ON METHODOLOGY AND METHODS FOR ANALYSING CLASSROOM DISCURSIVE INTERACTION: A DISCUSSION BETWEEN QUANTITATIVE AND QUALITATIVE APPROACHES

: This article presents a theoretical work aiming to discuss methodologies and methods in educational research, particularly those that analyse classroom discursive interactions. The first part examines the two main research paradigms and their ontological and epistemological bases: positivism and interpretivism. Next, two methods for analysing discursive interactions are presented and discussed, one for each paradigm. Systematic coding - within the quantitative context - is indicated to treat data from large samples, to describe general patterns, and to statistical comparisons or temporal analyses by transforming the discourse into variables. The method is applied in a set of 42 episodes of group dialogue. The results are discussed in light of the nature of the research questions, showing what types of statistical tests can be performed. Within the interpretivist paradigm, sociocultural discourse analysis is presented as an example of a qualitative method applied to excerpts from classroom dialogues. The main finding is the identification of typologies that describe how teachers and students construct scientific explanations. The final part discusses some possibilities and limits of each method. The conclusion defends that both are complementary for advancing knowledge in the educational field.


INTRODUCTION
Much research has focused on the role of language and discursive interaction in the knowledge-building process in classroom-based settings (Howe & Mercer, 2007;Mercer et al., 2020;Resnick et al., 2015;Schwarz & Baker, 2016).In the last 40 years, this research field has flourished under umbrella terms such as 'classroom dialogue' or 'dialogic teaching' (Kim & Wilkinson, 2019;Mercer & Dawes, 2014), encapsulating studies that aim to effectively use discursive interactions in the teaching context (Alexander, 2008).This stance in the educational field has been called the 'dialogic turn', emphasising students' voice, agency, and participation in the co-construction of knowledge (Wilkinson & Son, 2011).
The first large-scale study on classroom dialogue was probably Flanders's (1970), who depicted the two-thirds rule; that is, in regular classes, two-thirds of the time is filled by talk; two-thirds of this talk is teacher's talk, and two-thirds of this teacher's talk is lecturing.Another relevant finding is framed in the triadic discourse: initiation, response, and feedback or evaluation (Mehan, 1979;Sinclair & Coulthard, 1975).This pattern is possibly the most common feature in whole-class talk and can account for up to 70% of all teacher-student interactions (Cazden, 1988;Nassaji & Wells, 2000;Wells, 1993).Although many of these studies have taken place in Anglo-Saxon countries, this pattern is recognised worldwide (Alexander, 2001).
Interventionist research has investigated and identified forms of enhanced teacher-student interactions that support teaching and learning (e.g., Mercer, 2002).Perhaps the main conclusion is that forms of talking have profound implications for what is made available to learn (Howe & Mercer, 2007;Kelly, 2014;Lemke, 2001); "the quality of student learning is closely linked to the quality of classroom talk" (Nystrand, 1997, p. 29).
Review books on methods for analysing classroom dialogue have recently been published (Kershner, Hennessy, et al., 2020;Kumpulainen et al., 2009;Märtsin, 2012).Contributing to this proliferating field of research, this paper aims to reflect on the methodologies and methods to analyse classroom discursive interaction.However, instead of directly departing to analytical methods, we first offer a broad methodological discussion starting from research paradigms that consider elements such as epistemology and ontology.The second part of this paper presents methods and procedures for discursive data analysis in two veins: quantitative and qualitative approaches.Both methods are discussed and illustrated based on examples from previous research.

