Grammar and levels of addressivity

Evans et al. (2018a,b) introduce the notion of ‘engagement’ as a new grammatical domain related to intersubjective coordination of knowledge. The present paper applies this notion to data from the Australian Aboriginal language Ungarinyin. It identifies three markers/construction types in the language as expressions of engagement and develops a descriptive framework rooted in Bakhtinian Dialogism in order to demonstrate why these expressions represent the category. It is argued that the main problems that arise in the analysis of engagement are very similar to those that have been encountered in the description of (other) TAME-categories aswell, and that thesemay be overcomeby applyingMikhail Bakhtin’s idea of ‘addressivity’. It concludes that a better understanding of the category of engagement that explores its relation to addressivity may contribute to the development of an approach to grammar in which sociality takes priority, a Dialogic linguistics.


Introduction: observing and interpreting engagement
To say that the human being behaves individually at one moment and socially at another is as absurd as to declare that matter follows the laws of chemistry at a certain time and succumbs to the supposedly different laws of atomic physics at another' (Sapir, 1949, 545) In an agenda-setting discussion of epistemicity in grammar, Evans et al. (2018a,b) argue for the adoption of a new grammatical category, that of engagement. The main function of this category is to signal knowledge asymmetries between the speaker and an addressee: with an engagement marker the speaker indicates whether she expects the communicated content to reflect knowledge the addressee already shares or not. Engagement can take many forms, from dedicated inflectional categories or particles, as in (1), to expressions that would traditionally be associated with other functions, such as do 'speaker proximate' in (2), which doubles as a demonstrative pronoun. (1) Andoke (Colombia, unclassified) a. duiʌhʌ whites b-ə +spkr+addr.engag-3pl dã-ə-ʌ ingr-move-3 'It's the whites arriving (as we can both witness)' b. duiʌhʌ whites kẽ-ə +spkr-addr.engag-3pl dã-ə-ʌ ingr-move-3 'It's the whites arriving (which I know / can witness but you can't)' (Landaburu, 2007, 25, cited in Evans et al., 2018a (2) Abui (Indonesia, Timor-Alor-Pantar) na 1sg.a nala something.eat-phsl nee-ti bad beeka cannot do sp.prx 'I couldn't eat up (swallow) anything' (Kratochvíl, 2011, cited in Evans et al., 2018b In (1-a), the prefix b-indicates that the event of the 'whites arriving' is epistemically accessible to both the speaker and the addressee, whereas in (1-b) the prefix kẽ-signals that epistemic access is asymmetric: the speaker presents her/himself as having knowledge about the event, but not the addressee. The prefix casts the speaker as having 'epistemic authority' (Heritage and Raymond, 2005). In (2) the demonstrative do 'speaker proximate' 'stress[es] the speaker's foundation for his assertion in immediate experience' (Evans et al., 2018b, 160): 'members from the basic set [of Abui demonstratives] can be placed in sentence-final position to index the distribution and extent of knowledge among speech-act participants' (Evans et al., 2018b, 161). In other words, according to this analysis, in (2) do 'speaker proximate' signals that the speaker states that the information that s/he 'couldn't eat up anything' is grounded in knowledge/experience that the speaker has, but the addressee cannot (yet) ascertain. Evans et al. (2018a,b) argue 'that many languages have grammaticalised systems for monitoring and adjusting intersubjective settings; it is this grammaticalised intersubjectivity which we refer to as engagement, in much the same way as grammaticalised time representation merits the special metalinguistic term tense.' (Evans et al., 2018a, 113-114) 'Engagement refers to a grammatical system for encoding the relative accessibility of an entity or state of affairs to the speaker and addressee' (Evans et al., 2018b, 141) At the same time, Evans et al. (2018b) problematise the distinction between grammaticalised and nongrammaticalised expressions of engagement: 'The sorts of epistemic management mechanisms we have illustrated, in the pointedly grammaticalised forms we have been calling engagement, have been widely investigated in the conversation analysis literature, but in the languages examined there the formal coding is much more diffuse -involving prosody, gesture, tactical restatement, or the use of epistemic particles or adverbials like well or actually. What difference does this semiotic investment make? (Evans et al., 2018b, 166) These are important questions. To what extent are the knowledge asymmetries diagnosed through the methodology of conversation analysis in languages such as English comparable to strategies found in languages with 'coded engagement'? Is there a fundamental difference between a language that uses, e.g., a morphological engagement inflection and languages that use an epistemic particle, the adverb actually, or in which a speaker points to her forehead to express a similar intention; distinctions that Evans et al. (2018b) call differences in 'semiotic investment'? Are these fundamental differences? Answering such questions is a prerequisite for understanding the scope of the enterprise of examining engagement, and for demonstrating what morphological engagement systems can actually tell us about the nature of grammar.
In the present paper I have two goals in relation to the research programme Evans et al. (2018a,b) set out. First, I want to add to its empirical base by examining data from the Australian Aboriginal language Ungarinyin (Worrorran). After briefly introducing the language in section 2.1, in section 2.2 I will identify three morphemes and constructions in the language that broadly seem to fit with the description of engagement and raise the question how to determine whether these actually are expressions of the category.
