Revija Slovenskega društva učiteljev tujega strokovnega jezika Journal of the Slovene Association of LSP Teachers ISSN: 1854-2042 2024 Letnik XIX, številka 1 Volume XIX, Number 1 Scripta Manent Revija Slovenskega društva učiteljev tujega strokovnega jezika Journal of the Slovene Association of LSP Teachers Letnik XIX, številka 1/Volume XIX, Number 1 Založnik/Published by: Založba Univerze v Ljubljani/University of Ljubljana Press Slovensko društvo učiteljev tujega strokovnega jezika/The Slovene Association of LSP Teachers Za založnika/For the Publisher: Gregor Majdič, rektor Univerze v Ljubljani / the Rector of the University of Ljubljana & Mateja Dostal, predsednica Slovenskega društva učiteljev tujega strokovnega jezika Izdajatelj/Issued by: Znanstvena založba Filozofske fakultete UL/Ljubljana University Press, Faculty of Arts Za izdajatelja/For the Issuer: Mojca Schlamberger Brezar, dekanja Filozofske fakultete Univerze v Ljubljani/Dean of the Faculty of Arts, University of Ljubljana, Slovenia Glavna in odgovorna urednica/Editor in Chief: Violeta Jurković, Fakulteta za pomorstvo in promet, Univerza v Ljubljani/Faculty of Maritime Studies and Transport, University of Ljubljana Urednica/Editor: Sara Orthaber, Filozofska fakulteta, Univerza v Mariboru/Faculty of Arts, University of Maribor Gostujoča urednica/Guest Editor: Melita Koletnik, Filozofska fakulteta, Univerza v Mariboru/Faculty of Arts, University of Maribor Uredniški odbor/Advisory Editorial Board: Simon Borg, Western Norway University of Applied Sciences; Vesna Cigan, University of Zagreb, Croatia; Alejandro Blas Curado Fuentes, University of Extremadura, Spain; Danijela Djorović, Universityof Belgrade, Serbia; Jan Engberg, University of Aarhus, Denmark; Peter Franklin, Konstanz University of Applied Sciences, Germany; Pedro A. Fuertes Olivera, University of Valladolid, Spain; Nataša Gajšt, University of Maribor, Slovenia; Julio C. Gimenez, University of Westminster, United Kingdom; Ken Hyland, University of East Anglia, United Kingdom; Vita Kilar, University of Ljubljana, Slovenia; Melita Koletnik, University of Maribor, Slovenia; Jaroslaw Krajka, Maria Curie-Skłodowska University, Poland; Sara Laviosa, University of Bari “Aldo Moro”, Italy; Rachel Lindner, TU Dortmund University, Germany; Saša Podgoršek, University of Ljubljana, Slovenia; Mary E. Risner, University of Florida, USA; Michele Sala, University of Bergamo, Italy Naslov uredništva/Editorial Office Address: Scripta Manent Društvo učiteljev tujega strokovnega jezika (Slovene Association of LSP Teachers) Aškerčeva 2 SI-1000 Ljubljana Slovenia Publication is free of charge. ISSN: 1854-2042 Elektronska revija/Online https://journals.uni-lj.si/scriptamanent This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License (except photographs). / To delo je ponujeno pod licenco Creative Commons Priznanje avtorstva-Deljenje pod enakimi pogoji 4.0 Mednarodna licenca (izjema so fotografije). VSEBINA/CONTENTS EDITORIAL Sara Orthaber, Melita Koletnik 1 Izvirni znanstveni članki/Research Papers FACTORS DETERMINING THE EFFICACY OF AI-GENERATED WORD PROBLEMS FOR CONTENT-SPECIFIC MATH LANGUAGE COURSES IN HIGHER EDUCATION Karen Fleischhauer, Kate Friedrich 4 POSREDOVANJE, PODPRTO Z ORODJI UMETNE INTELIGENCE, PRI POUČEVANJU TUJEGA JEZIKA STROKE – PRIMERA POUČEVANJA ANGLEŠČINE ZA PRIHODNJE ZDRAVSTVENE DELAVCE IN ZA PREVAJALCE Barbara Majcenovič Kline, Melita Koletnik 25 MIT ChatGPT AGRAR-FACHWORTSCHATZ ERLERNEN: MÖGLICHKEITEN UND GRENZEN Katja Težak, Sara Orthaber 49 LES ENJEUX DE L'INTELLIGENCE ARTIFICIELLE : ENSEIGNER ET APPRENDRE A TRADUIRE SANS ORDINATEUR – UTOPIE OU DEFI ? Nadja Dobnik 63 1 Scripta Manent XIX/1 (2024) DOI: 10.4312/SM.19.1.1-3 EDITORIAL This issue of Scripta Manent brings together a selection of papers that examine the role of Artificial Intelligence (AI) in shaping the future of Languages for Special Purposes (LSP). As the educational landscape evolves, the application and integration of technology in LSP contexts – such as the use of corpora, wikis, chat platforms, and blogs – have been widely explored (see Dashtestani & Stojković, 2015, for an overview). However, the efficacy of these technologies and their learning outcomes for LSP instruction are not always clear and remain insufficiently supported by experimental studies. This divergence emphasises the need for continuous investigation into how the potential of AI can be explored and effectively tailored to meet the specific demands of LSP education. AI’s potential in this domain is vast, and the four papers featured in this issue explore its innovative applications in LSP learning and teaching. Each paper addresses a distinct appli- cation of AI, such as generating mathematical word problems for LSP instruction, integrating AI-assisted mediation tools in teaching English for Specific Purposes, using AI to extract do- main-specific terminology, and employing AI in professional translation education to ensure the authenticity of student translations. Together, these papers provide fresh insights into the opportunities and challenges that AI brings to LSP teaching and learning across various fields. The rapid advancement of technology, particularly in AI, however, means that findings to- day may become outdated as early as the next day. This presents a significant limitation in many studies exploring the affordances (Gibson, 1979; Hutchby, 2001) of specific AI tools like ChatGPT. Put differently, as AI tools evolve at a swift pace, their capabilities, limitations, and potential applications in LSP contexts require continuous re-evaluation and adaptation of research methodologies and conclusions. In light of these rapid advancements, each paper in this issue contributes to an evolving understanding of how AI can be effectively harnessed in diverse LSP contexts. The first paper by Karen Fleischhauer and Kate Friedrich titled “Factors determining the ef- ficacy of AI-generated word problems for content-specific math language courses in higher education” explores the potential of AI tools to assist in generating word problems for math language courses in higher education. Based on insight gained through instruction of STEM students, the study investigates how AI-generated problems, which are assessed by con- tent instructors, can help LSP educators, who often lack content-specific expertise, to devel- op relevant and diverse instructional materials. The study reveals that while AI-generated word problems can save time, they also present challenges related to accuracy and consist- ency. By analysing content experts’ feedback through screencasting and semi-structured 2 Orthaber, Koletnik / Scripta Manent XIX/1 (2024), 1–3 interviews, the paper highlights both the strengths and limitations of using AI in this con- text. While AI proves helpful in creating varied problem sets, the study cautions against relying solely on these tools without careful oversight. The paper recommends integrating AI-generated materials into LSP courses but emphasises the need for further refinement and collaboration between content experts and language instructors to ensure quality and effectiveness. The second paper by Barbara Majcenovič Kline and Melita Koletnik focuses on interlingual mediation, as defined in the CEFR model of communicative language activities and strategies, combined with the use of AI in teaching of English for Specific Purposes to healthcare and translation students. By exploring the vast and ever-growing landscape of available AI tools, it examines those that facilitate interlingual mediation for each group of students, and discuss- es their potential and challenges. While AI offers benefits such as accessibility, adaptability, instant feedback, and has motivational value, the challenges include, among others, biased representation of information, the loss of human interaction, excessive reliance on technolo- gy, and the overall “dehumanisation” of teaching (Aberšek, Flogie, & Kordigel Aberšek, 2023). Through two case studies, the paper then connects theoretical insights with practical AI ap- plications and presents activities that involve the use of AI tools to enhance interlingual and intercultural mediation competences, offering a blueprint that educators can benefit from. The paper concludes that although digital tools can enhance efficiency, it is essential to main- taining interpersonal relationships with students, and to promote intercultural understand- ing and critical thinking. The third paper by Katja Težak and Sara Orthaber examines how ChatGPT can be used to en- hance terminology acquisition in a German language course for agricultural science students. As the study highlights, while ChatGPT has the potential to foster learning autonomy and ease the workload of educators, its use in self-learning raises important concerns. The re- search found that ChatGPT-3 often misidentifies terms, leading to potential misunderstand- ings, especially in specialised contexts like agricultural science. For example, everyday verbs like “wirken” were incorrectly tagged as technical terms, posing risks of limited or inaccurate vocabulary use by students. The paper suggests that while ChatGPT can serve as a valua- ble tool, it should be used with caution and preferably within a guided classroom environ- ment. This recommendation applies not only to ChatGPT-3 but also to all future versions of ChatGPT, as the same concerns regarding accuracy and contextual understanding are likely to persist. This, in turn, allows students to critically assess the AI’s output, discuss its errors with instructors, and better understand the nuances of domain-specific language. The study ultimately stresses the need for continued support from language instructors when using AI tools like ChatGPT, as well as the importance of refining these tools for effective use in LSP contexts. The fourth paper by Nadja Dobnik investigates the impact of computer-assisted translation and machine translation tools on translator training and assessment. It addresses the chal- lenges of authenticating student translations in the context of the COVID-19 pandemic, dur- ing which the widespread availability of generative AI tools has increased students’ reliance on such technologies. Focusing on the French-Slovene language pair, the research compares student translations produced with and without computer assistance. The findings highlight the need for innovative approaches to ensure the authenticity of student translations while 3 Orthaber, Koletnik / Scripta Manent XIX/1 (2024), 1–3 preparing future translators for the dynamic and evolving demands of the translation indus- try of the future. In conclusion, this issue of Scripta Manent highlights the growing influence of AI in LSP educa- tion, demonstrating its potential to transform teaching and learning across various domains. While AI offers promising tools for generating instructional materials, enhancing interlingual mediation, facilitating terminology acquisition, and reshaping translation training, its use re- quires careful supervision. The papers collectively stress the importance of continuous evalu- ation of AI tools to ensure they meet the specific demands of LSP education. As AI technology rapidly evolves, future research must keep pace to address emerging challenges and opti- mise its educational impact. Sara Orthaber & Melita Koletnik Editors References Aberšek, B., Flogie, A., & Kordigel Aberšek, M. (2023). Transformation of education: from dehumaniza- tion to re-humaization of Society. In L. Lamanauskas (ed.), Science and technology education: New developments and innovations. Proceedings of the 5th International Baltic Symposium on Science and Technology Education (BalticSTE2023). (pp. 18–27). Scientia Socialis. https://doi.org/10.33225/ BalticSTE/2023 Dashtestani, R., & Stojković, N. (2015). The use of technology in English for specific purposes (ESP) in- struction: A literature review. The Journal of Teaching English for Specific and Academic Purposes, 3(3), 435–456. Gibson, J. J. (1979). The Ecological Approach to Perception. Houghton Mifflin. Hutchby, I. (2001). Technologies, texts and affordances. Sociology, 35(2), 441–456. DOI: 10.1017/ S0038038501000219 4 Scripta Manent XIX/1 (2024) Research Paper DOI: 10.4312/SM.19.1.4-24 Paper received: 15.12.2023 Paper revised: 10.04.2024 Paper accepted: 20.04.2024 Paper published: 30.11.2024 Karen Fleischhauer Technical University of Darmstadt, Germany Kate Friedrich Technical University of Darmstadt, Germany FACTORS DETERMINING THE EFFICACY OF AI-GENERATED WORD PROBLEMS FOR CONTENT-SPECIFIC MATH LANGUAGE COURSES IN HIGHER EDUCATION Abstract The use of artificial intelligence (AI) tools to generate content-specific instructional materials has at- tracted the interest of Language for Specific Purposes (LSP) educators in higher education, as language courses in this setting typically do not utilize a textbook, requiring the instructor to create independent materials. However, instructors are often not content experts. Collaboration between LSP instructors and content experts in the form of co-teaching is one way in which materials can be generated and benefit instructors and students alike. At the same time, creating instructional materials can be a time-consuming task and can detract from other areas of collaboration. The use of AI tools to gener- ate mathematical content could help instructors save time, enable real-world connections and offer a variety of materials to students. This study examines the generation of word problems with the help of AI tools for a content-based mathematics language course for first-semester bachelor students pursuing a Science, Technology, Engineering or Mathematics (STEM) degree. As they form part of the final exam, a new set of word problems needs to be generated each year. While recent studies (cf. Lu et al., 2022) found that AI methods were effective in generating math word problems that were diverse, relevant, and useable, there have been no studies examining the applicability of AI-generated word problems in terms of their efficacy in an LSP setting. The action research study used a screencasting to capture the math content tutors’ content analysis of AI output on math word problems and was followed by a semi-structured group interview. The results showed that word problems generated by AI were generally useful based on factors such as prompting techniques. However, limitations were observed in the areas of accuracy and consistency. Based on initial results, the report suggests first implications for use in LSP instruction and describes measures that need to be taken into account in further studies. Keywords: instructional materials, word problems, co-teaching, screencasting, mathematics 5 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 1 Introduction This paper describes an action research project in which language teachers and content tu- tors, working in a co-teaching setting, explored ways that the artificial intelligence (AI) tool ChatGPT 3.5 could support the generation of materials for a language-for-specific-purposes (LSP) mathematics course in higher education. The LSP math course designed for first-se- mester international bachelor students pursuing a bachelor’s degree in science, technology, engineering or math (STEM) is led by a co-teaching team consisting of a language teacher and a mathematics content tutor1. Its main goal is to help students improve their oral communi- cation skills in mathematics in an academic setting. Although international students achieve university-level language requirements, many of them are often unable to communicate ade- quately in a specific language, which is related to a complex interplay of subjective (e.g. study satisfaction, achievement of individual study goals, intentions to withdraw) and objective factors (e.g. successful completion of studies or (in)voluntary dropout, study grades, study progress), according to the results of recent comprehensive studies such as SeSaBa (Pineda et al., 2022, pp. 10, 14; Pineda, 2018, pp. 13, 21-25). In accordance with the principle of co-teaching (cf. Cook & Friend, 1995; Escobar Urmenata, 2020; Heydarpur & Umhauer, 2018), course planning, as well as the teaching of the course, take place collaboratively. Language teachers and math content tutors each assume expert roles and complement one another in these roles (Kricke & Reich, 2016, p. 9; Heydarpur & Umhauer, 2018). One of the main tasks of the math content tutors prior to the start of the course each year (the course is held annually) is to generate a large base of math word prob- lems. Some of these word problems are used as examples during lessons, most of these, however, serve as the basis for the graded oral presentations held by students at the end of the course. Presentations focus on the mathematical strategy employed and solution path and are held in pairs and not individually, primarily due to time constraints and the large number of students in each class (three parallel courses, each comprised of ca. 20-25 stu- dents). This amounts to one dozen word problems per course and in total three dozen new word problems each year. The word problems are divided into four different thematic areas: equations, geometry, set theory and functions. These themes were chosen in close cooperation with professors and instructors in the university mathematics department and were seen to best represent the thematic areas that students in the first-semester STEM target group would encounter in their first-semester math lectures and tutorials (cf. Helfrich-Schkarbanenko, 2023, p. 103). A vast pool of word problems can be found on the Internet, yet not all include the solutions as well. This may require the tutor to write out the proof themselves in order to judge its com- plexity and to later rate the mathematical correctness of the presentation. The task of finding adequate word problems for the target group is time-consuming for math content tutors and can reduce the amount of valuable, not to mention, limited collaborative planning time with the language teacher. Co-teaching is seen as a highly valuable form of teaching to students and teachers alike, yet it is more the exception than the rule in a high- er education setting (Rooks et al., 2022, p. 1). At the same time, the need for more explicit 1 Content math tutors are higher-semester university students studying a STEM degree. Many of the content tutors in this setting are also teachers in training (pre-service teachers). 6 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 treatment of specific language (especially English for Specific Purposes) in higher education is increasing (Kováčiková, 2020). Considering the abovementioned time constraints, juxtaposed with the positive response of students and teachers toward the co-teaching setting in this LSP course, the authors wanted to find a way to generate math word problems more efficiently in order to leave more time for collaborative course preparation. The release of ChatGPT 3.5 in November 2022 encouraged them to explore ways in which this AI tool could assist them in doing so. This paper describes initial attempts to use AI tools to help in materials development for a LSP math course for first-semester STEM students at the Language Resource Centre at the Technical University of Darmstadt (Germany). We begin with a brief overview of studies which have examined the efficacy of AI tool use for the generation of mathematical word problems and their solutions. Next, we explain the methods used for data collection in this action re- search study and offer a preliminary glimpse into the emerging themes and patterns dis- cerned through analysis of the gathered data. We conclude with a discussion of implications for further studies as well as relevance of the findings for the LSP classroom. 2 Background The number of studies pertaining to AI use in the foreign language classroom has increased steadily in the past two decades, and even more so since the release of ChatGPT at the end of November 2022 (see Huang et al., 2023). However, very few of these studies focus explicitly on the use of AI tools for LSP instruction, let alone on using LSP for mathematics. The interest in the use of artificial intelligence tools to help solve math problems began well over a half a century ago (Lu et al., 2022, p. 1). However, the launch of ChatGPT 3.5 marks a new phase for AI users, as it can be operated via natural language alone. This means that users do not have to rely on or to learn additional mathematical languages to interact with ChatGPT (Spannagel, 2023, p. 168). Word problems or questions pertaining to word problems can be entered directly into the interface via prompts. Paradoxically, the artificial intelligence of ChatGPT, a Large Language Model (LLM), does not lie with its comprehension or thought about the meaning of the natural language entered by humans in its prompts. It is a language model and not a knowledge model and is based on probability and not on truth (Spannagel, 2023, p. 168; Wessels, 2023). This means that it un- derstands the individual words in the prompt only as natural language symbols (tokens) and generates its output based on the mathematical probability of what an answer should look like and not on reason. The database it uses calculates this probability and hence the quality of its output depends by and large on the natural language texts it was trained with and the respective context of these texts (Arnold, 2023). The challenge of generating and solving word problems for AI tools like ChatGPT lies with “language comprehension, semantic parsing, and multiple mathematical reasoning skills” (Lu et  al., 2022, pp. 2, 21). Not only is sound mathematical knowledge required to solve math problems, but tacit, real-world knowledge is also essential. However, commonsense knowledge is often assumed and not explicitly addressed and constitutes a “largely unob- served foundation” (Davis, 2023, p. 2). This may be why no chatbot always provided the right 7 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 answer, even with simple mathematical operations (addition), and why there are no guaran- tees that the LLM will correctly solve a math problem (Plevris et al., 2023, p. 18; Spannagel, 2023, p. 168). Thus, as a communication partner, ChatGPT is not always dependable, as it can generate false output or “hallucinate” (Arnold, 2023; Ghassemi et al., 2023, p. 39; Wes- sels, 2023). Studies pertaining to solving word problems with ChatGPT have been published by profes- sors of mathematics or professionals with extensive mathematical experience (Frieder et al., 2023, p. 18; Helfrich-Schkarbanenko, 2023) and describe the nature of their communication via prompts. Not surprisingly, these studies do not address aspects for language learning. This raises the question as to how non-expert educators and researchers (i.e. language teach- ers) can efficiently use tools such as ChatGPT. The clear response among mathematicians regarding the use of ChatGPT in generating and solving word problems is positive (Helfrich-Schkarbanenko, 2023, pp. 33, 61; Spannagel, 2023, p. 168). While some studies compared the output generated from various versions of ChatGPT 3.5 and 4.0 (Spannagel, 2023, p. 168) and compared these with Google Bard (Plevris et al., 2023), most studies exclusively used ChatGPT 4.0 and were overwhelmingly positive in their critique (Helfrich-Schkarbanenko, 2023; Spannagel, 2023, p. 169). Several studies inte- grated ChatGPT 4.0 with “mathematical experts” (Spannagel 2023, p. 170) or plug-ins such as Wolfram Alpha, which can considerably improve the quality of the mathematical output generated (Helfrich-Schkarbanenko, 2023; Spannagel, 2023, p. 169). Access to ChatGPT 4.0 is only available with a Plus account for which users must pay a month- ly fee. Additionally, access to plug-ins is a service only available to users of a Plus account. Just how exclusive the access to these services is becomes even clearer when we learn that the number of users registered with a Plus account compared to the total number of ChatGPT users2 is less than 1% (Nerdynav, 2022). 3 Methodology Based on conclusions from the background, it becomes evident that existing research has largely centered on the content-related perspective of mathematicians. This study seeks to address a notable gap by focusing on language-related issues and frames these in the follow- ing research questions: 1) How do content tutors determine the applicability of math word problems generated by ChatGPT 3.5 for an LSP math course? 2) How might LSP teachers use word problems generated by AI for instructional purposes? Building upon the insights gleaned from the background, this study adopts a qualitative, exploratory approach and focuses on the language-related issues pertaining to the use of AI tools as they apply to the LSP math instructional setting. The nature of the research is inher- ently non-representative, focusing on depth rather than breadth. 2 ChatGPT has more than 180.5 million active users, whereas the number of ChatGPT Plus (4.0) subscribers is estimated between 230,000-250,000 as of October 2023 (Nerdynav, 2022). 8 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 3.1 Action research Action research enabled us to inquire into our own instructional practices (Somekh, 2008, p. 5) and coincided well with the collaborative nature of co-teaching in the LSP math course. Per definition, action research is a research methodology that involves a systematic and col- laborative inquiry conducted by educators within their own teaching contexts. It aims to im- prove teaching practices, enhance learning outcomes, and address specific challenges or issues in the language education setting (Cohen, 2017, pp. 440–456; Lew et al., 2018, p. 89; Banegas & Consoli, 2019, pp. 176–187). Action research typically follows a cyclical process of planning, acting, observing, and reflecting, allowing teachers to make informed and iterative adjustments to their instructional methods based on real–time feedback and insights gained from their own experiences and the experiences of their students. This approach emphasizes the practical application of research findings to bring about positive changes in the teaching and learning environment. According to Banegas and Consoli (2019), action research can be guided by three main as- pects: context, agents and issues (p. 177). Our inquiry was largely based on ways to change certain aspects of the instructional setting (context), as we (agents: here to be understood as co-teachers as well as co-researchers) (Somekh, 2008, p. 4) were looking for ways to in- crease the time for collaborative planning of the LSP math course (the issue). With the launch of ChatGPT 3.5 the co-authors saw the opportunity to explore whether the help of AI tools (possible intervention tool) would reduce the workload and allocate time for meaningful and cognitively more demanding activities, such as leveraging skills and levels of expertise of co-teachers in collaborative course planning. 3.2 Data Collection Data was collected within a one-week period from two different settings: a screencasting and a semi-structured group interview in the form of two recordings and the written chat history with ChatGPT. 3.2.1 Screencasting Screencasting is the process of recording the actions on a computer or a digital screen (e.g. a smart phone). It allows us to capture computer screen output, including audio narration. Screencasting can be valuable for studying learning and exploration processes that take place in the digital environment and is an instrument used to collect data for qualitative (as well as quantitative) research (see Lee, 2022; Mroz, 2012). Here researchers may gain great- er insights into materials development in the LSP setting through the computer-mediated setting (Hamel & Séror, 2016, pp. 138, 140) to be better able to understand the multimodal interaction in the digital setting (p. 140). Screencasts monitor user activities and can thus precisely show the steps the user(s) took at certain times during an online session, which an audio recording or even an audio-visual recording without screensharing could not provide. The actions taken during the screencasting can later be used as an impetus for questions to be asked during a follow-up interview. A screencasting can present a more dynamic view of 9 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 learning processes and help the researcher(s) gain valuable insights into strategies for using (in this case) AI tools for LSP materials development (p. 142). The time spent on various ac- tivities can also be recorded and measured and can provide insights into which activities or strategies are implemented most often (p. 142). A one-hour screencasting was led by two experienced math content tutors who had already taught in the LSP math course. Video recording of the session enabled with a digital video platform (Zoom®) was hosted by one of the content tutors (co-author) participating in the screencasting. This content tutor also shared their ChatGPT 3.5 (free version) screen during the session, entered the prompts and agreed to share the saved prompts as well as the out- comes generated for later transcription and data analysis. This content tutor (co-author) also moderated the screencasting and explained the pro- cedure of the screencasting to the other content tutor ahead of time. Both content tutors agreed on a selective inquiry of only three of the four thematic word problem areas: equa- tions, geometry, and functions, as they anticipated the generation of output in these areas to be most useful for the instructional setting and most likely applicable for use by non-experts (i.e., language teachers) in the co-teaching setting. They agreed that the focus should lie first on equations, as it is in many ways the most basic of the selected mathematical areas. It is versatile but manageable, the one that is taught in school the earliest, and its vocabulary, especially concerning solution strategies, is necessary for any mathematical conversation. Geometry word problems were to follow, which involve a whole new set of vocabulary and where the emphasis on visual representation is in stark contrast to the first subject. Then, if time allowed, they wanted to look at functions, a topic the tutors considered to be a sort of hybrid of the first two. The other co-author (a language teacher) took part in the planning phase prior to the screen- casting; however, they were not present during the screencasting. This mirrors the prepara- tion phase of the LSP course where the tutors generate a list of math word problems without their corresponding language teachers, because the content tutors inhabit the role of math experts. Here both participating tutors had plenty of experience with this process. Written consent of all participants was gathered prior to data collection. Audio files were transcribed according to the conventions of Dresing and Pehl (2018) and Kuckartz and Rädik- er (2023). 