Content uploaded by Christer Johansson
Author content
All content in this area was uploaded by Christer Johansson on Apr 17, 2024
Content may be subject to copyright.
115
doi:10.5128/ERYa20.07
croSS-linguiStic patternS of meta-
diScourSe: diSciplinary SimilaritieS
and Section-baSed differenceS
Djuddah Leijen, Helen Hint, Helena Lemendik,
Baiba Egle, Anna Ruskan, Christer Johansson
Abstract. This study examines metadiscourse markers across a corpus
of Estonian and Lithuanian journal articles in the field of linguistics.
We aim to 1) compare the global use of all the metadiscourse markers
across the languages and texts, making distinctions between these lan-
guages and specific academic journals, and 2) to discern whether similar
and/or different patterns can be identified across the languages and
whether such patterns also manifest across various academic journals.
We find that Estonian writers use self-mentions more frequently in
methods sections than Lithuanian counterparts. Comparing journals,
the Lithuanian journal Kalbotyra shows more transition markers, code
glosses, and endophoric markers, while the Estonian ERÜ aastaraamat
relies more on transition markers in results and discussion sections.
Despite discipline similarities, variations emerge in specific sections
and interpersonal categories across languages and journals. The study
provides insights into metadiscourse patterns and their role in different
languages and academic contexts, offering potential guidance for future
research and practice in non-English academic writing.*
Keywords: metadiscourse, research articles, IMRaD, Estonian,
Lithuanian
1. Introduction
In journal articles, metadiscourse markers play a crucial role in establishing a
balance in the discourse, aligning the writer's intentions with the reader's percep-
tions and interactions within the propositional content of the text. The concept of
metadiscourse encompasses various linguistic expressions that not only convey
the primary content of the text but also guide and engage readers through the dis-
course, addressing their needs and facilitating both the coherence and cohesiveness
* This work is supported by the project “Bwrite: Academic Writing in the Baltic States: Rhetorical Structures through
culture(s) and languages” (EMP475), funded by Iceland, Liechtenstein and Norway through the EEA Grants and
Norway Grants.
The colored versions of Figures 1–4 have been published in the online edition of the article (https://doi.
org/10.5128/ERYa20.07).
EESTI RAKENDUSLINGVISTIKA ÜHINGU AASTARAAMAT 20, 115–132
116
of the text (Crismore et al. 1993, Mauranen 1993, Vande Kopple 1985). The use of
metadiscourse markers, while well documented in English, also presents notable
patterns and variations across different languages (Dahl 2004, Fløttum et al. 2006,
Mur-Dueñas 2011), genres (Ädel 2018, Hyland et al. 2022), and disciplines (Birhan
2021, Hyland et al. 2022). Hyland (2005) has been pivotal in shaping our current
understanding and application of metadiscourse markers, particularly through the
application of his interpersonal model of metadiscourse, which has been widely
recognised and applied in studies investigating English academic writing, but also
across a variety of other languages (Cao, Hu 2014, Peng, Zheng 2021).
The interpersonal model (Hyland 2005), see also Table 1, bifurcates meta-
discourse into the interactive and the interactional dimensions, each serving a
distinct yet intertwined role in creating academic discourse. The interactive dimen-
sion, encompassing transitions, frame markers, endophoric markers, evidentials,
and code glosses, serves to guide the reader through the text. Conversely, the
interactional dimension, which includes hedges, boosters, attitudinal markers,
self-mentions, and engagement markers, mirrors the intention of the author to
involve the reader in the text (Hyland et al. 2022, Hyland, Tse 2004). As such, at
the macro level, i.e., at the whole text level, the use or lack of use of these meta-
discourse markers can offer a snapshot of how the text and the author interact
with the reader and, more specifically, where in the text, as has been highlighted
by (Ruskan et al. 2023). Building upon this framework, the present study empiri-
cally explores the use of metadiscourse markers across two languages: Estonian
and Lithuanian, found across various published journal articles in the discipline
of linguistics to 1) compare the global use of all the metadiscourse markers across
the languages and texts, making distinctions between these languages and specific
academic journals, and 2) discern whether similar and/or different patterns can
be identified across the languages and whether such patterns also manifest across
various academic journals.
Exploring metadiscourse markers across different languages and academic jour-
nals requires a closer examination of the linguistic and discursive choices embed-
ded within the editorial expectations of each respective journal context. Regarding
journal articles subject to stringent editorial and language requirements, selecting
texts that successfully pass through the peer review and editorial process may be
influenced. Consequently, these texts may exhibit similarities in style, thereby
validating the survivorship bias that can arise when we choose to investigate texts
that have undergone a rigorous process of writing, reviewing, revising, and editing
before publication. This process, particularly when assessed by a small group of
stakeholders, may exhibit a specific metadiscursive footprint. Moreover, studies have
indicated that stylistic expectations in language communities other than English
are often higher, assuming a very high language proficiency and rhetorical style
representative of the national culture and/or academic/discursive culture (Duszak
1994, Duszak, Lewkowicz 2008, Harbord 2018). Therefore, while we may have
some understanding of the intricacies involved in constructing texts, the editorial
process of journals, and the complexities of national cultures and languages, the
evaluation of these metadiscoursal patterns at the macro level (i.e., patterns observed
across the whole text), and the meso level (i.e., patterns observed across sections of
a text) remain underexplored (see also Leijen et al., Forthcoming). Furthermore,
117
to our knowledge, no studies have compared how these patterns compare across
languages and across journals published in other languages. Evaluating these pat-
terns at the macro- and meso- levels across languages and journals may help us
better understand whether the interpersonal model proposed by Hyland, specifically
the bifurcation interactive and interactional dimensions, show distinctive editorial
differences and/or language-related differences.