In aiming to answer this question, section 3.1 first takes a step back and lists some approaches in the literature that have attempted to deal with the problems the analysis of engagement raises. I conclude that a philosophy of language that has inspired each of these approaches is the philosophy of Dialogism, developed by Mikhail Bakhtin and collaborators. Section 3.2 first demonstrates that the main problems for the analysis of engagement are very similar to the ones debated in literature on tense, aspect, mood/modality, evidentiality (TAME) and then adopts a Dialogic approach. Specifically, I argue that Bakhtin's notion of 'addressivity' allows us to overcome these problems. This discussion relates to the second goal of the present paper: I aim to present a dialogic framework that can motivate and delimit the description of engagement as a grammatical category. Dialogism states that every action, including language action or 'utterance', is inherently addressed in the sense that it responds to an earlier or parallel action and anticipates other actions. These further actions may belong to concrete people who are present in the current speech situation, or are even imagined. But for Bakhtin this means that all human acts and utterances are necessarily dialogic, although, crucially, not all actions and uses of language demonstrate their dialogic nature with the same degree of explicitness. Some types of actions or language display their addressivity, i.e. the degree to which they are addressed to participants and acts, whereas others do not. I propose that this conceptualisation of language allows us to both acknowledge that addressivity is a pervasive feature of language, but also that it is variably relevant for describing grammar. I argue that grammatical structures that express engagement are types of grammar that necessarily display some degree of addressivity. With this definition in hand, section 3.3 aims to answer the question whether the Ungarinyin morphemes/constructions introduced in section 2 indeed are expressions of engagement, and why. In doing so, the section develops a descriptive framework rooted in Dialogism that, I argue, can be used to explore semantic distinctions in engagment more widely. Section 4 wraps up the paper with a brief conclusion.

Background
Ungarinyin is a non-Pama-Nyungan language of the Kimberley region of North-Western Australia of the Worrorran family (McGregor and Rumsey, 2009). The language is headmarking, mostly suffixing and most verbal constructions consist of a minimally inflecting preverbal particle followed by an inflecting verb (i.e. 'preverb'/'coverb' constructions). Nominal subjects and objects are unmarked, but the language contains several optional semantic cases (e.g. genitive, dative, commitative). There is one morphological subordinating strategy, one multifunctional reported speech construction (Rumsey, 1990;Spronck, 2015) in addition to a few conjunctive strategies, but simplex clauses are preferred in discourse. For a fuller description of Ungarinyin grammar the reader is referred to Rumsey (1982) and (Spronck, 2015).

The Ungarinyin definite subject marker and the overlapping functions in engagement
Ungarinyin has three clear candidates for morphemes/construction types that could qualify as expressions of engagement. The first of these is illustrated in (3), more particularly in (3-b), underlined in the pronominal prefix on wurrumarn 'she takes it'. Following Rumsey (1982, 80), and for reasons to be illustrated below I will refer to this marker as the 'definite subject marker'. Ungarinyin verbal affixes often show fusion, resulting in the complex prefix in (3-b), which consists of an object and subject pronoun, and the definite subject marker. (3) The story describes the behaviour between a mother-in-law and a son-in-law. They were not allowed to communicate or to have close contact under normal conditions, but in times of war the mother-in-law would hit the son-in-law with branches, giving him strength while preparing for battle. Since it is unspecified for gender, the third person subject of wurrumarn 'she takes it' in (3-b) is potentially coreferential with either one of the two protagonists of the story, the mother-in-law ('she') or the son-in-law ('he'). In this instance, the definite subject marker helps to disambiguate: it indicates that the subject pronoun in (3-b) is coreferential with the subject of the immediately preceding clause, which identifies the mother-inlaw as the subject of wurrumarn 'she takes it'. This function, signalling coreferentiality of the grammatical subject, motivates the gloss 'definite subject marker ' Rumsey (1982) gives to a form that in an earlier description Coate and Oates (1970) simply describe as the 'long form'.
The function of the definite subject marker is reminiscent of what Stirling (1993, 17-18) calls 'recapitulation clauses' in languages that morphologically mark 'same subject' versus 'different subject': 'In these cases the first clause in a new sentence is a recapitulation of the final clause in the previous sentence, and is marked for switch-reference in such a way as to connect the final clause of the previous sentence to the first full clause in the new sentence' (Stirling, 1993, 17) Like Stirling's (1993) recapitulation clauses, the definite subject marker (nearly) always re-introduces a subject from the immediately preceding clause and frequently combines with verbs that describe a similar or identical event. But there are important differences as well. In Ungarinyin there are no syntactic or paradigmatic constraints on the use of the underlined marker in (3-b). Significantly, absence of the definite subject marker in Ungarinyin does not indicate 'different subject'.¹ In addition, whereas recapitulation clauses could be described as a discourse or stylistic process to increase cohesion, the observation that the definite subject marker is used especially in referentially ambiguous sequences as in (3), signals that it takes the knowledge state of the addressee into account.
This brings the Ungariniyn definite subject marker into the realm of engagement: it signals the epistemic status of the subject referent as 'known'/'accessible' to both the speaker and the addressee. The interpretation of the marker in questions is also particularly relevant in this respect, compare (4).
(4) wujika wujika Q irroden a 1 -irra 2 -ode-n 3msg-DEFS-be.painted-PRS di? di n w .ANAPH 'Is it (really) his painting?' (Coate, 1970, line 57) Example (4) constitutes a turn in a dialogue about the location of a rock painting of a mythological ancestor, and the speaker seeks to confirm whether the discussed location is (indeed) the site where this ancestor is painted. The example further supports the analysis that the morpheme -irra 2 -primarily relates to epistemic access: posing a question about a referent marked with the definite subject marker signals 'do you share my judgment that we both have equal epistemic access to the described referent?'.
