On the demarcation of ecolinguistics

3.1 Cautious nominalism

First of all, in the context of methods that rely on bibliographic databases, an overall cautious remark is warranted. Thus, using any search string to elicit an overview of a given field is blatantly nominalist, as it presupposes that the text string covaries with a real entity “out there”. Lechevrel was the first to rightly warn against a tendency to rely on historiographic nominalism in ecolinguistic historiographies, where inferences about properties of a given concept are drawn from the usage of a given word form. This is for instance the case for those ecolinguists who have spent much time on finding the earliest occurrences of the term ‘ecolinguistics’ (e.g. Couto 2014; Fill 2018) or the earliest collocation of language and ecology, even if these occurrences never made it into the discipline as we know it today. Thus, what does it matter that Claude Hagège (1985), Kurt Salzinger (1979), or Jean-Baptiste Marcellesi (1975) used the term “ecolinguistics” if it did not affect the current discipline? After all, many of these sources are only identified long after ecolinguistics became a field in its own right, so by definition their work did not contribute to ecolinguistics. It is hard to deny the veracity of Ludwig, Mühlhäusler, and Pagel’s (2019b: 9) observation that “these names are often introduced as a means of demonstrating the legitimate roots of ecolinguistics”. This is particularly the case since the term was developed in the tradition of French sociolinguistics, a line of thought that have had zero influence on contemporary ecolinguistics. Oddly, the only early (i.e. pre-1990) use of the term “ecolinguistics” as a reference to the study of language in human-nature relations has so far not been included in these overviews. This early instance is P.A. Stott’s (1978a) edited volume on Nature and Man in South East Asia. The second section of the volume is on “Ecology and language”, and in his own contribution to this section, Stott states that “this essay is first and foremost an attempt to encourage the development of what may be termed ‘ecolinguistics’ or ‘phytolinguistics’” (Stott 1978b: 172).

In what follows, I will circumvent this historiographic nominalism in two ways. First, I will adopt Lechevrel’s methodological attitude which says that nominalism cannot be wholesale rejected, as long as it is used with care: “methodological awareness is necessary when doing the history of the ecological paradigm in linguistics, i.e. regarding the benefits and drawbacks of a nominalist approach to identify trends and/or delimit a field of research” (Lechevrel 2009: 4). Lechevrel pursues this modus operandi by widening her search from ‘ecolinguistics’ in a narrow sense to ‘ecological approaches in linguistics’ more widely. She does so by operationalising ‘ecological approach’ in terms of the five search terms mentioned in the previous section, namely (in English): linguistic ecology, ecological linguistics, the ecology of language/language ecology, the ecology of languages, and ecolinguistics (cf. Lechevrel 2010: 48).

Second, I adopt Ian Hacking’s (2006) concept of “dynamic nominalism”. Dynamic nominalism implies that whenever a person adopts (or is attributed with) a concept that functions as a description of the person, they will accommodate to the concept by adjusting their behaviour. This mechanism is nominalist because it assumes the existence of a category that corresponds to the concept, and it is dynamic because the person changes its behaviour in accordance with the categorisation. Dynamic nominalism accounts for the fact that ecolinguists can adopt “ecolinguistics” as a self-classification of their work, and in doing so they orient to a wider field of other scholars who have also adopted the term “ecolinguistics”. Dynamic nominalism differs from historiographic nominalism, as the latter relies on a post hoc classification based on the superficial use of a term. When Salzinger (1979), for instance, used the term “ecolinguistics”, he did not dynamically adapt to any ecolinguistic scholarship, so inscribing him in a long ecolinguistic tradition is a nominalist fallacy.

3.2 Bibliometric databases

Visualisation methods rely on text files extracted from scientific databases. Accordingly, the quality of the visualisation depends on the source from which the text file is built. Academic databases vary in quality, coverage, and in output formats. Therefore, one can rarely rely on a single database that one-sizely fits all one’s needs, so deciding on a database is a matter of making trade-offs.

