Evaluative stance in master's theses discussion sections: the case of Iranian graduate student writers

1. Introduction

Academic writing is a persuasive endeavor in which academics do not merely produce a text to represent an external reality (Hyland, 2011) but use language to build and negotiate social relations (Hood, 2004, 2010; Hunston, 1994; Hyland, 2005). In other words, to gain credibility and establish solidarity with the community and readers, academic writers need to present their voice and position themselves towards alternative views (Geng & Wharton, 2016; Gray & Biber, 2012; Hyland, 2005). Employing evaluative language can assist writers in fulfilling these requirements in their texts (Loghmani et al., 2020). Evaluative language involves expressing a speaker's attitudes, feelings, judgment, and commitment toward the message (Biber & Finegan, 1989), which entails appropriately presenting oneself and their argument to audiences with particular expectations (Hyland & Zou, 2021).

While a significant body of research has emphasized the importance of evaluative language in academic writing (Geng & Wharton, 2016; Gray & Biber, 2012; Hyland, 2005; Lam & Crosthwaite, 2018), there has been a paucity of research in this regard, and the little research shows that novice writers often struggle with constructing evaluative stances (Geng & Wharton, 2019; Hood, 2004; Hyland, 2012; Xie, 2016). Moreover, linguistic and rhetorical discrepancies make these challenges even more common for non-native novice writers (Hood, 2004, 2010; Loi et al., 2016). Those with different sociocultural backgrounds might have different conceptions about projecting a credible point of view and citing required resources (Hyland, 2010). Accordingly, the evaluation concept might appear odd for those educated in other educational cultures. As such, exploring the means through which L2 novice writers express evaluation in academic writing in English could be a resourceful avenue of research on learner writers. Despite some preliminary research on how English as a foreign language (EFL) learners express evaluation in English academic writing (e.g., Ansarin & Aliabdi, 2011; Gholami et al., 2014), many of these studies focused on exploring lexico-grammatical features, only a few have explored evaluative elements at discourse level. Further, relatively few studies have investigated master's theses as previous studies have mainly focused on either undergraduate argumentative texts and assignments or Ph.D. theses. A master's thesis, as Hyland (2004, p.134) identifies, can be seen as a "high stakes genre at the summit of a student's academic accomplishment", which is probably the "most significant" and "formidable" piece of writing with an "intimidating length", and "exacting expectations" that any student will ever do to display their competence and gain the acceptance of their discourse community. Hence, investigating engagement elements in master's theses can be beneficial in identifying patterns of use and challenges in the employment of these resources.

Martin and White's (2005) Appraisal framework was selected for assessing evaluative stance. This discourse-level framework divides evaluative resources into three broad areas: Attitude, Engagement, and Graduation. Attitude, comprises three interacting domains, looks for the expression of emotions, ethical judgments of behaviors, and aesthetic evaluation of things. Graduation deals with the intensity or amount of a gradable resource. It also deals with the degree to which a non-gradable phenomenon matches the prototypical instance of a category. Lastly, Engagement concerns writers' stances towards other viewpoints and their strategies to signal or predict the reader's reaction (Martin & White, 2005).

Although the appraisal framework integrates numerous evaluative categories (Martin & White, 2005), this study focuses solely on ‘engagement’ as a theoretical framework. The main reason for this focus is that epistemic aspects of evaluation, covered under the engagement sub-system, are viewed as "considerably more important in academic research writing than the attitudinal meanings" (Gray & Beaver, 2012, p.19). Additionally, while many evaluation frameworks prioritize the writer's perspective, the engagement sub-system considers both the writer's and reader's voices and those of other researchers, aiming to predict the reader's responses and make certain strategic inferences (Cheng & Unsworth, 2016).

Research results are not mere facts to be transferred; they must be explained, interpreted, and argued in a way the reader finds plausible (Hyland & Zou, 2021). Discussions are expected to encompass a wide range of engagement resources as their function in academic writing is to convince readers to agree or disagree with the presented positions (Swain, 2010). Writers in this section need to present the findings, make evaluative comments on them, relate them to theories or other research findings, and engage with alternative voices. One of the potential goals of producing a dissertation as a persuasive genre is constructing "a disciplinary aligned presence" and shaping "a credible self" (Hyland, 2012, p.135). To be taken seriously by disciplinary insiders (Geng & Wharton, 2016), graduate student writers need to implement evaluation and voice in their theses. The current study explores the discussion chapters of master's theses as one of the main venues for expressing stance (Hyland, 2012).

For the purposes of this research, a single field (i.e. applied linguistics) was selected to decrease the potential effects of disciplinary differences (Cominos, 2011). Among others, this field was selected firstly because the stance-taking most frequently occurs in soft sciences such as applied linguistics (Chan, 2015). Secondly, most students of applied linguistics across the world are taught through English as a medium of instruction (EMI) and write their assignments and theses in English. This would make it possible for researchers to look at authentic examples of textual productions in English by these learner writers. Discussion chapters in dissertations constitute interpretations of the results in the context of study questions and literature review, highlight the implications, state the limitations, and make recommendations. Students have been found to struggle with producing very successful discussion chapters as they seem to have more difficulty with understanding the function of the discussion chapter (Bitchener & Basturkman, 2006).  The current research aims to investigate the use of the evaluative stance in a corpus of master's theses' discussion chapters written by English L2 Iranian students. More specifically, it seeks to answer the following research questions:

1. Which engagement resources are used in the discussion chapters of master's theses written by English L2 Iranian students of applied linguistics?
2. Which evaluation patterns are salient in the discussion chapters of master's theses written by English L2 Iranian students of applied linguistics?