On the basis of these observations, the function of Ungarinyin definite subject marker in (3-b) appears close to that of the Andoke example (1-a); it signals a form of symmetric engagement.
The definite subject marker complements a number of other strategies for expressing definite reference in Ungarinyin, that show the degree of explicitness the language allows with respect to qualifying the accessibility status of referential terms, illustrated in (5). The examples in (5) each contain nominal constructions involving one or more demonstrative pronouns, with the exception of (5-e), which shows a bare noun (demonstrating that Ungarinyin does not require a specifier slot to be filled). The broad range of adnominal constructions in the language is motivated by the discourse properties each demonstrative is associated with, as summarised in Table 1.² Table 1: Discourse properties of demonstrative pronouns in nominal constructions (the labels 'anaphoric', 'ambiphoric' and 'demonstrative' are adopted from Rumsey, 1982) demonstrative type identifiable specific The two parameters 'identifiability' and 'specificity' are not mutually exclusive. With a nominal construction marked as identifiable, the main function of anaphoric pronouns, the speaker indicates to the addressee that s/he expects him/her to be able to recover the identity of the described referent, based on prior knowledge or discourse context. The opposite holds for nominals marked as not identifiable, with a ambiphoric pronoun, which signals that the referents are not presupposed to be commonly known. Specificity is a separate semantic feature that places a referent in space and time.
A construction as ari jirri 'the/some man' in (5-a) therefore signals a referent 'man' whose identity is supposed to be known to the addressee. In (5-b), the ambiphoric nominal construction signals that the referent andu orroli 'this dingo' is not supposed to represent a known entity, but in narratives these constructions are also frequently used to introduce a new scene in the story. The nominal construction jinda kundi 'this husband' signals that it refers to a specific entity, a man which in this particular case can quite literally be pointed at, since the speaker describes a picture from an elicitation task.
Each of the types in Table 1 can be combined in nominal constructions, which is perhaps most surprising for anaphoric and ambiphoric pronouns, as illustrated in (5-d). Although this combination may seem contradictory, its function is often to re-introduce entities into a narrative: here, the speaker signals that the referent yila 'child' in (5-d) may be identifiable to the addressee (consistent with the meaning of the anaphoric pronoun), but that it is perhaps not at the top of the mind of the addressee (consistent with the meaning of the ambiphoric pronoun).
Referential identifiability marking as illustrated in (5) shares an important function with the Ungarinyin definite subject marker: both indicate that the described referent can be identified as the same entity by both the speaker and the addressee.
One difference with the examples of engagement in (1-2) on the one hand and the Ungarinyin definite subject marker in (3-4) and complex referring expressions in (5) on the other, is their scope. Even though the speaker proximate marker in (2) seems to have originated as a demonstative pronoun, it has wide scope in (2) over the entire described event. In (1), even though the engagement marking follows the nominal element duiʌhʌ 'whites', the point of the sentence seems to be that the whites are coming (i.e. an event interpretation), not that they exist. The Ungarinyin markers can only have scope over a (pro)nominal referent: despite appearing as a bound marker on the verb the definite subject marker only has scope over the subject referent in the pronominal prefix, not the entire verb. Within the approach to engagement Evans et al. (2018a,b) this is not necessarily significant because, as they point out, definiteness inherently appeals to knowledge status (Evans et al., 2018a, 110, 117, 122). But it is a difference nonetheless.
A third and final strategy in Ungarinyin that can be related to engagement is the 'paragraph case'. Even though this marker, as the name suggests, originates as a case form, it typically has wide scope over the entire proposition. The paragraph case is a special use of a case form Rumsey (1982) and subsequent analyses gloss as 'lative', which carries a spatial interpretation illustrated in the case stacking construction in (6), in which the lative case combines with a locative case to express an inessive meaning. The non-spatial function of this marker, however, is much more common in contemporary Ungarinyin, as illustrated in (7), cited from Rumsey (1987 amerera a 1 -mara-ra west 'so then he asked me "when are you travelling westward?" (Rumsey, 1987, 607-608, glosses added) The lative case occurs three times in (7), twice in the construction yej-ju 'through laughing' in (7-c) and (7-d), but the relevant usage occurs in (7-f), where the marker indicates a new scene in the narrative, or an act that represents a new stage in the story. Rumsey (1987) comments on this use: 'The lative in this textual function is partially complementary to [...] the 'definite subject' marker which is exemplified in [line 4 of example (7)] above, which can be used for reverting to an old discourse topic, as opposed to introducing a new one' (Rumsey, 1987, 608) I believe that the analysis of partial complementarity between the paragraph case and the definite subject marker is exactly right: with respect to the engagement meaning the two are in direct opposition, signalling a knowledge asymmetry ('I estimate that you do not yet know about event p (of which I am about to tell you')) and a knowledge symmetry ('I estimate that you (also) know about referent x already'), respectively. However, the scope properties of both markers also mean that they apply to rather different domains: despite being affixed to the verb, the definite subject marker indicates the assumed knowledge status of a referent (the term 'topic' in Rumsey's description above is ambiguous in this sense), whereas the paragraph 'case' applies to events the speaker introduces into the narrative that are yet unknown to the addressee.
The image of Ungarinyin engagement painted in this section is diffuse. On the one hand the language has three clearly identifiable strategies dealing with the expression of knowledge status and intersubjective coordination of who-knows-what. On the other hand, these expression types are diverse and do not line up within a convenient system of oppositions. Can we qualify these three strategies as expressions of one single grammatical category of engagement? Is there a distinction between the free pronominal expressions and the two morphological forms, i.e. the definite subject marker and the paragraph case? These analytical questions will form the focus of the second part of this paper.

