When Renaming Replaces Modelling: Stratification, Instantiation, and the Flattening of SFL Architecture

Systemic Functional Linguistics (SFL) is not just a collection of analytic tools. It is a theory with a very specific architectural commitment: meaning is organised across different levels of symbolic abstraction, related by principled relations such as stratification, realisation, and instantiation. These relations are not optional metaphors; they are what give the theory its explanatory power.

In what follows, we want to examine a prominent reconstrual of SFL—associated most clearly with the work of J. R. Martin—not in order to dispute its descriptive utility, but to show that it rests on a fundamental misunderstanding of this architecture. The issue is not disagreement over details. It is a systematic flattening of relations that Halliday treated as categorically distinct.

This flattening has consequences: theoretical, methodological, and disciplinary.

---

1. Levels of symbolic abstraction are not modules

At the core of Halliday’s model is the idea that language is organised across strata that differ in level of symbolic abstraction. Semantics, lexicogrammar, and phonology are not parallel components that “interact”; they are asymmetrically related.

Lower strata realise higher strata.

Higher strata are realised by lower strata.

This is not a claim about where meaning “is”. It is a claim about how symbolic systems are organised so that meaning can be made at all.

A crucial distinction follows from this:

semogenesis (the fact that meaning is made across the system) is not the same thing as stratification (the organisation of symbolic abstraction).

When stratification is misunderstood as “all strata make meaning”, the architectural distinction collapses. Once that collapse occurs, all strata are treated as strata of meaning, rather than as differently abstract symbolic resources.

This is precisely where the reconstrual begins.

---

2. From stratification to interaction

In Martin’s work, strata and metafunctions are consistently treated as modules that relate by interaction rather than by asymmetrical realisation. This is visible in the claim that all strata “make meaning”, and more starkly in the treatment of phonology as a stratum of meaning.

From a Hallidayan perspective, this is a category error.

Phonology does not mean in the same sense that semantics does. It stabilises and transmits distinctions that are already semiotically differentiated at a higher level of abstraction. Treating phonology as a meaning stratum erases the very notion of symbolic abstraction that stratification was introduced to explain.

Once this move is made, the theory no longer has the resources to explain why different symbolic levels exist at all.

---

3. Instantiation is not descent

A parallel misunderstanding occurs with instantiation.

In Halliday’s model, instantiation is a perspectival relation between potential and instance. A system can be viewed from the pole of potential (as a system) or from the pole of instance (as a text or event). Nothing “moves” down a ladder. No stratum instantiates another stratum.

Instantiation is not a process. It is a way of seeing the same semiotic organisation from different perspectives.

In Martin’s reconstrual, however, instantiation is characterised as all strata instantiate, as if instantiation were a kind of downward movement of meaning. This confusion becomes decisive in his treatment of context.

---

4. How context became language

Halliday distinguishes context from language. Context is not a higher stratum of language; it is a different order of abstraction altogether. Field, tenor, and mode are contextual variables that are realised by semantic patterns in language, not systems within language.

Martin reconstructs this architecture by:

• redefining context potential (context of culture) as a stratum of genre

• redefining context instance (context of situation) as a stratum of register

• treating field, tenor, and mode as systems of register

This move converts a relation of instantiation into a relation of interstratal realisation, collapsing context into language. The result is a model in which register and genre are systems of context rather than ways of relating semiotic potential to situation and culture.

The cost of this move is high: the theory loses the ability to distinguish contextual organisation from linguistic organisation in principle.

---

5. Renaming as a method of theorising

These architectural confusions would be serious enough on their own. But they are compounded by a distinctive method of theorising.

Rather than modelling new phenomena from data, Martin repeatedly takes existing Hallidayan analyses and renames them, presenting the renamed constructs as novel theoretical advances.

The most striking example is the stratum of discourse semantics.

Halliday’s stratum is simply semantics. In Cohesion in English (Halliday & Hasan 1976), cohesion is analysed as a set of lexicogrammatical resources.

