A factor influencing CF effectiveness is CF type. In this study we decided to analyse one type in depth, instead of comparing CF types. Clearly, however, we can expect that a different CF type, or even different operationalizations of CF might have yielded different results. Fortunately, the CF type is concretely defined in the CALL system, which allows for cross-study comparison. In our setup, comparison of CF types is an attractive future possibility, or allowing learners to toggle CF on or off, or selecting the type of CF, which in turn could inform us on learner preference (cf. Hegelheimer & Tower 2004). With respect to the type of CF that we provided, we suggest a possible improvement, which is the possibility of repeating the correct utterance. Currently, learners may struggle to provide a correct answer. If they really do not know (guess), they may be surprised when they finally answer correctly. The system does not allow learners to retry a correct answer (in the CF condition). As a result, learners may sometimes miss the opportunity to understand the correct answer. From a procedural and noticing point of view, it may be beneficial for learners to be able to repeat a correct answer – in a way for them to make sure they understand.
8. Future research
As we have seen, a CALL system with ASR such as GREET is effective for research on oral CF, and for improving L2 proficiency. In our view, these systems hold great promise for future research.
In SLA it is becoming increasingly clear that the effect of CF is mediated and moderated by many variables, which requires a new approach to CF research. In our study, we have investigated one target feature and investigated in detail how learner variables interact with CF on learning that feature. In a similar setup with GREET, it is also possible to make comparisons of learning with different types of CF (e.g. recasts and metalinguistic CF), or with different linguistic targets. Since the system is otherwise consistent, the influence of CF type could be methodically investigated. Moreover, if the CALL system can register the learners affective- or motivational state, this can be related to CF (type) effectiveness, and inform learner behaviour (Bodnar et al. 2014; see for an example Bodnar, Cucchiarini, Penning de Vries, Strik & van Hout (forthcoming)).
By adding more instructional content to the system (or such a system), longitudinal studies in CALL are possible (see Heift 2010 for an example). Longitudinal studies allow for other related research questions to be examined: the
within-subjects interactions of linguistic targets and CF types, and how these change over time - and how this differs between subjects. Clearly, then, this would provide opportunities to test a system that automatically adapts to the learner by analysing learner behaviour. Adapting the difficulty level to the level of education and level of proficiency of participants offers new opportunities to research CF types
(effectiveness for learning new knowledge, or strengthening learned knowledge), as we mentioned in Chapter 4. In addition, this would allow investigation into the notion of the zone of proximal development ('ZPD', Vygotsky 1978). In our study, we have two measures that were seen to change over time and relate to L2 learning (ratio of correct answers, and attempts per question) These measures provided information about the relative difficulty of the task for the learner. A system could adapt the difficulty level of practice to keep learners in the optimal learning zone: for instance that, for a given time frame, they need on average 1.5 attempts to answer, or that 75% of their answers are correct. Setting these levels could inform us on the nature of the ZPD, and make treatment maximally efficient (and perhaps also maximally engaging). CF types could be compared in their efficiency in helping learners move to the next difficulty level. In addition, it will be interesting to research the relation of CF preference, CF perceived effectiveness and CF
effectiveness. Through close monitoring of learner behaviour and in-task reports of motivation, or through retrospective reports, the engagement of the learner could be related to CF effect.
Of key importance for the effectiveness of a CALL system is the learner’s motivation, or the willingness of the learner to continue practicing. A promising new development in the field of CALL is the growing interest in serious games. As seen in the study by Cornillie (2014), practice with these systems can provide effective CF on a language learning experience where the learner is immersed in the game environment. A combination of serious games and spoken practice as employed in GREET would allow to study CF in a context that is more meaning-based and probably engaging.
Other interesting directions for future research and CALL development concern relating CALL practice more closely to the language learning goals of the L2 learner, which should contribute to increasing motivation (Colpaert 2010). This would of course require the use of appropriate tools for eliciting those goals, and the development of content to address these goals. This is an exciting new direction for CALL research, and GREET could be employed in such research (Bodnar et al. (forthcoming)).
A current bottleneck for longitudinal studies with CALL is the creation of appropriate L2 content. The internet contains vast resources, such as film clips (with varying levels of meta-information, such as transcriptions), which presents us with the possibility of automatic generation of materials (cf. Presson, Davy &
MacWhinney 2013). It is clear that this holds great promise for expanding the possibilities for pedagogical CALL applications, and interrelated, for L2 research with CALL. In addition, experiments through the internet, such as sometimes already run through crowd sourcing (Rayner & Tsourakis, 2013), can boost the number of participants: more learners can be approached (even internationally) and
learners may be more inclined to participate, as they can simply work from home. For experimental control, this poses problems (e.g. learners may be discussing tasks with others, or not be focused on the task with a television in the background), and these need to be addressed in the experimental setup. The analysis of learner behaviour may be of use here, as the program can provide feedback to the participant (‘hey, make sure you answer your questions in time!’) or notify the researcher of irregular behaviour.
With respect to logging behaviour, the GREET system could be equipped with an eye tracking system (see for an example of a study using eye tracking, Smith 2012). These systems are becoming increasingly available, and could easily be attached to an experiment computer. For our experiment, this may add an extra layer of learner information: by studying eye movement, we could infer how they read the messages and the blocks on the screen, which could be analysed for signs of noticing the CF. Also for the proficiency tests this adds valuable extra information: especially for the GJT this may prove valuable (Godfroid, Loewen, Jung, Park, Gass & R. Ellis 2014).
The findings of our experiments were obtained in a highly controlled environment. In a classroom, such controlled settings are not possible, and CF will be provided and received in a much more diffuse manner (cf. Lyster et al. 2013). Our findings may therefore not transfer directly to the classroom. On the other hand, we have to consider that CALL and blended learning are becoming increasingly important in L2 pedagogy (Chapelle 2007, 2009b; Warschauer 2007) and the use of more advanced technologies will probably increase their functionalities and
applicability. It is therefore necessary to investigate L2 learning in this new context. With the advance of technological possibilities, however, it is necessary to
emphasise that the effectiveness of these technologies for L2 learning are dependent on the design in which it is implemented, and that evaluation and development of language learning systems is a crucial step in improving pedagogical design (cf. 'Educational Engineering', Colpaert 2014; Chapelle 2001).
Our experiment with the GREET system was performed mostly outside of the class. However, it is easy to see that practice with this system could make it into a language course. Particularly for speaking practice, which requires much repetition and CF on features such as pronunciation, or grammatical features such as articles, V2, or morphology, tutorial CALL systems such as ours can be very valuable for L2 learning. Ideally, in the future, systems such as GREET could be part of L2 learning courses for speaking practice, for learners of various educational backgrounds. The practice data of the learners could be analysed to inform SLA, and the design of the CALL system. In that manner, CALL systems will not only generate new findings for SLA, but also provide data that can be analysed to improve their effectiveness.