This study builds upon a prior investigation (Hint et al. 2022, Ruskan et al.
2023), which identified metadiscourse markers across 21 journal articles (seven
journal articles belonging to three specified journals in the field of linguistics). This
work is part of a larger project aiming to uncover the rhetorical structures of aca-
demic texts in Estonian and Lithuanian (Jürine et al. 2021, Leijen et al., Forthcom-
ing). For this study, we combined the research on the use of metadiscourse markers
across journal articles in Lithuanian (30 journal articles belonging to three specified
journals in the field of linguistics) with the previously mentioned prior investigation
of Estonian metadiscourse markers. The goal is to compare metadiscourse patterns
of Estonian and Lithuanian and to assess the metadiscourse usage in comparable
academic journal articles in the field of linguistics across these two languages.
Additionally, this research endeavours to identify potential universal metadiscourse
strategies that may transcend these languages and academic journal contexts. By
offering a nuanced, cross-linguistic, and cross-academic journal perspective, this
study contributes valuable insights to the existing knowledge on metadiscourse in
languages other than English. These insights may potentially inform and enhance
future research and practice.
2. Data and method
The data used for this study comprises two languages, Estonian and Lithuanian,
collected and coded as part of a larger research project investigating writing con-
ventions and rhetorical structures, that is, writing traditions (Hint et al. 2022,
Jürine et al. 2021, Leijen et al., Forthcoming). For this study, a corpus of academic
journals was used across the two languages each contributing to a comprehensive
analysis of all the metadiscourse markers contained and identified by Hyland’s
interpersonal model of metadiscourse (Hyland 2005). For a complete overview of
the metadiscourse markers, see Table 1, and for a complete description of metadis-
course markers in Estonian, see (Hint et al., submitted). The Estonian sub-corpus
comprises 21 journal articles with seven articles each from three journals: Keel ja
Kirjandus (‘ La ngu age a nd L it era tu re’ ), Eesti Rakenduslingvistika Ühingu aastaraa-
mat (‘Estonian Papers in Applied Linguistics’), and Emakeele Seltsi aastaraamat
(‘Yearbook of the Estonian Mother Tongue Society’), totalling 89,224 words. The
Lithuanian sub-corpus, somewhat larger than the Estonian sub-corpus, with a total
word count of 136,443, encompasses 30 articles, ten from each of the following
journals: Kalbotyra (‘Linguistics’), Lietuvių kalba (‘The Lithuanian Language’),
and Taikomoji Kalbotyra (‘Applied Linguistics’).
118
Table 1. An interpersonal model of metadiscourse by Hyland (2005: 49)
Category Function Examples
Interactive Help to guide reader through the text Resources
Transitions express relations between main clauses in addition; but; thus; and
Frame markers refer to discourse acts, sequences, or text
stages finally; to conclude; my purpose is
Endophoric
markers
refer to information in other parts of the
text noted above; see Fig; in section 2
Evidentials refer to information from other texts according to X; Z states
Code glosses elaborate propositional meanings namely; e.g.; such as; in other words
Interactional Involve the reader in the argument Resources
Hedges withhold commitment and open dialogue might; perhaps; possible; about
Boosters emphasize certainty or close dialogue in fact; definitely; it is clear that
Attitude markers express writer’s attitude to proposition unfortunately; I agree; surprisingly
Self-mentions explicit reference to author(s) I; we; my; me; our
Engagement
markers explicitly build relationship with reader consider; note that; you can see that
To ensure a thorough and accurate representation of metadiscourse markers across
the languages and journals, the initial phase of the data annotation was close reading
and manual annotation of a small portion of academic journal articles in the field
of linguistics from the corpus. Furthermore, both language teams negotiated their
language-specific understanding of what constituted metadiscourse markers in their
languages based on a shared understanding of English language descriptions (see
Table 1). The annotated markers spanned various linguistic units, including con-
structional, lexical, and grammatical elements, as well as punctuation marks. Sub-
sequently, a meticulous manual annotation phase was performed, which involved
a detailed review and, where necessary, adjustment of the annotated data (for a
comprehensive overview of the full data coding process see Hint et al., Forthcoming).
The manual annotation was conducted by at least two annotators for each language.
To ascertain the reliability of the annotations, Cohen’s kappa interrater reliability
measures were applied, revealing a high degree of agreement between annotators
across all categories (0.900 and above for both languages). Any discrepancies and
disagreements encountered during this phase were collaboratively discussed and
resolved, ensuring a refined and consistent annotation across the dataset, thereby
bolstering the reliability of the corpus of metadiscourse across the two languages.
