Mapping Language Model Failures Through Community Experience: A Study of Multilingual Researchers
PI: Dana Calacci (IST)
Plan for funding tuition for graduate students, or the remainder of the researcher’s salary for postdoc and research faculty: Dr. Calacci’s IST start-up funds
Dr. Dana Calacci, assistant professor in the College of Information Sciences and Technology (IST) and an ICDS Co-hire, studies the socio-technical and legal impacts of datafication and AI on communities, especially worker groups. Through collaborations like the Workers Algorithm Observatory, which she co-directs, she designs and deploys technologies with communities that aim to answer their most pressing questions about the impact of AI, new platforms, and surveillance on their lives. Dana received her PhD from MIT’s Media Lab in 2023, and a B.S. in computer science from Northeastern University in 2015. She also has experience as a startup co-founder and a mixed-media artist. Her writing and work has appeared or been featured in NPR’s Radiolab, Gizmodo, Wired, Reuters, The Atlantic’s CityLab, the New York Times, and other major publications.the New York Times, and other major publications.
Dr. Calacci will serve as mentor for the Junior Researcher.
Other Senior and Junior Team Members
Matt Viana – Graduate Student & Candidate for Junior Researcher Matt Viana is an incoming PhD student in Dana’s research group. Matt recently received his B.S. in Computer Science from Penn State’s College of IST and leads qualitative data collection and analysis for this project.
Team’s History of Interdisciplinary Engagement
As demonstrated by the publication history of the PI and other team members, the research team has extensive experience leading projects that cross discipline boundaries and engage expertise from diverse fields such as information science, law and law enforcement, critical algorithm studies, social sciences, human computer interaction (HCI), AI, and data stewardship and rights.
ICDS Engagement
As an ICDS co-hire, Dr. Calacci regularly engages with ICDS, including by providing content for the ICDS Staff Newsletter (March 2025) and collaborating with and engaging services from the Research Innovations with Scientists and Engineers (RISE) team and Project Management Office (PMO). Recently, Dr. Calacci lectured as part of the AI for Social Impact” series presented by the Center for Socially Responsible AI (CSRAI). Dr. Calacci’s talk, titled “How AI is Reshaping Your Paycheck: Personalized Wages in the Inference Economy,” discussed how companies are using AI to change how they set pay.
Project Description and Objectives
Project Description
This project investigates how English as a Second Language (ESL) graduate students interact with Large Language Models (LLMs) like ChatGPT, focusing on how language proficiency shapes their experience of model failures, biases, and harms. Through a mixed-methods design—combining surveys and semi-structured interviews with EFL and native English-speaking graduate students—we aim to document usage patterns, identify unique barriers faced by EFLRs, and surface the sociotechnical dynamics that influence how users trust, contest, or adapt to AI-generated outputs. The ultimate goal of this stage of the project is to inform the design of community-driven harm reporting mechanisms for future LLM development and contribute directly to an NSF proposal on equitable AI systems.
The study is structured in two phases. First, we will administer a screener survey to ~100 Penn State graduate students to collect demographic and background data (e.g., native language, language proficiency, LLM familiarity). From this pool, we will recruit ~40 participants (balanced between ESL students and native speakers) to complete a detailed survey about their LLM usage, including types of tasks (e.g., paraphrasing, summarizing, citation generation), perceived model reliability, and observed failures or limitations. In Phase 2, we will conduct semi-structured interviews with ~20 participants to explore specific failure cases in depth and understand participants’ strategies for evaluating or correcting problematic LLM output.
A Junior Researcher will support developing data collection instruments, performing data collection (including interviews and surveys), and analysis, working closely with the PI on each step.