2. Theoretical background: The engagement domain within appraisal theory

The Engagement system within Appraisal theory examines how writers adopt a stance toward the value positions presented in the text and toward readers. It distinguishes the monoglossic and the heteroglossic discourse (Martin & White, 2005). Unlike a monoglossic context, in which other voices or alternative positions are not overtly recognized or engaged, heteroglossic options are generally concerned with the writer's alignment towards other value positions presented.

The heteroglossic resources are further classified into two broad categories demonstrating whether an utterance allows for dialogically alternative positions and voices (i.e. dialogic expansion) or, in contrast, acting to challenge the scope of such voices (i.e. dialogic contraction). An overview of the contractive and expansive Engagement resources used in the current study is presented in Figure 1.

Figure.1. Engagement System (Adapted from Martin & White, 2005, p.134)

The Engagement resources for dialogic expansion include Entertain and Attribute, while those of dialogic contractions include Disclaim and Proclaim. Drawing on data from our corpus of discussion chapters, in what follows, we will elaborate the engagement system in more detail.

2.1. Evaluation studies within appraisal theory in academic writing

Exploring evaluation in expert writers' texts and comparing expert and novice writers' employment of these resources are two well-established lines of evaluation studies. Hood's (2004) early work compared the introductory sections of published academic research papers and undergraduate dissertations in the L2 context in terms of how each group attempted to bring objectivity to their arguments. The results revealed that while both student and published writers open their text by an "Observer Voice" (Hood, 2004, p. 198), student writers prefer to use this voice role more extensively, possibly where they were unable to offer a critique of other research, either due to the lack of access to relevant research or lack of confidence in critiquing it. However, a shift into the "Critic Voice" is evident in the published texts, even if the Observer Voice is the dominant voice. In her later works, Hood explored evaluative prosody in the introduction sections of research articles written by experts in applied linguistics (Hood, 2006), social sciences, and humanities (Hood, 2010). The results showed that academic writers preferred the explicit way when evaluating the object of their study but the implicit way when evaluating other research (Hood, 2010).

Comparison of EAL (English as an additional language) and English L1 learners to expert writer's use of engagement resources of reporting verbs use and building intertextuality through their reporting verbs choices showed that the reporting verb use of both EAL and English L1 learners were similar in that, they strongly tended to take more neutral attribute providing no overt stance (Liardét & Black, 2019). Pedagogically speaking, although comparative studies can help writing instructors pave the way for students in their transition toward making more efficient arguments and positioning themselves in their research (Hood, 2004), novice and professional writers seem to differ considerably in their understanding of academic conventions and practices, which might make direct comparisons unhelpful (Hyland, 2002, 2012). Moreover, one must be careful in comparing two different genres (e.g., Research articles and argumentative essays) since they might not be precisely identical in terms of organization (Hewings & Hewings, 2002). On the other hand, focusing on a single genre, several studies investigated intertextuality in the discussion section of doctoral dissertations written by native English speakers in applied linguistics (e.g., Geng & Wharton, 2019; Loghmani et al., 2020) and reported a tendency by the authors to employ more contractive resources than expansive ones.

Another wave of evaluative patterns studies emerged to track the potential influence of cultural factors. They specifically focused on identifying cross-cultural differences in strategies used by writers with the same expertise level but different language backgrounds. An analysis of stance patterns in the conclusion sections of English and Malay research articles suggests that these scholars experience evaluation differently due to linguistic, contextual, and potential social-cultural influences within their respective academic discourse communities (Loi, et al, 2016). Geng and Wharton (2016) investigated the discussion section of doctoral dissertations written by L1 Chinese and L1 English writers. However, they drew different conclusions, i.e. no statistically significant differences in evaluative resource choices were found, suggesting that first language background might not be a variable that influences evaluative language choices.

The findings on the effect of cross-cultural differences in the use of evaluation patterns are further supported by studies examining argumentative essays written by Asian L2 and English L1 writers (Lam et al., 2018), writers of English and Chinese research articles (Xu & Nesi, 2019), and research articles by expert writers of English L1, Spanish L1, and Spanish L2 backgrounds (Sheldon, 2018).