VOSviewer visualisations can be based on the direct import of data through an “application programming interface” that allows VOSviewer to extract relevant data directly from such applications as OpenAlex and Crossref, or it can be based on CSV files generated in bibliographic databases such as Web of Science, Scopus, Dimensions, Lens, or PubMed. The latter method has the advantage of securing that the included bibliographic data remain unaltered throughout the process of conducting and writing up the survey. Accordingly, the data underlying this survey is contained in CSV files extracted from the four databases listed in Table 1.^[1]

Each of the four sources has different strengths and weaknesses. Web of Science has a conspicuously bad coverage of ecolinguistics. To exemplify, Arran Stibbe who is a key figure in the field only has three items in Web of Science, and the database excludes his three landmark monographs (Stibbe 2015, 2021, 2024).

Purely in terms of coverage, the two open access databases, Dimensions and Lens, perform significantly better than their profit-making competitors, Scopus and Web of Science. However, the data quality in the two open access databases are at times very low, for instance in terms of missing author information and the permutation of information across data fields. In general, Lens has more data points (and is as such “better”), but it requires a more narrow inclusion of data in the visualisation. In contrast, Dimensions has fewer data points, but for that very reason more data points can be included in the visualisation. Overall, the outcomes are quite similar, and Dimensions is used as the primary database for outlining the structures of the overall field of ecolinguistics. However, when it comes to a more fine-grained visualisation of the content of research in ecolinguistics, neither of the two open access databases do well. For this reason, Scopus is used to create fine-grained networks of keywords.

As mentioned in the introduction, there are two competing definitions of ecolinguistics. The broad definition is here equated with Lechevrel’s “ecological approaches in linguistics” (2010) which translates into the five search terms mentioned above. In Table 1, this appears as the extended Scopus and Dimensions databases. Conversely, the narrow definition is operationalised through the two narrow databases.

3.3 Visualisation parameters

This subsection will present VOSviewer as the bibliometric tool used in the article. VOSviewer is developed by Nees Jan van Eck and Ludo Waltman, affiliated with the Centre for Science and Technology Studies at Leiden University. The tool allows for the visualisation of two kinds of maps: term co-occurrence maps (e.g. publication keywords) and maps of bibliographic data (e.g. authorships, affiliations, publication metadata, such as title, abstract, and outlet). In both cases, the maps contain three types of information:

1. information about the single items (represented by the size of the item)

2. information about the item’s similarity with other items (represented by distance to other items and by clustering using colours),

3. Information about the strength of inter-item connections (thickness of graphs/lines between items)

Consider Figure 1 as an illustration. This figure is based on the extended Scopus dataset (802 items), and it illustrates the largest co-authorship network in ecolinguistics, that is, those twelve authors that have co-authored ecolinguistic work with another author within the same network. The map depicts a number of relevant information that is coded in different ways:

1. Item size: in this map, item size indicates number of citations (the larger the circle, the more citations).

2. Graphs: in this map, graphs indicate co-authorship. Any two items connected with a graph has co-authored one or more publications; for instance, Mühlhäusler has co-authored with Nash, Ludwig, and Fill, but no other in the map.

3. Clusters: same colour indicates a cluster; for instance, Mühlhäusler, Nash, and Ludwig constitute one cluster. Exactly what constitutes this group cannot be read from the map.

4. Clusters and lines: Huang, Bao, Zhang, and Li are both clustered and interlinked, because they are thematically related (cluster) and have co-authored a number of articles (four, according to Scopus).

5. Distance: the Mühlhäusler cluster (green) and the Fill cluster (blue) are relatively more similar than the Fill cluster (blue) and the Huang cluster (red).

6. Inter-cluster connections: the three clusters are connected via inter-cluster connections between Mühlhäusler-Fill (who have published together) and Steffensen-Huang (who have published together). The two are linked because Steffensen and Fill are linked through their 2014 state of the art.

Figure 1:

The largest co-authorship network in ecolinguistics. Minimum four publications.

For each map or visualisation being produced, one has to select a number of inclusion criteria. For instance, the map in Figure 1 includes authors that have authored four or more publications, whereas the alternative version in Figure 2 includes authors that have authored two or more publications.

Figure 2:

Co-authorship network in ecolinguistics. Minimum two publications.

Obviously, the number of authors increase, as one lowers the criterion of numbers of publications. The figure is included because it illustrates some of the limitations of the visualisation. Thus, this figure includes Hermine Penz who branches out from Fill, as they have co-authored multiple pieces. However, Penz branches to yet another item called ‘Fill, Alwin F.’. This is obviously a database artefact, as Fill is both registered with and without his middle name (Frank). Unfortunately, VOSviewer allows for the removal of single items, but not the merging of two.