Given the objective to map the overall pattern of metadiscourse use on the macro
level (i.e., across the whole journal article), the text section variable was annotated
based on the classic IMRaD structure of a research article (Sollaci, Pereira 2004)
as much as the articles would allow. However, not all research articles in our cor-
pus adhered to a clear IMRaD structure. For example, in many cases, in addition
to the introduction, the journals would have an additional section to cover the
theoretical framework labelled literature review in Estonian journals or labelled
theory in Lithuanian journals, ostensibly covering the same type of information.
Additionally, in some instances the results and discussion sections were combined
119
into a single section, results and discussion, serving both functions. Therefore, and
regardless of the actual section headings in the research article, we used the unified
labels ‘introduction’, ‘literature review/theory’, ‘method’, ‘results and discussion’,
and ‘conclusion’, as much as possible. To conduct a comparative analysis across
the journal articles contained within the languages, and across the languages, we
collapsed some of the journal article sections. For example, in some instances
where there was a separate section for results and results and discussion, the
results would be included in the results and discussion section. Furthermore, some
sections which were coded in the original corpus were excluded from the analysis
of the cross-language and cross-journal article analysis, such as titles, footnotes,
and acknowledgments. Table 2 provides an overview of the included sections for
analysis of the language specific journals.
Table 2. Journals and included sections for comparative analysis
Journals Sections
Estonian
Keel ja Kirjandus (KK) Introduction; Literature Review; Method; Results;
Results & Discussion; Discussion; Conclusion
Eesti Rakenduslingvistika
Ühingu aastaraamat (ERÜ)
Introduction; Literature Review; Method; Results;
Results & Discussion; Discussion; Conclusion
Emakeele Seltsi aastaraamat
(ESA)
Introduction; Literature Review; Method; Results;
Results & Discussion; Discussion; Conclusion
Lithuanian
Kalbotyra (K) Introduction; Theory*; Data and Method; Results,
Results & Discussion; Discussion; Conclusion
Lietuvių kalba (LK) Introduction; Theory*; Data and Method; Results
&Discussion; Conclusion
Taikomoji Kalbotyra (TK) Introduction; Theory*; Data and Method; Results
&Discussion; Conclusion
*Theory, in the Lithuanian data, is the same as literature review in the Estonian data.
To provide a more in depth understanding of the articles within the specified
journals and to elucidate the extent to which variations can be accounted for, we
provide information of the journals pertaining to any editorial requirements. This
includes highlighting the explicitness of editorial guidelines related to accepted
article types and anticipated styles, the provision of editorial assistance, and the
extent of editorial editing post-article acceptance. This information serves to
ascertain whether editorial expectations might instigate a particular survivorship
bias within published articles, and subsequently, within the language-specific
corpus.
We used association plots to determine the patterns of the use of metadis-
course markers across the languages and across the journals. The vcd package
(Meyer et al. 2003) in the free statistical software R (R Core Team 2022) was
used for conducting the analysis and creating association plots. We further used
the reshape2 package (Wickham 2007) and ggplot2 package (Wickham 2016) to
further visualise the cross-linguistic and cross-journal article comparisons with
heatmaps. Specifically, the heatmaps facilitate the analysis of the Pearson residuals
presented in the association plots, revealing deviations from the expected statistical
120
independence of rows and columns, representing the sections (IMRaD type and
variations) and metadiscourse markers in this study, respectively. The plot visual-
ises the association and dissociation between the journal article sections and the
metadiscourse category. To further highlight these variations and compare any
editorial variation between the journals for each language, a baseline model was
calculated. The baseline model basically consists of the combined journal articles
representing a metadiscourse model, which you could expect when you observe
the use of metadiscourse across a larger dataset. The three separate journals in the
Estonian and Lithuanian sub-corpus are compared to the baseline metadiscourse
model to determine the location of variation (for example, variation of metadis-
course marker use) or at the level of the journal article sections, i.e., introduction,
method, etc. and finally, variation across languages.
3. Results and discussion
The first aim of this paper is to compare the prevalent use of metadiscourse markers
across languages and texts, differentiating among these languages and their corre-
sponding academic journals. The second aim is to underscore both the similarities
and differences within individual languages and genres, as well as between them in
the corpus. Before presenting the results as they relate to the aims, we checked the
specific journal websites for specific information pertaining to aspects related to,
1) text length, 2) macro formatting guidelines related to IMRaD structures, 3) style
guides, plus 4) specific linguistic style suggestions to add some additional contextual
information which may help us to determine whether editorial guidelines of the
specific journals may result in a specific pattern emerging in the metadiscourse.
The results are shown in Table 3.
The majority of the journals do not provide an explicit guideline related to sec-
tions and language-specific guidelines, except perhaps the Estonian language journal
Keel ja Kirjandus and the Lithuanian journal Kalbotyra which both explicitly state
that the journal reserves the right to edit the article in cooperation with the author or
requires editing by a native-language specialist. All other journal guidelines provide
a broader guideline related to text style guides, such as formatting and reference
style guides (i.e., APA, MLA, etc.).
