Large language models (LLMs) have shown considerable promise in clinical natural language processing, yet few domain-specific datasets exist to rigorously evaluate their performance on radiology tasks. In this work, we introduce an annotated corpus of 6,393 radiology reports from 586 patients, each labeled for follow-up imaging status, to support the development and benchmarking of follow-up adherence detection systems. Using this corpus, we systematically compared traditional machine-learning classifiers—logistic regression (LR), support vector machines (SVM), Longformer, and a fully fine-tuned Llama3-8B-Instruct—with recent generative LLMs. To evaluate generative LLMs, we tested GPT-4o and the open-source GPT-OSS-20B under two configurations: a baseline (Base) and a task-optimized (Advanced) setting that focused inputs on metadata, recommendation sentences, and their surrounding context. A refined prompt for GPT-OSS-20B further improved reasoning accuracy. Performance was assessed using precision, recall, and F1 scores with 95% confidence intervals estimated via non-parametric bootstrapping. Inter-annotator agreement was high (F1 = 0.846). GPT-4o (Advanced) achieved the best performance (F1 = 0.832), followed closely by GPT-OSS-20B (Advanced; F1 = 0.828). LR and SVM also performed strongly (F1 = 0.776 and 0.775), underscoring that while LLMs approach human-level agreement through prompt optimization, interpretable and resource-efficient models remain valuable baselines.

Evaluating natural language generation (NLG) systems in the medical domain presents unique challenges due to the critical demands for accuracy, relevance, and domain-specific expertise. Traditional automatic evaluation metrics, such as BLEU, ROUGE, and BERTScore, often fall short in distinguishing between high-quality outputs, especially given the open-ended nature of medical question answering (QA) tasks where multiple valid responses exists. In this work, we introduce MORQA (Medical Open-Response QA), a new multilingual benchmark designed to assess the effectiveness of NLG evaluation metrics across three medical visual and text-based QA datasets in English and Chinese. Unlike prior resources, our datasets feature 2-4+ gold-standard answers authored by medical professionals, along with expert human ratings for three English and Chinese subsets. We benchmark both traditional metrics and large language model (LLM)-based evaluators, such as GPT-4 and Gemini, finding that LLM-based approaches significantly outperform traditional metrics in correlating with expert judgments. We further analyze factors driving this improvement, including LLMs’ sensitivity to semantic nuances and robustness to variability among reference answers. Our results provide the first comprehensive, multilingual qualitative study of NLG evaluation in the medical domain, highlighting the need for human-aligned evaluation methods. We release our code and annotations to support future research.

Remote clinical care has significantly increased the workload for healthcare professionals managing digital inquiries. While automated systems aim to alleviate this burden, consumer health questions present unique challenges due to their linguistic complexity and the need for proactive clinical guidance, which traditional question-answering models often overlook. We introduce the medical Response Content Units (RCU) schema, a framework that facilitates automatic analysis to identify question-answer completeness and critical answer subparts, which can then be used as tools for supporting clinician response or for automatic metric evaluation. Our analysis using this schema reveals a 16.4% gap in response completeness in professional replies and demonstrates that essential medical directives are provided 2.4 to 12.1 times as frequently as direct answers. We provide baseline results and publicly release our annotations and source code to offer an evaluation framework that is more closely aligned with real-world clinical requirements.

Large language models (LLMs) have demonstrated great performance across various benchmarks, showing potential as general-purpose task solvers. However, as LLMs are typically trained on vast amounts of data, a significant concern in their evaluation is data contamination, where overlap between training data and evaluation datasets inflates performance assessments. Multiple approaches have been developed to identify data contamination. These approaches rely on specific assumptions that may not hold universally across different settings. To bridge this gap, we systematically review 50 papers on data contamination detection, categorize the underlying assumptions, and assess whether they have been rigorously validated. We identify and analyze eight categories of assumptions and test three of them as case studies. Our case studies focus on detecting direct, instance-level data contamination, which is also referred to as Membership Inference Attacks (MIA). Our analysis reveals that MIA approaches based on these three assumptions can have similar performance to random guessing, on datasets used in LLM pretraining, suggesting that current LLMs might learn data distributions rather than memorizing individual instances. Meanwhile, MIA can easily fail when there are data distribution shifts between the seen and unseen instances.

Automated Claim Verification (CV)—the task of assessing a claim’s veracity against explicitly provided evidence—is a critical tool in the fight against growing misinformation. This survey offers a comprehensive analysis of 198 studies published between January 2022 and March 2025, synthesizing recent advances in CV corpus creation and system design. Through two in-depth case studies, we illuminate persistent challenges in veracity annotation, limitations of conventional CV pipelines, and pitfalls in recent claim decomposition approaches. We conclude by identifying key unresolved challenges and proposing productive directions for future research.