Turning to Dialogism
What the existence of the grammatical category of engagement shows is that the intersubjective relation in which grammar is inherently embedded can be a central aspect of the conventional meaning of a category. One of the most fundamental and consistent philosophies of language to make the argument that a full un-derstanding of grammar depends on the relation between the speaker and (at least) one addressee is that of Dialogism, an approach that arose in the 20th century out of the work of Mikhail Bakhtin and collaborators (Bakhtin, 1981(Bakhtin, , 1986bVološinov, 1973;Erdinast-Vulcan and Sandler, 2015). As Sandler (2013) phrases it: '[t]he greatness of the Bakhtin Circle's linguistic works lies in the fact that they took nothing -not even school grammar-for granted. Instead, these works used Bakhtin's philosophical anthropology to make a fresh start, to seek new ways of classifying linguistic phenomena and of defining linguistic units: the utterance, delimited by the change of speaking subject [ (Bakhtin, 1986a, 71ff)], the inner dialogic division of the utterance following the "faint traces of changes of speech subjects that have furrowed the utterance from within" [ (Bakhtin, 1986a, 99)], and the different voices heard even in the very same stretch of discourse' (Sandler, 2013, 161).
The fact that Dialogism, as Sandler puts it, 'took nothing for granted' when it comes to grammar meant that it has long been disqualified in linguistics as postmodern and anti-linguistic, and Bakhtin's work is certainly not easily reconciled with many dominant views on language (cf. Stewart, 1983;Spronck, 2006Spronck, , 2019. But the three central ideas of Dialogism summarised in this quote demonstrate that it is highly relevant for understanding engagement. The first idea is understanding language, or 'the utterance', as a social construct: 'the speech act or, more accurately, its product-the utterance, cannot under any circumstances be considered an individual phenomenon in the precise meaning of the word and cannot be explained in terms of the individual psychological or psychophysiological conditions of the speaker. The utterance is a social phenomenon' (Vološinov, 1973, 82)⁴ While contributions in discourse alternate between speakers, understanding these contributions cannot be done by focusing on either of the two individuals within Dialogism.
The second, perhaps somewhat more cryptic point from Sandler's (2013) quote above, regarding the 'faint traces of changes of speech subjects' found in utterances touches even more directly upon engagement: what Bakhtin claims here is that the speaker herself is as much affected by the ensuing discourse as she affects it and actively tracks and anticipates how the addressee is affected. This suggests that a turn in language is not conceived and produced as a ready-made in isolation, but fundamentally shaped by a variety of subjective and intersubjective factors.
This view naturally leads up to the third point, the idea that any utterance combines multiple 'different voices'. These are not literally 'voices' in the sense that they are actual vocations, but reflections of what the addressee or anyone else present or not present in the speech situation might say that would affect the form of the utterance. Engagement is an example of multi-voicedness in Bakhtin's sense in that an engagement marker signals a property (assumed knowledge status) of both the speaker and the addressee. Therefore, an engagement expression is inherently Dialogically construed.
Dialogism provides a view of language in which many interactional aspects of language that are relegated to the fringes of most models of mainstream linguistics can take centre stage, and one in which engagement is not an anomaly, but a predictable consequence of the nature of language. But for Bakhtin (1986a), the utterance, i.e. language as part of its social context, is the only level at which these aspects of language can be studied: 'When one analyzes an individual sentence apart from its context, the traces of addressivity and the influence of the anticipated response, dialogical echoes from others' preceding utterances, faint traces of changes of speech subjects that have furrowed the utterance from within -all these are lost, erased, because they are all foreign to the sentence as a unit of language' (Bakhtin, 1986a, 99) Interestingly, this is exactly where the discovery of engagement as a grammatical category places an important footnote in the Bakhtinian programme: morphological markers such as in (1) and (2) demonstrate that with respect to knowledge status, in these languages 'the traces of addressivity and the influence of the anticipated response' can be studied at the grammatical level of the sentence.
Earlier attempts to implement Dialogism in grammatical analysis include francophone polyphonic linguistics, such as represented by Anscombre and Ducrot (1976) and Nølke et al. (2004), whose analysis often focuses on discourse particles (cf. Evans et al., 2018b, 165). But some Bakhtinian ideas seep through in Jakobson (1957), the first study to cite the most linguistically oriented work of the Bakhtin cicle, Vološinov (1973), in the West, and which put the idea of context-dependent grammar on the map in modern linguistics: Jakobson's (1957) famous description of shifters.⁵ The notion of shifters, combined with ideas about information structure as espoused by the Prague School (cf. Daneš, 1966), introduced a concern with intersubjective functions of language out of which emerged various flavours of functional grammar (cf. Halliday and Matthiessen, 2004;Hengeveld and Mackenzie, 2008;Van Valin, 2009).⁶ I would suggest that with each of these approaches productive dialogues about the nature of engagement are to be had. But in this section I would like to formulate a more direct proposal for how Dialogism can provide an analytical framework for defining engagement and apply it to answer the question raised in the previous section: are (some of) these three strategies examples of engagement or not?
Before addressing this question, in section 3.2 I take a step back and examine two problems that come up in the description of engagement, but which are in fact raised in many (other) parts of the literature on tense, aspect, modal and evidential categories (TAME). For reasons to become clear, I label these the 'form-function mismatch' and 'speaker reference' problems. After demonstrating that Bakhtinian addressivity can be used to defuse both of these problems, section 3.3 seeks to answer the question about the grammatical status of Ungarinyin engagement by relating the three described expression types to a Dialogically embedded view of the category. I propose that adopting a Bakhtinian view of engagement allows us to define and distinguish grammatical expressions of engagement in a principled way.