Adjusting the minimal cluster size is only one parameter that can be changed to affect the visualisation. To illustrate, Figure 3 is the same map as Figure 2, with the only difference that item size is not determined by number of citations, but by number of publications by an author. By comparing Figures 2 and 3, one can see that the visual appearance of the visualisations depends on which parameter the author selects. Any visualisation is as much a visualisation of author decisions as it is a visualisation of a given field.

Figure 3:

Co-authorship network in ecolinguistics. Number of documents as weight.

VOSviewer allows for another layer of information which replaces clustering information. Thus, one can use ‘overlay visualization’ to depict the temporal development within a network. This visualisation setting allows to illustrate how a field develops in generations. Thus, if one considers Figure 4, one can see that Mühlhäusler and Fill belong to a first generation (purple colour), as their average publication years are 2011.30 and 2011.00, respectively. The second generation consists of Nash (2015.71), Ludwig (2015.50), Steffensen (2017.25), Pagel (2018.00), and Penz (2018.67). Cowley (2020.75) and Huang (2021.00) constitute a third generation, and within Chinese ecolinguistics, Huang has given rise to a new generation of scholars, including Zhou (2020.00), Bao (2021.25), Zhang (2022.33), and most recently Chen (2023.50). These authors may well have published several pieces in other fields, so the generational categorisation pertains to ecolinguistics only.

Figure 4:

Co-authorship network in ecolinguistics. Overlaid with publication years.

Finally, to illustrate how different databases create very different results, Figure 5 is created using the exact same settings as Figure 2, with the only exception that the data file here is the extended Dimensions file rather than the extended Scopus file. Since Dimensions include many sources that are not indexed in Scopus, other author constellations come to the fore, and as one sees from the Figure, the largest co-authorship network in the Dimensions file is constituted by twenty-two Indonesian scholars. Ecolinguistics took off in Indonesia in 2017, and ninety-two out of the 1,525 items in the extended Dimensions file come out of Indonesia.

Figure 5:

Co-authorship network in ecolinguistics. Extended Dimensions data file.

Some of the finer distinctions in VOSviewer, souch as the difference between the use of full counting and fractional counting (Perianes-Rodriguez et al. 2016), will be ignored here. Further information on this difference, as well as on VOSviewer in general can be found on the website of the tool (www.VOSviewer.com), as well as in the numerous publications of the creators of the tool (e.g. Lamers et al. 2021; van Eck and Waltman 2009; van Eck and Waltman 2010; van Eck and Waltman 2023; van Eck et al. 2010; Waltman and van Eck 2012; Waltman et al. 2010).

4.1 The ecolinguistics-as-one view

Is the ecolinguistics-as-one view warranted? The three figures indicate a bipolar ordering of the field. Each of the three figures is organised around two red high-density areas with each their orange-yellow hinterland. Bear in mind that the three figures are based on the extended data files that included, in Lechevrel’s (2010) words, all “ecological approaches in linguistics”. Accordingly, a striking observation is that whether one attends to keywords, sources, or authors, the ecolinguistics-as-one view is unwarranted. There is not a single, coherent field of ecolinguistics that comprises ecolinguistics, linguistic ecology, ecological linguistics, ecology of language, language ecology, and ecology of languages (i.e., the ecological approaches as defined by Lechevrel’s search string).

The strong bipolarity in the maps suggest that ecolinguistics is not a single unified field. If one wants to pursue a unified definition of ecolinguistics, one is therefore forced to operate with a demarcation that excludes parts of what is mapped in Figures 6–8. Crucially, in the absence of empirical data that show an inherent unity, claims of unity are either observer-dependent projections (just as constellations in the night sky are projected by communities of earthly observers), or desiderata for a future development of the field.

The trend in contemporary ecolinguistics is to define ecolinguistics as the right pole in Figure 6. As a reviewer remarks in response to an earlier version of the article, this pole stands out as it attempts to overcome the anthropocentrism of mainstream linguistics by taking into consideration the natural ecology on which life depends. This contrasts with the left pole that explores sociolinguistic issues by taking an ecological perspective.

