arXiv subjects

Dorottya Demszky

Publications and source records attributed to Dorottya Demszky.

At least 19 recordsLinked to original sources

arXiv2026

IDEAlign: Comparing Ideas of Large Language Models to Domain Expert

Large language models (LLMs) are increasingly used to produce open-ended, interpretive annotations, yet there is no validated, scalable measure of idea-level similarity to expert annotations. We (i) introduce the content evaluation of LLM annotations as a core, understudied task, (ii) propose IDEAlign for capturing expert similarity judgments via pick-the-odd-one-out tasks, and (iii) benchmark various similarity methods (text embeddings, topic models, and LLM-as-a-judge) against these human ratings. Applying this approach to two real-world educational datasets (e.g., interpreting math reasoning and feedback generation), we find that most metrics fail to capture the nuanced dimensions of similarity meaningful to experts. LLM-as-a-judge performs best (11~18% improvement over other methods) but still falls short of expert alignment, making it useful as a triage tool rather than a substitute for human review. Our work demonstrates the difficulty of evaluating open-ended LLM annotations at scale, and positions IDEAlign as a reusable protocol for benchmarking on this task to help guide responsible deployment of LLMs.

cs.CL↗

arXiv2026

Mitigating LLM biases toward spurious social contexts using direct preference optimization

LLMs are increasingly used for high-stakes decision-making, yet their sensitivity to spurious context can introduce harmful biases. This is a critical concern when models are deployed for tasks like evaluating teachers' instructional quality, where biased assessment can affect teachers' professional development and career. We investigate model robustness to spurious social contexts about teachers using the largest publicly available dataset of U.S. classroom transcripts (NCTE) paired with expert evaluation scores. Evaluating seven frontier and open-weight models across seven categories of spurious contexts -- including teacher experience, education level, demographic identity, and sycophancy-inducing framings -- we find that irrelevant contexts can shift model-generated ratings by up to 1.48 points on a 7-point scale. Prompt-based mitigations and popular post-training methods, such as Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO), fall short of addressing the issue. We propose Debiasing-DPO, which combines contrastive reasoning-augmented DPO with SFT on expert labels to reduce the effect of spurious context while avoiding mode collapse. Applied to Llama and Qwen Instruct models of 3-8B parameters, Debiasing-DPO reduces bias by 84% and improves predictive accuracy by 52% on average across models. Our findings from the educational dataset highlight that stronger models may exhibit greater sensitivity despite higher accuracy, and Debiasing-DPO can improve both accuracy and robustness in prompt-based prediction tasks.

cs.AI↗

arXiv2026

Pigeonholing: how bad prompts hurt models, causing collapse and mistakes

While in-context learning is generally shown to be effective in Large Language Models (LLMs), bad contexts can cause performance degradation and mode collapse, a phenomenon we call "pigeonholing." **Unintentionally bad** contexts can happen without malicious jailbreaking intents: For example, a user asks the model to justify an incorrect math theorem or fails to correct the model's buggy code. Specifically, we investigate ``pigeonholing" in two scenarios: (1) when the user suggests a solution, and (2) when the conversation context includes the assistant's previous (incorrect) responses. Our experiments across 10 verifiable and open-ended tasks with 10 different models show that pigeonholing manifests in several ways: (1) repeating the incorrect answers from context (leading to 38-40% performance drop), (2) converging on a narrow set of answers in coding and text generation without exploring alternatives, and (3) flipping stance on controversial topics to align with the user or the assistant's previous claims. We find that pigeonholing worsens almost monotonically with the number of conversation turns (performance drops by additional 14+% as repeated mistakes increase from 1 to 5), and pigeonholing-induced mode collapse can happen even when the provided example is correct. As a step toward mitigation, we propose RLVR with synthetic errors which improves models by 43-60% under bad contexts compared to vanilla RLVR baselines.

cs.CL↗

arXiv2026

When Youth Enter The Chat: An Epistemic Shift in the Validation of LLM-Based Measures of Student Talk

