Supporting Multilingual Students with GenAI

Generative AI Series: Topic Intro | Part 1 | Part 2

Generative AI Series
Part 2:
Supporting Multilingual Students with GenAI

Overview

As Generative Artificial Intelligence (GenAI) becomes increasingly accessible for students, many students and instructors are exploring how GenAI tools can be particularly helpful for multilingual students. 

 — 

Who are Multilingual Students? 

Multilingual students are students who are fluent in language(s) other than English. Multilingual students are a vast and diverse group. Nearly 44% of Californians speak a language other than English at home; as such, many UC Davis students are multilingual (U.S. Census Bureau). In this resource, we use “multilingual students” to refer to students who learned English as an additional language, recognizing that many native English speakers are also multilingual. 

In CEE’s recent GenAI Student Survey, 53% of multilingual student respondents used GenAI overall multiple times per quarter (compared to 40% of all respondents). However, few multilingual students are using GenAI solely for linguistic purposes. Like most of the survey respondents, multilingual students report using GenAI to support and enhance their learning, in addition to occasionally using GenAI for language support. 

Like the students themselves, multilingual students’ uses of GenAI are diverse and rooted in individual experiences. When considering how to support multilingual students in the context of GenAI, it is critical to keep three important things in mind: 

  • Multilingual students, including multilingual international students, already have high degrees of English proficiency (Cox, 2018; Young, 2013). 
    Instructors should remember that GenAI does not need to teach English to multilingual students, but multilingual students may seek input from GenAI for linguistic and intercultural support in different ways than native English-speaking students do. 
  • Instructors must actively challenge implicit biases and deficit mindsets towards multilingual students and their assumed use of GenAI.
    Students who speak with nonnative accents of English and submit well-written papers should not be assumed to have used GenAI dishonestly any more so than students who are native English speakers. 
  • There is no singular “standard” English. Rather, there are multiple “Englishes” and sociocultural contexts for English, such as English for Academic Purposes (EAP). 
    Rather than thinking in terms of “proper” or “standard” English, instructors should consider the social and cultural contexts of English and develop students’ critical awareness of the conventions of American Academic English, or how English is used in American academic settings, by whom, and for what purposes (Greenfield, 2011). While GenAI may support certain rules-based features of English, it will be less capable in supporting students with the culturally salient contexts of American Academic English.  

 

Understanding the “Language” in Large Language Models 

Multilingual students may wish to engage with GenAI tools to support their ongoing development of language fluency and intercultural competence. To support multilingual students in these efforts, instructors should keep in mind the following limitations: 

GenAI has a U.S.-centric, English language bias. 

GenAI training data overwhelmingly consist of English, U.S.-based text (Rettberg, 2022), reflecting global disparities in language resourcedness. As a result, GenAI outputs are likely to reflect the linguistic and cultural conventions of U.S.-based English speakers. While most GenAI tools are capable of accurately translating from other languages into English, they are not as capable when translating English into other languages (Lai et al., 2033; Yong et al., 2023). 

GenAI is useful for some aspects of (English) language use, but less useful for others. 

GenAI may be helpful with tasks that require formal linguistic competence (the knowledge of rules and statistical regularities of language), but less helpful with tasks that require functional linguistic competence (the ability to use language in the world). Mahowald et al. (2023) argue that Large Language Models (LLMs), the computational models that power GenAI tools, have achieved high degrees of accuracy with formal linguistic competence, but their capabilities with functional linguistic competence remain “patchy.” GenAI produces convincing texts by drawing on rules-based, statistically mappable components of language. However, producing statistically-likely strands of words is not equivalent to generating culturally-relevant texts that achieve the writer’s communicative intent (Bender and Koller, 2020).  

GenAI’s proficiency with formal linguistic competence, as well as its “patchiness” with functional linguistic competence, are what make these tools simultaneously useful and insufficient for multilingual students at the college level. 

GenAI tools can help support multilingual students with rules-bound elements of language (vocabulary, grammar, or syntax), but can only go so far in helping a student communicate their intended meaning in culturally- and contextually relevant ways.  

Remember that “academic language…is no one’s mother tongue” (Bourdieu & Passeron 1994). 

Navigating the dual terrains of linguistic and cultural fluency are part of the advanced learning processes of multilingual students and, by extension, all students who are learning how to communicate in American Academic English. 

  — 

Teaching Strategies 

GenAI-informed strategies that support multilingual students are also broadly beneficial for all students navigating the linguistic and cultural terrain of American Academic English. 

Using GenAI to Support Formal Linguistic Competence. 

GenAI performs well at language tasks that map onto rules-based and statistically-patterned features of English, such as grammar, syntax, tense, and words that often appear near to each other (collocates).  