Although contrastive analyses in L1 and L2 contexts can prove beneficial in shedding light on the perceptions of novice writers of "academic conventions" and showing them how to accommodate their cultural practices (Hyland, 2002, p.1096), adopting an academic voice, still seems a highly complex concept for novice writers, since they tend to focus more on generating their ideas on the topic and may not be fully aware of the interactive aspects of the text and the role of various linguistic choices in constructing this interaction (Hyland, 2012; Liu and & McCabe, 2018). Due to these underlying difficulties, which put students in a rhetorically and interpersonally disadvantageous position (Hyland, 2002), a growing body of literature has explored evaluative pattern use in student genres. For example, Miller et al. (2014) conducted a 4-year longitudinal study in which Engagement elements of high and low-graded argumentative essays written by college-level students in the Middle East were explored. While higher-graded essays showed a more consistent argument and an objective voice, the low-graded essays were reported to prefer a single subjective voice.

While adapting the pragmatic conventions of academic discourse feels confusing even for native English students, it seems particularly challenging for non-native students who are socialized with different cultural norms that might conflict with their familiar conventions (Hyland, 2002, 2012). To explore these difficulties, Xie (2016) analyzed the Engagement resources in literature reviews of Chinese EFL learners' master's theses and revealed that Chinese students preferred to express evaluation more explicitly than implicitly. Swain's (2010) analysis of attitude and engagement elements in a small corpus of non-native English speaker (NNS) undergraduate discussion essays revealed that high-scoring essays employed a full range of engagement resources suggesting that making a balance between informative and evaluative propositions is a defining characteristic of a good essay. Moreover, Chung et al. (2021) compared Vietnamese students' evaluative expressions in using appraisal across two groups of texts written by the same writers, one in their mother tongue and the other in English as a second language. The results revealed that although multi-voiced reasoning was generally preferred, the writers frequently adopted bare assertions.

3. Method

Following the Engagement system of Appraisal framework (Martin & White, 2005), we conducted a functional analysis of available engagement categories and elements. A corpus consisting of discussion sections of 40 master's theses (over 65000 tokens) in applied linguistics written between 2010-2019 was collected from the theses databases of four top-ranking Iranian universities (University of Tehran, Tarbiat Modarres University, Shahid Beheshti University, and University of Isfahan). These top-ranked state universities which were chosen randomly offered graduate degrees in applied linguistics and made digital access to their theses available.

The location of the discussion chapter in the theses varied due to university policies (i.e., Chapter 3 or 5). Nevertheless, all the selected theses were checked to have a separate discussion header to allow for more consistency. The discussions were extracted into separate text files, and their extra parts, including subheadings, introductory sentences, and illustrations such as figures, tables, and endnotes, were eliminated (Xie, 2016). Moreover, direct quotations were excluded as they do not reflect the author's evaluations. (See Table 1 for descriptive statistics of the selected samples). Due to the varied length of the texts, the normalized frequency per 1000 words of Engagement categories was calculated.

Table 1- Selected Samples in the Corpus of Discussion Chapter

We conducted a pilot study to make decisions about the coding process. Hence, two theses were randomly selected. The related abstract, introduction, and discussion sections were read to understand author's general stance and engagement resources. Starting from engagement realizations as they emerged from the text and working back to what the writer probably had in mind (Martin & White, 2005), we employed a combination of bottom-up and top-down approaches for analyzing the samples (Uschold & Grüninger, 1996). After the pilot study, the analyzed samples were shared with an experienced analyst, and agreement was sought on the coding procedure and potential conflict areas. Then, a coding sheet was developed which included the Engagement system adapted from the original Appraisal framework. Finally, all the texts were coded manually according to the adapted framework (See Figure 1 above)

The unit of analysis was clause since, in many instances, elements were realized at the clause level. The Engagement resources within a single clause were assigned to their related categories separately (Martin & White, 2005, p.155). Moreover, following Bakhtin's (1981, cited in Martin & White, 2005, p.92) notion of dialogism and other studies (Geng & Wharton, 2016; Xu & Nesi, 2019), every statement was considered attitudinal in a way that all the clauses which did not fit in any of the heteroglossic categories, were counted as monoglossic.

Following White (2012) and Geng and Wharton's (2016, p. 86) proposal of a new dialogic contractive option labelled "Justify", and "Justify-by-Data", an additional category of Reference-to-Research-Data was added in the present study. This category involves authors' references to their research data and participants' viewpoints as external support for their assertions. This inclusion occurred after the second coding process by the authors which will be elaborated in the method chapter. During the initial analysis, these options were classified as Attributes as they conveyed external ideas. However, when analyzed by a second analyst, they were found to be Contractive. The emergence of this category was necessitated by the unique contractive behavior observed in the context, specifically in reference to research data conveyed as data. Thus, we proposed this extension to better capture the intricacies of our examined part-genre. Similar categories found in recent literature (e.g., Geng & Wharton, 2016; White, 2012) led to separating these options into separate categories. The reason for using this heading rather than Justify-by-Data is that not all instances of this category were used for justification. Instead, a group of the instances was used as evidence affirming the credibility of the presented viewpoint, thus closing the dialogic space, while the other group was intended to report or reference the data gathered through research merely. Although the formulations under this category looked pretty similar to those of Pronounce, in most of the Reference-to-Research-Data instances, there was a tendency by the author to employ reporting verbs typically associated with Distance formulations, which may indicate the author's reluctance to take responsibility for those values coming from participants. Pronounce, as Martin and White (2005) explain, involves an author's "intervention or interpolation" (p.127), which is why these instances are not classified under Pronounce. These instances also resemble the monoglossic statements used for reporting the research results. However, it must be mentioned that these options convey the external source's opinions, even though the external source is not from other scholars. Overall, these instances naturally signal the existence of alternative voices and cannot be categorized under the monoglossic group (Martin & White, 2005). Moreover, following Martin and White (2005), two extra options of 'hearsay' and 'covert attribute' were added to the 'attribution' category.

