Methodological considerations in language description: an interview with Christian M.I.M. Matthiessen

1 Introduction

Language description has been a core activity in linguistics since antiquity. Arguably, it is the foundation and powerhouse of linguistics science. Language description refers to the systematic, comprehensive account of the semantics, lexicogrammar and phonology of languages. We say lexicogrammar instead of the popular term “morphosyntax” or “morphology” and “syntax” because not all languages make a neat division between morphology and syntax. Chinese, for instance, is an isolating (non-inflectional) language, and it is not empirically rewarding to the grammarian to make a division between “morphology” and “syntax” in Chinese. Arabic, on the other hand, has a rich inflectional morphology and language description in Arabic can target Arabic morphology (word grammar), Arabic syntax (clause and group/phrase grammar), or Arabic lexicogrammar (a holistic view of wording in the language).

In principle, language description can cover any stratum or a combination of strata of a language, comprising semantics (i.e. content substance), lexicogrammar (i.e. content form), phonology (i.e. expression form), or phonetics (i.e. expression substance). In practice, however, the point of entry into the description of languages is the lexicogrammar (content form) and the phonology (expression form). Both semantics, i.e. meaning, and phonetics, i.e. the physical properties of speech sounds, are the substance of language. They are the outer strata of language, with semantics interfacing with the social environment of language and phonetics interfacing with the bio-physiological environment. As substance, semantics and phonetics can properly be observed by the analyst in their manifestations in form. Lexicogrammar and phonology, as the formal aspects of language, provide the window for linguists to observe and systematically analyse the meaning potential and sounding potential of languages. Thus, the descriptivist constantly keeps an eye on meaning in describing lexicogrammatical forms and an eye on phonetic properties of sounds in describing phonological forms. A good grammar will account for the meaning potential of a language just as a good phonology will account for the phonetic properties of sounds. The foregrounding of meaning in describing lexicogrammar has been very central and defining in functional linguistics, including systemic functional linguistics (SFL) and the Prague School (see Davidse 1986; Halliday and Matthiessen 2006).

In ancient times, language description had focused on accounts of individual languages such as Greek and Latin in Classical grammar of the 5th century BC, Sanskrit in India in the 2nd to 6th century BC, and Classical Chinese. Today, the scope of language description has widened greatly. While the term “language description” is still used in its narrow sense to refer to the description of individual languages, the core activity it entails has been extended to newer domains of research, comprising language contrast, language comparison and language typology. Language contrast deals with two languages, language comparison with a few languages and language typology typically focuses on a representative sample of a wide range of language families. In this paper, our primary focus is on language description, but we incorporate the other three areas of descriptive activity into the discussion.

The objective of the paper is to discuss pertinent methodological considerations in language description. We find this important because most descriptive grammars provide little information on methodology and many grammarians learn to describe languages through many years of practice and apprenticeship. We believe that both novice grammarians and experienced grammarians will find the information here useful pedagogical material. Although our discussion is principally reflections of our own practice as grammarians, we also draw on several sources and on accounts by other linguists. Most of the questions addressed are designed around a plenary speech given by Christian M.I.M. Matthiessen at the 48th International Systemic Functional Congress (ISFC-48) in Sousse, Tunisia, on 20th March 2023. The theoretical framework that guides our discussion is SFL and we hope to provide guidelines to researchers who are interested in producing systemic functional descriptions of languages or who are interested in using some of the ideas developed in SFL in their accounts of languages.

The paper is organized as follows. First, we will discuss the goals of language description in SFL under the notion of appliable linguistics. Second, we will contextualise language description within the evolution of language, addressing the notions of protolanguage, archaic language, and modern language. Third, we will discuss descriptive categories in language description, comparison, and typology by engaging the work of Martin Haspelmath on descriptive categories and comparative concepts. Next, we will introduce the notion of “theory as a guiding grid” in language description and show how the architecture of language according to SFL can guide and ensure comprehensiveness in language description. Then, we will demonstrate the need to shunt or move along different dimensions in the architecture of language in language description as opposed to being fixated as one point in the overall architecture of language. We will then discuss the different entry points in empirically describing a language and in writing a grammar. Practical guidelines will be given on how to structure a grammar. The next sections will proceed to highlight some common mistakes that tend to be made in language description and the role of register in sampling texts for language description. This paper concludes with final remarks and advice to novice researchers in language description.

2 Developing an appliable description of language

Isaac Mwinlaaru: We have just finished the ISFC-48 in Sousse, Tunisia.^[1] It was an exciting four-day event following the pre-congress institute at the University of Sfax. I am excited to be here and to have the privilege of listening to your plenary yesterday (Matthiessen 2023a). I have picked a few points from your talk and our previous discussions on language description. I think what came up quite strongly and importantly in your plenary was how to develop systemic functional descriptions of language, which I find interesting and fascinating. In the past few years, students also raised several questions to me on how to describe their native languages. Let me start with the following question: What will be the goals of describing the grammar of a language based on SFL?

Christian M.I.M. Matthiessen: Let me begin by noting the fundamental distinction in SFL between theory and description, and between theoretical categories and descriptive ones since this distinction will be relevant at various points throughout our conversation. It has been present in SFL since the beginning, even before SFL became SFL; it goes back to the insights by J.R. Firth and Louis Hjelmslev. I’ll quote an accessible account by Halliday (2003 [1992]: 201–202) (see also Butt 2019; Matthiessen 2023b; Matthiessen and Teruya 2024):

The categories used in the analysis are of two kinds: theoretical, and descriptive. Theoretical categories are those such as metafunction, system, level, class, realization. Descriptive categories are those such as clause, preposition, Subject, material process, Theme. (Halliday 2003 [1992]: 201–202, bold in original)

Theoretical categories are, by definition, general to all languages: they have evolved in the construction of a general linguistic theory. They are constantly being refined and developed as we come to understand more about language; but they are not subject to direct verification. A theory is not proved wrong; it is made better – usually step by step, sometimes by a fairly catastrophic change.

Descriptive categories are in principle language-specific: they have evolved in the description of particular languages. Since we know that all human languages have much in common, we naturally use the descriptive categories of one language as a guide when working on another. But, if a descriptive category named “clause” or “passive” or “Theme” is used in describing, say, both English and Chinese, it is redefined in the case of each language […].

So, for example, while system itself is a theoretical category, each instance of a system, such as mood, is a descriptive category; similarly, option (or feature) in a system is a theoretical category, while each particular instance of an option, like indicative or declarative, is descriptive.

Descriptive categories are thus of a lower order of abstraction. They can be defined in such a way as to make them subject to verification. For example, if in defining “passive” we include morphological criteria, saying that passive is distinguished from its alternative (“active”) by systematic variation in the morphology of the verb, then it becomes possible to say that a particular construction in a given language is not a passive, or that there is no passive in the given language at all (Note that, if it is claimed that some descriptive category is a “universal” of language, such a claim can only be evaluated if there is some explicit formal definition of this kind. A universal feature is different from a theoretical category: it is a descriptive category that is being said to be present in every language).

In our glossary of key terms in SFL, we indicate for every term whether it denotes a theoretical category or a descriptive one (Matthiessen et al. 2010). Now, against this background, let me return to your question. We can compare the question with the goals you set for yourself when you developed the first comprehensive text-based, meaning-oriented and contextually sensitive description of Dagaare (Mwinlaaru 2017). One of the very general goals for a systemic functional description of a language is that it should be an appliable one (e.g. Halliday 2008; Matthiessen 2014, 2023c; Matthiessen and Teruya 2024: 196–222). That means it should be a description that is powerful enough to be able to be applied in a wide range of institutional settings in a community. As an appliable description, for example, it should be a resource not only in institutions of academic research but also in institutions of education, of healthcare, or of the legal system. It should be a resource in any institution where the lives of people in the community are affected and where problems could be solved by having an appliable description of the language.

What does it mean to be appliable? The description should ultimately be comprehensive, instead of merely being concerned with fragments. It certainly should not be just fragments that are selected to explore aspects of a theory or questions that come from philosophy, say, trying to settle the issue between rationalism and empiricism. It should be comprehensive and be based on naturally-occurring texts in context selected from a number of registers in different types of context. We select samples of different registers (i.e. functional varieties of language). The description should also be meaning-oriented so that, among other tasks, it can be used in the analysis of texts in context (e.g. Matthiessen and Teruya 2024: 222–238), revealing patterns of meaning characteristic of particular texts but also of the registers that they instantiate. As is well-established in systemic functional areas of research and application, text (discourse) analysis is pervasive as a strategy used in addressing all sorts of problems that arise in communities.

These are the important goals for a description. Of course, they include goals that are likely to be internal to linguistics (like providing material for language comparison and typology), or just on the borderline between linguistics and other disciplines or professional communities, e.g. supporting translation and interpreting, supporting language comparison, typology.^[2] But these are goals that are more common across different descriptive traditions than the goal of supporting text analysis. Your description of Dagaare (Mwinlaaru 2017, 2018, 2021, 2025) was really a pioneering description in that while there had been some earlier descriptions of the language, they were not comprehensive, they were not text-based and meaning-oriented, and they were more theory-derived. For instance, Bodomo’s (e.g. 1997a) work drew on Lexical Functional Grammar (LFG), and the parameters or the boundaries of description were set by the vision in LFG of what a language or the grammar is like (but see Bodomo [1997b, 2000] for alternative descriptions of Dagaare).

Isaac Mwinlaaru: When I set out to describe the grammar of Dagaare, my orientation was basically towards a grammar that would be usable for various purposes in the community of the Dagaare speakers, primarily for language education. I also had discourse analysis in mind, and in relation to discourse analysis, the exploration of oral literature, because I had colleagues who were interested in documenting and describing proverbs and folktales. I had the opportunity to know their work (see Kyiileyang 2016; Kyiileyang et al. 2017; Kyoore 2018), but what I had seen was that they were applying traditional literary analysis and stylistic resources developed in the context of Western languages.