3.1. Language-specific metadiscourse baseline models
To compare the distribution of metadiscourse markers across journal articles for
each language, we explore the relationship between metadiscourse markers and
the journal article sections coded in the corpus. Table 4 presents the percentages
of the metadiscourse markers used for the purpose of this comparative analysis.
The percentages represent metadiscourse markers in the entire sub-corpus and
the percentages of all metadiscourse markers within each section of the journal
articles. Generally, in both the Estonian and Lithuanian datasets, interactive
markers – markers that assist in guiding the reader through the text, such as
transition markers, code glosses, endophoric markers, evidentials, and frame
121
Table 3. Journal editorial guidelines pertaining to any stylistic or discoursal guidelines
Journals Specific guidelines
KK • Recommendedlength40,000–50,000characters(incl.spaces).
• Thejournalreservestherighttoeditthearticleincooperationwiththeauthor,incl.
shortening it if necessary.
• Listedtextstyleguidelines.
ERÜ • Lengthupto40,000characters(excludingspaces).
• Thetextshouldbestructuredinsectionswithdecimalhierarchicalnumberingand
section titles. Excessive use of footnotes must be avoided.
• Instructionsforabbreviations,citations(in-text),references,glosses,andtextstyle
are included.
ESA • Recommendedlength35,000–50,000characters(incl.spaces).
• Textstylerequirementsandrecommendedstructuring(onlyonelevelofsub-
section), in-text citation instructions and reference guidelines.
K• Recommendedlength8,000words;inexceptionalcasesthepapercanbelonger.
• Ifthelanguageofthepaperisnotanativelanguageoftheauthor(s),thepaper
should be proof-read by a native-language specialist to check its correctness.
• Divisionintosectionsandsubsections.
• Otherlistedtextstyleguidelines.
LK • Therearenorequirementsforthelengthofanarticle.
• Recommendedstructureandform:theresearchquestion/problem,reviewof
previous research on the subject, data and methods, research findings/results
(evaluated and validated), evidence (documented), conclusions and references.
• Listedtextstyleguidelines.
TK • Thetotallengthofthepublicationshouldrangefrom10,000to80,000characters
with spaces. The recommended volume of scientific reviews is 16,000 characters. In
case the contribution exceeds the indicated length, it should be negotiated with the
Editorial Board.
• Otherlistedtextstyleguidelines.
markers – constitute the majority of metadiscourse markers in the texts. This
comprises approximately 75% of the Estonian data and 67.8% of the Lithuanian
data. Furthermore, the most frequent marker in both datasets is transition markers
(e.g., next, but, however, etc.), accounting for 32.3% and 22.7%, respectively. When
examining the distribution of metadiscourse markers across the various sections
(IMRaD) of the research papers, we observe that the majority of metadiscourse
markers are located in the results and discussion section for both the Estonian and
Lithuanian datasets. 54% of the metadiscourse markers are found in the combined
results and the results and discussion sections in the Estonian sub-corpus, and
51.1% of the metadiscourse markers are found in the results and discussion sec-
tion of the Lithuanian sub-corpus. The lowest number of metadiscourse markers
in both datasets is found in the discussion section in Estonian and Lithuanian:
7.2% and 3.3%, respectively.
122
Table 4. Percentage distribution of Estonian and Lithuanian metadiscourse markers across the various
sections of the journal articles
Sub-corpus Metadiscourse markers %
Metadiscourse markers
represented in each section
of the journals
%
Estonian
Transition markers* 32.3 Introduction 14.5
Code glosses* 16.8 Literature Review 7.8
Endophoric markers* 16.5 Data & Methods 8.1
Evidentials* 5.7 Results 30.2
Frame markers* 3.7 Results & Discussion 23.8
Hedges° 5.9 Discussion 7.2
Boosters° 6.5 Conclusion 8.5
Attitudinal markers° 3.5
Engagement markers° 1.4
Self-mentions° 7.7
Lithuanian
Transition markers* 22.7 Introduction 18.6
Code glosses* 12.2 Theory 12.2
Endophoric markers* 8.4 Data & Methods 6.6
Evidentials* 22.0 Results & Discussion 51.1
Frame markers* 2.5 Discussion 3.3
Hedges° 9.4 Conclusion 8.1
Boosters° 10.5
Attitudinal markers° 2.1
Engagement markers° 9.5
Self-mentions° 0.7
* Interactive category: metadiscourse markers that help to guide readers through the text.
° Interactional categor y: metadiscourse markers that involve the reader in the text.
To determine the patterns of metadiscourse marker usage across both languages
and academic journals, we employed association plots. The plot visualizes the
relationships between journal article sections and the metadiscourse categories,
both in terms of association and dissociation. In the association plots, a red cell
indicates under-representation (i.e., frequencies lower than expected under the
assumption of independence), whereas a blue cell indicates over-representation
(i.e., frequencies higher than expected under the assumption of independence).
The base of each bar represents the degree of support for the association, while
its height is proportional to its significance. In this context, significance denotes a
pattern that deviates significantly from random variation.