3.2.2 Semi-structured group interview A method of inquiry often used in qualitative studies is the semi-structured interview. Here the role of the interviewer is to provide an open-ended format in which thematic impulses should encourage the interviewees to elaborate in a detailed narrative on themes related to the research focus, which should also be in alignment with the research questions. The inter- action between the interviewer and the interviewees during the interview generates a com- municative text (Helfferich, 2019, p. 671), which can later serve as an important source for data analysis. The semi-structured interview is well-suited when the researcher (interviewer) has a solid understanding of the research setting (Dörnyei, 2012, p. 146), which applies well to action research, where teachers act as researchers in their own learning setting. 10 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 Transcription of audio data from the screencasting was useful in preparation for the semi-structured interview (Helfferich, 2019), which followed exactly one week later. The nar- row time lapse between the screencasting and interview allows for elicitation of retrospective accounts. The thematic impulses given by the interviewer (co-author and language teacher) were in alignment with the research questions and were designed to encourage participants (content tutors) to engage in detailed narrative about their experiences from the LSP course and during the screencasting. The one-hour interview was held online, an audio recording was made with the same transcription conventions as with the screencasting. 4 Data analysis In this section, we analyse the rich details based on qualitative data gathered through in- depth exploration and experiences of the co-authors (as co-teachers and co-researchers). The focus of our analysis centres on identifying and interpreting emergent themes gathered inductively that capture the dimensions of our research questions. A smaller quantitative component involves an examination of screencasting data, wherein the frequency of specific actions (prompting, evaluation and other tasks) was measured over the one-hour timeframe. The quantitative analysis facilitated the derivation of percentages and enables a nuanced portrayal of the prevalence and distribution of the observed actions and also offers a lens through which to comprehend and articulate the salient trends within this small data set. Based on the data, the following four categories emerged from the inter- action between the content tutors: prompting, AI output, interaction with the AI, and appli- cability for LSP math setting. 4.1 Prompting Prompting with a Large Language Model such as ChatGPT 3.5 means providing it with input to which it is meant to respond. Prompting is written in natural language, and responses from the LLM (known as output) also appear in natural language. During the screencasting, the content tutors wrote prompts which contained information about the type of word prob- lem desired, the level of the target group and the request for a solution. The content tutors considered this prompt information to be general. Example of a basic prompt: P2: Let’s start with a very general task on the topic of terms and equations. Then maybe the level? #00:00:50-6#3 Post-editing. After posting an initial prompt to ChatGPT and reviewing the output, changes can be made to the previous prompt(s) in order to specify details that were overlooked or to express doubt toward the given answer. In our case, post-editing steps often included the request to ChatGPT 3.5 to increase the complexity of the word problem. Content tutors’ post-editing prompting strategies show their knowledge, experience and consideration of 3 Original transcriptions (of screencasting and semi-structured interview) are in the German language. Ex- cerpts appear here in English and were translated by the co-authors. 11 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 the LSP math instructional setting, as they try to match the given output to their vision of the required course material. Example of post-edit: P2: Then we thought, yes, maybe we’ll just put the word “more complex” in the sentence [prompt] so that the task becomes a bit more difficult, so that the task explanation in the [oral] exam is a bit more detailed. #00:06:09-8# As a whole, the process of creating, wording, and discussing the prompts during the screen- casting was straightforward. The content tutors quickly agreed on each other’s suggestions and a successful approach was found early on; sometimes the goal of a new prompt was simply to generate more of the same output. Figure 1 Distribution of time spent discussing prompts, outputs, and other subjects during screencasting 18% 57% 25% Distribu�on of Time Prompt Output Other When looking at the mathematical topics of equations, geometry, and functions individually, one must first clarify that due to time constraints, the discussion of prompts to generate word problems on functions only briefly took place during the screencasting and remained on a the- oretical level, as no prompts on functions were entered. Overall, the prompting of equations was less time-consuming than with geometry word problems. Of the 15 prompts written in total within the one-hour screencasting, only 6 prompts were written for equations, whereas the content tutors wrote a total of 9 prompts for word problems having to do with geometry (and none for functions). This was due to a greater need for post-editing with geometry word problems. A more in-depth breakdown of prompt and output counts can be seen in Table 1. 4.2 Evaluation of output Over half of the time spent during the screencasting involved evaluation of the output generat- ed by ChatGPT 3.5 (see Figure 1). Content tutors reported being pleased with the generated out- put for equations. In contrast, geometry word problems generated by ChatGPT 3.5 contained 12 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 many errors. In addition, crucial visual elements (e.g., a sketch) were missing from the output for geometry problems, a feature only available with the Plus version of ChatGPT (4.0). Table 1 Evaluation of prompts and output according to suitability – terms and equations, and geometry Word Problem Type Prompts Output all with post-editing all unsuitable suitable no changes necessary with minor changes Terms and Equations 6 4 7 0 6 1 Geometry 9 7 8 5 1 2 As indicated in Table 1, there are more math language problems than prompts for terms and equations, whereas for geometry it is the opposite case. The generated output (i.e., word problems and their solutions) for terms and equations showed potential (see Figure 2), and so the content tutors requested the generation of multiple word problems at once. In the lat- ter part of the screencasting involving geometry word problems, the content tutors often dis- agreed with the output and spent more time responding to the prompts and asking ChatGPT questions about the output. Even though the number of math language problems generated was nearly the same for both word problem types (terms and equations, and geometry), the number of suitable word problems for geometry was considerably lower (see Figure 3). There was a higher need for post-editing the prompts with geometry problems; the content tutors tried specifying a ge- ometric shape, the required solving strategy, and a thematic context due to poor results with prior generic prompts. Here only one out of four generated language problems was suitable (with minor changes). Minor changes meant an adjustment of the wording of the word prob- lem or recategorizing the problem from an exam exercise to an example during class. Figure 2 Proportions of (un)suitable equations word problems 86% 14% Terms and Equa�ons suitable suitable (minor changes) unsuitable 13 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 Figure 3 Proportions of suitable geometry word problems 10% 20% 70% Geometry suitable suitable (minor changes) unsuitable The content tutors noticed a clear deficit with geometry problems due to the lack of images which could better describe spatial relationships. As was discovered, ChatGPT 3.5 is unable to generate images (unlike version 4.0). P2: But I think I’m just missing a sketch based on the text so that I can imagine it better. #00:37:29-1# Overall, content tutors evaluated the quality of the generated output in terms of a) whether the mathematics level was appropriate for the target group, b) whether the language level matched the level of the language course, c) whether the math language problem made sense as a question, and d) whether the solution was correct and comprehensible. Structure and correctness of the proof. Proofs represent an important part of LSP math instruction because they foster precision, logical reasoning, critical thinking, and effective com- munication. They serve as a bridge between language and mathematical concepts and can enhance the overall language proficiency and analytical skills of learners (cf. Rocha 2019, p. 1). Overall, content tutors were pleased with the structure of the proofs given as answers for equations based on their appearance, their complexity as well as their correctness. In mathe- matics a proof is considered correct when the final statement follows logically from the given propositions by deductive reasoning and the line of argument is exhaustive i.e. there are no gaps in the step-by-step explanation. An example of the content tutors’ evaluation follows: P2: The solution is so clearly presented. It actually contains all the important calcula- tion steps. It4 [ChatGPT] even named them. That’s good. But the task is just a bit simple. #00:02:26-8# 4 In the original version, the participant used the German masculine pronoun “er” to refer to ChatGPT as he. Use of the masculine pronoun could be attributed to a reference to the German masculine noun “der Chat- bot” (Duden, n.d.). However, ChatGPT is also referred to as a program (German: das Programm, a neuter noun, which corresponds to the neuter pronoun “es” or in English, “it”). 14 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 P2: Yes, it is also so well-structured, [the steps are] written so nicely one below the other. #00:04:37-8# The format of the answers to the word problems, i.e. calculations embedded into a coherent text, was evaluated positively by the math content tutors as being a textbook example/par- adigm. It met their expectations for an answer that could have been presented by a student in the LSP course. P1: I would really like people in the language course to say, “Um, I subtract 2x” and it’s really correct. #00:04:11-8# Comprehensibility of output. Critical review of output showed that the generated text did not always make sense and lacked commonsense or real-world knowledge, as the example in this output demonstrates: Figure 4 Prompt for geometry problem including “hallucinating” output generated by ChatGPT 3.5 Prompt: Create a word problem in which you have to calculate the volume and surface area of a composite geometric solid. Output from ChatGPT 3.5: In an art project, a sculpture park is being designed that consists of various geomet- ric solids. One sculpture has the shape of a rectangular prism measuring 6 meters in length, 4 meters in width and 3 meters in height. A hemispherical dome is placed on the top of the prism, which fits exactly on the rectangular prism. Calculate the volume and surface area of this sculpture. Solution: The sculpture consists of a rectangular prism and a hemispherical dome. P1: How can a hemisphere fit exactly on the rectangle? One is round and the other is rec- tangular. #00:56:10-7# The content tutors’ satisfaction with the output for geometry word problems was much lower due to factors such as lack of accuracy and a lack of visualizations. P1: The crux of the matter is that you cannot use the text alone, or because there is no illustration, it is difficult to check that the text is not accurate, i.e. that the text does not make physical sense. (...) And because no illustration is provided, it’s also not a good way to quickly create tasks that you can quickly check. Because I think that just eats up more time. #00:59:01-9# Time spent reviewing output. Because the generation of math word problems was time-con- suming for content tutors and reduced the amount of valuable collaborative planning time with the co-teacher (language teacher), the use of AI tools was seen as a possible way to make this part of course planning more efficient. 15 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 Prior to the screencasting, the content tutors said they were hoping to be able to save time by having the word problems generated by ChatGPT 3.5. Figure 5 Progression of screencasting session according to action and amount of time Subject Change Subject Change 00:00:00 00:10:00 00:20:00 00:30:00 00:40:00 00:50:00 01:00:00 1 Screencas�ng Progression The bar in Figure 5 represents the duration of the screencasting and is cut into time spans ac- cording to the topic of discussion. The red stripes indicate parts of the conversation referring to the prompt and the green stripes stand for discussion of the output. As indicated in the figure, the time needed to discuss the output for terms and equations was far less than in the second part about geometry (indicated by the first “Subject Change”). Theories on designing adequate prompts of a third thematic area (functions), designated in Figure 5 by the second “Subject Change”, was discussed by the content tutors, yet no actual prompts were generated during the screencasting. Reflecting during the interview on how their time during the screencasting was spent, the content tutors note that users of AI tools such as ChatGPT 3.5 must consider not only the writing of prompts but also the time needed for the critical proofing of the output. One con- tent tutor said: P1: Well, I mean, you have to take into account that the time you save when generating [the word problems], you still have to check the figures. #00:29:23-4# 4.3 Interaction with ChatGPT During the screencasting the content tutors discuss with one another the nature of their interaction with ChatGPT 3.5. They assume a critical stance towards the generated output throughout the screencast. They discuss the importance of reading through the material themselves and how complete trust of ChatGPT without the review of the (human) users is ill-advised. As one tutor put it: P2: But I still don’t think you can trust it blindly. I think you still have to look at the calcu- lation somehow just to have the personal feeling that you’ve checked it again. So I find it difficult to trust just that. #00:05:50-6# 16 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 During the screencasting, content tutors negotiated the value of the generated output. They interacted with one another (human to human) as well as together with ChatGPT 3.5 (human to machine), so that a multimodal interaction can be described within the observed setting. 4.3.1 Applicability for LSP math setting The content tutors are not only math experts for the target group, they are also thoroughly familiar with the LSP math course setting and assume an emic perspective of the target group during their use of ChatGPT 3.5. In their interaction during the screencasting and during the interview, they speak not only about the mathematics as such, they also often make reference to their experiences as content tutors in the LSP math setting as they de- termine whether the AI-generated output is applicable to the language learning setting. For example: P2: Yes, in my opinion, you also want language lessons to somehow focus not on applying a complex mathematical procedure, but on being able to express the approach you have, even if it is rather straightforward, so to speak. And that’s why there are only two steps, so to speak. And then in the steps, you can also describe the arithmetic operations actually exactly appropriate for this. #00:15:55-7# Comprehensibility. During the interview, the content tutors emphasized the significance of incorporating visual aids in addition to mathematical and language output, particularly in thematic areas such as geometry and certain aspects of functions. Determining whether an equation word problem is well-posed and whether its solution is correct is generally much easier. This field of mathematics is, therefore, more viable than the generation of geometry word problems with the help of ChatGPT 3.5. The following example illustrates this view. P2: Exactly. So if you read through it [output generated by ChatGPT 3.5] and say, that makes sense. But it’s hard to imagine, could you perhaps make an additional drawing for the students? And in the [mathematical] thematic area of functions, we haven’t created a task or had one created. But we have speculated a little that tasks that rely on a sketch are probably less useful to create. But those where you have to calculate an intersection point, for example, which is very similar to an equation, because what is actually an equation is very similar to the subject area of terms and equations, which is certainly also suitable for this. Exactly. And? Yes. It is also suitable for language teaching in that the an- swers that are given, as P2 had already said, are examples of how to describe what you have calculated, so to speak. #00:26:34-4# Complexity. Many of the word problems generated by ChatGPT lacked complexity and were described by the content tutors as being too easy and the solution as too straightforward. This must be taken into careful consideration when choosing teaching approaches for the LSP math course, as one content tutor mentions. P2: Um, so you have to be careful not to create a false expectation, so to speak, that you only do easy word problems in class and the exam it’s then more difficult, rather you do an easy word problem to get used to the vocabulary and the special expressions that are somehow special for the thematic area and then on the same day or in the days 17 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 preparing for the exam, then again provide a more complex word problem from the sub- ject area. #00:27:45-4# Register. Students need to be familiar with the variety of language or level of formality used in mathematics lectures and tutorials so that they can actively socialize in the univer- sity setting. Academic socialization requires knowledge, understanding as well as practice of register. The term “register” refers to a variety of language that is associated with a par- ticular communicative situation, context, or social setting. It involves the use of language that is appropriate and tailored to the specific requirements, norms, and expectations of a given context. Registers can vary based on factors such as formality, field of discourse, purpose of communication, and the relationships between interlocutors (Richards et al., 1999, p. 313). Many students enter their university studies unaware of the levels of (in) formality. Therefore, careful evaluation of a text’s register is an important part of materials development. In the LSP math course, content tutors are peer experts in two ways: they have expert knowl- edge in the mathematical content for the target group and are also peer experts, in that they already have experience within the university (mathematical) setting. For this reason, they can evaluate how well ChatGPT 3.5 performs concerning the mathematical language for the university setting. As one content tutor explains: P2: One aspect that the subject tutors are there for is to convey a bit of a feeling. That’s how you say it at university, especially in the setting or to decide that it might be either somehow outdated language or at school you could say it like that but at university you wouldn’t say it anymore. And in that aspect, I had the impression that ChatGPT delivered brilliantly. So phrasing was never weird or something. #00:29:57-9# The specialty of ChatGPT 3.55 is generating written text output. Hence, it is not surprising yet worth mentioning that during the screencasting no instance occurred in which the text for- mat requirements for an answer to a math word problem were not adhered to. 5 Discussion The discussion is based on the research questions posed at the beginning of the paper. Discussion of research question 1: How do content tutors determine the applicability of math word problems generated by ChatGPT 3.5 for a LSP math course? Content tutors judge the applicability of math word problems based on time saved, quality of output, and relevance for the target group. They underpin decisions on their expertise in mathematics as well as their prior experience as content tutors in the LSP setting. Content tutors evaluated the output based on level of difficulty and its applicability for the LSP math setting. Word problems can also be sorted according to where they would be need- ed, given the structure and organization of the LSP math course. 5 ChatGPT 4.0 18 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 The tacit, real-world knowledge in word problems presents a challenge to language learners and users of Large Language Models alike. Understanding what is meant in a word problem (what is given and what is asked for) is not always clear and can lead to false answers. LSP math learners need to be aware of the nuances of LSP math when solving word problems, just as users of AI tools such as ChatGPT need to be able to be explicit when writing their prompts. LLM users need to be aware that ChatGPT is a language and not a knowledge model and assumes (or calls for) a certain degree of digital competency. ChatGPT 3.5 can help reduce time spent with generating certain word problems, such as equations, yet this is not equivalent to the teaching of these word problems in the LSP set- ting, as one content tutor asserts: P1: I would like to point out that if the language teacher has the tasks created with ChatGPT in preparation for the course or perhaps selects tasks in some areas that is not an adequate means of replacing the language tutor’s task. Because on the one hand, the time. Yes, it’s just distributed to another person. And secondly, co-teaching is also relevant in the classroom and not just in preparation [emphasis from authors]. And that’s why, if someone creates the tasks, it would be best if it was a person with mathe- matical training. And it doesn’t necessarily have to be. In certain contexts, anyone can do it, but it still doesn’t excuse the nice word problem that you then have if the person who provides the impetus in class is missing, so to speak. #00:40:06-7# Discussion of research question 2: How may LSP teachers use word problems generated by AI for instructional purposes? As a non-expert in the field of mathematics, the language teacher (co-author) in this co-teaching setting has gained a better understanding of the content tutors’ decision-mak- ing process involved in generating word problems with AI tools such as ChatGPT 3.5. Here the screencasting enabled a multimodal observation of the various types of interaction taking place: not only among the content tutors themselves or between the content tutors with ChatGPT 3.5 as a learning partner, it also revealed the ways in which content tutors maneuver themselves within the digital interface of ChatGPT 3.5 and offered insights into the skills which teachers need to have when developing LSP materials with the help of an AI tool. ChatGPT 3.5 may partially aid in generating word problems and create more time for other collaborative tasks important for a co-teaching setting. Nonetheless, content tutors are clear that they themselves are irreplaceable6, despite positive and time-saving support, even from ChatGPT 4.0. The findings of this action research may also have value for LSP teachers who are not in a co-teaching setting, as they can offer general recommendations for use of ChatGPT 3.5 for the planning of LSP course materials in three areas: prompting, evaluation of output, and applicability for the LSP setting. 6 The UNESCO education 2030 emphasizes a “human-centered approach to AI” (2023, p. 10). See also Kasneci et al. (2023, Multilingualism and fair access section). 19 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 Prompting: 1. Use of LLMs should focus on thematic areas that do not heavily rely on visual informa- tion (such as geometry word problems). 2. The intended language and content level of the learner target group should be clearly indicated in the prompt. 3. The type of expected answer (e.g. a proof) should be clearly indicated in the prompt and should correspond to the thematic area. 4. The prompt should request the generation of more word problems than needed. This allows for choosing the best output without additional prompting. Evaluation of output: 1. The solution should contain the answer sentence, which allows for easy verification of the word problem. 2. The output should handle all parts of the word problem. A further prompt can be writ- ten to inquire about this. 3. Careful attention should be paid to real-world context when evaluating output by ask- ing the following questions: a) What is the order of magnitude of the figures? b) What kind of numbers are expected (e.g. natural numbers)? c) Do the units fit the item asked for? Applicability for the LSP course: 1. The answers in ChatGPT 3.5 tend to be long and detailed. This helps to determine whether they are appropriate for an oral exam (the answer needs to be long enough for a 10-minute oral presentation, for instance). 2. The lexical terms used should correspond to the vocabulary covered in the LSP course. 3. Our findings indicate that ChatGPT generates reliable output in terms of register, i.e. appropriate for university mathematics. 4. Request for output of oral texts is not recommended because GPT is trained on written texts. In an even broader scope, our findings offer insights into the development of LSP instruc- tional materials with AI tools and the importance of digital competency as a key skill for the professionalization of language teachers. While the authenticity of instructional materials remains a constant “driving force” (Arnó-Maciá, 2013, p. 5) for LSP teachers, the way in which these materials are found and furthermore used is fundamentally changing with the use of AI tools, as we see a shift from “materials as product [to] materials as process” (Mishan, 2022, p.19). With the help of AI tools such as ChatGPT, teachers no longer just locate instructional materials online, as they formerly would do in an online or analogue library. Instead, ma- terials are generated with the help of AI tools and do not just represent a digital version of print materials (cf. Mishan, 2022, p. 19). Furthermore, the materials that are generated are not a finished product, rather, they are generated within the process of human to machine 20 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 communication between the language teacher and the AI tool. With this understanding, we see that the use of AI tools here represents not merely an alternative to print materials, it offers a whole new dimension to materials development (cf. Mishan, 2022, p. 19), and repre- sents a “blurring of the materials-tools distinction” (Mishan, 2022, p. 20). As AI tools themselves play a more active role than ever, they are transforming the role of the (LSP) language teacher and elucidate the importance of digital (language) competency for (LSP) teachers. To critically engage with technology when using it for materials development in LSP settings, language teachers need to cultivate digital competencies (such as learning how to interact with ChatGPT) that prioritize pedagogical value and ethical use. Such competencies will enable them to “capitalize on its [ChatGPT] affordances” (Kohnke et al., 2023, p. 10). 