What I find missing mostly in such studies are how the resources of the language itself are used in these oral traditions. The following issues are missing, including (i) the use of serial verb construction, (ii) the use of resources of the nominal group and of the verbal group, and (iii) the phonological resources used for the different literary effects. So, when I set out to work on the description of Dagaare grammar, I aimed to provide a grammar that was powerful enough to aid applications in discourse analysis, language education, translation and other areas that may come up. That was my appliable motivation.

Christian M.I.M. Matthiessen: Absolutely. Your point about discourse analysis is very important. If you look around at applications, so many of them have discourse analysis as a way in, with discourse analysis being actually based on a comprehensive description of language, so that the discourse analysis can be both systemic and systematic. You can relate the instances of actual meaning to the meaning potential, as illustrated already by Halliday (1977). That is much more powerful than merely a kind of informed commentary based on an intelligent reading of the text along the lines of close reading or explication de texte (a point indicated already in Halliday et al. 1964). What we need is an appliable description, which then enables us to engage in appliable discourse analysis (ADA) (Matthiessen 2014). By engaging in ADA, we can address a whole host of language-based or language-related issues and problems that arise in a community.

3 Modern language, archaic language and protolanguage

Isaac Mwinlaaru: In the plenary talk you just gave at ISFC-48, you suggested that systemic functional descriptions or descriptive grammars basically focus on modern languages. Then you tried to make a distinction between modern language, archaic language and protolanguage. Would you talk about the different phases of language that you mentioned? In the first place, what is modern language? How is modern language differentiated from archaic language and protolanguage?

Christian M.I.M. Matthiessen: In a way, part of the context is the debate about language universals, but I will save that for a minute and then come back to it. In Michael Halliday’s (1975) study of how young children learn how to mean (i.e. ontogenesis), he identified the developmental phases, and I have used this account of increasing power coupled with increasing systemic complexity, and identified comparable phases in the evolution of language (i.e. phylogenesis), originally set out in Matthiessen (2004a). In Halliday’s account, ontogenesis begins with a pre-linguistic stage, but then learning how to mean starts with protolanguage in around five to eight months of age when the young child starts to construct a primary semiotic system in interaction with their immediate caregivers; this is a small meaning group of people who are very familiar with one another. Then at some point, she or he will move into the beginning of transition into the mother tongue. It is as if the child said: “Okay, I’ve grasped the basic principles of learning how to mean, but I have actually discovered there’s a perfectly good language spoken around me, so why reinvent the semiotic wheel? Let me make a transition into that language instead of continuing to try to construct my protolanguage interaction with others.” Of course, that transition takes time, but it is relatively short precisely because young children have a model to draw on and adopt – the adult language spoken around them.

In applying this to phylogenesis (Matthiessen 2004a), I imagine that the transitional phase, which I now call archaic language, lasts much longer – probably from around 1.8 million years ago (when Homo erectus emerges on the scene in Africa and then migrates to other parts of the world) to around 250–150 K years ago. One reason why it took so long was that there was no model around for speakers of archaic language to use as a bootstrap mechanism. There was no modern language around for our ancestors to draw on, so they had to work this out quite gradually. The transition to modern language probably took place around 150–250 K years ago as anatomically modern humans (AMHs), i.e. Homo sapiens sapiens, emerged out of archaic humans. As I have put it, it is a “package deal” together with Linguistically Modern Humans (LMHs).^[3] I have provided a very rough simplified summary in Table 1. That is based on evidence from various sources, including a spurt in brain development expansion based on indirect evidence. The previous one was perhaps closer to homo habilis two million years ago. That may then have been the transition from protolanguage to what I have called an archaic language. You can reason about this in terms of what the evidence suggested our ancestors were able to do at different points, like coordinated collective hunting, but also fire and what that fire implies in terms of the possibility for dialoguing around the fireplace. That is a hypothesis. The transition into archaic language would have been possibly around two million years ago. Now we can characterize these in terms of the architecture of language.

Modern language as opposed to protolanguage and archaic language differs in being quadri-stratal as opposed to bi-stratal. It is not just content and expression, but both content and expression are stratified. Content is stratified into semantics (meaning or content substance) and lexicogrammar (wording or content form). Expression, on the other hand, is stratified into phonology (expression form) and phonetics (expression substance). That constitutes a more complex but also a much more powerful meaning potential. Part of that meaning potential is metafunctional with simultaneous metafunctional strands in meaning (ideational, interpersonal, and textual strands); whereas protolanguage is micro-functional and the alternative modes of meaning of each micro-function is tied to particular type of context (situation type). Modern language is contextually variable; archaic language is a gradual transition towards further stratification, generalizations of the micro-functions into macro-functions, and then at some point, the macro-functions become simultaneous metafunctions.

People have debated the status of language universals for centuries, and they seem to go in and out of fashion. The second half of the 20th century came to be dominated by the Chomskyan research programme, where universals are given central prominence. However, as the description of languages around the world and both genetics and brain sciences have advanced to the point where biological innateness can be investigated empirically, the status of universals have changed fairly dramatically. There was a very important issue of a journal directed to cognitive scientists in 2009. The lead article was by Nick Evans and Stephen Levinson (2009); they deconstructed the myth of language universals with detailed empirical evidence. They took many of the universals that had been posited for human language since the conception of universals in Chomsky’s (1965) Aspects Model [Aspects of the Theory of Syntax]. According to their study, when you enlarge the set of languages on which generalizations are based, taking all the rich descriptive work that has been done into account, no universals remain. Instead, they suggest that there may be strong tendencies, but these tendencies are not universals (an approach taken already by Keenan 1976).

The position that Evans and Levinson (2009) took was discussed in the same issue of the journal by other linguists. Some supported their argument with further points, and some were critical. When I looked at the critical ones defending universals of the Chomskyan kind, I found the small residue that remained quite unimpressive. One universal that was, and has been, put forward is recursion. One has to be careful when interpreting Chomsky’s notion of recursion, as he has pointed out. But it seems to me that in SFL, we can transcend this particular debate by recognizing the complementarity between logical and experiential modes within the ideational metafunction. Logical recursion is true recursion, involving recursive systems that can engender linearly recursive or iterative structures (cf. Matthiessen 2024). In contrast, “experiential recursion” is actually technically not recursion: it is the opportunistic use of embedding – of downranking units (clauses, phrases, groups) to serve in the structure of other units as if they were of a lower rank (rank shift). When Daniel Everett (2005) points to Pirahã and suggests that there is no recursion, what you find instead is much more use of what we might interpret as logical, linear recursion. For the logical type of recursion, speakers do not embed, but they develop potentially unending tactic complexes (maybe mainly parataxis and possibly hypotaxis as opposed to the experiential mode with embedding), leading to more logical intricacy (characterized by Robert Longacre and other linguists in that broadly descriptive tradition as “clause chaining”). That is the kind of recursion that has been critical.

Thus, we are not saying that there are languages without any “recursion”. It is really a question of how they achieve the effect of “recursion” – logically and/or experientially (but as I have just noted, true recursion is logical, not experiential in terms of rank shift). This has much broader implications in terms of the complementarity of the logical and the experiential mode of meaning. But because linguists outside SFL do not really have this sense of the complementarity within the resources for construing experience, I do not think that has been foregrounded.

Coming back to the contrast between modern language and archaic language, we can reasonably ask: “Is the architecture of modern language universal?” In a way, the short answer is “yes, of course: all languages spoken now and all languages spoken since the emergence of AMHs and LMHs are organized along the same lines: they are quadristratal, metafunctional, and so on”. However, given the baggage that is attached to the term “universal” – or “semantic slime”, as Michael Halliday called it jocularly, maybe it is better to avoid the term “universal” altogether.^[4] As soon as we use the term, we need to anticipate and forestall misinterpretations since there are quite a few senses of “universal”. And long ago, J.R. Firth warned against what he called the “universalist fallacy” (see also Caffarel et al. 2004: 11–12).

Isaac Mwinlaaru: We have situated the description of languages within modern language, under the conception that language has evolved with increasing complexity (i.e. in metafunctional diversity and in stratificational complexity). Micro-functions in protolanguage transition into macro-functions in archaic language and macro-functions transition into metafunctions on modern humans. The content plane of language gradually stratifies into semantics (i.e. content substance) and lexicogrammar (i.e. content form) and the expression plane stratifies into phonology (i.e. expression form) and phonetics (i.e. expression substance). I find the issue of universals particularly interesting.

Christian M.I.M. Matthiessen: The notion of universal is very much dominated by Western Eurasian discussions in philosophy, gaining in prominence during various periods, like the medieval debate about universals and then again during the Enlightenment. I think it is helpful to distance ourselves from this strand in the “Western” tradition so that we view the issues more globally, and draw on the growing documentation of the incredibly rich diversity of human cultures and societies throughout our history (cf. Graeber and Wengrow 2021). I don’t think there can be any doubt that, in empirical linguistics, universalism is now giving way to relativism once again – a reconnection with e.g. B.L. Whorf and J.R. Firth – because of huge advances in empirical research in linguistics (just as in archeology and anthropology, as documented by Graeber and Wengrow 2021). We can note that over hundreds of years there’s a pendulum swing between more universalist and more relativist positions – a continuous succession of thesis and antithesis. So of course it is interesting to ask whether we can transcend these pendular swings, and arrive at a synthesis. I think it is, and I do think the Firth-Halliday tradition has the key to the synthesis. Fundamentally, this involves the metatheoretical distinction between theory and description – between the theory of the architecture of modern language in context (as opposed to archaic language, and by another evolutionary step, protolanguage) and descriptions of particular richly diverse languages in their contexts. I hope we can pursue this line of exploration, but we would need a great deal of time and discursive space.

