Introduction
Instruction practices and theoretical discourses in language education have focused increasingly on the individual language learner as a multilingual subject, considering different institutional and cultural–historical contexts. Classrooms are gradually moving away from regimented, teacher-centered approaches to language teaching and learning that employed mechanical metaphors such as input, output, and uptake and minimized and, at times, negated the human agency of language learners. In education in general, we have seen a paradigmatic shift toward learner-centered and learning-centered approaches. In applied linguistics, more integrative approaches, which ascribe agency to both the teacher and the learner, have gained traction. Sociocultural theory (Lantolf & Poehner, Reference Lantolf and Poehner2014) captures language learners as the central subject of a dynamic activity system; Larsen-Freeman (Larsen-Freeman, Reference Larsen-Freeman1997; Larsen-Freeman & Cameron, Reference Larsen-Freeman and Cameron2008) began to conceptualize the language learning process as a complex dynamic system, and this nonreductionist scientific theory and approach has been adopted by researchers worldwide. Societies in many countries and regions pay more attention to issues of accessibility, diversity, equity, and inclusion. All these advancements have impacted and even shaped the research, development, and praxis in computer-assisted language learning (CALL). This makes it timely to investigate adaptive technologies and adaptive instruction in the context of CALL.
The move to adaptive technologies and instruction has some inherent challenges. Individualized, learner-centered language teaching is resource-intensive. Ideally, groups would be small, language exposure would be expansive, and language use opportunities for learners would be manifold and varied. Obviously there are tensions between this idealism on the one hand and political and economic constraints and societal belief systems about language and learning on the other. However, more and more economies and states recognize the economic and societal need for many employees and citizens to have (professional) proficiency in other languages and transcultural awareness of the peoples who speak them. We would submit that these tensions cannot be resolved exclusively by turning to (digital) technologies but they can be addressed in part through appropriate technology mediation. We believe that CALL can make and has made a considerable contribution to furthering language education. And, to have more of a positive impact, it is worthwhile for researchers, developers, and teachers to reflect on adaptivity and to work with adaptive technologies.
Adolescents and young adults, in particular, use digital technologies in many domains of their life, and often for considerable lengths of time every day. Given the proliferation of digital artifacts, one could expect their prominence in language education, but as the sudden and rapid shift to emergency online teaching during the height of the global pandemic has shown, the use of these technologies is not yet understood sufficiently in the language teaching and language learning community, although there is considerable enthusiasm for technology-mediated language learning. On the one hand, CALL, in the forty years of its existence, has gained breadth and depth in language education in many countries. The many academic journals and books demonstrate the significant progress made in research and development over these decades. On the other hand, we still know relatively little about how exactly language learners use and engage with technology when conducting a learning activity or performing a communicative task. Why do some learners thrive in certain digital or online learning environments, while others suffer or fail? What do learners actually do when they engage with digital learning tools – as opposed to what developers think and students say they do (Chun, Reference Chun2013; Fischer, Reference Fischer2007, Reference Fischer and Stockwell2012)? What exactly happens when students learn and use the language on the computer – tutorial CALL (Heift & Schulze, Reference Heift and Schulze2015; Hubbard & Bradin-Siskin, Reference Hubbard and Bradin-Siskin2004; Schulze, Reference Schulze, Hampel and Stickler2024) – or in communicative interaction via the computer – computer-mediated communication (Kelsey & St. Amant, Reference Kelsey and St. Amant2008; Thorne, Reference Thorne and Hornberger2008)? And what is the role of adaptivity in these environments, activities, and interactions?