Martin elevates cohesion to a new stratum and relabels its systems:

• reference becomes identification

• lexical cohesion becomes ideation

• conjunction becomes connexion

• Halliday’s semantic system of speech function becomes negotiation

These renamings are not accompanied by a new theory of symbolic abstraction. On the contrary, they are enabled by misunderstandings—such as confusing reference with deixis—which create apparent novelty by differentiation.

Misunderstanding becomes a generative resource.

---

6. Why this matters

This is not a dispute about terminology, nor about whose framework is “better”. It is about what happens when a theory built on differentiated relations is reconstructed as a flat landscape of interacting modules.

When stratification, instantiation, realisation, and variation are all treated as the same kind of relation:

• explanatory power is replaced by descriptive proliferation

• architectural constraints disappear

• theoretical novelty becomes indistinguishable from relabelling

The result may be productive in the short term, but it is unstable. It cannot explain its own foundations.

---

7. A broader lesson

The deeper issue here is not confined to SFL.

This case illustrates how a theoretical architecture can be hollowed out while its vocabulary is preserved. Once the relations that give concepts their force are flattened, the theory may continue to circulate—but it no longer does the work it once did.

Recovering Halliday’s model is not a matter of loyalty or orthodoxy. It is a matter of taking symbolic abstraction seriously.

Without that commitment, we are no longer doing systemic functional theory at all—we are simply rearranging its names.

Anaphoric Reference

Yaegan Doran On Sundanese Nominal Groups

Sundanese nominal groups: A textual grammar

Yaegan Doran

The Australian Catholic University

This talk considers the nominal group in Sundanese, a Malayo-Polynesian language of West Java, Indonesia. In particular, it builds a meaning-based description of Sundanese nominal groups, focusing in particular on textual meaning. Born of educational concerns associated with literacy programs, this talk describes the nominal group not only in terms of the formal syntagms at play, but also their functions; not only the paradigmatic choices that are available, but how they are taken up in text; and not only the grammar by itself, but how it realises meanings in discourse. The aim is to develop a richly co-textualised and metafunctional description of Sundanese nominal groups that can explain text patterns from a range of genres and registers.

Blogger Comments:

To be clear, this paper did not discuss the textual meaning of the nominal group. Instead, it was concerned with reference, which is not a system of the nominal group, though the nominal group, along with the adverbial group, is a grammatical domain in which reference items are located.

One reason why reference is not a system of the nominal group is that it is not realised by a structural relation within the nominal group. Instead, reference is realised by a non-structural relation between a reference item and a referent which may not even be in the text (exophoric reference), let alone in the same nominal group.

This confusion was compounded by the fact that Doran used Martin's model of reference (identification), though without acknowledging the fact, which is (purportedly) discourse semantics, not grammar. Moreover, as explained in great detail here, Martin's model mistakes nominal groups for reference items, mistakes ideational denotation for textual reference, mistakes interpersonal deixis for textual reference, and mistakes non-reference ("presenting reference") for reference.

Ellipsis And Textual Prominence

Halliday & Matthiessen (2014: 635):

Ellipsis marks the textual status of continuous information within a certain grammatical structure. At the same time, the non-ellipsed elements of that structure are given the status of being contrastive in the environment of continuous information. Ellipsis thus assigns differential prominence to the elements of a structure: 

if they are non-prominent (continuous), they are ellipsed; 

if they are prominent (contrastive), they are present. 

The absence of elements through ellipsis is an iconic realisation of lack of prominence.

Deixis Vs Reference

Personal and demonstrative DEIXIS is a system of the nominal group, personal and demonstrative REFERENCE is not a system of the nominal group.

Personal and demonstrative DEIXIS serves an interpersonal function, personal and demonstrative REFERENCE serves a textual function.

Halliday & Matthiessen (2014: 367):

Many languages embody these two forms of deixis in the structure of the nominal group. The two are closely related, both being (as indicated by the term ‘deixis’) a form of orientation by reference to the speaker – or more accurately, to the ‘speaker-now’; the temporal-modal complex that constitutes the point of reference of the speech event.