4 Descriptive categories in language description, comparison and typology

Isaac Mwinlaaru: Our discussion now leads to the issue of the complementarity of description of a single language and language typology. There have been discussions recently, particularly by Martin Haspelmath (2021: 1), who has tried to emphasize that we should distinguish between descriptive concepts and comparative concepts, and between particular languages and human language in general:

In this paper, I address a core foundational aspect of linguistics: the difference between the study of a particular language (spoken at a particular time by a particular community) and the study of human language in general. I argue that this distinction is important both for particular linguistics and for general linguistics, and I note that it has often been unduly neglected. (Haspelmath 2021: 1)

If you pick a category like adjectival or relative clause, it may turn out to be different when you describe a language and refer to the English relative clause or the Dagaare relative clause or the Akan relative clause. The meaning of these terms will be different when we investigate language typology, where we may want to redefine the terminologies for comparative purposes. Do we really have to posit different concepts, different terminologies for description of languages and for typology or comparison of languages? This question is still related to the issue of how universals are related to the architecture of language, but I think within Martin Haspelmath’s framework, the focus is more on structural terminologies – the lower level categories, rather than semantics.

Christian M.I.M. Matthiessen: Let me check one thing first. Does he refer to Halliday’s work from the early 1960s (e.g. Halliday 1966 [1959–1960], 1967a, 1967b, 1968, 1981 [1965])?

Isaac Mwinlaaru: No.

Christian M.I.M. Matthiessen: Does he refer to Halliday et al. (1964)?

Isaac Mwinlaaru: No, obviously not.

Christian M.I.M. Matthiessen: It’s interesting that you say “obviously not”; but sadly for the discipline of linguistics, it’s predictable. If he had, then he would have seen that there has been a concern in what became SFL from the beginning, drawing on J.R. Firth, because Halliday et al. (1964) very much emphasized that these descriptive categories are based on evidence in the engagement with one language. If you want to compare languages, whether it is in linguistic comparison leading to typology or in translation studies, you have to establish comparability. Halliday, McIntosh and Strevens discuss the different ways in which you might establish comparability, translation being one possible way, but they also warn precisely against this. So, that has been quite central in the systemic functional work for the past 60 years. It is interesting and important that somebody in another tradition has come to this point, but it is also quite discouraging that all the work we have done over the decades is simply ignored.

Let me refer again to the distinction between theory and description, i.e. between the theory of modern language (as opposed to archaic language) and descriptions of particular languages, and add to the picture the consideration of the number of languages involved in description: Table 2. The difference between theory and description is one of abstraction: (i) the theory of modern language as a general human system, which emerged around 150–250 K years ago as part of the “modern package” as I put it above – the “architecture” of modern language, and (ii) descriptions of particular languages that are manifestations of this theory, this architecture. In terms of particular languages, we can focus on one language at a time, as we tend to do in descriptive linguistics; or we can take a larger sample of languages as our descriptive domain. In the table, I have included three domains that have emerged in different areas of linguistics – contrastive linguistics, comparative linguistics, and typological linguistics (or language typology). I have arranged them left to right in the table to reflect the increasing number of languages typically involved:

1. in descriptive linguistics, linguists posit descriptive categories based on empirical evidence specific to the language under description – centrally naturally occurring text in context;

2. in contrastive linguistics, linguists tend to contrast pairs of languages in order to identify areas of potential difficulties for language learners based on differences between their L1 and the L2 they are trying to learn. This approach was of course pioneered by Lado (1957) and remained popular through the 1960s. Here the descriptive categories may need to be generalized to accommodate both languages – even stretched; or they may need to be kept separate for each language where they are far apart (e.g. Chinese and English in the area of the grammatical systems for construing the unfolding of process through time, viz. aspect and tense respectively).

3. in comparative linguistics, linguists typically compare a larger number of languages, traditionally in order to establish genetic relationships (the comparative method in historical linguistics; for different types of comparative linguistics, see Ellis 1966); this often involves reconstruction of earlier forms, but linguists have also used mass comparison of items without reconstruction (e.g. Greenberg 2005; Hiraiwa et al. 2017) and of grammatical traits (e.g. Ceolin et al. 2020), using sophisticated clustering techniques;

4. in typological linguistics, linguists compile samples of languages in order to produce generalizations and typologies that apply to large numbers of languages, perhaps by extrapolation to all languages (for databases, WALS: see Dryer and Haspelmath 2013, and now the new resource Glottobank, including Grambank:^[5] see Skirgård et al. 2023); so criteria used in sampling are crucial since they determine the representativeness and reliability of such typological samples.

When we consider descriptive categories that have been posited in the description of particular languages and move from one particular language to more than one, along the horizontal dimension of Table 2, we will obviously find that they are gradually “attenuated”, probably to the point where they are no longer useful because they are not sufficiently alike in the languages we are dealing with. Let’s take the perennial example of Subject (for relevant discussion in SFL of the glossing and description of Subject, and of explanations of certain of its properties, see Halliday 2003 [1992], 2014; Halliday and Matthiessen 2014: 145–150; Matthiessen and Teruya 2024: 302–304). Like all categories (both theoretical and descriptive ones), it is a node in a network of relations along different dimensions – not a thing in itself. When we investigate patterns in a particular language that we may have reason to interpret as “Subject”, we view them trinocularly. As an illustration, I have adapted Halliday’s (2003 [1992]: 202–203 characterization in Table 2).

As illustrated in Table 3 , in the description of English, we can recognize a number of properties of Subject; let me list those in the table and add a few more: modal responsibility (= (i) in the table), target of modulation in the default case, orientation towards topical Theme in the first instance (rather than towards Actor or other participants) – & mediation between textual and experiential (in the system of voice), part of the Mood element together with the Finite and systemic significance of the relative sequence (Subject ˆ Finite vs. Finite ˆ Subject), repeat in the Moodtag. And we can add concord (agreement) within the Mood element with Finite and the realization by ‘nominative’ (vs. ‘oblique’) case in certain pronominals in finite (as opposed to non-finite) clauses, although these properties are not very prominent in Modern English.

As a very rough illustration of the variation in descriptive categories when we move from the description of one language to that of another, I have set out a number of potential properties of Subject as a descriptive category for a sample of languages (English, German, Spanish, [Modern Standard] Arabic, Chinese, Japanese, and Tagalog) in Table 4.^[6]

As Table 4 illustrates, Subject can be interpreted descriptively as an interpersonal cluster category, i.e. as a cluster of different properties (captured in discussions of prototypes in linguistics). But this is actually the nature of all descriptive categories that we posit in the description of particular languages, or in our generalizations across such descriptions. For example, this is brought out in the characterization of the terms in the system of process type in English in Halliday’s Introduction to Functional Grammar: see Halliday and Matthiessen (2014: 354); and Martin (1996) demonstrates how the different process types represent generalizations across a number of properties.^[7] This is of course reflected in the way that they shade into one another (cf. Halliday and Matthiessen [2006] on indeterminacy).

Returning to Subject, we can see that when we compare and contrast the languages in the sample, there is one language that is not characterized as having any Subject properties, viz. Tagalog – as shown by Martin (1990, 2004); Martin (2004: 295) offers the following comment:

My point here is that something that is negotiated in languages like English through the choice of Subject, where modal responsibility in invested, is handled differently in a language where the structural function Subject is very difficult to motivate on interpersonal grounds (Schachter 1976, 1977, 1996; Martin 2004: 295)

The languages other than Tagalog in the sample all have some cluster of such properties, according to my illustration here. Thus, they all operate with one interpersonal element of the structure of the clause assigned modal responsibility – the modal element called “Subject”. This is reflected in a number of different ways, e.g. in clauses of ‘imperative’ mood type realizing commands, where this is the element held modally responsible for complying with the command (or refusing to), as in Be guided by the experts! (Interestingly, this is also reflected in ‘imperative’ clauses in languages with ergative case-marking (alignment) systems according to my interpretation of Dixon [1994].) Uniquely among the languages in the sample, Japanese Subjects have a related property: they constitute the one element in the clause that is interpersonally “elevated” such that they play a role in the system of honorification: see Teruya (2004, 2007).

Some of the properties characterize the Germanic languages in the sample, i.e. English and German; they are only languages here where I think positing a Mood element is useful and justified, one piece of evidence being the deployment of the relative sequence of Subject and Finite within the Mood element used in the realization of the distinction in mood type between ‘declarative’ and ‘polar interrogative’, and in English of course also patterns of clausal ellipsis (cf. Teruya et al. 2007) and the “mirror image” of the Mood element in the Moodtag. This is in fact quite an exotic pattern around the languages of the world, largely confined to Germanic languages (and reflected to some extent as one realizational option in French: see Caffarel 2004, 2006).

In English, German and Spanish, there is concord in finite clauses between the Subject and the Finite part of the verbal group (whether it is separable as a Finite element in the clause or not); and we find concord in (Modern Standard) Arabic as well, but the patterns are quite complex. In Chinese and Japanese, there is no concord; in fact, Chinese hasn’t got a distinction in finiteness at all (e.g. Halliday and McDonald 2004).

In German and Spanish, there are so-called “dative subjects” (in clauses of mental or physiological experience), but not in Modern English – although this is a well-known phenomenon found in a variety of languages (some illustrated with references in Matthiessen 1995). The use of the dative case here instead of the nominative can be characterized experientially.

If we describe particular languages using system networks, we can then take one step further and unify these descriptions using multilingual system networks (e.g. Bateman et al. 1999; Matthiessen 2018). The description of each particular language is justified intra-linguistically based on empirical evidence drawn from authentic texts in context in the first instance, so when these descriptions are unified or mapped onto one another in a multilingual system network, we know the status of the descriptive categories, and we can use multilingual system networks to raise questions about the degree to which certain descriptive categories in one language are similar enough to those of another to justify treating them as the “same” category.

Do we need separate descriptive categories for each column in Table 2? As long as we are aware of the issue of how much we have to posit to refer to something as the same category, we are aware of the differentiation between descriptive categories and theoretical categories. It may be helpful to establish a further notion, a kind of descriptive generalization – descriptive categories, but maybe that is not necessary if you have this clear understanding of the basis of comparability and of the distinction between descriptive and theoretical categories. Haspelmath operates in another metalinguistic universe, as it were, and this clear distinction between theory and description does not seem to be part of it. He came from a tradition where they did not problematize the question of comparability about 60 years ago onwards.