Specific Objectives for Work
● Supporting recruitment and scheduling for interviews with graduate students
● Leading interviews with study participants
● Cleaning and analyzing survey data to identify patterns in LLM usage, failure experiences, and language background
● Assisting in transcription, qualitative coding (e.g., via MaxQDA), and thematic analysis of interviews
● Helping develop an initial taxonomy of EFLR-specific harms and LLM failure types
● Co-authoring or contributing to a research paper for submission to a venue such as AIES, EAAMO, or CHI
Medium to Long-term Goal(s)
● NSF CISE Proposal: Use results and taxonomies from this project, as well as other preliminary results, as core justification for a Spring 2026 NSF proposal focused on building multilingual, community-centered feedback mechanisms for foundation models
● Scholarly Publication: Submit a paper documenting the findings, failure modes, and user experiences to a top venue in AI ethics or HCI (likely TOCHI, maybe FAccT)
● Tooling and Infrastructure: Develop and publicly share interview codebooks, taxonomies of failure, and early prototypes for harm-reporting tools
Realistic Alignment of Effort/Objectives and Realistic Expectations of Junior Researchers
We have already designed much of the study, including surveys, interview guides, and scheduling procedures. The work is appropriate for a 25% RA over 1-2 semesters. We aim to have data collection partially complete by the start of the Fall 2025 semester, allowing the Junior Researcher to focus on analysis, remaining interviews, coding data, and writing. This will provide the Junior Researcher with enough scaffolding and structure to create a reasonable output through the project timeline.
Alignment with Junior Researcher Call & ICDS Mission
ICDS Mission
This project aligns directly with ICDS’s mission of advancing data-driven, computationally intensive research for public impact. We plan to engage with ICDS within this project by sharing findings through ICDS events and online (talk, blog post, symposium, etc). We also hope to contribute a curated dataset or other artifacts that can aid reproducibility and other research, which could be hosted on ICDS infrastructure. We will encourage the Junior Researcher to participate in ICDS seminars, events, and workshops, including the annual symposium.
Relevant ICDS Hub/Area
This work falls within the ICDS hubs/areas of Artificial Intelligence and Data Science.
● AI: This project focuses on evaluation and design of LLMs, using both technical and user-centered lenses.
● Data Science: The project combines mixed-methods analysis of structured (survey) and unstructured (interview) data to generate actionable insights and infrastructure.
ICDS Center Affiliation
This project is aligned with the Center for Socially Responsible AI (CSRAI)’s mission to promote equitable, ethical, and more transparent AI systems. This project centers the experience of a group often overlooked in AI design and evaluation—ESL students and researchers. It will contribute empirical evidence about how language and identity shape user experience with AI systems, and how those experiences can be used to guide more ethical design and evaluation. Our work advances CSRAI’s focus on fairness, inclusion, and, in the future, participatory AI.
Interdisciplinary Basis of Project
This is a fundamentally interdisciplinary project at the intersection of computer science, human-computer interaction, and social science. The PI brings expertise in community-centered AI auditing and data ethics, while the student(s) will bring experience in machine learning (ML) and qualitative methods, as well as a multi-lingual background. The project methodology integrates computational survey analysis with qualitative social research, and the outcomes will be relevant to scholars across AI, education, HCI, and algorithmic fairness.
Funding
Level of Effort & Tuition Funding Plan
The Junior Researcher will dedicate 25% effort to this project during the academic year and 100% effort to the project during Summer 2026. All tuition will be covered from the PI’s department startup funds.
Intended Pay Grade
The intended graduate student pay grade for the Junior Researcher is Grade 19.
Expectations of Junior Researcher
Expertise or Skills Sought
We are seeking a Junior Researcher with one or more of the following skills:
● Qualitative analysis experience—coding interview transcripts with tools like maxQDA or NVivo
● Familiar with survey design in tools like Qualtrics and statistical analysis of results using R or Python
● Knowledge of or interest in LLMs, especially related to fairness, multilingual usage, or user-centered evaluation
● Strong communication and demonstrated research skills, including the ability to synthesize findings into presentations or op-eds, etc. for academic and non-academic audiences
● Experience working with diverse or multilingual communities is a plus
We are open to mentoring candidates from either a technical or social science background.
General Expectations
For the duration of the Junior Researcher project, the Junior Researcher will allocate their time between research to support their dissertation, the ICDS Junior Researcher work outlined in this proposal, and regular engagement with ICDS (e.g. ICDS Symposium, talk series, lunches, and other events).
As outlined in the Junior Researcher Call, the Junior Researcher will submit a mid-semester project report and a ~1 page written report describing progress towards the proposed objectives and goals for each project/opportunity within two weeks of the end of their ICDS-supported appointment.
To the best of their ability and as schedules allow, the Junior Researcher should plan to participate in a one-hour meeting of Dr. Calacci’s lab group once per week.
The student may be pre- or post-comps. In-person availability is preferred due to the study population, and the student should possess the ambition to lead authorship on a research paper.