Engagement in light of the TAME-debates
The acronym TAME labels several linguistic categories that are often characterised as complex and confusing. This already begins with the acronym itself. The first two letters rather consistently denote Tense and Aspect in the literature, but the M can stand for either 'mood' or 'modality', and the E (only added over the past two decades) for the grammatical category of 'evidentiality' or 'epistemicity' in a broader sense. None of the categories involved are unambiguously defined. It has proven remarkably difficult to indicate consistently where, e.g., aspect ends and tense begins (Dahl, 1985), and even more so for the categories of modality and evidentiality. Despite fiercely argued positions that 'evidentiality' exclusively refers to a grammatical (morphological) category distinct from epistemic modality (Aikhenvald, 2004), or is a broader semantic category (Mushin, 2001), both interpretations remain in the literature.⁷ Epistemic modality has been argued not to constitute a separate category from evidentiality both because evidential meanings can derive from modal ones (Nuyts, 2001) or modal meanings from evidentials (Matthewson, 2011). And even if we can settle on discrete definitions for each of the categories involved, they customarily show deep cross-connections, such as 5 According to the translator's note Jakobson was also instrumental in having Vološinov (1973) translated, and recommended it highly to colleagues (Vološinov, 1973, vii). 6 In separate developments, cognitive linguistics has taken to debate the (often apparently loose) relation between intersubjectivity in interaction and grammatical form in the recent literature on viewpoint (Dancygier and Sweetser, 2012;Dancygier et al., 2016) and 'intersubjectification' (cf. Traugott, 1989;Davidse et al., 2010;Breban and Davidse, 2016). In logic and formal semantics Hintikka (1962) initiated a prolific debate about the connection between sentence type and knowledge assignment and 'much of the later interest in formal epistemic models derives from a concern with situations in which there are multiple knowers who may know or be ignorant about the knowledge and ignorance of the others' (Stalnaker, 2006, 174). 7 With the stricter definition along the lines of Aikhenvald (2004) most commonly accepted in typology and grammatical description, and the second interpretation more current in conversation analysis and Interactional Linguistics (Couper-Kuhlen and Selting, 2018). tense/aspect with modal interpretations (Abraham and Leiss, 2008;Brosig, 2015;Hacquard, 2009;Trnavac, 2006b), or evidential ones (Lauwers and Duée, 2011).⁸ The question of the semantic boundaries of TAME-categories is compounded by questions about what formally counts as a representative of the category. While it was never particularly controversial that tense, aspect and mood could be expressed through multi-word expressions in some languages and morphemes in others, the debate changed with respect to evidentiality following Aikhenvald (2004). Aikhenvald's emphasis on morphological expression and obligatoriness made the position that periphrastic constructions could equally express the category (cf. Diewald and Smirnova, 2010) a contested view.
Finally, a problem that often comes up in functionalist accounts of modality, in which the category is associated with attitudinal meanings, is whose attitude is exactly expressed. The most typical participant whose attitude is signalled by a modal (e.g. the entity whose doubt is expressed by using an epistemic modal adverb like 'maybe') is the speaker (Nuyts, 2005;Boye, 2012). But, problematically, not always. For example, consider the respective interpretations of 'must' in (8).
(8) a. You must take off your shoes b. Over the years, many people have written both positively and negatively about the NCFIC. Here are the seven most common mischaracterizations. [...] The NCFIC believes that the whole family must always be together for all gatherings. False. We have never said that the whole family must be together for all gatherings (Gentens, 2016, 106)⁹ All occurrences of 'must' in (8) reflect some wish or strong desire. Or, in more traditional terms used to characterise deontic modality, they indicate that 'p is necessary'. But what differs between these instantiations of 'must' is the understanding of the referential identity of the person(s) holding the desire or according to whom p is necessary. In (8-a) the most likely interpretation is that the speaker holds the expressed desire (or at least it would be highly irregular if the speaker does not share it). However, in (8-b) the 'desirer' is explicitly not the speaker/writer, this role is attributed to the 'many people' introduced in the first sentence of the example (and those who share their expectation). So the speaker is to some extent a relevant value for understanding the semantics of (most) modals, but it seems that there is no reliable referential relation to the speaker in modals across all sentence types and situations (for more examples and discussion, see Gentens et al., 2019).
In summary, the TAME-debates revolve around the two general problems in (9).
(9) a. Form-function problem: TAME-categories show imperfect form-function mapping. This results in two issues: (I) difficulty delimiting morphological categories, since meanings expressed through inflection in one language appear to be marked with particles and multi-word constructions in other languages (or even within the same language), and (II) difficulty distinguishing TAMEcategories from each other, both on semantic grounds and based on their distribution, since meanings tend to blend into each other and the markers involved are often portmanteau-morphemes; b. Problem of speaker reference: the referential value of perspective takers is not strictly fixed (the speaker is typically the main perspective taker, but not always).