Figure 1 presents the metadiscourse baseline models for Estonian and Lithu-
anian in each journal section of the corpus. In the introduction section, the Estonian
metadiscourse baseline model exhibits a positive association with the use of eviden-
tial markers (e.g., according to) and self-mentions (e.g., I), while it demonstrates
a negative association with transition markers (e.g., in addition). In the literature
review section, there is a positive association with evidential markers but a negative
123
association with self-mentions. In the method sections, there is generally a positive
association with self-mentions and negative associations with transition markers,
endophoric markers (e.g., in section one), hedges, and boosters. In the results
sections, there are generally positive associations with transitional markers and
endophoric markers but a negative association with evidential markers, hedges,
engagement markers, and self-mentions. In contrast, the results and discussion
section exhibits positive associations with engagement markers and attitudinal
markers but negative associations with frame markers and self-mentions. The
discussion section of articles shows a small positive association with boosters and
a small negative association with self-mentions. Finally, the conclusion section
displays a small positive association with transitional markers and hedges, as well
as negative associations with endophoric markers and evidentials.
The Lithuanian metadiscourse baseline model offers a more conventional over-
view of the journal sections. For the introduction section, a large positive association
is observed with the use of evidential markers and frame markers, but more nega-
tive associations with transitional, endophoric, hedges, boosters, and engagement
markers. In the theory section, there is also a positive association with the use of
evidential markers and code glosses. More negative associations are found among
the interactional markers, such as hedges, booster, and engagement markers, but
also with endophoric markers. In the data and method section, we find small posi-
tive associations among the interactive markers frame markers and code glosses,
and small negative associations amongst the interactional markers, hedges, and
boosters. In the results and discussion section, there are large negative associations
with evidential and frame markers, as well as a small negative association with
self-mentions. We find more positive association amongst transitional, endophoric
markers, hedges, boosters, and engagement markers. Overall, the discussion section
exhibits a positive association with markers found in the interactional category:
hedges, boosters, attitudinal markers, and self-mentions, but a negative associa-
tion with evidential markers. Similarly, the conclusion section also demonstrates
a positive association with the use of interactional markers: hedges, boosters, and
self-mentions, but a negative association among the interactive category markers:
evidentials, endophoric markers, and code glosses.
Figure 1. Estonian and Lithuanian metadiscourse baseline models (* – interactive category (guiding
the reader through the text); ° – interactional category (involving the reader in the text))
124
Overall, the metadiscourse baseline models of the use of metadiscourse markers
across the various sections of the text indicate significant variations. The most sig-
nificant variation is observed within the interactional category (involving the readers
in the text). More specifically, there is a negative association with metadiscourse
marker use in the Estonian corpus within the results and discussion sections, while
a more positive association is observed for the use of interactional category markers
in comparable sections in the Lithuanian corpus. This observation might imply that
Lithuanian journal articles in linguistics exhibit a more positive trend of writers
engaging readers in the text in comparison to Estonian writers.
3.2. Metadiscourse models of Estonian academic journals
versus baseline model
Since the corpus also distinguishes data from three different journals in both the
Estonian and Lithuanian corpus, we conducted further comparisons between the
specific metadiscourse patterns of these journals and how they compare to the
metadiscourse baseline model. The assumption is that if a specific journal shows
a large degree of variation to the baseline, we may assume that this variation is
because of the type of research these journals accept, which invariably has a dif-
ferent discoursal pattern. Alternatively, the journal may have a strong editorial
preference or pattern which is different from the baseline. As such, we calculated
the metadiscourse model for each journal for each language and compared these
to the baseline model using heatmaps.
Figure 2 shows the association plots and the heatmap of differences between
the Estonian journals: KK, ERÜ, and ESA, and the Estonian baseline model. On the
left side of the figure, the association plots for the specific journals are presented.
On the right side, the heatmaps of the differences which presents the calculation
of the expected values for each count of data in comparison to the baseline count.
Cells with residuals which are close to 0, which means that the observed count in
the specific journal and the expected values are close, are coloured white. Positive
residuals, coloured red, indicate that the observed values in the specific journal
are greater than expected in the baseline model, while negative residuals, coloured
blue, indicate that the observed values in the specific journal are less than what
would be expected based on the baseline model. Additionally, since the heatmap
calculates residuals from the bottom left to the top right, the sections in the heatmap
are presented in reverse order.
In general, KK does not deviate too much from the baseline model. The most
notable finding in the comparison is the negative residual (indicated in blue, sug-
gesting a lower count in the specific journal compared to the baseline model) of
transition markers in the results sections of KK. In other words, in our corpus, KK
has fewer transition markers (such as, but, in addition, and, etc.) in the results
section than one would expect based on the metadiscourse baseline model for
Estonian. ERÜ, like KK, closely aligns with the metadiscourse baseline model,
and the only noticeable deviation (indicated in blue) is also in the usage of transi-
tion markers. However, compared to KK, ERÜ exhibits a negative trend in using
transition markers, specifically in the results and discussion section. Finally, ESA
125
Figure 2. Estonian journal articles (KK, ERÜ, ESA) compared to the Estonian baseline model
126
shares commonalities with ERÜ in the negative residual for transition markers
in the results and discussion section, but overall, not too much variation with the
baselines. What sets ESA apart from the other journals and the baseline model is a
slight inclination towards a positive residual (indicating a higher frequency in the
journal compared to the baseline) of self-mentions (e.g., I, we, etc.) in the methods
section, although this tendency is not very pronounced.
