1. Introduction
Decades before educators were forced to confront the disruption posed by widely accessible generative artificial intelligence (AI) tools such as ChatGPT, language learners, instructors, and researchers began dealing with its game-changing predecessor: machine translation (MT). Researchers began assessing MT systems and proposing language teaching applications for them as soon as universities and schools gained access to them in the mid-1980s (Footnote *Anderson, 1995*; Ball, 1989*; Corness, Reference Corness1985; French Reference French, Brierley and Kemble1991; Lewis, Reference Lewis1997; Richmond, 1994*). These inquiries accelerated in the early 2000s, when internet-enabled computer labs and increasingly smarter devices put free online MT services such as Babel Fish and Google Translate (GT) at students' fingertips, triggering concerns over output quality, academic dishonesty, and the short-circuiting of actual learning. In recent years, there has been a veritable explosion of research on MT's role in and impact on language teaching and learning, with many dozens of peer-reviewed articles published in the past five years alone, as documented in a handful of comprehensive literatures reviews (Gokgoz-Kurt, Reference Gokgoz-Kurt, Qin and Stapleton2023; Jiang et al., Reference Jiang, Yu and Zhao2024; Jolley & Maimone, Reference Jolley and Maimone2022; Klimova et al., Reference Klimova, Pikhart, Benites, Lehr and Sanchez-Stockhammer2023; Lee, Reference Lee2023). The present article provides a timeline of this rapidly expanding research domain.
Hutchins and Somers (Reference Hutchins and Somers1992) define MT as “computerised systems responsible for the production of translations from one natural language into another, with or without human assistance” (p. 3). A detailed account of the history and evolution of MT technology, capabilities, and applications is beyond the scope of this timeline and available elsewhere (Garg & Agarwal, Reference Garg and Agarwal2019; Hutchins, Reference Hutchins2010; Németh, Reference Németh2019; Pestov, Reference Pestov2018). However, given the impact of MT's evolving functionality on research in the domain in question, familiarization with its major developmental milestones is important. Hutchins (Reference Hutchins2010) dates MT's origins to the late 1940s, coinciding with the invention of computers. During the 1950s and 1960s, the Cold War tech race and advances in computational linguistics fueled developments. By the 1980s, early commercial systems, such as SYSTRAN, Logos, and METAL, previously available only on mainframes in government or military facilities, could be installed on home computers and desktop workstations in schools and universities. SYSTRAN's Babel Fish, generally considered the first free online MT tool, was launched as a website in 1997, nine years before GT debuted in 2006. In the years since, advances in smartphone technology and network coverage have connected billions of people to MT and generative AI apps that translate just as well.
MT capabilities and quality have increased in step with these advances in accessibility. The earliest MT systems relied on technology known as rule-based MT, whereby automated transfer operations are executed using one-to-one lexical substitutions and preprogrammed morphosyntactic rules. By the 1990s, this rudimentary approach had largely been replaced by more sophisticated databank- and corpus-based approaches that target phrase-level equivalents, such as example-based MT, which identifies analogous phrases from aligned parallel text databanks, and statistical MT, which analyzes patterns in bilingual corpora and calculates probabilities to determine optimal word combinations and sequences. The most recent major advance in MT technology happened in 2016, when GT switched its underlying architecture to a neural machine translation (NMT) model. A powerful form of machine learning, NMT uses artificial neural networks to analyze large datasets, training itself to predict the most likely sequences of words in sentences. NMT has been shown to be substantially more accurate than previous systems. For example, Wu et al. (Reference Wu, Schuster, Chen, Le, Norouzi, Macherey, Krikun, Cao, Gao, Macherey, Klingner, Shah, Johnson, Liu, Kaiser, Gouws, Kato, Kudo, Kazawa and Dean2016) found that NMT-powered GT (GNMT) commits 60% fewer errors than the previous phrase-based model and approaches the accuracy levels of experienced human translators. Recent assessments of the translation capabilities of large language models (LLMs) have demonstrated that they are also highly accurate (Moslem et al., Reference Moslem, Haque, Kelleher and Way2023), with some suggesting that LLMs are the new paradigm for MT development (Xu et al., Reference Xu, Kim, Sharaf and Awadalla2024).