Adaptivity is both a prerequisite for and an outcome of human social interaction. When we interact in communication, cooperation, or collaboration, each actor adapts in one way or another to other actors, the context, the available tools, and the object or goal of the interaction. This is why, in complex dynamic systems research, we talk about coadaptation; all actors adapt to one another, in interaction. For coadaptation to occur in human–computer interaction, the technology needs to have some degree of adaptivity built in. Adaptive instruction, which is a term used in computer-assisted learning, is defined as the capability of a system to alter its behavior according to learner needs, prior individual performance, and other characteristics (Shute & Zapata-Rivera, 2008). A considerable amount of research has demonstrated that individualized instruction is superior to the uniform approach of one-size-fits-all teaching (Kulik, Kulik, & Bangert-Drowns, Reference Kulik, Kulik and Bangert-Drowns1990). Recent advances in technology and their integration in instructional design have facilitated individualization further. Personalized instruction can be offered simultaneously to large groups of learners (Lee & Park, 2008). Two questions then arise: How can the system or tool adapt to the learner, and how can the learner or teacher adapt to the technology (Vandewaetere et al., Reference Vandewaetere, Desmet and Clarebout2011) in the complex context of language education?
What can the human actors do? Instructors can adapt instructional sequences, pedagogic strategies and methods, and the digital, pedagogic, and linguistic affordances to individual students during or between iterations of learning processes, while students can learn to adapt to using appropriate digital affordances more effectively (Conole & Dyke, Reference Conole and Dyke2004). Developers of learning tools and systems can conduct thorough needs analyses during the conceptualization and design phases in order to achieve a better learner or teacher fit (or both), at least across specific (sub)groups. During the evaluation phase (Hubbard, Reference Hubbard and Pennington1996), the pedagogic evaluation is at least as important as the computational testing. It needs to include a valid empirical study of the tool’s or system’s efficacy in students’ second language development. Such efficacy studies need to look beyond student perception (did students appreciate the use of the digital system or tool, and do they believe they benefited?) and focus on improved factors of the learning processes and improved learning outcomes.
The focus of this chapter is on the following question: What can be done to facilitate the coadaptation in the technology-mediated teaching and learning process on the side of the digital tool or system? This research and development in computer-assisted learning in general is called adaptive instruction. The literature in this field identifies three major approaches (Lee & Park, Reference Lee, Park, Spector, Merril, van Merriënboer and Driscoll2007; see also Regian & Shute, Reference Regian and Shute2013):
1. Adapt instruction at the macrolevel by providing various alternatives for only a few main components of instruction. Instructional alternatives at the very broad level are selected mostly on the basis of the student’s instructional goals, general ability, and achievement levels in the curriculum structure. Within language education, the instructional goals can vary by primary focus on a specific sublanguage (e.g. language for business, language for the medical profession, language for the diplomatic service, language for heritage speakers). General ability is usually gauged using standardized proficiency levels and their indicators and test results. Achievement levels are often operationalized – at a very simple level – as prior or prerequisite courses. Thus, at the macrolevel, this adaptivity is done via scaffolded student selection. For example, they select one of a few available paths through an online course at the beginning, and then the chosen path remains static. To give a recent example, the results of the study by El-Sabagh (Reference El-Sabagh2021) indicate the potential of an adaptive e‐learning environment to engage students more in the teaching and learning process (p. 1). In this system, the level of instruction changes on the basis of a student’s predetermined learning style. Four different learning styles (visual, auditory, reading/writing, kinesthetic (VARK)) (p. 14) were determined, before the students were presented with the learning activities and materials. This early informed selection was more conducive to learning success for the student.
2. The second approach is to adapt specific instructional procedures and strategies to specific student characteristics. This approach necessitates the prior determination and testing of at least some relevant learner characteristics. These characteristics are also called aptitudes. Hence this approach to adaptive instruction is called “aptitude–treatment interactions” (ATI). We are not aware of this approach being employed in CALL.
3. The third approach is to adapt instruction at a microlevel, by diagnosing the student’s specific learning needs during instruction.
The predominant focus of this chapter is the third approach, micro-adaptive instruction, which focuses on information processing, diagnoses learners’ contingent learning needs during instruction at a fine-grained level, and subsequently provides appropriate instructional (re)mediation for these needs (Mödritscher, Garcia-Barrios, & Gütl, Reference Mödritscher, Gütl, García-Barrios and Maurer2004). Unlike macro-adaptive models, micro-adaptive approaches consider learner variables or characteristics. These are dynamic and use within-activity measures or fluid learner characteristics to define the most appropriate instructional intervention for a given micro-situation (Lee & Park, 2008).