LLMs are being used increasingly to measure aspects of student discourse (e.g. talk moves, collaboration, equity of voice) at scale. Typically, LLM-based measures of student talk use transcriptions of classroom conversations that only include verbal contributions, which de-contextualize student language. Common practices for validating these measures include comparing outputs against expert annotations by adults, using held out evaluation sets and F1 scores. We argue that these approaches are insufficient to ensure that such measures are meaningful and equitable for teaching and learning, particularly for racially and linguistically marginalized youth. In order to center the youth whose talk is being analyzed, re-contextualizing these classroom conversations and engaging youth in the research process is necessary. Sharing epistemic authority with youth, ultimately, centers their point of view and adds crucial nuance to the analysis of their talk that adult experts, researchers, and LLMs cannot provide. In a case study of multilingual youth in one 8th-grade math classroom, we address the epistemic exclusion of youth by employing multiple ethnographically-oriented methods to re-contextualize student conversations and center youth as epistemic authorities in conversation with researchers and LLMs. We conducted participant observations, interviews, focus groups, and member checks with four focal students. Findings reveal that there were misalignments between students' interpretations of their own math talk experiences and the LLM-based measures of their talk. Students contested both the LLM classifications and the coding scheme used to measure their talk, highlighting the need for youth to be involved in the epistemic process of producing knowledge about their experiences.

cs.CL↗

arXiv2026

EduCoder: An Open-Source Annotation System for Education Transcript Data

We introduce EduCoder, a domain-specialized tool designed to support utterance-level annotation of educational dialogue. While general-purpose text annotation tools for NLP and qualitative research abound, few address the complexities of coding education dialogue transcripts -- with diverse teacher-student and peer interactions. Common challenges include defining codebooks for complex pedagogical features, supporting both open-ended and categorical coding, and contextualizing utterances with external features, such as the lesson's purpose and the pedagogical value of the instruction. EduCoder is designed to address these challenges by providing a platform for researchers and domain experts to collaboratively define complex codebooks based on observed data. It incorporates both categorical and open-ended annotation types along with contextual materials. Additionally, it offers a side-by-side comparison of multiple annotators' responses, allowing comparison and calibration of annotations with others to improve data reliability. The system is open-source, with a demo video available.

cs.CL↗

arXiv2026

Mapping the Methodological Space of Classroom Interaction Research: Scale, Duration, and Modality in an Age of AI

Research on classroom interaction has long been divided between large-scale observation and in-depth ethnographic work. We propose a framework mapping this methodological space along three dimensions--scale, duration, and modality--where a study's position shapes what it reveals and obscures. We illustrate it through contrasting studies of dialogic teaching--Howe et al. (2019) and Snell and Lefstein (2018)--and an interview with the lead researchers, organized around three questions: what can be operationalized, what mechanisms become visible, and what translates to practice. We then examine how AI is expanding this space and how the framework can guide research and tool design.

cs.AI↗

arXiv2026

From Scoring to Explanations: Evaluating SHAP and LLM Rationales for Rubric-based Teaching Quality Assessment

Automated scoring models are increasingly used to assign rubric-based quality ratings to complex language performances, including classroom transcripts, yet they typically provide little insight into why a particular score is produced. We propose a general framework for sentence-level interpretability of rubric-based scoring that combines model-agnostic Shapley-value attributions with rationales generated by large language models (LLMs). Instantiated on the Quality of Feedback dimension of the CLASS framework using the NCTE corpus, the framework enables systematic comparison of fine-tuned pretrained language models (PLMs) and prompted LLMs on both scoring performance and explanation faithfulness. Across 6k annotated transcript segments, fine-tuned PLMs outperform LLMs in prediction accuracy but exhibit label compression toward mid-scale scores. Deletion-based tests show that SHAP identifies sentences that reliably drive model predictions, producing typically larger and more coherent prediction shifts than LLM-generated rationales. Cross-model analyses further reveal that SHAP attributions transfer robustly across architectures, whereas LLM rationales exert limited and inconsistent influence. Overall, the findings demonstrate that SHAP provides more faithful and transferable explanations for rubric-based scoring, and that the proposed framework offers a principled basis for evaluating both scoring models and their explanations in high-stakes educational settings and other rubric-based language assessment tasks.