4.2 The ecolinguistics-as-two view

We can now turn to the question of what the two poles of the bipolar order in Figures 6–8 refer to. In doing so, one attends to the ecolinguistics-as-two view by raising the question whether the distinction between Haugenian and Hallidayan ecolinguistics is still a relevant model for contemporary ecolinguistics, or if indeed the development since it was first proposed by Fill (1996) suggests that it should be abandoned.

Let us first consider Figure 6 in detail. As mentioned, the model shows how the keywords within ecolinguistics in the extended sense (cf. Fill 1996; Fill and Penz 2018) are organised. We see that the two poles correspond to two networks of terms, one pivoting on ecolinguistics, and another on language ecology. The former is encircled by such terms as biodiversity, ecology, nature, sustainability, environment, environmental change, climate change – and also such linguistic terms as (critical) discourse analysis, and multimodal analysis. The latter, in contrast, links to terms like multilingualism, language policy, language planning, language maintenance, language shift, and linguistic diversity.

This picture is confirmed if one turns to journals/sources and authors, as shown in Figures 7 and 8, respectively. The split between the two poles is to a large degree coextensive with the organisation of sources. Thus, Figure 7 affiliates the ecolinguistic pole with such outlets as Language Sciences, The Ecolinguistics Reader, Journal of World Languages, The Routledge Handbook of Ecolinguistics, Discourse and Communication, to name a few. Unsurprisingly, the other pole is associated with journals like Journal of Multilingual and Multicultural Development, World Englishes, and Internal Journal of Bilingual Education.

Likewise, Figure 8 exhibits a similar bipolarity, where authors in the ecolinguistic pole includes Stibbe, Huang, Steffensen, Fill, Nash, and Umiyati, whereas the language ecology pole exhibits names like Mühlhäusler, Mufwene, Philipson, Hornberger, Blackledge, Kramsch, and Pennycook. A few authors do indeed cross the boundary between the two. Peter Mühlhäusler is arguably the most prominent one, and the collaboration between Kramsch and Steffensen also bridge the divide (Kramsch and Steffensen 2008; Steffensen and Kramsch 2017). However, it does not follow from that observation that the two poles are inherently linked, as authors can contribute to different fields.

In conclusion, the three bibliometric maps indicate that on all relevant parameters – keywords, sources, and authors – ecolinguistics in Fill’s original sense comes over as two distinct areas of research. Superficially, this may seem to confirm the ecolinguistics-as-two view, according to which ecolinguistics comprises a Haugenian tradition and a Hallidayan tradition. Yet, while the maps show that the field is indeed organised in two poles, the main insight from these maps is that the interconnections between the two poles are virtually non-existing. This is a crucial observation, because the lack of interconnections between the two poles undermines the claim that ecolinguistics is a field that consists of two complementary poles. Rather, the ecolinguistics-as-two model is the result of an observer-dependent claim about the parallels between two separate fields.

Judged from the bibliometric maps, there are no indications that the practitioners in the field engage with work from the other pole. In conclusion, the two-traditions view is thus only warranted by how collected works in the field are organised (Fill and Mühlhäusler 2001; Fill and Penz 2018), which is a quite shallow criterion. Accordingly, the conclusion of this bibliometric analysis is that there is no empirical evidence for the claim that ecolinguistics comprises a language ecology pole and a natural ecology pole.

4.3 The ecolinguistics-as-four view

We can now turn to the most recent model of the organisation of ecolinguistics, the ecolinguistics-as-four view presented by Steffensen and Fill (2014). To do so, we have to change perspective. The heatmap visualisations in the previous analyses provided a coarse-grained understanding of how the field is organised. These visualisations were relevant for discussing the adequacy of the ecolinguistics-as-one and ecolinguistics-as-two views. However, the heatmap visualisations are not adequate for providing a more detailed overview of work being carried out within the different poles. Thus, while it is tempting to conclude that the two poles correspond to a Haugenian and a Hallidayan pole, a more fine-grained view of the two poles will show that they are in fact composed by multiple smaller areas.

Accordingly, in order to get beyond the Haugen-Halliday duality, this section establishes how ecolinguistics is constituted by attending to the 135 author keywords that are use three or more times in the extended Scopus dataset. The co-occurrence of author keywords is exhibited in Figure 9, which shows the “cluster density” of author keywords, and Figure 10, which shows the co-occurrence, overlaid with information about publication year.

