Intradisciplinarity: can one theory do it all?

James Martin^*

• The University of Sydney, Darlington, Australia

In recent papers (e.g., 2020a, b, 2021, 2022a, b) Bateman explores his vision for multimodality as an empirical discipline. In doing so he draws on sociological studies of knowledge structure, including work by Bernstein (2000) and Maton (2011, 2014, 2016, Maton & Chen 2016, Maton & Howard 2016, Maton et al. 2016). As part of this projection he warns against falling foul of "various flavours and variations of Saussure's well-known proposal of language (or any other system) as a 'master template' for semiotics as such" (Bateman 2022: 47) and what he calls 'linguistic imperialism' (Bateman 2022b: 63). In addition he notes that 'predatory' interdisciplinarity "will be rejected from the start" (Bateman 2021: 308).Read in tandem with Kress's many declarations of a new age of meaning making called 'Multimodality' (e.g., Kress 2003Kress , 2010Kress , 2015)), superseding language and the discipline of linguistics, serious questions have to be raised about work on multimodality informed by a theory of language such as Systemic Functional Linguistics (SFL) -work evolving into something we might call Systemic Functional Semiotics (SFS) via publications such as Kress & van Leeuwen's Reading Images (1990 and Accordingly in this paper I will draw on the sociological studies referred to above to explore the nature of SFL and SFS as knowledge structures, compare them with the model of empirical multimodality envisioned by Bateman and make some suggestions about how his ambitions for the field might be most effectively accommodated. I write as an SFL linguist (discourse analyst in particular), who has been drawn into work on multimodality by research students and colleagues over the past two and a half decades. As such, given the misgivings about the contribution of linguistics noted above, I should perhaps request readers' indulgence -as I suggest that an SFL/SFS perspective need not be read as the foul and predatory one that some of the more logophobic multimodalists apparently fear.In discussions of this kind it is important to distinguish multimodality as a field of research and multimodality as its object of study. Multimodalists (like psychologists) unfortunately tend to use the same term for both phenomena (cf., language and linguistics for linguists). Where confusion might arise I will refer to the field of research as Multimodal Studies below. By way of framing the discussion let's begin with Bernstein's (1996: 23) distinction between singulars and regions. For Bernstein a singular is "a discourse which has appropriated a space to give itself a unique name", for example, "physics, chemistry, sociology, psychology" and which "created the field of the production of knowledge". These he contrasts with regions, "a recontextualising of singulars", for example, "medicine, architecture, engineering, information science", noting that "any regionalisation of knowledge implies a recontextualising principle: which singulars are to be selected, what knowledge within the singular is to be introduced and related." Importantly he goes on to comment that "regions are the interface between the field of the production of knowledge and any field of practice." Had Bernstein's vision extended into the 21st century, he might well have added multimodality as an emerging region to his list, with media and communications as its field of practice.Seen in these terms SFL is a canonical singular (Martin 2014(Martin , 2016) ) and contrasts with its regionalisation in the Sydney School's well-known genre-based literacy programs (Rose & Martin 2012) -which tend to draw on a range of relevant singulars (including for example Bernstein and Maton's sociology of knowledge, neo/Vygotskyan social psychology and strands of Critical Discourse Analysis). One possible reading of Bateman's vision would entail, via design and/or evolution, the transformation of Multimodal Studies into a singularwith its own distinctive knowledge structure deploying an empirical methodology grounding theory and description.Bernstein's perspective is further elaborated in the distinction he draws between horizontal and vertical discourse (an opposition between what he earlier on referred to as common and uncommon sense). A horizontal discourse involves "a set of strategies which are local, segmentally organised, context specific and dependent, for maximising encounters with persons and habitats....This form has a group of well-known features: it is likely to be oral, local, context dependent and specific, tacit, multi-layered and contradictory across but not within contexts" (Bernstein 2000: 157). A vertical discourse on the other hand "takes the form of a coherent, explicit and systematically principled structure, hierarchically organised as in the sciences, or it takes the form of a series of specialised languages with specialised modes of interrogation and specialised criteria for the production and circulation of texts as in the social sciences and humanities" (Bernstein 2000: 157).In addition two forms of vertical discourse are distinguished -hierarchical knowledge structures vs horizontal ones. A hierarchical knowledge structure is "a coherent, explicit and systematically principled structure, hierarchically organised" which "attempts to create very general propositions and theories, which integrate knowledge at lower levels, and in this way shows underlying uniformities across an expanding range of