cs.CL↗

arXiv2026

Practitioner Voices Summit: How Teachers Evaluate AI Tools through Deliberative Sensemaking

Teachers face growing pressure to integrate AI tools into their classrooms, yet are rarely positioned as agentic decision-makers in this process. Understanding the criteria teachers use to evaluate AI tools, and the conditions that support such reasoning, is essential for responsible AI integration. We address this gap through a two-day national summit in which 61 U.S. K-12 mathematics educators developed personal rubrics for evaluating AI classroom tools. The summit was designed to support deliberative sensemaking, a process we conceptualize by integrating Technological Pedagogical Content Knowledge (TPACK) with deliberative agency. Teachers generated over 200 criteria - initial articulations spanning four higher-order themes (Practical, Equitable, Flexible, and Rigorous) - that addressed both AI outputs and the process of using AI. Criteria contained productive tensions (e.g., personalization versus fairness, adaptability versus efficiency), and the vast majority framed AI as an assistant rather than a coaching tool for professional learning. Analysis of surveys, interviews, and summit discussions revealed five mechanisms supporting deliberative sensemaking: time and space for deliberation, artifact-centered sensemaking, collaborative reflection through diverse viewpoints, knowledge-building, and psychological safety. Across these mechanisms, TPACK and agency operated in a mutually reinforcing cycle - knowledge-building enabled more grounded evaluative judgment, while the act of constructing criteria deepened teachers' understanding of tools. We discuss implications for edtech developers seeking practitioner input, school leaders making adoption decisions, educators and professional learning designers, and researchers working to elicit teachers' evaluative reasoning about rapidly evolving technologies.

cs.HC↗

arXiv2026

Marked Pedagogies: Examining Linguistic Biases in Personalized Automated Writing Feedback

Effective personalized feedback is critical to students' literacy development. Though LLM-powered tools now promise to automate such feedback at scale, LLMs are not language-neutral: they privilege standard academic English and reproduce social stereotypes, raising concerns about how "personalization" shapes the feedback students receive. We examine how four widely used LLMs (GPT-4o, GPT-3.5-turbo, Llama-3.3 70B, Llama-3.1 8B) adapt written feedback in response to student attributes. Using 600 eighth-grade persuasive essays from the PERSUADE dataset, we generated feedback under prompt conditions embedding gender, race/ethnicity, learning needs, achievement, and motivation. We analyze lexical shifts across model outputs by adapting the Marked Words framework. Our results reveal systematic, stereotype-aligned shifts in feedback conditioned on presumed student attributes--even when essay content was identical. Feedback for students marked by race, language, or disability often exhibited positive feedback bias and feedback withholding bias--overuse of praise, less substantive critique, and assumptions of limited ability. Across attributes, models tailored not only what content was emphasized but also how writing was judged and how students were addressed. We term these instructional orientations Marked Pedagogies and highlight the need for transparency and accountability in automated feedback tools.

cs.CL↗

arXiv2026

Protecting and Promoting Human Agency in Education in the Age of Artificial Intelligence

Human agency is crucial in education and increasingly challenged by the use of generative AI. This meeting report synthesizes interdisciplinary insights and conceptualizes four aspects that delineate human agency: human oversight, AI-human complementarity, AI competencies, and relational emergence. We explore practical dilemmas for protecting and promoting agency, focusing on normative constraints, transparency, and cognitive offloading, and highlight key tensions and implications to inform ethical and effective AI integration in education.

