Students’ Use of an AI Translation Tool in Multilingual Classrooms

Abstract

Artificial Intelligence (AI) translation tools hold strong potential for supporting multilingual classrooms. This mixed-methods study investigates lower secondary school students’ use of and experiences with a real-time AI translation tool embedded in an interactive presentation platform developed by Learnlab in Norway. The tool combines a fine-tuned version of ChatGPT aligned with the Norwegian curriculum and Learnlab AI Translator. We analyze survey data and system logs (N = 148) and four focus-group interviews (N = 19) from five lower secondary schools. The results indicate that the AI translation tool is beneficial for students overall, but particularly for multilingual students. Students used the tool not only to understand difficult words and sentences, but also to explore languages and engage with their peers’ texts across languages. They reported feeling more included in classroom activities suggesting that the tool enabled collaborative meaning-making across languages and perspectives. However, some monolingual students mainly perceived the benefits for others and did not recognize the tool’s potential for their own language learning. There were no statistically significant differences in perceived inclusion or usability across genders or grade levels. The results are discussed in relation to language support, multilingual learning, and inclusion and participation, with the aim of contributing to more equitable learning environments.

1. Introduction

Language diversity is an increasing characteristic of classrooms, raising questions about how all students can access learning content and participate meaningfully in shared activities. While multilingualism has been shown to support learning and identity development (Forbes et al., 2021; Jessner, 2008), language barriers can also limit students’ opportunities to engage with subject matter and with each other. This creates a need for tools and pedagogical approaches that support both access to content and participation in collective learning processes. AI (artificial intelligence) is rapidly shaping education by enabling personalized, real-time feedback and providing language and learning support for students. As in many other educational contexts, schools in Norway, the context of the present study, lack clear guidelines for AI use. Consequently, some schools use freely available open AI tools, some prohibit AI use, and others rely on “closed” versions aligned with the General Data Protection Regulation (GDPR). Little research has been conducted on the (mis)uses of AI in Norwegian classrooms. In one study, Elstad and Eriksen (2024) examined how 236 teachers at five high schools in Oslo experienced the use of AI. They found that individual teacher factors have greater impact than organizational factors when it comes to using AI or not. They also found positive relationships between professional teacher communities and teacher efficacy. In another more relevant study for our purpose, since the focus was high school students, 413 and 984 students in 2023 and 2024 were surveyed about their use of AI and their beliefs about AI and writing (Bueie et al., 2025). The students showed multiple reservations against the use of AI and they were reticent about using AI for writing, but students in 2024 were more positive than students in 2023 and boys were more positive than girls. In a third study, AI-generated feedback and its potential to improve formative assessments were investigated (Burner et al., 2025). In that study, we interviewed 13 teachers and 26 students and conducted unstructured observations at five secondary schools. The findings indicated that high-performing students are motivated by AI-generated feedback to further improve their writing. However, some students found the AI-generated feedback to be too general or too complicated to understand. The lack of language support led to unequal benefits among students. Some students struggled to access or make use of AI-generated feedback due to language barriers, highlighting the importance of integrating language support into AI-supported learning environments. A plethora of research has focused on AI translation tools for students, most commonly Google Translate before 2022 when AI technologies offering several new translation tools still had not conquered writing arenas in classrooms. Studies concluded with improved writing quality when students used Google Translate (Cancino & Panes, 2021). Research often focuses on translation as an individual support function and primarily within higher education, underscoring the need for more research at lower levels of education (Albadarin et al., 2024; González-Calatayud et al., 2021). There is limited knowledge about how AI translation tools can support multilingual learning and how such tools enable students to engage with each other’s inputs across languages in collective classroom settings. Thus, as new AI technologies are developing, there is a need for research on practical and pedagogical usage of them in classrooms, ensuring that they serve educational purposes (Uanachain & Aouad, 2025) and contribute to embedding AI tools in classroom activities (Pratschke, 2024) that support “transformative competencies” as part of 21st-century skills (OECD, 2023).

