Open generative AI changes a lot, but not everything

Technology is again a timely topic at this moment in our profession. The appearance of generative artificial intelligence (AI) has brought an unprecedented level of public awareness and appreciation of language—both English and other world languages. The November 2022 appearance of generative AI embodied in ChatGPT ignited a combination of curiosity, imagination, and trepidation among the public. The Economist named "ChatGPT" the word of the year for 2023, with the explanation that "nothing can stop technology from dominating this year's words" (The Johnson Column, 2023, para. 7). ChatGPT captivated the attention of the public with its language performance: "The breakthrough in particular of large language models (LLMs) has been stunning. They produce prose so human-like that they have ignited a debate about whether LLMs are actually thinking (and whether students will ever do homework without them again)" (The Johnson Column, 2023, para. 7). Similarly, The New Yorker's Sue Halpern summed up 2023 with a column entitled "The year A.I. ate the internet: Call 2023 the year many of us learned to communicate, create, cheat, and collaborate with robots" (Halpern, 2023). These are but two of the many examples of how the public media in the United States has weighed in on the publicly accessible generative AI unleashed over the previous months. need to articulate and communicate the value of language study in a social context, identify what technology offers that is positive for language education, rethink how we organize our teaching in light of technology's affordances, and be clear about what technology cannot do. (Kern, 2024, this issue, p. XX) He opens the conversation with a discussion of the new technological capabilities offered by generative AI—specifically, machine translation and ChatGPT—and with concrete suggestions for prompting students' "critical engagement with technology guided by human teachers" (p. XX). This inflection point suggests that it is time to grapple again with our understanding of the significance of language, language teaching, and our roles as language professionals. After all, language is at center stage. Google has been responding admirably to our queries for information for years with lists of sources, links, and more precise questions for us to explore. We already had access to copious information. ChatGPT captivated the world with its language: the grammatically idiomatic, responsively contingent turn taking in natural written conversation. It soothed our sensibilities that have been irritated for years by repetitive, irrelevant responses from robotic chatbots positioned between us and the personalized information we sought from websites built to mitigate the need for human communication. This is the time for us as language professionals to increase our own critical engagement with language and language technologies in a world never so fascinated by language. As a founding faculty member of the doctoral program in Applied Linguistics and Technology at Iowa State University, I am accustomed to watching the evolution of technology as it changes the ecology of language teaching, learning, and assessment. Over the past decades, technology has delivered a continuing stream of important issues for us to study and countless avenues for expanding learners' access to language and culture learning while developing agency as language users. The constantly evolving technologies of the past should have prepared us well for the entrance of ChatGPT, but it is necessary to stop and consider what we have learned from the past. Taking Kern's (2024, this issue) prompt to consider our preparation for this moment, I recall three episodes where I find clues for understanding the opportunity presented by the technical accomplishment manifest as generative AI. The first episode is the 50-plus years of navigating technology in language learning through research and practice. The Modern Language Journal has documented a fitting sample of this history, which is more fully displayed in journals focusing on technology and language learning, most notably CALICO Journal, CALL Journal, Language Learning Heift grammar checkers left errors unchecked or unresolved. They did not demand language teachers to fundamentally change how they teach. But they did contribute to the professional zeitgeist animating the study of the computer-assisted language learning of the time. This era left me with relevant takeaways including that (a) descriptive research is essential to document the contingencies in computer-mediated interactions, (b) students need guidance to learn effective strategies for using technology for language learning despite their ascribed identities as digital natives, (c) research can show benefits of technology-mediated pedagogies when they are carefully designed, appropriate to the needs of learners, engaged with over a sufficient span of time, and investigated with an appropriate methodology, and (d) perseverance is required to appreciate the unique contributions of each generation of technologies in language learning. During this period, the NLP capabilities in language learning tools improved incrementally, but the appearance of ChatGPT in November 2022 marked the beginning of a new era. A second episode that today's inflection point brings to mind was connected to the events of 9/11 in 2001. The Modern Language Association (MLA) Ad Hoc Committee on Foreign Languages referred to a "sense of crisis" at this time about the nation's "language deficit." In the 2007 MLA