Halliday & Matthiessen (2014: 623):

The textual status at issue in the system of reference is that of identifiability: does the speaker judge that a given element can be recovered or identified by the listener at the relevant point in the discourse or not? If it is presented as identifiable, then the listener will have to recover the identity from somewhere else. If it is presented as non-identifiable, then the listener will have to establish it as a new element of meaning in the interpretation of the text.

Theoretical Problems With Doran's 'Factoring Out Structure: Nuclearity, Linearity And Iteration'

Factoring out structure: Nuclearity, linearity and iteration

Yaegan Doran:

Like much of its framework, SFL offers a very elaborate model of structure. Its metafunctional distribution of periodic, prosodic and particulate structures is well known, with numerous distinctions within each type (Halliday 1979). Focusing on particulate structures, Halliday distinguishes between multivariate and univariate structures, and within univariate structures, between hypotactic and paratactic structures (Halliday 1965, Halliday and Matthiessen 2014), while Martin (1996) conceptualises particulate structures as involving orbital and serial relations (Martin 1996). Similarly, relations in discourse are alternatively conceptualised from the perspective of grammar as non-structural (Halliday and Hasan 1976) or from discourse semantics as involving co-variate structures (Martin 1992, adapted from Lemke 1985).  

Many of these types of structure are sharply distinct, while others have significant overlap. Together, they account for a wide range of linguistic phenomena and do a good job at describing the patternings that occur across regions of language. However as SFL has expanded its descriptive framework into a wider range of languages, it has been clear that there are grammatical constructions that are not quite accounted for. Building upon this work, this talk offers a generalised model of particulate structure, developed with Jim Martin, that aims to cover a wider range of grammatical patterns. 

This model does not start with types of structure, but rather with three sets of factors that can be combined relatively freely: whether the structures involves a nucleus and subordinate satellites (nuclearity), whether there is linear interdependency between the elements in the structure (linearity), and whether the relations can occur more than once (iteration). These factors generate a wider set of structures than so far acknowledged explicitly in SFL. Together, they offer a means of seeing both the differences between types of structure already acknowledged, as well as variation within them. They also generate structures that account for phenomena not yet considered structurally in SFL to date. 

The goal of this talk is build more explicitness in SFL’s model of structure as it continues its descriptive expansion and in doing so, help clarify issues that have often been left aside in SFL grammatical description. Where possible, the model will be exemplified through English but I will bring in other languages as need be.

Blogger Comments:

[1] To be clear, Halliday soon after revised his model of grammatical structure types in favour of the current typology. Halliday (1994: 36) :

[2] To be clear, as demonstrated here and here, Martin's orbital vs serial distinction purports to model the structural distinction for the experiential vs logical metafunctions, but actually models the distinction between hypotaxis (nucleus vs satellite status) and parataxis (nuclei of the same status).

[3] To be clear, these are cohesive relations at the level of lexicogrammar, and Halliday & Matthiessen (2014) provides an updated interpretation.

[4] To be clear, by 1992, Lemke had already recanted his view that 'covariate' was a type of structure. Lemke (1988: 159):

My own 'covariate structure' (Lemke 1985), which includes Halliday's univariate type, is for the case of homogeneous relations of co-classed units, and should perhaps be called a 'structuring principle' rather than a kind of structure.

[5] To be clear, this remains a bare assertion until it is supported by evidence. The previous attempt to provide such evidence, Martin's seminar of 13/8/21, was demonstrated to be based on theoretical misunderstandings (see here).

[6] To be clear, Doran's work builds on Martin's misunderstanding of structure; see [2].

[7] To be clear, here Doran is modelling metalanguage (SFL Theory), not language, by essentially devising a metalinguistic system of features ("factors") that generates types of structure, as if metalanguage were language.

[8] To be clear, each of these factors is theoretically problematic.