RESEARCH PARADIGM: A ONE-WAY RELATIONSHIP
A usual view relates methodology with selecting methods (techniques) for a study and their application for data analysis.However, a far more comprehensive notion of methodology comprises the researcher's attachment to different epistemological and theoretical traditions that influence not only the selection of methods, but also the framing of research questions and design (Mercer et al., 2004;Taber, 2012Taber, , 2013;;Treagust et al., 2014).Crotty (1998) argues that a research conceptualisation should entail four elements: epistemology, theoretical perspective, methodology, and methods.Taber (2013) considers similar elements when referring to ontological and epistemological assumptions in research: theoretical perspective, methodology, research design, and the coherent employment of techniques for acquiring and analysing data.
These elements jointly create what is often called the research paradigm, a worldview or basic belief that sets the value of research and guides the investigator in the choice of methods and some fundamental assumptions (Guba & Lincoln, 1994;Treagust et al., 2014).In other words, the research paradigm is seen as a 'reference point', 'vision', or 'philosophy' that amalgamates beliefs, values, and methodologies in research (Taber, 2013;Treagust et al., 2014).In practice, paradigms are how researchers think about and make sense of their study.
Different paradigms can employ very contrasting aspects, most fundamentally when considering the two broadest perspectives.The first paradigm is positivist, nomothetic, and confirmatory, which means that it deals with definitive and objective knowledge, seeks general patterns or laws, and tests hypotheses.The underlying assumption is that "it is possible to report unambiguous truth, in terms of observable phenomena and verified facts" (Taber, 2013, p. 49).Guba and Lincoln (1994) pointed out that this approach is grounded in realistic ontology and objectivist epistemology.Realistic ontology entails that "an apprehendable reality is assumed to exist, driven by immutable natural laws and mechanisms" (p.109).In contrast, objectivist epistemology assumes that "the investigator and the investigated 'object' are assumed to be independent entities, and the investigator to be capable of studying the object without influencing it or being influenced by it." (p. 110).This approach to social science is deemed objectivist and determinist because it sees human beings as products of the environment and responses to it (Cohen et al., 2007).
In more detail, positivist research "strives for objectivity, measurability, predictability, controllability, patterning, the construction of laws and rules of behaviour" and seeks causal explanations to produce claims (Cohen et al., 2007, p. 26;Treagust et al., 2014).Within this context, quantitative methods are more aligned with such a perspective, trying to capture social reality through predesigned categories and measurements and providing causal-effect explanations.
The second paradigm is interpretivist, idiographic, and involves discovery.As knowledge (or nature) is viewed as an inherently subjective human interpretation, this paradigm focuses on specific and contextual cases.It deals with "meanings that those participating in educational situations give to what they experience" (Taber, 2013, p. 52).Researchers under this paradigm believe that meanings are not pregiven but co-created through interaction.Therefore, they might engage with participants in activities and seek their views (Treagust et al., 2014).Guba and Lincoln (1994) said that this approach is grounded in a relativist ontology and subjectivist epistemology.Relativist ontology conceives that "realities are apprehendable in the form of multiple, intangible mental constructions, socially and experientially based" (p.110).In contrast, a subjectivist epistemology denotes that "the investigator and the object of investigation are assumed to be interactively linked so that the 'findings' are created as the investigation proceeds" (p.111).In this approach, human beings are seen as agents of their own actions and producers of their own environment.Because of this, this perspective is deemed subjectivist and voluntarist (Cohen et al., 2007).
Thus, interpretive researchers "strive to understand and interpret the world in terms of its actors" (Cohen et al., 2007, p. 26) and frame the situated meaning of human experience (Treagust et al., 2014).Employing mainly qualitative methods, such tradition interprets social situations when constructing meanings from the data.
There is criticism about both paradigms.Positivist researchers might be seen as superficial and limiting, as they do not grasp the inner contradictions and assume that the same finding or solutions can be applied in every context (Taber, 2013).At the same time, interpretivist studies have sometimes been critiqued for being anecdotal or not methodically rigorous, resulting in a lack of generalisation or scalability.
Notwithstanding the above, some researchers consider critical theory a third research paradigm (Cohen et al., 2007;Guba & Lincoln, 1994;Treagust et al., 2014).Researchers following this tradition emphasise the political and ideological stances in human interaction, arguing that power and inequality shape and forge how humans live, behave, and perceive reality (Cohen et al., 2007).
The existence of many research paradigms might not be a problem in itself, especially in educational research.Treagust and colleagues (2014) advocate that this diversity promotes the construction of more balanced knowledge and broader effort.In the same position, Taber (2013) acknowledges that any paradigm has strengths and limitations.Thus any of them can produce valuable knowledge and be employed in complementary ways.
At this point, at least three aspects must be considered concerning adopting a research paradigm.First, it must provide a consistent and coherent account across every study step (Taber, 2013, p. 55).Second, different paradigms have different purposes and methods that must lead to different outcomes (p.68).Third, and finally, any study refers to a paradigm.If it is not explicit, it is implicit in how the research questions are framed and addressed (p.68).These concerns stress the importance of thinking carefully about the relationship between the research aims and the most appropriate methods to find them.
All paradigms can be applied and generate new knowledge when analysing classroom interaction.For instance, the positivist approach demands simplifying social practices to make research manageable, generally selecting a set of dialogue variables and controlling others related to learning outcomes (Asterhan et al., 2020).This method allows us to state something about how dialogic teaching affects students' learning.On the other side, the interpretivist paradigm might provide a more nuanced frame on the dynamics of dialogue through, for example, ethnographic case studies that stretch the cultural context of the classroom setting (Asterhan et al., 2020).However, as stated above, different methods are required for each paradigm when addressing research questions of different natures.Ultimately, this is precisely the aim of this discussion paper.
The next sections are divided into quantitative or qualitative methods for didactic purposes.For each of them, there is a brief presentation of one analytical procedure and its application to real classroom discursive data from a previous project.The research questions and findings are presented here as an exemplary model of operationalising paradigms and methods.Hence, the goal is not to discuss the new knowledge on dialogic teaching generated from this data.

