Towards Reward Modeling for AI Tutors in Math Mistake Remediation

Evaluating the pedagogical quality of AI tutors remains challenging: standard NLG metrics do not determine whether responses identify mistakes, scaffold reasoning, or avoid revealing the answers. For the task of mistake remediation, we derive a hierarchy of pedagogical aspects from human pairwise preferences on MRBench, and synthesize minimally contrastive response pairs that differ along key aspects (e.g., mistake identification and location, targetedness, scaffolding, actionability, clarity, and coherence). We develop and release Bradley-Terry preference models trained on weighted-sum rankings that we automatically create from MRBench, synthetic pairs, and data combinations. Using only synthetic data, our best model reaches 0.69 pairwise accuracy on a human preference test, and combining weighted-sum data with targeted synthetic groups improves accuracy to 0.74, outperforming larger general-purpose reward models while using only a 0.5B-parameter backbone.

MzansiText and MzansiLM: An Open Corpus and Decoder-Only Language Model for South African Languages

Decoder-only language models can be adapted to diverse tasks through instruction finetuning, but the extent to which this generalizes at small scale for low-resource languages remains unclear. We focus on the languages of South Africa, where we are not aware of a publicly available decoder-only model that explicitly targets all eleven official written languages, nine of which are low-resource. We introduce MzansiText, a curated multilingual pretraining corpus with a reproducible filtering pipeline, and MzansiLM, a 125M-parameter language model trained from scratch. We evaluate MzansiLM on natural language understanding and generation using three adaptation regimes: monolingual task-specific finetuning, multilingual task-specific finetuning, and general multi-task instruction finetuning. Monolingual task-specific finetuning achieves strong performance on data-to-text generation, reaching 20.65 BLEU on isiXhosa and competing with encoder-decoder baselines over ten times larger. Multilingual task-specific finetuning benefits closely related languages on topic classification, achieving 78.5% macro-F1 on isiXhosa news classification. While MzansiLM adapts effectively to supervised NLU and NLG tasks, few-shot reasoning remains challenging at this model size, with performance near chance even for much larger decoder-only models. We release MzansiText and MzansiLM to provide a reproducible decoder-only baseline and clear guidance on adaptation strategies for South African languages at small scale.

LLM as a Meta-Judge: Synthetic Data for NLP Evaluation Metric Validation

Validating evaluation metrics for NLG typically relies on expensive and time-consuming human annotations, which predominantly exist only for English datasets. We propose \textit{LLM as a Meta-Judge}, a scalable framework that utilizes LLMs to generate synthetic evaluation datasets via controlled semantic degradation of real data, replacing human judgment. We validate our approach using \textit{meta-correlation}, measuring the alignment between metric rankings derived from synthetic data and those from standard human benchmarks. Experiments across Machine Translation, Question Answering, and Summarization demonstrate that synthetic validation serves as a reliable proxy for human judgment, achieving meta-correlations exceeding 0.9 in multilingual QA and proves to be a viable alternative where human judgments are unavailable or too expensive to obtain. Our code and data will become publicly available upon paper acceptance.

PET-F2I: A Comprehensive Benchmark and Parameter-Efficient Fine-Tuning of LLMs for PET/CT Report Impression Generation

PET/CT imaging is pivotal in oncology and nuclear medicine, yet summarizing complex findings into precise diagnostic impressions is labor-intensive. While LLMs have shown promise in medical text generation, their capability in the highly specialized domain of PET/CT remains underexplored. We introduce PET-F2I-41K (PET Findings-to-Impression Benchmark), a large-scale benchmark for PET/CT impression generation using LLMs, constructed from over 41k real-world reports. Using PET-F2I-41K, we conduct a comprehensive evaluation of 27 models across proprietary frontier LLMs, open-source generalist models, and medical-domain LLMs, and we develop a domain-adapted 7B model (PET-F2I-7B) fine-tuned from Qwen2.5-7B-Instruct via LoRA. Beyond standard NLG metrics (e.g., BLEU-4, ROUGE-L, BERTScore), we propose three clinically grounded metrics - Entity Coverage Rate (ECR), Uncovered Entity Rate (UER), and Factual Consistency Rate (FCR) - to assess diagnostic completeness and factual reliability. Experiments reveal that neither frontier nor medical-domain LLMs perform adequately in zero-shot settings. In contrast, PET-F2I-7B achieves substantial gains (e.g., 0.708 BLEU-4) and a 3.0x improvement in entity coverage over the strongest baseline, while offering advantages in cost, latency, and privacy. Beyond this modeling contribution, PET-F2I-41K establishes a standardized evaluation framework to accelerate the development of reliable and clinically deployable reporting systems for PET/CT.

Assessing the Effectiveness of LLMs in Delivering Cognitive Behavioral Therapy

As mental health issues continue to rise globally, there is an increasing demand for accessible and scalable therapeutic solutions. Many individuals currently seek support from Large Language Models (LLMs), even though these models have not been validated for use in counseling services. In this paper, we evaluate LLMs' ability to emulate professional therapists practicing Cognitive Behavioral Therapy (CBT). Using anonymized, transcribed role-play sessions between licensed therapists and clients, we compare two approaches: (1) a generation-only method and (2) a Retrieval-Augmented Generation (RAG) approach using CBT guidelines. We evaluate both proprietary and open-source models for linguistic quality, semantic coherence, and therapeutic fidelity using standard natural language generation (NLG) metrics, natural language inference (NLI), and automated scoring for skills assessment. Our results indicate that while LLMs can generate CBT-like dialogues, they are limited in their ability to convey empathy and maintain consistency.