apparently different phenomena" (Bernstein 1999: 161-162) -e.g., physics, chemistry or biology. A horizontal knowledge structure on the other hand is defined as "a series of specialised languages with specialised modes of interrogation and criteria for the construction and circulation of texts" (Bernstein 1999: 162) -e.g., linguistic theories which position themselves as functional, arguably West Coast Functionalism, Lexical Functional Grammar, Functional Grammar, Discourse Functional Grammar, Role and Reference Grammar or Systemic Functional Linguistics. Bernstein uses a triangle to symbolise hierarchical knowledge structures, since they attempt to create ever more general propositions which account for an expanding range of phenomena (e.g., Newtonian physics, superseded by Einstein's relativity, superseded by string theory and so on). Horizontal knowledge structures on the other hand are visualised by succession of 'Ls' since what counts as development is the introduction of a new perspective, typically by junior speakers who challenge the power and legitimacy of more senior ones (e.g., marxist history, feminist history, post-colonial history and so on). A synoptic overview of these distinctions is offered in Figure 1.Figure 1: Discourse and knowledge structure (after Bernstein 1999Bernstein , 2000) ) As exemplified above, in Bernstein's terms SFL is a canonical member of a horizontal knowledge structure comprising many different theories. Bateman's vision for Multimodal Studies is perhaps a more ambitious one, leaning towards the design and evolution of a hierarchical knowledge structure. This is a trajectory that linguistic theories have embraced, without success, since the modern discipline was founded by Saussure (1916). Wignell (2007a, b) examines the history of social science, focusing on the emergence of economics, political economy and sociology as "a hybrid of the language of the physical sciences and the language of the humanities" (Wignell 2007a: 202) -suggesting that the stronger the boundaries around one of these disciplines, the more it will evolve the characteristics of a hierarchical knowledge structure. In his 2004 conference presentation of Wignell (2007a) he in fact refers to social science knowledge structures as 'warring triangles', since they in general aspire to be recognised as hierarchical knowledge structures (viz linguists' claims for their discipline as the 'science of language'). What happens in practice however is that one or another linguistic theory gains institutional rather than intellectual control of the discipline, for a specific period of time, in a specific place (e.g., Chomskyan linguistics' supremacist control of American linguistics and its intellectual dominions in the 1960s, waning not long thereafter). Seen in these terms, Bateman's vision involves strengthening boundaries around what counts as empirical Multimodal Studies, thereby fostering its development as a hierarchical knowledge structure -occluding more 'weakly bounded' competing triangles as it does so and enjoying globalised longevity. Bernstein (2000: 132-134) probes more deeply into the characteristics of hierarchical and horizontal knowledge structures in his recognition of internal and external languages of description (which he labels L 1 and L 2 respectively). L 1 "refers to the syntax whereby a conceptual language is created" or how constituent concepts of a theory are interrelated; and L 2 "refers to the syntax whereby the internal language can describe something other than itself " (2000: 132) or how a theory's concepts are related to referents. Knowledge structures with a strong internal grammar (L 1 ) have concepts that are tightly interrelated; in hierarchical knowledge structures this facilitates the deployment of a strong external grammar (L 2 ) whereby concepts are related to data in relatively unambiguous ways. Muller (2007) elaborates these ideas, focusing on how knowledge structures progress. He introduces the term 'verticality' to focus on how internal grammars develop -via ever more general propositions accounting for a broader range of data (more verticality) or the addition of new incommensurable languages of description (less verticality). And he introduces the term L 1 L 2 L 3 L 4 …L n'grammaticality' to focus on how knowledge structures manage data -via testable hypotheses about a restricted set of referents (strong grammaticality) or via readings of a less restricted set of referents that are hard to disconfirm (weak grammaticality). An outline sketch of these ideas is presented as Figure 2, including a rough positioning of canonical knowledge structures along a hierarchical/horizontal knowledge structure cline. Seen in these terms Bateman's ambitions for Multimodal Studies involve strengthening internal and external grammars of description so that the field can progress via what Muller (2007: 71) refers to as 'explanatory sophistication' based on 'worldly corroboration'. As far as as grammaticality is concerned Bateman (2021: 302-303) draws attention to Maton & Chen's (2016) discussion and exemplification of mediating languages of description and external ones (termed Lfoot_0.5 and L 2 respectively). Mediating languages are designed to be more general and less data specific than external languages. In SFL, for example, mediating languages comprise what are generally referred to as 'descriptive motifs and generalisations' (Matthiessen 2004) -i.e., general categories such as transitivity, modality or tense (often presented