  • Translating primary language into English: 
    GenAI can translate languages into English but does not perform equally well when translating all languages. Multilingual students should be cautious when using GenAI to translate and remember GenAI cannot perceive their communicative intent. Encourage students to use GenAI as a linguistic aid throughout the writing process, rather than using GenAI as a translator for a finished manuscript. 
  • Translating complex language into plain English and/or summarizing texts: 
    Students may benefit from using GenAI tools to reduce the linguistic complexity and generate summaries of course materials. As always, students should be mindful that these tools may oversimplify concepts and may not capture nuanced meaning. Students should confirm with their instructors/TAs that the summaries generated by GenAI are accurate. 
  • Correcting certain grammar and syntax errors: 
    Students can use GenAI tools to identify, correct, and explain how to address errors in grammar or syntax, and GenAI may be especially helpful for identifying words that are particularly perplexing in the English language, such as prepositions and indefinite/definite articles (a/an, the). However, instructors should encourage multilingual students to review any suggestions made by GenAI and try to discern whether the suggested text accurately reflects what they are attempting to convey. 
  • Word choice suggestions: 
    GenAI can make some suggestions for alternative word choice. However, it is important to keep in mind that GenAI cannot perceive the student’s true intent or meaning behind the words, nor can it assess the cultural relevance or appropriateness of particular word choices.  
  • Generating English-language content for study and practice: 
    With careful prompting, students can use GenAI to generate sample quizzes, reflection questions, or practice scenarios to help them encounter and master vocabulary and idioms that may be used in future learning environments. 

 

Using GenAI to Support Functional Linguistic Competence. 

GenAI performs less well at language tasks that convey meaning in socially- or culturally-relevant ways. However, GenAI can be used as a learning tool that supports students’ ongoing fluency in American Academic English. 

  • Producing and reading mentor texts: 
    Most students benefit from critical reading of mentor texts to analyze how the writing and communication “works.” GenAI can be used to produce mentor texts to read and critically analyze.  
  • Drafting emails: 
    Because GenAI tools are trained on US-centric data, they are generally adept at composing genres particular to U.S. cultural conventions. Students may find GenAI helpful for drafting and refining emails and can prompt the GenAI tool to make suggestions on the tone, directness, formality, and vocabulary in the email.  
  • Getting feedback on writing: 
    Students can prompt GenAI to give feedback and/or offer suggestions on their writing, and can use specific prompts to ask for particular, actionable feedback. As with all GenAI use, students should always be critical of GenAI’s output and remember that the tool cannot grasp the deeper meaning behind their writing nor the context in which their writing is produced. 
  • Conversational partner: 
    Students can benefit from using GenAI for its original purpose: chatting! GenAI tools can provide students with a readily accessible conversation partner to practice different scenarios in English (e.g., preparing for office hours or attending a conference).  
  • Developing pragmatic competence: 
    Students can use GenAI to explore language strategies for specific situations (e.g., adjusting requests for a specific context or crafting an effective apology). GenAI may not be able to explain the linguistic features of these situational communications but may help students consider strategies and other questions related to these contexts. 

 

Equity Check 

The goal of these strategies should never be to make multilingual students “sound like native English speakers.” Such a goal advances an ethnocentric worldview and is not inclusive. Almost all multilingual people have some degree of influence from their first language for the remainder of their lives, despite being fully fluent in English (Valdés 1992). (Even monolingual English speakers write and speak with accents.) 

Instead, focus on helping students develop critical awareness of academic discourse by helping them perceive (and later replicate) the linguistic and rhetorical moves authors make and how scholars express voice, authority, and identity in their communications. 

GenAI tools should be used to enhance, activate, fine-tune, and further support multilingual students' advanced English proficiency. When used appropriately and with critical awareness of culture and context, GenAI tools can be one of many resources all students can use to enhance their learning, in English and beyond.   

Considerations for Instructors 

When considering how multilingual students use Generative AI, instructors should not make assumptions that multilingual students are somehow different from monolingual students; many students, regardless of linguistic backgrounds, may find GenAI tools helpful for supplementing and enhancing their learning. Multilingual students may also find these tools beneficial for supporting their ongoing process of developing language fluency. 

It is crucial for instructors to interrogate and disrupt implicit bias toward multilingual students and their perceived use of GenAI. Students who are perceived to be multilingual must not be subject to inadvertent, additional scrutiny about the potential use of GenAI. Instructors should consider using blind scoring of student work to disrupt the potential of inadvertently subjecting multilingual students to this kind of scrutiny. If an instructor has a multilingual student who makes some grammatical errors when speaking, but submits writing relatively free of grammatical errors, they should not assume that student improperly used GenAI. 