The problems in (9) are exactly the same as the ones Evans et al. (2018a,b) diagnose for engagement. They demonstrate that not only it is difficult to define engagement as a discrete (morphological) category, its 8 For further studies on connections between evidentiality and modality see de Haan (1999); Cornillie (2009), among many others; between aspect and evidentiality, see Lau and Rooryck (2017); Lauwers and Duée (2011); Sadanobu and Malchukov (2011); between valency structure and aspect, Spreng (2012); Tenny (1987), or valency structure and evidentiality Ørsnes (2011); between aspect and modality, see Brosig (2015); Gentens (2016); Nara (1999);Trnavac (2006a), between aspect, tense, mood, evidentiality on the one hand and discourse structure on the other, see Fedder and Wagner (2015); Fetzer and Oishi (2014); Hartzler (1983);Hopper (1979); Merlan (1981). 9 For more detailed discussion of this example, see Gentens et al. (2019, 7) association with articles and cohesive devices shows that its boundaries stretch even further than the range of structures commonly associated with TAME-categories. Similarly, the knowledge asymmetries reflected in engagement inevitably need to be interpreted with respect to relevant knowledge holders, typically the speaker and the addressee. This intersubjective dimension proves a rather confusing aspect of the semantics of engagement markers, since it is not necessarily clear how a (candidate) engagement marker establishes reference to either a speaker or an addressee. If the speaker and addressee are central to the semantics of engagement, should we be able to identify them as stable semantic values in reflexes of the category of engagement? Apart from for examples of engagement marking deriving from personal pronouns (e.g. Schultze-Berndt, 2017) no clear pathway towards connecting speaker-addressee reference to engagement has so far been shown.
I would suggest that Bakhtinian addressivity presents such a pathway. In the Dialogic view, address is not simply a reference to an addressee, but an act that gives intersubjective meaning to elements that may or may not carry addressivity as part of their conventional semantics. Compare Bakhtin's (1986a) description of address terms: 'Language as a system has an immense supply of purely linguistic means for expressing formal address: lexical, morphological (the corresponding cases, pronouns, personal forms of verbs), and syntactical (various standard phrases and modifications of sentences). But they acquire addressivity only in the whole of a concrete utterance. And the expression of this actual addressivity is never exhausted, of course, by these special language (grammatical) means. They can even be completely lacking, and the utterance can still reflect very clearly the influence of the addressee and his anticipated responsive reaction' (Bakhtin, 1986a, 99) The distinction Bakhtin (1986a) introduces here is important: yes, a pronoun like formal 'vous' and an imperative like 'eat up' suggest something about the relation between the speaker and the addressee as part of their conventional, grammatical meaning, but the exact nature of this relation is only established in the act of uttering these words. Traces of the intersubjective relation, however, are not limited to grammatical meaning. When talking to some addressees using the term 'tourbillon' is more effective than 'that little round thingy in my watch', and vice versa, which is a reflection of assumptions the speaker makes about the addressee.
Within this view, two observations can explain the TAME/engagment problems summarised in (9): (I) The effects of addressivity do not line up neatly with common distinctions such as morpheme, construction or word or even divisions such as morphosyntax, semantics and pragmatics. And (II): A speaker only becomes a speaker when addressing an addressee, and the addressee is an addressee by virtue of being addressed by a speaker. Within this mutual intersubjective relation, which forms the bedrock of the Bakhtinian utterance, only certain meanings become relevant (also cf. Sperber and Wilson, 1986;Dor, 2015). A linguistic element E (the words 'must', 'tourbillon' or examples of engagement such as 1-b) may or may not have a sufficiently interpretable meaning without being addressed (i.e. in the context of an utterance). But this does not mean that such elements cannot be described in terms of conventional semantics; it simply means that the meaning of instances of the respective element are inherently variable. That this becomes more striking when element E is a morpheme rather than a periphrastic construction or an ad-hoc expression (i.e. the form-function problem) is simply because in language description a morpheme requires a conventional semantic description, whereas looser structures do not. In this sense, both the form-function and speaker problems in (9) are analytical problems, rather than a more general assessment of intersubjective meaning in language. We may illustrate this situation as in Table 2. (1), (2), 'must'; all shifters 'tourbillon'; anything In order to fully appreciate the analysis sketched above it may be useful to compare the Dialogic view with the description of shifters/demonstratives and their relation to knowledge asymmetries Hanks (1990) proposes: 'The individuation of objects in conversation is an interactive achievement possible only within a socially constructed world in which coparticipants already share a frame of reference and a sense of typicality. At the same time, [...] coparticipants in talk are virtually always asymmetric in some senses. [...] [S]hifters tend to be especially sensitive to such asymmetries. For our purposes, what is most significant is the fact that social asymmetries among coparticipants in talk constrain the kinds of deictic categories they can use in making reference' (Hanks, 1990, 21) In Hanks's (1990) approach deictics have a meaning potential that is constrained by intersubjectivity. If I talk about 'this pen' and you can see it, the proximal demonstrative highlights common knowledge; if I use the same expression but hide the pen, 'this' marks an object only I have direct access to. If we therefore ask the question 'what kind of intersubjective meanings can element E have in context?' (a question we only ask for shifters because it does not occur to us with the word 'tourbillon', not because it is more relevant) the answers are bound to be contradictory. The Bakhtinian approach reverses this analysis: it accepts that a description of grammatical, conventional meaning inherently falls short of capturing the intersubjective meanings an element can have, but acknowledges that addressivity can also shape the conventional meaning of certain elements, particularly those we call shifters. As the discovery of engagement as a morphological category shows, such elements are more common at the level of grammar (Bakhtin's 'sentence level') than Bakhtin (1986a) or Vološinov (1973) knew. But it would not have surprised them.