as complementarities such as transitivity/ergativity, modality/assessment or tense/aspect). These help a linguist approach the description of the grammar of a language with relatively 'soft eyes' before locking into a more specific description of the data to hand. What ends up counting as L 1 , L 1.5 and L 2 is itself a process, unfolding over time, as L 1.5 motifs and generalisations are promoted to L 1 status or L 1 concepts are demoted to mediating L 1.5 language status (or perhaps relegated to L 2 external grammar). We focus more specifically on this process when we consider the evolution of SFS from SFL below. SFL itself comprises a number of different languages of description, as reflected in the Routledge and Cambridge handbooks (Bartlett & O'Grady 2017, Thompson et al. 2019). Here we will assume the model developed by Martin and his colleagues (1992Martin and his colleagues ( , 2010Martin and his colleagues ( , 2014)), which is the one that has most strongly influenced Bateman (e.g., 1998Bateman (e.g., , 2008Bateman (e.g., , 2020b) -hereafter Saussure (1916) it treats language as a system of signs. Following Firth (e.g., 1957) it takes the complementarity of paradigmatic and syntagmatic relations as fundamental. And following Halliday it skews this complementarity, privileging system over structure (e.g., 1966, 1992). This axial orientation underpins all language description, resulting in external grammars which formalise valeur in networks of options realised in structure. Over time SFL's internal grammar has expanded to include the notion of hierarchy -i.e., realisation (levels of abstraction), instantiation (a cline of subpotentialisation/generalisation) and individuation (a scale of allocation/affiliation). Of these, realisation has the strongest grammar, as systems in system networks bundle together in relation to the size of the structural unit realising options (rank), the ideational, interpersonal or textual meaning and corresponding types of particulate, prosodic or periodic structure involved (metafunction) and the level of abstraction (phonology/graphology/signology, lexicogrammar and discourse semantics). A synoptic overview of these dimensions (following Martin 2010) is presented in Figure 3 (using an English MOOD system to represent axis). Of these, both instantiation and individuation are underarticulated compared with realisation and constitute major challenges for future research. 2 From an axial perspective, for some languages there is no need to distinguish word and morpheme ranks (since there is no word structure -i.e., no words consisting of more than one morpheme); parentheses make room for this variability at the bottom of the rank scale in Figure 3. To this compilation I will add five elaborations which bear on the discussion. First, form vs substance. As clarified in Martin et al. (2013), the register of SFL at stake here follows Saussure (1916) and Hjelmslev (1961) in treating language as form not substance. This means that phonetics is not treated as a stratum of language in its own right. Rather it is a region in Bernstein's terms (interfacing with practices such as speech recognition or speech pathology), drawing on physics (acoustic phonetics) and neuro/biology (articulatory phonetics) -and thus deploying a set of internal and external grammars very different to those employed by linguistics proper (stronger grammars in fact). This is not to deny the relevance of phonetics and phonology to one another (linguistics students are generally trained in both), but simply to acknowledge the very different knowledge structures involved in the description of form as opposed to substance. 3Second, structure and syntagm. As far as the description of grammatical 'form' is concerned, SFL does not restrict its description to what Whorf (1945) called phenotypesi.e., single or multi-segment syntagms consisting (for grammar) of classes of morpheme, word, group/phrase or clause. In order to develop rich meaning-making grammatical descriptions SFL is also inspired by Whorf's notion of cryptotypes. Halliday and Matthiessen's (e.g., 2014) distinction between Epithet and Classifier in English can be used to illustrate this point here. From the perspective of system, English nominal groups make a distinction between describing and classifying. Both types of nominal group can be realised by the same syntagm (i.e., determiner ^ adjective ^ noun). But a covert distinction can be uncovered by asking whether the adjective in the relevant syntagm is gradable or not. Describing adjectives are gradable (a really lovely film) whereas classifying adjectives are not (*a very Korean film). 4 Accordingly the same syntagm is assigned different structures, as in examples ( 1) and ( 2) below. SFL grammar descriptions in other words are not simply a catalogue of syntagms; they build function structures on top of syntagms in order to reflect the meaning-making valeur at stake.