Analysis of teacher-learner and learner-learner dialogues, and their role in the aquisition of skills and understanding

Andrew Tolmie
Professor of Education Institute of Education, University of London

Methodological narrative

The fundamental question addressed by the research highlighted in this showcase – and one which is central to the process of education – is how the acquisition of new skills and understanding is managed psychologically by those working with learners, so as to obviate the need for knowledge to be reinvented from scratch. This question, and the methodologies that have been adopted to address it, both derive ultimately from the social constructivism of Vygotsky (1978), although some of the research outlined below was only distantly connected to this perspective in its original conception.

Vygotsky's account of learning holds that novices are guided by more expert others towards the completion of activities that are a little beyond the scope of their current personal capability. Guidance within this ‘zone of proximal development' provides learners with an external window on the structure of activities, and with the language necessary to organise and direct their action within it. By first appropriating and then internalising these means of controlling their behaviour, learners are inducted into existing practices and the ways of thinking that are associated with them. This makes it possible to adapt this understanding further, and to pass it on to others in turn (see Cole, 1996, for contemporary developments of Vygotsky's initial theoretical and methodological ideas).

This theoretical framework has led to much research on apprenticeship processes in learning, especially scaffolding (see e.g. Wood, 1986), and on the adaptation and reconstruction of acquired knowledge, especially via collaboration and discussion with others. Wood's research on mothers' direction of their child's actions in a problem-solving task established a key methodological principle, the detailed analysis of moment-to-moment interactions, and examination of the association of features of these with subsequent performance. However, it concentrated on examining the means by which control of action was shifted from mother to child, at the expense of consideration of how the language used affected children's understanding and performance. Given the role ascribed to language by Vygotskian theory in guiding or mediating action, and in shaping thought, this was an important limitation. Moreover, the focus of the Wood research on informal learning contexts undermined its apparent relevance to formal educational settings.

Our research during the 1990s on collaborative group work in science among primary, secondary and tertiary level students provided an arena for the gradual development of a quantitative methodology that addresses both concerns. This methodology (which is still evolving) focuses on detailed analysis of language content during interaction in relation to subsequent action and conceptual grasp (NB it is important to note that in virtually all this research, it is the group which forms the unit of analysis for dialogue variables, not individual members, even though dialogue is related to individual learning; this is because early analyses made it apparent that it was exposure to the dialogue that mattered, not whether the learner was speaker or listener). By framing analyses in quantitative terms, it becomes possible to build up generalised statistical models that can be applied and tested within classrooms as well as other contexts.

This approach was manifested initially in work by Howe, Tolmie & Rodgers (1992). This utilised a semi-naturalistic setting, in which groups of learners worked out of class, under light supervision, on an extended practical task which investigated the factors affecting the speed of toy cars down inclined slopes. This setting enabled good quality video records to be made of group interaction. These were used to score the quality of group members' explanations of the observations made during the task, and how far others agreed with them. These scores were then analysed for statistical correlation with indices of individual change in understanding of the factors affecting vehicle speed, between a pre-test and both an immediate and a delayed post-test. This made it possible to examine the extent to which conceptual grasp was influenced by the content of group dialogue.

Tolmie, Howe, Mackenzie & Greer (1993) further refined this approach, by using a more differentiated and quantified analysis of the content of group interaction, this time within a task that examined the floating and sinking of objects. The frequency of three types of statement was noted: explanatory references to the factors affecting flotation; predictions or propositions about whether an object would float or sink; and attempts to chair or organise group activity. The frequency measures were then analysed for variation across different task formats, as well as for their degree of correlation with changes in understanding.

Tolmie & Howe (1993) employed a still more detailed frequency-based coding of group dialogue. This focused on the predictions and explanations made at different stages of activity on a computer-based task (e.g. comparison of earlier individual predictions) that examined the trajectories followed by propelled objects. This coding provided the basis for a form of path analysis (statistical modelling of causal relationships among a set of variables) to examine how explanations arose during interaction, how they fed into changes in understanding, and how the pattern of dialogue varied according to group gender composition.

Following on from this work, interest shifted to the manner in which task resources might be used to nudge dialogue into forms that were now known to be productive. Howe & Tolmie (1998) summarised research which took the particular step of introducing computer-based scaffolding into the group work context. This was used as a means of supporting decision-making about the tests needed to determine the factors that affect either the size of shadows, or the pressure in a column of water. Analysis here focused on the frequency with which corrective prompts regarding test design were issued by the software (cf. Wood's mothers), and correlations between these prompts, the incidence of group-constructed ‘fair' tests, and the level of understanding exhibited by participants at post-test.

In a further shift of both focus and context, elements from all these approaches were brought together by Tolmie, Thomson, Foot, Whelan, Morrison & McLaren (2005). This research examined relationships between tutor dialogue, learner dialogue and subsequent performance , in both individual and small group training of road-crossing skills using computer simulations. The dialogue codes used here were chosen to reflect the full range of those that appeared influential in previous work on tutoring and group collaboration. A key innovation, though, was the use of a common code for explanations for both tutors and learners, and for both training sessions and individual test performance. This allowed correlational methods to be used to track the appropriation of tutor explanations by learners across four training sessions, on into their subsequent appearance during post-testing. In this way, the direct effects of tutor language on learners' understanding and performance could be examined, redressing one of the limitations of the earlier research by Wood.

Philips & Tolmie (2007) returned to the context of science, utilising similar methods to analyse the effects of parent-child interaction during scaffolding of a balance scale task. The approach taken here was more fine-grained in two important respects, however. First, parent and child explanations were differentiated into four types of increasing sophistication, allowing the appropriation of more specific forms of language content to be tracked. Second, the relationship of parent and child explanations to the latter's attempts to make the scale balance was examined problem to problem , rather than across whole sessions. This microgenetic approach derives directly from Vygotsky: “To encompass in research the process of a given thing's development in all its phases and changes…fundamentally means to discover its nature, its essence, for ‘it is only in movement that a body shows what it is'.” (Vygotsky, 1978, p65). This methodology made it possible to dispense with post-testing, whilst permitting the shifting nature of explanation-performance relationships to be examined in detail as appropriation proceeded, and in relation to children's initial level of understanding.

Finally, addressing the second limitation of the Wood research, Howe, Tolmie, Thurston, Topping, Christie, Livingston, Jessiman & Donaldson (2007) showed how pared-down versions of such coding systems could be successfully employed within classroom-based research. This research, which was conducted as part of TLRP, focused on teacher support for collaborative group work in two areas of science, forces and evaporation. Video recordings were replaced with live classroom observers , to reduce intrusion. These observers noted the frequency of seven types of pupil dialogue, including propositions and explanations , across fixed-length observation windows in both whole class and group work lessons. They also made ratings of the extent to which features supportive of productive group work were present, including nine elements of teacher scaffolding . Correlational methods, including regression , were used to examine how far pupil performance on post-tests of individual understanding was related to the incidence of both dialogue and teacher support.

TLRP Resources for Research in Education by Teaching and Learning Research Programme is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 2.0 UK: England & Wales License