Multifunctionality of Epistemic Stance Markers: Variation across Disciplines and Speaker Roles in Classroom Discourse

This study investigated the communicative functions of ‘I think’ and ‘I guess’ in classroom discourse and intra-register variation in their use. Building on research into the multifunctionality of epistemic stance markers, a coding scheme was developed and iteratively applied to a corpus, revealing five macro functions: epistemic stance, softening, personal perspective, discourse organizing, and deciding in real time. The frequency of each macro function differed between the two markers, as did the observed subfunctions. Statistical analyses comparing instructors and students in three disciplines indicated intra-register variation, linked to situational differences. ‘I guess’ was most frequent in student language, and ‘I think’ was most frequent in the humanities. Different patterns were also noted in the use of certain communicative functions. The results suggest the importance of accounting for register variation when describing epistemic stance markers.


Introduction
Epistemic stance (ES) markers, which display a speaker's level of commitment to the truth-value of a proposition (e.g., doubt or certainty), are an important

Background
The following sections discuss previously reported communicative functions of 'I think' and 'I guess' in a variety of registers, the importance of epistemic stance in classroom discourse, and existing evidence of intra-register linguistic variation across academic disciplines and speaker roles.

2.1
The Communicative Functions of 'I Think' As a stance verb, 'think' has been the focus of several studies. Some studies have focused on its epistemic function (e.g., Andersen and Fretheim, 2000;Biber, 2006a and2006b), others have focused on the function of "saving face" (e.g., Ginsburg et al., 2016), and some have reported on its multifunctionality (e.g., Sanchez and Vogel, 2015;Kaltenböck, 2010). Focusing on epistemic functions, Biber (2006a and2006b) described 'think that' as a frequently occurring example of stance verb + 'that'-complementiser in his investigation of university language, describing it as a marker of likelihood. He found variation across the university registers (e.g., classroom discourse, service encounters, and advising sessions) but did not explore potential differences between professors and students or variation across disciplines. Aijmer (1997) argued that 'I think' communicates epistemic meanings of both certainty and uncertainty, which correlate with prosodic differences. She also reported the functions of expressing intention or cogitation (i.e., the literal act of thinking) and softening bluntness.
Other scholars have also demonstrated that the communicative functions of 'I think' comprise more than epistemic distinctions. Kärkkäinen (2003) attributed epistemic, face-saving, personal perspective, and discourse organizing functions to the phrase. Kaltenböck (2010) also observed multiple functions, describing parenthetical 'I think' as flexibly accomplishing four main functions: 1.
shield function: shields the speaker by expressing lack of commitment to a proposition; 2. approximator function: displays inexactitude within the proposition; 3. structural function: acts as a filler or turn-taking signal; 4. and booster function: strengthens the speaker's position. Zhang (2014) similarly observed a tentative (shield) function, an emphatic (booster) function, and a discursive (structural) function in an Australian customs office, as well as mitigating (face-saving) and evaluative functions. She also observed multifunctionality in individual occurrences but suggested that a dominant function can be ascertained based on context and collocation.
Since existing studies have come to different, though overlapping, conclusions about the functions of 'I think,' the present study sets out to thoroughly describe its use in classroom discourse by integrating previously identified functions into one empirical analytical framework. It is important to note that existing studies on 'I think' were conducted in various types of discourse, such as written evaluations in the health sciences (Ginsburg et al., 2016), informal written conversations on the web (Sanchez and Vogel, 2015), and general British English (Kaltenböck, 2010), without consideration for potential register-based functional differences. Therefore, this study also aims to evaluate the appropriateness of the previously reported functional categories in the classroom discourse register.

2.2
The Communicative Functions of 'I Guess' Contrary to 'I think,' 'I guess' has been the subject of far fewer studies, although it is reportedly one of the most frequently used ES markers (see Section 1), suggesting that an empirical description of its functional uses would be helpful in advancing knowledge of this category of stance markers. The studies that do exist have attributed various discoursal functions to the phrase. Biber (2006a) classified 'guess that' with 'think that' as a common likelihood verb + 'that'-complement construction. That study did not discuss functions other than likelihood or grammatical constructions other than verb + 'that'-complement. Kärkkäinen (2007 and, on the other hand, looked at 'I guess' in detail through conversation analysis. She proposed that 'I guess' functions as an evidential marker, referring especially to a speaker's stance based on evidence that has surfaced during a speech event. Her research also suggested that 'I guess' functions as a discourse marker and as a marker of politeness and hedging for saving face. Other functions attributed to 'I guess' include expressing softening, caution, and humility by students in classroom discourse (Lin, 2015) and signaling that a statement is an opinion (Weisser, 2020).
As can be seen, similar to 'I think,' reported functions of 'I guess' also include epistemic, softening or face-saving, discourse organizing, and personal perspective, in addition to marking an evidence-based inference. Considering the similar categories that emerged in existing research, the frequent occurrence of both markers in academic discourse, the categorization of both as ES markers, and their shared grammatical features, the present study undertook a comparative analysis of the two markers in a corpus of classroom discourse. More details about developing and applying this framework are provided in Section 5 below. The comparison of these two epistemic markers is meant to reveal the degree to which they conform to a general ES marker framework or diverge into individual functional profiles.