( This means for example in relation to an SFS description of horizontally polarised images that an optional information value system can be set up realised by the function structure Given ^ New, without making the claim that all horizontally polarised imagic syntagms in fact realise this system. Relevant options are extended from Kress & van Leeuwen (2021: 2016-217) in Figure 4 below. The name of the system is INFORMATION VALUE; it is an optional system; if the feature [newsy] is selected, then the structural functions Given and New are present, in the sequence Given followed by New (with Given realised to the left and New to the right). This formalisation makes no claims about all horizontally polarised systems; it simply positions [newsy] ones as having a Given ^ New structure realised by a horizontally polarised imagic syntagm. Note in passing that this is perhaps too generous a reading of Kress &van Leeuwen's often criticised account of information structure in polarised images; but my point here is that SFS need not fall foul of their apparent overgeneralisations. This approach to axis leaves open the question of whether systems need to be set up to generalise the syntagms available for realising function structures in a given semiotic system -opposing for example all polarised images to non-polarised ones, and if polarised, all horizontally opposed images to all vertically opposed ones. This could be important if polarised images are used to realise different function structures (and thus different meanings) for a given semiotic system. Arrows (as opposed to lines) and grids (as opposed to vertical or horizontal alignment) are good examples of imagic syntagms that arguably need generalisation in such terms -since arrows are used to realise motion or links for example (not to mention the system network specific uses of arrows in Figure 4), and grids can realise cross-classification (as in linguists' paradigms) or momented activity (as in comics) for example (not to mention culturally specific arrangements such as that organising Shirley Purdie's remarkable artwork Goowoolem Gijam 'Gija plants' which features at the Museum of Contemporary Art in Sydney). 6 Martin & Unsworth (2024) take this step in their work on secondary school science infographics, drawing directly on work by Hiippala and colleagues (Hiippala et al. 2021). Martin and Unsworth's network for MACRO-GROUPING is presented in Figure 5. Therein, a square brackets means 'or' (as in Figures 2 and 3 above), a slanted square bracket indicates a cline, a brace means 'and', and a combination of brace and square bracket means 'and/or'). So for the CO-TEXT systems, we have the option of including a text block or not, and if we choose to do so, we can include a caption or an interpolation or both. And to follow one path in the DESIGN system, if we choose line, it can be more or less vertical or horitzontal or both (in the latter case we end up with a grid). If this syntagm oriented step is taken, then an analysis dedicated to such regularities of form can be established (e.g., Caple 2013Caple , 2022 on BALANCE systems for images), and some kind of stratification of 'meaning' and 'form' can potentially be brought into the description (as in He's 2021 work on animations). There is nothing in the knowledge structure of SFS, as informed by SFL, blocking stratified generalisations of this kind.A related point about knowledge structure and SFS can be made in relation to 'etics' and materiality. As van Leeuwen (1999Leeuwen ( , 2011) ) shows through his work on parametric systems for sound and colour, axis can be used to formalise descriptions which cover the material oppositions which afford traces of function structures and syntagms of the kind introduced above. I would hesitate to refer to these systems as a stratum of language or any other semiotic system since other than semiotic internal and external grammars inform their description; in Hjelmslev's terms we are dealing with substance, not form. As emphasised above in relation to phonetics and phonology, this is not to suggest that work on materiality is not relevant to semiosis. It is simply to restrict stratification to cases where we have bundles of interdependent systems at different levels of abstraction (in a pattern of pattern relationship referred to by Lemke 1984 as metaredundancy). The relation of 'emics' to 'etics' is not strictly speaking a pattern of this kind. Third, axis. In SFL other dimensions of internal grammar are all articulated in relation to axis. The concepts of rank, metafunction and strata in other words are all based on bundles of interdependent features (organised with respect to constituency, type of meaning or level of abstraction respectively). Instantiation has to do with the manifestation of system in text and the generalisation of instances as system, over time -as texts unfold (logogenesis), as speakers mature (ontogenesis) or as languages evolve (phylogenesis). And individuation has to do with the allocation of systems to members of a culture and their use of those systems to affiliate in social groups -once again, over time. Take away axis (e.g., O'Toole 1994) and you may arguably be left with one kind of functional theory or another, but not Systemic As suggested in Martin (2011), in multimodal studies which take axis as fundamental (i.e., SFS) it is critical not to make a priori assumptions about how systems will enter into interdependency relations with one another. Depending on the semiotic system in question, constituency (rank), kind of meaning and type of structure (metafunction) and level of abstraction (stratification) may shape external grammar (L 2 ), but may not. Looking across the SFS studies surveyed in Table 1, the constant L 1 notion is axis. Accordingly in SFS rank, metafunction and strata are clearly better positioned as mediating L 1.5 notions -possibly shaping the description (L 2 ), possibly not. Fourth, delicacy. Recognition of mediating languages of description (L 1.5 s) carries with it the idea that the relation between L 1 and L 2 can be treated as a