Based on these considerations the definition of engagement I would like to propose is the one in (10): (10) An engagement marker is a grammatical element that even without being embedded in an utterance can only be described as being addressed towards a knowing (or unknowing) addressee What the definition in (10) adds to the exposition above is that the addressee of an engagement expression is specifically addressed in the role of a knowing (or unknowing) entity.¹⁰ This opens up possibilities for developing a broader approach to TAME categories as well, in which the type of address is central to their definition (doing so here would lead too far away from the main point; but see Spronck, ms.). But most importantly, the approach avoids the rabbit hole Evans et al. (2018b) appear to dive into (along with much of the TAME-literature): once we start identifying intersubjectivity in language, the only conclusion that we can ultimately arrive at is that everything is intersubjective to some extent. Rather than asking what aspects of addressivity are possible for an engagement marker (an inherently open question in Dialogic terms and therefore unanswerable), the definition in (10) aims to specify what aspects are necessary for the description.
Using this definition, in section 3.3 I aim to answer the question raised in section 2: whether the Ungarinyin expressions are examples of engagement or not.

Engagement in Ungarinyin: reigning in the category
With the Bakhtinian distinction between sentence level and utterance level in hand, the first conclusion we can draw is that each of the three Ungarinyin expression types described in section 2.2 are examples of engagement at the level of the utterance: all indicate a knowledge status within the intersubjective relation between the speaker and the addressee. If we seek to define engagement, however, this cannot be our final analysis. Returning to the example of 'tourbillon' one last time, even using a specialist term suggests something about the assumed knowledge status of the addressee (as does the term bolijman 'policeman' in (7-a) rather than the non-borrowed lexeme yirrkalngarri 'policeman' (< yirrkal 'to tie up') or addressing someone in English rather than French or Ungarinyin). Examining engagement at the level of the utterance cannot inform us about grammar.
For convenience, (11) lists the three expression types discussed in section 2.2: (11) a. Definite subject marker (e.g. 3-b) b. Demonstratives and complex demonstratives (e.g. 5) c. Paragraph marker (e.g. 7) For each of these expression types we may now ask, following the proposed definition of engagement in (10), whether addressivity with respect to a knowledge role is a necessary semantic component of each of the types in (11). In all instances I would propose that it is.
The definite subject marker requires shared knowledge about a referent, specifically, the referent indexed by the grammatical subject of the inflected verb. The minimal assumption required about the semantics of the definite subject marker is that the referent is known, which implies it is known to someone. We may suggest that this implies that the referent is known to the addressee, since it can be assumed that a referent is necessarily known to the speaker. But the Dialogic approach reminds us that focussing on the addressee in our analysis is as restrictive as focussing on a speaker: neither exists without the other. Since the definite subject marker in Ungarinyin does not contrast with, e.g., a marker that indicates a knowledge asymmetry, viz. indicates that the referent is unknown to one participant, there is no grammaticalised opposition between knowledge holders.¹¹ The grammatical meaning of the definite subject marker only requires a semantic variable 'known to x', where x most likely becomes associated with the speaker and the addressee in the act of utterance. Consequently, in the absence of a coded knowledge asymmetry, the marker is interpreted as symmetric engagement.
There are two differences between the Ungarinyin definite subject marker and the Andoke/Abui examples introduced in section 1: the first difference, as pointed out in section 2.2 relates to the scope properties in the three languages. In (1) and (2) the engagement marker has scope over the described event, in Ungarinyin the definite subject marker can only apply to a referential entity. A second important distinction is that whereas in (1) and (2) the engagement markers signal an assessment of epistemic access both based on linguistic and extra-linguistic cues (background knowledge, shared attention), the Ungarinyin definite subject marker specifically signals epistemic access on the basis of the the immediate discourse context.
This description results in four variables that can capture Ungarinyin engagement at the grammatical level, as in (12) (12) the knowledge holder is indicated as 'symmetric epistemic authority', while the referential identity of the knowledge holder is unspecified and left to be interpreted at the level of the utterance. For the semantic qualification of the definite subject marker, I suggest, it is only necessary that the knowledge holder variable be filled, not that it specifically indexes, e.g., the speaker and the addressee.¹² 11 Note that in a language that distinguishes asymmetric engagement an opposition between (at least) two knowledge holders does need to be made at the grammatical level, although it needs to established for each specific language if these two grammaticalised entities are necessarily the speaker and addressee or more general knowledge holders that can only be fully interpreted at the level of the utterance. 12 I propose that whether the knowledge holder does or does not index a specific referent for a given marker requires a languagedependent descriptive judgement: there may well be languages in which, e.g., the knowledge holder may index two and only two With respect to the demonstratives in Ungarinyin the descriptive situation is slightly more diverse. In the analysis presented in section 2.2, bare nouns (e.g. 5-e) are not marked with respect to knowledge status in any way (i.e. absence of a demonstrative pronoun does not signal indefiniteness or non-specificity in the language). For this reason, this construction type does not require either a knowledge holder, an indication of knowledge status or a knowledge ground, and on this basis it is not an example of engagament marking. The claim that adnominal proximal/distal demonstratives (e.g. 5-c) in Ungarinyin relate to specificity¹³ means that these constructions fall outside the realm of engagement as well. Specificity is a qualification of how some entity exists in space and time: it could be encountered or experienced and is therefore knowable, but a qualification of specificity is not an assessment that the referent under scope is known in itself.Under this interpretation, the expressions in (5-c) and (5-e) do not represent engagement. ¹⁴ However, the two demonstratives that relate to identifiability, i.e. the anaphoric (5-a) and ambiphoric pronouns (5-b) cannot be described without indicating a knowledge holder, and an indication of knowledge status and a knowledge ground. I would suggest the following analyses for each of these construction types: Ungarinyin anaphoric demonstrative (e.g. 5-a) Scope: referential entity For each of these demonstratives/demonstrative combinations the scope applies to a referential entity. I would suggest that as with the definite subject marker it is necessary to assume that there is a knowledge holder, but not to specify its identity further for (13) and (15). For these constructions the knowledge holder will emerge as an intersubjective meaning in utterances, most likely applying to the speaker and the addressee. For the ambiphoric marker in (14) I will assume that there is an asymmetry since presenting a refreferents, or even, only the speaker and the addressee. My claim is that at least for the Ungarinyin definite subject marker this is not a necessary assumption about its semantic profile. 