cs.HC↗

arXiv2025

DiagramIR: An Automatic Pipeline for Educational Math Diagram Evaluation

Large Language Models (LLMs) are increasingly being adopted as tools for learning; however, most tools remain text-only, limiting their usefulness for domains where visualizations are essential, such as mathematics. Recent work shows that LLMs are capable of generating code that compiles to educational figures, but a major bottleneck remains: scalable evaluation of these diagrams. We address this by proposing DiagramIR: an automatic and scalable evaluation pipeline for geometric figures. Our method relies on intermediate representations (IRs) of LaTeX TikZ code. We compare our pipeline to other evaluation baselines such as LLM-as-a-Judge, showing that our approach has higher agreement with human raters. This evaluation approach also enables smaller models like GPT-4.1-Mini to perform comparably to larger models such as GPT-5 at a 10x lower inference cost, which is important for deploying accessible and scalable education technologies.

cs.AI↗

arXiv2025

TeachLM: Post-Training LLMs for Education Using Authentic Learning Data

The promise of generative AI to revolutionize education is constrained by the pedagogical limits of large language models (LLMs). A major issue is the lack of access to high-quality training data that reflect the learning of actual students. Prompt engineering has emerged as a stopgap, but the ability of prompts to encode complex pedagogical strategies in rule-based natural language is inherently limited. To address this gap we introduce TeachLM - an LLM optimized for teaching through parameter-efficient fine-tuning of state-of-the-art models. TeachLM is trained on a dataset comprised of 100,000 hours of one-on-one, longitudinal student-tutor interactions maintained by Polygence, which underwent a rigorous anonymization process to protect privacy. We use parameter-efficient fine-tuning to develop an authentic student model that enables the generation of high-fidelity synthetic student-tutor dialogues. Building on this capability, we propose a novel multi-turn evaluation protocol that leverages synthetic dialogue generation to provide fast, scalable, and reproducible assessments of the dialogical capabilities of LLMs. Our evaluations demonstrate that fine-tuning on authentic learning data significantly improves conversational and pedagogical performance - doubling student talk time, improving questioning style, increasing dialogue turns by 50%, and greater personalization of instruction.

cs.CL↗

arXiv2025

Tell, Don't Show: Leveraging Language Models' Abstractive Retellings to Model Literary Themes

Conventional bag-of-words approaches for topic modeling, like latent Dirichlet allocation (LDA), struggle with literary text. Literature challenges lexical methods because narrative language focuses on immersive sensory details instead of abstractive description or exposition: writers are advised to "show, don't tell." We propose Retell, a simple, accessible topic modeling approach for literature. Here, we prompt resource-efficient, generative language models (LMs) to tell what passages show, thereby translating narratives' surface forms into higher-level concepts and themes. By running LDA on LMs' retellings of passages, we can obtain more precise and informative topics than by running LDA alone or by directly asking LMs to list topics. To investigate the potential of our method for cultural analytics, we compare our method's outputs to expert-guided annotations in a case study on racial/cultural identity in high school English language arts books.

cs.CL↗

arXiv2025

Multi-Stage Speaker Diarization for Noisy Classrooms

Speaker diarization, the process of identifying "who spoke when" in audio recordings, is essential for understanding classroom dynamics. However, classroom settings present distinct challenges, including poor recording quality, high levels of background noise, overlapping speech, and the difficulty of accurately capturing children's voices. This study investigates the effectiveness of multi-stage diarization models using Nvidia's NeMo diarization pipeline. We assess the impact of denoising on diarization accuracy and compare various voice activity detection (VAD) models, including self-supervised transformer-based frame-wise VAD models. We also explore a hybrid VAD approach that integrates Automatic Speech Recognition (ASR) word-level timestamps with frame-level VAD predictions. We conduct experiments using two datasets from English speaking classrooms to separate teacher vs. student speech and to separate all speakers. Our results show that denoising significantly improves the Diarization Error Rate (DER) by reducing the rate of missed speech. Additionally, training on both denoised and noisy datasets leads to substantial performance gains in noisy conditions. The hybrid VAD model leads to further improvements in speech detection, achieving a DER as low as 17% in teacher-student experiments and 45% in all-speaker experiments. However, we also identified trade-offs between voice activity detection and speaker confusion. Overall, our study highlights the effectiveness of multi-stage diarization models and integrating ASR-based information for enhancing speaker diarization in noisy classroom environments.