cline. SFL's approach to axis is well adapted to a conception of this kind since systems are arranged from more general to more specific along a cline referred to as delicacy. Thus in Figure 6 for example indicative clauses are more general than interrogative ones, which are in turn more general than wh ones and so on. This makes it possible to be more and less specific about what gets treated as L 1 , L 1.5 and L 2 -perhaps treating the [indicative/imperative] opposition as L 1.5 , but remaining agnostic about more delicate options, pending construction of L 2 (i.e., a specific language's grammar of MOOD). It is important to keep in mind in relation to this point that positioning more or less general systems as L 1.5 can be done without making any claims at all about how such systems are realised in structure. Commitment to structural realisations of mood options needs to be withheld for L 2 , since structures realising mood vary considerably across languages (Martin 2018, Martin et al. 2021).Figure 6: English MOOD systems and delicacyfoot_6 Fifth, context. In SFL context is treated as form (Figure 7), not substance (as connotative semiotiocs in Hjelmslev's terms). In the model of SFL assumed here, context is stratified as register and genre (Martin 1992, Martin & Rose 2008); and register is modelled metafunctionally in proportion to its realisation in language -i.e., field is to ideational meaning, as tenor is to interpersonal meaning, as mode is to textual meaning. This move allows context strata to be treated as resources for making meaning and modelled axially, comparably to language ones (Doran & Martin 2021, 2024, Doran et al. 2024). These context dimensions of SFL's L 1 are less well articulated and more controversial than language onesprobably because of their level of abstraction and concomitant realisation as patterns of language patterns in addition to a lack of clarity in argumentation as far as the distinction between realisation and instantiation is concerned. We are now in a position to compare SFS with Bateman's proposals for the basic L 1 of what he refers to as a semiotic mode (a semiotic system in SFS). 8 In doing so we need to keep in mind the potential of SFS's L 1 in relation to extant descriptions of semiotic modes (L 2 ). External grammars (L 2 ) can all be critiqued and revised in ways that do not call for renovations or reconstructions of L 1 or L 1.5 . Some shortcomings, in other words, are more serious than others -as we shall flag below. SFL's stratified model of language and context, as presented in Figure 7, provides a useful point of departure for comparison with Bateman's diagrammatic illustrations of his modelling of semiotic modes. Bateman et al. (2017: 117) introduce the diagram in Figure 8, consisting of three strata -"the material substrate or dimension, the technical features organised along several axes of descriptions (abbreviated as 'form') and the level of discourse semantics." Compared to SFS this treats the 'etics' of materiality as a stratum proper and groups it together with the stratum of form -the two together realising discourse semantics.8 As noted above, the term mode is used for a register variable in SFL, so the term semiotic system is preferred to semiotic mode in SFS. "…meanings are only mediated by the application of the discourse semantics of the semiotic mode… Thus, as an example, whereas the often used classification of graphical resources set out in Kress andvan Leeuwen (2006 [1996]: 59-68) might classify graphical 'arrows' as 'narrative processes' (by virtue of their directionality as vectors), from the perspective of the approach adopted here this conflates two semiotic strata of description: the formal level at which visual properties of connection and directedness properly reside, and the discourse semantic level at which, under certain circumstances, it may be possible to abduce that the graphical connective is serving a 'narrative purpose' (but then, in other circumstances, it may not be). … meaning-making using semiotic modes is best characterised as discourse 'unfolding' and it is this that offers a higher 'unity' to any material regularities exhibited."This characterisation of meaning making in semiotic modes is helpfully reviewed in Bateman (2020b) in relation to Martin's (1992) model of discourse semantics. Seen in SFS terms Bateman's discourse semantics is strongly focused on instantiation -logogenesis in particular. From this perspective meaning is only ever something that can be abduced in relation to co-text and context as texts build meaning -rolling out a snowball of semiosis as they unfold. So what Bateman treats as a stratum called discourse semantics, SFS would interpret from the hierarchy of instantiation, not realisation. SFL's discourse semantic stratum, along with lexicogrammar and phonology, would all be treated as form in Bateman's modelling (i.e., "technical features organised along several axes of descriptions"; Bateman et al. 2017: 117). Of these, Martin's notion of covariate structure (e.g., Martin 2015) comes closest to Bateman's conception of discourse semantics -since covariate structures instantiate discourse semantic systems through a process of abducing relations of indefinite extent as texts unfold. And Martin's stratum of discourse semantics would have to be interpreted in Bateman's modelling as proposals for the systems of relations that can be so abduced.foot_7 In at least one articulation of Bateman's model (Bateman & Schmidt 2012: 81) discourse semantics