2.3
Epistemic Stance in University Classroom Discourse ES is a fundamental aspect of spoken communication (Gablasova et al., 2017). Through ES, speakers position themselves regarding the truth-value of the propositional content of an utterance. In addition, speakers use ES markers to express their level of confidence in an utterance, their identity, and their relationship with interlocutors through hedging and politeness (Boncea, 2013;Gablasova et al., 2017;Kirkham, 2011). Classroom discourse is an example of a spoken register in which ES is particularly important. Speakers frequently position themselves toward course content, for example presenting facts, opinions, evaluations, and uncertainty about new and/or complex ideas. At the same time, disagreements, confusion, and errors are addressed while identities such as expert, novice, and peer are negotiated (Friginal et al., 2017;Kirkham, 2011). Since ES markers perform functions related to all of these situational needs, their functional use should be included in linguistic descriptions of classroom discourse.
The target ES markers in this study, 'I think' and 'I guess,' are two of the most frequently occurring epistemic markers in English overall (Kaltenböck, 2010) and specifically in university registers (Biber, 2006b;Kärkkäinen, 2007;Lin, 2015). However, few descriptive studies on classroom discourse include these two markers. While Biber (2006a) included them among other ES markers in his study on university language (see above), he did not focus specifically on classroom discourse, interdisciplinary variation, or functions beyond epistemic stance. Poos and Simpson (2002) reported variation in the frequency of 'I think' across disciplines in classroom discourse (see below), but they did not address functional use. Similarly, Csomay (2007) reported higher frequency of 'I think' and 'I guess' in the humanities as compared to other disciplines, but, like Poos and Simpson, she did not discuss their functions. The current study seeks to build on the few existing reports of ES markers in classroom discourse by describing the functional use of two frequent ES markers in that register. Biber and Conrad (2019: 6) have defined register as "a [language] variety associated with a particular situation of use," for example university classroom discourse. In this framework, pervasive linguistic features occur in specific registers because of a functional relationship between such features and the situational context associated with the register. Based on this concept, sub-registers, such as specific disciplines within classroom discourse, may also vary linguistically. This is because while such sub-registers share many situational features as part of a larger register, they also differ from each other in systematic ways.

Intra-register Variation: Functional Differences across Disciplines and Speaker Roles
Research suggests that the multifunctionality of individual pragmatic markers, including ES markers, often varies based on context (e.g., Aijmer and Simon-Vandenbergen, 2011;Fischer, 2014;Garcia, 2004;He, 1998;Mauranen, 2004;Poos and Simpson, 2002). This suggests that ES marker use may vary across registers and sub-registers because of situational differences. Therefore, our understanding of ES markers may be incomplete without considering both register and intra-register variation. Within the classroom discourse register, intra-register variation in the use of 'I think' and 'I guess' is likely to occur across disciplines and speaker roles, since these are two important aspects in which classroom contexts differ.
Several studies have revealed linguistic variation related to communicative functions and frequency of linguistic features across disciplines. Kashiha and Heng (2014) found that the most common lexical bundles in university lectures were used for different discourse functions in chemistry lectures as compared to lectures in politics. Investigating the use of 'sort of' and 'kind of' in classroom discourse, Poos and Simpson (2002) reported a clear trend from the lowest frequency in the physical sciences to the highest in the humanities, with social sciences falling in the middle. They attribute this trend, at least in part, to the more subjective and abstract nature of humanities subjects and the more defined and precise technical language in the sciences. In addition, a keyword comparison in that study revealed that 'I think' and 'think that' both appeared in the top five most frequent two-word phrases in the social sciences, while those phrases did not even appear in the top ten most frequent two-word phrases in the humanities or the physical sciences. The current study similarly investigates frequency and functional differences in the use of 'I think,' as well as 'I guess,' in a corpus of classroom discourse. The results will also illustrate the extent to which the two markers are used similarly across disciplines or not.
In addition, the different needs, aims, and identities of different academic roles (i.e., students and instructors) may result in different language use. Biber (2006a) found that in university language the two verbs, 'think' and 'guess,' were most frequently seen in student registers, such as study groups. This suggests that in classroom discourse, they might also be used more frequently by students than faculty. Gablasova and Brezina (2015) found variation in candidates' and examiners' use of ES markers on different tasks during an exam of spoken English, and they attributed that variation in large part to speaker roles. Moreover, researchers have recommended that future studies investigate the use of stance markers from the perspective of speakers' roles and relationships (e.g., Kaltenböck, 2010;Poos and Simpson, 2002). This study answers that call by including speaker role as a variable.
In a study that took both speaker roles and disciplines into account, Csomay (2007) found large-scale differences in the ways that instructors and students in North American classrooms generally used language, as well as role differences moderated by discipline. Students' turns in all disciplines were shorter and more frequent than those of their instructors. In business, education, and engineering classes, instructors produced more language associated with contextual linguistic features, such as first and second person pronouns, activity verbs, and conditionals, while in the natural and social sciences, students used these features more frequently. Humanities instructors' language was characterised by conceptual, informational features, such as nouns and attributive adjectives, and both instructors and students in humanities classes produced Downloaded  significantly more language for personal framing, such as verb-controlled complement clauses, including 'I think' and 'I guess.' Disciplinary variation was attributed to differences in instructional focus and disciplinary style. The studies above describe linguistic variation related to disciplinary and role-based differences within academic discourse. The present study extends that research to empirically investigate such variation in the communicative functions of 'I think' and 'I guess' within classroom discourse.