METHODS FOR DISCURSIVE DATA ANALYSIS: A QUANTITATIVE AND QUALITATIVE COMPARISON
First, analysing discursive interaction is an inherently subjective exercise of building meaning from the thoughts and voices of others, which are constrained and affected by social, cultural, and situated contexts.Such an analytical process involves much interpretation related to reflexivity; researchers bring their own preconceptions, interests, biases, agenda, and so on (Cohen et al., 2007).
Researchers have also discussed competing tensions between deductive and inductive approaches in analysing data (Evans, 2013;Taber, 2013).While the former looks for evidence about preestablished themes, which undoubtedly steers analysis, the latter stems from open-minded inquiry, generates themes from the data and uses them as an analytical tool (Evans, 2013).In other words, the analysis can be drawn from the conceptual framework or be grounded in data (Taber, 2013).However, both cases involve instances of creating categories, grouping them under higher-order headings, and formulating a general description of the research topic (Elo & Kyngäs, 2008).
Despite there being many approaches to analyse discursive data and interaction, it might be argued that all of them assume that the researcher must "reflect on them [data] repeatedly and at length, to be able to fragment and manipulate them in the search for underlying patterns and meanings" (Evans, 2013, p. 158).The extension in which the data is fragmented may define the use of quantitative or qualitative methods.When analysing discursive interaction, in most cases, there is a moment in which the researcher will categorise or 'code' an utterance, turn of speech or episode (Hennessy et al., 2020).
Below, two examples are provided.The first illustrates a quantitative analysis regarding smallgroup dialogue during seven different tasks.The second is a qualitative analysis carried out in episodes of wholeclass teaching about evaporation.