To minimize subjectivity issues common in text analysis (Connor & Mauranen, 1999) and ensure coding accuracy, the entire analysis was repeated a couple of weeks after the initial coding. Intra-coder reliability was .91. Further, following Mackey and Gass (2005), who suggest “rater reliability can be established with as little as 10% of the data” (p. 243), nearly 20% of the corpus (i.e., two texts from each university) were analyzed by a graduate applied linguist who was familiar with the Appraisal System. Inter-coder reliability index was .73. To resolve conflict areas, the first researcher consulted the co-researcher and consensual agreement was reached. Where no agreement could be reached (3 instances), we consulted a professor of linguistics to resolve ambiguities based on his expert judgment. Based on our discussions, we took the following considerations into account in our analysis:

1- Although the elements were primarily classified based on their linguistic form, the co-text was considered in identifying relevant functions. For instance, the author's alignment did not always correspond to the selected citation verbs, necessitating consideration of the overall context rather than a single reporting verb. Consequently, verbs such as argue were sometimes classified as Distance, sometimes as Acknowledge, and in rare cases, Endorsement.

2- Data conveying participants' viewpoints were initially counted as Attribute as they convey an external idea; nevertheless, based on their contractive behavior in the context, they were later considered as a separate category labelled Reference-to-Data under contractive resources.

3- Statements like The results of this study are in line/in contrast to were marked as explicit Endorsement and Explicit counter samples, respectively.

4- Conclusions derived from the study's results were articulated via statements like The results show that, Data analysis revealed that were counted as Pronouncement as they somehow represent the author's commitment to the conveyed proposition.

3.1. Dialogic expansion

Entertain

Entertain refers to those resources which indicate the writer's presented position is associated with an individualized point of view and only one possible position among many others, thus opening up the space for dialogic alternatives. See the following statement from the corpus as an instance:

• The possible reason [Entertain: Reasoning] for such inconsistencies is maybe due to different levels of proficiency of learners. (Text 10: p.111)

Attribution

Attribution encompasses formulations dealing with external sources and is generally associated with direct and indirect speech reports. Attributions consist of formulations that refer to external sources and are mainly associated with direct and indirect speech reports. While in Acknowledge, no specification is available as to where the authorial voice stands concerning the presented proposition, in Distance, the author clearly attempts to disassociate themselves from the presented external material thus refuses to take responsibility for the conveyed proposition. Moreover, using Covert Attributes, authors can indirectly lead the readers towards appraising the attributed materials as either "highly credible and warrantable, or alternatively, as dubious and unreliable." (Martin & White, 2005, p.116). Other instances of Attribution include "Hearsay" in which no specific source is specified for the attributed material (See examples 2-5 from our corpus):

• "Non-semantic information is useful, and perhaps essential, in long-term memory" (p. 71) was what they concluded. [Attribute: Acknowledge] (Text4: p.69)
• Determining the effectiveness of GOs for vocabulary learning in Tsubaki‘s study was only mildly successful [Attribute: Distance] .... (Text 35: p.82)
• As regards the effect of song on retention of materials, almost any article or book [Covert Attribute: High credibility] that talks about the application of songs in the classroom… (Text12: p.85)
• Science content has been found [Attribute: Hearsay] to be particularly effective for engaging language learners. (Text 19: p.53)

3.2. Dialogic contraction

Disclaim

Disclaim resources, including two categories of Deny and Counter, refer to those meanings which close the space for alternatives by directly rejecting (i.e. Deny) or replacing them (i.e. Counter). The following example displays the use of Denial and Counter in the corpus:

• While [Disclaim: Counter] they wished they could speak English fluently, their L2 ideal self was not strong enough [Disclaim: Deny] to make them do any tangible action. (Text 40: p.87)

Proclaim

Using Proclaim resources, the authors display their position, emphasis, and interventions toward other positions. These resources are contractive in that the author emphasizes what the alternative positions might be against, thus reducing the scope of those opposing views instead of overtly rejecting or replacing them. Through Concur formulations, writers overtly present their agreement or support with alternative propositions. The Endorsement refers to the formulations through which the author confirms that external propositions are valid and maximally warranted. Pronounce formulations involve authorial emphases and interventions. Reference-to-Research-Data was included under Proclaim category (See examples 7-8 for some Proclaim resources):

• So, it is clear that [Proclaim: Concur] as the proficiency level of students goes higher, they need less CS. (Text 3: p.93)
• Studies in L2 listening by Berne, 1992; In'nami, 2006; Teng, 1999 also proved that [Proclaim: Endorse] answering multiple-choice items was easier than answering open-ended items. (Text 18: p.76)

4. Results

4.1 Engagement Resources

3533 Engagement resources were identified in the corpus (i.e., 54 instances per 1000 words). The frequency, percentage, and frequency per 1000 words of the main classes of these resources are presented in Table 2.