Several studies have shown that Large Language Models (LLMs) can answer medical questions correctly, even outperforming the average human score in some medical exams. However, to our knowledge, no study has been conducted to assess the ability of language models to validate existing or generated medical text for correctness and consistency. In this paper, we introduce MEDEC (https://github.com/abachaa/MEDEC), the first publicly available benchmark for medical error detection and correction in clinical notes, covering five types of errors (Diagnosis, Management, Treatment, Pharmacotherapy, and Causal Organism). MEDEC consists of 3,848 clinical texts, including 488 clinical notes from three US hospital systems that were not previously seen by any LLM. The dataset has been used in the MEDIQA-CORR 2024 shared task to evaluate seventeen participating systems. In this paper, we describe the data creation methods and we evaluate recent LLMs (e.g., o1-preview, GPT-4, Claude 3.5 Sonnet, Gemini 2.0 Flash, and DeepSeek-R1) for the tasks of detecting and correcting medical errors requiring both medical knowledge and reasoning capabilities. We also conducted a comparative study where two medical doctors performed the same task on the MEDEC test set. The results showed that MEDEC is a sufficiently challenging benchmark to assess the ability of models to validate existing or generated notes and to correct medical errors. We also found that although recent LLMs have a good performance in error detection and correction, they are still outperformed by medical doctors in these tasks. We discuss the potential factors behind this gap, the insights from our experiments, the limitations of current evaluation metrics, and share potential pointers for future research.

Gender bias has been extensively studied in both the educational field and the Natural Language Processing (NLP) field, the former using human coding to identify patterns associated with and causes of gender bias in text and the latter to detect, measure and mitigate gender bias in NLP output and models. This work aims to use NLP to facilitate automatic, quantitative analysis of educational text within the framework of a gender bias taxonomy. Analyses of both educational texts and a lexical resource (WordNet) reveal patterns of bias that can inform and aid educators in updating textbooks and lexical resources and in designing assessment items.

Recent progress in large language models (LLMs) has marked a notable milestone in the field of artificial intelligence. The conventional evaluation of LLMs primarily relies on existing tasks and benchmarks, raising concerns about test set contamination and the genuine comprehension abilities of LLMs. To address these concerns, we propose to evaluate LLMs by designing new tasks, automatically generating evaluation datasets for the tasks, and conducting detailed error analyses to scrutinize LLMs’ adaptability to new tasks, their sensitivity to prompt variations, and their error tendencies. We investigate the capacity of LLMs to adapt to new but simple tasks, especially when they diverge from the models’ pre-existing knowledge. Our methodology emphasizes the creation of straightforward tasks, facilitating a precise error analysis to uncover the underlying causes of LLM failures. This strategic approach also aims to uncover effective strategies for enhancing LLM performance based on the detailed error analysis of system output.

In many practical scenarios, contents from different modalities are not semantically aligned; for instance, visual and textual information may conflict with each other, resulting in non-compositional expression effects such as irony or humor. Effective modeling and smooth integration of multimodal information are crucial for achieving good understanding of the contrast across modalities. Being focusing on image-text matching, most current studies face challenges in identifying such contrast, leading to limitations in exploring the extended semantics when images and texts do not match. In this paper, we propose an LLM-based approach for learning multimodal contrast following the encoding-decoding paradigm, enhanced by a memory module with reinforced contrast recognition, and use a series of tasks that have the nature of multimodal contrast to verify our approach. The memory module learns the integration between visual and textual features with trainable memory vectors and the reinforced contrast recognition uses self-rejection sampling to optimize the memory to further enhance learning multimodal contrast. The resulted information, accompanied with visual and text features, is finally fed into the LLM to predict corresponding labels. We experiment our approach on four English and Chinese benchmark datasets, where it outperforms strong baselines and state-of-the-art studies.

End-to-end Aspect-based Sentiment Analysis (EASA) is a natural language processing (NLP) task that involves extracting aspect terms and identifying the sentiments for them, which provides a fine-grained level of text analysis and thus requires a deep understanding of the running text. Many previous studies leverage advanced text encoders to extract context information and use syntactic information, e.g., the dependency structure of the input sentence, to improve the model performance. However, such models may reach a bottleneck since the dependency structure is not designed to provide semantic information of the text, which is also important for identifying the sentiment and thus leave room for further improvement. Considering that combinatory categorial grammar (CCG) is a formalism that expresses both syntactic and semantic information of a sentence, it has the potential to be beneficial to EASA. In this paper, we propose a novel approach to improve EASA with CCG supertags, which carry the syntactic and semantic information of the associated words and serve as the most important part of the CCG derivation. Specifically, our approach proposes a CCG supertag decoding process to learn the syntactic and semantic information carried by CCG supertags and use the information to guide the attention over the input words so as to identify important contextual information for EASA. Furthermore, a gate mechanism is used in incorporating the weighted contextual information into the backbone EASA decoding process. We evaluate our approach on three publicly available English datasets for EASA, and show that it outperforms strong baselines and achieves state-of-the-art results on all datasets.

Relation extraction (RE) is an important natural language processing task that predicts the relation between two given entities, where a good understanding of the contextual information is essential to achieve an outstanding model performance. Among different types of contextual information, the auto-generated syntactic information (namely, word dependencies) has shown its effectiveness for the task. However, most existing studies require modifications to the existing baseline architectures (e.g., adding new components, such as GCN, on the top of an encoder) to leverage the syntactic information. To offer an alternative solution, we propose to leverage syntactic information to improve RE by training a syntax-induced encoder on auto-parsed data through dependency masking. Specifically, the syntax-induced encoder is trained by recovering the masked dependency connections and types in first, second, and third orders, which significantly differs from existing studies that train language models or word embeddings by predicting the context words along the dependency paths. Experimental results on two English benchmark datasets, namely, ACE2005EN and SemEval 2010 Task 8 datasets, demonstrate the effectiveness of our approach for RE, where our approach outperforms strong baselines and achieve state-of-the-art results on both datasets.