Quantitative approach to discursive data: systematic coding
In this section, we focus on the systematic coding method, as it is the most popular among researchers and examines interaction as a turn-taking system and categorises each of them.Within this method, the analytic process of discursive data involves reducing the data by marking relevant points for the study.This procedure is often called 'coding' and enables the researcher to "organise and structure the data" while translating "raw data into particular conceptual references" and then "identify links to the different categories" (Evans, 2013, pp. 158-159).Objectivity is a key aspect of systematic coding as it encompasses the development of a scheme that needs to be unambiguous; that is, the criteria for coding must be clear enough so that different observers assign the same code/category to the discursive unit (Galton et al., 1980).Quite frequently, codes come from a theoretical framework or previous research.In this case, this research may be labelled as a 'template' since the analytical framework is informed by the theoretical perspective "so that the analyst knows just what they are looking for in the data from the start" (Robson, 2002;Taber, 2013, p. 293).
When the codes' occurrences are counted, such a method provides a broad overview of the data sample and establishes patterns within and across events (Snell & Lefstein, 2011).However, this technique is not free of problems, as there is a trend to fragment the data and lose context or other meanings not captured in the codes (Cohen et al., 2007).Therefore, systematic coding might be considered reductionist (Mercer et al., 2004;Snell & Lefstein, 2011).The following subsection shows an example of the kinds of research questions and findings that systematic coding might provide.

Material and method
In the research reported by Lago (2021), there were three questions about group-work interaction: 1) How can small-group dialogue in a Brazilian primary school be characterised?, 2) How did it change throughout the intervention?, and 3) Did the groups become more dialogic?For context, the research project was based on a teacher professional development programme to promote dialogic talk in small-group work and whole-class teaching.
In total, 42 episodes of group talk were recorded, transcribed, and analysed.Due to the size sample and the nature of the questions, a quantitative analysis would be the most reasonable approach to find the general pattern of the discursive interaction that emerged in the groups and how such a pattern has potentially changed over time.
A coding scheme was devised based on previous works from the literature and comprised a set of ten codes relevant to dialogic classroom talk at the utterance level.The resulting scheme had ten categories divided into three contexts: content, task, and off-topic, which discriminate the utterance's tenor.That is, the codes can be arranged to distinguish if the utterance is related to the content proposed in the task (topic under discussion); if it is related to task procedures or class management; or, ultimately, if it is off-topic, talk neither focused on content nor task (Box 1).The utterances related to the task content were classified regarding their discursive functions inspired by the IRF pattern and based on the dialogic features that describe productive talk (Howe et al., 2019).This process generated six codes, two for each move of the triadic discourse: invitation (INV), dialogic invitation (D_INV), contribution (CON), dialogic contribution (D_CON), follow-up (FOL) and dialogic evaluation (D_EVA).This distinction was made only for content utterances because they are the medium in which knowledge construction and conceptual learning happen.

Code Definition and
The term 'dialogic' tries to frame the utterances in which participants make significant contributions by moving ideas forward by critically considering others' perspectives in building knowledge collectively (Kershner, Dowdall et al., 2020;Mercer, 2003).Given the interest, engagement in dialogic talk was operationalised by relatively frequent occurrences of the three dialogic codes.
To conclude, this scheme made it possible to describe classroom talk and evaluate dialogicity; both goals were at the core of the research questions.Classroom talk could be described by comparing the relative code occurrences within each context (content, instruction, off-topic) and dialogicity evaluated by considering the frequency of dialogic utterances in relation to simple ones.
This kind of positivist coding assumes that the codes can capture components or characteristics of the discursive interaction.Thus, aiming for objectivism, the coding scheme was tested for inter-reliability purposes.Three Master's students joined the researcher in piloting and refining the scheme, establishing coding rules, and achieving consistency.
In the first round, all raters coded the same group-talk episode and discussed the scheme and rules for alignment and convergence.After obtaining moderate scores, the scheme was simplified by dropping some codes, and rules were clarified and written in more detail.Two raters coded 28% of the episodes (12 out of 42), and inter-rater reliability was calculated using Cohen's Kappa scores through R (RStudio team, 2020).After the iterative process, an acceptable/excellent level of agreement was accomplished, as Cohen's Kappa scores averaged K = .79(Kappa values for each code are presented in Table 1).Then, the researcher finally coded the rest of the material .This procedure makes the scheme reliable.