When considering how to support multilingual students with GenAI, instructors should keep in mind that multilingual students already have advanced language proficiency, thus GenAI can supplement and enhance their language learning. Thus, strategies for supporting multilingual students can also be beneficial for all students, regardless of linguistic background. Instructors should also consider the ways that having multilingual students is an asset for their course and the learning environment and model this asset-framing to their students as much as possible. 

  • Acknowledgement
  • This resource was developed by Erica Bender (PhD, Assessment Specialist, UC Davis Center for Educational Effectiveness), Emily Montgomery (Graduate Program Coordinator of International and Academic English, Global Affairs, UC Davis), Dawn Takaoglu (Director of International and Academic English, Global Affairs, UC Davis), and Barbara Mills (Testing Specialist, UC Davis Center for Educational Effectiveness).
  • Citation
  • Center for Educational Effectiveness (CEE). (2024). Generative AI Series: Just-in-Time Teaching Resources. https://cee.ucdavis.edu/JITT
  • Additional Resources
  •  
    • Center for Educational Effectiveness Generative AI Student Survey
      UC Davis Center for Educational Effectiveness resource.
    • International Academic English Program
      UC Davis Global Affairs resource for multilingual and international student support.
    • AI Has a Language Diversity Problem. Humans Do, Too
      D’Agostino, S. Inside Higher Ed.
    • ChatGPT Is Cutting Non-English Languages Out of the AI Revolution
      Dave, P. Wired.
    • Language Versus Thought in Human Brains and Machines?
      Fedorenko, E. Video presentation from AMLD.
  • References
  • Bender, E. M., & Koller, A. (2020). Climbing towards NLU: On meaning, form, and understanding in the age of data. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. https://doi.org/10.18653/v1/2020.acl-main.463

    Bourdieu, P., & Passeron, J. C. (1994). Linguistic misunderstanding and professorial power. Stanford University Press.

    Caucheteux, C., & King, J.-R. (2022). Brains and algorithms partially converge in natural language processing. Communications Biology, 5, 134. https://doi.org/10.1038/s42003-022-03036-1

    Cox, M. (2018). “Noticing” language in the writing center: Preparing writing center tutors to support graduate multilingual writers. In S. Lawrance & T. Myers Zawacki (Eds.), Re/Writing the center: Approaches to supporting graduate students in the writing center. Utah State University Press.

    Ferris, D. (2009). Teaching college writing to diverse student populations. University of Michigan Press.

    Greenfield, L. (2011). The “standard English” fairy tale: A rhetorical analysis of racist pedagogies and commonplace assumptions about language diversity. In L. Greenfield & K. Rowan (Eds.), Writing centers and the new racism: A call for sustainable dialogue and change. Utah State University Press.

    Lai, V. D., Ngo, N. T., Pouran Ben Veyseh, A., Man, H., Dernoncourt, F., Bui, T., & Nguyen, T. H. (2023). ChatGPT beyond English: Towards a comprehensive evaluation of large language models in multilingual learning. arXiv Preprint Database. https://doi.org/10.48550/arXiv.2304.05613

    Mahowald, K., Ivanova, A. A., Blank, I. A., Kanwisher, N., Tenenbaum, J. B., & Fedorenko, E. (2023). Dissociating language and thought in large language models. arXiv Preprint Database. https://doi.org/10.48550/arXiv.2301.06627

    Rettberg, J. W. (2023). ChatGPT is multilingual but monocultural, and it’s learning your values. Jill/txt. https://jilltxt.net/right-now-chatgpt-is-multilingual-but-monocultural-but-its-learning-your-values/

    U.S. Census Bureau. (n.d.). Language other than English spoken at home, percent of persons age 5 years+, 2018–2022, California. U.S. Department of Commerce. https://www.census.gov/quickfacts/fact/table/CA/POP815222#POP815222

    Valdés, G. (1991). Bilingual minorities and language issues in writing: Toward profession-wide responses to a new challenge. Written Communication, 9(1). https://doi.org/10.1177/0741088392009001003

    Yong, Z. X., Zhang, R., Forde, J. Z., Wang, S., Subramonian, A., Lovenia, H., Cahyawijaya, S., Winata, G. I., Sutawika, L., Blaise Cruz, J. C., Tan, Y. L., Phan, L., Garcia, R., Solorio, T., & Aji, A. F. (2023). Prompting multilingual large language models to generate code-mixed texts: The case of South East Asian languages. arXiv Preprint Database. https://doi.org/10.48550/arXiv.2303.13592

    Young, V. A. (2013). “Keep code meshing”: Literacy as translingual practice. In S. Canagarajah (Ed.), Between communities and classrooms. Routledge.