6 Limitations and Concluding Remarks Due to the limited scope of this action research study, our findings cannot confidently be applied to more general contexts. Our sample was too small to be representative, as we only focused on the use of one AI chatbot, namely ChatGPT 3.5, and only one LSP envi- ronment, namely mathematics for university studies in STEM. However, the results may be valuable for teacher development within the LSP context, and potentially hold wider applicability. Without any practical experience with ChatGPT 4.0, it is difficult to evaluate the promising re- sults of studies performed by mathematicians working in a higher education setting. ChatGPT 4.0 is reportedly much better in solving word problems in mathematics, including geometry, set theory, and functions. This conclusion is based on the fact that ChatGPT 4.0 can be em- bedded into highly powerful mathematical tools such as Wolfram Alpha. In just over one year, the number of AI tools has exponentially increased as well as the range of application areas, not to mention the quality and accuracy of their output. Looking through a lens of optimism, the future of AI in the LSP classroom could hold many positive changes and opportunities for instructors, students, and researchers. However, the upward trend could also continue in an even more exclusive direction, if we consider the great disparity between promising studies mentioned in this paper that were performed using ChatGPT 4.0 by a selected group of users. When compared with 99% of users, who may be well-informed but who have no access to practical use of these tools, we witness a widening of the digital divide7 on two levels. On the one hand, the “access divide” (Iberdrola, n.d.), such as the access to (or lack thereof) ChatGPT 4.0 due to finances and infra- structure could have far-reaching consequences and result in inequality among researchers, educators and students alike. Another equally important form of the digital divide which de- serves attention refers to the types of digital skills that teachers have (or do not have) when using digital tools such as ChatGPT. Here the “use divide” (Iberdrola, n.d.) which refers to digital competency skills (or lack thereof) that language teachers need in order to use tech- nology in ways that are sustainable, fair, not to mention safe and pedagogically sound. This 7 Use of the term “digital divide” to describe household access (or lack thereof) to the Internet saw a notable increase in usage in the 1990s (Pierce, 2018, p. 1). However, it remains unclear as to who actually coined this term (cf. Gunkel, 2003, p. 501). 21 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 divide can also be attributed to factors such as age, academic background and professional development opportunities from one’s employer. The value and importance of action research amidst such major changes in the current ed- ucational landscape shows that slow policymaking with regard to the use of AI, coupled with rapid technological advantages in AI, can be countered by meaningful, contextually rich col- laboration among teachers (cf. Eisenmann & Steinbock, 2024, pp. 3,5). Teachers can be en- couraged to reflect on the current needs of their students and actively pursue ways in which to help them. It is, nevertheless, clear that it is crucial to continue to explore how AI tools such as ChatGPT 3.5 can promote effective and accurate language use within specific language contexts. References Arnold, T. (2023, February 10). ChatGPT für Nicht-Informatiker*innen [Video]. YouTube. https://www. youtube.com/watch?v=-c8ogAwX6KI&t Arnó-Macià, E. (2013). Information technology and languages for specific purposes in the EHEA: Op- tions and challenges for the knowledge society. In E. Bárcena, T. Read & J. Arús (Eds.), Languages for Specific Purposes in the Digital Era (pp. 3–25). Springer International Publishing. https://doi. org/10.1007/978-3-319-02222-2 Banegas, D.-L., & Consoli, S. (2019). Action research in language education. In J. McKinley & H. Rose H. (Eds.). The Routledge Handbook of Research Methods in Applied Linguistics (pp. 176–187). Rout- ledge. https://doi.org/10.4324/9780367824471 Cohen, L. (2017). Research methods in education (8th ed.). Taylor and Francis. https://doi. org/10.4324/9781315456539 Cook, L., & Friend, M. (1995). Co-teaching: Guidelines for creating effective practices. Focus on Excep- tional Children, 28(3), 1–16. https://doi.org/10.17161/foec.v28i3.6852 Davis, E. (2024). Mathematics, word problems, common sense, and artificial intelligence. Bulletin of the American Mathematical Society, 61(2), 287–303. https://doi.org/10.1090/bull/1828 Dörnyei, Z. (2012). Research methods in applied linguistics: Quantitative, qualitative, and mixed methodol- ogies [reprint]. Oxford Univ. Press. Dresing, T., & Pehl, T. (2018). Praxisbuch Interview, Transkription & Analyse. Anleitungen und Regelsysteme für qualitativ Forschende. 8. Auflage. Eigenverlag: Marburg. Duden. (n. d.). Chatbot. https://www.duden.de/node/237994/revision/1382866 Eisenmann, M. & Steinbock, J. (2023). Vorwort. ZFF – Zeitschrift für Fremdsprachenforschung, 34(1), 3–7. Escobar Urmeneta, C. (2020). Coteaching in CLIL in Catalonia. CLIL Journal of Innovation and Research in Plurilingual and Plurilingual Education, 3(2), 37–55. https://doi.org/10.5565/rev/clil.54 Frieder, S., Pinchetti, L., Chevalier, A., Griffiths, R. R., Salvatori, T., Lukasiewicz, T., Petersen, P. C., & Berner, J. (2023). Mathematical capabilities of ChatGPT. https://arxiv.org/abs/2301.13867 Ghassemi, M., Birhane, A., Bilal, M., Kankaria, S., Malone, C., Mollick, E., & Tustumi, F. (2023). ChatGPT one year on: Who is using it, how and why? Nature, 624(7990), 39–41. 10.1038/d41586-023-03798-6 Gunkel, D. J. (2003). Second thoughts: Toward a critique of the digital divide. New media & society, 5(4), 499–522. https://doi.org/10.1177/146144480354003 22 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 Hamel, M., & Séror, J. (2016). Video screen capture to document and scaffold the L2 writing process. In C. Caws & M.-J. Hamel (Eds.), Language studies, science and engineering: Vol. 2. Language-learner computer interactions: Theory, methodology and CALL applications (pp. 137–162). John Benjamins Publishing Company. https://doi.org/10.1075/lsse.2.07ham Helfferich, C. (2019). Leitfaden- und Experteninterviews. In N. Baur & J. Blasius (Eds.), Springer eB- ook Collection. Handbuch Methoden der empirischen Sozialforschung (pp. 669–686). Springer VS. https://doi.org/10.1007/978-3-658-21308-4_44 Helfrich-Schkarbanenko, A. (2023). Mathematik und ChatGPT: Ein Rendezvous am Fuße der technologis- chen Singularität. Springer Spektrum. https://doi.org/10.1007/978 Heydarpur, S., & Umhauer, A. (2018). Co-Teaching. Eine Strategie für Nachwuchswissenschaftler*in- nen zur Erleichterung des Einstiegs in die universitäre Lehre. (Hochschuldidaktischer Berater: Franz Kroath). Universität Innsbruck: Institut für Erziehungswissenschaft. https://www.uibk. ac.at/media/filer_public/92/61/9261372e-f445-4789-b4aa-384b468c3056/2018-12-04_umhaus- er_andrea_heydarpur_sepideh.pdf Huang, X., Zou, D., Cheng, G., Chen, X., & Xie, H. (2023). Trends, research issues and applications of arti- ficial intelligence in language education. Educational Technology & Society, 26(1), 112–131. https:// doi.org/10.30191/ETS.202301_26(1).0009 Iberdrola. (n. d.). Digital divide. Digital divide throughout the world and why it causes inequality. Iberdrola. https://www.iberdrola.com/social-commitment/what-is-digital-divide Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., Krusche, S., Kutyniok, G., Michaeli, T., Nerdel, C., Pfeffer, J., Poquet, O., Sailer, M., Schmidt, A., Seidel, T., & Stadler, M. (2023). ChatGPT for Good? On Oppor- tunities and Challenges of Large Language Models for Education. https://doi.org/10.35542/osf. io/5er8f Kohnke, N., Moorhouse, B. L., & Zou, D. (2023). Using ChatGPT for language teaching and learning. RELC Journal, 54(2), 537–550. https://doi.org/10.1177/00336882231162868 Kováčiková, E. (2020). English for specific purposes in higher education through content and language inte- grated learning. Cambridge Scholars Publishing. Kricke, M., & Reich, K. (2016). Teamteaching. Beltz. Kuckartz, U., & Rädiker, S. (2023). Qualitative content analysis: Methods, practice and software. SAGE. Lee, S.-M. (2022). L2 Learners’ strategies for using machine translation as a personalized writing tool. In J. Colpaert & G. Stockwell (Eds.), Smart CALL: Personalization, Contextualization, & Socialization (pp. 184–206). Castledown. 10.29140/9781914291012-pbk Lew, S., Yang, A. H., & Harklau, L. (2018). Qualitative Methodology. In A. Phakiti, P. I. de Costa, L. Plon- sky, & S. Starfield (Eds.), The Palgrave Handbook of Applied Linguistics Research Methodology (pp. 79–101). Palgrave Macmillan. https://doi.org/10.1057/978-1-137-59900-1 Lu, P., Qiu, L., Yu, W., Welleck, S., & Chang, K. W. (2022). A survey of deep learning for mathematical reasoning. arXiv:2212.10535. https://arxiv.org/pdf/2212.10535 Mishan, F. (2022). Language learning materials in the digital era. In J. Norton, & H. Buchanan (Eds.), The Routledge Handbook of Materials Development for Language Teaching (pp. 17–29). Routledge. Mroz, A. P. (2012). Process research screen capture. The Encyclopedia of Applied Linguistics, 1–7. https:// doi.org/10.1002/9781405198431.wbeal1450 Nerdynav. (2022, December 13). 73 Important ChatGPT Statistics & Facts For Mid Feb 2023 + An Infograph- ic – Nerdy Nav. Nerdynav. https://nerdynav.com/chatgpt-statistics/ 23 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 Pierce, J. (2018). Digital Divide. In R. Hobbs, & P. Mihailidis (Eds.) The International Encyclopedia of Media Literacy. John Wiley & Sons. https://doi.org/10.1002/9781118978238.ieml0052 Pineda, J. (2018). Problemlagen und Herausforderungen internationaler Studierender in Deutschland. Ergebnisse einer qualitativen Vorstudie im Rahmen des SeSaba-Projekts. https://www2.daad.de/me- dien/der-daad/analysen-studien/sesaba/vorstudie_pineda_2018.pdf Pineda, J., Kercher, J., Falk, S., Thies, T., Yildirim, H. H., & Zimmermann, J. (2022). Accompanying inter- national students in Germany to academic success: Results and recommendations from the SeSaBa project (DAAD Studien). Bonn. https://static.daad.de/media/daad_de/pdfs_nicht_barrierefrei/der- daad/analysen-studien/daad_sesaba_final_report.pdf Plevris, V., Papazafeiropoulos, G., & Rios, A. J. (2023). Chatbots put to the test in math and logic prob- lems: A preliminary comparison and assessment of ChatGPT-3.5, ChatGPT-4, and Google Bard. arXiv preprint arXiv:2305.18618 Richards, J. C., Platt, J. T., & Platt, H. (1999). Longman dictionary of language teaching and applied linguis- tics (New ed., 9. impr). Longman. Rocha, H. (2019). Mathematical proof: From mathematics to school mathematics. Phil. Trans. R. Soc. A 377: 20180045. http://dx.doi.org/10.1098/rsta.2018.0045 Rooks, R. N., Scandlyn, J., Pelowich, K., & Lor, S. (2022). Co-teaching two interdisciplinary courses in higher education. International Journal for the Scholarship of Teaching and Learning, 16(2), 1–8. https://doi.org/10.20429/ijsotl.2022.160208 Somekh, B. (2008). Action research. In L. M. Given (Ed.), The SAGE Encyclopaedia of Qualitative Research Methods. Volumes 1 & 2 (pp. 4–7). SAGE. Spannagel, C. (2023). Hat ChatGPT eine Zukunft in der Mathematik? Mitteilungen der Deutschen Mathe- matiker-Vereinigung, 31(3), 168–172. https://doi.org/10.1515/dmvm-2023-0055 UNESCO. (2023). Education 2030 agenda. https://www.unesco.org/en/digital-education/artificial-intel- ligence Wessels, D. (2023, January 27). Was ist ChatGPT und wie funktioniert es? – und welche ähnlichen Tools gibt es? [Video]. YouTube. https://youtu.be/cMuBo_rH15c?si=QuQ_oEgREPXVgTVw 24 Fleischhauer, Friedrich / Scripta Manent XIX/1 (2024), 4–24 Povzetek Dejavniki učinkovite priprave besedilnih nalog, generiranih s pomočjo umetne inteligence, pri poučevanju jezika za potrebe matematične stroke na visokošol- ski ravni Uporaba orodij umetne inteligence (UI) za generiranje učnih gradiv s strokovnimi vsebinam je vzbu- dila precejšen interes visokošolskih učiteljih tujega strokovnega jezika, saj pri jezikovnem pouku pra- viloma ne uporabljajo učbenikov. Gradiva velikokrat pripravljajo sami ali v sodelovanju s področnimi strokovnjaki, s katerimi so lahko vpeti v sodelovalno poučevanje. Priprava gradiv od učiteljev tujega strokovnega jezika zahteva veliko časa, ki bi ga lahko namenili drugim oblikam sodelovanja. Prednosti uporabe orodij UI, s katerimi si pomagajo pri pripravi matematičnih vsebin, zato med drugim vključu- jejo prihranek časa, boljše povezovanje z aktualnim dogajanjem v stroki in bolj raznolika učna gradiva za njihove študente. V prispevku proučujemo generiranje besedilnih nalog s pomočjo orodij UI pri poučevanju tujega stro- kovnega jezika za potrebe študentov matematike, ki so bili vpisani v prvi semester dodiplomskega štu- dijskega programa Naravoslovje, tehnika, inženiring in matematike (STEM). Ker so naloge del zaključ- nega izpita, vsako leto pripravimo nov nabor. Nedavna študija Lu in soavtorjev (2022) je pokazala, da je mogoče z UI uspešno pripraviti raznolike, primerne in uporabne matematične besedilne naloge, a zaznavamo odsotnost prispevkov, ki bi raziskali uporabnost besedilnih nalog, generiranih z IU, z vidika njihove učinkovitosti pri pouku tujega strokovnega jezika. V pričujoči raziskavi, ki uporablja načela akcijskega raziskovanja, smo s snemanjem računalniškega zaslona spremljali področne strokovnjake pri vsebinski analizi odziva UI na pripravo matematičnih besedilnih nalog, z njimi pa smo v nadaljevanju opravili še polstrukturirani skupinski intervju. Rezultati so pokazali, da so besedilne naloge, ki jih je ustvarila UI, opredelili kot načeloma uporabne, denimo z vidika tehnik oblikovanja pozivov UI (ang. prompting techniques), a smo kot pomembni omejitvi pre- poznali nenatančnost in nedoslednost. Na podlagi preliminarnih rezultatov smo v nadaljevanju osvet- lili, kako lahko uporaba UI prispeva k poučevanju tujih strokovnih jezikov, in predlagali vidike, ki naj jih upoštevajo nadaljnje raziskave. Ključne besede: učno gradivo, besedilne naloge, sodelovalno poučevanje, zajemanje zaslona, matematika 25 Scripta Manent XIX/1 (2024) DOI: 10.4312/SM.19.1.25-48 Research Paper Paper received: 15.06.2024 Paper revised: 10.08.2024 Paper accepted: 21.09.2024 Paper published: 30.11.2024 Barbara Majcenovič Kline Univerza v Mariboru, Slovenija Melita Koletnik Univerza v Mariboru, Slovenija POSREDOVANJE, PODPRTO Z ORODJI UMETNE INTELIGENCE, PRI POUČEVANJU TUJEGA JEZIKA STROKE – PRIMERA POUČEVANJA ANGLEŠČINE ZA PRIHODNJE ZDRAVSTVENE DELAVCE IN ZA PREVAJALCE Povzetek Skupni evropski jezikovni okvir (SEJO, 2001; slov. prevod, 2011) v svoji zadnji konceptualni različici (na- zadnje posodobljena v Dodatku k SEJO, 2018; slov. prevod, 2023) kot odmik od sheme štirih spretnosti k štirim načinom sporazumevanja poleg sprejemanja, tvorjenja in interakcije vključuje tudi posredova- nje (angl. mediation). V okviru akcijsko usmerjenega pristopa k učenju in poučevanju tujih jezikov (Svet Evrope, 2023, str. 23–34) pripisuje posredovanju tudi vse vidnejšo, samostojno vlogo, saj praviloma združuje elemente vseh sporazumevalnih načinov. (Medjezikovno) posredovanje kot eden od štirih načinov sporazumevanja je bistvenega pomena pri študiju angleščine kot tujega jezika stroke, ki v študijske procese vse pogosteje vključuje uporabo inovativnih tehnologij. Te se pri pouku tujih jezikov zdijo pričakovan pa tudi nujen korak v svetu, kjer je umetna inteligenca (UI) naš vsakdanji spremljevalec. Prednosti UI pri učenju in poučevanju tujih jezikov med drugim vključujejo neomejeno dostopnost in prilagodljivost vsebin in pristopov, takoj- šen odziv in povratne informacije, imajo pa tudi motivacijsko vrednost in omogočajo prilagajanje različnim psiho-socialnim potrebam. Prispevek združuje posredovanje z uporabo inovativnih tehnologij, kot so orodja umetne inteligence, pri pouku angleščine kot tujega jezika stroke (TJS) dveh skupin študentov: prihodnjih zdravstvenih delavcev in prihodnjih prevajalcev. Osredotoča se na orodja, ki podpirajo predvsem medjezikovno posredovanje, in predstavlja nekaj izbranih orodij (prevajalniki in jezikovni modeli UI), in sicer njihov potencial pri učenju in poučevanju TJS in s tem povezane težave. S prikazom dveh študij primera nazor- no povezuje teoretična izhodišča s praktično rabo UI. V ta namen sva zasnovali dve učni aktivnosti, ki vključujeta uporabo orodij UI za krepitev kompetenc medjezikovnega in medkulturnega posredovanja v najinih specializiranih kontekstih, tj. zdravstveni negi in prevajanju in tolmačenju. Kritična presoja in prizma etične refleksije nas nujno privedeta do spoznanja, da digitalne tehnologije in program- ske rešitve sicer optimizirajo in pospešijo delovne oziroma učne procese ter pogosto omogočajo višjo učinkovitost, vendar pa morajo medčloveški oziroma medosebni odnosi, medkulturni vidiki, dimenzije subjektivnosti, razumevanja drugega in kritičnega mišljenja ostati na prvem mestu. Ključne besede: posredovanje, medjezikovno posredovanje, umetna inteligenca, pouk angleščine, tuji jezik stroke, zdravstvena nega, prevajanje in tolmačenje 26 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 Abstract AI-Assisted Mediation in the Teaching of Languages for Specific Purposes – Case Studies in ESP for Future Healthcare Workers and Translators The Common European Framework of Reference for Languages (CEFR, 2001), in its latest conceptual version (last updated in the CEFR Companion Volume, 2018), replaced the traditional model of the four skills with the model of communicative language activities and strategies. In addition to reception, production, and interaction, these now include mediation as a distinct component. Within the CEFR’s action-oriented approach to language teaching and learning, mediation is recognized as an important mode of communication, which frequently combines the elements of reception, production and inter- action (Council of Europe, 2018, p. 35). (Interlingual) mediation is essential in studying English for specific purposes (ESP), too, where innova- tive technologies are increasingly integrated into the learning process. These technologies are seen as both intuitive and expected in language education, as well as a necessary step in a world where artifi- cial intelligence (AI) is a constant companion. The advantages of AI in language learning and teaching include, among others, unlimited accessibility, content and approach adaptability, instant responses, and feedback. They also offer motivational value and can cater to various psycho-social needs. The paper combines mediation with the use of innovative technologies, e.g., AI tools, in teaching ESP to two student groups: future healthcare professionals and future translators. It focuses on tools that primarily support interlingual mediation and presents selected tools (translation engines and AI lan- guage models), discussing their potential in learning and teaching ESP, as well as the related challeng- es. Through two case studies, the paper clearly links theoretical foundations with the practical use of AI. For this purpose, two learning activities were designed that involve the use of AI tools to enhance interlingual and intercultural mediation competencies in our specialized contexts, i.e., healthcare and translation and interpretation. Critical assessment and ethical reflection bring us to the conclusion that, while digital technologies and software solutions may optimize and accelerate work and learning processes, often achieving higher efficiency, interpersonal relationships, intercultural aspects, dimen- sions of subjectivity, understanding others, and critical thinking should remain paramount. Keywords: mediation, interlingual mediation, artificial intelligence, English language teaching, lan- guage for specific purposes, healthcare, translation and interpretation 27 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 1 Uvod Stopati v korak z razvojem komunikacijske tehnologije in vseh oblik umetne inteligence (UI) v kontekstu terciarnega visokega izobraževanja ni samo potreba, ampak predvsem nuja za vse udeležence tega procesa. Če upoštevamo, da je današnja mladina zelo vpeta v uporabo informacijsko-komunikacijske tehnologije, je to po eni strani izziv, po drugi pa priložnost, da izkoristimo potencial, ki jih nudijo te tehnologije in jih uvajamo v učenje in poučevanje tujih jezikov. To velja tudi za tuje jezike stroke (TJS), ki so prav tako povezani z razvojem informa- cijsko-komunikacijske tehnologije in UI ne samo na jezikovnem področju, temveč tudi pri spremljanju razvoja stroke. Kot navaja Unescovo Globalno poročilo o izobraževanju za leto 2023 o uporabi novih tehnolo- gij v izobraževanju, ima izobraževanje posebno nalogo, da se »popolnoma uglasi« s tveganji, ki jih prinaša uporaba UI – tako s tistimi, ki jih že poznamo, kot s tistimi, ki se šele nakazujejo (Unesco, 2023, str. 3–4). Med glavnimi sporočili Globalnega poročila je izpostavljeno, da je tehnologija v izobraževanju lahko vključujoča (milijonom omogoča izobraževanje) in izključu- joča (predvsem za najrevnejše), obenem pa izobraževalni sistemi težko dohajajo hiter tempo njenih sprememb. Globalno poročilo še dodaja, da primanjkuje nepristranskih dokazov o vplivu tehnologije na izobraževanje, saj se ta razvija hitreje, kot jo lahko evalviramo (Unesco, 2023). Kljub temu so prav empirični vpogledi v to, kako učitelji dejansko uporabljajo tehnolo- gijo, še zlasti pomembni, če jih želijo približati drugim učiteljem in jih usposobiti za delo z UI (Luckin in soavtorji, 2022). Ta prispevek prinaša tak vpogled v uporabo orodij umetne inteligence pri pouku angleškega jezika kot TJS pri pouku dveh skupin študentov: prihodnjih delavcev v zdravstveni negi in prevajalcev in tolmačev. Pri tem se osredotoča na dejavnost (medjezikovnega) posredovanja, pri katerem študenti pomagajo oblikovati ali prenesti pomen iz enega jezika v drugega, in predstavlja nekaj izbranih orodij (prevajalniki in jezikovni modeli UI), njihov potencial pri po- učevanju tujega jezika strok in s tem povezane težave. V članku uvodoma predstaviva in pojasniva temeljne koncepte in termine, ki jih uporabljava, ter podava kratek pregled najnovejše literature na obravnavanih področjih. Teoretični okviri, ki jih podajava, omogočajo globlje razumevanje obravnavane tematike in njeno uvrstitev v širši kontekst. Nato opredeliva pomen in vlogo UI v okviru orodij za (medjezikovno) posredovanje ter predlagava primera, kako jih je mogoče vključiti v poučevanje tujega jezika stroke. Kot znanstveno raziskovalno metodo uporabljava deskriptivno študijo primera, ki vključuje natančno in podrob- no preučitev preučevanega področja in se pogosto uporablja v družbenih vedah in humanistiki. Na podlagi pregleda literature slednjič izpeljeva smernice za učitelje za vključevanje tehnolo- gije v pouk ter nasvete, kako lahko izboljšamo učinkovitosti poučevanja tujega jezika ali TJS s pomočjo orodij UI. 2 Razumevanje posredovanja in medjezikovnega posredovanja, raznojezičnosti in raznokulturnosti Skupni evropski jezikovni okvir (SEJO, 2001; slov. prevod, 2011) v svoji zadnji konceptualni različi- ci (nazadnje posodobljena v Dodatku k Skupnemu jezikovnemu okvir, 2018; slov. prevod, 2023) 28 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 kot odmik od sheme štirih spretnosti k štirim načinom sporazumevanja poleg sprejemanja, tvorjenja in interakcije vključuje tudi posredovanja [angl. mediation]. V okviru akcijsko usmer- jenega pristopa k učenju in poučevanju tujih jezikov mu SEJO (2023) pripisuje tudi vse vidnejšo vlogo, saj pogosto združuje elemente vseh (drugih) sporazumevalnih načinov (tj. sporazume- valnih jezikovnih dejavnosti in strategij). Posredovanje pomeni, poenostavljeno, omogočanje sporazumevanja med posamezniki v situacijah, kjer bi bilo to sicer težko, tj. z vidika okoliščin, ali težavno, tj. z vidika tistega, ki posreduje (Schmidhofer, Cerezo in Koletnik, 2022), kljub oboje- stranskemu trudu, ki ga sogovorci vlagajo v uspešnost komunikacije. V primerjavi s sprejema- njem, tvorjenjem in interakcijo, kjer kot odziv na nekaj, kar je učenec1 slišal, prebral ali videl, pri pouku nastajajo praviloma nove vsebine, gre pri posredovanju za … »obdelavo [angl. pro- cessing] in prilagajanje že obstoječih vsebin« (University of Cambridge, b. d.2). Posredovanje, kot ga opredeljuje Dodatek k SEJO (Svet Evrope, 2018; slov. prevod, 2023), sicer ni omejeno zgolj na posredovanje med (dvema ali več) jeziki ali na medjezikovno posredova- nje. V sodobnem svetu, ki ga zaznamujeta razmah informacij in vse hitrejši tempo bivanja, omogoča tudi ljudem, ki govorijo isti jezik, (boljše) razumevanje besedil in konceptov, torej izboljšano sporazumevanje nasploh. Kot poudarja Kerr (English with Cambridge, 2022 in Kerr 2022), smo posredniki vsi, tudi učitelji, ki vsebine povezujemo s predhodnim znanjem učen- cev, prilagajamo jezik sporazumevanja njihovi stopnji izobraževanja in razumevanja, saj zanje razčlenjujemo zapletene informacije in jih po potrebi razširimo ali okleščimo (kar Dodatek k SEJO imenuje »posredovalne strategije«): t. i. procesi potekajo ne glede na to, v katerem jezi- ku se odločimo sporazumevati. V globaliziranem večjezičnem in večkulturnem okolju je medjezikovno posredovanje nujnost. Poleg zgoraj navedenih strategij Dodatek k SEJO kot dejavnost vključuje tudi elemente pre- vajanja, tj. govornega in/ali pisnega posredovanja sporočil iz tujega jezika v maternega in/ ali obratno, ustvarjanja pogojev za sporazumevanje in sodelovanje, soočanje z občutljivimi situacijami in njihovo blaženje, ki od udeležencev sporazumevanja poleg jezikovne zahteva- jo še socialne, raznokulturne in raznojezične zmožnosti (Svet Evrope, 2023). Medjezikovno posredovanje se torej lahko razume tudi kot »posebna kompetenca v komunikaciji dvo- in večjezičnih učencev« (Buchwald in soavtorji, 2019, str. 2). V izogib nesporazumom naj opozoriva, da SEJO razlikuje med »večjezičnostjo«, ki jo opre- deljuje kot soobstoj različnih jezikov predvsem na družbeni ravni, in »raznojezičnostjo«, kjer gre za dinamičen in nenehno spreminjajoč se individualni jezikovni repertoar posameznika. Koncepta raznojezičnosti in raznokulturnosti sta bila razvita kot »oblika dinamičnega ustvar- jalnega procesa »jezikovnega prehajanja« [angl. languaging] med jezikovnimi različicami, kot metodologija in kot cilj jezikovne politike« (Svet Evrope, 2023, str. 27). Beacco in soavtorji (2016) o tem zapišejo: Raznojezična zmožnost je opredeljena kot sposobnost uporabe obsežnega reper- toarja jezikovnih in kulturnih virov za zadovoljevanje sporazumevalnih potreb v in- terakciji z ljudmi iz drugih okolij in kontekstov, ki se med sporazumevanjem bogati. Repertoar sestavljajo viri, ki so jih posamezni učenci pridobili v vseh jezikih, ki jih 1 V prispevku uporabljava za učečo se osebo poimenovanje »učenec«, in sicer predvsem v teoretičnem delu in pri dobesednih citatih, ter »študent«, ko gre za najina specializirana konteksta. 2 Vse prevode neposrednih citatov v dokumentu sva pripravili avtorici. Prav tako se uporabljene moške slovnične oblike nanašajo na oba oziroma vse spole. 29 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 znajo ali so se jih (na)učili, in ki se nanašajo tudi na kulture, povezane s temi jeziki [...] V središču večjezične perspektive so učenci in razvoj njihovega individualnega repertoarja (str. 20). Pomembno je, da je v jedru obeh učenec kot posameznik, ki pri sporazumevanju za ustvarja- nje pomena uporablja vse svoje razpoložljive (jezikovne, kognitivne, semiotične itd.) vire in se prožno opira na ves svoj raznokulturni in raznojezični repertoar. Pri tem prestopa meje med »nacionalnimi« jeziki in njihovimi različicami ter med jeziki in drugimi semiotičnimi sistemi (García in Lin, 2017). Verjetno najbolj viden koncept, ki je tesno povezan z raznojezičnostjo, je »medjezikovno pre- hajanje« [angl. translanguaging], ki ga SEJO opredeljuje kot »ravnanje raznojezičnih oseb v situacijah, v katerih se pojavlja več jezikov« (Svet Evrope, 2023, str. 28). S pedagoško-didak- tičnega zornega kota je medjezikovno prehajanje uporaba pristopov, katerih cilj je izkoristiti prakse prožnega oblikovanja pomena za podporo učenju. Pedagoško medjezikovno prehaja- nje »se lahko nanaša na uporabo različnih jezikov za sprejemanje in/ali tvorjenje ali na druge načrtovane strategije, ki temeljijo na uporabi virov učenčevega celotnega jezikovnega reper- toarja« (Cenoz, 2017, str. 194). Vsi zgoraj omenjeni koncepti (medjezikovno posredovanje, medjezikovno prehajanje, razno- jezičnost, raznokulturnost, itn.) so prisotni v pedagoškem procesu in najinem delu s študenti. Ena od avtoric namreč poučuje angleščino na Fakulteti za zdravstvene vede v Mariboru (pro- gram Zdravstvena nega), druga poučuje jezik in prevajanje na Filozofski fakulteti v Mariboru (program Prevajalske študije). Oboje sodi v domeno učenja oziroma poučevanja tujega jezika stroke, zato naj na tem mestu na kratko osvetliva še termin »(tuji) jezik stroke, TJS« [angl. Lan- guage for Specific Purposes, LSP] ali »(tuji) strokovni jezik« ali »tuji jezik za potrebe strok(e)«3. Na kratko TJS opredeljujeva kot specializiran diskurz, ki zadovoljuje specifične sporazume- valne potrebe določenih jezikovnih uporabnikov na specializiranem področju. TJS pojmujeva torej precej široko, obenem pa dovolj ozko, da ne sodi več v okvir v tujega jezika za potrebe splošnega sporazumevanja. Kot za angleščino, ki je sploh najbolj preučevani jezik na tem področju, pravi Hyland (2019), se TJS nanaša na poučevanje in učenje jezika na posebnih področjih uporabe [angl. domains of use]. Pri tem se učitelj osredotoča na specifične sporazumevalne potrebe posebnih sku- pin, sporazumevanje pa je odvisno od specializiranega znanja. Gandin (2016) navaja, da ima- jo TJS »specifične leksikalne, skladenjske, morfološke, slogovne in pragmatične značilnosti, ki se uporabljajo predvsem (ali skoraj izključno) v specializiranih komunikacijskih situacijah« (str. 6). Van Dijk (2003) meni, da je najbolj očitna raba terminologije, vendar je treba upošte- vati še več drugih vidikov, npr. »prednostne teme, splošno obliko ali shemo besedila, slog, re- toriko (vključno z zanjo značilnimi metaforami), vzorce argumentacije, metode dokazovanja« in celo nebesedne elemente in značilnosti, kot so uporaba tabel in slik (str. 22). Z vidika poučevanja Dudley-Evans in St. John (2002) poudarjata, da mora poučevanje za pot- rebe TJS upoštevati in odražati med drugim tudi metodologijo disciplin oziroma področij, na katerih bodo učenci jezik uporabljali, pri čemer se (lahko) zelo razlikuje od pouka »splošnega« 3 V slovenskem jeziku se po pribl. 