as well as form is presented as involving paradigmatic systems of choice and organisation imposing structure. This modelling informs Bateman's (2007) work on semantic relations between shots in film (his 'grande paradigmatique') and Tseng & Bateman's (2011) description of filmic identification -with systems formalising relations to be abduced. In later work, perhaps because of reservations about the synoptic, non-dynamic nature of system networks,foot_8 axis seems to be reserved for the stratum of form. In its place, at the level of discourse semantics, lists of relations, elaborated from Mann & Thompson's (e.g., 1986) Rhetorical Structure Theory (RST), regularly function as reservoirs of meaning to be abduced (e.g., Bateman 2008, Hiippala 2015).As acknowledged in Section 3, compared to realisation, instantiation is to date a still developing hierarchy in SFL. Suggestive explorations include Martin (2006) on intralingual re-instantiation, de Souza ( 2006) on interlingual re-instantiation, Martin (2010) on coupling and commitment, Painter et al. (2013) on intermodal convergence (concurrence, resonance and synchronicity), Martin & Matruglio (2013) on presence, Martin (2017) on mass, Martin & Doran (2024) on context as realisation vs instantiation, and Martin & Unsworth (2024) on syndromes of instantiation referred to as mass, presence (and association in Martin & Doran 2024). None of these approaches comes anywhere near the level of explicitness and detail underpinning work by Bateman and his colleagues (e.g., Bateman & Rondhuis 1997, Wildfeuer 2021), inspired as it is by Asher & Lascarides' work on the logic of abduction (e.g., 1991, 1993, 2003). This conceptual shortcoming is clearly one area where SFL and SFS are certain to be positively influenced by Bateman's modelling of multimodality (discourse semantics in particular).In addition to the three strata outlined in Figure 8, alternative L 1 imagings of Bateman's model include a more abstract stratum comprising "context, social norms & values" (Bateman & Schmidt 2012: 81) or "context/register/situation" (Bateman 2022b: 69). And in related work (e.g., Bateman 2008Bateman , 2016) ) the notion of genre is brought into the picture. This work resonates with SFL work on modelling context as register and genre, and there is clearly room for ongoing collaboration in this area. That said, one weakness of some SFS work on semiotic systems such as that noted in Table 1 is the lack comprehensive treatments of register and genre -especially for descriptions clearly inspired by Kress & van Leeuwen's (1990 and subsequent editions) grammar of images. Further work, modelled on the attention paid to field and genre in Doran's work on physics and mathematics (2017) or to field in Yu's work on chemistry (forthcoming) is clearly in order. As foregrounded in Maton's Legitimation Code Theory (LCT) disciplinarity needs to be considered from the complementary perspectives of knowledge and knowers (Maton 2014). Accordingly LCT's legitimation code referred to as Specialisation takes into account both epistemic relations (between knowledge and what it describes) and social relations (between knowledge and who is producing it). Based on the strength of epistemic and social relations LCT establishes a topology of legitimation codes, with four principal 'modalities'knowledge codes (ER+, SR-) for which legitimacy depends on what you know; knower codes (ER-, SR+) where what matters is who you are; elite codes (ER+, SR+) where it matters both what you know and who you are; and relativist codes (ER-, SR-) through which everyone's voice and knowledge is equally valid (Maton 2014: 30-33). This framework is exemplified in Figure 9 with reference to speaking in 19th century debates about evolution. The key players were Watson (knowledge code), a middle class biologist who made a living by selling specimens to private collectors and museums, Darwin (elite code), an wealthy biologist who married into the Wedgewood family pottery fortune, and Bishop Wilberforce (knower code), a gifted public speaker and high ranking clergyman. They are positioned according to specialisation in Figure 9 below. Tellingly, it is Darwin (not Watson) who comes readily to mind when we think of evolution, along perhaps with Bishop Wilberforce who played the role of Huxley's protagonist in the famous debate at a meeting of the British Science Association in 1860. To complete the picture I've imagined a know-it-all nameless blogger (relativist code), opining about evolution in the 21st century. Turning to multimodality, we might position Bateman as legitimised by a knowledge code, van Leeuwen (a professional jazz musician and film maker) by an elite code, someone like Manuel Lima 11 by a knower code, and our blogger (still blogging) by a relativist code. 12 Kress's positioning would have to be a more transitional one, beginning ER+/SR-with the publication of Reading Images (1990), but sliding towards ER-/+SR thereafter -viz publications such as Literacy in the New Media Age (2003) and Multimodality: a social semiotic approach to contemporary communication (2010) which typify his later work, rarely draw on any analysis at all from Reading Images and popularise his thinking. With his passing in 2019, the field of multimodality lost its most influential knower. This is significant because turning a region into a singular with a strong internal and external grammar requires a range of voices (including popularising knowers), even if legitimation via a knowledge code is what protagonists like Bateman ultimately have in mind.Digging deeper into epistemic relations, Maton (2014: 175-177) distinguishes relations between knowledge practices and the part of the world they are oriented to (ontic relations, OR) and relations between knowledge practices (discursive relations, DR). Strong ontic and discursive relations (OR+/DR+) establish a purist code, which emphasises both the object of study and how it is studied. Weak ontic relations but strong discursive relations (OR-/DR+) establish a doctrinal code, which legitimate a multiplicity of objects of study but foregrounds 11 Lima is the only 'multimodalist', as far as I am aware, who has been called on to give a TED talk. His bestknown publications (2011,2014,2017) are popularisations. 