cs.SD↗

arXiv2025

From Weak Labels to Strong Results: Utilizing 5,000 Hours of Noisy Classroom Transcripts with Minimal Accurate Data

Recent progress in speech recognition has relied on models trained on vast amounts of labeled data. However, classroom Automatic Speech Recognition (ASR) faces the real-world challenge of abundant weak transcripts paired with only a small amount of accurate, gold-standard data. In such low-resource settings, high transcription costs make re-transcription impractical. To address this, we ask: what is the best approach when abundant inexpensive weak transcripts coexist with limited gold-standard data, as is the case for classroom speech data? We propose Weakly Supervised Pretraining (WSP), a two-step process where models are first pretrained on weak transcripts in a supervised manner, and then fine-tuned on accurate data. Our results, based on both synthetic and real weak transcripts, show that WSP outperforms alternative methods, establishing it as an effective training methodology for low-resource ASR in real-world scenarios.

eess.AS↗

arXiv2025

CPT-Boosted Wav2vec2.0: Towards Noise Robust Speech Recognition for Classroom Environments

Creating Automatic Speech Recognition (ASR) systems that are robust and resilient to classroom conditions is paramount to the development of AI tools to aid teachers and students. In this work, we study the efficacy of continued pretraining (CPT) in adapting Wav2vec2.0 to the classroom domain. We show that CPT is a powerful tool in that regard and reduces the Word Error Rate (WER) of Wav2vec2.0-based models by upwards of 10%. More specifically, CPT improves the model's robustness to different noises, microphones and classroom conditions.

cs.CL↗

arXiv2024

Problem-Oriented Segmentation and Retrieval: Case Study on Tutoring Conversations

Many open-ended conversations (e.g., tutoring lessons or business meetings) revolve around pre-defined reference materials, like worksheets or meeting bullets. To provide a framework for studying such conversation structure, we introduce Problem-Oriented Segmentation & Retrieval (POSR), the task of jointly breaking down conversations into segments and linking each segment to the relevant reference item. As a case study, we apply POSR to education where effectively structuring lessons around problems is critical yet difficult. We present LessonLink, the first dataset of real-world tutoring lessons, featuring 3,500 segments, spanning 24,300 minutes of instruction and linked to 116 SAT math problems. We define and evaluate several joint and independent approaches for POSR, including segmentation (e.g., TextTiling), retrieval (e.g., ColBERT), and large language models (LLMs) methods. Our results highlight that modeling POSR as one joint task is essential: POSR methods outperform independent segmentation and retrieval pipelines by up to +76% on joint metrics and surpass traditional segmentation methods by up to +78% on segmentation metrics. We demonstrate POSR's practical impact on downstream education applications, deriving new insights on the language and time use in real-world lesson structures.

cs.CL↗

arXiv2024

Kid-Whisper: Towards Bridging the Performance Gap in Automatic Speech Recognition for Children VS. Adults

Recent advancements in Automatic Speech Recognition (ASR) systems, exemplified by Whisper, have demonstrated the potential of these systems to approach human-level performance given sufficient data. However, this progress doesn't readily extend to ASR for children due to the limited availability of suitable child-specific databases and the distinct characteristics of children's speech. A recent study investigated leveraging the My Science Tutor (MyST) children's speech corpus to enhance Whisper's performance in recognizing children's speech. They were able to demonstrate some improvement on a limited testset. This paper builds on these findings by enhancing the utility of the MyST dataset through more efficient data preprocessing. We reduce the Word Error Rate (WER) on the MyST testset 13.93% to 9.11% with Whisper-Small and from 13.23% to 8.61% with Whisper-Medium and show that this improvement can be generalized to unseen datasets. We also highlight important challenges towards improving children's ASR performance. The results showcase the viable and efficient integration of Whisper for effective children's speech recognition.

eess.AS↗