Another linguist in the Firthian-Hallidayan tradition who wrote about language comparison is Ellis (1966, 1987). I do not think Haspelmath has referred to Ellis either (certainly not in his 2021 article). I have read Haspelmath too and he does seem to have an important sense of the need to “weed the garden” – to make sure that things are clear in terms of proceeding, which is very laudable and welcome. However, coming from the Firthian, Hallidayan and the systemic functional traditions, we already know the need for this kind of clarity – it is a central part of our upbringing and has been part of how we proceed. So we must consider the question of dialogue within the discipline of linguistics. But to have a dialogue, the potential dialogue partners must actually be interested in other traditions and positions and in reading their work – i.e. going beyond their own tradition or traditions and engaging with speakers of other metalanguages. Just over 20 years ago, Haspelmath (2000) wrote an article, entitled “Why can’t we talk to each other” – about why it should be so difficult to agree with scholars from other traditions; he wrote it as a review of Newmeyer (1998), so in a sense Newmeyer set the agenda in his discussion of form and function, but, in view of his title, it is still remarkable – and arguably quite telling – that Haspelmath does not mention SFL.

Isaac Mwinlaaru: I think it is a response to the generative literature.

Christian M.I.M. Matthiessen: Yes indeed, that was the context – Haspelmath’s review of Newmeyer (1998). I would just like to add one point directly relevant to the discussion of categories. In SFL, language is viewed as a resource for making meaning – a meaning potential, so language is theorized as such; it is theorized systemically in the first instance rather than structurally as a vast paradigmatic network of options in meaning. Similarly, descriptions of particular languages are systemic, and comprehensive descriptions are thus vast system networks. Therefore, every category, whether descriptive or theoretical, is a node in such system networks, related to other categories; its nature is relational, and it is interpreted in terms of its valeur in the system. Thus, you cannot interpret it just in terms of the glossing of it as a “thing in itself”. If you posit the category of Subject, its nature inheres in the relations it enters into; and this is difficult to capture in a single gloss or definition, as Halliday (2002 [1984]) highlights in his discussion of the ineffability of grammatical categories; so we approach it triconularly, as illustrated above in Table 2, showing that it is a “node” that enters into different relationships. In Halliday and Matthiessen (2014: 147–150), we outline what a “thick description” of Subject in English needs to cover, including the trinocular vision of it based on the different relevant dimensions (axis, rank, stratification, metafunction, instantiation). Here it is important that in SFL, Subject is a descriptive category, so for each language under description if we posit it, it must be justified empirically, and it is not the starting point of the description: it will be introduced if it turns out that the interpersonal grammar of the language warrants it: see Matthiessen and Teruya (2024: 302–304), and references therein to systemic functional descriptions. The relational network of theoretical and descriptive categories as nodes in networks is not foregrounded in the same way in the tradition that Haspelmath is part of.

Incidentally, the point about the relational nature of categories came up in the last few days during ISFC-48 in Tunisia, in particular when participants gave presentations on nominalization, typically process nominalization in Modern Standard Arabic and Tunisian Arabic. Here I think it is important to recognize the number of strategies in Modern Standard Arabic of construing experience in the environment of non-finiteness. It is interesting to observe that if we just take what seems to be the process of nominalization, we will then just map it onto what we understand in English or other languages. But in modern standard Arabic, there is no direct equivalent of the infinitive in English, so you really have to see it in terms of the whole set of resources of construing things in a non-finite environment.

Isaac Mwinlaaru: One crucial distinction will also be the orientation of comparison. Maybe for most of the comparative notions Haspelmath (2000: 663) mentioned (e.g. “adjective, passive voice, accusative case, future tense, second person, subject, affix, clitic, phrase, WH-movement”), they are structural categories. If we focus the comparison from the orientation of systems as has been the agenda in SFL description of languages, then we can see how different languages may realize different systems structurally.

Christian M.I.M. Matthiessen: That is a very important point. That makes me think of work that goes back even to the 1980s in the context of translation studies: Erich Steiner (1988) used the notion of diathesis as a way of exploring strategies in languages for different thematic informational assignments (cf. Ellis 1987). It may be through a system of voice. It may be through a system of theme that is not linked to voice. We need to be aware of the extent to which we have a basis for comparison and of the extent of over-extending a category like voice, as pointed out already by Halliday (1966 [1959–1960]). For example, both “voice” and “Subject” have been misapplied in the description of Tagalog and other Philippine languages, as shown by Martin (2004) – and in the discussion of “Subject” in the typological literature, Schachter (1976, 1977, 1996) had already problematized its status, as Martin has pointed out.

I was very lucky to have Paul Schachter as a teacher at UCLA, including during a one-year course of field methods. I’m sure that if he’d been fully aware of Halliday’s theory of metafunction, he would have used it in his work problematizing the status of subject in Philippine languages – as Elizabeth Traugott did insightfully in her work on the stages of grammaticalization, having met Michael Halliday when he was at Stanford University. After all, Paul Schachter took notice of Richard Hudson’s (1976) version of systemic functional grammar from the second half of the 1970s, i.e. “Daughter Dependency Grammar” (DDG). His interest in and even championing of DDG grew out of his honest intellectual curiosity. It was certainly not a move designed to promote his standing in US linguistics. I remember visiting MIT in 1979 on my way to UCLA, and Jay Keyser kindly took time to see me. He gave me a rapid profile of linguistics at UCLA, noting that a very good linguist, Paul Schachter, seemed to have derailed, taking an interest in DDG; so he recommended a linguist in a language department at UCLA (I remember – I hope I’m not mistaken – a photo in Keyser’s office of him and fellow faculty members in sports gear as members of a sports team, including the labels “Killer Keyser” and “Mighty Morris” [Morris Halle]. This suggested to me a supportive atmosphere of intellectual comradery. When I visited, probably in August 1979, Noam Chomsky was not around; Keyser told me that he was at his vacation home in Cape Cod).

5 Theory as a guiding grid in language description

Isaac Mwinlaaru: The point I was trying to raise is the notion of “theory as a guiding grid” in language description. Could you explain what it means?

Christian M.I.M. Matthiessen: The systemic functional theoretical architecture of language in context can be thought of as a grid or template or blueprint that guides us as we develop descriptions of particular languages – another related lexical metaphor that seems to me to have become more common is that of “lens”, the lens or really array of lenses through which we view the phenomena under investigation. The architecture defines the overall semiotic space of language in context, so it shows us where we need to do descriptive work – globally, covering all the strata of the content (i.e. semantics and lexicogrammar) and expression (i.e. phonology and phonetics) planes and the spectrum of metafunction within the content plane; and accumulating instances of texts in their contexts of situation at the instance pole of the cline of instantiation, inducing or distilling patterns in this textual data set to move towards systemic generalizations.

There are ways of displaying the overall architecture or parts of it that bring out the “grid” – viz. matrices (As far as their use in observing data goes, we have used an analogy with Albrecht Dürer’s famous drawing machine; see Matthiessen and Teruya 2024: 182 for example) Globally, we have the function-instantiation matrix and the function-stratification matrix; and locally, within the content strata, we have the function-rank matrix, and the phonological domain (prosodic/articulatory)-rank matrix for the stratum of phonology within the expression plane (obviously with analogical matrices for graphology and sign). The function-rank matrix simply locates the systems in terms of metafunction and in terms of rank. As a way of getting started, we can seek inspiration in function-rank matrices of well-described languages (see e.g. the function-rank matrices of Chinese [Halliday and McDonald 2004], Japanese [Teruya 2004, 2007], English [Halliday 2005 [1970]; Halliday and Matthiessen 2014]). We can of course sample such matrices systematically and try to identify systemic motifs and produce generalizations, e.g. trying to distinguish between systems that are highly likely to be useful in the description of one language after another like theme, mood and transitivity and systems that are much more variable, like tense and aspect, which may occur in different combinations.

But at the same time, it is actually also helpful to remove the systems located in the matrix of a particular language altogether, so that we can focus on the dimensions of the metafunctional spectrum and the compositional scale – rank, where importantly you have to allow for variation. For example, in the division of labour across the ranks, is there a rank below the word? That is a typological variable. How much work is done at the rank between clause and word? That is also a typological variable, and it is variable within the history of a language. But in any case, an “empty” function-rank matrix will invite us to explore the lexicogrammatical system of a particular language in terms that are abstract enough that we do not foist descriptive categories based on the established description of other languages. At the same time, it’s an invitation to look upwards for discursive guidance – at the instance pole of the cline of instantiation, where we find “data”, viewed as texts (semantic patterns) in contexts of situation. For example, we can examine texts instantiating narrative registers, and ask how the episodes in the narratives are “chunked” into grammatical units or unit complexes, and also ask how dramatic dialogue is constructed.

When you begin to collect naturally occurring texts in context, or when you begin to elicit examples from yourself if you are a native speaker or from a language consultant, then you can explore fundamental architectural characteristics, e.g. what seems to be the maximal domain in the grammar – the clause (complex) or something else? You can explore this issue interpersonally, starting with dialogue, with fairly quick turn taking. Or you can begin to explore it experientially, starting with texts that seem to construe fairly different kinds of experience. So, you develop it in this way, guided by what you can expect to find in a language, without positing descriptive categories borrowed from the description of other languages such as Theme or Subject or Actor, Process or Predicator or anything along those lines. Such categories may emerge as generalizations you can induce from the patterns you observe in text, but they are not the starting point in the description of a given language. We posit them based on patterns that we observe, guided by the “empty” function-rank matrix but also of course by the motifs and generalizations that I referred to earlier (cf. Matthiessen 2004b).

