistic variation of epistemic markers. His proposal is based on the observation that epistemic markers can be used to mark any kind of proposition in some languages ("epistemic proposition marking") but are tied to specific arguments or semantic roles in others ("epistemic argument marking"). 11 We essentially adopt Bickel's approach for the following discussion of egophoricity marking, but use a somewhat different terminology. Instead of his term "scope" (2008), we adopt the term "involvement", a notion that has figured prominently in Creissels' (2008)  Drawing on Bickel's (2008) ideas, we assume that egophoricity markers can express two different types of the assertor's involvement in an event: (i) "actional involvement" and (ii) "epistemic involvement". In the former case, egophoric markers indicate that the assertor has privileged access to the knowledge about an event because (s)he was directly involved in it. In other words, such constructions profile the assertor in her / his role as an event participant. In the case of epistemic involvement, egophoric markers indicate that the assertor has privileged access to the knowledge about an event because (s)he is well-acquainted with the relevant facts. Accordingly, such constructions profile the assertor in her / his role as a "knower" without specifying whether the assertor directly participated in an event.
The distinction between these two types of involvement is illustrated in Figure 1 based on a semantic representation in the form of a logical structure. The logical structure describes a semantically unspecified event predicate, which takes the arguments ARG 1 and ARG 2 (and possibly further arguments). These arguments are directly involved in the event in question. The functional predicate KNOW, in turn, takes the two arguments: ARG 0 and the event predicate. The first argument of KNOW corresponds to the knower role and can be thought of as an experiencer at the metalinguistic level. Egophoric markers can either portray the assertor as an argument of the event 11 A somewhat similar approach has been developed by Tournadre (2008), who uses the term "scope" to describe the distribution of egophoric markers in Lhasa Tibetan, distinguishing between "narrow scope" and "wide scope" egophorics. Tournadre's dichotomy comes close to Bickel's (2008) distinction between "epistemic argument marking" and "epistemic proposition marking", but is not identical with it. 12 We do not use the term "scope" because this term is already established in the literature with a rather different meaning. In formal semantics and other areas of theoretical linguistics, the concept of scope is commonly used to denote the modification relationship that a semantic operator bears to specific constituents within an utterance (see Cann 1993: 8-9). Bickel (2008) uses the term "scope" in a different sense to describe whether an egophoric marker is tied to a specific argument role within a proposition or to the proposition as a whole.
predicate, in which case they express privileged access due to actional involvement, or they can portray the assertor as an argument of the predicate KNOW, in which case they express privileged access due to epistemic involvement.  In (19b), in which the assertor expresses the fact that (s)he is currently sick, egophoric marking is not possible in spite of the fact that the assertor is directly involved in the relevant event. This is because the form -kī=joe can exclusively express privileged access due to actional involvement if the relevant event involves a volitional agent. Since the semantics of the verb na̠ ː-'be sick' is not compatible with such an interpretation, the imperfective allophoric form -kì has to be used. The use of -kī=joe with the verb na̠ ː-'be sick' only becomes possible if the event of being sick is portrayed as the consequence of a causal relationship. These conditions are met in (19c), in which the assertor describes the fact that (s)he becomes sick when eating a specific kind of food. In this case, however, the privileged access that licenses egophoric marking is not given through the assertor's actional involvement in the event but through her / his epistemic involvement. In other words, the assertor has privileged access to this knowledge not because (s)he is currently sick but because (s)he became sick in the past after having eaten the respective kind of food. The fact epistemic involvement does not necessarily entail direct participation in an event is evidenced by (19d). Here, the assertor describes a child's natural reaction to being afraid. In this example, the assertor's privileged access cannot be due to actional involvement, as (s)he is not an argument of the predicate. Accordingly, the assertor's privileged access is again due to epistemic involvement, which naturally gives rise to a habitual interpretation. 