Figure 9:

Co-occurrence of author keywords in the extended Scopus data file. Only keywords with three or more occurrences are included. Labels indicate keywords; size of labels indicate number of occurrences. Colours show the clustering of keywords, based on strength of co-occurrence. Not all keywords appear in the map, as some are hidden under others. Files that allow for a zoomable map in VOSviewer are provided as supplementary material.

Figure 10:

Co-occurrence of author keywords in the extended Scopus data file. Only keywords with three or more occurrences are included. Labels indicate keywords; size of labels and nodes indicate number of occurrences. Colours of the nodes indicate the average publication year for each keyword. Not all keywords appear in the map, as some are hidden under others. Files that allow for a zoomable map in VOSviewer are provided as supplementary material.

A first observation from Figures 9 and 10 is that the bipolar maps (Figures 6–8) do not give a full picture, as it gives the impression of a symmetry between the two poles. In contrast, Figures 9 and 10 are highly asymmetric. The language ecology pole (to the left) is both more varied and older than the ecolinguistics pole (to the right). The variability is shown in Figure 9 where more clusters inhabit the left half of the map than the right half. Likewise, Figure 10 shows that the average publication year is earlier at the left pole than at the right.^[2] These observations support the conclusion that any description of ecolinguistics in terms of two subfields, two approaches, or two traditions is too superficial.

If we want a better understanding of the field, we need to pay closer attention to the configuration of the 135 author keywords, by considering the two measures in VOSviewer: the distance between nodes and the clustering of nodes.

Figure 11 depicts these keywords including their clustering (i.e. the colour of the nodes) and distance (i.e. the spatial distribution). In contrast to the ‘cluster density’ map in Figure 9, Figure 11 is technically a network map with inter-node links omitted, just as label names are omitted for clarity (the label names are identical to the names in Figures 9 and 10). This view allows for a manual bundling into four areas, based on three criteria:

1. Clusters can be only bundled if they occupy the same space, or when they occupy adjacent spaces

2. Clusters can be bundled if they exhibit thematic coherence

3. A single cluster is by default held together, unless it is discontinuous, that is, if a single cluster occupies two distinct spaces with one or more clusters occupying the space between these.^[3]

Figure 11:

Co-occurrence of author keywords in the extended Scopus data file. Size of nodes indicate number of occurrences. Colours show the clustering of keywords. Black lines indicate the superimposed bundling of keywords into four areas, depending on criteria of clustering and distance.

The four areas of bundled clusters cover a wide academic terrain. First, to the west on the map, one finds a bundle of brown, green, orange, yellow, and purple clusters. This is the sociolinguistic tradition of documenting the “language ecology” of languages that coexist in the same area. This is basically a continuation of Haugen’s project, and it has given rise to a rich tradition of studying language ecologies (or linguistic ecologies) characterised by language contact on multiple levels from individual speakers to speech communities. For a contemporary overview of this work, the reader is referred to the contributions in Ludwig, Mühlhäusler, and Pagel’s (2019a) excellent edited volume. Steffensen and Fill (2014: 8–9) referred to this line of work as the “symbolic ecology of language”.

The second area is in the middle of the figure, and it comprises the blue, indigo, light red, and light green clusters. It consists of scholars with an interest in language learning, socialisation, and acquisition. As Kramsch (2002a: 3) argues, they have sought to overcome the unhelpful juxtaposition between language acquisition and language socialisation by rallying around the ecology metaphor. By referring to social, cultural, institutional economical, and political forces as constituting an “ecology of language learning” (cf. Kramsch 2002b; Leather and Van Dam 2003), this tradition is the one which lies closest to the traditional preoccupations of applied linguistics (cf. Kramsch 2020; Steffensen and Kramsch 2017). In the 2014 state of the art, Steffensen and Fill referred to this body of work with the term “sociocultural ecology of language” (Steffensen and Fill 2014: 12–14).