Research Questions
This study aims to: (1) describe the communicative functions of 'I think' and 'I guess' in classroom discourse by using a framework based on existing studies in other registers and (2) investigate intra-register differences in the use of these ES markers across disciplines and speaker roles. In addition to extending descriptive linguistic knowledge of the classroom discourse register, the results also shed light on the extent to which functional use of 'I think' in this register mirrors previous studies in other registers and the extent to which 'I guess' functions similarly to 'I think.' Two research questions were addressed: 1. What communicative functions do 'I think' and 'I guess' perform in a corpus of university classroom discourse? 2. To what extent does use of 'I think' and 'I guess' vary in university classroom discourse a. across disciplines (humanities, business, and natural science)? b. across speaker roles (i.e., instructors and students)? c. as a function of the interaction of discipline and speaker role? The next section (Section 4) describes the corpus used in this study. This is followed by a discussion of the communicative functions of 'I think' and 'I guess' attested in the corpus with qualitative analyses of corpus excerpts (Section 5). Finally, an intra-register situational analysis and statistical analyses are presented and interpreted (Section 6). Limitations and suggestions for future research are also addressed (Section 7).

Corpus Description
This study analyzed a subcorpus of the TOEFL 2000 Spoken and Written Academic Language Corpus (T2K-SWAL) (Biber et al., 2002) with the text (i.e., transcript) as the unit of observation. T2K-SWAL includes 2.7 million words of spoken and written language from four U.S. American universities. The subcorpus is comprised of transcripts of class sessions from business, the humanities, and natural sciences categorised by high and medium interactivity (10 or more turns per 1000 words). Instructor-led labs in the natural sciences were also included because they represent a similar situational context to class sessions, they are a major part of natural sciences curricula, and their inclusion brings the number of natural science texts closer to that of the other two disciplines. The subcorpus consists of 55 texts and approximately 370,000 words. It was evaluated for representativeness based on domain considerations and linguistic distribution (Egbert et al., 2022). The corpus metadata and a subsample of transcripts revealed a close match with the target domain: university students and faculty engaged in a variety of classroom activities with data stratified across the three target disciplines and four types of universities in different geographic regions. Relative standard error (RSE) for the mean rate of occurrence of each linguistic feature was calculated to estimate the precision of the corpus for the research questions, meaning the likelihood that the corpus would reliably reflect the distribution of the features in the actual domain. For 'I think,' which occurred 549 times, RSE was 0.10, which suggests an error rate of +/− 20%. This is not highly precise, but considering the qualitative representativeness of the corpus, this was considered acceptable for an exploration into this feature in this register. However, findings should be interpreted cautiously with this in mind. For 'I guess,' which occurred only 102 times, RSE was 0.17. For this reason, quantitative results related to this marker should be considered provisional. However, considering the paucity of descriptive research on this marker and the value of its comparison with 'I think,' its inclusion in the study was considered beneficial. Table 1 shows the composition of the corpus. For quantitative analysis, each transcript was separated into two files, one with instructor-produced and one with student-produced language, generating a total of 110 transcripts. Qualitative coding was completed using the original, fully intact files. As can be seen, the number of texts is nearly equal between humanities and science, while humanities texts account for nearly 40% more words. The business section comprises more texts and more language than the other two sections, but the ratio of words to number of texts is higher in the humanities. This indicates that humanities classes generated more language than the other disciplines. Also, even though only interactive texts were included, instructor language constitutes roughly 70% of the word count. Business and humanities texts show this trend of instructor language dominance, with approximately 84% and 65% of words in the respective sections uttered by instructors. However, in the natural sciences, the word count is closer to 50/50 due to the inclusion of Contrastive PragmaticS 4 (2023) 213-242 lab sessions during which instructors rarely spoke. These differences resulted in unequal corpus sections but may also be indicative of situational disciplinary differences, which are further discussed in Section 6.1.

Communicative Functions
The following subsections address the first research question regarding the communicative functions of 'I think' and 'I guess' in classroom discourse. An explanation of the coding methodology is followed by descriptions of the functions observed in the corpus and comparisons between the two markers. The results confirm the presence of the four macro-functional categories reported in the literature along with additional findings.

5.1
Coding Communicative Functions A coding scheme was developed by synthesizing the previously reported functions in Section 2 into four macro-functional categories: epistemic stance, softening, personal perspective, and discourse organizing, with a fifth open category for emergent functions. The functions were operationalised with definitions and examples of subfunctions, as described in previous literature (see Appendix). For example, epistemic stance was defined as relating to the truth-value of the proposition, and examples of subfunctions of epistemic stance included likelihood, certainty, uncertainty, and inexactitude.
Using a custom Python program, concordances containing 'I think' and 'I guess' were extracted from the corpus, written to a spreadsheet with 25 words to the left and right, and coded for file name, discipline, and speaker role. 'I [adverb] think' (e.g., 'I really think') and 'I would think/guess' were also included, following previous researchers (Gablasova et al., 2017), Table 1 Number of texts and words by discipline and role as well as 'I think so' and 'I guess so,' since they seemed to accomplish the same functions. The coding scheme was piloted on 10% of the concordances, and through this process, prototypical examples were collected and added to the rubric. In addition, substitution tests were developed for each functional category based on operational definitions and example occurrences. If substituting the provided word or phrase for the target feature in an utterance did not change the meaning, this would provide additional evidence for determining its function. An element of subjectivity is always present in this type of coding, but the addition of examples and substitutions was intended to increase reliability.
A second coder coded 5% of the data to check the reliability of the coding scheme. Cohen's kappa showed a substantial level of initial agreement at .769 (Phakiti, 2014), after which the two coders discussed disagreements until 100% agreement was reached. In some cases, utterances were too ambiguous to code confidently, and they were excluded from the analysis, resulting in 549 occurrences of 'I think' and 102 of 'I guess.'1 Coders relied on situational context and linguistic co-text for evidence to support inferences about speaker intention. Occurrences that seemed to accomplish multiple functions were coded for all appropriate functions but, following Zhang (2014), eventually assigned a primary function based on contextual evidence and only counted in that category. This means that the multifunctionality of individual occurrences was lost in the analysis, as this study chose a middle level of granularity.