Firstly, nuclearity corresponds to Martin's orbital structure, which purports to be a model of experiential structure, but is actually a model of logical hypotaxis (independent nucleus vs dependent satellites). That is, nuclearity misconstrues a multivariate structure as a univariate structure.

Secondly, linear interdependency is just interdependency. The word 'linear' is redundant, since all structures are linear. (For Halliday (1965), univariate structures are lineally recursive.) Linear interdependency corresponds to Martin's serial structure, which purports to be a model of the univariate structure of the logical metafunction, but is actually just a model of one of its dimensions: parataxis (multiple nuclei of the same status).

Thirdly, in SFL Theory, iteration describes the type of structure favoured by the logical metafunction (Halliday (1994: 36). The purpose of introducing Halliday's term here is to provide the possibility of 'iterative nuclearity' and 'non-iterative linear interdependency'; see [9].

To summarise, Doran's three factors actually concern:

• hypotactic interdependency
• paratactic interdependency
• interdependency in general.

[9] To be clear, this "wider set of structures" includes iterative multivariate structures and non-iterative univariate structures (subjacency duplexes), neither of which withstands close scrutiny, as demonstrated here.

[10] To be clear, these phenomena include structure markers misconstrued as units in unit complexes, as in Martin's notion of a subjacency duplex.

Postscript:

The theoretical misunderstandings in this seminar turned out to be so complicated, that the paper is now being examined, slide by slide, here.

Halliday's Definition Of Mode

Halliday (2002 [1981]: 225):

Halliday has suggested (1975) that the “textual” properties of a text – the cohesive patterns and those of ‘functional sentence perspective’ – tend to be determined by the “mode”, the function ascribed to the text in the given context of situation, the purpose it is intended to achieve.

External vs Internal Cohesive Conjunction

Halliday & Matthiessen (2014: 611):

Relations between representations of segments of experience are called external relations, and conjunctions marking such relations are called external conjunctions. … 

Relations linking text segments in their interpersonal guise are called internal relations – internal to the text as a speech event, and conjunctions marking such relations are called internal conjunctions.

Today's 'Typology Forum' Paper By David Rose

The textual metafunction: making matter mean

David Rose

University of Sydney

The textual metafunction is sometimes presented as if it were an afterthought, there merely to organise interpersonal and ideational meanings into convenient structures of clause and text. But in the first stages of language, and in much of everyday discourse, it is the primary means of turning the world into meaning, with exophoric reference to the shared sensations of me and you, this and that, here and there. Once the world gets classified, and configured into clauses, and strung out into texts, the same resources used to indicate the world around us are re-deployed to indicate the semiotic world within a text. Where once there were only signs, now meanings come in messages, each of which must be sensible in the unfolding world of the text, and manage listeners’ attention to the information they present. 

Textual typological studies might start from observing these three general functions: introducing and tracking meanings through a text, or phoricity, contextualising messages, or thematicity, and managing information prominence, or newsworthiness. Resources for realising these functions are prototypically distributed between language strata: phoricity in the reference chains of discourse, thematicity in segmental structures of clauses, and newsworthiness in periodic structures of tone groups. But while they are in principle independently variable, both the functions and their realisations interact with each other, and variations in textual functions proliferate as a result. Perhaps also as a consequence of these interactions, their realisations come to vary across languages and registers. 

This paper will make some general observations about interactions between phoricity and thematicity and between thematicity and newsworthiness, and will exemplify variations in these interactions between languages. In the process, it will also illustrate textual variations in register, from dialogue to monologue, and from accompanying to constituting fields.

Blogger Comments:

[1] Rose's negative appraisal — implied by 'afterthought' and 'merely' (counter-expectancy: limiting) — of an unsourced characterisation of the textual metafunction is unjustified. Halliday & Matthiessen (1999: 398):

The textual metafunction is second-order in the sense that it is concerned with semiotic reality: that is, reality in the form of meaning. This dimension of reality is itself constructed by other two metafunctions: the ideational which construes a natural reality, and the interpersonal, which enacts an intersubjective reality. The function of the textual metafunction is thus an enabling one with respect to the rest; it takes over the semiotic resources brought into being by the other two metafunctions and as it were operationalises them.