The introduction and experimentation of AI tools in schools, we believe, need to focus on both “qualification”—that students gain access to subject matter and are able to work with content—and “socialization” and “subjectification”—that students take part in the wider sociocultural setting and become more independent in their thinking and acting (Biesta, 2011, pp. 19–21). In this study, these dimensions are collectively reflected in what we refer to as inclusion (OECD, 2024), operationalized through students’ reported experiences of access to content, participation in shared activities, and ability to engage with peers on reasonably equal terms.

During the previous study published in the present journal (Burner et al., 2025), we noticed that a lack of language support makes some students benefit far more from AI use than others. Thus, the present study examines students’ use of and experiences with a real-time AI translation tool embedded in an interactive presentation, and how such use may support qualification, socialization, and subjectification in multilingual classroom interaction.

The study addresses the following research question:

How do students use and experience an AI translation tool in an interactive classroom setting, and how may such use support inclusion and participation across language backgrounds?

2. Materials and Methods

2.1. The AI Translation Tool

The tool used in this study is a web-based interactive presentation and survey platform with integrated AI capabilities, including facilitator support, automated reporting, and real-time translation. It builds on Learnlab AI Translator and OpenAI’s ChatGPT-4o, fine-tuned to align with the Norwegian national curriculum and core educational values such as equity, democracy, and participation (OECD, 2023). While participating, students can choose to translate content for comprehension, write in their own language, or translate peers’ contributions, allowing students to co-construct meaning across languages and perspectives. Both translations and feedback remain accessible, context-sensitive, and aligned with national standards for inclusive and progressive education. The system is fully GDPR-compliant and consistent with the principles outlined in the EU AI Act, promoting safe, transparent, and responsible use of AI in schools. For a detailed description of the tool, its technical architecture, and pedagogical foundations, see Lindvig et al. (2025).

2.2. Sample and Data Collection

Data collection took place during eight lessons in eight different classes at five different lower secondary schools with a total number of 148 students (see

Table 1. Overview of the sample.

School A	School B	School C	School D	School E
- 16 students in a Norwegian Language Arts class - three boys and two girls in focus-group interview, all of them multilingual	- 14 students in a Norwegian Language Arts class - two boys and two girls in focus-group interview, two of them multilingual	- 19 students in a Social Studies class - three boys and three girls in focus-group interview, four of them multilingual	- Nine students in a Social Studies class - two boys and two girls in focus-group interview, all of them multilingual	- 90 students in various classes - no qualitative data were collected due to practical reasons

below; for a detailed overview, see Appendix A).

Three data sources were used to capture usage patterns and students’ experiences with the AI translation tool: survey (see Appendix B), system logs, and focus-group interviews. The students were familiar with the AI-assisted interactive presentation tool, but not the real-time translation feature. Each session began by asking students for consent to participate in the study. Those who did not consent to participate in research were later deleted from the system logs. Next, the students filled out a simple survey about their background (gender, whether they speak another language than Norwegian at home—and if yes, which language, and grade level in the subject “Norwegian”). The lessons where the AI translation tool was used were either about “Samis in Norway” (Sami are indigenous peoples of Norway) (four lessons) or gender perspectives when choosing professions (four lessons). The lessons were conducted similarly with certain sequences and uses of the AI translation tool. For example, the students had to read some history and reflect on the Samis’ situation today by responding to a set of questions. Moreover, they had to respond to statements about the Samis, for example, “Sami culture is well-preserved in Norway today”. They were challenged to like and comment on the best arguments from their peers. There were three such tasks. Finally, they were asked to summarize in their own words what they had learnt during the session. All the students could see each other’s answers (anonymously) during the whole session. At the very end, they answered survey questions related to perceived inclusion and usefulness of the tool, using a 5-point Likert scale. They were also asked what they used the tool for: “understand difficult words”, “understand longer text”, “understand peers’ text”, “write a text in another language than Norwegian”, “general translation” or “I did not use the tool”.

The surveys were supplemented with system logs that showed which languages the students chose during which tasks and the languages they used in the interactive steps. The participants were grouped in several ways to enable more granular analyses of usage patterns and perceptions across meaningful constellations. Specifically, students were categorized as monolingual or multilingual (based on whether they reported speaking another language at home), by gender, and by self-reported achievement level.