13 The difference between privileged access due to actional involvement and privileged access due to epistemic involvement also plays an important role in the use of the past actor egophoric -et ~ -men in Bunan. However, epistemic involvement has a somewhat different semantic connotation in the case of the endings -et ~ -men, as the marker is associated with an experiential or an assertive reading rather than a habitual reading in such contexts. Consider the following example sentences.  13 Two reviewers raised the question of whether it is sensible to distinguish between actional involvement and epistemic involvement in combination with events in which the assertor is directly involved as a participant, and whether it would not be more adequate to interpret actional and epistemic involvement as a function of the assertor's participation or nonparticipation in an event. We believe that such an approach would fall short for the following reason: If one puts actional and epistemic involvement on a level with event participation and event non-participation, one can no longer explain why the imperfective egophoric form -kī=joe can have a habitual reading in combination with all predicate classes, while a progressive reading is restricted to prototypically controllable predicates. By contrast, assuming that the distinction between actional and epistemic involvement is independent of the assertor's participation or nonparticipation in an event makes it possible to explain these aspectual differences in terms of the construal of the egophoric marker. The approach that we argue for has therefore more explanatory power than a model that treats even participation and involvement as interdependent notions. In addition, the fact that Bunan shows a very similar pattern of variation (see below in the main body of text) suggests that this distinction is not an idiosyncracy of Shigatse Tibetan but is also of importance for other Tibeto-Burman languages. To be sure, it is conceivable that there are certain constructions in a language in which the distinction between actional and epistemic involvement is neutralized. It is equally conceivable that there are languages in which egophoric markers can either only express actional or epistemic involvement but not both types. However, the data currently available for Shigatse Tibetan and Bunan suggest that it is justified to treat the assertor's involvement in an event and the assertor's participation in an event as independent notions. In (20a), the egophoric marker -et can express privileged access both due to actional involvement and epistemic involvement. In the first case, the assertor is portrayed as the volitional instigator of a specific past event, which results in the past reading captured by translation (i). In the second case, the assertor is portrayed as a knower who has experienced the relevant event before in her / his life, which results in the experiential reading captured by translation (ii). In (20b), the assertor reports describes her / his falling from a tree as a specific and individual event. Since the egophoric marker -et can only express privileged access due to actional involvement in combination with prototypically controllable events and certain experiencer events (see § 3.2), egophoric marking is not possible when conveying this meaning. However, egophoric marking is possible if the assertor wants to express that (s)he has intimate knowledge about the relevant facts. This is the case in example (20c), which was uttered by a language consultant on asking whether he had ever fallen from a tree when working in his apple orchard. With his answer, he portrayed himself as having intimate knowledge about this event not because he had recently fallen from a tree but because he had fallen from trees many times in his life. As in Shigatse Tibetan, egophoric markers can also express privileged access due to epistemic involvement in combination with events in which the assertor was not directly involved. 14 This is illustrated in (20d), in which the assertor states that a given monk had joined the local monastery much later than the addressee had previously claimed.
As noted above, such statements sometimes display an emphatic assertive connotation, which is captured by the expression [I know that] in the English translation.
Our survey of selected constructions in Shigatse Tibetan and Bunan demonstrates that the distinction between actional and epistemic involvement is crucial for the egophoricity systems of those two languages. At the same time, it is evident that the notion of involvement is not sufficient to adequately describe the functional range of individual egophoric markers. We additionally have to take into account what kind of semantic role the assertor plays in a given event. This leads us to the next parameter of variation, which will be discussed in the following section.