Results
In addressing the first research question, the result is presented considering the entire dataset.Firstly, group talk as a whole was considered.Relative code frequencies measured the number of contentrelated (codes 1 to 6), instruction-related (code 7), and off-topic utterances (code 8) divided by the sum of all of them.Figure 1 shows that groups spent almost half (56%) of the utterances talking about the content of the task, around one-quarter negotiating the task-instruction (25%), and a reasonable number off-topic (19%).Only half of group work was devoted to content, the kind of talk more directly related to conceptual-disciplinary learning gains.The data could be further explored by searching for variations in the codes' frequencies across tasks (question 2) and using statistical analysis to find significant variances.The dataset was consistent for non-parametric tests to compare three or more populations (Fligner-Killeen's test was above the significance level for all but the D_EVA code when comparing tasks or groups, p > 0.05).This result allowed non-parametric tests to analyse comparisons among groups employing the Kruskal-Wallis and post-hoc Nemenyi tests.The 'other' and 'inaudible' codes were discarded and not considered for analysis.
Figure 2 shows code distributions for contexts of group talk as a whole and discursive functions across tasks.Only a small number of significant differences between groups were found.For instance, the analysis for context showed a significant difference in content (H(6) = 17.575, p = .007)between Tasks 3-4 (p = .037)and Tasks 3-5 (p = .022),as well as in off-topic (H(6) = 14.007, p = .029)between Tasks 3-5 (p = .046).No difference was found in instruction.Broadly speaking, students talked more about the content and less off-topic in Tasks 2 and 3.In order to answer whether the groups became more dialogic (question 3), it was compared the aggregated results for non-dialogic utterances (the sum-up of codes 1, 3 and 5) with those considered dialogic 2, 4 and 6).As the dialogic and non-dialogic measures are mirrored images, it suffices to conduct the statistical analysis for just one of them.
There are just three task-pairs that show significant differences (H(6) = 24.881,p < .001);tasks 1-2 (p = .008),tasks 2-6 (p = .001)and tasks 4-6 (p = .04).In Figure 3, it can be seen that tasks 2 was the most dialogic, followed by task 4, while tasks 1 and 6 showed moves with more non-dialogic utterances.Therefore, the hypothesis of enhancing dialogic utterances throughout tasks was not confirmed.There were only a few differences between tasks that cannot be classified as an effect of the intervention.
It is worth noting that these findings on classroom interaction were obtained without presenting any discursive extract to the reader.In some cases, researchers present a small extract to illustrate the scheme application, not as part of the analysis itself.Thus, the result is based on the codes' explanatory power and the coding process's reliability.Moreover, considering the nature of the questions, it is clear that a qualitative approach would neither offer the generalisation for the description nor objectivism to compare the change in the discursive variables over time.For example, Extract 1 shows an example of dialogic talk that did emerge in these groups: an episode carefully selected to illustrate how students interacted in high levels of dialogicity.Identifying such an episode among the others was only possible after the systematic codification and frequency counting of the entire dataset.Quantitatively, more than half of the utterances were considered dialogic (17 out of 30; around 57%).For a brief comparison, the dialogicity average across the seven tasks was around 33% (Figure 3).