12 I should perhaps emphasise here that codes do not ascribe value to speakers in different ways; they simply characterise the factors that legitimise a given voice. Darwin Wallace Bishop Wilberforce Blogger particular way of studying them. Strong ontic but weak discursive relations (OR+/DR-) allows for a situational code, whereby a specific object of study is in focus but it can be approached from multiple points of view (or as Maton allows, no clear knowledge code at all). Weak ontic and discursive relations (OR-/DR-) legitimate a approach which is unlimited with respect to both what is studied and how. As suggested in Figure 10, a framework of this kind would position SFL as purist (since it studies language from one theoretical perspective), SFS as doctrinal (since it studies any semiotic system but always from the perspective of SFL), Social Semiotics (a la Kress & van Leeuwen 2005) as DR-/OR-(since it encourages the study of multimodality from different points of view) and something like information visualisation as DR-/OR+ (since it focuses on graphic representations of complex data by whatever means afford a clear 'synoptic' overview). Seen in these terms one of Bateman's objectives is to re-orient the current trajectory of Multimodal Studies, which at some conferences seems to sprawl towards ever weaker discursive and ontic relations; he wants to shift its trajectory towards stronger discursive relations whatever its object of study (without, we might reiterate here, falling foul of the linguistic imperialism and predatory interdisciplinarity that SFS's doctrinal stance might be accused of). The main message to take from Maton's work on specialisation and epistemic relations is that disciplines involve both knowledge and knowers. Prescribing strong internal and external grammars for Multimodality Studies is not enough; a given field needs knowers as well. In this regard SFS has the advantage of being able to recruit both knowledge and knowers from SFL, since in practice SFL informed discourse analysis cannot avoid bumping into multimodal texts and the technicality of SFL's internal and external grammar is already in play. Given Batemans' vision for Multimodal Studies, recruiting knowers from Social Semiotics is perhaps more of a challenge since its relatively weak internal and external grammar and its multidisciplinary stance make stronger grammars a harder sell. Work which draws on Lascarides & Asher (1991, 1993) to formalise the complexity of discourse semantic abduction seems certain to frighten large numbers of OR-/DR-multimodalists well away. In this paper I have drawn Bernstein and Maton's sociology of knowledge to explore SFL and SFS in relation to Bateman's vision for empirical multimodality research. There is of course much more to survey. In closing let me just highlight three main points here. First, there is the question of which theoretical dimension is privileged as fundamental. For SFL/SFS this is axis; all other dimensions of the theory, including stratification, depend on a specific conception of paradigmatic relations underpinning and underpinned by syntagmatic ones. For Bateman the fundamental dimension is stratification, further specified as materiality, form and discourse semantics. So where SFL/SFS derives strata from axis depending on the interdependency of systems according to levels of abstraction in a particular semiotic system, Bateman's vision assumes three strata and in recent work uses axis to characterise just one of these (i.e., form). Related to this point is Bateman's treatment of materiality as a stratum, whereas in the model of SFL/SFS assumed here it would be treated as 'etic' substance and explored through knowledge structures which have evolved for the study of physical and biological reality (as opposed to those which have evolved for exploring semiotic reality, i.e., systems of meaning).Second, and perhaps most crucially from Bateman's perspective, discourse semantics is approached from a dynamic perspective by Bateman and his colleagues -with form imbued with meaning through a process of abduction as texts unfold. This is compatible with SFL/SFS's approach to instantiation (logogenesis in particular) and its conception of covariate structure. But as noted above, SFL/SFS description (L 2 ) has not caught up with theory (L 1 ) as far as instantiation is concerned. Bateman's misgivings about SFL/SFS's many promissory notes in this regard are right on target.Third, in SFL/SFS key concepts are deployed across strata. One dimension, axis, is fractal; all strata, ranks and metafunctions are explored axially, and systems of choice shape SFL/SFS's conception of hierarchy (realisation, instantiation and individuation). This axial orientation