Review of Automated speaking assessment: Using language technologies to score spontaneous speech

In this dynamic technology world, where innovative ideas and algorithms evolve almost daily, the landscape of computer-based language testing over the years has been dramatically reformed (Suvorov & Hegelheimer, 2013). One unique challenge that remains, however, is to effectively assess speaking proficiency (Chapelle & Voss, 2016), and since reliability plays a fundamental role in test score interpretation, the degree of error in speaking scores should be taken into account by test developers (Chapelle, 2021, p. 72). In oral speaking assessment, human rating, as traditionally the only means to evaluate constructed responses, can pose a threat to validity because of factors such as rater fatigue and rater bias (Zechner & Evanini, 2020, p. 3). A possible approach to addressing these potential threats is automated speech recognition (ASR). Recent developments such as the use of Deep Neural Network algorithms for training ASR and growth in the performance of scoring models have considerably advanced automated speech processing technology. Yet, there is still a lot to be discovered because of the lack of knowledge about how these systems work. Responding to this need, Automated Speaking Assessment: Using Language Technologies to Score Spontaneous Speech, edited by Zechner and Evanini, aims to bridge the gap between stakeholders such as scientists and engineers who may not have any expertise in reliability, validity, and fairness issues and stakeholders in academia who may not have expertise in speech processing technology. In doing so, the book unveils the state-of-the-art automated speech scoring technology, SpeechRater, used at Educational Testing Service (ETS). This book aims to be easily accessible to graduate students and applied linguists as well as testing professionals and measurement specialists.