[2] Here Rose confuses the ideational and textual metafunctions.  It is the ideational metafunction that construes experience as meaning ("turns the world into meaning").

[3] As Halliday & Matthiessen (1999: 530) point out, the textual metafunction only operationalises the resources already created by the other metafunctions:

Reference is a way of referring to things that are already semiotically accessible: either actually, in the text, or potentially, in the context of situation.

[4] To be clear, "classifying the world" is construing experience as ideational meaning.

[5] To be clear, it is experiential options that are "configured into clauses", and "configured" takes a structural (syntagmatic) rather than systemic (paradigmatic) perspective on the clause.

[6] To be clear, it is meaning that is "strung out into texts", through the process of instantiation during logogenesis.

[7] To be clear, these resources are ideational potential, construed of experience, and on the SFL model, they do not "indicate" the world, since there are no meanings outside semiotic systems that can be "indicated".

[8] To be clear, in theoretical terms, this "redeployment" is the process of instantiation during logogenesis.

[9] To be clear, in theoretical terms, "indicate" here refers to the logical projection of second-order meaning.

[10] To be clear, in SFL terms, this is the semogenic distinction between protolanguage ("signs") and language ("messages"), though there is no reason to think that this is Rose's meaning.

[11] Here Rose switches to promoting Martin's work on textual systems, as if the preceding paragraph leads naturally to them, and as if they address the (confused) issues he has raised.

[12] This misrepresents the model in Martin (1992).  The meanings that are introduced and tracked are participants, and the system for doing so is termed IDENTIFICATION, not phoricity.  Martin's model is Halliday & Hasan's (1976) model of cohesive reference, misunderstood, rebranded and relocated from lexicogrammar to discourse semantics (evidence here).  Essentially, Martin confuses

• textual reference with reference in the sense of ideational denotation, like his source Du Bois (1980),
• interpersonal deixis of nominal group structure with non-structural textual reference, and, as a result,
• nominal groups realising participants with reference items,

which is why the experiential feature 'participant' is presented (Martin 1992: 325) as the unit and entry condition of this textual discourse semantic system.

[13] These refer to Martin's (1992) reworkings of Halliday's THEME and INFORMATION, which confuse linguistic theory (Theme, New) with writing pedagogy, namely:

• introductory paragraph, rebranded as Macro-Theme, 
• topic sentence, rebranded as Hyper-Theme (misunderstood from Daneš (1974)), 
• paragraph summary, rebranded as Hyper-New, and 
• text summary, rebranded as Macro-New.

Evidence here.

[14] In SFL theory, these are all systems of the lexicogrammatical stratum.

[15] In SFL theory, reference chains represent (non-structural) cohesive relations within the lexicogrammar.

[16] The claim that thematicity is "distributed" in segmental structures is not only inconsistent with SFL theory, but also with Martin (1992), and most notably with Martin & Rose (2007), where such structures are modelled as waves in a chapter titled Periodicity.

[17] This follows Martin (1992: 384) in confusing the content plane (INFORMATION is a lexicogrammatical system) with the expression plane (the tone group is a rank unit of the phonological stratum).

[18] The notion of "interaction" between strata — different levels of symbolic abstraction — repeats Martin's (1992: 488) misunderstanding of the dimensions of the SFL theoretical architecture as modules.  See the explanatory critiques here.

[19] To be clear, in SFL theory, because 'register' refers to a functional variety of language, it is modelled as a point on the cline of instantiation.  Here Rose follows Martin (1992) in misconstruing register as a stratum of context, the culture as semiotic system.  See the critiques here.

[20] To be clear, in SFL theory, the dialogue vs monologue distinction is one of MODE, a system of context (culture), not register (language).

[21] The notion of 'constituting field' derives from Martin (1992: 185, 193) where it applies to activity sequences, which are modelled in that work as context (field), misconstrued as register, but in Martin & Rose (2007: 100-3) as discourse semantic.  To complicate matters further, Rose, like Martin, confuses the ideational semantics (that realises a field) with field (misconstrued as register).