Right after the lessons in schools A–D, students participated in focus-group interviews. The teacher was told to select a variety of students, including factors such as gender, home language, and achievement level. Nineteen informants in total were selected. We asked them four questions: how was your experience of the tool, how did you use it (for what and when), to what extent did you use the tool to [understand text, translate, understand subject matter, work with your own text/other’s text…], and to what extent would you say that the tool makes you feel more included [give examples]. The interviews lasted for 20–30 min and were recorded and transcribed. No qualitative data were collected from School E due to practical constraints.

2.3. Data Analysis

System data were analyzed using a thematic behavioral approach combining reported home languages, languages accessed within the translation tool, the frequency of language changes, and the number of interactive steps in which the translation tool was used with fewer than four non-Norwegian languages. This threshold was set to allow for meaningful multilingual use, as some students reported using more than one home language in addition to Norwegian, while filtering out interactions with large number of translation languages for that student. We then triangulated these behavioral indicators with students’ writing data to capture how translation was integrated into their learning activity.

A combination of descriptive and inferential statistics was applied to examine relationships between tool use, inclusion, and perceived usefulness. Descriptive statistics summarized overall patterns of translation use, writing languages, and survey responses. Paired-samples t-tests were used to compare students’ mean ratings of inclusion and usability both across the entire sample and within subgroups (monolingual/multilingual, gender, grade, and multilingual usage types). Independent samples t-tests compared mean inclusion and usability scores between monolingual and multilingual groups, and between genders. One-way ANOVAs were used to compare inclusion and usability ratings across grade levels and among the five multilingual subgroups. Chi-square tests of independence were conducted to examine associations between categorical variables, including: Language background (mono/multi) and type of translation use, grade level and multilingual subgroup membership, and inclusion versus usability responses across groups. For proportional data (e.g., frequency of tool use between groups), two-proportion z-tests were performed to determine whether observed differences (such as general use and use for understanding difficult words) were statistically significant. Together, these analyses provided both descriptive insight into usage patterns and statistical evidence regarding whether perceptions of inclusion and usability varied across key demographic and behavioral groups (see table in Appendix C for an overview). Finally, interview transcripts were coded and categorized using the constant comparative method (Strauss & Corbin, 1998) in the software program NVivo version 15. Each meaningful unit was coded by the authors, and the codes were merged into categories, constantly comparing them across the four sets of interviews. The interviews were translated from Norwegian into English.

Results from surveys, system logs and interviews were compared systematically to identify convergences and paradoxes, strengthening the study’s validity. All four authors participated in the interpretation of the results, strengthening the study’s reliability.

3. Results and Discussion

3.1. Group Composition

The gender composition was comparable across all schools, meaning it was relatively balanced. The composite inclusion score did not differ significantly between girls and boys. However, one item showed a robust difference: girls reported significantly higher agreement with the statement that they participated “on equal terms” with other students. This difference was statistically significant even after correction for multiple testing, though it did not extend to other items or the overall composite score. Across all classrooms, 50% of the students reported that they speak one or more languages other than Norwegian at home. Altogether, 27 distinct home languages were identified. The most frequent was English, followed by Bosnian and Somali, and Albanian, Hindi, Tigrinya, and Serbian.

The majority of the students reported they are on the higher achievement level. Analyses across grade bands (1–2, 3–4, 5–6) revealed statistically significant differences on some participation-related items, as well as on the composite inclusion score. These differences were primarily driven by the lowest-achieving group (1–2), which reported lower perceived participation on equal terms and lower feelings of safety in writing and sharing texts. In contrast, students in the 3–4 and 5–6 bands reported similar and generally high levels of inclusion. Importantly, overall satisfaction with the AI translation tool remained high across all groups, and differences were more pronounced in how students experienced participation than in whether they found the tool useful.