Isaac Mwinlaaru: Actually, I found the function-rank matrix a powerful tool in my own description. I often see it as a kind of a roadmap to the description as I proceed with the analysis of the language and try to see what roads are there in this language. Of course, I start the analysis with hypotheses I develop based on typological guidance and transfer comparison. That is, I arrive at the hypotheses based on available descriptions of other languages, especially genetically and typologically related languages to the target language. The hints for these hypotheses are however ultimately provided by my preliminary examination of the data I have. If I am able to proceed this way, I view the ranks against the metafunctions, then I begin to look at every rank very closely and see what systems of wording for creating meanings are emerging. That can take a long time. In my own work (see Table 5), I have always had to revise the function-rank matrix, removing some systems and adding others over and over again as I proceed with the analysis of more data. But then it becomes, as I said, a roadmap for the description and allowing for comprehensiveness. You posit some systems, you test them, they may not be tenable in the language, or they may belong to other bigger systems, and then you have to merge them. In Table 5, I have provided alternative labels for some systems for Dagaare – countability (=number) and reduplication (=emphasis) – because at this point of my analysis of the language, these systems are still being explored and further insights are required to determine what systems are actually at work.

Christian M.I.M. Matthiessen: The summary of your descriptive journey using the function-rank matrix as a roadmap is very instructive. The point you made regarding the function-rank matrix also encourages your description to be comprehensive because you need to consider all cells of the matrix.

Isaac Mwinlaaru: Yes, of course. It is like this because at every rank of the language under description, you want to exhaust the systems as much as possible based on the data you have. We are able to test this, because once you posit a system, you have to develop a system network for it, look at the structural realizations of features or terms in your system network and so on. We also make sure that every systemic feature in the system network has a realization in syntagmatic patterns. In this sense, we are guided not just by the system, but also by the structural realizations. Once we do this, we normally will do some kind of discourse analysis of the texts we are working with to see how well the system applies to data in the language, and to see whether the system is something productive or something that needs revising. We keep adjusting the system as more data accrue and reveal new insights until we get to a saturation point, a point where no new patterns are emerging from new data.

Christian M.I.M. Matthiessen: That is crucial because we can consider deductive and inductive moves along the cline of instantiation. Deduction means you have a description of the potential, and then you look for examples to illustrate or to test it. Inductive means you move from the instance towards the systemic potential. Deduction and induction are reflected in a distinction made in corpus linguistics, viz. between corpus-based versus corpus-driven (e.g. Tognini-Bonelli 2001; for comments, see e.g. Halliday and Matthiessen 2014). But I think that overlooks precisely what you said. You have the emergent description of the system, and you test it against text by means of systemic text analysis, i.e. text analysis based on the emergent description of the system. It seems to me one could call that an abductive approach, taking the term from C.S. Peirce.

6 Shunting along the semiotic dimensions

Isaac Mwinlaaru: Your discussion here also leads to the term shunting, which came from Michael Halliday in 1961 (see e.g. Halliday 1961, 1978, 2002 [1996]; Matthiessen 2007). What is shunting? Is it just shunting between different ranks? Can shunting go beyond ranks in language description?

Christian M.I.M. Matthiessen: I would say yes. For Michael Halliday, shunting came from his love of trains and railway systems. It is a movement along any of the semiotic dimensions that are the domain of relations. You can shunt in terms of rank. You can shunt in terms of axis (i.e. paradigmatic and syntagmatic). You can shunt across the metafunctions. You can shunt stratally up and down (across semantics, lexicogrammar, and phonology). You can shunt along the cline of instantiation (e.g. from the meaning potential through register and text-type to text): you make it a bit more systemic, you test the data, and you look for further instances. You can shunt in terms of delicacy. So, any dimension that is part of the architecture – the multi-semiotic space defined by the intersecting dimensions (Matthiessen 2007, forthcoming) is open to shunting.

It goes with the trinocular vision (e.g. Halliday 1978; Halliday and Matthiessen 2014): viewing any category “from below”, “from above” and “from roundabout” in terms of the different semiotic dimensions, as illustrated for Subject in Table 2 above – i.e. in describing a category in a language, you shunt along the relevant dimension to adopt these different vantage points. The danger in description is to get stuck in one place without shunting, so shunting is to be encouraged. Another related danger to rely on the gloss of a particular category, like Theme or Subject, instead of viewing it as a node in different relationships along different semiotic dimensions. Witness, for example, the long list of failed attempts in “traditional grammar” to define central categories like sentence, noun, verb, subject, predicate; cf. Halliday (2002 [1984]).

Isaac Mwinlaaru: I have also realised that shunting allows the descriptivist or analyst to be more critical of their own analyses and of themselves, so that they can always be in constant doubt and try to test items further. For instance, if you are working with a system of the nominal group, then you want to move to the clause rank and see how this system functions within the unit of the clause. Or you want to investigate further, to find out whether there are other realizations beyond those you have identified at the group/phrase rank. For example, tense is realised by the finite element in the verbal group but it can also be realised below the group by verbal morphology. If you are dealing with a system network, you want to see how a feature/term (e.g. ‘future’ in the system of tense in Dagaare) relates to other systemic oppositions in the network (e.g. ‘non-future’) and how two systems such as Dagaare polarity and tense are related to each other in terms of common or shared realizations (see Figure 1). I see it as a tool for reflection and also for being self-critical, leading to more accurate descriptions.

Figure 1:

Dagaare polarity and tense represented as a system network.

Christian M.I.M. Matthiessen: Absolutely! You could also envisage interpersonal meta-shunting. You shunt from yourself to a colleague who is also in a position of experience.

Isaac Mwinlaaru: Yes. That is important. I have also been asking other native speakers. Sometimes I would check with a friend or a relative who is a native speaker of Dagaare and see whether there is agreement with my interpretations. With appropriate hints, you can normally find relevant information from (other) native speakers.

Christian M.I.M. Matthiessen: Indeed. This raises another interesting methodological point. Because of the Chomskyan generative approach, there is a kind of naïve understanding that anybody can make grammatical judgments; this was certainly the general view when I grew up linguistically in the 1970s. But in fact, this is not the case; empirical linguists have noted that speakers will exclude examples as ungrammatical even though they may themselves produce them under natural conditions, as in spontaneous casual conversation (cf. Halliday 2005 [2002]). When we are trained as descriptive linguists, we actually need to increase our ability to imagine possible co-texts and contexts for examples that might at first seem “ungrammatical”: they may be highly marked and/or highly infrequent, but these are different judgements. If you are being trained as a linguist, you are probably likely to expand your tolerance, not because you are being worn out (although that can also happen!), but because you are better able to imagine possible contexts and also to become more critical and self-aware, so that you are less likely to rash in your judgements. Also, we can ask to what extent grammars have evolved to support their speakers in making such judgements: I remember Martin Kay making this point at an invited talk at the Information Sciences Institute/USC in the first half of the 1980s – the background being his long-term project of developing a grammatical formalism (at the time called Functional Unification Grammar) to be used in computational linguistics for both generation and parsing (e.g. Kay 1979). He emphasized that grammar has not evolved to be used to make grammaticality judgements.

7 Structuring a grammar

Isaac Mwinlaaru: In your ISFC-48 presentation, you cited The Art of Grammar: A Practical Guide (Aikhenvald 2015). By the way, I find her many publications impressive contributions and educative. You indicated that at the beginning of her book, she discusses how to structure a grammar, starting from morphological and phonological issues, and then gradually building towards all of grammar. What would be your own position on structuring a grammar? How can we structure a grammar? What would be the effective way of presenting a grammar to our readers?

Christian M.I.M. Matthiessen: In my presentation, I talked about the structuring of grammar and the procedures for discovering it. The term “discovery procedure” was introduced as a characterization of Neo-Bloomfieldian descriptivism by Chomsky, and then rubbished by him and his followers.^[8] But Longacre (1964) bravely wrote a book on discovery procedures; and we can review the methods by him and other descriptive linguists of that era in view of the development of machine learning techniques and the use of neural networks: see e.g. Goldsmith’s (2006) long-term project on the description of morphology based on large volumes of text. I think we can recognize this as part of the methodology of how to discover and how to develop a description of a new language. This approach is, naturally, different from how we organize the description that we developed using “discovery procedures” or other methods (cf. Halliday 1961):^[9] normally, we organize descriptions to bring out the functional organization of the resources of a language, not to document our journeys of discovery (although that is of course also possible, say for pedagogical reasons in a manual designed to be used in training descriptive linguists).

In her account of the organization of a descriptive grammar, Aikhenvald (2015) starts with something about the background of the language, the cultural setting and so on. Then she descends into the phonology and the morphology. I have diagrammed her recommendation in Figure 2, interpreting it in terms of SFL as moves along the hierarchy of stratification and within the stratum of grammar (lexicogrammar) along the rank scale. I have drawn the diagram so that we can take different considerations into account. In a systemic functional description, I would certainly also prominently foreground the function-rank matrix: the spectrum of metafunction across the ranking domains, and along the paradigmatic axis, the cline of delicacy. The general challenge is of course how to present a linear description of a phenomenon, the resources of language in context, that is not linear but multidimensional. Here we can also learn from alternative descriptive sequences used by systemic functional linguists in their descriptions of different languages (e.g. contrast the organization in Halliday’s IFG with that adopted in my LexCart ^[10]).

Figure 2:

Aikhenvald’s (2015) recommendation for the organization of the description of the grammar of a language (“how to structure a grammar”) interpreted systemic-functionally in terms of the hierarchy of stratification and (within grammar) the rank scale.

In terms of a systemic functional presentation, it is very important from the start to give a sense of what kind of phenomenon grammar is: it is a resource for creating meanings as wordings. If you do not provide that sense, you have not been able to lead your readers into an understanding of the grammar, because the grammar is descriptive, but a comprehensive description of the grammar as a resource also enables us to explain it (cf. Halliday 2014). Explanations may be external, e.g. the view “from above” relating to context, but they may also be internal in terms of the view “from roundabout” (i.e. from the grammar itself). The different ways in which the concerns, the systems, the metafunctional systems are all mapped onto one another and are unified. There are different possibilities here (as illustrated in Matthiessen 2004b).