report, Foreign Languages and Higher Education: New Structures for a Changed World, the ad hoc committee described the crisis in much different terms than those Kern (2024, this issue) used to characterize the circumstances giving rise to today's inflection point. The authors wrote, "In the context of globalization and in the post-9/11 environment, the usefulness of studying languages other than English is no longer contested" (Modern Language Association of America, 2007, para. 4). The challenge was seen as the need to stitch together the instrumental with the humanistic goals of world language teaching to build students' "translingual and transcultural competence" (Modern Language Association of America, 2007, para. 4). The proposed solution was for language courses to "incorporate cultural inquiry at all levels" (Modern Language Association of America, 2007, para. 11) and the study of more subject areas in advanced courses. Cultural inquiry in language study would require that courses "situate language study in cultural, historical, geographic, and cross-cultural frames within the context of humanistic learning" (Modern Language Association of America, 2007, para. 11), which helps students to grasp their own subjectivity. Any movement toward curricula and materials targeting these goals has undoubtedly been useful for addressing today's challenges, which Kern (2024, this issue) sees as an opportunity for promoting critical engagement with technology, language, content, and culture. However, barriers to promoting cultural inquiry informed by cultural, historical, geographic, and cross-cultural frames may remain and perhaps have grown even stronger today. Kramsch (2012) identified a range of challenges including the following: a division of labor exists in language departments, where faculty teach culture only in the upper levels; some faculty do not support the goal of helping students understand their American identity; some language teachers are unprepared to discuss historical events; students' lack of knowledge of history in general provides an insufficient basis for learning cultural histories; and textbooks do not support the goals. These issues remain significant for the field even as Kern sees the need to use AI as a means for supporting learners' cultural inquiry at all levels. Despite the scholarship in world language teaching that has been supportive of efforts to promote critical cultural inquiry across all levels of language instruction, Kern (2024, this issue) points to some countervailing indicators. First, folk wisdom about language promotes claims that language instruction can be adequately obtained by technology-delivered drills and tutorials and that language performance does not require language knowledge in view of intelligent language generation and translation tools. Second, enrollments in language courses other than Korean and Hawaiian are decreasing—a trend highlighted by the closure of one university language department in the United States. Third, the view of some in higher education is that language teaching is too basic for research universities. The decreasing enrollments in world language courses may be part of the larger observed downward trends across humanities courses, possibly unrelated to generative AI. What Kern (2024, this issue) refers to as folk wisdom about language learning might be seen as a consequential component of "folk linguistics," the knowledge and beliefs about language held by people who are not linguists. Folk linguistic beliefs affect what people think and say about language as well as their actions pertaining to language (Niedzielski & Preston, 2000). Folk linguistic beliefs about language learning are presumably shaped by students' experience in language classes, but the large majority of language study in US higher education is in first- and second-year classes. These language classes provide the opportunity to demonstrate to students the breadth and depth of knowledge and understanding conveyed through language study, but in view of the barriers to teaching cultural inquiry noted above, it is not clear how well the opportunity is exploited. This is the issue raised in the 2007 MLA report. It is the challenge explored in the 2012 special issue of L2 Journal on history and memory in foreign language study (Kramsch, 2012). It is the goal targeted by many scholars in applied linguistics over the past 20 years. This challenge remains a high-priority topic for understanding how generative AI can improve language learners' experience with critical cultural inquiry beginning in their first language course. The near future will undoubtedly reveal fascinating new ideas for increasing beginning-level learners' engagement in critical cultural inquiry while conveying a more sophisticated understanding of language learning than what is currently expressed by current folk wisdom. In other words, the current inflection point brings tools that could help to address the previous crisis. Generative AI supplies beginning language students with an unprecedented degree of power over their target language by supplying them with tools for multimodal translanguaging through translation, text to speech synthesis, synchronous interaction, composition of text, and multimodal forms of cultural expression. These tools can unlock barriers to language understanding and interaction that frustrate beginners. They can empower learners with language for expressing their own meaning and creativity. They can produce artifacts for critical analysis that requires gaining knowledge of historical frames of