Often under different labels, adaptive instruction using the microlevel approach has been applied in a number of subfields of CALL:
Tutorial CALL: sophisticated branching algorithms (see Heift & Schulze, Reference Heift and Schulze2015; Schulze, Reference Schulze, Hampel and Stickler2024), which relied on regular expressions to analyze student input in simple language practice software in the 1980s and 1990s, guided students from one (set of) activity item(s) to the next, depending on their score for or answer to this item, to the next. In other words, each student who answered differently saw a different subsequent item.
Online language learning: The student results for pre-task quizzes, for example, can determine when students are deemed to be ready for tackling a (graded) learning task or whether remedial or review activities on a different dynamic webpage are necessary first. Although such adaptive instruction happens at the microlevel, the range of learner characteristics considered is often very limited. Frequently, it is only one prior performance that is used as determining information at key moments in the course.
Intelligent CALL (ICALL): ICALL uses techniques, approaches, and tools from artificial intelligence (for an overview, see Heift & Schulze, Reference Heift2007; Schulze, Reference Schulze, Chapelle, Chung and Xu2008). The inclusion of natural language processes allows for a fine-grained structural analysis of learner texts. The results of this analysis are then the basis for contingent corrective feedback, (linguistic) learner guidance and help, and remedial activities. Learner data are analyzed and structured in individual learner profiles, which in turn form the basis for a learner model. Learner models (Bull & McKay, Reference Bull, McKay, Lester, Vicari and Paraguaçu2004; Schulze, Reference Schulze, Chapelle, Chung and Xu2008) inform adaptive instruction by drawing inferences from the structured learner data in the profile. These inferences “estimate” learner beliefs at a fine-grained level.
Intelligent Language Tutoring Systems (ILTS): These rely on natural language processing and include a learner model. Heift (Reference Heift2010) is a good example of only a handful of ILTS research projects that resulted in a system that has been used consistently by groups of students. Systems such as the e-Tutor (Heift, Reference Heift2010), Tagarella (Amaral, Meurers, & Ziai, Reference Amaral, Meurers and Ziai2011), and RoboSensei (Nagata, Reference Nagata2009), emerged from PhD projects, which are then challenging to sustain under very different institutional and entrepreneurial constraints and in diverse and ever-changing curricular contexts. Other factors that have thus far impeded the broad adoption and implementation of these systems, which are highly conducive to adaptive instruction, are their inherent features. Intelligent Language Tutoring Systems are (a) complex, in that they contain an expert model with a comprehensive information structure about the domain to be learnt – this is often the natural language parser with its grammar and lexicon; (b) expensive both to develop and to maintain, given their computational, linguistic, and pedagogic complexity; and (c) narrowly tailored to very specific learning contexts, which are designed to curtail their complexity and render them feasible. They are language-specific (e.g. specific to German, Portuguese, Japanese), connected to a specific course textbook, and limited to specific course or language aspects and components.
Adaptive digital textbooks: To our knowledge, these have not been researched in CALL, although the general literature suggests that adaptive digital textbooks enhance students’ learning, and textbooks are used widely in language education. Chau et al. (Reference Chau, Labutov, Thaker, He and Brusilovsky2021) includes an excellent overview of past and current developments in this general area. It is our hope that this research into adaptive technologies will also be carried out in CALL.
In the following sections, our focus will be exclusively on adaptive instruction at the microlevel. First we will discuss commensurate theoretical approaches to the analysis and design of adaptive instruction. This will be followed by methods that are appropriate for the empirical research that needs to accompany further development, implementation, and evaluation. The chapter will conclude with an overview of the practical use of adaptive instruction in CALL and its impact.