Table 2- Distribution of Monoglossia vs. Heteroglossia in the Corpus of Discussion Sections

Note. f= Frequency; p=Percentage; f/1000 = Frequency per thousand words

We discovered that discussion sections of examined master's theses include almost three times more heteroglossic (74%) than monoglossic (26%) resources. Heteroglossic realizations occur, on average, every 25 words. To discover any significant differences between monoglossic and heteroglossic frequencies, related-samples Wilcoxon Signed Rank test was run (Table 3), and the difference between monoglossic and heteroglossic resource frequencies was statistically significant.

Table 3- Related-Samples Wilcoxon Signed Rank Test (Monoglossia vs. Heteroglossia)

The frequencies of various heteroglossic resources (See Table 4) show that dialogic contractions (53%) are slightly more favored than dialogic expansions (47%), although the variation is not substantial. Although the difference was close to significance (i.e., p=.056), no significant differences were found between contractive and expansive resources in the examined corpus (See Table 5).

Table 4- Distribution of Heteroglossic Resources in the Corpus of Discussion sections

Note. f= Frequency; p=Percentage; f /1000W = Frequency per 1000 words

Table 5- Related-Samples Wilcoxon Signed Rank Test Summary Contraction vs. Expansion

Despite the higher frequency of contractive resources, the results showed that the two most often used heteroglossic categories were among the Expansive resources, i.e., Acknowledge and Entertain. Figure 2 gives a more detailed schema of heteroglossic subcategories' distribution in the current corpus.

Figure 2- Frequency per 1000 words of Heteroglossic Resources in the Corpus of Discussions

Acknowledge as an Expansive resource ranked highest in terms of frequency (20%) in the whole discussion corpus, while Entertain (17%) and Pronounce (17%), with subtle difference in frequency, were the second and third most commonly used resources respectively. Concur (%1), Hearsay (2%), and Covert Attribute (3%) were respectively the least frequently used.

Except for Concur and Hearsay (See Table 6), no significant differences were found between high-frequency (i.e., Acknowledgement, Entertain, and Pronounce) and low-frequency resources (i.e., Hearsay and Covert Attribute) in the corpus.

Table 6- Related-Samples Wilcoxon Signed Rank Test for Highest and Lowest frequencies

• Functional analysis of patterns of engagement resources use

In what follows, a more detailed examination of patterns of use of engagement resources is presented and discussed.

4.2.1 Monoglossic resources' use

The current study distinguished monoglossic resources into two broad categories, non-propositional (occupying 52% of instances; Example 9) and propositional monoglossic resources (48%; Example 10). The former constitute metalanguage, intertextual elements, and rhetorical strategies which do not mainly convey a kind of assertion, viewpoint, or interpretation, and the latter include those single-voiced utterances in which a viewpoint is conveyed, or the author potentially had the chance of employing heteroglossia, but preferred monoglossia.

• Having collected the intended data and analyzed them, a set of findings were resulted which would be discussed and compared with the results of previous studies in the present section [Mono: Non-Propositional]. (Text40: p.82)
• The higher the age, the more developed the cognitive capacity, so the less gestures and speech are synchronized [Mono: Propositional]. (Text 34: p.73)

Our finding show a relatively high rate of propositional monoglossia which reflects the general findings in the literature on the prevalence of monoglossic resources in academic writing of student writers (Xu & Nesi, 2019; Geng & Wharton, 2016). This consistency suggests that academic writers, regardless of their linguistic and cultural backgrounds, tend to use monoglossic resources extensively, indicating their appreciation of the value of these resources in conveying authoritative knowledge claims in academic writing. As Xu and Nesi (2019) note, not all arguments in academic writing are expected to be disputable, and monoglossic resources may play a crucial role in presenting facts and knowledge.

4.2.2 Heteroglossic resources' use

Attribution

Attribution was the most frequently used heteroglossic resource in the corpus occupying 29% of all heteroglossic instances (See Table 4). Along with the typical reporting verbs such as said, reported, and stated (e.g., in Example 11), and direct and indirect quotes (e.g., in Examples 12 & 13), Acknowledge instances mostly (Nearly %75) appeared with verbs like distinguished, recommended, concluded, put forward, clarified, encouraged, discussed (e.g., in Example 13), and phrases such as according to X, In X's view, based on X, to X's belief, reached to the conclusion that.