Who can we trust? LLM-as-a-jury for Comparative Assessment

Large language models (LLMs) are increasingly applied as automatic evaluators for natural language generation assessment often using pairwise comparative judgements. Existing approaches typically rely on single judges or aggregate multiple judges assuming equal reliability. In practice, LLM judges vary substantially in performance across tasks and aspects, and their judgment probabilities may be biased and inconsistent. Furthermore, human-labelled supervision for judge calibration may be unavailable. We first empirically demonstrate that inconsistencies in LLM comparison probabilities exist and show that it limits the effectiveness of direct probability-based ranking. To address this, we study the LLM-as-a-jury setting and propose BT-sigma, a judge-aware extension of the Bradley-Terry model that introduces a discriminator parameter for each judge to jointly infer item rankings and judge reliability from pairwise comparisons alone. Experiments on benchmark NLG evaluation datasets show that BT-sigma consistently outperforms averaging-based aggregation methods, and that the learned discriminator strongly correlates with independent measures of the cycle consistency of LLM judgments. Further analysis reveals that BT-sigma can be interpreted as an unsupervised calibration mechanism that improves aggregation by modelling judge reliability.

Online Domain-aware LLM Decoding for Continual Domain Evolution

LLMs are typically fine-tuned offline on domain-specific data, assuming a static domain. In practice, domain knowledge evolves continuously through new regulations, products, services, and interaction patterns. Retraining or fine-tuning LLMs for every new instance is computationally infeasible. Additionally, real-world environments also exhibit temporal dynamics with shifting data distributions. Disregarding this phenomenon, commonly referred to as concept drift, can significantly diminish a model's predictive accuracy. This mismatch between evolving domains and static adaptation pipelines highlights the need for efficient, real-time adaptation without costly retraining. In response, we introduce Online Domain-aware Decoding framework (ODD). ODD performs probability-level fusion between a base LLM and a prefix-tree prior, guided by adaptive confidence modulation using disagreement and continuity signals. Empirical evaluation under diverse drift scenarios demonstrates that ODD consistently surpasses LLM-Greedy and LLM-Temp Scaled across all syntactic and semantic NLG metrics. It yields an absolute ROUGE-L gain of 0.065 and a 13.6% relative improvement in Cosine Similarity over the best baseline. These results demonstrate ODD 's robustness to evolving lexical and contextual patterns, making it suitable for dynamic LLM applications.

Order in the Evaluation Court: A Critical Analysis of NLG Evaluation Trends

Despite advances in Natural Language Generation (NLG), evaluation remains challenging. Although various new metrics and LLM-as-a-judge (LaaJ) methods are proposed, human judgment persists as the gold standard. To systematically review how NLG evaluation has evolved, we employ an automatic information extraction scheme to gather key information from NLG papers, focusing on different evaluation methods (metrics, LaaJ and human evaluation). With extracted metadata from 14,171 papers across four major conferences (ACL, EMNLP, NAACL, and INLG) over the past six years, we reveal several critical findings: (1) Task Divergence: While Dialogue Generation demonstrates a rapid shift toward LaaJ (>40% in 2025), Machine Translation remains locked into n-gram metrics, and Question Answering exhibits a substantial decline in the proportion of studies conducting human evaluation. (2) Metric Inertia: Despite the development of semantic metrics, general-purpose metrics (e.g., BLEU, ROUGE) continue to be widely used across tasks without empirical justification, often lacking the discriminative power to distinguish between specific quality criteria. (3) Human-LaaJ Divergence: Our association analysis challenges the assumption that LLMs act as mere proxies for humans; LaaJ and human evaluations prioritize very different signals, and explicit validation is scarce (<8% of papers comparing the two), with only moderate to low correlation. Based on these observations, we derive practical recommendations to improve the rigor of future NLG evaluation.

From NLG Evaluation to Modern Student Assessment in the Era of ChatGPT: The Great Misalignment Problem and Pedagogical Multi-Factor Assessment (P-MFA)

This paper explores the growing epistemic parallel between NLG evaluation and grading of students in a Finnish University. We argue that both domains are experiencing a Great Misalignment Problem. As students increasingly use tools like ChatGPT to produce sophisticated outputs, traditional assessment methods that focus on final products rather than learning processes have lost their validity. To address this, we introduce the Pedagogical Multi-Factor Assessment (P-MFA) model, a process-based, multi-evidence framework inspired by the logic of multi-factor authentication.

CTest-Metric: A Unified Framework to Assess Clinical Validity of Metrics for CT Report Generation

In the generative AI era, where even critical medical tasks are increasingly automated, radiology report generation (RRG) continues to rely on suboptimal metrics for quality assessment. Developing domain-specific metrics has therefore been an active area of research, yet it remains challenging due to the lack of a unified, well-defined framework to assess their robustness and applicability in clinical contexts. To address this, we present CTest-Metric, a first unified metric assessment framework with three modules determining the clinical feasibility of metrics for CT RRG. The modules test: (i) Writing Style Generalizability (WSG) via LLM-based rephrasing; (ii) Synthetic Error Injection (SEI) at graded severities; and (iii) Metrics-vs-Expert correlation (MvE) using clinician ratings on 175 "disagreement" cases. Eight widely used metrics (BLEU, ROUGE, METEOR, BERTScore-F1, F1-RadGraph, RaTEScore, GREEN Score, CRG) are studied across seven LLMs built on a CT-CLIP encoder. Using our novel framework, we found that lexical NLG metrics are highly sensitive to stylistic variations; GREEN Score aligns best with expert judgments (Spearman~0.70), while CRG shows negative correlation; and BERTScore-F1 is least sensitive to factual error injection. We will release the framework, code, and allowable portion of the anonymized evaluation data (rephrased/error-injected CT reports), to facilitate reproducible benchmarking and future metric development.