We should adopt an ecological approach to language, in the sense of a contextual one, where we view language in relation to its semiotic environment. Just as scientists developing ecological thinking began to theorize ecosystems in the 1930s, thus moving even further away from the treatment of biological organisms as specimens in a museum only analysed in terms of their morphology, we need to approach language within its semiotic ecosystem – which is the contextual view initiated in the 20th century by Malinowski, developed by Firth and interpreted semiotically by Halliday.

Depending on the language, you may need some kind of preamble just to give people a sense of what you need to understand examples in terms of phonology and in terms of morphology. It does not mean that it is the full-blown account of the phonology or the graphology before presenting a comprehensive description of the content plane, or at least of the lexicogrammar. This is in line with what Aikhenvald (2015) suggests, i.e. we start with the general cultural context, then we begin to sketch the description of the language, quite possibly including phonological (and if relevant, morphological) characteristics readers need to make sense of the description of the lexicogrammar, including the interlinear glossing and analysis of examples. But then it will almost certainly make sense to return to a more comprehensive description of the phonological resources against the background of an understanding of how they are deployed by the lexicogrammar (this is what I mean by a functional approach to the phonology: Matthiessen 2021, 2023b). I have a vision of a linguist’s workbench, and such a workbench would include the “infrastructure” for storing linguistic description and for navigating through them in different ways – i.e. not confined to the linear technology of a book: see Matthiessen (2023b: 146–149); and we can compare this vision with Bateman’s (e.g. 1997) KPML system for describing computational systemic functional grammars.

8 Mistakes that tend to be made

Isaac Mwinlaaru: This now leads us to the grammarian’s work itself. As we mentioned, the presentation of grammar is different from description. When you are presenting, you are thinking about the readers; but when you are going into the language, you want to find the best way in into the language. When I started to describe the grammar of Dagaare (Mwinlaaru 2017), I was new to this research area, so it was quite confusing for me. But my initial training was in applied linguistics and that is the context I come from, so the immediate way in for me was text, which I think is compatible with SFL. Different grammarians may have different ways into grammar. The way you enter into grammar may end up influencing the kind of discoveries you make in the language. You have worked on Akan and some other languages (e.g. Matthiessen 1987, 2004b; Mwinlaaru et al. 2018). Also, through your supervision of students, they might have adopted different approaches at the beginning. What are the effective ways of entering into grammar in terms of starting the description? What are the mistakes young scholars often made that can be avoided?

Christian M.I.M. Matthiessen: Whatever the enterprise is, it is always useful to have some exploratory phase – a kind of pilot study. It is interesting to entertain the notion that we could take as a pilot study something like Pike’s monolingual demonstration in front of an audience of exploring the “basics” of phonology and grammar. Now, the method does not have to be monolingual, but Pike’s monolingual is a thought-provoking, stimulating example, and Daniel Everett and one or two other linguists have maintained the tradition. Pike’s monolingual was designed as an illustration of how you can begin to engage with the language as a linguist. It typically lasted an hour and a half. He was presented with a speaker of a language, knowing nothing about the identity of the language or the speaker, and then he would use a language that he knew, Mixtec, to make the interaction verbal as well as non-verbal. He would use gestures, pointing to concrete objects and so on. That is why it was called a monolingual. He demonstrated convincingly that it was possible to develop some initial sense of the language in an informed way.

Interestingly here, I was lucky enough to be in the audience for one monolingual. Pike was in his 70s by then. In any case, he revealed different facets of the language. Moving in “from below”, he captured aspects of the phonology. He was able to make certain guesses about the sound system, but also some other aspects of the language, maybe a little bit of morphology if it is a language with elaborate morphology, a bit of syntax and so on. His monolingual was based on elicitation in the sense that he was not observing language unfolding as text in a natural setting, but it seemed quite useful as a way of first familiarizing oneself with the language.

Depending on the conditions when one sets out on a descriptive journey, one might try something like that. But it depends a great deal on the circumstances: if you are a native speaker yourself, do you look in the mirror and pretend? – you don’t, of course. But something like a Pikean approach can serve as a warm-up. Alternatively, you may have the opportunity to observe language users for a while in natural contexts, allowing yourself to get an initial impression of the language by “eaves-dropping” and gradually attending to patterns that seem prominent to you. If you approach a language that is well-described, of course you go to the secondary sources – the literature. If you have a language that has only been described in some field notes, then of course you adjust accordingly (There are many languages only characterized by means of sketchy descriptions. Field notes may be based on work by a linguist observing a language for an afternoon or a day. There are also languages that have not been described at all. Returning to Grambank, it is interesting to note that the developers have been able to include a little over 2,400 languages, which is impressive and gives an indirect indication of the frontier of the description of languages spoken around the world).

In terms of using primary sources, you move in from two angles – text collection and elicitation. You start collecting texts, looking for contexts of situation where you are likely to get texts that are accessible – ones that lend themselves to exploratory description. So, you do not start with esoteric texts, highly theoretical texts, highly stylized ritual texts and so on. You try to start with texts that are likely to be naturally simple as a way in, e.g. ordinary dialogue, possibly with fairly young children being involved in the dialogue.

The way in may be through the interpersonal, just to get a sense of how people interact using the grammar of exchange. At some point, in principle, you postulate and then probe the paradigm of intersecting interpersonal options. You begin to elicit examples according to some guesses as to what might be relevant distinctions, and importantly then intersecting them. With the help of a speaker of Kannada, which is a Dravidian language spoken in South India, I did some exploratory fieldwork. At some point, I needed to probe the interpersonal grammar, so I produced a simple intersection of polarity and mood type, and was able to elicit a paradigm of the combinations of systemic options. Of course, I might have had to adjust my assumption that these two systems are simultaneous; in fact, one reason for probing paradigms is precisely to determine whether systems are simultaneous or ordered in delicacy (If I had worked on Mandarin, I would hopefully have realized at some point that while the systems are simultaneous, the realizations of ‘negative’ polarity are conditioned by mood type, process type and aspect. And if I had worked on Modern Standard Arabic, I would have found even more complex conditioning of the realization of ‘negative’ polarity. Here it is very helpful to be able to draw on a library of descriptions of various languages, and descriptive generalizations based on them). Naturally, we will need to interleave the analysis of natural texts that we have sampled and the probing of paradigms by means of elicitation. As Michael Halliday has said (e.g. Halliday 2005 [1998]), people do not speak in paradigms, so you can infer the paradigms from succession of terms in dialogic moves. But at the same time, we also need to proceed using elicitation, in order to probe the paradigms we begin to postulate.

Isaac Mwinlaaru: When we talk about text-based descriptions, people tend to confuse description with discourse analysis. With discourse analysis, you already have the system – the resource available, and then you apply this. But if you are describing a language and it is the first time to discover a system, it is still discourse-oriented and you may have to start with lower-level items such as function words and affixes and identify the patterns of all the grammatical systems of say the nominal group or some other grammatical class of interest. If you have an inflectional language – for example, the “noun” class system in Niger-Congo languages – you may want to look at instances of nouns and look at the different inflectional patterns; you may want to group the nouns into morphological paradigms, then you want to see the shared meaning you can induce from each paradigm.

In Table 6, I have provided a paradigm of the system of thing type (“noun classes”) in Dagaare (Niger-Congo: Mabia/Gur). “H” means “human” and “NH” means “non-human” and I have excluded proper nouns and non-count nouns because these are not marked morphologically in Dagaare (see Mwinlaaru 2023 for details). The table presents the system from the point of view of meaning (type of thing) and then the morphological realisations are specified and illustrated for both singular and plural nouns in the classes. But before we get to this point, the data analysis starts from the other way round: you first observe nominal items occurring in nominal groups and compile them into their morphological paradigms, or in some instances morphophonological paradigms. It takes some time to sort nouns into the appropriate paradigms and some nouns may take different morphological forms in different contexts and all these variations must be considered and sorted out. As you proceed with this kind of sorting, you also begin to deduce the semantic motivations for the morphological groupings. Thus, while the initial consideration of classification will tend to be formal as the paradigms form and become clearer, semantic criteria come into the picture and work simultaneously with the formal criteria. In Table 6, you will also realise that items in No. 5, 6, and 9 belong to the same morphological paradigm by taking on a similar suffix (-rɪ/ri), but they split in terms of semantic and phonological criteria as well as in terms of differences in agreement patterns in the nominal group. This split is ultimately due to borrowing, which is an external factor that perturbs the system (see Mwinlaaru 2023, for details). So this is explainable. But we need to sort out the items from different points of view to be able to explain the patterns.

Again, I like the notion of shunting. As you are looking at the forms, you have the meaning in view, and at the same time you want to see how these are functioning in discourse or in context. That can be a very interesting exercise.

Christian M.I.M. Matthiessen: Indeed, if you look at Pike’s monolingual, he would begin to hypothesize paradigms, probably phonological paradigms. Given the development of linguistic T-&-T (i.e. tools and technology), it would be very useful if you have a fairly quickly compiled corpus that is reasonably sizable. You can also expose your corpus to machine learning techniques, so as to give you a suggestion for patterns to be explored further. Particularly if you have texts that are likely to be naturally fairly simple lexicogrammatically, you will then get some preliminary suggestions. For example, John Goldsmith (2006) has worked with a morphological analyser developed over the last 15 years or so. That can produce certain discoveries of morphological patterns in different languages. If such tools and techniques are available and feasible, it is certainly a good complement to what you can do as a manual analyst.

Isaac Mwinlaaru: I especially like Michael Halliday’s metaphor – walk before you run. In terms of mistakes made by junior scholars, what mistakes do you think people tend to make before they are ready to run?