Theoretical Approaches in Analysis and Design
Language development and language use in CALL are mediated by computational technologies. In computer-mediated communication (Kelsey & St. Amant, Reference Kelsey and St. Amant2008), learners interact with other learners, instructors, and first language speakers via digital artifacts; in tutorial CALL (Heift & Schulze, Reference Heift and Schulze2015, Hubbard & Bradin Siskin, Reference Hubbard and Bradin-Siskin2004, Schulze, Reference Schulze, Hampel and Stickler2024), learners interact directly with socially, culturally, and cognitively imbued digital artifacts. In other words, digital components are “added” to the complexity of language use and second language development. This results in increased levels of complexity, but also facilitates the unobtrusive recording of the teaching-and-learning process through the tracking of learner behavior and the documenting of learning outcomes over time. Both tracking learner behavior and documenting learning outcomes offer windows into the processes of technology-mediated language learning. We are suggesting two complementary theoretical lenses – sociocultural theory (Lantolf & Poehner, Reference Lantolf and Poehner2014; Lantolf & Thorne, Reference Lantolf and Thorne2006) and complex dynamic systems theory (Larsen Freeman, Reference Larsen-Freeman1997; Larsen Freeman & Cameron, Reference Larsen-Freeman and Cameron2008) – for theoretical and empirical research, system and tool development, and pedagogic evaluation. Both theories have explanatory rather than predictive power, and therefore lend themselves particularly to post hoc data analysis and sense-making. They are nonreductionist, in the sense that they neither require nor allow the isolation of only one or more variables when the complex process of a learning activity is investigated in context and over time.
Sociocultural theory traces its provenience to (cultural–historical) activity theory (Rubinstein, Reference Rubinstein1984; Vygotsky, Reference Vygotsky1978). Vygotskyan activity theory is essentially a theory of the development of the mind. Its conceptualization is based on the concept of activity systems (Figure 12.1).
An activity is a bounded set of actions, which have an intention, and operations, which are triggered by an environmental condition. Actions and operations are closely intertwined and pursue one (complex) intension, the abstract object of the activity. Objects are not unlike learning goals and objectives, which students have internalized. In an activity system, language learners are conceptualized as the collective subject. This also means that all learning activities are social interactions with multiple actors, who all (and not only the subject) have agency. Within an activity, the learners – subject – pursue an abstract learning goal – object – and there is an interaction between subject and object. This interaction is mediated by abstract or material tools, which are often called artifacts. Examples of abstract artifacts are the learnt language itself, an idea or concept, and one more or less complex knowledge item or belief. In CALL research, material artifacts are digital tools and systems. Of course, other tools of the language classroom such as textbooks, speaker systems, recording devices, and whiteboards are also material artifacts that mediate the interaction between subject and object. If these tools are well designed and appropriately integrated, they help learners achieve their learning goal. The social interaction in each activity, with its material manipulations of speaking and writing, with its use of (digital) artifacts, also results in a material outcome when the object (goal) of the activity has been met. All activities are performed and take place in a concrete context. The activity system captures the main components of this context: community, rules, and division of labor. The division of labor denotes the various ways of work-sharing and the team roles that the collective subject, a group of language students, performs during an activity. The rules come from cultural and societal norms and conventions, regulations of educational systems and institutions, curricular stipulations and guidelines, and classroom management and teacher instructions. In CALL, they can also stem from software license and user agreements and regulations for appropriate technology use. Last but not least, the community comprises actors who are peripheral to the activity – they might not be present during the activity – but are relevant to it. These are teachers and teaching assistants, students’ guardians, caretakers, friends, and students from other groups.
These six components of an activity system are relevant to the research on adaptive instruction because they facilitate the comprehensive conceptualization of the place, function, and impact of digital artifacts in teaching-and-learning processes in CALL when these artifacts are developed, used, or evaluated. During the dynamic interaction of the components with one another during the activity, tensions arise between various components. It is through the gradual and often contradictory resolution of such system-inherent, ever-changing tensions that the activity systems move forward and an outcome is created, the object can be met, and learning takes place.
This sketch of an activity system highlights its complexity. This complexity is suitably captured and comprehensively considered through a dynamic-systems lens.