Communicative Functions of 'I Think' and 'I Guess'
In response to the first research question, the following communicative functions were observed in the corpus: epistemic stance, softening, personal perspective, and discourse organizing. In addition, an analysis of the 'other' category revealed a fifth macro-function: deciding in real time. Personal perspective was only carried out by 'I think'; both markers served all other functions. However, different subfunctions were observed between the markers within macro-functional categories, and proportional distribution of macro-functions also differed. These differences are discussed and supported by textual examples below.

5.2.1
Epistemic Stance An utterance fulfilling an ES function, as previously defined, expressed the speaker's commitment to a proposition's truth or accuracy. This was the most Contrastive PragmaticS 4 (2023) 213-242 frequent function of both markers accounting for 53% of occurrences of 'I think' and 42% of 'I guess. ' ES subfunctions ranged from highly uncertain to highly certain based on evidence or logic. For this reason, a range of substitutions were included such as 'maybe,' 'probably,' 'I expect,' and similar (see Appendix). Frequent subfunctions for 'I think' in this category included likelihood, expectation, uncertainty, and inexactitude. These reflect the epistemic functions reported in previous studies, in particular the flexibility of expressing both tentativeness and confidence (e.g., Aijmer, 1997;Kaltenböck, 2010;Zhang, 2014). On the other hand, 'I guess' served a more limited range of subfunctions, more frequently signifying uncertainty and speculation, though previous studies did not report this.
Linguistic co-text helps to illuminate variations in the level of certainty suggested by each marker. For example, in nearly 4% of concordances containing 'I think,' 'probably' co-occurred, while it only co-occurred once with 'I guess' (less than 1%). In addition, 'I think' co-occurred with contextual evidence of the speaker's belief in the likelihood of a proposition. In (1) the class is discussing an essay that suggests high school graduates should engage in 'maturing activities' before attending university. They are debating who should be responsible for deciding what activities qualify as 'maturing.' The instructor rejects two potential responsible groups: students, using 'I don't think,' and parents, stating 'well possibly but.' This is followed with 'I think he has the schools in mind,' in which the instructor explains what they believe the author intended. When the instructor asks a question, the student replies with a similar level of confidence, using a combination of 'I think' and 'probably.' (1) Instructor: Yeah, but at any rate, I don't think he thinks the students should decide what a maturing activity is. Parents? Well possibly, but I think he has the schools in mind. Let's go to this one for a minute, uh, that students might forget what they learned in high school during this two-year period. How might he respond to this? Student: I think he would probably say that if they're gonna forget it after two years, then they'd probably forget it after the first month in college anyway.
On the contrary, 'I guess' more frequently co-occurred with 'I don't know' and 'you know' as a hedge or filler, as well as other linguistic and paralinguistic evidence of a lack of commitment to a proposition. In (2), during an English class discussion, the instructor is encouraging students to interpret an author's Contrastive PragmaticS 4 (2023) 213-242 intended meaning. Student 2's use of rising intonation, denoted with a question mark, contributes to the sense of uncertainty, in addition to a general lack of student responses (the instructor rephrases the question several times) and another student answering 'I don't know.' (2) Instructor: What's that scene about? Where he's just going mad buying, buying, buying? What is that about? Anyone say something? Student 1: No, I just don't know. Student 2: Compulsive spending? Our society, I guess?
In (3) and (4), students in two different science labs express uncertainty, one with 'I think' and the other with 'I guess.' In (3), Student 1 is trying to understand how an experiment works, asking Student 2 for confirmation. Student 2 provides an affirmative answer, but then expresses uncertainty with 'I think so,' co-occurring with 'you know' and 'I don't know,' and followed by an admission that they need to ask the instructor.
(3) Student 1: It breaks the cell wall? Student 2: Uh huh. I think so. I have to ask her, you know, I don't know, there's no guidebook.
Similarly, in (4), two students are struggling to understand a lab experiment. Student 2 first expresses uncertainty about the shape of the material they are working with, posing a question that suggests they do not see the same shape as their lab partner ('You can see it's round?'). The student's next utterance contains 'I guess not' with a long pause in the middle (represented by …). This is followed by yet another question.
(4) Student 1: Oh, this looks round to me. Student 2: You can see it's round? Student 1: Well, it doesn't look flat. Student 2: No, yeah, I guess … not. Does it look triangular? Student 1: Uh, it's kinda hard to tell. Student 2: Yeah.
As (3) and (4) illustrate, both markers denoted uncertainty, but the most cautious expressions of epistemic stance were typically rendered by 'I guess.' Moreover, the strongest indications of likelihood were expressed with 'I think.'