Semantic roles
In Shigatse Tibetan and Bunan, egophoric markers cannot express actional and / or epistemic involvement in any given context. Rather, they can only be used if the assertor assumes certain The volitional agent is the agentive participant of a prototypically controllable event (e.g. run, make, give), while the nonvolitional agent is the agentive participant of a prototypically noncontrollable event (e.g. fall, forget, lose). The role of the endopathic experiencer has been adopted form Tournadre (2008), and we here additionally introduce the role of the exopathic experiencer as an antonym. The former is the experiencer in a state of affairs that involves a mental state or process that is only directly accessible to the experiencer herself / himself (e.g. be hungry, be thirsty, be exhausted); we also include states of affairs in which the participant engages in a cognitive activity (e.g. think, presume) or experiences a bodily process that is related to some internal stimulus (e.g. vomit, cry, laugh). The exopathic experiencer, in turn, is the experiencer in an event that involves an external stimulus that is equally accessible to other persons (e.g. see, hear, smell). Lastly, the undergoer corresponds to all other non-agentive roles, most importantly patients, themes, and recipients. 16 There is evidence that these five semantic roles can be located on a hierarchy that describes the relative probability of a given semantic role to be associated with the expression of privileged access due to actional involvement. This hierarchy is given in Figure 2 below. 17 This hierarchy emerges from a comparison of selected Tibeto-Burman languages in which egophoric markers express privileged access due to actional involvement in combination with different semantic roles. The languages in question are Shigatse Tibetan, Bunan,Kathmandu Newar,and Galo. 18 Example sentences Shigatse Tibetan and Bunan have already been given in the preceding sections and will be discussed in more detail in § 4.2 below. In the following, we will provide some more discussion of egophoricity marking in Kathmandu Newar and Galo to illustrate how egophoricity marking in those languages relates to the proposed hierarchy.
Consider the following sentences from Kathmandu Newar, which illustrate the use of the past egophoric marker -ā expressing actional involvement. As the example sentences illustrate, the Kathmandu Newar egophoric marker -ā expresses actional involvement in combination with agent arguments (21a) and certain endoceptive experiencers (21b), although some endoceptive experiencers are not associated with egophoric marking (21c).
The following example sentences illustrate the use of the perfective egophoric marker -tó in Galo expressing actional involvement. As the examples illustrate, the perfective egophoric marker -tó expresses actional involvement with volitional agents (22a), endoceptive experiencers (22b), and nonvolitional agents (22c). Post (2013) does not provide examples of the perfective construction in which the assertor assumes the role of an exopathic experiencer or an undergoer. However, he states that verbs like káa-'see' and nampáa-'catch (the) smell (of something)' (i.e. verbs taking an exopathic experiencer) require egophoric marking if their experiencer argument and the assertor are coreferential. On a more general note, Post notes that the egophoric marker occurs whenever the speaker acts as the "subject" of a predicate in a declarative clause. His account of the system thus suggests that egophoric marking is obligatory whenever the assertor assumes the role of the most agent-like participant of a predicate but impossible if the assertor assumes the role of the most patient-like participant of a (bivalent) predicate.
The following table illustrates which semantic roles are associated with privileged access due to actional involvement in combination with past tense / perfective egophoric markers in the relevant languages. As previously mentioned, Shigatse Tibetan and Bunan each display an actor-egophoric and an undergoer-egophoric marker in the past / perfective paradigm.
As can be seen from Table 8, the functional scope of individual egophoric markers varies considerably across these four languages. In Shigatse Tibetan, the perfective actor-egophoric marker only expresses privileged access due to actional involvement in combination with volitional agents. In Kathmandu Newar and Bunan, the corresponding egophoric markers have a somewhat wider functional range, comprising both volitional agents and at least some endopathic experiencers. In Galo, finally, the perfective egophoric marker covers volitional agents, endopathic experiencers, exopathic experiencers, and nonvolitional agents (i.e., it expresses actional involvement in combination with the most agent-like argument of a predicate). Shigatse Tibetan and Bunan additionally display undergoer-egophoric markers, which again strongly differ from each other in terms of the semantic roles they can cover. The Shigatse Tibetan undergoer-egophoric marker expresses privileged access due to actional involvement in combination with all semantic roles except volitional agents, while the Bunan undergoer-egophoric marker only covers undergoers.
In spite of the striking differences between these four languages, there are some potential distributional tendencies emerging from Table 8. First, if a language has an actor-egophoric marker, the volitional agent appears to be the semantic role that is most likely to be associated with the marking of actional involvement. Second, if a language only has an undergoer-egophoric marker, the undergoer appears to be the semantic role that is most likely to be associated with the marking of actional involvement. Third, the existence of an undergoer-egophoric marker appears to presuppose the existence of an actor-egophoric marker. It goes without saying that such tendencies should be confirmed or contradicted by a much more comprehensive in-depth survey of the Tibeto- Burman family than what we sketch in the present article.