Like Larsen-Freeman, Ellis, de Bot, van Geert, Verspoor, and others (see e.g. de Bot, Lowie, & Verspoor, Reference de Bot, Lowie and Verspoor2007; Ellis & Larsen-Freeman, Reference Ellis and Larsen-Freeman2006; Larsen-Freeman, Reference Larsen-Freeman1997; Larsen-Freeman & Cameron, Reference Larsen-Freeman and Cameron2008; Verspoor, de Bot, & Lowie, Reference Verspoor, De Bot and Lowie2011), we view second language development as a complex, dynamic subsystem within a social system (de Bot et al., Reference de Bot, Lowie and Verspoor2007, p. 14; Larsen-Freeman & Cameron, Reference Larsen-Freeman and Cameron2008, p. 35). Complexity here refers to the multitude of variables that affect the open system, that is, the teaching-and-learning process. These variables are often dependent upon each other and/or interact in other ways and thus change and/or are being changed in the process of learning and using a second language. This ongoing change is the main reason for the system’s being described as dynamic. From a complexity science perspective, second-language development processes are complex, sensitive to initial conditions, nonlinear, and nonmonotonic, as well as fractal and nonperiodic (Lorenz, Reference Lorenz1993, pp. 161–179; Schulze, Reference Schulze, Chapelle, Chung and Xu2008). Subprocesses such as developmental spurts, backsliding, and fossilization are evidence that a second language is being acquired at varying speeds, generating nonlinear developmental trajectories of individuals. As a result of individual language learner differences, this diachronic variation is compounded by the synchronic variability that exists within groups of language learners.
Researching Adaptivity: Methods
When investigating the teaching and learning of a language through the lens of both complex dynamic systems (Larsen-Freeman, Reference Larsen-Freeman1997; Larsen-Freeman & Cameron, Reference Larsen-Freeman and Cameron2008) and sociocultural theory (Lantolf & Poehner, Reference Lantolf and Poehner2014; Lantolf & Thorne, Reference Lantolf and Thorne2006), qualitative and quantitative methods of analysis need to be employed (a) to yield rich data from the interaction of members of the collective subject with one another, with others, and, importantly in CALL, with (digital) artifacts in specific and complex learning contexts and (b) to structure these data for a meaningful and pedagogically fruitful interpretation. Then these learner data provide insight into individual learner trajectories in their second language development and into concrete language learning processes, as well as into the characteristics of suitable and effective learning artifacts and their affordances.
To capture, describe, and interpret individual learner trajectories,
these data can be plotted on a time series graph, with the relevant learner variable on the y-axis and time on the x-axis, to show development over time;
the dynamic change of the variable that is researched can be visualized in a phase space diagram, which has the value of the variable at time t on the x-axis and the value of the same variable at time t+1 on the y-axis, to show change over time;
the initial conditions – the many learner variables at the very beginning of the teaching-and-learning process under investigation – need to be recorded in great detail; and
and each individual trajectory needs to be considered in the context of the group (collective subject).
For example, Scholz and Schulze (Reference Scholz and Schulze2017) employ surveys, data tracking, and clustering and pairwise comparison to analyze students’ individual trajectories in the context of digital game-based language learning. In their study, the characteristics of the complex dynamic system are the basis for their selection of suitable variables, which in turn inform the path of analysis of both process and product data.
To identify the components of a complex system in language learning, we employ for our analysis the activity system (Engeström, Reference Engeström1987; Rubinstein, Reference Rubinstein1984; Vygotsky, Reference Vygotsky1978), which is more widely known in applied linguistics as sociocultural theory (Lantolf & Thorne, Reference Lantolf and Thorne2006; Thorne, Reference Thorne2005). Describing a teaching-and-learning process, we use observational data to capture the dynamic components – subject (digital artifacts), object, rules, division of labor, community, and outcome – systematically and in fine detail. Each component changes over time also in its relationships and its interaction with other components of the activity system. We investigate the resulting tensions between the components, because it is these tensions that move the system forward. The second language development of the subject is the nonlinear result of their processing the tensions in the activity system.