Moreover, in presenting this distinction as a textual distinction at the level of context, it invites confusion with Hasan's (1985/9: 58) feature opposition of ancillary vs constitutive LANGUAGE RÔLE in her model of MODE, the textual system at the level of context (culture), not register (language):

First, there is the question of the LANGUAGE RÔLE — whether it is constitutive or ancillary. These categories should not be seen as sharply distinct but rather as two end-points of a continuum.

Halliday On 'Collocation'

Collocation Defined

Halliday & Matthiessen (2014: 648-9):

At the same time there are other instances of lexical cohesion that do not depend on any general semantic relationship of the types just discussed, [repetition, synonymy, hyponymy, meronymy] but rather on a particular association between the items in question – a tendency to co-occur. This ‘co-occurrence tendency’ is known as collocation.

The Semantic Basis Of Collocation

Halliday & Matthiessen (2014: 649):

In general, the semantic basis of many instances of collocation is the relation of enhancement, as with dine + restaurant, table; fry + pan; bake + oven. These are circumstantial relationships (for collocations involving Process + Manner: degree, e.g. love + deeply, want + badly, understand + completely, see Matthiessen, 2009b), but as the example with smoke + pipe illustrates, participant + process relationships also form the basis of collocation – the most important ones involving either Process + Range (e.g. play + musical instrument: piano, violin, etc.; grow + old) or Process + Medium (e.g. shell + peas, twinkle + star, polish + shoes); and there are also combinations involving functions in the nominal group, in particular, Epithet + Thing (e.g. strong + tea, heavy + traffic, powerful + argument) and Facet + Thing (e.g. pod + w[h]ales, flock + birds, school + fish, herd + cattle, gaggle + geese).

The Measure Of Collocation

Halliday & Matthiessen (2014: 59):

The measure of collocation is the degree to which the probability of a word (lexical item) increases given the presence of a certain other word (the node) within a specified range (the span). This can be measured in the corpus.

SFL Work On Collocation

Halliday & Matthiessen (2014: 60n):

The notion of collocation was first introduced by J.R. Firth (1957) (but note Hoey, 2005), and gained wide acceptance, particularly in work based on corpus analysis, as in the Birmingham tradition, e.g. Sinclair (1987, 1991), Coulthard (1993), Hoey (2005) and Cheng et al. (2009). For further systemic functional accounts of collocation, see e.g. Halliday (1966b), Halliday & Hasan (1976: Section 6.4), Benson & Greaves (1992), Gledhill (2000), Tucker (2007) and Matthiessen (2009b); Matthiessen (1995a) relates collocational patterns to structural configurations such as Process + Medium, Process + Range, Process + Degree; Thing + Epithet (for a corpus-based study of Process + Degree, see Matthiessen, 2009b).

Collocation Vs Other Types Of Lexical Cohesion

Halliday & Matthiessen (2014: 644):

Meaning "Beyond" The Clause

Halliday & Hasan (1976: 10):

It will be clear from what has been said above that cohesion is not just another name for discourse structure.  Discourse structure is, as the name implies, a type of structure; the term is used to refer to the structure of some postulated unit higher than the sentence, for example the paragraph, or some larger entity such as an episode or topic unit.

Note that the units postulated by Martin (1992: 325, 385) for discourse semantic structure — move, participant, message, message part — are not 'higher than the sentence', and have no postulated internal structure.

As Halliday & Hasan (1976) theorised, on the SFL model, "meaning beyond the clause" is realised by the systems of cohesion, the non-structural resource of the textual metafunction.

Do Ellipsed Subjects Function As Theme?