Softening
In utterances performing a softening function, speakers appeared certain about the truth of the proposition, but they used 'I think' or 'I guess' to mitigate the strength of a claim or a potentially face-threatening act, displaying politeness and distancing. This was the second most frequent function of 'I guess,' at 27% of occurrences, but infrequent for 'I think' at only 8%.
Like epistemic stance, this category also included a range of subfunctions. This range of subfunctions resulted again in the need for several substitution options (see Appendix). In many cases, direct substitutions were difficult to find in this category, but removing the marker to note the qualitative change in directness or politeness was also used as a test for this function. Both markers were used to soften the strength of a claim, mitigate a potential face-threat (e.g., making a suggestion; asking questions in class), and hedge uncertain language choices and concepts that were difficult to explain.
An example of mitigating a face-threat while showing humility towards an instructor can be seen in (5). In the accounting class where the exchange occurred, the instructor is leading a review of course concepts. A student asks a clarifying question, followed by an explanation of the source of their confusion.
(5) Student 1: Buying stock would be financing? Instructor: No. Buying stocks, again, is part of the investment.
[additional explanation removed] If it is part of the liability, that means it is part of the financing activities, just, just, a rough, a kind of, of um, of criteria. Several students continue to explain their confusion and seek clarification. It seems the student is certain about the source of confusion, especially considering their last utterance ('That's confusing'), but the face-threat of pointing this out to the instructor is mitigated with 'I think.' Contrastive PragmaticS 4 (2023) 213-242 'I guess' also appeared to suggest a special form of softening: a reluctant claim or agreement, which seemed to say, 'this is not ideal' or 'this wouldn't be my first choice.' For instance, a student in (6) softens the strength of a claim that describes a less-than-ideal situation in a book the class is discussing.
(6) Student: […] even though he's sort of at the end frozen and not happy to be there at least he did have that experience. I guess it's better to have loved and lost than uh […] In (7), Student 2 suggests reluctance about giving her talk, prompting surprise and laughter. Student 1 has just finished a presentation, and the instructor calls the next student. The instructor uses 'I think' in the mitigating manner of 'if you don't mind' in the first turn and, interestingly, in the last turn, combines a boosting 'for sure' with a mitigating 'I think.' As Student 2 immediately begins her presentation, it seems clear that she was prepared to speak, just not particularly willing. Reluctance occurred somewhat frequently with 'I guess,' also as a secondary function with epistemic stance, but no occurrences of 'I think' performed this subfunction. It appears that 'I guess' may not only be more tentative than 'I think,' but also provides distancing in relation to unpleasant or unwanted situations, which 'I think' does not.

5.2.3
Perspective Marking Perspective marking utterances introduced the speaker's personal viewpoint, opinion, or interpretation. This was the second most frequent function of 'I think,' at 31%, supporting the findings of Kärkkäinen (2003) and Zhang (2014). Combined ES and perspective marking accounted for 85% of instances of 'I think.' However, in contrast to Weisser's (2020) findings, no examples of this function were identified among occurrences of 'I guess.' This is a significant distinction between the two markers, and it is unclear whether this function did not emerge due to the small number of occurrences of 'I guess' in the corpus, as a reflection of use patterns in the register, or because this function is truly infrequent.
Perspective marking often occurred with evaluative language, such as evaluative adjectives, as in (8) ('weird'), in which 'in my opinion' is a suitable substitution. The student expressed this sentiment to their lab partners during a science experiment. This utterance exemplifies the straightforward nature of most of these occurrences.

(8) Student:
This whole banana thing is just kind of weird, I think. We're taking DNA from a banana.
In (9), an instructor provides insights on the style of a poem the class is discussing. In this case, an evaluative adjective occurs in an earlier utterance ('simple'), and 'I think' marks the subjective nature of the entire comment. This usage was frequent in humanities classes where subjective course topics were frequently discussed, such as poetry and literature.
(9) Instructor: It sounds like a very simple kind of poem. And maybe it is, but I'm just wondering, you know, th-the style, I think, adds something to it. If she had written it differently, it might've had a different effect.

5.2.4
Discourse Organizing Utterances that seemed to perform a structural discourse function, such as filling pauses and marking false starts, were coded as discourse organizing. Although several previous studies described this as a function of 'I think' (Kaltenböck, 2010;Kärkkäinen, 2003;Zhang, 2014) and 'I guess' (Kärkkäinen, 2007 and, it was among the least frequent functions for both markers, 7% of 'I guess'; 6% of 'I think.' These instances were often accompanied by other fillers and repetition. For example, in (10), the student uses 'I think' and repetition to hold the floor while formulating a response.
(10) Instructor: […] Ok. You were gonna say something, George? Student: Oh. Yeah, I was thinking of, I think, I-I call this the big lie of, uh, the American dream. […] Contrastive PragmaticS 4 (2023) 213-242 5.2.5 Deciding in Real Time An additional function emerged from the analysis, which was not reported in any of the previous studies. Labeled 'deciding in real time,' this function occurred when speakers verbalised coming to a decision or realization 'on the spot.' This accounted for 23% of instances of 'I guess,' the third most frequent function, but was the least frequent function of 'I think,' at only 2%. In (11), a student is setting up recording equipment in a science lab, deciding where they will place the camera.
(11) Student: It's awkward but it's a really nice one which is why we use it. Um, I think what I'm going to do is stick it up there and then our microphone […] Similarly, in (12), Student 2 seems to decide in the moment to use an Australian accent, before realizing that they lack the ability to do so. This is a conversation among students during a science lab and represents the informality and humor characteristic of lab sessions.
(12) Student 1: You going to switch accents on me? Student 2: I guess I'll speak Australian. Uh, I don't know the accent. Student 3: I'll speak Spanish.