• Nikoopour, Farsani & Neishabouri (2011) stated that [Attribute: Acknowledge] existing research on LLS has heavily relied on learners’ strategy use in second language contexts. (Text31: p.70)
• Chief among these considerations is whether or not to provide direct or indirect feedback in error correction (Ferris & Hedgcock 2005) [Attribute: Acknowledge]. (Text39: p.76)
• Generally, in any model at least two measures per factor are required, although three measures per factor are usually recommended: “Two might be fine, three is better, four is best, and anything more is gravy” (Kenny, 1979, p. 143; see also Kline, 2015). [Attribute: Acknowledge] (Text 2: p.74)

Nevertheless, similar to monoglossic resources, nearly 25% (131 out of 537) of Acknowledge instances did not reflect any viewpoint but rather used for providing peripheral information about the background of the external resource, reporting its results and findings, describing methodological procedures, providing a brief intertextual introduction for the proposition, or even an evaluative comparison with the current study. Of these, nearly 7% (18 of them) were in the form of Borrowing terms or concepts, and 17% of instances included all other examples of non-propositional acknowledgments (See Examples 14 &15):

• Nagata (1999) found significant differences between MCG group and single gloss group. [Acknowledge: Results and Findings] He has conducted the study in a multimedia setting in which feedback was given to student after they chose the wrong options. [Acknowledge: Procedure] (Text 10: p.112)
• Truscott published an article in response to her criticisms [Acknowledge: introduction to debate] maintaining that Chandler's study provided further support for his original thesis. [Acknowledge: Propositional] (Text 5: p.80)

The findings reflect realizations of Attribution (e.g., of borrowing terms, concepts, models, and theories) found in doctoral dissertation discussions (Geng & Wharton, 2019) and in research articles by Persian and English writers wherein writers tended to provide background information rather than resolve conflicts with previous researches (Sadeghi and & Alinasab, 2020). 

These frequent propositional and non-propositional Acknowledgements can be seen as graduate students’ attempts to display their work as credible by linking them to previous similar works, but may also show their inadequate experience in employing influential citations. 

For Distance Attribution, writers generally preferred more varied and context-dependent means by employing signaling terms, which are not generally associated with the Distance subcategory, but in some way evoke an attitude and convey a sense of disalignment (See Examples 16 & 17). Realizations through reporting verbs were surprisingly rare and revolved mainly around a limited number of verbs, such as claimed and asserted.

• … Nevertheless, most of the criticisms that have been levelled against Truscott are rooted in a misunderstanding [Attribute: Distance] of his 1996 case against grammar correction. (Text5: p.79)
• As mentioned earlier, in the literature review section, Razmjoo (2008), Mahdavy (2008) or Saricaoglu and Arikan (2009) could hardly prove [Attribute: Distance] MI to be a significant predictor of English language proficiency … (Text30: p.83).

These infrequent and limited ranges of Distance realizations may show student writers’ hesitation about overt disalignment with other external voices, mainly to fend off the risk of being challenged by examiners or readers. Realizations of covert attributes mainly denote high credibility via getting support from multiple external sources (Example 18), except for a few instances that showed that the debatable topic gets support from a few studies (Example 19).

• The use of technology in language teaching has generated extensive debate in the last twenty years. Many educators advocate the enhancement of learning with technology [Covert Attribute: High credibility, multiplicity of sources]. (Text24: p.63)
• To the best of researcher‘s knowledge, no literature was found concerning this question [Covert Attribute: Low credibility]. (Text3: p.97)

 Finally, Hearsay was among the least frequently used heteroglossic resources in the corpus (47 instances; 2%), which was anticipated as it is not a typically expected engagement element within academic and formal texts (Loghmani et al., 2020).

Entertain

Entertain as the second most often used heteroglossic resource constituted 17% of all heteroglossic resources (see Table 4). Little diversity was evident in the employment of Entertain elements which were more often conveyed via modal auxiliaries (e.g., may, could), auxiliaries of obligation (e.g., need to, should), or sometimes through mitigation devices or postulations (e.g., it seems, apparently); and modal adjuncts (e.g., perhaps, probably), or modal attributes (e.g., possible, likely), when writers tried to interpret the findings and/or present reasons (Example 20), conclude (Example 21), make generalized claims based on the findings, and finally provide suggestions and directions.

• The possible reason [Entertain: Reasoning] for such inconsistencies is maybe due to different levels of proficiency of learners. (Text 10: p.111)
• So it could be concluded [Entertain: Conclusion] that gestural input was more effective in transferring information. (Text 34: p.73)

Moreover, a paragraph-level pattern was also detected in which the author presented a challenge as a potential alternative view (i.e., Entertain), attributing it to an imaginary external voice called One, and most often employed Distance to "detach [themselves] from responsibility for what is being reported." (Martin & White, 2005, p.113). This way, the author indicates that the presented challenge "is one of a number of possible positions" but at the same time, refused to explicitly confirm it (Martin & White, 2005, p.104). As can be seen in example 22, the author seems to Deny that the input enhanced based on the guidelines provided by Sharwood-Smith (1993) can lead to the right level for noticing; and appears to believe that the failure in learning the target speech act in the current study might be the result of applying Sharwood-Smith's guidelines. Meanwhile, the author refuses to explicitly take responsibility for the validity of the presented view, thus prefers to attribute it to an imaginary external source, even though it is originally an authorial statement.

• Alternatively, one could [Expansive: Entertain] argue that [Expansive: Attribute: Distance] the input enhanced based on the guidelines provided by Sharwood-Smith (1993) and used in the present study do not amount to the right level for noticing [Contractive: Disclaim: Deny] and consequently learning of the target speech act to take place. [Contractive: Disclaim: Deny] (Text 38: p.83)

This strategy can be seen as a politeness and face-saving strategy (Izadi, 2014) in which the author prefers not to disaccord with previously presented viewpoints but to respect the potential opposite view and to give space to them (Izadi, 2023).