The third area to the east (Red, light purple, light blue, and yellow clusters) is constituted by a large group of scholars who pursue Halliday’s question of how language relates to environmental and climatic processes, either by being shaped by these processes or by influencing them directly or indirectly. On the one hand, this tradition has taken up the baton from Sapir (1912: 229) who argued that languages depend on “fauna or topographical features,” depending on “the interest of the people in such environmental features.” Thus, through a filter of social relevance, languages are influenced by topographic and biological factors. On the other hand, under the influence of Halliday’s work, many scholars argue that the relationship is reciprocal, in the sense that meaning-making processes also shape our environmentally embedded behaviour. As such, language plays a role in the current ecological crisis (cf. Halliday 1992). This area correlates with Steffensen and Fill’s (2014: 9–12) “natural ecology of language”.

Finally, to the south one finds a single purple cluster of ecolinguists who focus on the cognitive aspects of the language-behaviour-environment nexus. Without explicitly evoking Salzinger’s (1979) behaviourism, they focus on how human agents engage with their sociomaterial environment, primarily by using methods from cognitive science, including ecological psychology (Steffensen and Baggs 2024), autopoietic biocognition (Kravchenko 2016), radical embodied cognitive science (Steffensen and Cowley 2021), and distributed cognition (Li et al. 2020). While this development is quite recent (cf. Steffensen et al. 2024b), it was foreshadowed by Lechevrel (2010: 69) who observed that “Étendues aux sciences cognitives, les approches écologiques concernent les domaines de la cognition située et distribuée, et on les retrouve dans le traitement des affordances, du pointage, et des interaction multimodale” [“Extended to cognitive sciences, ecological approaches concern the domains of situated and distributed cognition, and we find them in the treatment of affordances, pointing, and multimodal interactions”]. The first full attempt at establishing an ecological-cognitive version of ecolinguistics appeared in Steffensen (2008), and later in the second half of the Steffensen and Fill’s state of the art (2014: 16–22). Most readers of this work, however, have focused on the four strands model, and not on how the cognitive perspective in fact dissolves and unifies the four strands.

If one considers the boundaries between the language ecology (to the west) and the sociocultural ecology (in the middle), it meanders between three large keywords that are placed in great proximity to one another: ‘language ecology’ (blue), ‘multilingualism’ (green), and ‘language policy’ (brown). This proximity indicates that the boundary between these two areas is somewhat fluid. However, if one changes the lens from an author keyword map (as in Figure 11) to a citation map (Figure 12), the language ecology area and the sociocultural ecology area are indeed two distinct areas. A citation map structures publications within a given dataset according to the citation links between two publications (where the one cites the other or vice versa). Figure 12 maps the 245 publications which are part of the largest interconnected network of publications in the extended Scopus dataset.

Figure 12:

The largest citation network in the extended Scopus dataset. Proximity indicates citation relations. Colours indicate VOSviewer’s clustering of publications, and size of labels indicate number of citations.

Crudely speaking, the map in Figure 12 has three sections:

1. In the middle, stretching toward the bottom left corner, there is a large area of many (albeit less cited) articles that would be categorised as the natural ecology of language and the cognitive ecology of language, or simply: ecolinguistics.

2. To the upper left, one finds a yellow area which corresponds to the symbolic ecology of language. It includes many publications on multilingualism and the linguistic ecology of select geographical areas.

3. To the right, the green area pertains to the sociocultural ecology, represented by work by such authors as Kramsch, van Lier, Creese, and Philipson.

Accordingly, this map documents that the symbolic ecology and the sociocultural ecology are in fact distinct. No single visualisation functions as the holy grail of bibliometrics, so it is no surprise that one has to draw on several figures to come up with an empirically based understanding of how a given field is organised.

In conclusion, the ecolinguistics-as-four view seems to find support in the bibliometric evidence. However, the premise of this conclusion is the broad Lechevrel-inspired search string, and the maps in this subsection suffer from the same limitations as the maps in the previous subsections, namely that the interconnections between the four areas are limited. While it is indeed warranted to claim that the overall category of “ecological approaches in linguistics” are organised in four different strands, it is not necessarily warranted to claim that all four areas contribute to a single, unified ecolinguistics. To do so is to merge this perspective with an ecolinguistics-as-one view, and as discussed in Section 4.1, this can either be an observer-dependent projection or a desideratum for the future. This conclusion leaves us with one important question that still needs to be considered: Faced with the bibliometric evidence, how do we demarcate ecolinguistics in a way that provides a sound basis for future work in the field? I will turn to this question in the next section.