3.3. Metadiscourse models of Lithuanian academic journals versus
baseline model
The same comparative analysis was carried out for the Lithuanian journals. Figure 3
shows the association plots and the heatmap of differences between the Lithuanian
journals K, LK, and TK and the Lithuanian baseline model. Again, the association
plots are presented on the right side of the figure and the heatmaps of differences
(i.e., comparing the journal metadiscourse model with the Lithuanian baseline
model), are presented on the right. Overall, the heatmaps do not display significant
deviations from the specific journal article metadiscourse models in comparison to
the Lithuanian metadiscourse baseline model. However, across the journals, again,
the largest variation is observed in the results and discussion section. Among the
three journals, K, LK, and TK, the most significant variation from the baseline model
is observed in the usage of evidentials (referring to information from other texts,
e.g., according to X, Z states), which generally appears less frequent (indicated by
blue) compared to the baseline model. Given the nature of the results and discus-
sion section, this is not surprising, but interesting, nonetheless. Specifically, you
would expect a lower frequency of references to information from other texts in
the presentation of results, but the inclusion of a discussion could require a few
evidentials to be included. Comparatively, the journal LK, when contrasted with K
and TK, appears to exhibit a lower residual for transition markers in the results and
discussion section, as has also been noted in the Estonian metadiscourse models
of the specific journals. Lastly, the metadiscourse model for the journal K, overall,
exhibits some minor positive residuals (although only slightly) for certain markers
(mainly in the interactional category involving readers in the text) in the discus-
sion section, confirming an earlier observation when comparing the metadiscourse
baseline models of Estonian and Lithuanian, which showed some variation in the
association plots within the interactional category.
3.4. Metadiscourse models across languages and academic journals
To perform a cross-language comparison of metadiscourse markers, we included
the sections from the language-specific metadiscourse baseline models that allowed
for a direct comparison. In other words, we examined introductions, literature
review/theory, methods, results and discussions, discussions, and conclusions.
To make a cross-language specific journal comparison, we took the metadiscourse
model of the Estonian journal ERÜ and the metadiscourse model of the Lithu-
anian journal K, two journals which in the corpus had comparable sections. The
127
Figure 3. Lithuanian journal articles (K, LK, TK) compared to the Lithuanian baseline model
128
purpose of these comparisons is to determine whether the metadiscourse models
exhibit any similarities, suggesting a more universal metadiscourse pattern across
Estonian and Lithuanian, or if they demonstrate variations, either across sections
of a journal, metadiscourse markers, or both.
Figure 4. Estonian metadiscourse model compared to Lithuanian metadiscourse model and
Lithuanian journal K compared to the Estonian journal ERÜ
Figure 4 shows the heatmap of differences between Estonian and Lithuanian.
The heatmap on the left of Figure 4 shows the comparison of the metadiscourse
baseline model of Estonian to the metadiscourse baseline model of Lithuanian.
The heatmap on the right of Figure 4 shows the comparison of the metadiscourse
model of the Lithuanian journal K and the Estonian journal ERÜ. In comparison to
the language-specific comparison, the heatmaps show a less uniform comparison
and similarity between the language-specific baseline model and language-specific
journal. Initially, a direct comparison of the Estonian and Lithuanian models reveals
less variation, with the only significant positive deviation observed in the use of self-
mention (e.g., I, we, etc.) in the methods section. In other words, Estonian writers
tend to include self-mentions much more in methods sections when compared to
their Lithuanian colleagues. Interestingly, when comparing the Lithuanian journal
K to the Estonian journal ERÜ, a different trend emerges. When comparing these
two journals, the majority of positive variations, red-coloured cells, are observed
in the results section, with K having more observations of transition markers, code
glosses, and endophoric markers, within the interactive category of metadiscourse
markers aiming to help guide the reader through the text, and more observations
of self-mentions in the results section in comparison to ERÜ. The first three in the
interactive category might be explained by the type of texts K publishes in their
journal, which might contain more linguistic examples which would need more
writer guidance such as, in addition to .., as seen in Table 2, e.g., and in other
words. ERÜ, on the other hand, might publish articles which more frequently
present results without the use of guidance by authors and reserves the transition
markers for the results and discussion or discussion section, as one would expect.
129
4. Conclusion
Overall, the results of the cross-linguistic and journal comparisons underscore
the value of employing metadiscourse markers. When comparing metadiscourse
models across journals within a specific discipline, such as linguistics, the variation
in the model is relatively small, with only certain sections of the journal displaying
deviations from the general baseline model. It would be intriguing, however, to
extend this analysis to metadiscourse models in journals from neighbouring and
other disciplines, like social science or medical sciences. This exploration might
reveal entirely different patterns or reinforce existing ones. Moreover, when com-
paring across languages, the comparison of disciplines may not yield significant
variations. Instead, most variations could stem from specific cultural or rhetorical
expectations. For instance, the perennial question of whether to use self-mentions
such as ‘I’ or ‘we’ might manifest differently across languages. Nevertheless, des-
pite the disciplinary similarities across languages, comparing similar discipline-
specific journals across languages may reveal more significant variation within
specific sections of a paper and across the various categories of the interpersonal
metadiscourse model. As demonstrated in the comparison between Lithuanian and
Estonian journals, specific strategies to guide readers through the text may be more
essential for certain types of articles that are commonly accepted and published in
those journals, effectively overcoming bias. To validate this assumption, a broader
and more extensive selection of journals may be required.