Summary
The functions attested in the corpus confirmed and extended previous research into the multifunctionality of both ES markers. Furthermore, although the two markers share similar functions, they differ in their distribution. As Figure 1 shows, ES was the most frequent function for both. Softening and deciding in real time were both frequent for 'I guess' but quite infrequent for 'I think.' Conversely, personal perspective was a frequent function for 'I think' and non-existent for 'I guess.' Discourse organizing was relatively infrequent for both.
The subfunctions of each marker suggest additional similarities and differences. Within the epistemic function, 'I guess' suggested less certainty than 'I think.' Both markers performed softening functions, but only 'I guess' displayed reluctance. The subfunctions of discourse organizing and deciding in real time were similar across the two markers. 'I guess' was not observed expressing personal perspectives at all. These results demonstrate that the framework synthesising previous studies identified most functions of 'I think' and 'I guess' in classroom discourse but not all. An additional function emerged Contrastive PragmaticS 4 (2023) 213-242 for both markers (deciding in real time). Moreover, the macro-functional framework captures generalities but misses the nuanced differences between the two markers.

Variation across Disciplines and Speaker Roles
The second research question concerned variation in the use of 'I think' and 'I guess' based on disciplines, speaker roles, and the potential interaction between these variables. A situational analysis was conducted to describe the different disciplines and speaker roles. Then, statistical analyses explored variation in use, and that variation was interpreted through the situational analysis.

Situational Analysis
The classroom discourse register is characterised by multiple participants communicating orally in real time in a shared classroom space. While instructors are positioned as experts and experienced academic communicators, students are typically novices. In this corpus, all speech events were at least moderately interactive, but the level of individual participants' interactivity varied. The communicative purposes of instructors and students generally differed, and their relationships may have been influenced by power distance. As expected, instructors' apparent purposes were to explain material and check comprehension. To this end, they lectured, elicited responses, evaluated student comments, made corrections and suggestions, and in some cases attempted to stimulate and guide discussion. Students' main communicative purposes appeared to be showing their understanding, co-constructing knowledge, and overcoming confusion about course content. They answered instructor questions, asked questions, and sometimes engaged in discussions or gave presentations. Unsurprisingly, topic was a key difference across the disciplines, and such topic variation may have been a source of variation in the use of ES markers. Business was comprised of accounting, business administration, business communication, management, and marketing. Some of these classes focused on mathematics, while others were more concerned with topics like decision-making and communication. Humanities sub-disciplines included English, classical studies, history, and philosophy, with topics such as poetry and literary analysis, historical events, ancient societies, and cultural criticism. Science encompassed biology, botany, chemistry, geology, and physics, with a focus on natural laws, characteristics of plants and animals, mathematics, and experimental procedures and outcomes. Differences in the level of abstraction and the amount of specific, technical, discipline-specific language typically used to discuss these topics could engender variation in the use of pragmatic markers, including 'I think' and 'I guess. ' The nature of the topics also resulted in different types of classroom activities. Interactive lectures were frequent in business and science classes in which instructors explained and demonstrated concepts and processes, checking students' understanding and answering questions. In addition to lectures, humanities classes consisted of frequent class discussions between instructors and students, small group discussions among students, and student presentations. Some business classes, such as management, also included class discussions, sometimes based on case studies. Different communicative functions are emphasized in these different activities, though the functions can overlap to some extent. Discussions typically required the sharing of interpretations and opinions, while lectures were focused more on factual information. Finally, in science labs students actively engaged in experiments, discussing the process and the results with their peers, and only occasionally received guidance from instructors. Since different communicative functions can result in different language choices, types of classroom activities may influence the use of 'I think' and 'I guess. ' As seen in Section 4, instructors typically produced more words than students, except in science because of the labs, and humanities classes generated a higher ratio of words to transcript than the other disciplines. Humanities Contrastive PragmaticS 4 (2023) 213-242 instructors produced the most language in the corpus. It may be that the interpretive and abstract nature of humanities topics and lack of standardised, technical vocabulary resulted in more language compared to scientific and quantitative business topics, as others have found (Poos and Simpson, 2002). Business students produced the smallest number of words, probably due to the high number of lectures in which student turns were generally brief clarification questions or short responses to display questions in which they demonstrated knowledge. The amount and type of language produced by each group of speakers may also help to explain differences in the use of the target features in this study.
This situational analysis provided a framework for interpreting the linguistic differences revealed through statistical analyses, following Biber and Conrad's (2009) register analysis procedure. The statistical analyses are described in the next section, followed by quantitative results and the situational interpretation.

6.2
Statistical Analyses After all occurrences were coded for communicative function, normed rates per 1,000 words were calculated for each marker and each function of each marker (epistemic, softening, personal perspective, discourse organizing, and deciding in real time). Statistical analyses were conducted to answer the second research question regarding variation across disciplines, speaker roles, and their interaction. Descriptive statistics were calculated to investigate overall frequency patterns, followed by a series of 2 × 3 factorial ANOVAs on 'I think' and Kruskal Wallis tests on 'I guess.' All analyses were carried out in R (R Core Team, 2014), and the results were interpreted through the situational analysis described in Section 6.1.

