Autorin: Dr. phil. des. Johanna Oeschger
Gutachtende: Prof. Stefan Keller (PH FHNW), Prof. Dr. Miriam Locher (Universität Basel)
Projektdauer: 2018-2023

Dialog systems are a promising technology to maximize opportunities for learner interaction in foreign language (FL) classrooms. Research has provided initial evidence for the effectiveness of learner-system interaction to promote FL development and for the occurrence of learning-facilitating interactional features, but in-depth investigations into the specific nature of FL learner-system interaction, particularly task-based interaction, are still outstanding. The present study explores the potential of dialog systems for FL learning further by describing the nature of task-based interaction between English as a foreign language (EFL) learners and a written dialog system compared to written synchronized computer-mediated communication (SCMC) learner-peer interaction. The study was situated in the Swiss commercial vocational education and training (VET) context where dialog systems may offer a particular benefit due to limited classroom time and self-organized computer-assisted learning.

The sample was collected in three commercial VET EFL classes using a written dialog system developed for the study based on a curriculum task (system group) and a written chat application (SCMC group), yielding 18 learner-system interactions and 21 learner-peer written SCMC learner-peer interactions. The two sets of interaction scripts were coded for interactional features and analyzed for group differences with respect to the learners’ task performance in terms of complexity, accuracy, and communicative adequacy.

The findings revealed the occurrence of negotiation of meaning, recasts, and lexical and structural alignment in learner-system interaction at rates that were generally higher than in learner-peer SCMC and marked by characteristics unique to the learner-system interaction type. With regard to learners’ task performance, the system group compared to the SCMC group produced language that was significantly less complex and showed a non-significantly higher rate of inappropriate requests but was lexico-grammatically and orthographically non-significantly more accurate. Learners from the system group also had a significantly lower task completion rate, but their generic structure and style ratings did not differ from those of the SCMC group.

Overall, the study’s findings strengthen the case for the potential of dialog systems for certain aspects of FL learning, while also pointing to some important implications and limitations: Learner-system interaction may engage learners in frequent interactional features whose unique characteristics may be theoretically linked to FL learning functions and benefits as well as disadvantages. Compared to learner-peer SCMC, learner-system interaction may encourage learners to produce and practice language with higher accuracy, but it may offer reduced opportunities for syntactically and lexically complex as well as pragmatically fully appropriate language use. Finally, learners interacting with a dialog system may be able to engage with the sociolinguistic and pragmatic competences targeted by the task’s genre requirements in comparable ways as those interacting with a peer through SCMC; however, their ability to engage with the complete set of targeted competences might be reduced as they were more often unable to complete the task.