As the most recent stages outlined in this MT history overview unfolded, four clearly discernible strands or subdomains of the research into MT and language education have emerged. Research in this field rests on foundational articles that describe the history and capabilities of the first MT systems available in instructional settings, often suggesting potential applications in translator training and language programs (Anderson, 1995*; Ball, 1989*; Corness, Reference Corness1985; French, Reference French, Brierley and Kemble1991; Hutchins & Somers, Reference Hutchins and Somers1992; Lewis, Reference Lewis1997; Richmond, 1994*; Somers, Reference Somers, Forcada and Pérez-Ortiz2001). A second clearly identifiable strand involves a cluster of survey-based studies designed to gauge how often, in what ways, and for what reasons students use MT in their language learning activities, as well as the perceptions, beliefs, and attitudes learners and instructors hold regarding various aspects of MT, including specific kinds of uses (Case, 2015*; Clifford et al., 2013*; Farzi, 2016*; Hellmich & Vinall, 2023*; Jolley & Maimone, 2015*; Knowles, 2016*; Larson-Guenette, 2013*; Niño, 2009*; *O'Neill, 2019a*; and White & Heidrich, 2013*, among others). This research relates to a third strand, which centers around questions of academic dishonesty. Many publications in this strand argue that unauthorized MT use is, in fact, cheating and should be discouraged through detection and prevention strategies (Correa, Reference Correa2011, 2014*; Harris, Reference Harris2010; Luton, Reference Luton2003; McCarthy, Reference McCarthy2004; Steding, 2009*). As a corollary to this line of thought, a handful of empirical studies have sought to determine whether instructors are able to reliably detect MT use and to identify the telltale signs of MT as compared with unassisted translation or direct second language (L2) writing (Innes, 2019*; Maimone & Jolley, 2023*; O'Neill, 2012*; Somers et al., 2006*; Stapleton & Leung, 2019*). Finally, the bulk of publications in this research field may be categorized as pertaining to a strand focused on the uses of and implications for MT in formal language learning contexts. These articles discuss ways in which MT may be or has been used to support language learning, as well as its impact on performance (particularly on L2 writing output), and often discuss pedagogical implications (Ducar & Schocket, 2018*; Garcia & Pena, 2011*; Fredholm 2015*, 2019*; Lee, 2020*; Niño, 2004*, 2008*, O'Neill, 2012*, 2019b*; Vold, 2018*; and Williams, 2006*, among others), including the importance of MT literacy (Bowker, Reference Bowker2020; Loock et al., 2022*; Pellet & Meyers, 2022*). Individual learner differences, such as proficiency level (Chung, 2020*; Chung & Ahn, 2021*; Lee, 2022*; Mujtaba et al., 2022*; Shin & Chon, 2023*), learner strategies (Lee, 2020*; Ryu et al., 2022*; White & Heidrich, 2013*), and motivation (Tsai & Liao, Reference Tsai and Liao2021), as well as the role of MT and translation more broadly in supporting translanguaging approaches (Beiler & Dewilde, Reference Beiler and Dewilde2020; Hell et al., 2022*; Heugh et al., Reference Heugh, French, Arya, Pham, Tudini, Billinghurst, Tippett, Chang, Nichols and Viljoen2022; Jiang et al., Reference Jiang, Yu and Zhao2024; Kelly & Hou, 2021*; Rowe, Reference Rowe2022; Zhou et al., Reference Zhou, Zhao and Groves2022), are also frequent themes in this strand. Overall, the studies in this timeline suggest many practical applications for MT in language teaching, even if findings have been somewhat contradictory (Lee, 2022*), and more studies investigating whether MT use actually supports language development or durable proficiency gains are needed (Jolley & Maimone, Reference Jolley and Maimone2022).
The present timeline includes 63 publications. One challenge in providing a representative sampling of notable articles from each of the subdomains summarized above is that many empirical studies in this field are relatively recent. As a whole, this domain has seen a gradual progression from articles grounded in personal perspectives or anecdotal observations to small-scale exploratory experiments to more rigorous empirical studies. In selecting articles for inclusion, we prioritized empirical studies published after the advent of GT. However, we also opted to include a handful of earlier and non-empirical articles that have been especially influential or that propose systematically designed pedagogical uses for MT. We elected to consider influential doctoral dissertations but to exclude master's theses. To maintain a focus on research with an explicit emphasis on the intersection of MT and language teaching and learning, we disregarded studies more narrowly focused on translator training. We also excluded non-English sources. The publications included in this timeline are categorized according to the following themes:
-
A. Use of and perceptions about MT and language learning
1. Learner use: frequency, types, and reasons
2. Learner perceptions of MT: accuracy, usefulness, appropriateness, etc.
3. Instructor use: frequency, types, and reasons
4. Instructor perceptions of MT: accuracy, usefulness, appropriateness, etc.
-
B. MT and academic dishonesty
1. Belief that unauthorized use of MT is a form of cheating
2. Detection: emphasis on detection or instructor ability to detect MT use
3. Signs: textual features characteristic of MT use
4. Response to unauthorized use of MT (resist vs. accept, policies, etc.)
-
C. Instructional applications of MT to promote leaning
1. Proposal of or report on MT-enhanced learning activity
2. Use of pre-editing or post-editing activities to raise linguistic awareness
3. Proposal of or report on MT-enhanced pedagogical strategies or model
4. Focus on MT training or MT literacy
5. Other pedagogical implications or recommended bast practices
-
D. Impact of MT on L2 writing (products, process, overall quality, textual features, etc.)
-
E. Contributions of MT use to language development or proficiency gains
Jason Jolley is a professor of Spanish at Missouri State University. He holds a Ph.D. in Spanish from The Pennsylvania State University. His teaching and research interests include Latin American literature, translation and machine translation in language learning, motivation, and self-directed language learning.
Luciane Maimone is an associate professor of Hispanic Linguistics at Missouri State University. She holds a Ph.D. in Applied Linguistics from Georgetown University. Her research focuses on psycholinguistic aspects of non-native language acquisition, language pedagogy and assessment, and the educational affordances of online environments, including machine translation and telecollaboration.
