Here we present the research questions and report on our answers to the questions.
RQ1 When the articles in the Global Environmental Change journal are ‘clustered’ according to linguistic features, do the clusters that are identified represent communities of practice, disciplinary or otherwise?
We hypothesized that the clusters of papers identified would be based on linguistic features and these clusters would coincide with disciplines.
The multidimensional analysis model we developed identified six dimensions of linguistic features. When we clustered the 675 files in the Global Environmental Change journal corpus according to their dimension profiles, six ‘constellations’ emerged. Each of the six constellations represents different approaches to research and to knowledge building, rather than to disciplines or formalized groupings.
The six constellations are:
Constellation 1: Site- or target- specific narrative and quantification
This constellation has a concern with action or events rather than with system; there is use of implicit rather than explicit argumentation; a relative absence of features associated with spoken-ness; a focus on space and place rather than on text; and a concern with the non-research world. The papers belonging to this constellation tend to focus on specific sites of interaction between people and the environment (e.g. forest, coastal cities, individual countries or regions), often coupled with specific influences on environmental change. Most of the papers in the constellation give quantified data about changes in aspects of the environment and construe human societies as abstractions defined by environment-related activity.
Constellation 2: Theory
The largest constellation in the journal, with 169 papers, tends to be action-oriented but also discursive, with an emphasis on building an argument around other researchers’ contributions. Some papers in this constellation construct a history of research into environmental change, others address theoretical stances taken by various schools of thought, while others conduct more traditional meta-analyses of existing data.
Constellation 3: Modelling
This constellation features: a concern with system rather than action but a complementary concern with the non-research world; relatively little use of explicit argumentation; little use of features connected with spoken-ness or conceptual discourse; relatively little text focus. Constellation 3 is more system-oriented whereas constellation 1 is more action-oriented. The papers in this constellation are mostly about the activity of modelling environment change.
Constellation 4: Researching people
This constellation is action-oriented and concerned with the process of research itself and with the world of things rather than abstractions. It focuses on people, and includes histories of academic and political approaches to issues of environmental change as well as surveys of public attitude.
Constellation 5: Personal voices
This constellation has a focus on system rather than action; a concern to make arguments explicit; a relatively high proportion of features associated with spoken-ness and with a text focus; and a concern with the research world. In summary, there is a focus on the abstract but also on engagement with a number of voices and with explicit argumentation. The papers belonging to this constellation deal with human perspectives on the environment, including perception studies, and also with social perspectives of science.
Constellation 6: Modelling human beings
It is similar to constellation 3 in terms of dimensions 1 (high in both) and 4 (low in both). It is also similar to constellation 5 in terms of dimension 2 (high in both). It contrasts with constellation 1 in all dimensions except 4. The dimensional profile suggests: a focus on system rather than action; a concern for explicit argumentation; a focus on text but not on conceptual discourse. Like constellation 3, these papers explore models and uncertainty, but the models have a more human focus.
Do the clusters remain the same over time?
Proportions of the constellations vary in the period 1990-2010 but are overall noticeably consistent. In the first years, papers on theory (that is, constellation 2) were predominant (in a period where theorising and agenda setting was necessary) but the proportion subsequently dropped from over 40% to around 15-20%. GEC is a resilient journal (in its twenty year history – as represented in the corpus – it has grown steadily and has a healthy impact factor score; in the period 2011-2015 it has grown even faster), and this balanced and consistent profile over time may be indicative of the reasons for its success.
Do the general linguistic profiles of GEC articles (established through multidimensional [MD] analysis) resemble those of other IDR journals? Are they distinct from those of specialist disciplinary journals?
No. There is no simple grouping of IDR journals that is distinct from the grouping of MD journals, and the general linguistic profiles of GEC resemble some other ID journals but not all of them. It is not possible, therefore, from the multidimensional analysis, simply to describe the linguistic profiles of ID journals in contradistinction to monodisciplinary journals. A possibility we considered was that the constellations would distinguish between MD and ID, raising the possibility that there is a distinctive ID ‘style’, but this turned out not to be the case. On reflection this is not altogether surprising – different forms of interdisciplinary enterprise will bring researchers together from different research backgrounds and while some may be predominantly physical science oriented, others may mix social sciences with life sciences in relatively equal proportions, for example.
RQ2 What is the phraseological profile of the GEC articles?
This question was approached by identifying all the 3-, 4- and 5 grams (in other words, sequences of three, four or five ‘words’) for the GEC corpus, and also finding sequences in which one slot is variable (for example, the pattern it is * to where the slot indicated by the asterisk could be filled by one of several adjectives). The overall phraseological profile was not distinct from other academic writing and so we reframed the other questions to ask which phrases were typical of a constellation and which ones were shared across constellations.
The phraseologies for each constellation tended to be oriented mainly towards either organizing the text (providing what are called ‘transition’, ‘resultative’, ‘framing’ signals) or structuring the foci of the research (explaining location, time, procedure, quantities, or topic). Only in the case of Constellation 6 were there significant uses of phrases that focus on the relationship between reader and writer (what are termed stance or engagement features). Each constellation was characterized by certain categories of phrase function: Constellation 1, for example, containing phrases relating to historical periods (eg, ‘for the period 1960-1972’), to textual entities (eg, ‘as shown in Table 2’) or measurements (‘more than 30% of’). Constellation 6 has a set of evaluative frames that are not as frequent in the other constellations: ‘it should be * that’, for example, where the asterisk could be replaced by ‘noted’, ‘observed’ or ‘argued’.