The ranking of semantic roles on the hierarchy given in Figure 2 can be explained in terms of the cognitive exclusiveness of mental states that they are associated with (see Widmer & Zemp 2017).
The highest-ranked semantic roles are volitional agents and endopathic experiencers, that is to say, arguments that are associated with mental states that are only directly accessible to the person experiencing them. As one moves further down the hierarchy, the semantic roles become associated with less exclusive and less self-initiated mental states. These lower-ranked semantic roles may still be the associated with the marking of actional involvement, as the cases of Shigatse Tibetan, Bunan, and Galo illustrate. However, if they are, they tend to be associated with a dedicated undergoeregophoric marker that covers non-agentive semantic roles.

Modeling
In the following subsections, we use the hierarchy introduced in Figure 2 to model how the distribution of egophoric markers maps onto the identified semantic roles and the explicit participation of the assertor. For this purpose, we combine the hierarchy of semantic roles with the distinction between events in which the assertor is a participant (ASS [+PART] ) and events in which the assertor is not a participant (ASS [-PART] ) and plot it against the distinction of actional involvement vs.
epistemic involvement, as in Figure 3 below. column "semantic roles" allows us to specify in more detail the contexts in which egophoric markers occur. This row is relevant for both Shigatse Tibetan and Bunan, in which egophoric markers expressing actional and -to a lesser extent -epistemic involvement are sensitive to semantic roles (see § 4.1.2). The column "involvement", finally, allows us to specify whether an egophoric marker can express actional involvement, epistemic involvement, or both types of involvement in a given context. 19 We use this representation in the next subsection to compare the egophoricity markers of Shigatse Tibetan and Bunan.

Modeling Shigatse Tibetan egophoricity markers
In Shigatse Tibetan, the imperfective egophoric form -kī=joe can either express actional or epistemic involvement. The expression of actional involvement is restricted to prototypically controllable events, in which case the marker profiles the assertor as the volitional agent of a specific, individual event (4a). In such contexts, -kī=joe receives a progressive interpretation. The expression of epistemic involvement is possible in combination with all event types and profiles the assertor in her / his role as a knower who is well-acquainted with the habitual occurrence of an event (4a)-(4b), (5b)-(5c). In such contexts, -kī=joe usually receives a habitual interpretation. Note, however, that the epistemic construal of -kī=joe is only possible if it can be construed as the result of a cause-effect relationship with which they are well-acquainted. Consider Figure 4 below. AGT [+VOL] -kī=joe -kī=joe Egophoricity marking is more complex in the perfective aspect, where we encounter two egophoric forms: -pa=jĩ and =tɕu. The former expresses privileged access due to actional involvement in combination with prototypically controllable events and profiles the assertor in the role of a volitional agent of a specific, individual event. The resulting verb forms are thus associated with a punctual-perfective reading (9). Note that the marker cannot express epistemic involvement. This is evidenced by the fact that the marker cannot be used to profile the assertor in the role of a knower who has privileged access to the knowledge about an event. The marker =tɕu has a much wider distribution; it can express privileged access due to actional involvement in combination with the semantic roles EXP ENDO (10a), EXP EXO (10b), AGT [-VOL] (10c), and UND (10d). In such contexts, it profiles the speaker as an undergoer of a specific perfective event. Accordingly, this use of the marker =tɕu is associated with a punctual-perfective reading. The marker can also express epistemic involvement in contexts in which the assertor does not assume the semantic role AGT [+VOL] (11a)-(11b). In such contexts, =tɕu profiles the assertor in the role of a knower who has privileged access to the knowledge about an event because (s)he has experienced the relevant process in the past. This often presupposes that the assertor has repeatedly observed the event in question. As a consequence, this use of =tɕu is usually associated with a habitual past reading. Note that the relevant event does not have to be construed as the result of a cause-effect relationship for =tɕu to express epistemic involvement as is the case with the imperfective marker -kī=joe (see above).