grounds decisions for a specific semiotic system -with respect to how many strata, how many ranks and which metafunctions (if any) are presumed as L 1 , suggested as L 1.5 or described as L 2 . Bateman's model is more modular in design, with distinct internal and external grammars proposed for each stratum (and for context it would appear, once we move beyond his semiotic modes). The accessibility of work on materiality by Bateman and his colleagues (i.e., their slices of canvas) contrasts markedly with the technicality of their adoption of Lascarides & Asher's formalisation of the logic of abduction. To be frank, it is clear to me that such an approach potentially formalises the complexity of what is going on ideationally in logogenesis as far as the snowballing of meaning is concerned; but I am much less clear about how it manages this complexity for descriptive or applied purposes (especially once we scale up and move beyond the fragments of exemplificatory discourse used as illustrations of the approach). This may simply be a matter of unfamiliar technicality and the challenge it imposes on outsiders (such as myself). But it may be more than that. As I often tell my research students when they are feeling overwhelmed by the phenomena they are describing, there is a difference between documenting complexity and managing it. The job of internal and external grammars in any discipline is to manage complexity, not simply catalogue it. I'll leave much needed discussion of this instantiation modelling crisis to another time (to another generation perhaps, who can come to our rescue in this regard).Overall my comment would be that SFL/SFS's internal grammar is by and large compatible with Bateman's vision, even if its external grammars fall short in so far as extant descriptions of one semiotic system or another are concerned. This is hardly surprising, since SFL has had considerable influence on Bateman's thinking. Where differences arise, I think that by and large the models can learn from each other -provided suitable contexts for working together are formed.Drawing on one knowledge structure to position two others as I have done is a challenging task. But it has the salutary advantage of drawing attention to the demands of coming to grips with the technicality of incommensurable L 1 s -which can be forbidding given the everyworsening time constraints of academic life and the challenge of practical applications which demand solutions yesterday for what really needs to happen tomorrow. This takes me to my final point, appliability, which in my experience bears critically on what it takes to come to grips with an unfamiliar L 1 and the L 2 descriptions it affords. Halliday (2008: 7) coins the term 'appliable linguistics' for the dialectic of theory, description and practice informing his linguistic work. An outline of this problem oriented perspective is outlined in Figure 11 (with multiple Ls allowing for the probability of a number of singulars influencing a given region of practice). This orientation to linguistics was an unusual one in the 20th century, the closest parallel being Pike's tagmemics (viz. Pike's tagmemics and his 'maxim' from1988: "I wanted a theory that would allow one to live outside the office with the same philosophy one uses inside it"). 13 The challenge for SFS and Bateman's Multimodal Studies, as I see it, lies precisely in finding contexts of application (in educational semiotics, clinical semiotics, forensic semiotics etc.) which foster a dialectic of theory, description and practice. It is in these contexts, especially if knowledge workers share a politics in relation to a specific problem, that the challenge of incommensurable technicality can be overcome. This dialectic works best when theory and description provide complementary perspectives on the same data. I am not in other words talking about interdisciplinarity (i.e., "you do your bit, I'll do mine"), but rather about transdisciplinarity (i.e., "this is how I see it, how about you?"). By approaching the same data from a different point of view, with a practical politically-charged challenge in mind, theories can learn from one another -if they can make explicit what another theory is interested in 13 From 'Kenneth L. Pike Maxims'; https://www.sil.org/about/klp/pike-maxims. Pike's Christian motivations were of course very different from Halliday's Marxist ones (Halliday 2015). and/or draw attention to regularities another has missed (Martin 2011, Maton at al. 2016, 2020). 14 It is for this reason that I have done what I can to encourage the evolution of SFS out of SFL -so it can 'trespass' into conversations of this kind. Dialogue between SFS and Bateman's vison for an empirical multimodality is likely to be far more productive than one involving SFL rather than SFS, precisely because SFS and Multimodal Studies can focus on the same data from complementary points of view. For me then it is important to avoid dialogue in which linguists focus on language and multimodalists deal with everything else. Multimodal discourse needs to be the focus of all parties in the conversation.In saying this I hope I am allaying fears about linguistic imperialism and have presented a less than predatory vision of SFS and SFL. One abiding concern I have is the 'logophobia' generated by multimodality 'knowers' through their by now rather dated proclamations of a new multimodality age and the striking absence of any language analysis to speak of in key multimodal conferences and publications.