15–20 let uporabe v sicer maloštevilnih člankih v slovenščini ustreznica za TJS še ni ustalila, čeprav v okviru terminološke sekcije Slovenskega društva učiteljev tujega strokovnega jezika (SDUTSJ) potekajo prizadevanja v tej smeri. 30 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 jezika oziroma jezika za splošne potrebe. Kot dodajajo Cerezo Herrero, Schmidhofer in Kole- tnik (2021), se učitelji TJS pri tem osredotočajo na potrebe učencev, tj. na vrzel med njihovim trenutnim in želenim (ciljnim) jezikovnim znanjem ter prihodnjo rabo jezika, skupaj s specifič- nimi jezikovnimi sredstvi in dejavnostmi, ki jih bodo uporabljali v svojem poklicnem življenju. Posebno pozornost je pri tem treba nameniti tudi učiteljem TJS, saj »večina učiteljev TJS v mednarodnem prostoru poučuje z zelo malo ali celo povsem brez začetnega izobraževanja« (Podgoršek in soavtorji, 2021, str. 95), njihovo izobraževanje in nenehno poklicno usposa- bljanje pa je nujno za kakovostno poučevanje TJS (več o tem v Jurkovič in soavtorji, 2023, ter Jurkovič, Mertelj in Podgoršek, 2024). 3 Poučevanje in učenje TJS v dobi tehnološkega napredka, umetne inteligence in globalnih skupnosti Uvajanje sodobne tehnologije v visokošolski študijski proces je postalo nujnost, saj je, kot na- vaja Alič (2023), »[t]ehnološki napredek preoblikoval naš način življenja, dela in komunikacije. Prav je, da se odraža tudi v načinu, kako se izobražujemo« (str. 10). Sodobne tehnologije med drugim omogočajo bolj dinamično poučevanje, ki prispeva k motivaciji študentov, obenem pa jih pripravljajo na delo v tehnološko vse bolj zahtevnih okoljih, kjer bosta poznavanje in kompetentna uporaba teh orodij pogoj in ne zgolj zaposlitvena prednost. Učitelji smo sicer na tem področju pogosto v situaciji, ko je naše znanje korak za znanjem učencev in sami po svojih najboljših močeh iščemo načine, kako se v čim krajšem času čim bolje seznaniti s čim več uporabnimi orodji. Kot namreč trdi Kirovska-Simjanoska (2020), so življenja študentov TJS nenehno vezana na uporabo različnih digitalnih naprav in so zato močno odvisni od teh- nologije. Torej, nadaljuje Kirovska-Simjanoska, »[p]osledično študenti prinesejo s seboj svoje tehnološke izkušnje, kot tudi njihova prepričanja in dojemanja, da tehnologija mora imeti pomembno vlogo tudi v njihovem izobraževalnem procesu« (str. 424). Nedavni izziv – vendar prav zagotovo ne zadnji – je umetna inteligenca. Ravnovesje med upo- rabo UI kot načinom dopolnjevanja tradicionalnega izobraževanja oziroma podporo dosega- nju zastavljenih izobraževalnih ciljev in uporabo UI, ki ni osmišljena ali je zgolj šibko smiselno vpeta v učne cilje (torej je predvsem sama sebi namen), je zapleteno in zahteva temeljit raz- mislek ter skrbno načrtovanje, preden se UI na kakršenkoli način uporabi. Ko sva orodje umetne inteligence samo vprašali, kakšne so prednosti uporabe UI pri po- učevanju (OpenAI, 2024),4 je med drugim navedlo individualno prilagojeno in prilagodljivo učenje, ki vključuje tudi neomejeno dostopnost in prilagodljivost vsebin in pristopov,5 takoj- šen odziv in povratne informacije, imelo pa naj bi tudi motivacijsko vrednost in omogočalo prilagajanje različnim psihosocialnim potrebam učencev. Med pomanjkljivostmi je naštelo izgubo stika oziroma interakcije med ljudmi ter preveliko zanašanje na tehnologije in nasploh »razčlovečenje« poučevanja, izkrivljeno predstavljanje informacij, diskriminacijo in nekatera druga etična vprašanja (npr. ali lahko algoritmi odločajo o učnih ciljih, neenakost pri dostop- nosti do orodij). Ta seznama bi lahko na pozitivni strani razširili s hitrostjo, s katero so na voljo informacije, ki jih ponujajo orodja UI, ter na negativni strani med drugim s pogosto precej 4 ChatGPT je na poziv [angl. prompt] pripravil tudi citatni vnos. 5 Sal Khan, ustanovitelj spletne učne platforme Khan Academy, ki med drugim uporablja tudi posebej v ta namen razvit chatbot, Khanmigo, jih predstavlja v svojem TED-govoru (Khan, 2023). 31 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 prepričljivimi napakami in občasno celo »halucinacijami« ali izmišljenimi informacijami, ki jih proizvajajo orodja UI, še posebej veliki jezikovni modeli.6 Slednjič učitelji opozarjajo na to, da lahko orodja UI za učence predstavljajo tudi precejšnjo »skušnjavo« in možnost za goljufanje, ki ji le s težavo pridejo na sled, seznam pa še zdaleč ni popoln in se dnevno daljša. Zaradi globalizacije je znanje tujih jezikov in posredovanje med njimi ter med kulturami vse bolj postavljeno v ospredje. To velja tudi za TJS. Skupnosti namreč pospešeno postajajo glo- balne in za »uspešno delovanje v večjezični skupnosti je medjezikovno in medkulturno posre- dovanje (prevajanje, tolmačenje, povzemanje itn.) nujno potrebno in bistvenega pomena […]« (Mikolič Južnič in soavtorji, 2014, str. 636). Na področju zdravstvene nege se zavedamo po- večane potrebe po znanju in zmožnostih medjezikovnega in medkulturnega posredovanja zaradi povečanega števila tujcev, ki potrebujejo ali iščejo pomoč v zdravstvenih ustanovah (turisti, ki počitnikujejo pri nas, poslovneži, ki so pri nas zaradi službe, študenti na izmenjavah itn.). Iz izkušnje ene od avtoric, ki sem v čakalnici Urgence v UKC Maribor preživela štiri ure in opazovala dinamiko prihajanja in sprejemanja pacientov, sem se o tem lahko prepričala sama, saj je v tistem času pet tujcev iskalo zdravstveno pomoč, pri čemer so se morale me- dicinske sestre na sprejemu zelo potruditi, da so na učinkovit način uporabile znanje tujega jezika ob tankočutnem upoštevanju tako zdravstvenega stanja kot tudi kulturnega ozadja teh pacientov. Zmožnost (kompetenca), ki so jo medicinske sestre v zgornjem primeru potrebovale in uporabljale, je bila dvojna: a) posredovanje besedil (prenašanje, pojasnjevanje in zgošče- vanje specifičnih podatkov iz jezika A v jezik B ter nazaj v jezik A) in b) posredovanje pri sporazumevanju (spodbujanje raznokulturnega prostora in prevzemanje vloge posredni- ka).7 Uspešno so se spopadale z jezikovnimi in kulturnimi vrzelmi tujcev, da so ti prišli do svojega komunikacijskega cilja: zagotovitve zdravstvenih storitev. Pri tem so uporabljale posredovalne strategije: informacije so povezovale s predhodnim znanjem (npr. navajale so zglede) in prilagajale svoj jezik ravni svojih komunikacijskih partnerjev in situaciji, razčle- njevale so zapletene informacije, prilagajale dolžino povedanega in podobno. Pri tem se je kot lingua franca uporabljal angleški jezik, saj ta ni bil jezik A ne njih in ne pacientov, ki so jim pomagale. Podobno velik pomen ima posredovanje (v skladu s Sejo 2023) pri učenju tujega jezika za potrebe prihodnjih prevajalcev in tolmačev, saj predstavlja bistvo njihovega prihodnjega strokovnega dela. Dodatek k SEJU (Svet Evrope, 2023, str. 91) v poglavju o posredovanju sicer izrecno poudarja, da so »zmožnosti in strategije prevajanja in tolmačenja […] povsem drugo področje« od jezikovnih zmožnosti, ki jih lahko študent na tem področju naredi v neformalnih, vsakodnevnih situacijah, a je mogoče z vključevanjem posredovanja v pouk tujega jezika prihodnjih prevajalcev in tolmačev že zelo zgodaj ustrezno »naravnati« nji- hovo poučevanje in pri tem prihraniti precej časa pri razvijanju prevajalskih in tolmaških zmožnosti. Tudi tukaj imajo veliko vlogo novi, sodobni načini poučevanja, ki so tehnološko kar se da napredni in ponujajo podporo učiteljem pri poučevanju kot tudi študentom pri usvajanju znanja. 6 Na podlagi študije Magesh in soavtorji (2024), ki je ocenila zanesljivost nekaterih vodilnih orodij UI za upo- rabo v pravu, naj bi stopnja haluciniranja za nekatera med njimi dosegala tudi več kot 30 %. 7 Za podrobnejše opisnike glej SEJO Dodatek (Svet Evrope, 2023), Poglavje Posredovanje. 32 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 4 Tehnologije in TJS Hkrati z globalizacijsko pogojenimi več- in raznojezičnimi ter več- in raznokulturnimi vidiki, ki naj bi jih učitelji približali svojim učencem, pa se poučevanje tujega jezika spreminja tudi zaradi dejstva, da učitelji želijo (in so včasih kar prisiljeni) znanje prenašati in razvijati na nove načine. Računalniško posredovano poučevanje in učenje jezikov premošča vrzel med priča- kovanji učencev in potrebami globalne dobe. Če želijo učitelji učence voditi skozi vedno nove tehnološke pripomočke in orodja ter jih obenem usposobiti v njihove kompetentne uporab- nike, to od njih zahteva, da spremljajo novosti in spremembe ter upoštevajo njihov vlogo in pomen tudi pri medjezikovnem in medkulturnem posredovanju. Digitalna pismenost je danes za učitelja TJS torej nuja, saj lahko le tako sledi potrebam današnjih učencev in razvoju digitalne tehnologije. V svoji raziskavi, ki jo je izvedla med učitelji angleščine kot jezika stroke [angl. English for Spe- cific Purposes, ESP], je Benabdallah (2023) ugotavljala, v kakšne namene učitelji uporabljajo UI. Rezultati so pokazali, da anketiranci najpogosteje uporabljajo UI za povzemanje besedil (18 %), prilagajanje učbenikov [angl. personalized textbooks] (12 %), ustvarjanje jezika [angl. language generating] (8 %), parafraziranje (16 %), prevajanje (28 %), ocenjevanje učencev in vrednotenje njihovega dela [angl. evaluate and assess learners] (30 %) ter za ocenjevanje izpitov (14 %). Smernice za uporabo in vključevanje tehnologije v učenje in poučevanje TJS je zaradi njenega hitrega razvoja nehvaležno in nemogoče napovedati. Trenutni trendi nakazujejo, kot med an- ketiranimi učitelji TSJ ugotavlja Benabdallah (2023), da bi bila UI najkoristnejša pri ocenjeva- nju in vrednotenju (20 %), načrtovanju celotnega učnega procesa [angl. course design] (18 %), prepoznavanju in analizi potreb (16 %) ter pripravi učnih načrtov (10 %). Za potrebe jezikovne industrije (torej tudi prevajalcev in tolmačev) spletna stran nimdzi8 vsa- ko leto sestavlja »tehnološki atlas«, v katerem poskuša čim bolj izčrpno zajeti in predstaviti specializirana orodja, namenjena prevajalcem, tolmačem in sorodnim poklicem, ter splošna tehnološka (jezikovna) orodja, s katerimi se utegnejo srečati v poklicnem življenju. Čeprav po lastnih podatkih niso (in ne morejo) zajeti vseh, je bilo v letu 2023 v atlas vključenih več kot 920 orodij, razvrščenih v naslednjih deset kategorij: sistemi za upravljanje prevodov [angl. translation management systems], sistemi upravljanja prevajalskih storitev [angl. translation business management systems], tolmaški sistemi, sistemi za prepoznavo govora, sistemi za avdiovizualno prevajanje, strojni prevajalniki, integrirani sistemi, sistemi za nadzor nad kako- vostjo, platforme in spletne tržnice in sistemi za generiranje večjezičnih vsebin [angl. multilin- gual content generators]. Kategorijo sistemov za generiranje večjezičnih vsebin so v atlas dodali v letu 2023, vključuje pa splošna UI orodja, kot sta ChatGPT podjetja OpenAI in Googlov Bard (sedaj Gemini), in specializirana UI orodja, ki na primer podpirajo pisanje besedil, kot so AI Writer, Writeseed, Writesonic, Magic Write ali Ailaysa Writer. S prihodom ChatGPT novembra 2022 in podobnih velikih jezikovnih modelov [angl. Large Language Models, LLMs], ki so namenjeni verjetnostnemu modeliranju jezika, so jezikovne 8 https://www.nimdzi.com/language-technology-atlas/ 33 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 tehnologije v sodobni družbi dobile novo vlogo, povečal pa se je tudi strah, kaj utegnejo po- meniti za prihodnost številnih poklicev, ne zgolj jezikovnih. Z razmahom jezikovnih tehnologij, ki uporabljajo UI, zlasti generativnih orodij UI, je oblikovanje večjezičnih vsebin postalo sploš- no dostopno. Kot navajajo na spletni strani nimdzi, lahko generativna orodja UI ustvarjalcem vsebin pomagajo pripraviti prvo različico besedila, to prilagoditi želenemu žanru in slogu, lahko ga jezikovno pregledajo in še veliko več. Po zadnjih dostopnih podatkih iz januarja 20249 ChatGPT podpira več kot 8010 jezikov, kar pomeni, da lahko poleg pisanja vsebine uporabniki vsebine ustvarjajo, obdelujejo in prede- lujejo, po potrebi hkrati v več jezikih, tudi v slovenščini. Načeloma je ChatGPT na voljo v dveh različicah: brezplačni ChatGPT 3.5, ki se ne posodablja več od septembra 2021, in plačljiva verzija/različica ChatGPT 4, ki je bila nazadnje posodobljena decembra 2023. Med prednosti plačljive različice sodi predvsem večja zasebnost, omogoča pa tudi nekatere dodatne funkci- onalnosti, kot je denimo generiranje slik in grafik. Zadnja sprememba se je zgodila v začetku maja 2024, ko je OpenAI predstavil novo različico z imenom GPT-4o (4 omnia), ki je prinesla znatno izboljšanje za vse uporabnike ChatGPT, tudi tiste, ki uporabljajo brezplačno različico. Za didaktične namene dveh skupin specializiranih jezikovnih uporabnikov, ki jih predstavljava v nadaljevanju, sva uporabili brezplačno različico ChatGPT 3.5. Ko gre za orodja UI, ki so uporabna za medicinske namene, obstaja kar nekaj aplikacij, ki so prosto dostopne tudi širši, laični javnosti in jih pogosto uporabljamo v vsakdanjem življenju. To so zdravstvene aplikacije za mobilne naprave, ki so namenjene spremljanju telesne aktivnosti in vadbe in jih redno uporablja že skoraj vsak posameznik, na primer Fitbit ali MyFitnessPal. Nadalje so na voljo orodja, ki pomagajo pri prepoznavanju simptomov in postavljanju možnih diagnoz na podlagi podanih simptomov. Sicer pa bolj specializirana orodja UI, ki jih upora- bljajo strokovnjaki, nudijo pomoč pri elektronski hrambi in urejanju podatkov pacientov, pri čemer se poraja veliko etičnih dilem, ki jih avtorici obravnavava v šestem poglavju. Za medicinske namene ChatGPT po lastnih podatkih (OpenAI, 2024) poleg zgoraj omenjenih ponuja še možnost dela z radiološkimi in obdelovalnimi informacijskimi sistemi, na primer PACS [angl. Picture Archiving and Communication System] ter Siemens Healthineers syngo.via, ki jih zdravniki in radiologi uporabljajo za ogled in analizo rentgenskih, MR- in CT posnetkov. Zdravstveni delavci se lahko nadalje poslužujejo specializiranih sistemov za upravljanje z od- merki zdravil, ki poleg naprav za avtomatsko izdajanje zdravil ponujajo tudi povezane aplikacije za spremljanje odmerkov in opomnike. Medicinske sestre lahko obenem z zgoraj omenjenimi orodji UI uporabljajo tudi sisteme, ki jim omogočajo avtomatizacijo rutinskih nalog (npr. obve- ščanje o urnikih bolnikov, opomniki za jemanje zdravil, spremljanje vitalnih znakov, ipd.). V pedagoške namene na področju zdravstva (torej za izobraževanje in usposabljanje prihod- njih zdravstvenih delavcev) se uporabljajo simulacijski programi in interaktivni učni vmesniki, ki poleg e-izobraževanja omogočajo delo z virtualnimi bolniki in različnimi medicinskimi sce- nariji. Pomoč pri medjezikovnem in medkulturnem sporazumevanju jim nudijo tudi chat boti (t. i. klepetalni roboti, pogovorni roboti, klepetalniki ali navidezni asistenti), ki olajšajo spora- zumevanje med medicinsko sestro in pacientom v smislu osnovnega svetovanja o zdravju, 9 Podatki so bili nazadnje preverjeni ob oddaji članka v uredniški postopek junija 2024. 10 https://www.tomedes.com/translator-hub/chatgpt-language-capabilities, na nekaterih spletnih straneh sicer navajajo, da naj bi bilo podprtih jezikov več kot 100. 34 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 zdravilih, postopkih in prepoznavanju simptomov, vendar le do mere, ki je še v pristojnosti medicinske sestre. Tovrstna tehnologija v zdravstvu lahko optimizira poslovanje, skrajša čakalne vrste, zniža stroške in posledično izboljša kakovost in varnost oskrbe. Rožanec in Lahajnar (2019) povze- mata Microsoftovo vizijo našega vsakdana, v katerem bodo pametne naprave (osebni digital- ni asistenti) nadzorovale naše zdravje: Nosljiva naprava bo izmerila krvni tlak, analizirala koncentracijo kisika in druge para- metre ter rezultate poslala zdravniku, ki bo med klicem analiziral podatke. Umet- na inteligenca, vgrajena v pametno rešitev, bo zdravniku pomagala pri diagnostiki in predpisovanju pacientu prilagojene terapije. V nekaj urah bo brezpilotni letalnik dostavil zdravilo, na katerega jemanje nas bo opozarjal osebni digitalni asistent. Ta bo tudi spremljal napredek, občasno obveščal zdravnika in če ne bo ustreznega iz- boljšanja, tudi zaprosil za dovoljenje, da rezervira pregled pri zdravniku. Osebni digi- talni asistent bo spremljal in načrtoval tudi rutinske preglede, cepljenja in različna testiranja ter poskrbel za pravočasno naročanje pri zdravniku (str. 46). Razmah sodobnih jezikovnih in drugih tehnologij, ki so vse pogosteje podprte z najnovejšo različico UI, torej poteka z nesluteno hitrostjo, v prihodnosti pa lahko pričakujemo, da se bo le še pospešil. Kot sem ena izmed avtoric ugotavljala že leta 2011, ko sem raziskovala upora- bo prosto dostopnega prevajalnika Google Translate med študenti programa Medjezikovnih študij, »tehnologija v današnjem svetu ni le možnost, ampak nuja« (Koletnik Korošec, 2011, str. 17), zato sem menila, da bi jih morali učitelji upoštevati pri pripravljanju didaktičnih dejav- nosti. Opozorila sem na spremenjeno vlogo učiteljev, ki bomo oziroma smo že prisiljeni svojo vlogo »podajalcev« znanja zamenjati z vlogo »sooblikovalcev« znanja. Spremenila se bo oziroma se je že tudi vloga študentov, saj niso več zgolj v vlogi »sprejemni- kov« znanja, temveč pri sooblikovanju znanja delujejo proaktivno in odgovorno. Pri tem je pomembno, kot ugotavljata Drašler in Podgoršek (2022), da so študenti vključeni v načrtova- nje in oblikovanje učnega procesa, kar med drugim pomeni tudi načrtovanje in oblikovanje učnega gradiva. Več kot deset let po raziskavi iz 2011 sva sicer obe soavtorici še vedno prepri- čani v potencial in neizogibnost tehnologije, zato v nadaljevanju predstavljava dva primera didaktičnih aktivnosti, ki vključujejo uporabo orodij UI pri pouku angleščine kot TJS s ciljem krepitve kompetenc medjezikovnega in medkulturnega posredovanja v najinih specializira- nih kontekstih: a) zdravstveni negi in b) prevajanju in tolmačenju. 5 Študiji primerov z integriranimi orodji UI v procesu poučevanja in učenja TJS s poudarkom na medjezikovnem in medkulturnem posredovanju 5.1 Primer 1 – zdravstvena nega Prva študija primera prihaja s področja poučevanja angleščine za študente zdravstvene nege. Študenti Fakultete za zdravstvene vede Univerze v Mariboru, smer Zdravstvena nega, se z angleščino kot TJS (Tuji jezik – angleščina) srečajo v 1. letniku.11 Izbirni predmet (izbirajo med 11 Nazadnje v štud. letu 2023/2024. 35 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 dvema tujima jezikoma, angleščino in nemščino) obsega 45 kontaktnih ur (za izredne študen- te 23). Poleg klasičnih načinov poučevanja se pri tem predmetu ena od avtoric poslužujem med drugim Mentimetra,12 Kahoota,13 Canve,14 Kiala15 in prevajalnika DeepL Translate.16 Kljub temu, da so uporabe zgoraj naštetih aplikacij študenti vajeni in so jih sprejeli kot del procesa učenja, je opaziti upad interesa: zdi se, kot da so dela s temi orodji naveličani. Da bi ohranila motivacijsko vrednost za posredovanje kot eno izmed štirih sporazumevalnih jezikovnih de- javnosti, in z zavedanjem, da znanje TJS bodoči zdravstveni delavci pojmujejo bolj kot nepo- trebno obvezo kot koristno kompetenco, se kot učiteljica TJS zato vedno znova poslužujem uporabe novih spletnih orodij. Ob tem se zavedam tveganj, pomanjkljivosti in drugih negativ- nih vidikov, poleg tega pa je vsaka uporaba nove tehnologije UI v današnjem hitro spreminja- jočem se okolju za učitelja dokajšen organizacijski zalogaj. Da bi pouk angleščine kot TJS še bolje prilagodila potrebam svojih študentov, v razredu upo- rabljam tudi specializirano platformo Glass,17 ki temelji na UI za upravljanje medicinskega znanja. Glass pomaga pripraviti diferencialno diagnozo za kompleksne bolnike, pri čemer analizira povzetek pacientovih podatkov in predlaga diagnoze, ki se lahko raziščejo. Uporaba te programske opreme za bodoče medicinske sestre in zdravstvene tehnike je lahko koristna v smislu kritičnega razmišljanja in pridobivanja besedišča v TSJ, za pojme, ki jih poznajo v prvem jeziku. Pomemben je tudi komunikacijski vidik ter nadgrajevanje čustvene inteligence skozi učenje TJS z uporabo UI. Ena od aktivnosti, ki se izvaja na začetku semestra, je povezana s pridobivanjem bolnikove anamneze. Naloga medicinske sestre18 je, da ob sprejemu spozna pacienta in pridobi na- tančno anamnezo. Pri tem se za potrebe vaje uporablja obrazec, ki se nahaja v gradivu (Allum in McGarr, 2008, str. 8). Naloga študentov je, da v paru odigrajo vlogo diplomirane medicin- ske sestre oziroma diplomiranega zdravstvenega tehnika – ta postavlja vprašanja, s katerimi pridobiva informacije, ki so potrebne, da se izpolni dani obrazec – in vlogo pacienta, ki na ta vprašanja odgovarja. Glavni pravili sta: a) vsak par razmisli o zdravstveni težavi (stanju, bolez- ni), zaradi katere je eden od njiju, torej pacient, prišel v bolnišnico na sprejem, in b) odgovora »ne« pacient v dialogu ne sme uporabiti (npr. »Ste imeli v preteklosti kakšne operacije?« »Ne.« ali »Ali imate kakšne alergije?« »Ne.«). Ko spraševalec obrazec izpolni z odgovori pacienta, študenta vlogi zamenjata. V razredu nato sledi predstavitev parov in njihove komunikacije pri pridobivanju anamneze. Glede na čas, ki so ga študenti potrebovali za prvo aktivnost to pomeni, da se bodisi predstavijo vse dvojice bodisi le nekaj naključno izbranih.19 Vsi študenti poslušajo, spremljajo predstavljene dialoge in si hkrati zapisujejo svoja mnenja in predloge: 12 https://www.mentimeter.com/ Mentimeter je spletna storitev, ki omogoča takojšnje pridobivanje mnenj ozi- roma povratnih informacij udeležencev, kar lahko vodi v takojšnjo ustrezno reakcijo učitelja glede na izid. 13 https://kahoot.it/ Kahoot je spletna aplikacija, ki se uporablja kot popestritev učenja tudi na terciarni ravni izobraževanja. 14 https://www.canva.com/sl_si/ Canva je aplikacija, ki med drugim omogoča grafično oblikovanje predstavitev. 15 https://www.kialo-edu.com/ Kialo edu je spletna aplikacija, ki se lahko uporablja tudi za diskusije, kjer se jasno izriše osnovna teza in nato se kot veje razvejajo mnenja »za« in »proti«, kar omogoča jasnejši pregled nad argumenti. 16 https://www.deepl.com/translator 17 https://glass.health/ 18 Izraz »(diplomirana) medicinska sestra« se nanaša tudi na moško poimenovanje »(diplomirani) zdravstvenik«. 19 Kljub temu da na začetku dobijo navodila in časovno omejitev, je aktivnost zasnovana dovolj prilagodljivo, da delo študentov spremljam in dovolim, da aktivnost zaključijo tudi tisti z nekoliko šibkejšim znanjem, ki so morda počasnejši. 36 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 kaj je bilo po njihovem mnenju zelo dobro, kaj se je izvedlo odlično in kaj bi se dalo še izboljša- ti, spremeniti ali popraviti. Ta mnenja študenti podajo in prediskutirajo po vsaki predstavitvi ali na koncu, odvisno od skupine in razpoložljivega časa. Naslednji korak je, da vsak par v aplikacijo Glass vnese svojo zgoraj omenjeno zdravstveno te- žavo [angl. Reason for Admission], ki je lahko kronična bolezen, akutno zdravstveno stanje ali poškodba, aplikacija pa nato poda diferencialno diagnozo. Eden od pomislekov pri uporabi te aplikacije bi lahko bil, da je tovrstna diagnoza uporabna le za zdravnike in ne za medicinske sestre, vendar pa je ena od pomembnih nalog medicinskih sester tudi ta, da skrbno sprem- ljajo delo zdravnikov in, če je potrebno, opozorijo na morebitna neskladja ali nepravilnosti, kar od njih zahteva določeno mero dodatnega znanja in lastne angažiranosti. Ko je podana diferencialna diagnoza, se študenta v paru lotita ponovne diskusije, pri čemer razmišljata, kateri od predlogov diagnoze bi bil sprejemljiv in kateri ne, ter ob svojem razmišljanju podata argumente za ali proti. Naslednji korak je razširjena diskusija in plenum o vsakem individu- alnem primeru: vsako izbrano diagnozo in argumente zanjo vsak par predstavi ostalim pri pouku, kar je osnova za nadaljnjo diskusijo o ustreznosti diagnoze. Medjezikovno posredovanje v tem učnem kontekstu zavzema posebno mesto, otežuje pa ga okoliščina, da so v skupini študenti z zelo različnimi ravnmi znanja maternega in tujega jezika. Tuji študenti, ki prihajajo iz različnih jezikovnih okolij (bodisi študenti na izmenjavi Erasmus bodisi tujci, ki študirajo na mariborski univerzi), pa težavo mešanih skupin [angl. mixed-ability classes] še poglabljajo. Kljub temu oziroma prav zato si pri poučevanju angleščine kot TJS za delavce v zdravstveni negi prizadevam spodbujati raznojezičnost, ki je, kot jo pojmuje SEJO, (dinamičen jezikovni repertoar, ki ga uporabnik/učenec nenehno širi) […], predstav- ljen[a] kot neenakomerna in spreminjajoča se zmožnost, kjer se lahko uporabnikovi/ učenčevi viri v nekem jeziku ali njegovi različici močno razlikujejo od virov v drugem jeziku ali njegovi različici (podpoglavje 6.1.3.2) (Svet Evrope, 2023, str. 27). Ker se zavedam, da imajo »raznojezični posamezniki en sam, sicer v sebi povezan jezikovni repertoar, ki ga pri izvajanju opravil povezujejo z drugimi splošnimi zmožnostmi in različnimi strategijami« (Svet Evrope, 2023, str. 27), praviloma raznojezično skupino študentov vodim skozi raznolike vidike (kulturni, strokovni, starostni, vezani na spol, ipd.) ter primerjam obsto- ječe jezikovne in kulture različice in jih integriram v druge splošne zmožnosti, v vse večji meri s pomočjo UI, ki pomembno spreminja način dela. Učna aktivnost je didaktično zasnovana na naslednji način: 37 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 Utemeljitev Različni načini uporabe orodij umetne inteligence omogočajo kakovostnej- še učenje in poučevanje tujega jezika stroke ter boljše razumevanje obrav- navanih vsebin. Pri tem študenti lažje povežejo svoje znanje z drugimi stro- kovnimi temami ter aktivirajo povezave med njimi. Aktivnosti poučevanja so strukturirane tako, da omogočajo nadzorovano rabo TJS in študentove materinščine, npr. z uporabo t. i. »trenutkov za L1« [angl. own language moments] (Kerr 2014) in pedagoškega prevajanja (Cenoz, 2017). Učna enota Angleščina kot izbirni predmet, 1. letnik dodiplomskega študija Zdravstvene nege Cilji Cilj je na študenta osredinjeno učenje TJS z odkrivanjem in raziskovanjem: a) kritično razmišljanje ob problemskem pristopu pri delu s pacienti in drugi- mi strokovnimi sodelavci v zdravstveni negi, b) utrjevanje, ponavljanje in širjenje študentovega razumevanja angleškega jezika (strokovne in splošne angleščine) na višji ravni, c) spodbujanje in razvijanje sposobnost razumevanja različnih govorcev an- gleškega jezika in besedil v različnih kontekstih na višji ravni za strokovne potrebe, d) sistematično razširjanje oz. bogatenje strokovnega besednega zaklada, e) aktivno kritično ustno sodelovanje v razpravah, pri čemer se uri pravilna izgovarjava in naglaševanje ter pravilnost pri rabi jezika, f) razvijanje zmožnosti empatije in razumevanja raznokulturnosti pri sodelo- vanju z bolniki in drugimi sodelavci, g) spodbujanje tehnično-tehnološke kompetence. Učni izidi Študenti: 1. znajo osmisliti in dejansko uporabiti strokovna jezikovna sredstva, ko v ustni interakciji od bolnika pridobivajo natančne informacije, 2. znajo prepoznati in povzemati glavne argumente v strokovnem besedilu (medicinska dokumentacija – anamneza) in se nanje jasno in jedrnato pisno odzovejo, 3. nadgrajujejo mehke veščine (čustveno inteligenco) in raznokulturno zmo- žnost. Dejavnosti Sprejemanje (slušno in bralno razumevanje), tvorjenje (govor in pisanje), interak- cija, posredovanje (medjezikovno in raznokulturno) Jezikovna raven A2–B2 38 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 Postopek/ aktivnost Aktivnost se izvaja na podlagi obrazca Patient Admission Form iz obveznega učne- ga gradiva avtorjev Allum in McGarr (2008) Cambridge English for Nursing, ki ga je treba izpolniti s pridobljenimi podatki. 1. Študenti preberejo obrazec in preverijo razumevanje strokovnega besediš- ča. Nato v paru izvedejo igro vlog – prvi prevzame vlogo pacienta, ki je tujec in govori angleško, drugi prevzame vlogo medicinske sestre, ki pridobljene informacije sproti prevaja v slovenščino in vstavlja v sistem. Začneta s pri- dobivanjem anamneze: s postavljanjem vprašanj in podajanjem odgovorov, pri čemer po potrebi uporabljata spletni prevajalnik DeepL. 2. Sledi zamenjava vlog in vaja se ponovi, nato se vsak par predstavi ostalim v skupini. Učitelj in ostali študenti sproti podajajo dvojicam povratne infor- macije o predstavljenem. Če študent svojih misli ne zna izraziti v angleščini kot TJS, mu drugi študenti ali učitelj pomagajo bodisi s parafraziranjem ali s prevodom terminov. 3. Nato vsak par v aplikacijo Glass v angleščini vnese »svojo« zdravstveno te- žavo, ki je lahko kronična bolezen, akutno zdravstveno stanje ali poškodba. Ta aplikacija poda diferencialno diagnozo. Študenti so ob natančnem bra- nju predlogov pozorni na podrobne informacije, s pomočjo katerih potem podajo argumente v prid podanemu predlogu možne diagnoze ali proti. Ob tem se učijo natančnega branja, natančnega strokovnega izražanja, in osnov aktivnega poslušanja in kulture dialoga. 