The markers cannot express privileged access due to epistemic involvement. This is evidenced by the fact that the markers cannot be used to profile the assertor in the role of a knower who has privileged access to the knowledge about an event. This is depicted in Figure 6. AGT [+VOL] -ek / -h ek - In the past tense, there is variation in the use of egophoric forms between the younger speaker generations and the oldest speaker generation. In the variety of the former, the past-tense egophoric marker -et ~ -men can express privileged access due to actional involvement in combination with the semantic roles AGT [+VOL] (14a) and EXP ENDO (14b). The resulting verb forms express a past tense.
In addition, the marker can express privileged access due to epistemic involvement in combination with propositions in which the assertor acts as the most agent-like participant (15a)-(15b). In such contexts, the assertor is portrayed in the role of a knower who has privileged access to the knowledge about an event because (s)he experienced the relevant process in the past. As a consequence, the resulting verb forms express an experiential aspect. Consider Figure 7 below. In the more conservative variety of the oldest speaker generation, the past-tense egophoric marker -et ~ -men can express privileged access due to epistemic involvement in combination with any proposition in which the assertor occurs as a participant (15c)-(15d). As in the innovative variety spoken by younger speaker generations, the past-tense egophoric marker expresses an experiential aspect in such contexts. Furthermore, the variety of the oldest speaker generation features an undergoer-egophoric marker -ku. This marker expresses actional involvement in combination with the semantic role of an undergoer (16a)-(16b), but cannot express epistemic involvement. Note that the suffix is not obligatory and is in fact often omitted. Consider Figure 8 below.

A comparison
In the present tense / imperfective aspect, Shigatse Tibetan and Bunan egophoric markers differ strongly from each other with regard to the parameter of involvement. The Shigatse imperfective egophoric form -kī=joe can express privileged access due to actional involvement in combination with the semantic role AGT [+VOL] and privileged access due to epistemic involvement if the relevant event can be construed as the result of a cause-effect relationship. The Bunan present-tense egophoric forms -ek / -h ek, in turn, can only express privileged access due to actional involvement in combination with the semantic roles AGT [+VOL] and EXP ENDO but not privileged access due to epistemic involvement. Accordingly, the Bunan egophoric forms -ek / -h ek can express privileged access due to actional involvement with a wider range of semantic roles, but are at the same time more restricted in the sense that they cannot express privileged access due to epistemic involvement.
These differences most probably reflect the distinct diachronic origins of the two egophoricity systems. The Shigatse Tibetan egophoric form goes back to a periphrastic construction that consisted of a nominalizer and a copula. The Bunan egophoric endings -ek / -h ek, in turn, were once first person subject agreement forms (Widmer 2015;Widmer & Zemp 2017). The more restricted functional scope of -ek / -h ek is a consequence of the fact that these morphemes were already tied to specific semantic roles when they were still subject agreement markers. The Shigatse form -kī=joe, which never expressed agreement, is not restricted in such a way.
With regard to the past tense / perfective aspect paradigm, both Shigatse and (conservative) Bunan have two egophoric markers, viz. an actor-egophoric form and an undergoer-egophoric form.
The two languages differ considerably as to the functional range of these markers. In Shigatse Tibetan, the actor-egophoric form -pa=jĩ is exclusively associated with agentive contexts and expresses privileged access due to actional involvement in combination with volitional agents. The undergoer-egophoric form =tɕu, in turn, can express privileged access due to actional involvement in combination with all semantic roles except volitional agents and privileged access due to epistemic involvement in combination with propositions in which the assertor does not assume the role of a volitional agent. In Bunan, we encounter a rather different situation. Here, it is the undergoer-egophoric -ku form that is restricted in the sense that it can only express privileged access due to actional involvement in combination with undergoer participants. The actor-egophoric form, in turn, can express privileged access due to actional involvement in combination with volitional agents and endopathic experiencers as well as privileged access due to epistemic involvement in all other contexts. The innovative paradigm of younger Bunan speakers differs from the paradigm of old speakers in two respects. First, the former has lost the undergoer-egophoric form -ku. Second, in this variety the marker -et ~ -men can only express privileged access due to actional involvement in contexts in which the assertor acts as the most agent-like participant.

