Innovation in Language Learning

Based on a recent language center and departmental implementation (see e.g. Workshop), this paper will give a brief overview over the current research status of automatic speech recognition (ASR), i.e. "to convert speech into a sequence of words by a computer program", from introduction of the "statistical paradigm" in the 70s (Huang & Deng (2010), 339, 351). We will then compare various current product implementations that are more or less readily available to the end user. We discuss their possible practical application in SLA programs for speaking practice. We will demonstrate how currently widely available automatic speech recognition technology for 5 to 7 (depending on definition of language variants) of the most popular languages in current SLA can be implemented in many SLA programs around the world in an economically (practically no extra cost other than for local installation; individual, short, flexibly scheduled speech assignments mean no need for funneling entire classes hogging the language center) and in a pedagogically viable way. We used Windows 7 Enterprise and up ASR (likely available to you already in your institutional language program/center) with Microsoft Language Packs (free) which gives you high-quality ASR for a subset of display languages. However, this still required not only carefully controlling the technical aspects that - even though low-difficulty mostly - make such projects often fail in language programs with their limited resources. We also carefully managed expectations by clarifying that we are not pretending to replace the teacher in an unrealistic way, but only blend AI with human intelligence, in a domain-specific way, in order to widen the "expert bottleneck" for the learner and relieve the teacher from routine tasks, like having to schedule meetings (even if online) and/or listen to the entirety of a student's oral production, while maintaining the capability of providing personalized feedback where needed. We also carefully designed the task: Teacher has to create speaking practice assignments based on syllabus, correlated with assigned textbook activities (e.g. practice reading not quietly, but aloud a target culture website the textbook sends you to for information extraction), and with corresponding upload assignments in the (optional) LMS (email could replace). Easy integration of speaking practice tasks into the syllabus is a major advantage over the usually intractable problems with language learner progression when trying to integrate language learning material providers’ ASR solutions if the course syllabus is not primarily based on those materials. While going through the individual steps of the speaking practice task, we will give samples of the training and documentation created to guide the users through this ASR task. We also included a number of example ASR task completions (as screencasts) by teachers and learners of different languages which demonstrate actual recognition quality (and some remaining issues). Variations of assignment will be discussed also (immediate student reaction to bad recognition by using of built-in speech correction; simpler GUI control through speech, based more restrictive vocabulary). We finally recommend integration of the task results into ePortfolios to show off language learner achievement.