6.2.1
Comparison of 'I Think' and 'I Guess' The descriptive statistics revealed one outlier transcript in which 'I guess' and 'I think' occurred significantly more frequently than in other transcripts in the discipline. It was removed prior to statistical analyses. Overall, 'I think' occurred much more frequently than 'I guess'; 501 occurrences were analyzed compared to 81. Figure 2 illustrates the usage frequency of each marker by instructors and students without considering disciplinary variation. Students (M = 1.48; SD = 1.47) used 'I think' at a similar rate to instructors (M = 1.44; SD = 1.74), but they used 'I guess' more than three times as often (students M = 0.36; SD = 0.57; instructors M = 0.11; SD = 0.18). The fact that the standard deviations in most cases were greater than the means suggests substantial variation among speakers.
Contrastive PragmaticS 4 (2023) 213-242 Descriptive statistics for the overall use of 'I think' and 'I guess' by discipline and by role within discipline can be seen in Tables 2, 3, and 4. 'I think' was consistently most frequent in the humanities and least frequent in business, regardless of speaker role. 'I guess' was consistently infrequent, but students used it more frequently than instructors in every discipline. There was variation in the use of both markers across transcripts, indicated by the range, and individual variation, indicated by the high standard deviations in relation to the means. This means that although general trends were observed, individual speech patterns and specific communicative events also contributed to variations in use.

Figure 2 Use of 'I think' and 'I guess' by role
Note: This figure presents normed frequency (per 1,000 words). Table 2 Descriptive statistics for 'I think' and 'I guess' by discipline Note: Frequencies were normalized to 1,000 words. Two outlier files were removed from business (instructor and student files from the same transcript).
Contrastive PragmaticS 4 (2023) 213-242 Table 3 Descriptive statistics for 'I think' and 'I guess' for instructors by discipline  Table 4 Descriptive statistics for 'I think' and 'I guess' for students by discipline 6.2.2 Variation of 'I Think' A factorial ANOVA was conducted for overall frequency and the frequency of each function of 'I think.' Main effects analyses revealed significant differences among the disciplines regarding overall use (F(2, 102) = 4.62, p = .01), softening (F(2, 102) = 7.77, p = .001), and personal perspective (F(2, 102) = 13.23, p < .001). The ANOVA results can be viewed in Tables 5 and 6. Role did not significantly impact use, and there were no statistically significant interactions between disciplines and roles. No analysis was conducted for deciding in real time, since there were so few occurrences. A power calculation using the pwr2 package2 in R (R Core Team, 2014) indicated that the minimum sample size (27) was not reached in the discipline groups. Because of this, non-parametric Kruskal Wallis tests were also run, and the results confirmed the ANOVA. Only the ANOVA results are included here for brevity.
Contrastive PragmaticS 4 (2023) 213-242 Post hoc pairwise comparisons indicated that overall use was significantly higher for the humanities compared to business. For the softening and personal perspective functions, use was significantly higher for the humanities compared to both business and science. Moreover, comparing the two roles across disciplines, the humanities instructors engaged in softening significantly more frequently than every other group except students in their discipline. The humanities students employed personal perspective significantly more frequently than every group except their instructors. In each case, the humanities speakers used 'I think' more frequently than others.

6.2.3
Variation of 'I Guess' Since the frequency of 'I guess' was quite low and occurrences were not normally distributed, only non-parametric Kruskal Wallis tests were conducted on these data. Overall use of 'I guess' differed significantly across roles (H(1) = 3.60, p = .05). This is not surprising considering that descriptive statistics indicated a consistently higher mean for students in every discipline. Additionally, the Contrastive PragmaticS 4 (2023) 213-242 softening function was significantly different across disciplines (H(2) = 8.78, p = .01). Similar to 'I think,' humanities had the highest mean (0.11, SD = 0.20), in this case contrasting strongly with science (M = 0.01, SD = 0.05); business occupied a middle position (M = 0.07, SD = 0.22). It is possible that additional significant effects could appear with a larger sample size.