References
Ädel, Annelie 2018. Variation in Metadiscursive “You” Across Genres: From Research Articles
to Teacher Feedback. – Educational Sciences: Theory & Practice, 18 (4), 777–796.
https://doi.org/10.12738/estp.2018.4.0037
Birhan, Amare Tesfie 2021. An exploration of metadiscourse usage in book review
articles across three academic disciplines: A contrastive analysis of corpus-based
research approach. – Scientometrics, 126 (4), 2885–2902. https://doi.org/10.1007/
s11192-020-03822-w
Cao, Feng; Hu, Guangwei 2014. Interactive metadiscourse in research articles: A comparative
study of paradigmatic and disciplinary influences. – Journal of Pragmatics, 66, 15–31.
https://doi.org/10.1016/j.pragma.2014.02.007
Crismore, Avon; Markkanen, Raija; Steffensen, Margaret S. 1993. Metadiscourse in persuasive
writing: A Study of texts written by American and Finnish university students. – Writ-
ten Communication, 10 (1), 39–71. https://doi.org/10.1177/0741088393010001002
Dahl, Trine 2004. Textual metadiscourse in research articles: A marker of national culture
or of academic discipline? – Journal of Pragmatics, 36 (10), 1807–1825. https://doi.
org/10.1016/j.pragma.2004.05.004
Duszak, Anna 1994. Academic discourse and intellectual styles. – Journal of Pragmatics, 21
(3), 291–313. https://doi.org/10.1016/0378-2166(94)90003-5
Duszak, Anna; Lewkowicz, Jo 2008. Publishing academic texts in English: A Polish per-
spective. – Journal of English for Academic Purposes, 7 (2), 108–120. https://doi.
org/10.1016/j.jeap.2008.03.001
Fløttum, Kjersti; Dahl, Trine; Kinn, Torodd 2006. Academic Voices: Across Languages and
Disciplines. Pragmatics & Beyond New Series, 148. Amsterdam–Philadelphia: John
Benjamins. https://doi.org/10.1075/pbns.148
130
Harbord, John 2018. Language policy and the disengagement of the international academic
elite. – Science Editing, 5 (1), 32–38. https://doi.org/10.6087/kcse.115
Hint, Helen; Leijen, Djuddah A. J.; Jürine, Anni 2022. Eestikeelse akadeemilise teksti tun-
nustest. – Keel ja Kirjandus, 4, 327–353. https://doi.org/10.54013/kk772a3
Hint, Helen; Lemendik, Helena; Johansson, Christer; Leijen, Djuddah A. J. (Forthcoming).
Decoding metadiscourse markers in Estonian academic texts: A language-specific
perspective. – Written Communication.
Hyland, Ken 2005. Metadiscourse: Exploring Interaction in Writing. London–New York:
Continuum.
Hyland, Ken; Tse, Polly 2004. Metadiscourse in academic writing: A reappraisal. – Applied
Linguistics, 25 (2), 156–177. https://doi.org/10.1093/applin/25.2.156
Hyland, Ken; Wang, Wenbin; Jiang, Feng (Kevin) 2022. Metadiscourse across lan-
guages and genres: An overview. – Lingua, 265, 103205. https://doi.org/10.1016/j.
lingua.2021.103205
Jürine, Anni; Leijen, Djuddah; Sinkuniene, Jolanta; Laiveniece, Diāna; Johansson, Christer;
Groom, Nicholas 2021. Academic writing in the Baltic States: Introducing the Bwrite
project. – Educare, 1, 27–37. https://doi.org/10.24834/educare.2021.1.3
Leijen, Djuddah; Hint, Helen; Jürine, Anni (Forthcoming). Academic writing tradition and
how to model it. – C. Donahue, C. Gannett (Eds.), International Writing Studies
Research: Cultivating Collective Research Capacity through International Exchanges
about Higher Education Writing Research. The WAC Clearinghouse International
Series.
Mauranen, Anna 1993. Contrastive ESP rhetoric: Metatext in Finnish-English eco-
nomics texts. – English for Specific Purposes, 12 (1), 3–22. https://doi.
org/10.1016/0889-4906(93)90024-I
Meyer, David; Zeileis, Achim; Hornik, Kurt 2003. Visualizing independence using extended
association plots. – Proceedings of the 3rd International Workshop on Distributed
Statistical Computing. https://www.r-project.org/conferences/DSC-2003/Proceed-
ings/MeyerEtAl.pdf (27.3.2024).
Mur-Dueñas, Pilar 2011. An intercultural analysis of metadiscourse features in research
articles written in English and in Spanish. – Journal of Pragmatics, 43 (12), 3068–3079.
https://doi.org/10.1016/j.pragma.2011.05.002
Peng, Jian-E.; Zheng, Yuting 2021. Metadiscourse and voice construction in discussion sec-
tions in BA theses by Chinese university students majoring in English. – SAGE Open,
11 (2), 215824402110088. https://doi.org/10.1177/21582440211008870
R Core Team 2022. R: A Language and Environment For Statistical Computing. R Foundation
for Statistical Computing. Vienna: R Foundation for Statistical Computing. https://
www.R-project.org/ (19.3.2024).