In CALL, we pay particular attention to the role of the various digital artifacts, for example language learning apps, components of online language courses, social media tools and sites, and web services, because these artifacts create additional affordances for learners. Affordances emerge “in a three-way interaction between actors, their mediational means, and the environments” (Kaptelinin & Nardi Reference Kaptelinin and Nardi2012, p. 974, as quoted in Blin, Reference Blin, Caws and Hamel2016). The diversity of educational contexts, the wealth of digital technologies, and their successful integration into diverse language education contexts require careful analysis and evaluation (Colpaert, Reference Colpaert2006) and motivate the increasing need for empirical study of the complex interaction of subjects, artifacts, and contexts (Lafford, Reference Lafford2009; Verillon & Rabardel, Reference Verillon and Rabardel1995). The results of such studies can inform the design of digital tools and systems in adaptive instruction. Within the field of educational ergonomics, researchers have argued that a holistic approach to learning design will help us understand better what learners actually do when they are working with technology (Bertin and Gravé, Reference Bertin, Gravé, Bertin, Gravé and Nancy-Combes2010; Raby, Reference Raby, Egbert and Petrie2005). Within this approach, recommendations have been made to focus the research not solely on the design of the system but, more importantly, on the learner and the learning task (Chapelle, Reference Chapelle2001; Colpaert, Reference Colpaert2006; Felix, Reference Felix2005). We take the artifacts as our point of departure and focus on how their design affects both the user and the task. Here we follow Norman (Reference Norman and Carroll1991), who argued that a clear understanding of the role played by the cultural artifacts – in our case, a variety of CALL tools and resources – is critical to the improvement of their design. He states that
[e]very artifact has both a system and a personal view, and they are often very different in appearance. From the system view, the artifact appears to expand some functional capacity of the task performer. From the personal view, the artifact has replaced the original task with a different task, one that may have radically different cognitive requirements and use radically different cognitive capacities than the original task.
This concept is essential for the development, evaluation, and implementation of such digital artifacts in educational contexts because the ways in which our learners are affected by new learning environments and learning tasks and their effect on task performance need to be considered both cognitively and functionally.
These factors play a key role in our definition of different learner types, which are often called learner personas in adaptive instruction. In CALL, personas are archetypal users of a learning tool who represent the needs of larger groups of users in terms of their goals and personal characteristics. Personas capture and cluster similarities among learners and are based on interconnected clusters of learner characteristics. The usefulness of personas in defining and designing interactive applications is based on ideas advanced by Cooper (Reference Cooper1999). In contrast to iterative user prototyping, the more powerful method is to make up “pretend users and design for them” on the basis of in-depth ethnographic data (p. 123). Clearly, it is impossible to capture each and every trait of an individual learner. However, by creating distinct personas (Heift, Reference Heift2007), we can capture and cluster essential similarities and differences among learners who warrant individualization (see Colpaert, Reference Colpaert2006; Heift, Reference Heift2002, Reference Heift2008; Levy & Stockwell, Reference Levy and Stockwell2006) and that result in more adaptive instruction. Once the similarities and differences have been determined, the learning process can be modeled to enhance the learner–computer interaction with an individualized, adaptive CALL tool or system. For instance, instructional alternatives with respect to learning objectives, tasks, and media, or the use of learning tools, can be provided. We can decide whether this information is static and hard-wired into the learning tool or is dynamic – that is, changes over time and adjusts to our learners as they develop.