Christian M.I.M. Matthiessen: One example would be to try to begin to explore texts that are very pregnant with ideational grammatical metaphor or texts that deal with a lot of abstraction such as scientific journal articles or sacred texts that are far into adolescence or early adulthood. Maybe such texts are of particular interest to the descriptivist, but one must start with naturally occurring simpler texts. Again, if you can find texts that are naturally simple rather than edited or elicited, that is good too and that is why I come back to texts for (or by) fairly young children (but not so young that have not mastered the basic resources). But instead of starting with such texts, many will make the mistake of starting with “adult” texts because they are of interest in terms of considerations other than language description.

Another arguably related mistake is linguists start with some quite specific area before they have a sense of the overall system. They may transfer descriptions from a well-described language, e.g. English, Spanish, Chinese, or Japanese. But another major mistake or temptation is to ignore earlier descriptions of the language. If you have a good coverage in secondary non-systemic functional sources, then of course you make the best of it. You make as much use of it, you get as much mileage out of it as you can.

When I find good non-systemic functional descriptions with reasonable coverage (as in “comprehensive” or “reference” grammars), I create a function-rank matrix and then I insert the sections, subsections, chapters that I can locate in this description. This step will immediately give me a sense of what the degree of coverage of different areas and maybe major gaps that are simply not covered. Again and again, I have found such gaps in coverage even in reference grammars – gaps that can be broadly characterized as gaps in metafunctional coverage and in systemic coverage because these aspects of the grammar are not theoretically visible to the linguist who has produced the description; they are simply not on their descriptive agenda. Very often their angle of approach to the grammar is still dominated by the view “from below” as a way in, as in “traditional grammar”. And if the description of the grammar is not seen as an integral part of the overall description of language in context, then the linguist won’t be “invited” to adopt a view “from above”. To illustrate this point, I contrast the linguistic domains “visible” to traditional grammar versus systemic functional grammar in Figure 3. This is obviously not to claim that traditional grammar in the Graeco-Roman tradition is not a remarkable achievement of ancient and medieval scholarship – it clearly is, and the same applies to the great Arab-Islamic (e.g. Owens 1988) and Indian traditions (e.g. both Pāṇini’s description of Sanskrit and the Tolkappiyam description of Tamil);^[11] but they are like pre-Newtonian physical theory – before the breakthrough to modern science.

Figure 3:

Phenomenal domains of language “visible” to traditional grammar versus systemic functional grammar.

Then at some point, it is important to try to liberate oneself from existing descriptions, and perhaps even from the whole descriptive tradition that they are part of. This will be particularly important if you are working with languages that have a long rich tradition of descriptions, e.g. Greek, Latin, Arabic, Tamil or Chinese. Once you have “harvested” the descriptive insights from such a description, set it aside and begin to explore the languages systemic-functionally based on a registerially rich and varied corpus of texts in context.

In descriptions informed by traditional grammar (or more recent “mainstream” derivatives), linguists tend to be constrained by the traditional descriptive categories, and they are likely to approach grammar syntagmatically from below – that is, from the combinations of axis: syntagmatic (rather than paradigmatic) and rank: morpheme/word (rather than clause). Informed by systemic functional theory and meta-theory, their constraints are perfectly understandable although I would argue strongly that they are limiting: like traditional grammarians, they are very likely to focus on what is easy to observe – the more exposed aspects of the language under description.

9 Registerial consideration in text collection

Isaac Mwinlaaru: Regarding your long-term project on registerial cartography, which you have proposed for collection of texts and building a corpus or text archive for the description of languages (e.g. Matthiessen 2015a, 2015b). Why is it important to focus on registerial variation when describing a language?

Christian M.I.M. Matthiessen: That is a crucial question (see also Matthiessen and Teruya 2024: 208–221). I will say something about it and then ask how you have deployed this in your work on Dagaare. All the contextual parameters, i.e. field, tenor and mode, are relevant, but to address various tasks, I have started with field of activity – with what’s going on, the socio-semiotic activities. If you approach language from above, i.e. from context, you can have certain expectations as to what aspects of the grammar are likely to be “at risk”, to be doing most of the semiotic work. To return to what we just talked about, based on the accumulation of correspondences between contextual values and choices in linguistic systems within the different metafunctions that we have observed in the study of a growing number of languages, we choose texts instantiating registers that are likely to illuminate areas we have set out to describe:

1. If you are interested in the interpersonal grammar, clearly it makes sense to start with dialogic rather than monologic text – fairly casual dialogue with short turns, so that the dialogue will bring out basic properties of the interpersonal grammar of interaction.

2. If you are interested in the textual grammar, then you choose texts from registers that are likely to have quite dramatically different “methods of development”, so you immediately get a contrast in choices in the textual grammar.

3. If you study the experiential grammar, then you choose texts that are likely to have different characteristics in terms of the kind of experience being represented. Is it very fluid, volatile, or is it sort of a verbal picture of something fairly static, like a taxonomy or a verbal of a region?

So the heuristic is to look for different gateways into the grammar depending on the nature of the systems we want to begin to sketch as we develop our description. I have set out examples in Table 7. These examples are just the initial ways in. For example, if we have started to describe the interpersonal grammar of a language that we know or have reason to believe has grammaticalized resources for enacting tenor distinctions having to do with power and familiarity (“speech level”), we need to add another step in the selection of texts from different registers operating in contexts with different settings of the power and familiarity variables. But as far as possible, we try to sequence the texts we analyse to control for complexity – one way being to take developmental stages into consideration, as noted above. Needless to say, in the study of any particular language, we may find registers in the total registerial make-up of the language that can be useful also at an early stage, e.g. folk plays, oral laws. If we are really working in the field, in a community, then it may be possible to use the discourse diary method with the help of language consultants (see e.g. Ure and Ellis 1977; Halliday 1985), and we can try to chronicle the use of registers during a day in a community, as illustrated by Wiessner (2014), in her study of Ju/’hoansi Bushmen, with a focus on “firelight talk” at night.

Isaac Mwinlaaru: For my description of Dagaare, collecting texts in terms of register was my first consideration. Even before I came to Hong Kong, I started to collect texts from different registers. There were some constraints. One was the constraint of time. When doing a PhD, I had just three years and needed more funding, so I could not exhaust the registerial variations as much as I wanted to. The other constraint was what was available in the language, because Dagaare is largely spoken rather than written.

Following a sketch in Ernest Akerejola’s (2005) thesis on Ò̩ko̩, I wanted to have texts from the eight fields of activities in your registerial cartography and also vary them across modes (see Table 8). But I realized that most of the texts were spoken rather than written, and that reflects the state of Dagaare in terms of literacy development. So, for written texts I could only find biblical passages either written for children or for adults. School texts were quite challenging because the dialect I studied (i.e. Dagara-Lobr) was not the dialect engaged in mother tongue education in Ghana, but was rather prominent in Burkina Faso, which was Francophone and I did not speak French. So it was not easy for me to access written materials on Dagaare from Burkina Faso beyond Catholic liturgical materials, which are also used across the border in my hometown in Ghana. So, that was one challenge in collecting material.

The written texts I had were fairly small in number. I had more spoken texts – texts that are more regular and characterize more of the community discourse, which I think was also good because then it allowed me to look carefully at interactions. As I said, because of time, there were some gaps, which I continue to fill up even after my initial description in the form my PhD thesis. If I had focussed on just some registers or aspects, I would tend to miss some aspects of the whole picture. For instance, if I did not have a dialogic text or naturally occurring conversations, what I would have missed would be the final particles or initial particles, which are very diverse and very important in the interpersonal grammar of Dagaare. They are more delicate realizations of mood. Some of them are obligatory mood markers and others are optional attitude markers (see Mwinlaaru 2018). On the other hand, if I had not looked at narrative texts, I would have also missed out on the use of not only the textual resources of Theme and Rheme, but also the rich resources of tense and modality within the verbal group. The selection of texts according to registers was very effective in this sense.

Christian M.I.M. Matthiessen: Your reflections on your sampling of Dagaare texts from different registers, compilation of a Dagaare text archive and the investigations you undertook using texts from different registers can actually be turned into a general guideline. I remember one of your external examiners was Bernd Heine, a great German linguist who has contributed much to African linguistics (e.g. Heine and Nurse 2000), and also to studies of grammaticalization (e.g. Heine and Reh 1984; Heine 1993), informed by his approach to language and cognition (e.g. Heine 1997). He commented specifically on this being a quite unique feature on your description of Dagaare – your attention to the interpersonal resources, noting that they did not tend to be given enough attention in descriptions of languages around the world.

It is interesting to consider the selection of registers during different phases of language description, comparison and typology. From the 1970s, 80s into the 90s, I remember in general descriptive typological linguistics in the US, linguists often used narratives because they were easy to find or elicit in the languages they investigated, and importantly they were effective in studying certain areas, e.g. tense and aspect systems (e.g. Hopper 1982), linguistic conceptualization of experience, including the construal of motion through space (e.g. Chafe 1980; Berman and Slobin 1994; Slobin 2004). However, not surprisingly, they invited certain lines of interpretation, like the interpretation of the category of subject as a grammaticalization of topic since narratives tend to display extended referential chains. In the case of the category of subject, we can supplement narrative registers with other selections – importantly, dialogic ones with relatively short turns: they make it possible to discern the interpersonal nature of subject (in descriptions of languages where it makes sense to posit such a category), i.e. Subject as the nub of the argument, as the element held modally responsible, as the element elevated interpersonally. Interestingly, when descriptive and typological linguists began to turn their attention to dialogic interaction (e.g. Ochs et al. 1996), they do not seem to have considered subject as an interpersonal category in interactive terms because they were concerned with other aspects of grammar and interaction, often oriented towards the sociological framework of Conversation Analysis. In any case, in order to develop well-rounded thick descriptions of different grammatical categories, we need to investigate and illuminate them based on textual evidence from diverse registers.

10 Final remarks: some general advice

Isaac Mwinlaaru: Based on our discussions above, do you have any general advice for scholars working on language description?