No — this would be a contradiction.  Ellipsis marks elements as textually non-prominent, whereas Theme marks elements as textually prominent. Therefore, ellipsed elements do not function as Theme.  Halliday & Matthiessen (2004: 563):

Ellipsis marks the textual status of continuous information within a certain grammatical structure. At the same time, the non-ellipsed elements of that structure are given the status of being contrastive in the environment of continuous information. Ellipsis thus assigns differential prominence to the elements of a structure: if they are non-prominent (continuous), they are ellipsed; if they are prominent (contrastive), they are present. The absence of elements through ellipsis is an iconic realisation of lack of prominence.

Meaning "Beyond" The Clause

In SFL theory, the two enabling metafunctions, the logical and the textual, provide the means of realising meaning "beyond" the clause.

The logical metafunction does this structurally, by relating clauses in complexes by means of interdependency and the logico-semantic relations of expansion and projection.

The textual metafunction does this non-structurally, through the cohesive relations of reference, substitution–&–ellipsis, conjunction and lexical cohesion.

Anyone who claims that the grammatics does not model meaning "beyond" the clause does not understand the theory (or is being dishonest).

Conjunctive Adjuncts

	Type	Meaning	Examples
I elaborating	appositive	‘i.e., e.g.’	that is, in other words, for instance
	corrective	‘rather’	or rather, at least, to be precise
	dismissive	‘in any case’	in any case, anyway, leaving that aside
	summative	‘in short’	briefly, to sum up, in conclusion
	verificative	‘actually’	actually, in fact, as a matter of fact
II extending	additive	‘and’	also, moreover, in addition, besides
	adversative	‘but’	on the other hand, however, conversely
	variative	‘instead’	instead, alternatively
III enhancing	temporal	‘then’	meanwhile, before that, later on, next, soon, finally
	comparative	‘likewise’	likewise, in the same way
	causal	‘so’	therefore, for this reason, as a result, with this in mind
	conditional	‘(if…) then’	in that case, under the circumstances, otherwise
	concessive	‘yet’	nevertheless, despite that
	respective	‘as to that’	in this respect, as far as that’s concerned

(Halliday & Matthiessen 2004: 82)

Expansion Type	Sub-Types			Items
elaboration	apposition	expository		in other words, that is (to say), I mean (to say), to put it another way
		exemplifying		for example, for instance, thus, to illustrate
	clarification	corrective		or rather, at least, to be more precise
		distractive		by the way, incidentally
		dismissive		in any case, anyway, leaving that aside
		particularising		in particular, more especially
		resumptive		as I was saying, to resume, to get back to the point
		summative		in short, to sum up, in conclusion, briefly
		verificative		actually, as a matter of fact, in fact
extension	addition	positive		and, also, moreover, in addition
		negative		nor
		adversative		but, yet, on the other hand, however
	variation	replacive		on the contrary, instead
		subtractive		apart from that, except for that
		alternative		alternatively
enhancement	temporal	simple	following	then, next, afterwards [including correlatives first…then]
			simultaneous	just then, at the same time
			preceding	before that, hitherto, previously
			conclusive	in the end, finally
		complex	immediate	at once, thereupon, straightaway
			interrupted	soon, after a while
			repetitive	next time, on another occasion
			specific	next day, an hour later, that morning
			durative	meanwhile, all that time
			terminal	until then, up to that point
			punctiliar	at this moment
		simple internal	following	next, secondly (‘my next point is’) [including correlatives first…next]
			simultaneous	at this point, here, now
			preceding	hitherto, up to now
			conclusive	lastly, last of all, finally
	manner	comparison	positive	likewise, similarly
			negative	in a different way, contrariwise
		means		thus, thereby, by such means
	causal-conditional	general		so, then, therefore, consequently, hence, because of that, for
		specific	result	in consequence, as a result
			reason	on account of this, for that reason
			purpose	for that purpose, with this in view
			conditional: positive	then, in that case, in that event, under the circumstances
			conditional: negative	otherwise, if not
			concessive	yet, still, though, despite this, however, even so, all the same, nevertheless
	matter	positive		here, there, as to that, in that respect
		negative		in other respects, elsewhere

(Halliday & Matthiessen 2004: 542-3)