4. Zaključna aktivnost je refleksija v obliki pisnega sestavka v angleščini: kaj se je posameznemu študentu zdelo zanimivo, zahtevno, kaj se je naučil nove- ga. To se izvede po metodi »refleksije 3-2-1«: napiši tri stvari, ki si se jih nau- čil, dve vprašanji, ki se ti ob tem zastavljata, in eno stvar, ki ti je bila v veselje. Učno gradivo Obrazec 39 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 5.2 Primer 2 – prevajanje in tolmačenje Druga študija primera prihaja s področja poučevanja angleščine za potrebe prihodnjih preva- jalcev in tolmačev. Razvijanju jezikovne zmožnosti v tujem in maternem jeziku je pri poučeva- nju študentov Filozofske fakultete Univerze v Mariboru, dodiplomski programi20 Prevajalske študije, namenjena osrednja pozornost, saj je jezik njihovo prihodnje profesionalno orodje. Vsebine, ki so predstavljene v nadaljevanju, sodijo v okvir obveznega predmeta Razvijanje je- zikovne zmožnosti 2 – angleščina, ki ga študenti obiskujejo v drugem semestru prvega letnika, kjer si prizadevamo, da bi poučevanje prilagodili njihovim prihodnjim poklicnim potrebam. Obenem z razvijanjem slušne, govorne, pisne in bralne zmožnosti s poudarkom na slušnih in govornih dejavnostih ter sistematičnim pridobivanjem novega besedišča ta predmet utrjuje že predhodno pridobljeno slovnično znanje na zahtevnejših primerih besedil. Ker program omogoča kombinacijo z drugimi jezikovnimi programi (npr. Nemščina), je med udeleženci veliko raznojezičnih študentov. Če gre pri poučevanju angleškega jezika za potrebe prihodnjih zdravstvenih delavcev za »stan- dardni« pouk TJS, povezan s specifičnim področjem, pa je poučevanje angleščine za potrebe prihodnjih prevajalcev in tolmačev »nestandardni« TJS, saj ni specializiranih (žanrskih) komu- nikacijskih situacij. Z vidika TJS je poučevanje tujega jezika za prihodnje prevajalce in tolmače prilagojeno v tej smeri, da vključuje besedila, s katerimi se utegnejo študenti srečati v prihod- nosti, jezikovne dejavnosti (npr. medjezikovno prehajanje in posredovanje besedil vključno s prevajanjem) in aktivnosti (npr. povzemanje in senčenje21), teoretično in praktično primerjavo obeh jezikovnih sistemov in podobno (za več o poučevanju tujega jezika za potrebe prihodnjih prevajalcev in tolmačev glej npr. Cerezo Herrero, Schmidhofer in Koletnik, 2021). Za zgodnje razvijanje jezikovne in prevajalske zmožnosti pri predmetu uporabljamo spletne vire in pripomočke, npr. kolokacijska orodja v slovenskem in angleškem jeziku (npr. OZDIC22 in CJVT Kolokacije23), vire za iskanje sopomenk (npr. Thesaurus24 in CJVT Sopomenke25) in kor- pusna orodja (npr. COCA in BNC26 ter CJVT Gigafida27). Za angleški jezik pa sicer kot »korpus« pogosto uporabljamo kar svetovni splet oziroma iskalnike, kot je Google. Tudi pri tem predmetu uporabljamo spletne pripomočke, kot so Mentimeter, Slido,28 Kahoot ali Socrative.29 Uporabo strojnih prevajalnikov še načeloma odsvetujemo, ker se temu po- dročju strukturirano posvetimo pri drugih predmetih v okviru prevajalskega študija (npr. pri predmetu Lektorat iz prevajanja). Aktivnosti, ki smo jih zasnovali za prihodnost, pa 20 Do študijskega leta 2024/25 so se lahko študenti na Oddelku za prevodoslovje FF UM vpisali v tri programe Prevajalskih študij, in sicer Angleščina, Nemščina in Madžarščina. Od leta 2024 naprej so programi združeni v enega s tremi smermi: angleščina, nemščina in madžarščina. 21 Senčenje je didaktična tehnika učenja jezika, ki temelji na tem, da študent posluša zvočni posnetek in s čim manjšim zamikom glasno ponavlja slišano. 22 https://ozdic.com/ 23 https://viri.cjvt.si/kolokacije/slv/ 24 https://www.thesaurus.com/ 25 https://viri.cjvt.si/sopomenke/slv/ 26 https://www.english-corpora.org/ 27 https://viri.cjvt.si/gigafida/ 28 https://www.slido.com/. Storitev je v prvi vrsti namenjena interaktivnim srečanjem, omogoča pa tudi kratke kvize in preverjanje odziva. 29 https://www.socrative.com/. Gre za didaktično orodje, ki med drugim omogoča preverjanje znanja. 40 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 predvidevajo tudi uporabo orodij UI, med njimi takih, ki pomagajo študentom pri pisanju in urejanju besedil, npr. QuillBot30 ali Grammarly,31 ter velikih jezikovnih modelov, v konkretnem primeru ChatGPT in Gemini, ki ju v omenjeni nalogi uporabimo za povzemanje. Pri tej dejavnosti je bil uporabljen odlomek iz podkasta From Our Own Correspondent za BBC Radio 4 o rokoborbi in veroizpovedih v Mongoliji, njegov avtor je Robert Pigott (BBC News, 2009).32 Besedilo je zahtevnejše in od študentov zahteva zgodovinsko in geografsko znanje, znanje o verstvih, obenem pa jim približuje elemente precej oddaljene mongolske kulture in tamkajšnjega sodobnega življenja. Spodnji primer je prirejen po primeru, ki je predstavljen v Cerezo Herrero in soavtorji (2024). Primer ima sicer dva dela, a) poslušanje za prepoznavanje bistva in b) poslušanje za prepozna- vanje podrobnih informacij. Zaradi omejenega prostora drugega dela tukaj ne predstavljamo. Učna aktivnost je didaktično zasnovana na naslednji način: Utemeljitev Z aktivnostmi medjezikovnega prehajanja [angl. translanguaging] učitelj prisotnost večjezičnosti v razredu (tj. več raznojezičnih študentov) izkoristi za opolnomočenje in aktiviranje celotnega jezikovnega potenciala študen- tov (Donley, 2022). Uporaba spletnih orodij in orodij UI študente pripravlja na prožno uporabo tehnologije v poklicnem življenju v prihodnosti in razvija njihovo prilagodljivost in odzivnost na spremembe. Učna enota Razvijanje jezikovne zmožnost 2 - angleščina, 1. letnik dodiplomskega študija Pre- vajalske študije, 2. semester Cilj Razvijanje sposobnosti razumevanja različnih govorcev angleškega jezika in be- sedil v različnih kontekstih na zahtevnejši ravni; sistematično razširjanje oz. bo- gatenje besednega zaklada, aktivno sodelovanja v ustnih razpravah, spodbujanje tehnično-tehnološke kompetence. Učni izidi Študenti: 1. razumejo ključne informacije v poslušanem besedilu, znajo pripraviti za- piske, ustno povzamejo glavne argumente in jih posredujejo sošolcem; 2. kompetentno uporabljajo orodja UI pri širjenju jezikovnega znanja. Dejavnosti Sprejemanje (slušno razumevanje), tvorjenje (govor in pisanje), posredovanje Jezikovna raven B2-C1 30 https://quillbot.com/. 31 https://www.grammarly.com/ 32 http://news.bbc.co.uk/2/hi/programmes/from_our_own_correspondent/7860539.stm 41 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 Aktivnost/ postopek Prva faza (poslušanje za prepoznavanje bistva) 1. Učitelj pripravi krajši do srednje dolg avtentičen posnetek o aktualni temi v angleščini, ki ga študenti samostojno poslušajo doma kot dejavnost pred poukom. Težavnost besedila učitelj v procesu izbiranja lahko prila- godi stopnji njihovega jezikovnega znanja (zmožnosti). Študentom naroči, naj med poslušanjem prepoznajo glavne teme (2–3) in si jih v poljubnem jeziku zapišejo ter jim dodajo nekoliko »ozadja«, tj. dodatnih informacij (pisanje zapiskov), ki jih slišijo in razumejo. 2. V razredu študente razdeli v pare in jih pozove, da se v angleščini pogo- varjajo o glavnih temah besedila, ki so ga poslušali. Pri tem naj se opirajo na zapiske. 3. Nato razredu omogoči dostop do elektronske različice besedila in ga raz- deli na dvoje: polovica razreda bo pri delu uporabljala orodje ChatGPT, druga polovica pa orodje Gemini. Delo z obema orodjema zahteva regi- stracijo. Učitelj skupaj s študenti razmisli o iztočnici [angl. prompt], s kate- ro bodo orodjema UI naročili, da povzameta glavne ideje besedila. Mi smo orodju naročili, da pripravi »miselni vzorec«.33 4. Učitelj nato parom naroči, da s pomočjo oz. na podlagi jezikovnih informa- cij orodij UI na list papirja narišejo pravi miselni vzorec. Različica: Študenti bi lahko pripravili tudi risbo in/ali karikaturo o zgodbi. Risba naj ima angle- ški naslov in 5 ključnih besed v angleščini in v njihovem maternem jeziku. 5. Nato učitelj pare pozove, naj pripravijo kratko poročilo, v katerem v ne več kot treh povedih v slovenščini (ali drugem jeziku, ki ga razume večina) sošolcem prenesejo glavne teme in vsebine besedila. Ko poslušajo poro- čila drugih skupin, naj bodo pozorni na razlike v izbiri in predstavitvi tem, dejstev in argumentov in si naredijo zapiske (govorna obdelava besedila). Sledi krajša vodena razprava v razredu. 6. Če čas ne zadošča, študenti pripravijo kratko govorno individualno poro- čilo (posnetek) kot domačo nalogo, ki jo učitelj pregleda, poda povratne informacije in razpravo uvrsti v naslednjo uro. 33 Nobeno od orodij v primerih, ki smo jih preizkusili, ni pripravilo »grafičnega« miselnega vzorca, temveč zgolj iztočnice v obliki seznama. Na voljo so sicer tudi orodja UI, ki to pripravijo (npr. GPT Mind Maps Ma- ker, https://shorturl.at/D2QCk), vendar so v brezplačni različici večinoma zgolj omejeno funkcionalna, npr. omogočajo zgolj eno brezplačno uporabo. 42 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 Primer Primer 1: Gemini Primer 2: ChatGPT Nadaljevanje je zaradi pomanjkanja prostora izpuščeno. 43 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 6 Etika in kritična refleksija uporabe tehnologije pri poučevanju in učenju Globalizacija in hiter razvoj tehnologije ne vplivata zgolj na napredek v strokah in pri global- nem komuniciranju, kjer se pojavlja vedno večja potreba po večkulturnih in večjezičnih kom- petencah, ampak spreminjata tudi dinamiko dela v razredu, v našem primeru pri pouku TJS. Namen učiteljev je, da se tovrstne kompetence razvijejo do te mere, da posamezniku (tj. štu- dentu kot bodočemu strokovnjaku) kar najbolj koristijo. Le tako bo namreč lahko sledil hitrim znanstvenim in tehnološkim spremembam ter se prilagajal zahtevam svoje stroke, obenem pa ohranjal svojo konkurenčnost na trgu delovne sile. Digitalna tehnologija spreminja tako način življenja in dela kot tudi način učenja in poučevanja. Njene potencialne koristi pa po drugi strani prinašajo tudi pasti in nevarnosti. Ker stremimo k vse učinkovitejšim delovnim postopkom in vedno višji delovni učinkovitosti, v hitrem vsakodnevnem delovnem tempu pozabljamo na marsikatere etične dileme, ki jih tovrsten napredek prinaša s seboj. Hiter razvoj digitalnih tehnologij spreminja način dela in življenja posameznikov, organiza- cij in družbe kot celote. »Z uporabo digitalnih tehnologij in programskih rešitev lahko delo opravimo hitreje, ceneje in bolj kakovostno,« menita Rožanec in Lahajnar (2019, str. 45). Hkrati s tem je treba pozornost posvetiti tudi izboljšanju digitalnih kompetenc in digitalne- mu opismen jevanju uporabnikov storitev. Področje zdravstva po njunem mnenju na globalni ravni zaostaja za drugimi panogami na področju modernizacije in digitalnega prehoda. Ker zdravstveni sistem ne sledi digitalnim novostim in zdravstveni informacijski sistem ne deluje ustrezno in tudi ne skladno s pričakovanji zaposlenih in bolnikov, lahko to povzroči tudi manj- šo motiviranost osebja za tovrstno delo in izobraževanje, ne nazadnje pa sta lahko ogrožena tudi zajem podatkov in njihova varnost. Ker je vse več (strogo) zaupnih in občutljivih podatkov v digitalni obliki, je ob nepravilni upo- rabi ter nezadostni (nad)gradnji varnostnih sistemov občutno zmanjšana njihova varnost. Pri- dobitev podatkov o pacientih je tako pogosto razlog za vdore v sisteme zdravstvenih evidenc, nesmotrno ravnanje z zaupnimi podatki in njihovo nenadzorovano vnašanje v prevajalnike in chatbote pa je problematično tako za prihodnje zdravstvene delavce kot tudi za prihodnje prevajalce in tolmače. Nadaljnji pomisleki nenadzorovane in nezadostno premišljene uporabe tehnologij in orodij UI se na področju zdravstvene nege izražajo ob vedno več primerih »samodiagnosticiranja«, pri čemer pacienti sami v spletu raziščejo simptome in si na podlagi tega sami postavijo »dia- gnozo« ter dvomijo v pravilnost presoje in zatem delovanja zdravstvenih delavcev. Resda so bolnikom tovrstne informacije lahko v pomoč, a ob nekritični uporabi katerihkoli dostopnih podatkov in aplikacij ter tudi že kar mnogih možnostih t. i. »samozdravljenja«, ki jih nudi sple- tno okolje, prihaja do vse več nesporazumov in nezadovoljstva med zdravstvenimi delavci in pacienti. Ob vsem doslej navedenemu velja predvsem poudariti jezikovne ovire, a) ko pacienti niso vešči tujih jezikov in si pri prevajanju pomagajo z neverodostojnimi prevajalskimi orodji, kar dodatno vodi do nepravilnosti uporabe prejetih informacij, in b) ko so pacienti prepričani, da razumejo zdravstveno terminologijo tudi v maternem jeziku, pa to najpogosteje ne more biti res, saj ne delujejo na področju zdravstva. Da bi se v čim večji meri izognili tovrstnim »nespretnim« uporabam dostopnih informacij, že učence, dijake in študente navajamo pra- vilne rabe spleta ter UI, pa tudi izobraževanje odraslih in starejših ne zaostaja v tem pogledu. 44 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 Molek (2023) ugotavlja, da postajajo sistemi UI, ki temeljijo na velikih jezikovnih modelih, vse bolj razširjeni kot izobraževalna tehnologija tudi na področju visokošolskega izobraževanja. Visokošolski učitelji smo njihov prihod in uporabo v razredu sprejeli z mešanimi občutki, saj kot novost predstavljajo izziv, a smo se morali v relativno kratkem času prilagoditi in naučiti, kako jih vključiti v pedagoški proces kot »dodano vrednost«, ki je povezana z etičnimi in mo- ralnimi pomisleki. Eden prvih je po mnenju Molek »... pristop, ki se osredotoča na vidike, kot so učinkovitost, uspešnost, tehnična zmogljivost in druga ‚objektivna‘ merila, pri čemer so človeški, etični, odnosni ali družbeni vidiki zanemarjeni« (str. 301). Ta vidik velja na področjih, kot sta humanistika (v našem primeru prevajanje in tolmačenje) in medicina (v našem primeru zdravstvene vede), postaviti v ospredje, saj gre primarno za vzpostav- ljanje in ohranjanje medčloveških in medosebnih odnosov. Pri posredovanju, tudi medjezikov- nem, pri delu s pacienti je človeški vidik v ospredju in ga ne more oziroma ne sme prevzeti UI. Večje dileme pri njeni uporabi so med drugim subjektivnost, odsotnost empatičnega razume- vanja, upoštevanja kulturnih vidikov in človečnosti, na katerega velja opozarjati ob uporabi UI v razredu. V akademskem okolju, kjer študente TJS pripravljamo na bodoče delo v stroki in na morebitno nadaljnje znanstveno udejstvovanje, se, tako Molek, zato morajo zavedati, da prekomeren poudarek na paradigmah računalniških znanosti pogosto pozabi na etične izzive in vplive UI na človeštvo in na razumevanje dejanj subjektivnih posamez- nikov […] saj se predpostavlja, da razvoj UI poteka v nevtralnih, objektivnih scenarijih brez [...] kulturnih, družbenih, gospodarskih in političnih posledic. Taki pristopi obi- čajno pustijo v ozadju osrednje legitimno in človeško vprašanje etičnih in akadem- skih dilem povezanih z uporabo UI tehnologije v univerzitetnem okolju, vključno s preglednostjo, odgovornostjo, pravičnostjo, varstvom podatkov, ki vplivajo na delo- vanje izobraževalne kulture visokošolskih ustanov (2023, str. 301). Akademska integriteta, tj. preplet etike, kritične presoje in odgovornosti glede odločitev, mora še naprej ostati osnovno vodilo dela na vseh ravneh poučevanja in učenja tujih jezikov, tudi TJS. Seveda nam uporaba UI pri pedagoškem delu po eni strani olajša delo: omogoča hitrejši dostop do gradiva, večjo prilagodljivost (predvsem v smislu hitrejše in lažje priprave dodatnih oz. diferenciranih gradiv za učence na različnih ravneh znanja), večjo kreativnost (UI spodbuja učiteljevo kreativnost, ker so mu lažje dostopne sveže ideje za izvedbo učnega procesa na inovativnejši način), boljšo interakcijo med učiteljem in učenci (zaradi privlačnej- šega pristopa v smislu uporabe virtualnih scenarijev in simulacij, kvizov, ipd.) in še marsikaj. Ob tem ne smemo pozabiti na ustrezno izobraževanje in stalno strokovno izpopolnjevanje za učitelje TJS, ki naj vključuje tudi pridobivanje zadostnega strokovnega znanja za učinkovi- to uporabo tehnologij UI, da lahko sledimo novostim na področju dela in izobraževanja, ter razvijanje znanj, ki naj omogočijo povečano varovanje osebnih podatkov, spletno varnost ipd. Kot že omenjeno, je UI vnesla novo dimenzijo dela s študenti, saj se med drugim tudi veli- ko prej doma opravljenega dela vrača v razrede. S tem zagotavljamo določen nadzor nad njihovim delom, ko jih za opravljanje nekaterih nalog v razredu prepustimo same sebi in njihovemu obstoječemu znanju, brez prekomerne rabe tehnologije v domačem okolju. Z izo- gibanjem pretirani rabi UI preprečujemo pretirano odvisnost od tehnologije, ki lahko zmanjša ustvarjalnost in sposobnost kritičnega mišljenja, preprečujemo pasivno sprejemanje infor- macij in spodbujamo pismenost, ki je na alarmantno nizki ravni. 45 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 7 Zaključek Posredovanje, ki omogoča sporazumevanja v situacijah, kjer bi bilo to sicer težko ali težavno, pri učenju in poučevanju TJS zaradi vse bolj povezanega globalnega sveta in dejstva, da v sebi združuje elemente vseh drugih sporazumevalnih dejavnosti (tj. sprejemanja, tvorjenja in in- terakcije), pridobiva vse vidnejšo vlogo pri pouku in pri strokovni komunikaciji. Posredovanje, ki sicer ni omejeno zgolj na posredovanje med (dvema ali več) jeziki, temveč vključuje tudi po- sredovanje med kulturami in posredovanje med posamezniki, ki govorijo isti jezik, omogoča v sodobnem svetu, ki ga zaznamujeta razmah informacij in vse hitrejši življenjski tempo, boljše razumevanje besedil in konceptov, kar posledično vodi k učinkovitejšemu (spo)razumevanju nasploh. V povezavi s sledenjem novostim in tehnološkim inovacijam, s katerimi učitelji na vseh ravneh učenja in poučevanja jezika, tudi na univerzah, stopajo v korak s časom, zna biti mnogim precejšen izziv: učitelji morajo pridobivati nova znanja na področju, ki jim pogosto ni blizu, podrobno in kritično morajo presojati prednosti in pomanjkljivosti tehnologij, jih premi- šljeno in kreativno uporabljati ter obenem znanje o njih prenašati svojim učencem. Avtorici meniva, da sta razvijanje zmožnosti posredovanja pri učencih in uporaba sodob- nih tehnologij pri pouku vključujoča elementa, pravzaprav je njun cilj maksimiranje principa vključevanja s številnih vidikov. V ta namen se mora poučevanje prilagoditi in spremeniti, v korak s tem pa gremo učitelji, ki si prizadevamo znanje prenašati in razvijati na nove načine. Če želimo študente učinkovito usposobiti za medjezikovno in medkulturno posredovanje in jih voditi skozi vedno nove tehnološke pripomočke in orodja UI ter jih usposobiti v njihove kompetentne in strokovne uporabnike, moramo sami spremljati novosti in spremembe ter upoštevati njihovo vlogo in pomen. Pri pouku TJS dveh skupin študentov – ki se zdita na prvi pogled sicer različni, vendar ju druži to, da bosta v svojem poklicnem življenju posredovali med jeziki in kulturami ter vzpostav- ljali in ohranjali medčloveške in medosebne odnose (empatijo, razumevanje, sprejemanje drugačnosti ipd.), ki sva ga predstavili v tem prispevku, je potencial uporabe nekaterih izbra- nih in predstavljenih orodij (prevajalnikov in jezikovnih modelov UI) visok, a ne brez pasti in težav. Zato naj pregled literature na tem področju in predstavljena primera, kako ta orodja vključujeva v svoje delo, služijo predvsem kot zgled in spodbuda za učinkovitejše poučevanje s pomočjo orodij UI. Zavedava se, da se lahko zgodi, da bo še pred objavo članka zaradi hitrega tehnološkega razvoja kakšen vidik, ki sva ga vključili, morda zastarel, a so nekatere temeljne podstati v globaliziranem večjezičnem in večkulturnem okolju postale stalnica, med drugim pomen in vloga medjezikovnega in medkulturnega posredovanja, ki terja akcijsko usmerjen pristop k poučevanju in učenju TJS in vključevanje novih orodij, ki jih omogoča tehnološki napredek. Pri delu s študenti prevajanja in tolmačenja in zdravstvene nege je še posebej pomembno proaktivno (so)ustvarjanje ustreznih pogojev za učinkovito in uspešno sporazumevanje in sodelovanje, saj se utegnejo študenti v svojih bodočih poklicih pogosto znajti v občutljivih situacijah, v katerih bodo morali uporabiti niz (razno)kulturnih zmožnosti ter pri sporazume- vanju in posredovanju vpreči ves svoj repertoar tujih jezikov. Uporaba tehnologij, tudi UI, pri poučevanju in učenju TJS v obeh predstavljenih primerih omogoča prav vse prej našteto in ob pregledu t. i. »tehnološkega atlasa«, ki vsako leto kar se 46 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 da ažurno in izčrpno predstavi specializirana orodja, namenjena prevajalcem, tolmačem in sorodnim poklicem, je jasno, da je izbira ogromna. Podobno velja tudi za specializirana orod- ja, ki so namenjena prihodnjim zdravstvenim delavcem. Nekatera so morda bolj znana širši javnosti – večina ljudi pozna mobilne aplikacije za spremljanje telesne aktivnosti in vadbe – druga so bolj ozko specializirana in pomagajo pri prepoznavanju simptomov ter postavljanju možnih diagnoz na podlagi podanih simptomov, ali so pomoč pri organizaciji in izvajanju dela strokovnih delavcev. S tehnološkimi novostmi, zadnja je bila denimo prihod velikih jezikovnih modelov, ki podpira- jo številne jezike, se spreminjajo tudi didaktični pristopi. Generativna orodja so lahko učitelju v pomoč, če jih le zna osmisliti in uporabiti: na zdravstvenem področju olajšajo medjezikovno in medkulturno komunikacijo, saj omogočajo lažje sporazumevanje med medicinsko sestro in pacientom, npr. pri osnovnem svetovanju o zdravju, zdravilih, postopkih in prepoznavanju simptomov, prihodnjim prevajalcem in tolmačem pa pomagajo povzemati besedila in razis- kovati njihovo ozadje, izluščiti bistvo in podobno. Predstavitev dveh učnih aktivnosti, ki vključujeta uporabo orodij umetne inteligence pri po- uku angleščine kot TJS s ciljem krepitve kompetenc medjezikovnega in medkulturnega po- sredovanja v najinih specializiranih kontekstih – zdravstveni negi in prevajanju in tolmačenju – ter njunih didaktičnih zasnov plastično ilustrira, kako se spreminjajo naše vloge; učitelj ni več le »podajalec znanja« in študent ni več zgolj njegov »sprejemnik«, temveč prevzemata oba vlogi sooblikovalcev v okviru učnega procesa, ki je osredinjen na učenca. Kritično gledano skozi prizmo etične refleksije se moramo zavedati, da digitalne tehnologije in programske rešitve sicer optimizirajo in pohitrijo delovne procese ter deloma nadgrajujejo učne procese, a pri tem ne gre zgolj stremeti k višji učinkovitosti, ampak je treba – še posebej na področjih humanistike in zdravstva – na prvo mesto postaviti medčloveške in medosebne odnose, več- oziroma med-kulturne vidike ter vidike subjektivnosti, razumevanja in kritičnega mišljenja. Slednjič ne smemo pozabiti na akademsko integriteto na vseh ravneh poučevanja in učenja TJS, ki naj ostane osnovno vodilo našega znanstvenoraziskovalnega dela, ne glede na mamlji- vo lahkotnost na številnih področjih, kar zares ali zgolj navidezno nudi UI. Na vseh področjih je namreč treba primarno upoštevati dejavnika človečnosti in človeškosti, ki ju umetna inteli- genca – vsaj zaenkrat – še ne premore. Viri in literatura Alič, P. (2023). Uporaba orodij umetne inteligence za generiranje slik v zdravstveno-vzgojnem delu. V A. Pušnik in M. Rajh Jager (ur.), Zbornik prispevkov 4. Nacionalne strokovne konference Kreativna učna orodja (str. 9–16). Šolski center Celje, Medpodjetniški izobraževalni center. Allum, V. in McGarr, P. (2008). Cambridge English for Nursing, Intermediate +. Cambridge University Press. BBC News. (2009). Wrestling and faith in Mongolia. http://news.bbc.co.uk/2/hi/programmes/from_our_ own_correspondent/7860539.stm Beacco, J.-C., Byram, M., Cavalli, M., Coste, D., Egli CUemat, M., Goullier, F. in Panthier, J. (2016). Educa- tion Policy Division, Language Policy, Education Department, Directorate of Democratic Citizenship and Participation, DGII – Directorate General of Democracy, and Council of Europe. Guide for the 47 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 Development and Implementation of Curricula for Plurilingual and Intercultural Education. Council of Europe. Benabdallah, A. (2023). Investigating the Future of ESP Teaching in the Age of Artificial Intelligence. An- nals of Ştefan cel Mare University of Suceava. V B. Popoveniuc in M. Cucu, (ur.), Philosophy, Social and Human Disciplines Series (str. 77–93). Ştefan cel Mare University of Suceava Press. Buchwald, S., Hartmann, E., Kelih, M. in Wutti, D. (2022). Jaz bi rada znala veliko jezikov. Privlačnost kot dejavnik učenja jezikov v večjezičnih biografijah. DiSlaw – Didaktik slawischer Sprachen 2(2), 1–17. https://doi.org/10.48789/2022.2.2. Cenoz, J. (2017). Translanguaging in School Contexts: International Perspectives. Journal of Language, Identity & Education, 16(4), 193–198. https://doi.org/10.1080/15348458.2017.1327816 Cerezo Herrero, E., Schmidhofer, A. in Koletnik, M. (2021). An LSP Framework for Translation and Inter- preting Pedagogy. Revista de Lenguas Para Fines Específicos, 140–156. https://doi.org/10.20420/ rlfe.2021.441 Cerezo Herrero, E., Schmidhofer, A., Ureel, J. in Koletnik, M. (2024). Navigating linguistic contrastivity in additional language teaching for translation and interpreting. V C. Perez in E. Cerezo Her- rero (ur.) Methodological Innovation in Higher Education: Language Teaching and Translation (str. 117–142). Tirant Lo Blanch. Council of Europe. (2018). Common European framework of reference for languages learning, teach- ing, assessment. Companion volume with new descriptors. Council of Europe. https://rm.coe.int/ cefr-companion-volume-with-new-descriptors-2018/1680787989 Donley, K. (2022). Translanguaging as a theory, pedagogy, and qualitative research methodology. NABE Journal of Research and Practice, 1–16. https://doi.org/10.1080/26390043.2022.2079391 Drašler, A. in Podgoršek, S. (2022). Stališča študentk in študentov do rabe avtentičnih gradiv in opravil pri pouku tujih jezikov stroke. Journal for Foreign Languages, 14(1), 325–341. https://doi.org/10.4312/ vestnik.14.325-341 Dudley-Evans, T. in St. John, M. (2002). Developments in English for Specific Purposes: a multi-disciplinary approach. Cambridge University Press. English with Cambridge. (2022a, October 12). Bite-sized essentials: what, why and how? Interlingual medi- ation - Philip Kerr - Cambridge Live. YouTube. https://www.youtube.com/watch?v=FkX88wcdWS8 Gandin, S. (2016). Translating and Learning the Language of Tourism as LSP: Corpus-based Approach- es. LCM - La Collana / the Series, 9788879167918, 65–82. https://doi.org/10.7359/791-2016-gand García, O. in Lin, A. (2017). Translanguaging in Bilingual Education. V O. García, Lin in May (ur.), Bilingual and Multilingual Education (str. 117–130). Springer International Publishing. Hyland, K. (2019). English for Specific Purposes: Some Influences and Impacts. V X. Gao (ur.) Second Handbook of English Language Teaching (str. 337–353). Springer International Publishing. Jurkovič, V., Dostal, M., Podgoršek, S. in Mertelj, D. (2023). LSP teacher professional development: needs-based curriculum design. V A. Memišević in M. Matešić (ur.) Linguistic and extralinguistic in interaction (str. 207-222). Peter Lang. Jurkovič, V., Mertelj, D. in Podgoršek, S. (2024). Further step toward a definition of the core profession- al development needs of teachers of languages for specific purposes in the European higher education area. Revista española de lingüística aplicada, 37(1), 299–334. https://doi.org/10.1075/ resla.21039.jur Kerr, P. (2014). Translation and own-language activities. Cambridge University Press. Kerr, P. (2022). Philip Kerr’s 30 Trends in ELT. Cambridge University Press. 48 Majcenovič Kline, Koletnik / Scripta Manent XIX/1 (2024), 25–48 Khan, S. (2023, May 1). Sal Khan: How AI could save (not destroy) education. Www.ted.com. https://www. ted.com/talks/sal_khan_how_ai_could_save_not_destroy_education/transcript?subtitle=en Kirovska-Simjanoska, D. (2020). Teaching ESP in the Digital World – Developing a Blended Learn- ing Environment for Computer Science Students. BELLS90 Proceedings, 423–438. https://doi. org/10.18485/bells90.2020.1.ch25 Koletnik Korošec, M. (2011). Applicability and Challenges of Using Machine Translation in Translator Training. ELOPE: English Language Overseas Perspectives and Enquiries, 8(2), 7–18. https://doi. org/10.4312/elope.8.2.7-18 Luckin, R., Cukurova, M., Kent, C. in du Boulay, B. (2022). Empowering educators to be AI-ready. Com- puters and Education: Artificial Intelligence, 3, 100076. https://doi.org/10.1016/j.caeai.2022.100076 Magesh, V., Surani, F., Dahl, M., Suzgun, M., Manning, C. D. in Ho, D. E. (2024, May 30). Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools. ArXiv.org. https://doi.org/10.48550/ arXiv.2405.20362 Mikolič Južnič, T., Žigon, T. in Zlatnar Moe, M. (2014). Visoka jezikovna kompetenca v maternem jeziku kot temelj uspešne večjezičnosti v Evropski uniji. Teorija in praksa, 51(4), 636–653. http://www. dlib.si/details/URN:NBN:SI:doc-3LY8JE1K Molek, N. (2023). Etični izzivi uporabe umetne inteligence v visokošolskih zavodih. Revija za univerzalno odličnost / Journal of Universal Excellence, 12(4), 300–318. OpenAI. (2024). ChatGPT. (različica z dne 15. junija) Veliki jezikovni mode. https://chat.openai.com/chat Podgoršek, S., Jurkovič, V., Dostal, M. in Mertelj, D. (2021). Analiza izobraževalnih potreb visokošolskih učiteljev tujih jezikov stroke v Sloveniji. Sodobna pedagogika, 72(3), 94–110. Rožanec, A. in Lahajnar, S. (2019). Digitalne tehnologije za zdravstvene storitve prihodnosti. Infor- matica Medica Slovenica, 24(1–2), 45–52. https://www.dlib.si/stream/URN:NBN:SI:DOC-C1BZA5JF/ e2324baa-43a5-4f80-8632-61b59a0b2753/PDF Schmidhofer, A., Cerezo Herrero, E. in Koletnik, M. (2022). Integrating Mediation and Translanguaging into TI-Oriented Language Learning and Teaching (TILLT). Lublin Studies in Modern Languages and Literature, 46(2), 99–109. https://doi.org/10.17951/lsmll.2022.46.2.99-109 Svet Evrope. (2001). Skupni evropski jezikovni okvir: učenje, poučevanje, ocenjevanje. Ljubljana: Minis- trstvo RS za šolstvo in šport. https://centerslo.si/wp-content/uploads/2015/10/SEJO-komplet-za- splet.pdf Svet Evrope. (2023). Skupni evropski jezikovni okvir: učenje, poučevanje, ocenjevanje – dodatek. Ljubljana: Zavod RS za šolstvo. www.zrss.si/pdf/SEJO.pdf UNESCO. (2023). Global Education Monitoring Report, 2023: technology in education: A tool on whose terms? Pariz, UNESCO. https://doi.org/10.54676/UZQV8501 University of Cambridge. (b. d.). Mediation: What it is, how to teach it and how to assess it. Cambridge Eng- lish Language Assessment. https://www.cambridgeenglish.org/Images/664965-mediation-what- it-is-how-to-teach-it-and-how-to-assess-it.pdf van Dijk, T. A. (2003). Specialized discourse and knowledge: A case study of the discourse of modern genetics. Cadernos de Estudos Lingüísticos, 44, 21–55. https://doi.org/10.20396/cel.v44i0.8637063 49 Scripta Manent XIX/1 (2024) DOI: 10.4312/SM.19.1.49-62 Research Paper Paper received: 15.06.2024 Paper revised: 10.08.2024 Paper accepted: 21.09.2024 Paper published: 30.11.2024 Katja Težak Universität Maribor, Slowenien Sara Orthaber Universität Maribor, Slowenien MIT ChatGPT AGRAR-FACHWORTSCHATZ ERLERNEN: MÖGLICHKEITEN UND GRENZEN Zusammenfassung Die Einbeziehung künstlicher Intelligenz (KI) – wie ChatGPT – in den Sprachunterricht öffnet neue Pers- pektiven für den Erwerb eines Fachwortschatzes und von Sprachkompetenzen. Die vorliegende Studie widmet sich, im Deutschunterricht an einer slowenischen Universität für zukünftige Expert/innen der Ag- rarwissenschaften mithilfe von ChatGPT-3 Termini und deren grammatische Eigenschaften zu extrahie- ren. Dabei wird untersucht, wie ChatGPT-3 diesbezüglich bei öffentlich verfügbaren wissenschaftlichen Beiträgen zum Thema Landwirtschaft, grüner Wandel und Nachhaltigkeit vorgeht. Die Ergebnisse haben gezeigt, welche Begriffe ChatGPT als bedeutsam für den Deutschunterricht identifiziert und wie sie von dem KI-Instrument dargestellt werden. Außerdem wird aufgezeigt, wie hoch die Fehlerrate ist und wel- che Terminologie für Studierende der Agrarwissenschaften einerseits hilfreich und relevant, andererseits jedoch potenziell irreführend sein kann. Die Studie könnte zur Erstellung zielgerichteter Lehrmaterialien für Studierende der Agrarwissenschaften genutzt werden, indem Lehrende die von ChatGPT generierten Termini als Ausgangspunkt präsentieren, die von den Studierenden verbessert werden können. Alterna- tiv könnte sie auch in der Form einer „Finde die Fehler“-Aufgabe gestaltet werden. Schlüsselwörter: ChatGPT-3, AI, DaF, Fachwortschatz, Agrarwissenschaften Abstract Learning Agricultural Vocabulary with ChatGPT: Possibilities and Limitations The integration of artificial intelligence (AI) such as ChatGPT into language teaching, offers new per- spectives for acquiring specialized vocabulary and language skills. This study explores the use of ChatGPT-3 to extract terms and their grammatical properties in German lessons at a Slovenian uni- versity for future agricultural science experts. It examines how ChatGPT-3 processes publicly available scientific articles on agriculture, green transition, and sustainability. The findings have shown which terms ChatGPT identifies as agricultural terminology in German language instruction and how they are presented by the AI tool. Additionally, the error rate is highlighted, as well as which terminology is helpful and relevant for agricultural science students on the one hand, but potentially misleading on the other. This research could inform the development of targeted teaching materials for agricultural science students, for instance, by using ChatGPT-generated terms as a starting point for student cor- rections. Alternatively, it could be adapted into a ‘find the errors’ exercise. Keywords: ChatGPT-3, AI, German as a foreign language, specialized vocabulary, agricultural science 50 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 1 Einführung Künstliche Intelligenz (KI) hat verschiedene Bereiche unseres Lebens revolutioniert, darunter auch die Bildung und das Erlernen von Fremdsprachen. Durch ihre Fähigkeit, enorme Daten- mengen zu verarbeiten, aus Mustern zu lernen und intelligente Entscheidungen zu treffen, kann KI dabei helfen, zeitsparend ein höherer Lernfortschritt zu erzielen, als wenn die Stu- dierenden alle Informationen selbstständig recherchieren müssten. Die Stärke von KI zeigt sich in der Bewältigung komplexer Aufgaben in bisher ungekanntem Ausmaß und Tempo. So werden Abläufe effizient optimiert und die Gesamtproduktivität verbessert. KI setzt mensch- liche Ressourcen frei, indem sie repetitive Aufgaben automatisiert und ihnen die Möglichkeit einräumt, sich auf höherwertige, anspruchsvolle und auf kritischem Denken beruhende Tä- tigkeiten zu konzentrieren. Einige Forscher und Forscherinnen weisen darauf hin, dass KI-Al- gorithmen Bildungsinhalte personalisieren können, um den individuellen Bedürfnissen und Lernstilen der Studierenden gerecht zu werden. Dies hilft Pädagogen und Pädagoginnen, Verbesserungsbereiche zu identifizieren und Bildungsprogramme an spezifische Anforde- rungen anzupassen (AI, 2024). Im Bereich des Sprachenlernens ist KI noch nicht so weit, eine herausragende Lernerfahrung zu schaffen. Laut einer Studie (Mykhalevych, 2024) ist dies vor allem auf die fehlende menschliche Bindung, Orientierungshilfe und Motivation durch eine Lehrkraft zurückzuführen. In diesem Sinne meint Damberger (2017), „dass Schule nicht nur dem Namen nach eine Bildungseinrichtung ist, dass Bildung Menschenbildung und nicht Systemoptimierung meint, und dass das Menschliche am Menschen sich ausschließlich in der menschlichen Begegnung und nicht in ihrer Simulation zu entfalten vermag.” Ein weiterer nennenswerter Konflikt zeigt sich darin, dass ChatGPT noch die Kompetenz fehlt, Daten an- gemessen zu extrahieren, sodass der Lehrkörper unbedingt benötigt wird, um Probleme im Lernprozess zu vermeiden. Die vorliegende Studie widmet sich der Thematik, im Deutschunterricht an einer sloweni- schen Universität für zukünftige Experten und Expertinnen der Agrarwissenschaften mithilfe von ChatGPT-3 Termini zu extrahieren. Dabei wird untersucht, wie ChatGPT-3 diesbezüglich bei öffentlich verfügbaren wissenschaftlichen Beiträgen zum Thema Landwirtschaft, grüner Wandel und Nachhaltigkeit vorgeht. Schlussendlich soll aufgezeigt werden, welche Termino- logie ChatGPT für den Deutschunterricht bedeutsam findet, wie sie von dem KI-Instrument präsentiert wird und was im Rahmen dieser Studie für Studierende einerseits hilfreich und relevant ist, andererseits aber irreführend wäre. So könnte die Studie für die Erarbeitung ziel- gerichteter Lehrmaterialien für Studierende der Agrarwissenschaften angewendet werden (vgl. Meršnik, Đorđević & Anđelković, 2023), indem sie eine Brücke zwischen fachspezifischen Termini und effektivem Sprachunterricht mit ChatGPT baut. Die gewonnenen Ergebnisse ge- währen einen wertvollen Blick auf den Einsatz von KI im Bildungsprozess und tragen zur erfolgreichen Integration von KI in den Lehrplan bei. 2 Was ist ChatGPT Die Entwicklung von Sprachmodellen stellt einen bedeutenden Meilenstein im Bereich der KI dar. Bemerkenswerte Fortschritte lassen die GPT (Generative Pre-trained Transformer)-Archi- tektur als bahnbrechenden Ansatz für die natürliche Sprachverarbeitung herausragen. GPT hat 51 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 mehrere Iterationen durchlaufen, von denen jede einen Fortschritt in der Fähigkeit, mensch- liche Sprache zu verstehen, zu generieren und darauf zu reagieren, darstellt (Mehmke, 2023). Die erste Version von GPT führte das Konzept des Vortrainings und der Feinabstimmung ein. Sie nutzte eine auf Transformer basierende neuronale Netzwerkarchitektur, die in der Lage ist, kontextuelle Beziehungen zwischen Wörtern zu erfassen und zusammenhängende Texte zu generieren. GPT-2 markierte einen bedeutenden Fortschritt in der Entwicklung der Architek- tur, indem es globale Abhängigkeiten erfasste, kohärentere und kontextuell relevantere Texte generierte und somit beeindruckende Fähigkeiten zur Sprachgenerierung zeigte. Ein ganz be- sonderer Aspekt war seine Modellgröße, die 1,5 Milliarden Parameter umfasste.1 Diese Para- meter ermöglichten dem Modell, aus einem umfangreichen Korpus von Textdaten zu lernen. Mit beeindruckenden 175 Milliarden Parametern hat die kostenlose GPT-3-Version jedoch die- se Größenordnung übertroffen und kann eine Vielzahl von Aufgaben durchführen. Sie ist in der Lage, Chatbots und virtuelle Assistenten zu erstellen, maschinelle Übersetzungen von Tex- ten zu generieren oder Inhalte auszuarbeiten. Mittlerweile hat auch ChatGPT-4 eine erstaun- liche Kapazität von 10 Billionen Parametern erreicht und zeigt die außergewöhnliche Fähigkeit, menschenähnliche Texte zu verstehen und zu generieren und übertrifft somit die vorherigen Versionen in Bezug auf Kohärenz, Kontextverständnis und kreative Antworten (Mahnke, 2023). Mit der regelmäßigen Anwendung von ChatGPT haben viele Nutzer und Nutzerinnen ge- lernt, dass er bei korrektem Einsatz ein nützliches Werkzeug sein kann. Die Basis für die von ChatGPT verfassten Texte liegt in der korrekten Formulierung des Prompts2. Im Internet fin- den sich zahlreiche Regeln zur richtigen Verwendung von Prompts (z. B. Noack, 2023; Mey, 2023). Beispiele für ChatGPT Anfragen: Einfache Anfrage: Fachspezifische Anfrage: Interaktive Aufgaben: 1 Parameter sind die Werte innerhalb eines Netzwerks, die eine ungefähre Vorstellung von der Größe und Komplexität und damit auch Leistungsfähigkeit einer KI bestimmen. Sie enthalten die Informationen, die von einer KI, die auf künstlichen neuronalen Netzen basiert, verwendet werden (Seemann, 2023). 2 Ein „Prompt“ kommt aus dem Englischen und ist eine Ableitung des Verbs „to prompt“, was ungefähr bedeutet, einem Computer oder einer KI-Plattform eine Aufgabe zu geben, die sofort zu erledigen ist. Laut Gabler Wirtschaftslexikon (o. D.) kann ein textbasierter Prompt Wörter, Buchstaben, Sonderzei- chen, Zahlen und Links umfassen. Für ein zielführendes Ergebnis sollte der Prompt klar und umfassend formuliert sein. 52 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 Vordergründig müssen präzise und für die KI klar verständliche Anfragen gestellt werden (das bedeutet, mit ausführlichen Informationen sowie korrekter Rechtschreibung und Gram- matik). Je mehr Kontext die KI erhält, desto besser, auch weitere Folgefragen können dabei sehr hilfreich sein (vgl. Luge, 2023). Eindeutig erkennbar ist, dass KI-Instrumente – wie ChatGPT – ein umfassendes Training benötigen, was für Nutzer und Nutzerinnen bedeutet, Zeit und Wissen zu investieren, wie man so ein Instrument für die gewünschten Ergebnisse richtig einstellt, da sie auf Daten- optimierung basieren. Übertragen auf unsere Situation mit Studierenden, die mithilfe von ChatGPT selbständig lernen sollen, zeigen sich potenzielle Probleme, weil es zu fehlerhaften Informationen bei der Vermittlung kommen kann. Aus diesem Grund gehen wir später auch darauf ein, dass die Entwicklung von ChatGPT noch nicht abgeschlossen ist. Außerdem be- nötigen Studierende zunächst ein umfassendes, sehr zeitintensives Training im KI-Gebrauch, um ChatGPT richtig und effektiv anzuwenden. Zudem fehlen ihnen auch die nötigen Sprach- kenntnisse, um die KI korrekt zu trainieren, was eine noch größere Herausforderung dar- stellt. Es ist wichtig zu betonen, dass ChatGPT trotz fortgeschrittenem Training weiterhin Ein- schränkungen aufweist. Insbesondere im Bereich der wissenschaftlichen Erklärung können potenzielle Risiken auftreten, da das Modell nicht nur zur Texterstellung, sondern auch zur Darstellung wissenschaftlicher Zusammenhänge verwendet wird. Dabei besteht die Möglich- keit, dass neben fehlerhaft generierten Zitaten oder Quellen auch inkorrekte Informationen bereitgestellt werden. Dies liegt daran, dass der Chatbot in manchen Fällen sogenannte „Hal- luzinationen“ produziert, um eine konsistente Antwort zu generieren, die den Nutzer zufrie- denstellt, ungeachtet ihres Wahrheitsgehalts (Dongmo, Krüßmann & Weimann, 2023). Bei der Extrahierung von Termini zeigt sich das in falsch angewandten Wortarten und Zeitformen und in unaufgefordert fehlerhaften Grammatiktipps. 3 ChatGPT im Fremdsprachenunterricht Es ist mittlerweile weitläufig bekannt, dass mit ChatGPT Texte erstellt, überprüft und verbes- sert oder sogar in andere Formate umgewandelt werden können. Außerdem kann ChatGPT Aufsätze bewerten, Verbesserungsvorschläge machen, Unterrichtspläne erstellen sowie Aufgaben, Klausuren, Fragen, Szenarien usw. generieren (Rudolph, Tan & Tan, 2023). Ob- wohl bereits viele Lehrer und Lehrerinnen ChatGPT bei der Optimierung ihrer Arbeit und zur Reduzierung der Lehrerarbeitsbelastung nutzen, mangelt es noch an empirischen Studien, die seine Verwendung im Fremdsprachenunterricht erforschen. Die Untersuchung von Hein et al. (2024) über das Potenzial von ChatGPT zur Unterstützung von Lehrkräften hat gezeigt, dass er bei der Erstellung von Unterrichtsmaterialien, beim Zugang zu Informationen oder zur Entwicklung von Lerninhalten Hilfe leistet und somit ihre landesweit gestiegene Arbeits- belastung inklusive vermehrter Überstunden reduziert. In seiner Studie über die Auswirkungen von ChatGPT auf den Fremdsprachenunterricht und den Erwerb einer Fremdsprache stellte Hin Hong (2023) fest, dass Lehrkräfte dazu ermutigt werden, ChatGPT zur Minimierung ihrer Bewertungsarbeit anzuwenden, um mehr Zeit in die Unterrichtsvorbereitung investieren zu können. Gleichzeitig wurden jedoch auch die da- mit verbundenen Einschränkungen und Probleme, wie beispielsweise der unethische Einsatz von ChatGPT durch Studierende identifiziert. ChatGPT zeichnet sich durch die Integration des 53 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 breiten Spektrums an Lernmaterialien und -ressourcen für Nutzer und Nutzerinnen aus und ermutigt Lehrkräfte und Bildungseinrichtungen dazu, ihren Unterricht und ihre Bewertungs- methoden so anzupassen, dass die durch neue Technologien entstehenden Vorteile und Möglichkeiten genutzt werden können. In dieser Studie werden jedoch keine spezifischen Beispiele angeboten. Auch Huang und Li (2023) kommen zu der Feststellung, dass mittels KI immens viel Zeit und Mühe eingespart werden, welche die Lehrkräfte für die eigentlichen Unterrichtsaktivitäten intensiver nutzen und die Qualität sowie Effektivität ihres Unterrichts verbessern können, damit das Ziel personalisierter und effizienter Bildung erreicht wird. Koraishi (2023) fügt hin- zu, dass ChatGPT konsequent unter der Aufsicht von Experten und Expertinnen eingesetzt werden sollte, da die Möglichkeit von Fehlern, unsinnigen Ausgaben oder ungenauen Infor- mationen aufgrund seiner Grenzen stets präsent ist. In seiner Studie untersuchte Koraishi (2023) die vielseitigen Anwendungen von ChatGPT als wertvolles Tool im Werkzeugkasten von Englisch als Fremdsprache, wobei der Fokus insbesondere auf der Materialentwicklung und Bewertung lag. In Bezug auf den Wortschatzerwerb wurde festgestellt, dass Lehrer und Lehrerinnen durch das Entwerfen geeigneter Anweisungen und die Bereitstellung der not- wendigen Parameter die Fähigkeiten von KI nutzen und zahlreiche Zielvokabeln in generierte Texte integrieren können. Dies reduziert nicht nur die zuvor mit der Anpassung von Materia- lien verbundene Zeit und Mühe, sondern ermöglicht es Lehrer und Lehrerinnen auch, effi- zient auf die vielfältigen Wortschatzbedürfnisse ihrer Schüler und Schülerinnen einzugehen und so letztendlich das gesamte Lernerlebnis zu verbessern. Wie aus Studien zu ChatGPT im Fremdsprachenunterricht ersichtlich ist, sind diese rar und beschränken sich hauptsächlich auf die allgemeinen Vorteile und Möglichkeiten, ohne eine fundamentierte Analyse über die Leistung von ChatGPT darzubieten. Gerade in diesem Kon- text zeigt sich die Notwendigkeit, eine gezielte Untersuchung der Leistungsfähigkeit von ChatGPT-3 vorzunehmen, um spezifische Herausforderungen beim Lehren und Lernen von Fachterminologie im DaF-Unterricht zu identifizieren und anzugehen. 4 Zielgerichtetes Lehren und Lernen der Terminologie im DaF als Fachsprache mit KI Terminologie bezieht sich auf den „spezifischen Wortschatz eines bestimmten Gebiets“ (Arntz, Picht & Schmitz 2014, S. 11), obwohl sie oft verständlich erscheinen kann, selbst wenn man sich mit dem jeweiligen Fachgebiet nicht auskennt. Breite Fachgebiete, wie beispielswei- se Agrarwissenschaften oder Technik, reichen aber oft nicht aus, um wahrscheinliche Bedeu- tungen zu identifizieren. Die Terminologie ist auch nicht Teil der ‚muttersprachlichen‘ Kom- petenz. Sie muss gemeinsam mit dem Fachgebiet erlernt werden. Um Fachtermini zu identi- fizieren, kann man in Fachwörterbüchern oder einschlägigen Fachtexten nachschauen. Alles, was dort definiert wird, kann als Terminus bezeichnet werden (Hagemann, 2024, S. 145-146). Von zahlreichen DaF-Experten und -Expertinnen wird darauf hingewiesen, dass der Erwerb von Terminologie das Alpha und Omega des Fachsprachenunterrichts ist (Benouddane, 2016, S. 56). Obwohl viele das auch als eine Einschränkung des Fachsprachenunterrichts be- trachten, wird es meistens noch so interpretiert. Weil der Erwerb von Fachtermini als so 54 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 wichtig empfunden wird, steht auch die Frage im Vordergrund, wie und wie gut KI uns dabei unterstützen kann und wie hilfreich sie für die Studierende als alleinstehendes Lernhilfsmit- tel ist, wobei natürlich weiterhin die Sorge um die Richtigkeit des KI-produzierten Lerninhal- tes besteht. Im Sprachunterricht der Agrarwissenschaften haben wir traditionellerweise Fachsprachen- lehrer und Fachsprachenlehrerinnen, die sich mit den Themen der Landwirtschaft und der relevanten Fachsprache befassen und somit ein breites und verknüpftes Netzwerk an Kennt- nissen erwarben, die sie auch kreativ anwenden und den Studierenden vorstellen. In diesem Artikel wird untersucht, inwieweit KI diese Aufgabe bereits übernehmen kann. Die potenziel- le Zielgruppe für den Einsatz der Terminologie-Erstellung oder -Extraktion sind Studierende der Agrarwissenschaften, die im Rahmen ihres Studiums einen Deutschkurs für Fachsprache auf Niveau B1 bis C1 besuchen. Die Gruppen enthalten meist 6 bis 10 Teilnehmer. Der Einsatz von KI ist derzeit noch nicht direkt im Curriculum dieser Kurse verankert, wird jedoch zumin- dest erwähnt und mit den Studierenden diskutiert, insbesondere im Zusammenhang mit der eigenständigen Kurs-Lernplanung. 5 Untersuchungsgrundlagen In diesem Kapitel wird die Herangehensweise an die Datenerhebung für die vorliegende Stu- die beschrieben. Ziel der Studie war es zu untersuchen, wie gut ChatGPT in der Lage ist, Fachterminologie aus den Bereichen Landwirtschaft und Klimawandel zu erkennen und wie präzise ChatGPT-3 das grammatische Geschlecht (Maskulinum, Femininum oder Neutrum) und den Numerus von Substantiven angeben kann. Mit „gut“ ist der Prozentsatz der Richtigkeit der von ChatGPT-3 generierten Daten gemeint. Dadurch wollten wir auch herausfinden, wie verlässlich ChatGPT-3 als Selbstlern-Instrument für Studierende der Agrarwissenschaften ist. Die Forschungsfragen lauteten daher: (1) Wie präzise kann ChatGPT-3 Fachterminologie aus den Themenbereichen Landwirtschaft und Kli- mawandel erkennen? (2) Wie genau kann ChatGPT-3 das grammatische Geschlecht (Genus) oder die Anzahl (Numerus) von Substantiven angeben? Unter Einbeziehung der Daten zu Frage 1 und 2 wollten wir zudem herausfinden: Was sind die Vor- und Nachteile der Nutzung von ChatGPT-3 als Selbstlern-Instrument für Studierende der Agrarwissenschaften? Unsere Hypothese war, dass ChatGPT-3 Fehler machen würde und daher derzeit noch nicht als voll- ständig verlässliches Selbstlern-Instrument eingesetzt werden kann. Die Ergebnisse bestätig- ten diese Hypothese. Die Daten für die Analyse wurden sechs sorgfältig ausgewählten Artikeln aus Zeitungen und Fachzeitschriften entnommen, nämlich News.at, Bauernzeitung, umweltbundesamt.de, Wo- chenblatt und 3drei3.de (Professionelle Schweine Community). Alle Artikel stammen aus den Jahren 2022 und 2023, haben eine ungefähre Länge von 1000 Wörtern und wurden von ChatGPT-3 bezüglich des Sprachniveaus B2/C1 (Europäischen Referenzrahmen) bewertet. Thematisch konzentrieren sich die ausgewählten Zeitschriftenartikel auf die Landwirtschaft in Zeiten des Klimawandels. Dieses Thema ist aktuell und behandelt landwirtschaftliche Fra- gen im Kontext des grünen Wandels und der Klimakrise. Alle Texte sind öffentlich zugäng- lich und kostenlos verfügbar. Fachsprachenlehrer und Fachsprachenlehrerinnen erkennen 55 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 Wörter aus Bereichen wie Umweltverschmutzung, Wetter, Tierphysiologie usw. sofort als landwirtschaftliches Vokabular, da diese Bereiche eng mit der Landwirtschaft verknüpft sind. Die Daten aus den Zeitschriftenartikeln wurden innerhalb der ersten zwei Wochen im Novem- ber 2023 von ChatGPT-3 und den beiden Autorinnen bearbeitet. ChatGPT-3 hat für alle Artikel den folgenden Prompt erhalten: „Extrahiere aus diesem Text die deutschsprachige landwirt- schaftliche Fachterminologie unter Berücksichtigung des grammatischen Geschlechts und des Numerus der Substantive. Auch Adjektive und Verben sind systematisch einzubeziehen.“ Hier ist ein Beispiel was ChatGPT-3 produziert hat: • Veränderungen (noun): changes • Niederschlags (noun): precipitation • Klimawandels (noun): climate change • wirken (verb): have an effect, • Landwirtschaft (noun): agriculture Hier ist zu beobachten, dass ChatGPT-3 in Gegensatz zum Prompt den Artikel nicht ange- wandt hat. Beispiel 2: • Mais-Erträge (die, plural): maize yields • Weizenerträge (die, plural): wheat yields • zurückgehen (verb): decrease • steigen (verb): increase Das zweite Beispiel stammt aus einem anderen Text als das erste. Beide wurden mit dem- selben Prompt bearbeitet, jedoch hat ChatGPT-3 nur beim zweiten Text den Artikel des Subs- tantivs angegeben. Das Ziel der vorliegenden Forschung war es, die von ChatGPT-3 extrahierten Daten hinsicht- lich ihrer Eignung für das einwandfreie Lernen der Studierenden zu überprüfen und ihre Richtigkeit festzustellen. Die von ChatGPT-3 produzierten Daten wurden sowohl auf ihre Be- deutung als Agrartermini als auch auf oben erwähnten Grammatikkategorien (Genus und Numerus) geprüft. Nachdem ChatGPT-3 die Ergebnisse generiert hatte, haben wir die Richtig- keit der extrahierten Substantiv-, Verb- und Adjektiv-Fachtermini überprüft. Bei Unstimmig- keiten haben wir diese gemeinsam besprochen und bei Bedarf eine dritte Person zur Klärung hinzugezogen. Anschließend haben wir alle von ChatGPT-3 als Fachtermini identifizierten Be- griffe gezählt und überprüft. Dabei haben wir sorgfältig die Anzahl der korrekten und fehler- haften Auswahl ermittelt sowie die Fälle, in denen ChatGPT-3 das Genus und den Numerus richtig bestimmt hat. Auf dieser Grundlage wurde die Fehlerquote von ChatGPT-3 berechnet. Der Numerus war ursprünglich nicht im Prompt verlangt, wurde dennoch numerisch erfasst, da ChatGPT-3 diese Angabe von sich aus angeboten hat. Bei der Datenanalyse haben wir zudem verschiedene Beobachtungen festgehalten, die unserer Meinung nach für ein Selbst- lern-Instrument wichtig sind. Zum Beispiel zeigte ChatGPT-3 6bei demselben Prompt in un- terschiedlichen Texten teils unterschiedliche Ergebnisse an – in einem Fall gab es den Artikel an, im nächsten Fall jedoch nicht. Dies könnte für die Studierenden verwirrend sein. 56 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 Vor der Datenanalyse mit ChatGPT-3 haben sich die Autorinnen mit dem Tool vertraut ge- macht und das Verständnis für seine grundlegenden Möglichkeiten und Grenzen kennenge- lernt. Dies wurde durch die Erkundung der OpenAI-Dokumentation, Richtlinien und Beispiele durchgeführt, um das potenzielle Anwendungsspektrum des Modells zu erfassen. Wir haben gelernt, dass es vor dem Einsatz von ChatGPT-3 wichtig ist, das Gespräch durch klare An- weisungen und Kontext zu strukturieren. Begonnen wurde mit einer kurzen Einleitung, der Angabe des Zwecks und der Festlegung eventueller Einschränkungen, um die Antworten der KI zu steuern. Im Voraus festgelegte Erwartungen gewährleisten, dass das Gespräch auf Kurs bleibt und sich mit spezifischen Anforderungen abstimmt. Um die Qualität der Antworten von ChatGPT-3 zu verbessern, müssen präzise und gut kontextualisierte Prompts formuliert werden. Es empfiehlt sich, den Kontext explizit anzugeben und komplexe Fragen in kleinere, leichter handhabbare Teile aufzuschlüsseln, damit ChatGPT sie besser verstehen und effektiv beantworten kann. Beispiele angeben oder gewünschte Ergebnisse spezifizieren, kann eben- falls die Nützlichkeit der Antworten verbessern. Wir haben beim Überblick der Lernfähigkei- ten von ChatGPT-3 auch Daten und Systeme gefunden, wie die KI „trainiert” werden kann, um genauere Antworten zu erhalten. Dennoch haben wir uns bewusst entschlossen, dies nicht zu tun, um das realistische Szenario zu erhalten, wie ein Agrarstudent bzw. eine Agrarstu- dentin KI anwenden würde, um relevante Fachterminologie in Texten zu finden. Es ist höchst unwahrscheinlich, dass die Studierenden eine „Trainingsphase” von ChatGPT durchführen würden, um die Terminologie besser auszulesen. Nachdem die Anfrage an ChatGPT-3 gestellt wurde, hat die KI-Software eine Liste von Fach- termini aus den Artikeln herausgefiltert und sie unterschiedlich geordnet. Bei manchen Ter- mini wurde die Wortart bestimmt, aber nicht bei allen, bei einigen wurde der Artikel angege- ben (manchmal auch falsch, wie zum Beispiel „der Sorge”), bei anderen wieder nicht, was wir in der anschließenden Datenbearbeitung detaillierter vorstellen. 6 Ergebnisse Eine der Säulen des fachbezogenen DaF-Unterrichts ist das korrekte Erlernen des neuen Vo- kabulars. Diesbezüglich haben wir bei den ChatGPT-Extrahierungen (die Fachtermini und de- ren grammatische Eigenschaften, die von ChatGPT aus den Texten extrahiert und angegeben wurden) folgende Kategorien beobachtet, die für den Erwerb des Vokabulars und dessen spätere richtige Anwendung von großer Bedeutung sind: bestimmter Artikel (Genus), Wort- artangabe, Numerus, Bedeutungsrelevanz und nützliche Nominalkomposita. Dabei mussten wir in Erwägung ziehen, dass Kategorien wie Numerus und Nominalkomposita nicht explizit im Prompt verlangt, sondern trotzdem von ChatGPT angegeben wurden. Da es sich hierbei um grundlegende Daten für das präzise und erfolgreiche Lernen der Studierenden handelt, war es unser Ziel, deren Genauigkeit zu evaluieren. Die von ChatGPT-3 generierten Daten wurden sowohl hinsichtlich der korrekt extrahierten Grammatikkategorien, wie der Angabe und Richtigkeit des Artikels, als auch der Wortartzuordnung und des Numerus überprüft, wie in Tabelle 1 dargestellt. Tabelle 2 analysiert die Fachrelevanz der extrahierten Terminologie im Kontext der Agrarwissenschaften. In Tabelle 1 wird die Zuordnung grammatikalischer Kategorien für Fachtermini aus allen sechs untersuchten Zeitungsartikeln dargestellt. Sie enthält sowohl die Gesamtanzahl der 57 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 von ChatGPT-3 extrahierten Kategorien als auch die von den Forscherinnen bzw. Fachspra- chelehrerinnen festgelegten Kategorien zur Überprüfung der Ergebnisse. Tabelle 1 Summe der grammatikalischen Kategorien für Fachtermini Grammatische Kategorien Gesamtanzahl der Kategorien von ChatGPT-3 Gesamtanzahl der Kategorien ermittelt durch Forscherinnen/ Fachsprachelehrerinnen Die korrekte Angabe des bestimmten Artikels von der Gesamtzahl aller extrahierten Nomen 387 411 Die korrekte Angabe der Wortart von der Gesamtzahl aller extrahierten Wörter 183 552 (nur bei Adjektiven oder Verben wurde die Wortart angegeben, sonst keine) Die korrekte Angabe des Numerus von der Gesamtzahl aller extrahierten Nomen 178 430 Aus Tabelle 1 geht hervor, dass ChatGPT-3 nur bei 94,16 Prozent der extrahierten Substanti- ve das korrekte Genus angibt. Zudem treten bei den angegebenen Artikeln 9,2 Prozent Fehler auf, wie beispielsweise „das CO2-Bilanz“ oder „der Treibhausgas“. Bei der Kategorie „Wortart von der Gesamtzahl aller extrahierten Wörter“ wurde die Wortart in 33,15 % der Fälle korrekt angegeben. Dabei war die Wortart ausschließlich bei Adjekti- ven und Verben spezifiziert. In einem anderen Text, der mit demselben Prompt bearbeitet wurde, wurden hingegen keine Wortarten angegeben. Bei der Kategorie „Numerus von der Gesamtzahl aller extrahierten Nomen” wurde die Wortart in 41,39 % der Fälle korrekt ange- geben. Der Numerus wurde lediglich bei zwei von drei Texten angegeben, obwohl alle Texte mit demselben Prompt bearbeitet wurden. In Tabelle 2 sind alle relevanten Fachtermini für Agronomie aus den sechs Artikeln zusam- mengefasst, ebenso wie die Gesamtzahl der von ChatGPT-3 extrahierten Fachtermini. Tabelle 2 Darstellung der relevanten Agronomie-Fachtermini Relevanz der Fachtermini ChatGPT-3 (Gesamtzahl der Fachtermini) Gesamtzahl der Fachtermini ermittelt durch Forscherinnen/ Fachsprachelehrerinnen Relevant für Agronomie von allen relevanten Termini 452 602 Irrelevant für Agronomie 52 / Angegebene relevante Nominalkomposita im Verhältnis zu allen relevanten Nominalkomposita 234 388 58 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 In Tabelle 2 wird ersichtlich, dass ChatGPT-3 lediglich 75,08 Prozent der relevanten Fach- termini aus den Texten korrekt erkannt hat. Zudem hat ChatGPT 52 Wörter als Fachtermini klassifiziert, die nicht als solche einzustufen sind, wie beispielsweise „geringeres“ oder „kom- pensieren“. Besonders bei Nominalkomposita hat ChatGPT nur 77,08 Prozent als relevant identifiziert. 