reference. In other words, AI tools provide concrete, accessible devices that can contribute to solutions to the crisis, but they may also remind us that the crisis was and is bigger than the most powerful AI tools. The crisis identified in 2007 was created from institutional structures, knowledge deficits, and pedagogical beliefs about the methods and goals of language instruction. Recalling the inflection point marked by recognition of the need for critical cultural inquiry in the first years of language study, I expect that the opportunities made feasible by generative AI will contribute to defining a 2020s version of the aughts crisis in world language teaching. The third episode related to today's inflection point took place as a hybrid event originating at Iowa State University in October 2023. Our annual Technology for Second Language Learning Conference (TSLL) provided a glimpse of the types of questions and issues that may be on the agenda for applied linguistics going forward. We invited abstracts describing research investigating the potentials, uses, and implications of AI technologies for language teaching, learning, assessment, and research. The conference theme was "Advancing Technologies—Expanding Research," signifying the multiple strands of research and practice in applied linguistics affected by current and future AI technologies. Applied linguists throughout the world responded with a range of fascinating studies that offered a glimpse into potential directions. Distinct categories were difficult to discern in the array of research directions represented by the presentations at the conference, but they generally fell into four groups. One group followed a tradition of descriptive research on technology for language learning with investigations of learners' generative AI use primarily, but not only, for their writing. A second group continued the strand of research on technology for language assessment with studies of automated evaluation of learners' writing and speaking and on language assessment literacy, but also included investigations of AI-generated language test items and test preparation. A third group explored a range of generative AI capabilities for applied linguistics, including automated assessment of accuracy for L2 writing research, detecting pragmatic competence for L2 Chinese teaching, promoting learner–computer interaction with question–answer technologies, and a corpus-based study of the discourse styles of ChatGPT. A fourth group investigated implications of generative AI for language teacher education to begin to understand the new knowledge that language teachers need to develop and the methods for teacher education. The conference program, which includes the abstracts and videos of some keynotes, is at https://apling.engl.iastate.edu/conferences/technology-for-second-language-learning-conference/tsll-2023/ Plenary speakers demonstrated the diverse experience and perspectives that will inform the study of AI in language learning in the future. Mike Sharples presented a vision of a new science of learning with AI, Abram Anders gave a glimpse into the practices in writing instruction that will be part of the new science, and Mark Warschauer provided examples of current research on generative AI for literacy development. Xiaoming Xi and Evgeny Chukharev each described how generative AI would be leveraged in architectures engineered for assessment and learning within an explicit pedagogical design. Michael Thomas demonstrated that AI itself should be the object of critical examination in view of the global ecology in which generative AI is constructed and used. Peter Crosthwaite highlighted the significance of this moment of public awakening to the view that language is data. The diverse expertise shared by these speakers masterfully ushered in this new generation of technology for language learning. This brief summary cannot do justice to the range and richness of the presentations and conversations generated among participants. Overall, the conference papers revealed that generative AI had not changed everything, even though it was clear that much remains to be learned, said, and done as we continue to grapple with the implications and opportunities presented by generative AI for applied linguistics. But given the novelty of the entangled issues and their immediate relevance, where can one begin? The predecessors of today's AI researchers examined laboratory accomplishments in language recognition and generation for what they might reveal about the cognitive linguistic, psychological, and philosophical dimensions of language and thought (e.g., Haugeland, 1981). Now that AI is out of the laboratory, we need to understand what open generative AI can reveal about the human and social dimensions of language use in addition to their significance for language teaching and learning. As Kern (2024, this issue) puts it, "language educators have a responsibility to think through the linguistic, social, and ethical issues related to AI—and other forms of technological mediation—with their students" (p. XX). Three directions in addition to the paths charted by previous research seem worthwhile. First, applied linguists are well positioned to study the language of generative AI following in the path of discourse analysts who study language on the Internet (e.g., Bou-Franch & Blitvich, 2019). What are the linguistic patterns the AI was trained to produce to make the Internet seem "nearly animate" (Halpern, 2023, para. 2) and "responsive and improvisational" (Halpern, 2023, para. 3)? If language learners