Christian M.I.M. Matthiessen: I remember one observation Michael Halliday made over 40 years ago when talking about the description of different languages and his experience in supervising PhD theses concerned with linguistic description, beginning with the early contributions in the 1960s (Bamgboṣe 1963; Barnwell 1969; Mock 1969). After PhD scholars had passed the examination process successfully, Michael would say: “Right, that was good. Now set that aside and start over again and pretend English never existed.” He said this somewhat jocularly, but what he meant was the ability to stand back and to go through this process of liberating oneself. Linguists in the generative tradition have proposed many constraints in the pursuit of universals. But they tended to be premature, being based on very small databases of languages (cf. Evans and Levinson 2009), I felt that the more urgent need was to expand our imagination of how languages might organize their multi-metafunctional resources for making meaning in context. Of course, this was part of the tagmemic concern with the development of etic pools of linguistic properties.

Even in SFL, the description of languages has been opportunistic in the sense that scholars are interested in describing languages often as native speakers. However, it is good to have a complementarity in teams with both native speakers and non-native speakers, because they can adopt different perspectives on the language. And such teams can provide models of collaboration and complementarity for linguists setting out to describe new languages.

If we consider language description in relation to comparison and typology, it is important to be humble and to recognize how little we know. According to the Ethnologue, maybe there are still around 7,500 languages spoken around the world. Among those, a few hundred at most have been comprehensively described. In SFL, a comprehensive description of a particular language would cover all the dimensions of the architecture of language in context: all the strata – from context to phonetics (or graphetics or sign), the full metafunctional spectrum, all the systems with structural realizations of systemic terms and probabilistic specifications; and the registerial composition of the language would also be covered. In addition, the coverage would extend from the most general, least delicate systems to systems of far greater delicacy – in lexicogrammar, from the grammatical zone into the lexical zone. This degree of coverage has yet to be achieved for any language, even for very well-described languages documented in reference grammars based on very large corpora. But we can locate the descriptive frontiers for any language, identifying gaps; and we can be guided in how to stage the gradual expansion of the descriptive coverage – centrally, by moving from less delicate descriptions towards more delicate ones.

So when we evaluate the descriptive database for doing generalized language comparison – for formulating typological generalizations, we need to consider the degree of coverage of each language included in the sample in the database. From a systemic functional point of view, there will be fairly few descriptions with a reasonably high degree of coverage. And we need to reason about the number of languages to include in typological generalisations and how to weigh factors: genetic relationships, areal contact, typological characteristics.

About 20 years ago, Pagel (2000) calculated that probably the maximum of languages spoken around the world would have been half a million, given the fact that we have been losing languages probably for about the last 10,000 years. His estimate does bear on attempts to create universals (cf. Evans and Levinson 2009) or even robust typological generalizations.

Still, a linguist setting out to begin to describe a “new” language will clearly find it very helpful to check to see if it is included in the WALS or Grambank typological databases or other databases or overviews, or if it is not, if there are generalizations relevant to the development of that language. This can certainly help us to avoid modelling the description of the new language on just a single or small set of well-described languages. Researchers starting out will be tempted to model their emergent description on a well-described language, very often English. But, as I said above, often you can get important help from looking at a language that is similar to the language and the description in one respect to another. If you are describing a language that is verb-initial, go to descriptions of other languages that are verb-initial. Or if the language is verb-medial, consult descriptions of other verb-medial languages, compare and contrast with verb-initial or verb-final ones.

For example, based on a sample of 100 languages, Blake (2001) shows that there are correlations between different positions of the “V” and use of case marking: see Table 9. The sample includes only a few VSO languages, so the most robust comparison is between “SVO” and “SOV” languages. Here it would seem that in “SOV” languages (e.g. among languages described systemic functionally, Korean, Japanese, Telugu), where “S” and “O” are not separated by “V”, the presence of case marking is very common; but in “SVO” languages (e.g. among languages described systemic functionally, Chinese, Vietnamese, Thai, Finnish, English, Swedish, Danish, German, Akan, Dagaare), where “S” and “O” are separated by “V”, lack of case marking is much more common, presumably precisely because “S” and “O” are separated by “V” in the (thematically) unmarked clause. It is important to note two points: (1) While the identification of “V” and “O” is probably reasonably clear, the status of “S” is of course quite problematic. (2) There are strategies other than sequence and case marking for distinguishing participants serving as “S” and “O”, viz. adpositions and head-marking (for the distinction between dependent-marking and head-marking, see Nichols 1986). Thus, the view “from below” of alternative/complementary syntagmatic patterns in the clause is not complete.

Now, thanks to the ongoing development of the WALS typological database and now also Grambank, it is possible to find more detailed information based on a larger sample of languages. Searching the WALS database, I combined the two relevant “features” (in their sense), sequence of elements in the clause (“word order”), specifically S, V, and O (WALS feature 81A; Dryer 2013) and the number of cases used by languages, from 0 to over 10 (WALS feature 49A; Iggesen 2013): see Table 10.

Here “exclusively borderline case-marking” refers to case marking of only circumstances, so we can group it with 0 cases (no case marking) from the point of view of the differentiation of “S” and “O”. The most common sequences of the elements of the clause are in descending order: SOV (101) > SVO (68) > no dominant order (37) > VSO (17). Let me simplify the table by leaving out the least common sequence patterns, i.e. “VOS”, “OVS”, and “OSV” (which there are no instances of at all), combine 0 cases and “exclusively borderline” for the reason mentioned above, and combine all the rows from 2 cases to 10 or more: see Table 11, visualized by the radar chart in Figure 4. Comparing Table 11 with Table 9, we can see that the tendencies are quite similar: the strongly preferred combinations are “SOV” and “2 or more cases” and “SVO” and “0 and exclusively borderline cases”. In the case of “VSO”, there are more languages in the sample that are “0 and exclusively borderline cases” than “2 or more cases”, but the difference is less marked than in the sample in Table 9. In the WALS sample in Table 11, there is also the languages characterized as having “no dominant order”; but while there are slightly more languages with case marking (21) than without (16), it is not immediately clear what this tells us. Here I need to make another important point: in typological overviews of the sequence of elements in the clause (“word order”), we do not actually know if any metafunctional factors have been considered and whether the classifications are at all based on the investigation of clauses in naturally occurring text. See further Matthiessen (2024).

Figure 4:

Visualization of the numbers in Table 11 by means of a bar diagram.

I have undertaken this brief little typological excursion for a couple of reasons:

1. It illustrates what aspects of grammar are available for investigation based on large-scale typological databases such as WALS, which is the leading resource: essentially, these are the aspects we can view when we approach the grammar of a language “from below”, e.g. segmental marking (case marking, adpositions, question particles and the like) and sequence (such as SVO).

2. By the same token, it suggests what a systemic functional typology of languages would contribute – centrally including the adoption of trinocular vision in relation to the semiotic dimensions that make up the architecture of language in context. For example, the “S” in SVO would be problematized metafunctionally and more generally the expressive resources of sequence and segmental marking that are viewed “from below” in Table 10 and Table 11 interpreted metafunctionally in their relevant systemic environments.

3. It shows how we can develop certain expectations based on typological generalizations when we undertake the description of a “new” language – as in this case, by intersecting two WALS “features”, so that we can be guided by the picture that emerges.

The point about typological guidance is, naturally, quite important. The value of such guidance would obviously increase significantly if it could be based on rich comprehensive meaning-oriented descriptions of particular languages. However, one challenge in the continued development is obviously to increase the sample of languages described in such terms significantly to ensure that the considerations involved in compiling such samples are taken into account. Given enough funding and person-power, this should certainly be possible. At the same time, it is important to remember that systemic functional linguists develop descriptions of particular languages for a whole range of reasons, crucially to empower (applied) linguists to undertake extensive analysis of texts in context to address needs in different institutions in a community – as you noted above, Isaac, in relation to your goals when you set out to develop your systemic functional description of Dagaare.

Isaac Mwinlaaru: I would like to add that one crucial consideration in describing a language is to be patient, because language description is laborious and a very gradual patient process. My advice is that we should not rush. We should be patient and go step-by-step. Always be self-critical because we should always allow the possibility that we are wrong. We should find proofs to whatever systems we are positing by assigning realization statements to features in the system and test these against the data we have. Allow some time. Come back to the analysis. Come back to the data. We should allow this time frame between the initial discoveries and our final acceptance of systems in the language.

Christian M.I.M. Matthiessen: Yes, maybe the three Ps – persevering patiently and painstakingly. I have an interesting example from Akan – a West African language we have both investigated. It took time to breakthrough to an understanding of a certain construction, which is now known as serial verb constructions. In the mid-1870s, when describing varieties of Akan, Christaller (1875) had not yet cracked the code really because he came from the Western European tradition that had auxiliaries, and that did not quite work (and coming to terms with auxiliaries in language descriptions is a challenge in its own right, as Heine [1993] shows). Then it took another 30 years or so until Westermann (1907, 1930), when he engaged with Ewe, went closer to some notion of serial verb construction. However, it was not really named and recognized until almost 60 years later by Stewart (1963), who called this pattern serial verb construction.

Then there was a period where people worked on constituency, and they tried to interpret it in terms of IC (immediate constituent) analysis. I remember in the 1980s, coming across this line of analysis, linguists were still trying to base the description on constituency (cf. Lord 1993). It was only thanks to Michael Halliday’s (1967a, 1967b, 1968, 2002 [1979]) work on logical recursive systems and univariate tactic structures that I realized that this was the way forward in coming to an understanding of serial verb constructions. And having located them in this way in the overall system of the grammars of languages, we can then begin to see that they are agnate with constructions with “converbs” (e.g. Bisang 1995), with verb compounding, and of course with clause complexing (cf. Matthiessen forthcoming). At the same time, we can begin to discern the complementarity of the experiential and logical models of construing our experience of the flow of events and the division of labour between these two models in any given particular language.

In short, it may take a long time before we gain enough descriptive experience with a diverse range of languages to allow us to observe, analyze, describe, understand and explain patterns that come across as unfamiliar to us. I put it like this: like beauty, un/familiarity is in the eye of the beholder.