Qualitative approach to discursive data: sociocultural discourse
Qualitative methods underpinned by the interpretivism paradigm usually consider the cultural and social context and practices that shape classroom interaction (e.g., Gee & Green, 1998).It is said that knowledge is socially constructed through language.These methods can include linguistic ethnography, sociolinguistics, and discourse analysis, among others (Hennessy et al., 2020).In most cases, they stress the contextual and sociocultural dimensions rather than the verbal exchange per se (turn-taking) mainly because discourse is framed in a broad sense: "the text (speech, writing, or image), the discursive practice of analysis within the text, and the social practices and structures bound to notions of power and knowledge."(Gregory, 2020).
Researchers dealing with discursive data and qualitative methods employ a variety of procedures such as identifying themes, generating units, classifying and categorising these units, structuring narratives to describe the content, interpreting scenarios and constructing theory (Cohen et al., 2007;Robson, 2002).
As an example of a qualitative method, sociocultural discourse analysis is presented and applied to trace relationships between classroom interaction and conceptual learning.Such analysis involved closely examining the episodes to generate categories and illustrate them with selected extracts.This focus enabled researchers to search not simply for a specific discursive function but to investigate whole forms of interactions that can be used to generate typologies and evaluate their impact on particular educational purposes.In addition, this method deals with language "content, function, and the ways shared understanding is developed in social context, over time" (Mercer, 2010, p. 9).
One of the strengths of sociocultural discourse analysis is that "the actual talk remains the data throughout the analysis and so the processes of the joint construction of knowledge can be examined in detail" (Mercer, 2004, p. 143).Thus, here, the focus is not the discursive functions or the language itself but the content of the talk and the pursuit of joint intellectual activity (Mercer, 2004).It is an inductive approach to inquiry from which themes or categories should emerge from the data (Evans, 2013;Taber, 2013).
Such analysis is often illustrated by selected extracts followed by a commentary regarding prior knowledge of the field and the context.It can be concerned with the syntax and the cohesive structure of language to represent how knowledge is socially built in the classroom (Mercer, 2004).As dialogue remains untouchable throughout the analysis, the emerged typologies offer a heuristic device for making sense of the talk regarding one particular issue (Mercer, 2004).For instance, Mercer's studies identified three types of group talk: disputational, cumulative, and exploratory.According to the author, the latter positively impacts student learning.Thus, the point is not to reduce the data to a categorical tally but set broad features that involve the context and the discursive dynamic regarding an educational learning goal (Mercer et al., 2004).The next subsection shows an example of the kind of research questions and finding that sociocultural discourse analysis might provide.

Material and method
Again, the data comes from the same previous project that consisted of a classroom-based intervention to promote dialogue.The second module of the TPD programme was based on predefined classroom material that included considerable student talking and thinking while allowing students to develop their own theories about water evaporation based on observations and everyday experiences (SPRinG, n.d.).
A research question was proposed to investigate how classroom talk supported students' scientific understanding.It was argued that much of the conceptual learning in primary science occurs through constructing explanations.Thus, the point was to frame and analyse how teachers and students jointly construct the scientific explanation of evaporation through talk.
However, defining what counts as an explanation or framing is not easy, despite the consensus in considering it an intellectual elaboration that accounts for a cause-effect relationship and uses a logical connective like 'because' to connect the cause and the result.Within this perspective, the data corpus was read to search for structures that resemble scientific explanations.
Considering the interpretivism nature of this analysis and the specificities of the study (primary science, scientific discourse and evaporation), a further framework was required during the data interrogation.It was needed to define 'what counts as an explanation in classroom talk within primary science'.Two structural aspects were set after looking at the data and theory reiteratively: forms of explanations and causation (Braaten & Windschitl, 2011).Thus, the analysis would search for sequences of utterances that convey instances of explanation (definitions, descriptions, and reasoning) and causation (cause-effect accounts of an observable event).Finding examples of these occurrences might make it possible to generate typologies, comment on and describe them, and discuss potential outcomes.