are expected to learn from the language of generative AI, its similarity to human language for specific purposes needs to be assessed. Second, the language knowledge, performance, and training of large language models opens a workshop for the study of constructs and questions in applied linguistics. For example, what does the architecture and training of large language models reveal about the need for universal grammar in human language acquisition (e.g., O'Grady & Lee, 2023)? Such a workshop might also be suited to consider, for example, how metalinguistic knowledge may or may not interface with a model responsible for language performance, and how particular characteristics of prompts used in language assessment influence the AI's language performance. Such research questions can be pursued without adopting a strong version of AI (Searle, 1980) that treats an AI as a theory of human language knowledge and performance. Instead, applied linguists can benefit from the use of generative AI as a tool for gaining a more precise understanding of factors that may affect language performance of both machines and humans. Third, sociolinguists investigate language use on the Internet because "the largest social space on earth these days is the virtual space" (Blommaert & De Fina, 2017, p. 13). Interlocutors on the network create multimodal contexts in time and space based on their own knowledge and interests. This social space is a rich source of data for both sociolinguists and language learners. But now the question is how the social order has been disrupted by participation of an AI that feigns humanity while telling lies using words and ideas stolen from others in the social space. How can applied linguists' research on the pragmatics of interactional language be used to understand the new conventions of communication engendered by the new participant? As we wrestle with the concrete implications of our ecological, sociocultural, social–semiotic, social–material, and other perspectives on language teaching and learning, we are not alone in recognizing our discomfort with technologies that not only pass the Turing test but also attempt to make friends with our students (Brooks, 2023), kindly building their trust while pretending to help them using stolen goods. In the introduction to The Cambridge Handbook of Responsible Artificial Intelligence: Interdisciplinary Perspectives, the editors present the issue facing all facets of social and intellectual life: "The enormous potential for innovation and technological advances and the chances that AI systems provide come with hazards and risks that are not yet fully explored, let alone fully understood" (Voeneky et al., 2022, p. 1). A familiar refrain about technology being ahead of regulation was recited again and again during 2023, but at last in the final days, it was reported that the New York Times was suing ChatGPT creator OpenAI and Microsoft for copyright infringement (Allyn, 2023). Despite the year-long strand of critical commentary about the dubious practices of ChatGPT, the public charges resulting from New York Times research punctuated the year in a comforting way. As I was completing my comment for the MLJ Perspectives column in the last days of 2023, I was listening to British Broadcasting Corporation (BBC) newscaster Julian Marshall interviewing a professor who was explaining how he had trained his AI to sound like the newscaster based on the plentiful accessible archived recordings of Marshall's voice. Interviewer and interviewee listened with us to the fake Julian Marshall, which sounded like a good, but not perfect, replica. The two men agreed that the AI replica was good, but they thought it gave Marshall a bit of an American accent. They discussed what other capabilities the AI would need to actually replace Marshall in his long-held job of news anchor, ending the year with the idea that Marshall's job was safe for the time being. Marshall concluded that his contribution to the larger enterprise extended far beyond what ChatGPT and voice synthesis could do. Can language teachers be counted among the professions that contribute more to education than what ChatGPT and voice synthesis could do? Professor Sharples addressed this question at the 2023 TSLL conference by predicting that professions such as nursing and teaching that are built upon the human capacities of caring, kindness, and trust cannot be replaced by AI. That is an encouraging starting point, but in our case, our kindness, care, and honesty require that we caution students not to let the AI befriend them, not to believe its lies, and not to take its stolen language. We will need to accompany these warnings with concrete strategies for using AI as a tool to help learn about language and culture. Like previous generations of technology, generative AI will require us to learn from descriptive research investigating how it can contribute to language learning goals and to provide students with guidance about how to use it to build their own language competence without letting it steal their opportunity to learn from their language class. I have confidence that we will be able to work together and with our colleagues across disciplines to negotiate the opportunities and risks presented by generative AI. Our profession has laid the groundwork to do so through years of professional engagement with technology for language learning as well as our recognition of the need for teaching critical cultural awareness across all levels of language study. Open access funding provided by the Iowa State University Library.