13 For fuller argumentation for this analysis, see Spronck (2015). 14 A fundamental difference between specificity and identifiablity, for example, is that under this interpretation it is nonsensical to state that a referent is specific to someone. Rather, it places a referent in the actual, knowable world (as opposed to an imagined, unknowable world). Once it has been established that access to a referent depends on a real vs. imagined world this then implies different types of access to the imagined referents, but establishing this distinction (real/imagined) does not relate to epistemic access. Obviously the communicative decision to present the referent as specific or not ultimately sits with the speaker, but this decision is also not a requirement for the semantic characterisation of this construction: it is a type of (inter)subjectivity that comes in at the level of the utterance. erent as unknown necessarily creates an opposition between an entity who does not know about a referent and an entity (presumably the speaker) who does.¹⁵ The complex demonstrative construction (15) further specifies a knowledge ground in the discourse context, serving to re-introduce a referent. For Ungarinyin anaphoric demonstratives the knowledge ground is unspecified, since it may either be based on previous discourse, exophoric reference or world knowledge (in this sense the label 'anaphoric' is slightly misleading). For the 'unknown' ambiphoric marker the knowledge ground is equally unspecified (but knowledge of the referent can be asymmetric on the basis of the same wide range of epistemic grounds).
Free demonstrative pronouns are flexible in their placement and combinatorics in Ungarinyin: appearing before or after a noun and in different pronominal stacking constructions can result in subtle meaning distinctions in the language (Spronck, 2015). Therefore, the expression of engagement through demonstratives in Ungarinyin is less restrictive, and more flexible than the morphological expression of engagement through the definite subject marker. For most 'known' engagement expressions with scope over referential entities the knowledge ground is further restricted to the discourse context, rather than other reasons for which referents can be part of shared knowledge. In this sense the Ungarinyin system is narrower than engagement systems in which knowledge can be shared on other grounds, as in (1) and (2).
Finally, the paragraph case, as in (7-f) is the only type of engagement marking in Ungarinyin that has scope over a full described event or proposition, as demonstrated in section 2.2. With respect to its other semantic properties I would suggest that it parallels the ambiphoric marker.
(16) Ungarinyin paragraph case Scope: described event / proposition Knowledge holder: asymmetric epistemic authority (realised at the level of the utterance) Knowledge status: asymmetrically known Knowledge ground: unspecified The descriptions of the Ungarinyin engagement markers in this section have implemented the Dialogic definition of engagement proposed in section 3.2. The descriptive format shown in (12) -(16) was further able to specify differences in engagement both within Ungarinyin and in a cross-linguistic perspective and demonstrate why the Ungarinyin expressions introduced in section 2.2 could be classified as examples of engagement. I therefore hope that in addition to presenting an argument about engagement in Ungarinyin the approach taken here could be expanded for exploring engagement as a grammatical category both for descriptive purposes and in a comparative perspective.

Conclusion
As Sapir writes in the quotation at the beginning of this paper, human beings and language behaviour are inherently social. The fact that some aspects of language appear more social than others is not because some parts of language are governed by social 'laws' and other parts are not; it is because many aspects of language viewed in isolation can be described in non-social terms. This is not an assessment of the sociality of language, it is an assessment of our descriptive practices.
15 An alternative analysis would be that the ambiphoric marker does not specify a knowledge holder and simply carries a knowledge status as 'unknown' that is realised in the utterance as a knowledge asymmetry, because speakers cannot talk about referents they do not have in mind/do not know. Under this analysis the semantic representation of ambiphoric pronouns does not require a specified knowledge holder. I believe that the possibility of raising these questions on the basis of the format presented for the semantic analysis of engagement in (12) -(15) above illustrates its potential for further examining engagement along these lines, but will leave resolving this issue to further research.
The observation that engagement markers cannot be described without recourse to the sociality of the speech situation demonstrates that social meaning can shape the conventional semantics of grammar. But, as Evans et al. (2018a,b) point out, once we accept this fact, restricting the morphosyntactic boundaries of social meaning soon becomes problematic. Bringing forward new data relevant for the study of engagement and applying methodologies sensitive to interaction (such as Conversation Analysis and Interactional Linguistics; Couper-Kuhlen and Selting, 2018) are two ways of exploring these morphosyntactic boundaries. But as much as this empirical effort matters, recognising the boundaries of engagement also depends on questioning what we choose to describe.
With respect to this theoretical aspect, the further development of a Dialogic linguistics presents a pathway towards integrating social meaning in grammatical description. While much work in this area remains to be done, here it is possible to find partnerships within linguistics with schools that have already implemented parts of the Bakhtinian programme, as suggested in section 3.1. By combining these descriptive and theoretical approaches and adopting ideas from Dialogism that once seemed beyond the realm of linguistics, the exploration of engagement paves the way towards an integrated characterisation of sociality in grammar.