Lilley, Pyper, and Attwood (Reference Lilley, Pyper and Attwood2012) make a distinction between ad hoc personas and data-driven personas. Ad hoc personas are defined during the conceptualization phase and are based on preconceptions of what software designers think users might be like. In contrast, data-driven personas are established through data collection from actual users. This collection includes data on user demographics gathered through user surveys and concurrent system interactions. Although both ad hoc and data-driven personas are abstract, with the concept of personas we will be able to cluster learners into meaningful groups. Data-driven personas are created by considering the similarities and differences among users from the angle of their demographics (the initial conditions for our learning activities) and observed behavioral patterns during these activities as well as the digital, pedagogic, and linguistic affordances that accompany them. The differences among distinct personas must be based on essential issues, for instance what users do (actions or projected actions) and why they do it (goals and motivations), and not too much on who those users are (see also Calabria, Reference Calabria2004). Once the similarities and differences have been determined, affordances for learner interaction can be adapted in areas that are relevant to and appropriate for a particular digital tool and environment (Caws & Hamel, Reference Caws and Hamel2013, Reference Caws and Hamel2016; Hamel, Reference Hamel2013a, Reference Hamel2013b).
The goal of such analyses is twofold: enabling students to improve their learning outcomes and facilitating and improving the learning process. Since second language development in general proceeds on a nonlinear trajectory, it is unrealistic to predict learning outcomes and results a priori. Instead, after analyzing individual learner trajectories a posteriori – by paying close attention to initial conditions, self-similarity at different scales, and growth conditions of developmental change, for example – we identify the various factors that contributed to achange in learner behavior. These interaction data are equally pertinent to improving the artifact and the language learning process (Caws & Hamel, Reference Caws and Hamel2013; Hamel, Reference Hamel2012).
Such a longitudinal investigation of computer-mediated learning processes, analyzing and interpreting the change in learners’ activities and language learning outcomes, provides more accurate insight than the snapshots of pre-tests and post-tests. The theoretical perspective of complex systems and activity systems allows us to combine qualitative and quantitative research methods and to consider each changing variable in context. The observational multimodal and multilingual data from language learning processes and outcomes can be augmented with survey and interview data from students’ and instructors’ perception and retrospection.
The observation of learning processes – with the goal of facilitating adaptive instruction in CALL – can rely on digital logs (e.g. tracking of system logins, page views, keyboard input), eye tracking (e.g. attention focus and noticing), chat and interaction logs (e.g. written peer-to-peer chats in the foreign language, written interaction with the computer program), iterations of textual learning outcomes (e.g. discussion board or wiki submissions and text versions over time), and screen capture videos of entire learning activities, which are often supplemented by think-aloud protocols or stimulated recall. In CALL research and development, computer logs have been used for some time for data collection and analysis of learner–task–artifact interactions (Fischer, Reference Fischer2007). Logs have provided valuable insights, for instance, into learners’ navigation patterns (e.g. Desmarais et al., Reference Desmarais, Duquette, Renié and Laurier1997; Heift, Reference Heift2002) in tutorial CALL as well as their linguistic performance in computer-mediated communication. The collection of tracking data is well understood, yet we still have to learn more about the analysis and interpretation of such large longitudinal data sets (Chun, Reference Chun2013). The analysis of learner texts can use discourse-analytical methods. The analysis of learner corpora (i.e. principled, electronic collections of texts produced by language learners) is another methodological cornerstone. The detailed analysis of textual learning outcomes produced over a period of time provides a window into the underlying processes during that time. User walkthroughs (Hémard, Reference Hémard and Felix2003) allow a specific focus on the students’ use of a new CALL tool or system. Video-recording individual and small-group digital language learning activities provides insight into the ongoing use of digital artifacts. Advances in computer technology facilitate the observation and capture of user–task–tool interactions in a more natural and less intrusive manner. Tools such as Camtasia Studio (computer screen video capture) and Morae (live observation and video capture of computer screen and user, data management, and analysis) were employed in CALL studies (e.g. Caws, Reference Caws2013; Hamel, Reference Hamel2012, Reference Hamel2013a, Reference Hamel2013b; Hamel & Caws, Reference Hamel and Caws2010). These tools are specifically designed to measure the usability (Rubin & Chisnell, Reference Rubin and Chisnell2008) of computer applications and facilitate the analysis of the quality of the learner experience with CALL resources.