6.3
Interpretation and Discussion In overall use, 'I think' was much more frequent than 'I guess,' despite existing studies suggesting both are frequent in academic discourse (Biber, 2006b;Kärkkäinen, 2007;Lin, 2015). General usage patterns emerged for both markers, but the analyses also indicated a fair number of individual differences among speakers. This may reflect the impact of ideolects on communicative functions (Kirkham, 2011), as well as contextual factors related to specific situations (e.g., impact of specific classes, interlocutors, etc.). As with all linguistic analyses, individual differences should be acknowledged while larger observed patterns are explored.
The observed patterns of both markers in this corpus can be explained through the above-described situational analysis. Roles seemed to predict use of 'I guess' but not 'I think,' although both had previously been most frequently observed among students in academic registers (Biber, 2006a). The situational analysis suggests a possible reason for students' greater use of 'I guess.' The uncertainty associated with this marker's epistemic function would be useful to students, who have less knowledge and experience than instructors. An example is found in the last line of example (2) in Section 5.2.1. The instructor elicits student interpretations of the meaning of a passage, to which a student replies: 'Compulsive spending? Our society, I guess?' The student's communicative purpose is to attempt to answer a question with newly constructed knowledge, a frequent student purpose that engenders tentativeness. Conversely, the communicative purposes of instructors in the corpus were generally to impart knowledge and help students understand course content, requiring a more confident delivery. In addition, student contributions are often evaluated by instructors and classmates, which could encourage a framing of uncertainty to avoid a full commitment to a claim, echoing Kaltenböck's (2010) shield function.
The softening and personal perspective functions were more frequent in the humanities than the other disciplines. In addition, the observed subfunctions of softening, such as weakening the strength of a claim, mitigating a potential face-threat, and hedging uncertain language choices and difficult concepts, expand on previous studies, which mentioned saving face as a function of both markers (Kärkkäinen, 2003(Kärkkäinen, , 2007(Kärkkäinen, and 2010Zhang, 2014) and, in the case of 'I guess,' demonstrating humility (Lin, 2015). The frequency of these two functions in this discipline supports similar findings related to other linguistic features (e.g., Kashiha and Heng, 2014), high levels of hedging (Poos and Simpson, 2002), and an abundance of personal framing (Csomay, 2007) in the humanities. The frequency of softening and perspective marking in the humanities may be related to the subjective content of humanities courses, such as the literary interpretation in the example above. On the contrary, the natural sciences, such as chemistry, botany, and biology, focused on observable phenomena, and business classes, such as accounting, focused on facts and mathematics. For instance, the following exchange (13)  In these types of exchanges, common in science and business, softening and personal perspective marking are unnecessary and in fact inappropriate. Disciplinary tendencies toward certain class activities may have also influenced the functional variation observed in this study, since engaging in different types of tasks can result in different language use (Gablasova, et al., 2017). Discussions were frequent in humanities classes, naturally encouraging frequent use of perspective marking because of the many personal views being shared. Discussions also encourage softening while offering an opinion, disagreeing, or criticizing someone's views. On the other hand, more than half of the science classes took place in the lab, where students carried out experiments. Epistemic stance was the primary function in this context, especially expressing uncertainty about how to perform the experiments, the purpose of the experiments, the potential outcomes, and the meaning of the outcomes. Example (3) in Section 5.2.1, in which two students attempted to understand a biological process in an experiment, captures this pattern well. Non-lab science classes and most business classes consisted of lectures, during which few subjective discussions emerged.
It is not surprising that epistemic stance was the most frequent function for both markers with no significant variation across disciplines or roles. Instructors and students across disciplines are engaged in the dissemination, analysis, and evaluation of knowledge, positioning themselves towards that knowledge through epistemic stance markers. Stance markers are also a prominent feature of interactive, oral communication (Biber and Corad, 2009). Interactivity as a result of the synchronous nature of classroom discourse helps to explain the use of discourse organizing and deciding in real time functions, which support online language processing. This intra-register analysis illustrates the impact of disciplines and roles on the use of the two focal ES markers within the classroom discourse register, extending previous studies in this register. In line with the theoretical underpinnings of register studies, variation can be linked to the different communicative needs of instructors and students in general and the different communicative needs in the disciplines based on topics and class activities.

Conclusions and Implications
The goals of this study were to apply a functional framework based on existing descriptive research to describe the communicative functions of 'I think' and 'I guess' in a corpus of classroom discourse and to explore intra-register variation. Hand-coding each utterance revealed several functions in the corpus, thus answering the first research question. 'I think' performed epistemic stance, softening, personal perspective, discourse organizing, and real-time decision functions. Most of these functions were reported in previous studies on 'I think,' but they were not all reported in all studies. In addition, deciding in real time was not reported in previous studies. Since situational context impacts language use, it may be that certain functions only occur in specific registers or occur more frequently in those registers based on communicative needs. Studies on the frequency and functional use of ES markers should account for register and intra-register differences to provide the most accurate descriptions. This analysis also provided a description of the functions of 'I guess,' expanding on the few existing studies on this marker. Previous studies on 'I think' provided a starting point for this analysis and a point of comparison. 'I guess' performed all of the functions of 'I think' except personal perspective. Though the sample size for 'I guess' was quite small, the results preliminarily suggest these two ES markers share most macro-functions but differ across subfunctions, at least in classroom discourse. Future studies can test the extent to which a general framework describes the ES category by analyzing other ES markers while exploring additional registers.
The second research question focused on intra-register differences in terms of disciplines and speaker roles. While students used 'I guess' more frequently than instructors, both groups used the softening and perspective marking functions of the markers more frequently in humanities classes. This further advances the argument that use of ES markers is influenced by situational context and related communicative needs and is therefore reflected in register variation.
The current study aimed to balance detailed functional analyses with an exploration of statistical usage patterns, illustrating the link between multifunctionality and register variation. The results suggest that our understanding of ES markers may be incomplete without considering both register and intra-register variation. Although the sample size was relatively small, limiting the generalizability of the results, the size of the corpus allowed for a careful qualitative analysis of each transcript, which was necessary for the detailed functional analysis in this study. This would not have been feasible with a very large corpus. However, future studies may wish to apply the coding scheme developed in this study to a larger corpus to triangulate the results in classroom discourse and perhaps extend into other registers. Future research could include more detailed analyses of linguistic context, for example including utterance position and collocation as predictors of function, which was beyond the scope of the present study. Prosody is another important linguistic feature, which could not be evaluated due to the written mode of the corpus but could be included in future studies. Pragmatics researchers would benefit greatly from wider availability of multimodal corpora, at the very least containing audio files, to facilitate such analyses.

Appendix: Coding Scheme
Epistemic stance: related to the truth-value of a proposition, e.g., expressing likelihood, certainty, uncertainty, or inexactitude Softening: mitigates the force of a proposition or a potential face-threat, e.g., giving advice, correcting someone, softening something unpleasant Substitution tests -If you don't mind -I don't want to make you feel bad/embarrass you/offend anyone, but -I don't want to appear too direct/overconfident/bossy -I'm not sure of the right word / I'm not sure how to say/explain this -I'm reluctant to say / This is not ideal, but Examples -I: I think that um, we're scheduled to do evaluations.