In the system logs we manually identified five distinct subgroups of multilingual students based on their observed translation and writing behavior:

• Bilingual users—Students who consistently alternated between two or more relevant languages and frequently wrote in at least two languages.
• Translation-only users—Students who primarily translated between Norwegian and another relevant language but wrote exclusively in Norwegian.
• L1-dominant users—Students who translated to and wrote primarily in their main home language.
• Situational users—Students who used the translation tool strategically for specific actions, such as translating or composing parts of the task or presentation.
• Exploratory users—Students who experimented across multiple or unfamiliar languages, sometimes producing short text segments in languages they had not reported speaking (e.g., Japanese).

The subgrouping provided a nuanced understanding of how multilingual learners engaged with the translation tool, balancing linguistic intention, task strategy, and exploratory use.

3.2. System Logs and Survey

From the system logs, we can see that all participants tried the AI translation tool. Most students used it more than ten times during the single classroom session (45 min). Moreover, the system logs showed a pronounced difference in translation use between language groups. Across all recorded steps (≤3 languages per step), translation was used in half of them. Multilingual students engaged with the AI translation tool far more often than their monolingual peers. This may not be surprising, as multilingual students use one or more language(s) at home in a society where the majority language is Norwegian. However, system logs show that multilingual students not only used the AI tool more often but also sustained usage across more steps. The proportion of steps with targeted translation was higher in the multilingual group. A high proportion of the multilingual students wrote in languages that they did not report as their home language, meaning the languages are their third or fourth language or that they use the AI tool much more frequently to explore different languages. These patterns suggest that multilingual students demonstrate greater curiosity and willingness to explore language learning than monolingual students (Burner & Carlsen, 2022; Forbes et al., 2021; Jessner, 2008).

The students evaluated the AI translation tool highly across all aspects of perceived performance. They reported high levels of agreement with all attitude statements related to inclusion, participation, and usefulness. Mean scores on individual items (1 = strongly disagree, 5 = strongly agree) were consistently above 4, and the composite variable Sum inclusion showed a mean of 4.3, indicating a strongly positive overall evaluation. The highest-rated items concerned instructional facilitation and adaptation, suggesting that students perceived the learning design and the tool as enabling meaningful participation. These findings point to the AI translation tool functioning as an inclusive support embedded in the instructional context, rather than as a compensatory add-on for a limited group of students. Students who reported lower Norwegian proficiency tended to rate the AI translation tool as more accurate and useful. This suggests that the translation tool may have helped these students access the teaching content more effectively and participate on more equal terms with their peers, i.e., adaptation as an educational purpose (Uanachain & Aouad, 2025). Finally, regarding perceived inclusion and participation (OECD, 2024), a large majority of students agreed that the AI translation tool enhanced their ability to participate in classroom activities. The numbers are more pronounced among multilingual students (Jessner, 2008). A particularly strong quantitative finding concerns actual use: all of the multilingual students stated that they switched to their home language when using Learnlab. This demonstrates that the tool’s multilingual functionality is actively taken up by those who are likely to benefit most from it.

Overall, survey and system log data indicate that the AI translation tool is beneficial across the student sample, although multilingual students benefit more than monolingual students in terms of semantic use and perceived inclusion. The findings suggest that the tool has the potential to support language learning, while multilingual students engage more actively with multiple languages, indicating increased curiosity and language awareness (Jessner, 2008). At the same time, their reported sense of inclusion suggests that the tool supports not only access to content, but also fuller participation in shared classroom activity (Forbes et al., 2021; Biesta, 2011).

3.3. Focus-Group Interviews

The analyses of the interview transcript resulted in these categories: “Actual use of AI tool”, “Perception of the AI tool”, and “Inclusion and participation”.