These data collection methods yield learner data of different types: products, processes, and perceptions. And the data need to be structured for analysis and triangulated to produce robust results that can be interpreted to inform adaptive instruction. Products stem from the outcomes of each learning activity. Product data are often in the form of multimodal learner texts. These texts are then structured in learner corpora (Granger, Reference Granger2003). Learner corpora provide a window into the underlying cognitive processes across groups and over time. Data from the interaction processes, in the form of tracking data, recordings, and observations, are structured in tables, graphs, and chronological logs. Student perceptions of satisfaction, subjective learning success, and learning preferences are elicited through surveys, interviews, think-aloud protocols, and stimulated recall. All data are best analyzed individually, comparatively, and cross-sectionally, resulting in a mixed-method study.
From a methodological point of view, we are not experimenting when observing language learning processes in computer-mediated education, we do not employ a reductionist approach by eliminating variables from rich educational context through research design or inferential statistics, and we do not conflate individual developmental learning trajectories into large leveled groups. Instead, through a complex systems lens, we focus on individual learners in their specific social, cultural, and educational contexts. The complex systems approach enables us to combine the advantages of qualitative research methods by paying attention to rich contexts and “thick description” (Geertz, Reference Geertz2017), with the advantages of their quantitative counterparts of generalizability and theoretical prediction through formal systematicity. Our focus in the investigation of computer-mediated learning processes is on the detection, analysis, and interpretation of change in the learner’s activity and outcomes, in line with the dynamically changing variables of the system. Therefore all such studies are longitudinal. The data are gathered in naturalistic, educational contexts and obtained as unobtrusively as possible. Observational data from learning processes that occurred over longer periods of time are analyzed in their respective educational contexts. These data are triangulated with data from learning outcomes and supplemented with qualitative data from students’ perception and retrospection.
Through an exploration of individual and collaborative language learning processes, the broad goals of such a comprehensive analysis of technology-mediated interactions are to:
define different learner types (personas) that are relevant to individualized learning and teaching approaches;
map educational contexts with their learning artifacts so as to determine their suitability for a given educational and digital context; and
offer a set of design criteria for learning objects, language learning tasks, and digital educational resources that reflect the needs and electronic literacy skills of individual learners and instructors.
In today’s language education settings, digital technologies, and especially internet-based resources and artifacts, are used as an integral component of many language learning environments. Their successful and effective integration requires careful analysis, design, implementation, and evaluation (Colpaert, Reference Colpaert2006).
Impact
Several practical questions arise when probing into adaptive instruction. Can we create a technology-mediated learning environment where students have access to digital resources that reliably facilitate their language development rather than impede their learning path? In what ways can we adapt instruction to individual learners to achieve this goal? How can we support learners so that they benefit most from adaptive instruction? Experienced teachers adapt their instruction to individual students. Yet a fully individualized approach is not possible with large groups of learners. What is the exact beneficial role of a digital technology in a specific context? How can the benefits of this approach with this technology be transferred to other contexts and groups? How can they be generalized?
CALL research with a specific focus on ways of adaptive instruction will contribute to a better understanding of technology-mediated language learning and language education in general by providing detailed, evidence-based suggestions and guidance on the implementation of language learning technology. This research also has applications that go beyond language learning contexts. It investigates new ways of learning by examining the roles that information and communication technologies play in second language learning and by focusing on the affordances of digital resources for adaptive instruction. More importantly, such research addresses the need for equitable access to information and communication technologies and ways to foster digital literacy.
Of course, learners can also negotiate adjustments and adapt (to the technology) autonomously, after critically reflecting on learning processes and developmental trajectories (Hampel & Hauck, Reference Hampel and Hauck2006; Hauck, Reference Hauck, Egbert and Petrie2005). However, we would submit that it is the teacher’s responsibility to guide and facilitate the students’ adaptation to their learning environment and to ensure that the most suitable, most adaptive technology is employed, so that the students’ focus can be on (language) learning and does not have to be on the struggle to adapt. It is the role of the researcher in CALL, and specifically in adaptive instruction, to provide teachers with a deeper understanding of the complexity of individual student trajectories and of the role of learning technologies in these complex contexts and to develop, sustain, and evaluate technologies that contribute to adaptive instruction.