Proclaim

Along with a small number of typical Pronouncement realizations via intensifiers like really, indeed, in fact (Martin & White, 2005, p.127; Example 23), we identified several recurrent statements by authors to promote their findings while simultaneously reducing the dialogic space for others who may challenge the author's proposition (Example 24):

• Indeed, [Proclaim: Pronounce] they were more divergent with regard to syllabus. (Text 23: p.62)
• We found that [Proclaim: Pronounce] teacher' social and professional identity, concerning teachers' perceptions of teaching, is seen by the students to be a predominantly practical endeavor. (Text 21: p.96)

The frequent use of second type Pronouncement statements is consistent with the findings of Sadeghi and Alinasab (2020), and their frequency over the typical realization through intensifiers may indicate the examined students' preference to use these options primarily in reporting their new knowledge or findings in their study (Sadeghi & Alinasab, 2020), avoiding bold authorial interventions which according to Swain (2010) can be seen as a characteristic of weak student essays.

Endorsement instances in the current corpus were mainly conveyed via the typical Endorsement verbs such as demonstrated, showed, indicated, found, and proved (Examples 25 & 26), often along with an explicit announcement by the author at the beginning or end of their reference to external voices, mainly to report that the results of a set of studies are in favor of their findings. An example of these 'Proof Statements' (Parkinson, 2011, p.166) is presented below:

• First of all, the findings are in agreement with the findings of other attachment priming studies [Proof Statement] (Shaver & Mikulincer, 2007 for a review) that have demonstrated [Proclaim: Endorse] the activation of the secure base schema in participants through priming procedures. (Text 25: p.65)
• The results of this study corroborate the findings of previous studies [Proof Statement] which indicated that [Proclaim: Endorse] learners had positive attitude toward English (Falout et al., 2009; Kaivanpanah & Ghasemi, 2011) (Text 40: p.84)

A potential reason for these learners' tendency to relate their claims to the already published literature is adding credibility to the current work and building on previous works to reinforce one's arguments (Murray, 2011; Parkinson, 2011). Hence, they may want their readers to accept their standpoints readily.

Among the instances found for the Reference-to-Research-Data category (%3.5), almost half of the instances (i.e., 44 out of 91 instances) were used as evidence affirming the credibility of the presented viewpoint, and thus for closing the dialogic space, while the other half (i.e., 47) was merely used to report or reference the data gathered through research:

• As Eslami (2010) claimed [Attribution: Distance], students' low proficiency in English … are likely to affect teachers' practice. Many of our interviewed teachers claimed [Proclaim: Reference by Data, Attribution: Distance] that since their students were not proficient in English, they could not utilize many strategies .... (Text 17: p.66)

Finally, the least frequent resource in the Proclaim category was Concur (0.7%). Considering the high degree of the author's commitment to the conveyed value in Concur (Martin & White, 2005), its low frequency in the corpus was unsurprising and might be because of examined master's students' limited experience and confidence in constructing assertive arguments.

Also, as Swain (2010) argues, this may be due to these student writer's reluctance to take a clear position in discussions, which might originate from their misconception that more citations are favored in discussion sections.

Disclaim

Little variety was observed in the discussions of Disclaim realizations (i.e., Deny and & Counter), both in their lexico-grammatical structure and functions. 

The instances of Deny were mainly restricted to typical negation forms like be not, do not, no…, and lack of (Example 28). As for functions, writers often used denials to report their research results. Similarly, in addition to typical realization via conjunctions like although, however, while, and but, Counter instances were sometimes conveyed explicitly (Example 29), and occasionally together with Deny:

• It is assumed that implicit knowledge is more likely to be drawn on under time pressure, but it is not a sufficient condition [Disclaim: Deny - Propositional]. (Text 2: p.71)
• Contrary to this study [Disclaim: Counter], de la Fuente (2006) argues that in terms of vocabulary retention, planned focus-on-form, which involves the use of focused tasks, is more effective than traditional PPP lessons. (Text 16: p.74)

Our observation showed that the examined Iranian dissertation writers in English preferred to frequently illustrate these types of inconsistency between their work and previous works in literature aligns with previous studies (Cheng & Unsworth, 2016; Sadeghi & Alinasab, 2020) and may suggest that examined Persian novice writers may be sensitive to the need to position their work within the existing literature.

In addition, the higher frequency of Counter (almost twice as frequent) over Deny in the corpus suggests the cautiousness of examined writers in their tendency to rely more on comparison strategy in correcting and disagreeing with an alternative view rather than directly challenging or rejecting it.

5. Discussion and conclusion

Graduate student writers in this study demonstrated conformity with current practices of academic writing, such as those in discussion sections of doctoral theses (Geng & Wharton, 2016) and introduction and conclusion sections of published research articles (Xu & Nesi, 2019) in referring to alternative voices or viewpoints.