7 Diskussion und Handlungsempfehlungen Die oben dargestellten Daten verdeutlichen sowohl positive als auch negative Aspekte der Verwendung von ChatGPT 3.0 im fachbezogenen DaF-Unterricht. Es sollte jedoch berück- sichtigt werden, dass inzwischen neuere Versionen der KI, wie ChatGPT 4.0 oder ChatGPT 4.1, verfügbar sind, die möglicherweise zu veränderten Analyseergebnissen führen könnten. Die von ChatGPT bereitgestellte Unterstützung kann auch motivierend wirken, da der erste Schritt, das Heraussuchen der Termini, eingeleitet wurde. Unverzichtbar ist eine sorgfältige Überprüfung, vorzugsweise durch einen Fachdeutschlehrer oder eine Fachdeutschlehrerin. ChatGPT eignet sich auch hervorragend als Unterrichtsinstrument, um die Studierenden zum kritischen Nachdenken anzuregen. Es fördert die Auseinandersetzung mit den tatsächlichen Möglichkeiten der KI und ermutigt sie, ihre eigene Informationsbasis bezüglich Fachtermini zu überprüfen. Andererseits ist festzustellen, dass ChatGPT gemäß den oben genannten Daten weiterhin zu viele Fehler bei der Extraktion von Fachtermini im Bedeutungsrahmen sowie bei der Bestim- mung von grammatischen Kategorien wie Genus oder Numerus aufweist. ChatGPT hat zwar bei 94,16 Prozent der extrahierten Nomen das Genus korrekt angegeben, jedoch bedeutet dies, dass 5,84 Prozent der Genera potenziell falsch von den Studierenden erlernt werden könnten. Das bedeutet wiederum, dass ChatGPT im Moment noch nicht als selbstständige Lernhilfe von Studierenden benutzt werden kann. In der Kategorie „Wortart von der Gesamt- zahl aller extrahierten Wörter” wurde die Wortart lediglich bei 33,15 Prozent der Fälle korrekt bestimmt. Bei den verbleibenden 66,85 Prozent besteht die Möglichkeit von Fehlern, falls die Studierenden die Wortart eigenständig ermitteln müssen. Ähnliches gilt für die Kategorie „Numerus von der Gesamtzahl aller extrahierten Nomen”, in der die Wortart nur bei 41,39 Prozent der Nomen korrekt bestimmt wurde. In den Bedeutungskategorien hat ChatGPT lediglich 75,08 Prozent aller relevanten Termini aus den Texten identifiziert. Darüber hinaus wurden 52 Begriffe von ihm fälschlicherweise als Fachtermini klassifiziert, obwohl sie nicht als solche gelten können. Zudem hat ChatGPT nur 77,08 Prozent aller relevanten Nominalkomposita angegeben. Diese Ungenauigkeiten könnten zu Problemen bei der Interpretation von Bedeutungen und beim Leseverständnis der Studierenden führen, da Bedeutungen häufig aus dem Kontext der Wörter erschlossen werden. Natürlich weist auch diese Studie, wie jede andere, gewisse Einschränkungen auf. Der ers- te Aspekt, den wir hervorheben möchten, ist die begrenzte Anzahl der analysierten Tex- te. Eine umfassendere Textbasis würde ein realistischeres Bild liefern und zu fundierteren Schlussfolgerungen führen. Daher empfehlen wir, eine ähnliche Studie in einem erweiter- ten Rahmen durchzuführen. Ein weiteres Problem betrifft die Auswahl der Termini und der 59 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 grammatikalischen Merkmale durch die Forschenden und Lehrenden. Es besteht die Mög- lichkeit, dass auch Wissenschaftler und Wissenschaftlerinnen bei der Auswahl von Termini Fehler machen. In unserer Studie haben wir jedoch sichergestellt, dass die Auswahl der Ter- mini durch gegenseitige Überprüfung unter den Forschenden validiert wurde. Zusätzlich gibt es Herausforderungen bei der Definition des „Fachterminus“. Die Forschenden benötigen einen klareren Konsens darüber, welche Begriffe als Fachtermini betrachtet werden sollen. Im Folgenden werden wir uns den Schlussfolgerungen zuwenden. 8 Schlussfolgerungen Im schnelllebigen Technologiebereich ist es selten, dass eine neue Entwicklung in so kur- zer Zeit für Aufsehen sorgt wie ChatGPT. In nur wenigen Monaten hat dieses revolutionäre KI-Sprachmodell die Welt im Sturm erobert. Während die technikaffinen und engagierten Lehrer und Lehrerinnen ihre beeindruckende Funktionalität feiern, betrachten viele dies mit Staunen und zugleich mit Bedenken. Die EU hat politische Maßnahmen und Initiativen zur Verbesserung der digitalen Kompe- tenzen von Arbeitnehmern und Verbrauchern entwickelt. Digitale Kompetenzen sind nicht nur im Privatleben, sondern auch am Arbeitsplatz von entscheidender Bedeutung, da über 90 Prozent der Berufe in Europa digitale Grundkenntnisse erfordern. Trotzdem fehlen diese rund 42 Prozent der europäischen Bevölkerung, darunter 37 Prozent der Erwerbstätigen. Es ist ebenso von Bedeutung, unsere digitalen Fähigkeiten über das Grundniveau hinaus zu er- weitern, die in verschiedenen Bereichen, wie digitalem Marketing und sozialen Medien, von großem Nutzen sein können. (Europäische Kommission, o. D.) Zusätzlich gewinnen digita- le Kompetenzen am Arbeitsplatz zunehmend an Bedeutung, weshalb auch Studierende der Agrarwissenschaften ihre Kompetenzen verbessern sollten. Obwohl KI-Tools wie ChatGPT die Lernautonomie der Studierenden fördert und Lehrenden die Unterrichtsvorbereitung er- leichtert, sollte man sie verantwortungsvoll nutzen, um u. A. keine Urheberrechtsverletzun- gen zu begehen. Die weltweite Rechtslage zu ChatGPT und ähnlichen KI-Technologien variiert stark; einige Länder haben sich für strikte Regulierungen oder sogar Verbote entschieden (z.B. China), während andere einen freizügigeren Ansatz verfolgen (siehe „Ist es verboten, ChatGPT zu nutzen? Verständnis der Regulierungen“, o. D.). In letzter Zeit wurden auch Sank- tionen für die Reproduktion von Materialien anderer ohne entsprechende Quellenangabe eingeführt. Terminologie ist für angehende Agrarwissenschaften-Experten und -Expertinnen nach wie vor sehr bedeutsam, da sie einen der Bausteine ihres Verstehens und ihrer Kommunikation im Arbeitsbereich darstellt. Deshalb muss beachtet werden, dass KI von den Studierenden nicht eigenständig angewendet werden kann, bis wir sicher sind, dass sie Terminologie ein- wandfrei erkennt. Der Einsatz von ChatGPT kann beim Selbstlernen erhebliche Nachteile mit sich bringen, da die KI die Termini offensichtlich nicht miteinander verknüpft und sie daher nicht als zusam- menhängend wahrnimmt. Infolgedessen ist es sehr wahrscheinlich, dass die Studierenden sich im Falle der Nutzung von ChatGPT nur unzureichend oder gar nicht mit der relevanten Terminologie auseinandersetzen würden. 60 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 Ein Beispiel hierfür ist das Verb „wirken“, welches von ChatGPT in unserem Fall fälschlicher- weise als Fachterminus identifiziert wurde. Studierende könnten dieses Verb ausschließ- lich im Fachkontext verwenden, was zu einem begrenzten und ungenauen Sprachgebrauch führen würde. Zudem könnte eine Kollokation wie „sorgfältig untersuchen“, die sowohl in wissenschaftlichen als auch in alltäglichen Kontexten verwendet wird, von den Studieren- den aufgrund der fehlerhaften Zuordnung durch ChatGPT nur in wissenschaftlichen Zusam- menhängen verwendet werden. Dies würde ihren Wortschatz unnötig einschränken und die Sprachkompetenz negativ beeinflussen. Unsere Empfehlung wäre, ChatGPT für Übungen im Unterricht zu nutzen, in denen die Stu- dierenden mit den Lehrenden diskutieren können, welche Fehler bei der Terminologie-Ex- trahierung von KI vorkommen und wieso es möglicherweise dazu kommt. Es wäre zudem sinnvoll, detaillierter darüber zu diskutieren, wie KI trainiert werden kann, um Fehler zu ver- ringern, oder Vergleiche anzustellen, wie die KI lernt und wie die Studierenden selbst Sprache erwerben. Die Erfolgsquote von ChatGPT bei der Extrahierung von Fachtermini ist also unserer Meinung nach noch ungenügend, um ihn als ein Selbstlerninstrument einzusetzen. Die Studierenden, die ihn nutzen möchten, benötigen für den Erwerb korrekter Informationen und Interpretatio- nen noch viel Unterstützung durch Fachsprachenlehrer und Fachsprachenlehrerinnen. Gemäß dieser Studie kann ChatGPT eine interessante Möglichkeit für die Entwicklung von Lehrmate- rialien sein, deren zielgerichteter Einsatz bei den Studierenden der Agrarwissenschaften eine Brücke zwischen fachspezifischer Terminologie und effektivem Sprachunterricht mit ChatGPT baut. Die gewonnenen Erkenntnisse liefern wertvolle Einblicke in den Einsatz von ChatGPT im Bildungskontext und unterstützen eine realistische Integration von KI in den Lehrplan. Literatur AI. (2024). AI generierte Inhalte für Bildung und Lernen – FasterCapital. FasterCapital. https://fasterca- pital.com/de/inhalt/AI-generierte-Inhalte-fuer-Bildung-und-Lernen.html Arntz, R., Picht, H., & Schmitz, K. (2014). Einführung in die Terminologiearbeit. Olms. Benouddane, A. (2016). Beitrag der Fachsprache zum Wortschatzlernen im DaF-Unterricht. Traduction et Langues, 15(2), 55–63. https://doi.org/10.52919/translang.v15i2.687 Damberger, T. (2017). Von der Abschaffung des Lehrers Die Zukunft des selbstgesteuerten Lernens mit digi- talen Medien. Friedrich Verlag GmbH. https://www.friedrich-verlag.de/friedrich-plus/schulleitung/ digitale-schule/von-der-abschaffung-des-lehrers-1449 Dongmo, J., Krüßmann, M., & Weimann, F. (2023). ChatGPT – Dein Freund und Helfer im Hochschulalltag? https://www.fh-muenster.de/itb/downloads/lehre-neu/michel-fabian/ChatGPT_-_Dein_Freund_ und_Helfer_im_Hochschulalltag.pdf Europäische Kommission. (o. D.). Digitale Kompetenzen. https://digital-strategy.ec.europa.eu/de/ policies/digital-skills Gabler Wirtschaftslexikon. (o. D.). Prompt. Definition: Was ist „Prompt“? Springer Gabler. https://wirt- schaftslexikon.gabler.de/definition/prompt-125087 Hagemann, S. (2024). Recherche im Translationsprozess: Ein Lehr- und Studienbuch. Arbeiten zur Theorie und Praxis des Übersetzens und Dolmetschens, Band 129. Frank & Timme GmbH. 61 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 Hein, L., Högemann, M., Illgen, KM., Stattkus, D., Kochon, E., Reibold, M.-G., Eckle, J., Seiwert, L., Beinke, J. H., Knopf, J., & Thomas, O. (2024). ChatGPT als Unterstützung von Lehrkräften – Einordnung, Analyse und Anwendungsbeispiele. HMD Praxis der Wirtschaftsinformatik, 61, 449–470. https:// doi.org/10.1365/s40702-024-01052-9 Hin Hong, W. C. (2023). The impact of ChatGPT on foreign language teaching and learning: Opportuni- ties in education and research. Journal of Educational Technology and Innovation, 3, 37–45. https:// doi.org/10.61414/jeti.v5i1.103 Huang, J., & Li, S. (2023). Opportunities and challenges in the application of ChatGPT in foreign langua- ge teaching. International Journal of Education and Social Science Research, 6(4), 75–89. https://doi. org/10.37500/IJESSR.2023.6406 „Ist es verboten, ChatGPT zu nutzen? Verständnis der Regulierungen”. (o. D.). Herfurtner. https://kanz- lei-herfurtner.de/chatgpt-verboten/ Koraishi, O. (2023). Teaching English in the Age of AI: Embracing ChatGPT to Optimize EFL Materials and Assessment. Language Education & Technology, 3(1), 55–72. Luge, R. (2023). Die 50 besten ChatGPT Prompts. Kopf & Stift. https://kopfundstift.de/chatgpt-prompts/ Mahnke, F. (2023). GPT-4: Der neue Heilige Gral der KI-Sprachmodelle? OMR Reviews. https://omr.com/ de/reviews/contenthub/chatgpt-4 Mehmke, S. (2023). ChatGPT von OpenAI: Weitere Verbesserungen angekündigt. ComputerBild. https:// www.computerbild.de/artikel/ChatGPT-deutsch-kostenlos-testen-35038619.html Meršnik, M., Đorđević, D., & Anđelković, J. (2023). Tailor-made ESP teaching materials for HEIs – two case studies. Scripta Manent, 18(1), 93–102. Mey, S. (2023). Chat GPT: Mit diesen Tipps schreibt die künstliche Intelligenz bessere Texte. Der Stan- dard (05.01.2023). https://www.derstandard.de/story/2000142293035/chat-gpt-mit-diesen- tipps-schreibt-die-kuenstliche-intelligenz-bessere Mykhalevych, N. (2024). Menschliches Lernen vs. KI: Die Bedeutung von sozialer Interaktion. Preply. https://preply.com/de/blog/lernen-mit-ki/ Noack, A. (2023). Wie man perfekte Prompts für ChatGPT verfasst. Künstliche Intelligenz: Ratgeber. https://digital-magazin.de/perfekte-prompts-fuer-chatgpt-verfassen/ Rudolph, J., Tan, S., & Tan, S. (2023). ChatGPT: Bullshit spewer or the end of traditional assess- ments in higher education? Journal of Applied Learning and Teaching, 6(1), 342–363. https://doi. org/10.37074/jalt.2023.6.1.9 Seemann, M. (2023). Künstliche Intelligenz, Large Language Models, ChatGPT und die Arbeitswelt der Zu- kunft. Working Paper Forschungsförderung. Hans-Böckler-Stiftung. 62 Težak, Orthaber / Scripta Manent XIX/1 (2024), 49–62 Povzetek Poučevanje kmetijskega strokovnega besedišča s pomočjo umetne inteligence ChatGPT: prednosti in pomanjkljivosti Vključevanje umetne inteligence (UI), kot je ChatGPT, v poučevanje tujega jezika stroke ponuja bodočim strokovnjakom in strokovnjakinjam kmetijstva nove možnosti za usvajanje strokovnega besedišča in razvijanje jezikovnih zmožnosti. V prispevku proučujemo uporabo ChatGPT-ja za usvajanje strokovne terminologije pri pouku nemščine kot jezika stroke na eni izmed slovenskih univerz. S pomočjo analize smo poskušali ugotoviti, kako uspešen je pri jezikovni obdelavi strokovnih besedil s področja kmetij- stva, zelenega prehoda in trajnostnosti. V tem okviru smo proučili, katere izraze ChatGPT-3 prepozna kot pomembne za poučevanje nemškega jezika stroke. Pri tem smo izračunali stopnjo zanesljivosti uporabe tega orodja, in sicer tako, da smo terminologijo, ki jo je na podlagi ukaza ponudil ChatGPT-3, primerjali z naborom terminov, ki so jih iz istega gradiva izluščili raziskovalci. Ugotovili smo, da Chat- GPT-3 sicer ponudi termine, ki so pomembni za študente in študentke kmetijskih ved, a tudi takšne, ki so lahko zavajajoči. Obenem smo preučili tudi koristnost in zanesljivost dodatnih informacij, ki jih je izpisal ChatGPT, kot na primer slovnični spol in število. Študija bi lahko prispevala k razvoju ciljno usmerjenih učnih gradiv za študente in študentke kmetijskih ved, pri čemer bi študenti in študentke sami izboljšali seznam terminov, ki jih je izbral ChatGPT, ali poiskali napake. Ključne besede: ChatGPT, umetna inteligenca, nemščina kot tuji jezik, strokovno besedišče, kmetijske vede 63 Scripta Manent XIX/1 (2024) DOI: 10.4312/SM.19.1.63-80 Research Paper Paper received: 25.02.2024 Paper revised: 12.04.2024 Paper accepted: 05.05.2024 Paper published: 30.11.2024 Nadja Dobnik Faculté d’Économie, Université de Ljubljana Département de traduction, Faculté des Lettres, Université de Ljubljana LES ENJEUX DE L’INTELLIGENCE ARTIFICIELLE : ENSEIGNER ET APPRENDRE A TRADUIRE SANS ORDINATEUR – UTOPIE OU DEFI ? Resumé Dans le domaine de la didactique de la traduction professionnelle, les outils de traduction automatique font partie intégrante de l’enseignement et de l’évaluation depuis de nombreuses années. Plus les outils de traduction automatique se sont développés et sont devenus indissociables des métiers de la traduc- tion, plus l’enseignement s’est appuyé sur l’utilisation de ces outils. En conséquence, il est devenu de plus en plus difficile d’évaluer l’authenticité des traductions et la compétence réelle de l’étudiant. Ces ques- tions sont devenues particulièrement pertinentes pendant et après la pandémie du Covid-19 et l’arrivée des systèmes de traduction neuronale lorsque les étudiants se sont habitués à faire leurs traductions à l’aide des outils de traduction automatique. Cet article présente les résultats d’une recherche visant à évaluer les traductions du français vers le slovène faites par les étudiants sans ordinateur, comparées à des traductions effectuées à l’aide d’outils de traduction automatique. Les résultats sont intéressants et ouvrent certaines pistes de réflexion sur les méthodes d’enseignement à appliquer. Mots clés : didactique de la traduction, traduction automatique, intelligence artificielle, évaluation en traduction, TA neuronale Abstract The Challenges of Artificial Intelligence: Teaching and Learning to Translate Without Computers – Utopia or Opportunity? In the field of translation didactics, computer-assisted and machine translation tools have long been an integral part of translator training and the assessment of translation competence. The more such translation tools have become developed, the more prominent role they have been assigned not only in professional translation practices, but also in translator training. As a result, assessing the authenticity of students’ translations and their actual translation competence has become increasingly difficult. These issues have become particularly challenging during the Covid-19 pandemic. The advent of generative AI translation tools has resulted in students’ excessive and uncritical reliance on these tools. This paper presents the findings of a case study whose aim was to compare and assess student translations from French into Slovene made without computers with translations generated with the assistance of com- puters and the full range of functionality that this affords. The results are highly informative and reveal a number of points that merit reflection on the teaching methods in translator training. Keywords: translation didactics, machine translation, artificial intelligence, translation assessment, generative AI translation tools 64 Dobnik / Scripta Manent XIX/1 (2024), 63–80 1 Introduction La révolution numérique et l’informatisation des services linguistiques ont profondément changé le secteur des métiers de la traduction. En l’espace d’une décennie, les traducteurs automatiques sont passés du statut d’outils obscurs et maladroits à celui d’outils fiables et largement utilisés qui augmentent considérablement la productivité, réduisent les coûts et permettent à de nombreux utilisateurs ne maîtrisant pas la langue d’accéder à des contenus en langue étrangère et de communiquer dans cette langue. Les progrès incontestables de la traduction automatique (TA) depuis l’arrivée de la TA neuronale obligent les formations universitaires à s’adapter et à repenser leur façon de former les étudiants aux futurs métiers de la traduction. A l’heure actuelle, de nombreuses études, recherches et expériences en matière de TA et, notamment, de l’enseignement de la TA permettent d’imaginer l’ampleur des enjeux (Loock, 2019 ; Massey et Ehrensberger-Dow, 2017 ; Moorkens, 2018 ; Faria Pires, 2018 ; Guerberof Arenas et Moorkens, 2019 ; Martikainen, 2019). Les questions se multiplient et les pistes de réflexion sont nombreuses. Néanmoins, il incombe à chaque institution de formation et à chaque formateur de trouver et appliquer les méthodes les plus adaptées à leur situation d’enseignement, en fonction du profil des étudiants, du niveau d’études, des langues ensei- gnées et d’autres spécificités propres à la formation. Dans cet article, nous présentons les résultats d’une recherche visant à évaluer la faculté des étudiants en troisième année de licence à traduire sans outils de TA par rapport aux tra- ductions faites avec des outils de TA. L’idée de l’analyse est née de l’observation de plusieurs générations d’étudiants en licence et, tout particulièrement, de la génération ayant étudié à distance pendant la pandémie du Covid-19. Pour cette génération, nous avons constaté une forte dépendance à l’égard des outils de TA, qui se traduisait principalement par une attitude non critique à l’égard de la qualité de leurs traductions et une confiance excessive dans leur capacité à réviser les traductions réalisées à l’aide d’outils informatiques. L’obser- vation de cette génération nous a amenés à différentes questions : Faudrait-il repenser les compétences de traduction en fonction des progrès de la TA neuronal ? Comment préparer les étudiants aux futures réalités du secteur de la traduction et quelles compétences déve- lopper ? Autoriser, limiter ou interdire le recours à la TA lors des examens ? Et finalement, comment évaluer les réelles compétences de traduction d’un étudiant qui traduit avec les outils de TA ? 2 La didactique de la traduction face à la traduction automatique neuronale Les développements technologiques ont eu un impact significatif sur la profession de traduc- teur dès l’arrivée des outils informatiques et l’avènement d’internet. Parallèlement à l’infor- matisation des métiers de la traduction dans les années 1990 et 2000, les progrès de la tra- duction automatiques et son intégration dans les institutions et entreprises internationales ont progressivement modifié les conditions de travail des traducteurs et les exigences en matière de maîtrise des outils de traduction. Ainsi, depuis les années 2000, les employeurs – agences de traduction, grandes entreprises, organisations internationales – s’attendent à 65 Dobnik / Scripta Manent XIX/1 (2024), 63–80 ce que les traducteurs maîtrisent les technologies de traduction et font des qualifications formelles une condition d’embauche. Force est de constater que les employeurs de traducteurs, tels que les organisations internationales, relèvent globalement une différence notable entre les traducteurs diplômés et les autres, en faveur des premiers, constate Christine Durieux (Durieux, 2005, 36). L’enjeu est double car, d’une part, les employeurs commencent à exiger des can- didats une formation spécifique voire un diplôme d’études supérieures et, d’autre part, les formations sont censées répondre à cette exigence et former leurs étudiants aux réalités en matière de la traduction pratique dans de nombreux secteurs. Le grand changement qui s’est produit avec les progrès de la TA neuronale a imposé au secteur de la traduction, en l’espace de quelques années, une nouvelle réalité : dans de nombreux cas, la traduction automatique a commencé à remplacer le travail des traduc- teurs humains. C’est le cas de grandes institutions internationales telles que la Commission européenne et sa Direction générale de la traduction (DGT) de la Commission européenne qui a mis en place, en 2017, son traducteur neuronal eTranslation. Les nouvelles méthodes standard de travail dans le secteur de la traduction professionnelle et technique sont ainsi devenues post-édition et révision, avec de multiples interrogations non seulement pour les traducteurs mais également pour les formateurs et chercheurs, comme en témoignent les résultats de l’enquête annuelle sur l’industrie linguistique européenne (ELIS, 2022). Parmi les principales préoccupations figurent la crainte que la TA remplace des emplois de traducteurs, mais aussi la diminution du nombre d’étudiants dans les programmes de traduction (alors que l’enquête confirme qu’il y a même un besoin croissant de linguistes dans le secteur). Depuis plus d’une décennie, ces interrogations se reflètent dans le nombre croissant et très varié de recherches et de publications sur la didactique de la traduction face au «mariage for- cé» entre la traduction automatique et la traduction humaine (Breyel-Steiner et Grass, 2021). Dans le contexte de notre recherche, certains domaines de recherche sont particulièrement pertinents : I. Intégration de la traduction automatique et de la post-édition dans les pro- grammes de formation ; II. Évaluation des compétences des étudiants utilisant la traduction automatique ; III. Développement de méthodes d’enseignement innovants et spécifiques à la traduction automatique. Une grande partie des études se penchent sur les modalités à adopter pour intégrer les outils de la TA dans l’enseignement, afin de permettre aux étudiants d’avoir à la fin de leurs études «toutes les cartes en main pour entrer le plus sereinement possible sur le marché professionnel de la traduction» (Faria Pires, 2018), de mettre en lumière ce que la machine peut faire et ne peux pas faire (Martikainen, 2019; Loock, 2019), de sensibiliser les étudiants aux avantages et aux risques de l’utilisation des outils de traduction en ligne; de démystifier la TA dès le début de la formation (Moorkens, 2018; Loock, 2019). De nombreuses études portent aussi sur l’impact de l’utilisation de la TA sur les compétences des traducteurs, telles que leur capacité à comprendre et à analyser les textes sources, leur précision dans la production de traductions et de la post-édition, leur sensibilité aux nuances linguistiques et culturelles. Les recherches portent également sur les nombreux effets néga- tifs d’une attitude passive et non critique à l’égard de la TA. Une place particulière est accor- dée à la formation dans le domaine de la révision et de la post-édition, aux typologies des erreurs (Vaupot, 2021 ; Hirci et Pisanski Peterlin, 2020). 66 Dobnik / Scripta Manent XIX/1 (2024), 63–80 Malgré un certain pessimisme, de nombreuses expériences témoignent d’une grande vitali- té dans la mise en place de démarches pédagogiques innovantes, notamment de coopéra- tions entre universités, de mise en place de cours de post-édition (Faria Pires, 2018) ou de simulation de contextes professionnels de post-édition (tel que l’organisation de skills labs), y compris l’utilisation de scénarios d’apprentissage basés sur des projets (Hirci, Mikolič Južnič et Pisanski Peterlin, 2019), des études de cas et des exercices pratiques utilisant des outils de traduction automatique; ces démarches sont d’ailleurs parfaitement en ligne avec l’enseigne- ment de la traduction par approche communicative, consistant à mettre les étudiants dans des situations de simulation des conditions d’exercice de la profession (Durieux, 2005) et tout particulièrement avec l’approche interprétative, mettant l’accent sur la compréhension (Lederer, 2020 ; Dancette, 2003). 3 Contexte La recherche est basée sur les traductions de trois groupes d’étudiants, au total 20 étudiants, de 3ème année du Département de traduction de la Faculté des Lettres de l’Université de Lju- bljana, inscrits au cours Traduction du français vers le slovène 3. Il est important de souligner qu’il s’agit du premier cycle d’études donc de la formation initiale en début d’études, destiné à enseigner aux étudiants l’ensemble des compétences requises pour la traduction, y compris la langue de départ, la culture générale, une bonne capacité rédactionnelle en langue d’arri- vée, ainsi que la compétence traductionnelle proprement dite. En plus des matières linguistiques générales (grammaire, langue, civilisation), tous les étudiants ont suivi un certain nombre de cours portant sur la traduction (traduction de l’anglais, introduc- tion aux technologies de la traduction et à la traduction automatique). Le cours de Traduction du français vers le slovène est un cours de traduction générale, non spécialisée, enseigné dès la première année et portant sur des thématiques très variées. Les étudiants s’initient à la fois aux techniques de la traduction et à la connaissance de différents types de documents : textes commerciaux, juridiques, journalistiques, littéraires etc. La traduction spécialisée (droit, écono- mie, médecine, sous titrage) est enseignée par la suite au niveau de master. L’analyse que nous avons entreprise s’inscrit dans le cadre d’un suivi attentif du travail des étudiants pendant le Covid-19. Surpris par une grande dépendance des étudiants à l’égard des outils de TA, en particulier DeepL, une attitude non critique à l’égard de la qualité de leurs traductions et même l’incapacité de traduire sans recours aux outils de TA, nous avons introduit quelques changements dans le travail en classe. Avec deux générations d’étudiants en traduction (3eme année de licence) inscrites après le Covid-19, nous avons entrepris en classe de traduction du français vers le slovène une approche beaucoup plus active et «in- terprétative» (selon la démarche de D. Seleskovitch et M. Lederer), basée sur un plus grand nombre d’exercices oraux - traduction à vue, résumé d’articles de presse, exercices d’impro- visation, de reformulation et de mise en scène - avant de passer à la traduction sur l’ordina- teur. En plus de créer une ambiance agréable et détendue, ces courts exercices oraux ont eu un impact positif sur la confiance en soi des étudiants et ont considérablement dissipé leur peur de se produire devant les collègues et d’être jugé pour leurs erreurs. En complément de la méthode principale qui consiste dans la traduction de divers textes et la discussion sur les enjeux rencontrés, ces exercices représentaient jusqu’à un quart du cours. Ont fait 67 Dobnik / Scripta Manent XIX/1 (2024), 63–80 partie de ces activités aussi des exercices de traduction de courts textes sans l’utilisation de l’ordinateur. Encouragés par les retours positifs de la part des étudiants, nous avons décidé de faire écrire aux étudiants de la plus récente génération les deux partiels sans l’utilisation de l’ordinateur. Les résultats des deux partiels nous ont positivement surpris à plusieurs égards. Premiè- rement, les résultats étaient tous positifs et plutôt bons. Deuxièmement, les étudiants ont tous relevé le défi de traduire sans ordinateur sans le moindre signe de mécontentement ou d’angoisse. Ils ont tous traduit le document entier et n’ont omis aucune partie (phrase ou paragraphe). Troisièmement, les traductions témoignent d’une lecture plus attentive et d’une compréhension plus précise ; les reformulations sont bonnes et exactes, même si les étudiants ne connaissaient pas tous les mots ou choisi le mauvais sens d’un mot ou d’une expression. Et finalement, les étudiants n’ont pas inventé de solutions incohérentes et impro- visées quand ils ne comprenaient pas un mot ou un passage mais ont mis dans la traduction soit l’expression française soit un espace vide entre parenthèses. Ce premier constat, basé sur l’observation et en partie intuitif, nous a amené à l’idée de comparer ces traductions avec celles des générations précédentes faites sur l’ordinateur. Il se trouve que les deux documents ont été traduits par les générations précédentes, notam- ment par la génération du Covid-19. 4 Recherche L’objectif de cette recherche est de déterminer dans quelle mesure les traductions réalisées par les étudiants sans utiliser d’ordinateur diffèrent des traductions faites par deux autres groupes d’étudiants, dans des conditions comparables, avec l’utilisation de l’ordinateur et avec le recours (limité) aux ressources sur internet. Nous avons émis l’hypothèse que les tra- ductions réalisées sans ordinateur ne diffèrent pas significativement de celles réalisées avec l’ordinateur, et que les résultats de cette analyse pourraient nous aider à repenser l’intégra- tion de la TA dans le cours. Notre recherche vise à répondre aux questions suivantes : 1. Les traductions effectuées sans l’aide de l’ordinateur diffèrent-elles sensiblement des traductions effectuées à l’aide de l’ordinateur ? 