3.3.1. Actual Use of AI Tool

The languages the students reported that they translated were Bosnian, Hungarian, Tamil, Turkish, French, Somali, English, Mandarin, and Serbian. Some students used the AI tool to test different languages, among others their own home language, as expressed by the following students: “To Faroese, so it was like… a bit cool to check out how it was translated” (all student quotes are our translations from Norwegian); “I have lived in Scotland, so I just had to translate to English and check if it was correct”. The tool was used for semantic reasons, but at the same time to study grammatical structures: “I varied a bit between different languages, so mostly to see… in a way find out what different words mean, and in which sequence they are set up in sentences”. In this sense, the tool aided language comprehension and appeared to stimulate active engagement with language learning. A small number of students reported using it to reinforce vocabulary or grammar in foreign language subjects such as German and French. As one participant noted: “If I’m studying German, I can translate everything I read to German and use it for practice”. Moreover, multilingual students used the tool not only as practical aid but also as a resource for vocabulary development and contrastive analysis in language learning (Jessner, 2008). One participant reported: “I had the text in French and compared it to the Norwegian version. It helped me learn new words and how they are structured in sentences”. Another remarked: “It was better than Google Translate, the grammar and gender endings were more correct”. Some multilingual students master only their home language orally but found the AI translation tool helpful because of the audio possibility. A student with limited proficiency in her home language, Tigrinya, used the AI tool to listen to the translation from Norwegian to Tigrinya. Finally, the students reported that they used the AI tool to understand other students’ texts when they saw languages they did not understand. Their curiosity and language awareness seemed to increase since their peers were the ones creating the texts they wanted to understand (cf. Biesta’s notion of socialization), in contrast to texts provided by the teacher.

In sum, these results indicate that students used the AI translation tool both functionally and exploratively: to improve comprehension of taught content, to engage more actively with peers’ contributions, and to explore language independently. In Biesta’s (2011) terms, this points to processes of qualification, socialization, and subjectification occurring simultaneously within the same classroom activity. The combination of instrumental and creative use suggests that AI-based translation can simultaneously support linguistic, exploratory language awareness, and greater student agency in how meaning is accessed and expressed (Jessner, 2008; Biesta, 2011). The tool enables students to engage in collaborative learning activities where translation becomes a natural and integrated part of communication and understanding, in line with 21st-century skills (OECD, 2023).

3.3.2. Perception of the AI Tool

Students emphasized the AI translation tool’s ease of use while doing the various tasks. Most students described the translation process as instantaneous, with only minor inaccuracies in short phrases or idiomatic expressions. Several students noted that “[it] translated much faster and more correct than Google Translate” and that it is “convenient to use”. One student said, “You have more perspectives on learning languages, and you can manage by yourself before needing to ask the teacher”. However, not all monolingual students or multilingual students who had grown up in Norway saw the potential in the translation tool: “I don’t really need it, because I know Norwegian very well”. Interestingly, the monolingual student underscored the AI tool’s usefulness for their multilingual peers: “It is a good tool for example for immigrants who do not know the language [Norwegian] yet”. An important aspect mentioned by some students was learning from peers’ texts: “… I learnt a lot more, and I can also learn from the others”.

In sum, students’ qualitative evaluations corroborate the responses from the survey. They are mostly positive, but some perceive the tool as something to be used only if you do not know Norwegian well enough, failing to observe the multilingual benefits for all students regardless of home language. This is in line with previous research on teachers’ attitudes towards multilingualism (Burner & Carlsen, 2022).

3.3.3. Inclusion and Participation

One student referred to the ongoing war in Ukraine: “Now that there is a war… if they [Ukrainian refugees in Norway] haven’t learnt Norwegian yet, then it will be much easier for them to participate in teaching and express their opinions when using this tool”. Several other students emphasized the usefulness of the AI translation tool for students who speak other languages than Norwegian at home: “If you struggle with Norwegian, then you can choose the language you want. Then you understand and then you feel more included because you follow what is going on”; “Scary and a bit foreign to come to another country, to be uncertain about how to speak the language and receive individual teaching, so it’s very good to be included in the same classroom with texts in your home language”; and “If I had been an immigrant, I’d appreciate this app a lot because I wouldn’t understand Norwegian and I would use it to understand what people say”. The students agreed that the AI tool contributes to a more inclusive and participatory classroom because they will be “more in sync with the rest of the class” and “feel more a part of the class”, not only because of translation but also that “you can read others’ ideas and languages”. Thus, from the students’ perspective, the two most prominent factors supporting inclusion and participation are language support, which enables them to follow both the content and the pace of classroom activity, and opportunities to engage with peers’ texts, ideas, and languages. The tool also appears to support self-expression by allowing students to use languages they can express themselves in with greater confidence. These findings align with Biesta’s (2011) concepts of qualification, socialization, and subjectification.