Results
The data corpus was analysed to select instances of explanations being built as a collective activity between teachers and students.In the 18 episodes of whole-class teaching recorded in this module, five types of explanatory sequences were elaborated and arranged into three broad discursive acts depending on whether teachers were exploring, guiding, or providing an explanation.
No clear, straightforward distinction exists between the sequences of exploring and guiding explanations, as real classroom interactions have unplanned utterances, deviations, and uncontrolled flows that make it hard to frame the sequences under a single category (Barnes, 1976).Thus, these kinds of explanatory sequences should not be taken as neat or crystalline; different categories overlap in some utterances.Mercer (1995) already raised this concern when he proposed the three kinds of group talk.
Whereas 'exploring explanations' involved advancing, extending, and adding to students' ideas, 'guiding explanations' marked sequences headed towards the scientific explanation, including an effort to use scientific terms more precisely.These two categories were further divided into four granular typologies.Finally, 'providing explanations' labelled sequences in which the teacher delivered an explanation to the students.In this sample of talk-intensive lessons, this type of direct telling was highly interactive, and students contributed with some words, but the teacher did most of the conceptual development.
The full research report (Lago, 2021) discusses these forms of collective construction while proposing a framework of reference to each typology, providing examples from lessons, and commenting on them.In this paper, for space and scope, I aim to illustrate and discuss only one of these typologies; guiding explanations through narrowing responses.
Here, two extracts illustrate the excessive use of cued elicitation (Mercer, 1995) and Socratic questioning (Roth, 1996), respectively; that is, the use of limiting invitations that constrain the discourse to a narrow line of reasoning.Students just needed to give a few-word response to follow the explanation across the scientific point of view.In the first case, the lesson aim was to compare water and perfume evaporation.As students could not see or feel the water evaporating, the perfume showed that something was leaving the water and spreading into the air.The teacher developed this reasoning (Extract 2).
In the teacher's third move (line 82), she used a strong intonation ("SEE") to reject a student's answer.In the next move (line 84), she reinforced the correct answer with a justification: "We don't see them, these particles are tiny, they are invisible".She asked the reason for bringing the perfume to the classroom.Knowing that the visual effect was not the answer, a student argued about the feeling.
Although this was a why question, there was so much indication of the correct answer that it was not coded as a dialogic invitation.Then, in lines 87 and 88, the teacher developed a largely complete explanation, leaving only the final word unsaid for student completion-'vapour'.Having this expected response, she moved on with the explanation relating to vapour, the scent, and its spreading around the room.It is an explanative construction, despite the narrow, inducing line of reasoning the teacher pursued.In the following extract (3), students had just read the brief text about particle theory and were answering handout questions collectively.During the interactions, the teacher generally accepted responses like 'separate', 'get further away from each other', and 'move faster'.She also explicitly asked for two other 'things' (lines 50 and 53) whereby she readily obtained 'energy' and 'freely'.The teacher seemingly aimed to find keywords from the text in students' voices.With these expected answers at hand (or their synonyms), the teacher provided the entire explanation in the last line.Again, these questions were not coded as dialogic (lines 50, 53, and 55) because expected answers were in the handout; students were merely reproducing them aloud and, as line 56 shows, hesitantly.It might the said that students' text comprehension was explored, but it is harder to say that explanations were explored.Neither example shows clear evidence of exploring or deepening students' ideas.Instead, it might be said that the teachers had clear objectives and tried to achieve them with students.I would say that they were successful in highlighting the comparative analogy between water and perfume evaporation and emphasising a microscopic explanation of the evaporation process.The student contributed with few words and, apparently, followed the teacher's line of reasoning.
This type of micro-analysis of the utterances' content in these extracts has allowed us to explore some factors in more depth, like the building of conceptual understanding and results of each discursive strategy employed by the teacher.It was possible to highlight a particular classroom activity and discursive techniques that had an impact on shaping classroom talk.