Ruskan, Anna; Hint, Helen; Leijen, Djuddah A. J.; Šinkūnienė, Jolanta 2023. Lithuanian
academic discourse revisited: Features and patterns of scientific communication. –
Open Linguistics, 9 (1), 20220231. https://doi.org/10.1515/opli-2022-0231
Sollaci, Luciana B.; Pereira, Mauricio G. 2004. The introduction, methods, results, and
discussion (IMRaD) structure: A fifty-year survey. – Journal of the Medical Library
Association, 92 (3), 364–371.
Vande Kopple, William J. 1985. Some exploratory discourse on metadiscourse. – College
Composition and Communication, 36 (1), 82–93. https://doi.org/10.2307/357609
Wickham, Hadley 2007. Reshaping data with the reshape package. – Journal of Statistical
Software, 21 (12), 1–20. https://doi.org/10.18637/jss.v021.i12
Wickham, Hadley 2016. ggplot2: Elegant Graphics for Data Analysis. 2nd ed. Cham: Springer.
https://ggplot2.tidyverse.org
131
metadiSkurSuSe muStrite keeltevaheline
võrdluS: valdkondlikud SarnaSuSed
ja artiklioSade erinevuSed
Djuddah Leijen1, Helen Hint1, Helena Lemendik1,
Baiba Egle2, Anna Ruskan3, Christer Johansson4
Tartu Ülikool1, Liepāja Ülikool2, Vilniuse Ülikool3, Bergeni Ülikool4
Artikkel käsitleb metadiskursuse markereid eesti ja leedu keeleteaduslikes aja-
kirjades. Meie eesmärk on 1) võrrelda kõigi metadiskursuse markerite üldist kasutust
tekstides keeliti, otsides nii keeltes kui ka ajakirjade kaupa ilmnevaid erinevusi, ning
2) leida keeliti võimalikud sarnased ja/või erinevad mustrid ja selgitada välja, kas
need mustrid tulevad esile ka eri ajakirjades. Tulemused näitavad, et eesti autorid
kasutavad artiklite meetodiosas enesele osutamisi (ingl self mentions) sagedamini
kui leedu autorid. Ajakirjade võrdluses leidub leedu ajakirjas Kalbotyra rohkem
sidususmarkereid (transition markers), täpsustavaid markereid (code glosses) ja
tekstisiseseid viiteid (endophoric markers), samal ajal kui Eesti Rakenduslingvistika
Ühingu aastaraamatus jäävad sellised lugejat juhatavad markerid rohkem tulemuste
ja arutelu osadesse. Vaatamata valdkondlikele sarnasustele ilmneb siiski erinevusi
teatud artikliosades ja interpersonaalse metadiskursuse kategooriates nii keeliti kui
ka ajakirjade lõikes. Uurimus heidab valgust metadiskursuse kasutusmustritele
ja nende rollile eri keeltes ja akadeemilistes kontekstides ning võib olla edaspidi
suunanäitajaks mitteingliskeelsete akadeemiliste tekstide uurijatele ja praktikutele.
Võtmesõnad: metadiskursus, teadusartiklid, IMRaD, eesti keel, leedu keel
Djuddah Leijen (University of Tartu) is interested in a range of research topics, including
research on text production, writing research, genre studies, PhD writing processes, peer
review, and modelling writing structures.
Lossi 3, 51003 Tartu, Estonia
djuddah.leijen@ut.ee
Helen Hint (University of Tartu) is interested in a range of research topics, including the
pragmatics of referential devices, writing studies, linguistic features of Estonian academic
texts, and Estonian language didactics.
Jakobi 2, 51005 Tartu, Estonia
helen.hint@ut.ee
Helena Lemendik (University of Tartu) investigates the specific features of Estonian academic
texts, focusing primarily on the linguistic aspects from a metadiscursive perspective.
Jakobi 2, 51005 Tartu, Estonia
helena.lemendik@ut.ee
Baiba Egle’s (Liepāja University) research interests include the Latvian language of science,
multilingualism in science, language policy, as well as various aspects of text linguistics and
translation studies.
Lielā iela 14, 3401 Liepāja, Latvia
baiba.egle@venta.lv
132
Anna Ruskan’s (Vilnius University) research interests include evidentiality, epistemic
modality, grammaticalization, corpus linguistics and contrastive linguistics.
Universiteto str. 5, LT-01513 Vilnius, Lithuania
anna.ruskan@flf.vu.lt
Christer Johansson’s (University of Bergen) research interests include experimental
psycholinguistics, statistical data analysis, and computational linguistics. He investigates relevant
factors for human language processing, with one goal being to aid language users through
the design of language tools adaptable to individual needs. His current research focuses on
codeswitching and how the interaction of vision and speech affects visual change blindness.
Sydnesplassen 7, 5007 Bergen, Norway
christer.johansson@uib.no