2. Quelles sont les différences et les côtés positifs de la traduction sans ordinateur par rapport à la traduction avec les outils de TA ? 4.1 Profils des étudiants Les étudiants appartiennent à trois générations différentes : le premier groupe de 6 étu- diants (ci-après G1) ayant suivi le cours en 2018/19 ; le deuxième groupe de 9 étudiants (ci- après G2) ayant suivi le cours en 2021/22 ; et le troisième groupe de 5 étudiants (ci-après G3) ayant suivi le cours en 2023/24. En ce qui concerne l’attitude des étudiants à l’égard de la traduction automatique, les trois groupes présentent certaines différences : 68 Dobnik / Scripta Manent XIX/1 (2024), 63–80 • Le groupe G1 est la génération inscrite en 3ème année en 2018/19, juste avant le Co- vid-19 et avant l’arrivée des traducteurs neuronales (DeepL) ; bien que fortement atta- chée à l’utilisation des ressources de l’internet, ces étudiants avaient une distance voire une méfiance considérable par rapport à l’utilisation de Google Traduction ; • Le groupe G2, c’est la génération du Covid-19 qui a étudié pendant un an et demi à distance (la 1ère et 2ème année de leurs études) et qui est retournée à l’université en 2021/22. Cette génération se caractérise par une très forte dépendance à l’égard du TA neuronale (DeepL), qui se traduit par une attitude peu critique à l’égard de leurs traduc- tions et par une surestimation de la capacité de «post-édition». En 2021 et en 2022, lors des entretiens avec les étudiants à la fin des évaluations, environ un tiers d’entre eux ont déclaré qu’ils ne se sentaient pas capables de traduire sans recours à la TA ; • Le groupe G3 a commencé à étudier immédiatement après Covid-19 et, malgré son attachement au numérique, n’a pas développé une dépendance excessive à l’égard des outils de TA au cours des deux premières années, ce qui est certainement dû aux mé- thodes de travail en cours de traduction. 4.2 Corpus pour l’analyse et le déroulement Les textes pour l’analyse sont des traductions de deux documents authentiques que les trois groupes d’étudiants ont traduits dans le cadre d’un partiel. Les deux documents sont thé- matiquement liés aux spécificités linguistiques, textuelles et pragmatiques couvertes par le cours. «Nantes, une usine à culture» est un texte journalistique authentique publié en 2008 dans le magazine Label France (20 février 2008). Le texte traite de l’Histoire, du patrimoine cultu- rel et industriel, de l’architecture et présente des activités proposées dans l’ancienne usine des biscuits reconnue comme patrimoine industriel. Le texte ne contient pas de références extralinguistiques-textuelles spécifiques ou de termes techniques qui rendraient difficile la compréhension de l’ensemble. «Accès et stationnement des autocars de tourisme dans les villes et sites touristiques», document authentique conçu par l’Union internationale des transports routiers (IRU), est par sa forme un texte juridique qui a pour objectif de sensibiliser les professionnels du tou- risme, administrations locales et autorités publics aux atouts du transport par autocars de tourisme. Le traducteur est censé reconnaître immédiatement la structure et la prendre en considération, le document est caractérisé par les termes et des formules juridiques. Le contexte de la rédaction de traductions est le suivant : 1. les deux documents (textes de départ) ont été traduits par tous les étudiants dans le cadre d’un partiel, la note obtenue ayant compté dans la note finale (les deux représen- tant 30% de la note finale) ; 2. les deux documents ont été traduits en premier lieu en classe (ci-après Version-classe) ; ensuite, les étudiants ont été invités à refaire la traduction à la maison (ci-après Ver- sion-maison) et la soumettre le même jour dans Moodle, la note finale est basée sur les deux versions ; 69 Dobnik / Scripta Manent XIX/1 (2024), 63–80 3. seules les traductions ayant obtenu une note au-dessus de la moyenne ont été prises en compte dans l’analyse ; 4. le texte de départ a été remis aux étudiants uniquement sur papier, avec l’interdiction, en classe, de le recopier dans l’ordinateur ; 5. les groupes G1 et G2 ont écrit, en classe, la première version de la traduction sur papier, les étudiants ont pu utiliser l’ordinateur (internet, dictionnaires en ligne) mais étaient autorisés à entrer au maximum 3 mots dans les traducteurs automatiques ; au bout de 60 minutes, ils ont disposé de 15 minutes pour transcrire la traduction sur l’ordinateur et la soumettre dans Moodle (version-classe) ; la copie écrite à la main a été remise au professeur ; 6. le groupe G3 a écrit en classe une seule version, sur papier, sans aucune aide (ordina- teur, téléphone mobile, dictionnaire papier), donc la version-classe ; les étudiants ont également soumis dans Moodle une traduction faite à la maison avec recours à tous les outils informatiques (version-maison). La recherche s’est déroulée en deux étapes : I. Analyse empirique des traductions Version-classe : 1. dans chaque traduction (version-classe), nous avons sélectionné trois éléments (terme technique, passage ou éléments de structure) présentant une difficulté spécifique et étant susceptible de témoigner de la compétence de l’étudiant à trouver une solution de traduction adéquate ; 2. les trois éléments de chaque traduction ont été évalués sur le critère d’acceptabilité : 2 points pour une traduction exacte et réussie ; 1 point pour une traduction maladroite mais acceptable du point de vue sémantique (par ex. reformulation exacte) ; 0 points pour une traduction erronée. II. Analyse comparative des versions-classe avec les versions-maison : • nous avons comparé les versions-classe et les versions-maison afin d’évaluer, très som- mairement, l’engagement des étudiants à améliorer leurs traductions. Les éléments choisis pour l’analyse : 70 Dobnik / Scripta Manent XIX/1 (2024), 63–80 Tableau 1 Éléments analysés Document 1 : Nantes, une usine à culture Difficulté : Exemple 1 couple de fondateurs duel, connaissances extralinguistique Exemple 2 Après avoir fermé et été rachetée dans les années 70 par un grand groupe, l’usine reste vide (...) syntaxe et vocabulaire de spécialité Exemple 3 La culture est (...) un moyen de reconquête de lieux inattendus. compréhension et habileté à reformuler Document 2 : Accès et stationnement des autocars de tourisme dans les villes et sites touristiques Exemple 4 L’IRU, considérant que /…/ recommande que / 1. soit établie… structure (document juridique) Exemple 5 10 recommandations / IRU recommande vocabulaire (cohérence) Exemple 6 IRU recommande que - soient sauvegardées les retombées économiques locales, voir nationales favorisant l’emploi, engendrées directement ou indirectement par l’autocar de tourisme en atténuant la saisonnalité touristique compréhension et habileté à reformuler 4.3 Analyse empirique 4.3.1 Document 1 : Nantes, une usine à culture Exemple 1 : expression «couple de fondateurs» L’expression «couple de fondateurs» avec la mention de l’année 1886 se réfère à Louis Lefèvre-Utile, le fils des fondateurs de la pâtisserie artisanale qui est à l’origine de la marque LU, et à son beau-frère Ernest Lefèvre. Ils ont créé la société LU en 1887. «Couple de fonda- teurs» désigne alors deux personnes, mais pas les époux ayant fondé la pâtisserie d’origine. Les groupes G1 et G2 ont pu vérifier cela sur internet, le groupe G3 ne pouvait se fier qu’à une lecture attentive. Pour le groupe G1, le taux d’acceptabilité est de 66,60%, 3 étudiants ayant traduit correcte- ment, 2 ayant proposé une solution acceptable mais pas tout à fait correcte, et 1 étudiant avec la traduction incorrecte. Pour le groupe G2, le taux est de 50%, 3 étudiants ayant traduit l’expression correctement, 2 ayant proposé une traduction pas tout à fait correcte et 3 étudiants n’ayant pas traduit cette expression correctement. Pour le groupe G3, le taux est de 80%, 4 étudiants ayant traduit l’expression correctement et un étudiant n’ayant pas traduit correctement. À titre d’exemple : 71 Dobnik / Scripta Manent XIX/1 (2024), 63–80 • Traduction correcte : ustanovitelja (fr. deux fondateurs) • Traduction acceptable : ustanoviteljski par (fr. couple fondateur) • Traduction erronée : ustanovitelji, dva ustanoviteljev (fr. les fondateurs, deux des fondateurs) Exemple 2 : le passage «Après avoir fermé et été rachetée dans les années 70 par un grand groupe, l’usine reste vide /…/» Cette partie du texte comprend deux difficultés : d’une part, le sens de l’apposition verbale à la forme passive, d’autre part, la signification des expressions «grand groupe» et «racheter» qui relèvent du vocabulaire des affaires. Pour le groupe G1, le taux est de 41,60%, un seul étudiant ayant traduit le passage correcte- ment et 3 d’une manière acceptable. Pour le groupe G2, le taux est de 55,50%, 5 étudiants ayant traduit le passage correctement et 2 l’ayant traduit d’une manière acceptable et suffisamment compréhensible. La compa- raison a montré que pour la traduction-maison certains étudiants ont simplement pris la version incorrecte proposée par le traducteur automatique. Pour le groupe G3, le taux est de 60%, deux étudiants ayant traduit le passage correctement et un étudiant l’ayant traduit de manière acceptable (traduction correcte de l’apposition, mais maladroite pour le «grand groupe»). De manière générale, les traductions du groupe G3 révèlent une lecture plus attentive et une reformulation plus précise de ce passage. À titre d’exemple : • Traduction correcte : Po tem, ko so tovarno zaprli in jo je v 70. letih kupilo veliko pod- jetje, je ostala prazna /…/ (fr. Après avoir fermé et été rachetée dans les années 70 par une grande société, l’usine est restéee vide) • Traduction acceptable : Po tem, ko se je zaprla in jo je v sedemdesetih letih kupila velika skupina, je bila tovarna zapuščena /…/ (fr. Après son fermeture et après avoir été rachetée par un grand groupe, l’usine reste vide) • Traduction erronée : Potem, ko je bila tovarna v 70. letnih skupno zaprta, je ostala prazna /…/ (fr. Après l’ensemble de la fermeture de l’usine dans les années 1970, elle est resté vide). Exemple 3 : Le passage «la culture est à la fois un vecteur de cohésion sociale et de dévelop- pement économique, ainsi qu’un moyen de reconquête de lieux inattendus» fait partie d’une déclaration du directeur du LU et relève d’un registre administratif-politique, avec une langue clichée et élevée. La deuxième partie («reconquête de lieux inattendus») demande du tra- ducteur slovène une certaine habileté de reformulation (donc éloignement d’une potentielle traduction mot à mot) pour que le registre et le sens restent acceptables. Pour le groupe G1, le taux est de 41,60%, aucun étudiant n’ayant bien réussi la traduction, mais 4 étudiants ont livré une reformulation acceptable. Pour le groupe G2, le taux est de 44,40%, 2 étudiants ayant livré une traduction réussie et 4 une reformulation acceptable. 72 Dobnik / Scripta Manent XIX/1 (2024), 63–80 Pour le groupe G3, le taux est de 40%, aucun étudiant n’ayant bien réussi la traduction, mais 4 étudiants ont néanmoins fait une traduction acceptable. À titre d’exemple : • Traduction correcte : kultura je /…/ sredstvo za ponovno odkrivanje skritih kotičkov. (fr. un moyen de redécouvrir des coins cachés) • Traduction acceptable : kultura /…/ način ponovnega osvajanja nepričakovanih krajev. (fr. une façon de conquerir de nouveau les lieux inattendus) • Traduction erronée : /…/ sredstvo ponovne obuditve izgubljenih krajev (fr. un moyen de faire revivre les lieux perdus) Tableau 2 Résultats de l’analyse du Document 1 73 Dobnik / Scripta Manent XIX/1 (2024), 63–80 4.3.2 Document 2 : Accès et stationnement des autocars de tourisme dans les villes et sites touristiques Exemple 4 - Structure du document. Le document est structuré à la manière des documents juridiques, un règlement par exemple, afin de donner aux recommandations proposées plus de poids. Pour le groupe G1, le taux est de 83,30%, tous les étudiants ayant reconnu la structure, mais 2 étudiants n’ont pas entièrement respecté les paramètres de la structure. Pour le groupe G2, le taux est de 72,20%, 6 étudiants ayant très bien reproduit la structure dans leurs traductions et un étudiant pas entièrement ; 2 étudiants n’ont pas reconnu la structure et n’en ont pas tenu compte dans leurs traductions. Pour le groupe G3, le taux est de 80%, 4 étudiants ayant bien reproduit la structure dans leurs traductions mais un étudiant n’en a pas du tout tenu compte. À titre d’exemple : • Traduction correcte : IRU, ob upoštevanju, da / priporoča, da /…/ (fr. IRU, considérant que / recommande que) • Traduction acceptable : IRU meni, da / zato priporoča, da /…/ (fr. IRU considère que / et recommande que) • Traduction erronée : Na IRU menijo, da / upoštevati je treba / Priporočamo, da /…/ (fr. Chez IRU on pense que / il faut considérer que / Nous recommandons que) Exemple 5 - Mots «recommandation» et «recommander» Dans le contexte de ce document, le terme «recommandation» et sa forme verbale conjuguée «IRU recommande» ont une signification précise, il est donc important d’en tenir compte et utiliser le terme slovène exacte et de l’appliquer de manière cohérente dans l’ensemble du document. Pour le groupe G1, le taux est de 66,60%, 4 étudiants ayant correctement traduit et appliqué ce terme dans la traduction, contre 2 étudiants dont la traduction n’est ni correcte ni appli- quée de manière cohérente. Pour le groupe G2, le taux est de 66,60%, 6 étudiants ayant correctement traduit et appliqué le terme, contre 3 traductions erronées. Pour le groupe G3, le taux est de 60%, 2 étudiants ayant correctement traduit et appliqué le terme dans le texte traduit ; 2 étudiants ayant utilisé un synonyme acceptable ; 1 étudiant n’a pas livré une traduction acceptable. À titre d’exemple : • Traduction correcte : 10 priporočil / IRU priporoča (fr. 10 recommandations / IRU recommande) 74 Dobnik / Scripta Manent XIX/1 (2024), 63–80 • Traduction acceptable : 10 predlogov / IRU predlaga (fr. 10 propositions / IRU propose) • Traduction erronée : 10 nasvetov / predlagamo (10 conseils / nous proposons) Exemple 6 - Paragraphe 7 : /IRU recommande que/ - soient sauvegardées les retombées économiques locales, voir nationales favorisant l’emploi, engendrées directement ou indi- rectement par l’autocar de tourisme en atténuant la saisonnalité touristique. Ce paragraphe exige du traducteur une bonne compréhension et habilité à reformuler le contenu pour qu’il soit compréhensible. Pour le groupe G1, le taux est de 41,60%, 2 étudiants ayant traduit le paragraphe correcte- ment, 1 ayant rendu une version acceptable, contre 2 traductions erronées. Pour le groupe G2, le taux est de 50%, 4 étudiants ayant fait une traduction correcte, 1 étu- diant une reformulation très maladroite mais acceptable, contre 4 étudiants n’ayant pas réussi à faire une traduction pertinente ou sensée. Pour le groupe G3, le taux est de 50%, les 5 étudiants ayant réussi à faire une traduction suf- fisamment compréhensible et acceptable bien que maladroite. À titre d’exemple : • Traduction correcte : /IRU priporoča, / da se zaščitijo lokalne in nacionalne gospodarske koristi, povezane z zaposlovanjem, ki jih neposredno ali posredno ustvarjajo turistični avtobusi, ki prispevajo k zmanjševanju sezonske narave turizma • Traduction acceptable : /…/ se s spodbujanjem novih zaposlitev ohranja ekonomski do- biček na lokalni in državni ravni, ki je neposredna ali posredna posledica avtobusnega turizma in njegovega zmanjševanja turistične «sezonskosti» • Traduction erronée : /…/ se prepreči ponovni padec lokalnega gospodarstva med slabšanjem turistične sezone, z državno spodbudo za zaposlovanje, ki se ustvari posre- dno ali neposredno s turističnim avtobusom 75 Dobnik / Scripta Manent XIX/1 (2024), 63–80 Tableau 3 Résultats de l’analyse du Document 2 4.3.3 Résultats de l’analyse et observations Les résultats de l’analyse révèlent que les traductions du groupe G3 sont comparables aux traductions des groupes G1 et G2. En ce qui concerne le Document 1 (Nantes), les trois exemples choisis pour l’analyse néces- sitent principalement une lecture attentive et il s’est avéré que le groupe G3 a lu plus attenti- vement que les groupes G1 et G2 : dans l’Exemple 1, la plupart des étudiants G3 ont identifié le duel, contrairement aux groupes G1 et G2. Les étudiants G3 ont également été plus précis dans leurs traductions de l’Exemple 2 (apposition verbale). Quant à l’Exemple 3, les étudiants 76 Dobnik / Scripta Manent XIX/1 (2024), 63–80 G2 ont montré plus de créativité mais l’écart entre les traductions correctes et erronées (2 bonnes traductions et 3 erronées) est plus grand que pour les groupes G1 et G3. Le résultat combiné des trois exemples analysés montre que le groupe G3 a été le plus performant. En ce qui concerne le Document 2 (Autocars), les trois exemples choisis pour l’analyse sont liés aux notions étudiées en cours (documents administratifs et juridiques), la traduction nécessi- tant une lecture attentive (Exemples 4 et 5) et une bonne compréhension (Exemple 6). Dans l’exemple 4, il s’agit d’identifier la structure du document : les traductions des trois groupes sont bonnes, avec de meilleurs résultats pour les groupes G1 et G3. L’Exemple 5 concerne la langue administrative, les résultats sont bons et comparables pour les trois groupes. L’exemple 6 concerne la traduction d’un paragraphe compliqué qui nécessite une bonne compréhension et l’habilité à reformuler. Le groupe qui a rencontré le plus de difficultés est le G1, où seuls 2 étudiants ont rendu une traduction correcte. Les groupes G2 et G3 ont obtenu un résultat plus élevé, mais avec une différence dans la distribution : dans le groupe G2, 4 traductions correctes contre 4 traductions erronées, alors que dans le groupe G3, la plupart des étudiants (4 sur 5) ont réussi à faire une reformulation correcte bien que maladroite. Le résultat combiné des trois exemples analysés (Exemples 4, 5 et 6) montre une meilleure performance pour le groupe G2 (69,40%), alors que les groupes G1 et G3 ont obtenu le même niveau de performance (63%). Finalement, la comparaison des résultats des six exemples analysés montre que les per- formances du groupe G3 sont non seulement comparables à celles de G1 et G2, mais sont même meilleures. Toutefois, il convient de souligner que ces résultats sont donnés à titre indicatif et destinés principalement à une réflexion sur la démarche à suivre. 4.4 Analyse comparative des versions-classe et versions-maison En plus de l’analyse empirique, nous avons comparé les traductions faites en classe (ver- sion-classe) et celles rédigées avec tous les outils numériques disponibles (version-maison). Nous avons procédé à une comparaison très sommaire, sans évaluation approfondie, avec pour objectif d’observer des améliorations apportées à la première traduction (version-classe). Il faut préciser que les groupes G1 et G2 ont gardé dans l’ordinateur leur version-classe et pouvaient simplement apporter des corrections. Par contre, le groupe G3 n’a fait en classe qu’une seule version écrite à la main, qu’ils ont rendu au professeur, ils devaient donc refaire la traduction. À première vue déjà, nous avons observé que, dans certains cas, la version-classe et la ver- sion-maison diffèrent considérablement et que, dans d’autres, les deux versions diffèrent très peu. Les résultats de cette comparaison se résument de manière suivante : • Dans le groupe G1, 2 étudiants ont amélioré leur traduction et ont rendu une ver- sion-maison correcte voire bonne ; 2 étudiants ont amélioré leur traduction, mais ont aussi remplacé certaines solutions tout à fait correctes par des solutions erronées ; 1 étudiant a rendu la traduction identique à la version-classe. 77 Dobnik / Scripta Manent XIX/1 (2024), 63–80 • Dans le groupe G2, 2 étudiants ont amélioré leur traduction et ont rendu une bonne version-maison ; 4 étudiants ont amélioré leur traduction, mais ont aussi remplacé cer- taines reformulations exactes par des solutions erronées ; 1 étudiant n’a rien changé dans la version finale ; 2 étudiants ont envoyé la traduction faite par la TA (DeepL). • Dans le groupe G3, 2 étudiants ont amélioré leur traduction et ont rendu une ver- sion-maison correcte, 3 étudiants ont amélioré leur traduction, mais ont aussi rempla- cé certaines reformulations exactes par des solutions erronées. Tableau 4 Résultats de l’analyse comparative Catégorie 1 : version-maison corrigée et améliorée Catégorie 2 : version-maison en partie améliorée, avec des corrections erronées Catégorie 3 : version-maison identique à la version-classe Catégorie 4 : version-maison est la version TA (DeepL) G1 2 2 1 G2 2 4 1 2 G3 2 3 4.4.1 Observations Cette comparaison soulève un certain nombre de questions, y compris celles ne concernant pas directement la qualité de la traduction voire les compétences langagières et traduction- nelles de l’étudiant (voir le référentiel des compétences, Loock, 2019), mais qui relèvent d’un manque de motivation pour mener jusqu’au bout la traduction, d’une «faiblesse dans la mé- thode» (Gile, 2005 : 133). Il convient de noter qu’il nous est difficile voire impossible de déterminer dans quelle mesure la version-maison a été faite en totale autonomie et pas «en collaboration» (entre collègues ou avec un traducteur expérimenté). Il convient également de souligner qu’il est difficile de juger si les traductions correctes des parties difficiles du texte reflètent la véritable compréhension de l’étudiant ou si elles sont le résultat d’une coïncidence heureuse et de bonnes solutions proposées par la TA. C’est no- tamment le cas de la catégorie 2, où nous avons remarqué plusieurs cas de compréhension exacte dans la version-classe qui ont été remplacés, dans la version-maison, par une solution incorrecte (Kussmaul, 1995). En revanche, les cas relatés dans les catégories 3 et 4 révèlent une attitude qui mérite une at- tention particulière. L’indifférence à l’égard de la possibilité d’améliorer la traduction (catégo- rie 3) et même l’utilisation d’une traduction du TA témoignent du manque de responsabilité et de professionnalisme. Les raisons sont diverses mais doivent être prises en compte pour de futures réflexions sur les méthodes de travail en classe. 78 Dobnik / Scripta Manent XIX/1 (2024), 63–80 Nous avons néanmoins constaté que l’engagement des étudiants à l’égard de la qualité des traductions est proportionnel à leur niveau de compétences et leur résultats scolaires. 5 Discussion La recherche que nous avons effectuée visait principalement à identifier les enjeux de l’im- pact de la TA sur trois générations d’étudiants par rapport à l’utilisation des outils de TA. Les résultats de l’analyse empirique (première partie de notre recherche), dans laquelle nous avons évalué six exemples tirés des traductions, montrent que les performances du groupe qui a traduit sans ordinateur sont non seulement comparables à celles de deux groupes ayant traduit avec ordinateur, mais sont même meilleures. Les résultats de l’analyse comparative (deuxième partie de la recherche), dans laquelle nous avons observé les traductions faites en classe et celles faites le même jour à la maison, montrent un plus grand engagement des étu- diants ayant traduit en classe sans ordinateur à mener jusqu’au bout leur travail de traduction. Dans l’ensemble, ces résultats nous permettent de constater certains aspects très positifs d’une approche à la traduction sans utilisation d’un ordinateur : • les traductions révèlent de manière plus fiable la réelle compréhension des étudiants, ce qui nous semble particulièrement important pour toute évaluation des fautes et maladresses individuelles des étudiants ; • les fautes de sens sont principalement liées à l’absence de source et à une connais- sance insuffisante de la langue de départ (dans notre cas le français) ; • la lecture du texte de départ est plus attentive, les étudiants n’étant pas distraits par un grand nombre de possibilités ; • la reformulation est plus précise, même si la langue est souvent plus faible et moins fluide ; les traductions faites à la maison, par contre, témoignent de l’éloignement de cette compréhension initiale exacte vers des reformulations influencées par les diction- naires et les ressources trouvées en ligne. 6 Conclusion La recherche que nous avons effectuée s’inscrit dans le cadre d’une réflexion permanente sur les méthodes d’enseignement de la traduction et tout particulièrement en formation ini- tiale à la traduction. Comment préparer les étudiants en traduction aux futures réalités du marché et quelle est la place du traducteur vivant face à ta traduction automatique neuro- nale, ce sont des questions au cœur des réflexions de très nombreux formateurs et cher- cheurs. Et même si ces futures réalités sont en partie inconnues, elles seront sans aucune doute numériques, il est donc impératif de «faire prendre conscience aux étudiants - futurs professionnels - de ce que la machine peut faire et ne peut pas faire» (Loock, 2019), de «dé- mystifier» les outils de TA (Moorkens, 2018) et d’amener les étudiants à prendre conscience de leur plus-value par rapport à la machine. Enseigner et apprendre à traduire sans ordina- teur est sans doute un défi, mais nullement une utopie, c’est une des approches essentielles pour développer la compétence de traduction et renforcer la confiance en soi des étudiants. 79 Dobnik / Scripta Manent XIX/1 (2024), 63–80 Dans cette optique, il est important d’appliquer à tout niveau de formation des méthodes innovantes et stimulantes et de faire exister en bonne intelligence l’intelligence artificielle et l’intelligence naturelle des futurs traducteurs. Références Breyel-Steiner, C., & Grass, T. (2021). Traduction automatique et biotraduction : le mariage forcé. Tra- duire, 244, 94–106. https://doi.org/10.4000/traduire.2350 Dancette, J. (2003). L’0élaboration de la cohérence en traduction; le rôle des référents cognitifs. TTR, 16(1), 141–159. https://doi.org/10.7202/008560ar Durieux, C. (2005). L’enseignement de la traduction : enjeux et démarches. Meta, 50(1), 36–47. https:// doi.org/10.7202/010655ar ELIS. (2022). European Language Industry Survey 2022 : trends, expectations and concerns of the European language industry. https://fit-europe-rc.org/wp-content/uploads/2022/03/ELIS-2022_survey_re- sults_final_report.pdf?x85225 Faria Pires, L. (2018). Intégration de la traduction automatique neuronale à la formation universitaire des futurs traducteurs : pistes d’exploration. Myriades, 4, 53–65. Gile, D. (2005). La traduction. La comprendre, l’apprendre. PUF. Guerberof Arenas, A., & Moorkens, J. (2019). Machine translation and postediting training as part of a master’s programme. Journal of Specialised Translation, (31), 217–238. https://www.jostrans.org/ issue31/art_guerberof.php Hirci, N., Mikolič Južnič, T. & Pisanski Peterlin, A. (2019). Enriching Translator Training with Interpreting Tasks: Bringing Sight Translation into the Translation Classroom. Translation and interpreting: convergence, contact and interaction (26), 213-234. Hirci, N., & Pisanski Peterlin, A. (2020). Face-to-face and Wiki revision in translator training: exploring the advantages of two modes of collaboration, The Interpreter and Translator Trainer, 14(1), 38– 57. https://doi.org/10.1080/1750399X.2019.1688066 Kussmaul, P. (1995). Training the translator. John Benjamins Publishing Company. Lederer, M. (1994). La traduction aujourd’hui. Le modèle interprétatif. Hachette. Lederer, M., & Stratford, M. (2020). Culture et Traduction. Au-delà des mots. Classiques Garnier. Loock, R. (2019). La plus-value de la biotraduction face à la machine. Traduire, 241, 54–65. https://doi. org/10.4000/traduire.1848 Martikainen, H. (2019). Post-Editing Neural MT in Medical LSP: Lexico-Grammatical Patterns and Distortion in the Communication of Specialized Knowledge.  Informatics,  6(3), 26. https://doi. org/10.3390/informatics6030026 Massey, G., & Ehrensberger-Dow, M. (2017). Machine learning: Implications for translator education. Lebende Sprachen. 62(2), 300–312. https://doi.org/10.1515/les-2017-0021 Moorkens, J. (2018). What to expect from Neural Machine Translation: a practical in-class translation evaluation exercise. The Interpreter and Translator Trainer, 12(4), 375–387. https://doi.org/10.108 0/1750399x.2018.1501639 Vaupot, S. (2021). Analyse des erreurs de traduction automatique pour la combinaison de langues slovène-français et perspectives pour une formation en post-édition. Matices En Lenguas Extran- jeras, 14(2), 83–110. https://doi.org/10.15446/male.v14n2.91299 80 Dobnik / Scripta Manent XIX/1 (2024), 63–80 Povzetek Izzivi umetne inteligence: poučevanje in učenje prevajanja brez računalnika – utopija ali priložnost? Na področju didaktike strokovnega prevajanja so orodja za strojno prevajanje že vrsto let sestavni del poučevanja in vrednotenja prevajalske kompetence. Bolj ko so se orodja za strojno prevajanje razvi- jala in postajala sestavni del prevajalskih poklicev, pomembnejše mesto so dobivala tudi pri pouku prevajanja. Obenem pa je postajalo vse težje ovrednotiti pristnost prevodov in resnično prevajalsko kompetenco študentov. Ta vprašanja so postala izrazito pereča v času pandemije Covida-19 in s priho- dom generativnih orodij za prevajanje, ki so pri študentih prerasla v pretirano in nekritično uporabo prevajalnikov. V članku so predstavljeni rezultati raziskave, katere namen je bil primerjati in ovrednotiti prevode iz francoščine v slovenščino, ki so jih študentje naredili brez računalnika, v primerjavi s prevo- di, narejenimi z uporabo prevajalnikov in interneta. Rezultati so zgovorni in odpirajo številna izhodišča za razmislek o metodah poučevanja, ki bi jih bilo smiselno aplicirati pri nadaljnjem delu s študenti. Ključne besede: didaktika prevajanja, strojno prevajanje, umetna inteligenca, vrednotenje prevodov, generativni prevajalniki