All in all, the qualitative data elaborate on the quantitative data when it comes to ways of using the AI translation tool. Students show both instrumental and creative uses, particularly the multilingual ones (Forbes et al., 2021; Jessner, 2008). Despite the tool being as relevant for monolingual students as for multilingual students regarding language awareness and “playing around” with language structures, its usefulness is attributed mostly to multilingual students, cementing the myth that multilingual students’ home language is just a means to learn Norwegian rather than being an end in itself (Burner & Carlsen, 2022). An interesting finding is the tool’s potential to enhance collaborative learning by providing the possibility for all students to see each other’s ideas, suggestions and evaluations. The tool is therefore not limited to adapting texts or tasks for individual comprehension. It also supports a broader educational function: students gain better access to content (qualification), greater scope to act and express themselves through language (subjectification), and more opportunities to participate in shared classroom meaning-making (socialization) (Biesta, 2011; Uanachain & Aouad, 2025).

3.4. Limitations

One limitation of the study is that School E constituted a large proportion of the quantitative sample, while no interview data were collected from that school due to practical constraints. The qualitative findings should therefore be interpreted as reflecting the four other school contexts, while the quantitative results are more strongly shaped by the larger School E sample. Moreover, it was the teachers who selected the students for focus-group interviews. While this is a common procedure in classroom research, it may yield some bias, for example if the teacher purposefully selects students who are usually positive to AI-enhanced tools. On the other hand, one could argue that the teacher knows the students best, and thus is more competent to select them for interviews regarding their home language(s) and their achievement level at school. Finally, the AI translation tool was used in the classes for a short period of time. The study does not investigate the long-term experiences and uses of the tool.

4. Conclusions

In this study, we investigated how students use and experience an interactive presentation tool with fine-tuned real-time AI translation capabilities, and the extent to which it contributes to perceived inclusion and participation among all students by creating equitable learning conditions through language support. We used surveys, system logs, and focus-group interviews based on data from eight classrooms across five lower secondary schools, with qualitative interview data drawn from four of the schools. The quantitative and qualitative data provide converging evidence across usage patterns and student experiences. We draw the following four conclusions, taking into consideration the short-term and context-specific nature of data:

(1)

Gender and achievement level did not show significant differences in students’ use of the AI translation tool, whereas multilingual students used it more than monolingual students, across both usage patterns and reported experiences. Students used the tool for multiple purposes, including understanding content, interpreting peers’ contributions, and exploring languages.

(2)

Most students reported that the AI translation tool was useful, accurate, and efficient. There were high levels of perceived inclusion and participation, although some monolingual students primarily perceived the tool as beneficial for others rather than for themselves. The findings suggest that the tool can stimulate curiosity and language awareness, and may support students’ agency in how they access, interpret, and express meaning across languages.

(3)

In this classroom context, the findings indicate that the AI translation tool supported inclusion in collective learning settings by enabling access to words, sentences, and texts (both written and via text-to-speech). Students were able to engage in collaborative learning activities where translation became part of communication and understanding. The tool supported co-construction of meaning by enabling students to read and respond to each other’s contributions across languages and appeared to support both qualification through access to content and socialization through participation in shared classroom activity.

(4)

The tool provided opportunities for sharing ideas and languages among students, as they could read and respond to each other’s contributions regardless of language. This appears to support a more equitable and collaborative learning environment, where students reported feeling better able to follow classroom activity and see their contributions as valued and understood.

Finally, this study underscores that the impact of AI technology depends on the purpose of the tool’s design, how usable and accessible it is in fulfilling that purpose, and how teachers and students engage with the tool to achieve pedagogical goals. In this sense, the study contributes by showing how the design and integration of real-time translation within shared classroom activities may support inclusion and engagement across language backgrounds. Future research should further investigate these interactions across different age groups, subjects, and cultural contexts, examining how usability, ethical design, and pedagogical integration can foster inclusion, participation, and learning in multilingual classrooms.