Language learning through task-based interaction with dialogue systems
Considering the existing evidence of the effectiveness of synchronous computer-mediated communication on the development of L2 oral proficiency (Lin, 2015), combined with the scarcity of opportunities to interact with native speakers in most foreign language learning contexts, ICALL researchers have long been interested in developing systems that would allow a learner to have a meaning-focused interaction in L2 with a conversational agent (e.g. Underwood, 1982; Wachowicz & Scott, 1999; Fryer & Carpenter, 2006). During the last three decades, various dialogue systems have been developed as proofs of concept, using different natural language processing techniques to accept unconstrained learner production, either spoken (see Eskenazi, 2009 for an overview) or written (e.g. Price, Bunt, & McCalla, 1999; Wolska & Wilske, 2011). In the conceptualization of those conversational systems, a computer-guided approach of dialogue has proven both more effective and more feasible than the open-ended one commonly seen in chatbots (e.g. Petersen, 2010).
Our first aim is to develop a written dialogue system allowing learners of French as a foreign language to practice the target language through task-oriented interactions with a conversational agent. On the technological side, we hope to be able to implement the latest advancements in dialogue systems development, using a statistical approach based on a wizard-of-oz or crowdsourced interaction corpus (Orkin, 2013), eventually combined into a hybrid approach based on probabilistic rules (Lison, 2015).
Our following objectives relate to the measurement of the user acceptance (perceived ease-of-use and perceived usefulness) and effectiveness of the developed system for language learning. To date, very few studies have done experimental research on the impact of dialogue systems on L2 acquisition. Notable exceptions are e.g. Vlugter et al. (2009), Petersen (2010) and Wilske (2014), who focused on the effectiveness of the dialogue system on the acquisition of specific structures (respectively in Maori, English and German), with a particular interest in feedback provision by the system. Our intention, by comparison, is to assess the capability of such a dialogue system to promote learning by allowing the learner to practice the target language in an interactional and spontaneous setting. We will evaluate the system’s effectiveness on the development of L2 oral proficiency, especially on fluency, with an experimental research, by comparing it to a prescripted (“canned”) dialogue system through utterance selection (without unconstrained input) and with an open-ended French-speaking chatbot.