The phraseological analysis tended to confirm, rather than extend, what had been captured in the multidimensional analysis. We decided, therefore, to take a different approach to the data, which would allow us to see how words co-occur with frequency in different parts of papers. The journal articles in GEC do not all follow a conventional form of organization (such as ‘Introduction – Methods – Results – Discussion’) so it was necessary to find a different way to identify what is happening at different parts of an article through lexical analysis. We used an approach called ‘Topic Modelling’ that is widely used in computational research (though relatively novel in corpus linguistic work) and added an innovation to it by first slicing papers into smaller chunks (one or more complete paragraphs with at least 300 words in total) and then estimating the so-called ‘topic’ probabilities for each chunk. The word ‘topic’ is not to be understood here in its everyday sense, but as a label for a set of words that co-occur in chunks of text. Our TM model was based around 60 topics, and we were able, among other things, to see how topics changed in dominance over the twenty period. The topic of ‘planning and agenda setting’ was strong in the first ten years, for example, then ‘mitigation and adaptation’ in the period 1996-2005, and in the last five years, there was a shift towards studies relating to network actor analysis, vulnerability, ecological systems, or with a focus on households and villages. We were also able to see how different topics feature at different points within articles.
Details of the topic modelling approach can be found in a forthcoming paper in Corpora, to be published in the second half of 2016.
RQ3 To what extent do GEC articles cite sources from across cluster divides?
Our original hypothesis that the constellation divides would reflect disciplinary boundaries was not confirmed, and so the question of whether GEC articles cite from across constellation divides was no longer important to the research. Anecdotally we can confirm that many of the papers do cite work from constellations other than their own, but further citation analysis work would be needed to calculate the comparative extent of this among the constellations.
What and how much do authors cite? How have these practices changed over time?
The average number of citations per paper in GEC has risen. In the period 1997-1999, for example, the average was 47 references per paper and in the period 2008-2010 this had increased to 66. Of the six constellations, 2 and 5 are the constellations with the highest averages with 64 and 62 respectively, and 3 (42) and 6 (48) the lowest. The broad disciplinary domains that papers in GEC cite and are cited by (measured by what the authors cite and also by the source of citations of GEC papers) has shown a shift in the period 1990-2010 with fewer citations of and to physical science journal articles and more to life sciences. However, it should also be noted that the social sciences (26%)and the arts and humanities (5%) are also represented in the data.
In terms of whether the citation appears in brackets after a part of a sentence (a non-integral citation) or the citation plays an explicit syntactic role within the sentence (an integral citation), the former is more common in constellation 1, 2 and 3 papers and the latter for 3 and 6, with constellation 5 having no bias in either direction.
RQ4 How are writers, researchers and readers represented in GEC articles, and how does this change over time?
Constellations differ in their preference for the passive over the active and this leads to varying proportions of use of first person pronouns. In the early years of the journal, the use of ‘we’ was often associated with the generic human sense (as in ‘We cannot survive’) but in later years this use almost disappears and ‘we’ is used to refer either to the writers themselves, or to the GEC research community. A distinction seems to be made between scholars (a part of the academic community, usually working in universities) and researchers who form a broader community, based in a range of institutions.
We have evidence that writers and researchers address an interdisciplinary audience by stating the extent to which they diverge and converge. In interdisciplinary research, the disciplines converge, addressing common questions and problems. They also diverge, adopting different stances towards those questions. It is clear, though, that the practice is not uniform.
Do ‘interdisciplinary researcher’ identities emerge over time?
There was no clear evidence to support this, although there is discussion of a shared agenda and in some instances a distancing of the GEC researcher from the hard physical sciences which are seen as detached from applied research work.
Has the imagined readership changed? How does GEC compare to other IDR journals? How do the IDR journals compare to specialist disciplinary journals?
One aspect of what has been termed ‘metadiscourse’ that is distinctive of GEC (compared to the other journals) is the frequent use of ‘code glosses’. These are expressions that introduce an elaboration, an explanation or a reformulation – such as ‘which is to say’, ‘for example’ or ‘in other words’. This suggests that GEC writers are aware of the need to unpack ides and to illustrate points for a broader audience. This can lead to lengthier articles – the two journals that feature the lowest proportions of code glosses are ‘Plant Sciences’ and ‘Environmental Pollution’ where the average article is four to five thousand words whereas GEC is 7.5 thousand words. A further point is that when GEC authors provide examples, using ‘e.g.’, ‘for example’ or ‘for instance’, it is usually in reference to a real-world entity or event, while in other journals the example is a reference to a research world entity such as another paper (‘see, for example, Reardon et al 2005’). The imagined readership appears to be one that is primarily concerned with phenomena beyond the research world, albeit through the lens of research.
RQ5 What do journal editors, reviewers and authors perceive the distinctive features of an interdisciplinary journal to be, in comparison with a discipline-specific journal?
They indicated that a paper for an interdisciplinary journal should demonstrate the applicability of the research beyond its own discipline (more than 80% of respondents indicated that demonstrating the applicability of their research was important for getting the paper published); a monodisciplinary paper needs to demonstrate a methodological, theoretical or empirical contribution to the field/discipline. An ID paper should draw on a broad range of literature (but not necessarily in depth). Authors have more freedom of expression when publishing in interdisciplinary journals, as there is less of a push to conform. Discipline-specific papers are harder for non-specialists to read than ID papers.
What practices and what linguistic features do they identify as key to successful IDR communication in journals?
The language used should be accessible to a broad audience, and technical terms have to be explained. The paper has to elaborate on methodological decisions and procedures to suit audiences unfamiliar with the approach.
How have practices changed over time?
There have been increases in the range of disciplines published in the journal, the range of methodological approaches, the range of research objects/questions and the number of papers that focus on policy.