DISCUSSION AND CONCLUSION
In this paper, we have discussed two research paradigms and related them to two methodological approaches to analysing educational dialogue; systematic coding (positivism-quantitative) and sociocultural discourse analysis (interpretivism-qualitative).The first consisted of a coding scheme in which utterances were allocated to predefined categories that could be broadened or further broken down.Hennessy et al. (2020) list some advantages of frequency counting, such as processing large quantities of data, highlighting key markers, searching the dataset efficiently to find how specific acts correlate with others, detecting patterns, measuring change in practice and making comparisons subjected to statistical analysis.
Such a method allowed us to frame a general pattern of the small-group talk that emerged in a Brazilian primary school and search for differences across time.This last procedure was done by measuring the relative frequencies of the occurrence of codes.Thus, the findings represented here have illustrated the positive aspects highlighted above.The great loss in this analysis is the temporal development of meanings and ambiguity because utterances can have multiple functions.In some cases, coding is said to be atheoretical or dogmatic in its conclusions (Hennessy et al., 2020).
The second method explored in the paper is sociocultural discourse analysis.It highlights the contextual and situatedness of turns of speech while deepening the meaning of contributions.Qualitative analysis can highlight the "participants' underlying intentions and responses to others.Moreover, it enables the researcher to go beyond the data, for instance, to identify missed opportunities for extension or challenge."(Hennessy et al., 2020, p. 6).Categorising extracts as a whole can build models rather than recording the presence/absence of target discourse moves in a way in which categories/typologies are generated as outcomes rather than predetermined (Mercer, 2010).
This method made it possible to see how teachers and students work together to build scientific explanations through classroom talk.From the five typologies identified in the original work, only one was presented here.We could see how teachers can guide students' ideas through a line of reasoning while building explanations interactively.Strategies like confrontation, response narrowing, and questioning were employed.
Although quantitative and qualitative methods have been contrasted, both methodologies play relevant roles and can be complementary (Snell & Lefstein, 2011).Systematic coding is a method for managing an extensive data set and identifying general patterns across multiple lessons /tasks while allowing statistical comparisons and developmental analysis (Mercer et al., 2004).On the other hand, sociocultural discourse analysis can explore fine-grained and offer a multi-dimensional understanding of the complexities of classroom dialogue (Asterhan et al., 2020).While the former constricts the discursive interactions into measured variables (positivist ideal), the latter reveals the complexity and richness of the discursive exchanges with a greater interpretative vein.
From a theoretical point of view, Filho (2013) brings an instigating theoretical discussion on the incommensurability, complementary, and unity of the paradigms.Researchers from the first thesis argue that the realism and interpretivism views are in strict opposition and therefore are incompatible.The ones who advocate in favour of complementarity say that both paradigms are legitimate as they are not necessarily in conflict.Moreover, empirically, each paradigm aligns with goals that another could not tackle.Finally, the unity between these paradigms is considered mainly by pos-positivist and critical philosophers who defend that there is no logical or consistent way to divide knowledge into radical terms.
Here, I argue for a complementarity stance mainly in the analytical-empirical dimension due to the nature of the research questions.Both quanti-and qualitative methods can be required for educational studies (Souza & Kerbauy, 2017).Using a mixed-method approach to methodology (Tashakkori & Teddlie, 1998) allows the juxtaposition of different languages to answer research questions properly and coherently.As a result, while exploratory research questions privilege a qualitative analysis, others present a confirmatory bias and should employ quantitative characterisations and comparisons.

Figure 1 :
Figure 1: Small-group talk divided among the three contexts.

Figure 2 :
Figure 2: Variation of the three contexts of talk across tasks.

58
When they turn to vapour, they gain energy.They detach from each other.It is what happened in the dramatisation.They detach from each other and are free in their surroundings.They move more quickly.
Because the wind blows and dries it', 'Vapour rises up and makes clouds'.Assess the status of the answer by stating if it is wrong and must be followed by grounds (reasons, evidence, or elaboration).Examples: 'No, because the wind makes the water moist and the sun turns it into smoke', 'You are wrong, the water evaporates, it turns into tiny droplets that form all the clouds'.Box 1. Coding scheme comprised of ten codes: description and examples (The numbers in the first column are the Cohen's Kappa scores of the interrater reliability).