We found a preference by thesis writers to restrict dialogic space rather than open it up. A similar pattern of dialogic contraction was noticed in the discussion sections of doctoral theses in applied linguistics (Geng & Wharton, 2016, 2019; Loghmani et al., 2020) and discussion and conclusion sections of research articles in psychology (Loi et al., 2016), and applied linguistics (Cheng & Unsworth, 2016). Although making direct comparisons of these groups of writers due to the different sizes of the corpus and frequencies of Engagement resources is difficult, this similarity might be because the present study's writers modeled more professional genres to express their stance in their academic writings. The other possible reasons may be related to their demonstration of confidence in their arguments or their lack of experience in managing academic conflicts leading to the exclusion or at least restricting the space for contrary views.

The dominance of dialogically contractive resources within professional academic genres like published research articles is expectable, given that a contractive stance helps increase the argumentative power of propositions and enhance 'writer-reader solidarity' "when readers have already been convinced that the author has greater expertise in some area than them […]; but it can reduce writer-reader solidarity when readers are resistant, [or] more knowledgeable than the author …" (Loghmani et al., 2020, p.11). In master's theses, however, the primary readers (i.e., the supervisor, advisor, and referees) have higher expertise and authority. Instead, in such a context, employing more Expansive resources seems an effective rhetorical strategy that can implicitly activate readers' positive response to the valued position (Cheng & Unsworth, 2016) and decrease the risk of rejection by the reader. In the present corpus, however, this strategy did not occur, which might reflect a lack of adequate awareness for examined master's students, hence running the risk of being challenged by more expert audiences. Nevertheless, considering this discordance, it must be mentioned that although the number of contractive resources in the current corpus seemingly overcame the expansive resources, the frequency gap was insignificant, and the examined authors in the corpus adopted a more balanced stance. A balanced distribution of expansive and contractive resources is believed to increase texts' persuasiveness and effectiveness (Swain, 2010; Xie, 2016).

Highest and lowest frequencies

The employment of Acknowledge and Entertain as the first and second most common heteroglossic resources in the first place indicate that the present study's master's students tend to take a neutral position when referencing other voices and stay cautious when presenting their propositions. Although the tendency to take a neutral position among non-native learners and novice writers in various contexts has been commonly reported in the literature (e.g., Geng & Wharton, 2016; Liardét & Black, 2019; Swain, 2010; Xie, 2016; Xu & Nesi, 2019), the role of culture could not be overlooked. The examined Iranian students' high reliance on Acknowledge and Entertain might also be due to cultural and social factors. Obeying rules and traditions without questioning or expressing doubt about them could be traced to Iranian culture, tradition, and philosophical precepts (Abdollahzadeh, 2011); Iranian students are also reported to be dependent on authority figures in the class and obedient and conforming to rules. (Rahimi & Asadollahi, 2012).

Moreover, their inclination to avoid certainty may be a sign of conformity to conventions of academic writing, which restricts the author's freedom in positioning themselves interpersonally, especially in the discussion genre (Hyland, 2005; Swain, 2010). In addition, students' limited experience and knowledge in making critical evaluations might prevent them from freely expressing their contrary and supportive opinions, mainly due to concerns about being proven wrong and uncertainty about the validity of their findings (Salmani Nodushan & Khabbaz, 2011).

Our target graduate student writers tend to use more explicit formulations of Endorsement, while for criticizing (i.e., Distance), they relied more on implicitly evoking negative attitudes. Likewise, they prefer more or less to contrast two opinions and thereby signal to the reader the unexpectedness of an opinion rather than overtly rejecting it. Further, a closer examination of the corpus revealed a tendency among examined graduate writers to combine Pronounce and Entertain in presenting their immediate results. Thus, they opted for overt intervention, showing their presence in the text. However, when interpreting and making generalizations based on their findings, they expanded the dialogic space via Entertain to make a reader-friendly context, show politeness and modesty, and ultimately lower the possibility of rejection.  The low frequency of Concur category might imply master's students' insufficient experience and confidence in constructing assertive arguments. Further, these graduate students prefer positive alignment (Pronounce and Endorse) over opposing alignment (Deny and Distance). This tendency to refrain from negative evaluations and explicit critiquing of alternative voices might be due to their general misconception that teachers, books, and published articles are "fount of all the knowledge" (Littlewood, 2000, p.31) and thus unchallengeable.

A general distinction between information and evaluation was evident among most heteroglossic elements: Acknowledge, Endorse, Reference-to-Data, and even Deny. Overall, they employed less diverse linguistic and rhetorical devices.

Our findings imply that appraisal-based research needs to investigate stance-taking across other dissertation subgenres (e.g., literature reviews, methodology) and in other disciplines to discover the potential range of various engagement resources of the appraisal system. We speculate that the subtle superiority of Expansive resources in the literature review section of Chinese master's theses (Xie, 2016), introduction and conclusion sections of RAs by native English and Chinese writers (Xu & Nesi, 2019), and reporting verb choices in EAL and L1 learners' corpora (Liardet & Black, 2019) might reflect genre-specificity of evaluation resources which makes the frequency of their types and tokens vary across different text-genres with different communicative purposes.