Concluding remarks
Assuming that egophoricity and evidentiality are two distinct grammatical categories, we have argued that some important sources of the cross-linguistic variation found in egophoricity systems of Tibeto-Burman languages can be captured with two analytical tools, viz. an epistemic role that we refer to as the "assertor" (see Creissels 2008) and a parameter of variation that we refer to as "involvement" (see Bickel's 2008 related "scope"). The notion of the assertor allows us to capture the fact that egophoricity markers may relate to different SAPs depending on the type of speech act under scrutiny. The involvement parameter allows us to capture the fact that egophoricity markers may express privileged access due to both actional and epistemic involvement. In order to model the two types of involvement in a more detailed manner, we have additionally postulated a set of five roles based on evidence from the Tibeto-Burman languages Shigatse Tibetan, Bunan, Kathmandu Newar, and Galo. The relevant roles are volitional agent, endopathic experiencer, endopathic experiencer, nonvolitional agent, and undergoer. Equipped with these notions, we have explored egophoricity in Shigatse Tibetan and Bunan in more detail and shown that (i) that egophoric markers vary significantly in terms of their functional scope between the two languages and (ii) that there is considerable variation between different TAM paradigms within each language.
The functional properties of Bunan egophoric forms additionally vary across varieties spoken by different speaker generations. This indicates that the typological model developed here can be employed to capture synchronic variation on the one hand and to model egophoricity in a diachronic perspective on the other.
Overall, the results of our study imply that the notion of the assertor and the parameters of involvement and semantic roles are crucial concepts for comparing the egophoricity systems of Tibeto-Burman languages. In addition, our study suggests that it is helpful to treat egophoricity and evidentiality as separate grammatical categories when conducting typological research on egophoricity. This makes it possible to directly compare egophoricity systems that are based on a binary contrast of egophoric vs. allophoric (e.g. the Bunan present paradigm) to egophoricity systems that are structurally more complex and additionally involve evidential contrasts (e.g. the Shigatse imperfective and perfective paradigms).
While our model is able to capture basic differences between individual egophoricity markers and their interaction with evidentials, future research will undoubtedly uncover further parameters along which egophoricity markers in Tibeto-Burman languages may vary, which will call for a refinement of the approach presented in this paper. It remains to be seen whether the proposed set of semantic roles will have to be extended, and whether the syntactification of this parameter (i.e. the expression of privileged access due to actional involvement in combination with specific grammatical relations rather than semantic roles) is also relevant for Tibeto-Burman. It also remains an open question whether the notion of epistemic involvement should be split up into further subparameters. Our study suggests that there is a considerable amount of crosslinguistic variation in this subdomain of egophoricity marking, both with regard to the aspectual and pragmatic connotations that are associated with egophoric markers expressing epistemic involvement and with regard to the semantic restrictions that such markers may be subject to. Further research is needed to gain a better understanding of these phenomena and how the can be integrated into a more comprehensive typological model.
Going beyond the Tibeto-Burman language family, it remains to be seen whether the typological approach proposed here can be extended to egophoricity systems in non-Tibeto-Burman languages, e.g. languages from the Caucasus, northwestern South America, and Papua New-Guinea. In this context, a particularly interesting question is whether there are large-scale distributional trends of the typological parameters proposed in this article. Is the distinction between actional and epistemic involvement is equally important in all languages that display egophoricity systems? Or are there areas / language families in which one type of involvement is clearly favored over the other?
Another open question is whether the tendency of egophoricity markers to be associated with agentive semantic roles is a characteristic trait of Tibeto-Burman languages or whether this trend is can also be found in other languages. Future research will hopefully allow us to resolve these questions and to refine our understanding of the typological variability of egophoricity systems.