Automated extraction of core scientific claims, concise statements of a paper’s primary contributions, is critical for navigating the growing scientific literature. We present a scalable framework that leverages citances, sentences from other papers citing the target work, as natural supervision, removing the need for costly manual labelling. Our method filters citances with a claim-focused rubric and aligns them with candidate claims to train two pipelines: an unsupervised extractor and a weakly supervised model. Experiments show our approach outperforms existing baselines, achieving up to 18% higher precision and 22% greater coverage. We further analyse claim distributions across paper sections and introduce a taxonomy of claim types, providing new insights into the rhetorical structure of scientific discourse.

Mitigating elderly loneliness requires policy interventions that achieve both adaptability and auditability. Existing methods struggle to reconcile these objectives: traditional agent-based models suffer from static rigidity, while direct large language model (LLM) controllers lack essential traceability. This work proposes a three-layer framework that separates diagnosis from control to achieve both properties simultaneously. LLMs operate strictly as diagnostic instruments that assess population state and generate structured risk evaluations, while deterministic formulas with explicit bounds translate these assessments into traceable parameter updates. This separation ensures that every policy decision can be attributed to inspectable rules while maintaining adaptive response to emergent needs. We validate the framework through systematic ablation across five experimental conditions in elderly care simulation. Results demonstrate that explicit control rules outperform end-to-end black-box LLM approaches by 11.7% while preserving full auditability, confirming that transparency need not compromise adaptive performance.

Recent advances in deep learning have greatly enhanced the accuracy and scalability of animal re-identification by automating the extraction of subtle distinguishing features from images and videos. This enables large-scale, non-invasive monitoring of animal populations. This article proposes a segmentation pipeline and a re-identification model to identify animals without ground-truth IDs. The segmentation pipeline isolates animals from the background using bounding boxes and leverages the DINOv2 and Segment Anything Model 2 (SAM2) foundation models. For re-identification, Recurrence over Video Frames (RoVF) is introduced, a novel approach that employs a recurrent component based on the Perceiver transformer atop a DINOv2 image model, iteratively refining embeddings from video frames. The proposed methods are evaluated on video datasets of meerkats and polar bears (PolarBearVidID). The proposed segmentation model achieved high accuracy (94.36% and 97.26%) and IoU (73.14% and 92.77%) for meerkats and polar bears, respectively. RoVF outperformed frame- and video-based re-identification baselines, achieving a top-1 accuracy of 46.5% and 55% on masked test sets for meerkats and polar bears, respectively, as well as higher top-3 accuracy. These results highlight the potential of the proposed approach to reduce annotation burdens in future individual-based ecological studies. The code is available at https://github.com/Strong-AI-Lab/RoVF-Meerkat-Reidentification.

Predicting athletes’ behavior is important in designing game strategies and enhancing audience engagement in sports analytics. While machine learning has enabled advancements in action prediction across various sports, most existing models primarily capture correlations rather than underlying causal relationships. Here, we study long-term player action prediction in team sports (e.g., Rugby League) by integrating causal behavior modeling with Graph Neural Networks (GNNs). We assessed the model’s performance over extended video sequences, evaluating its resilience to varying observation window sizes. Our results demonstrate that incorporating causal structures significantly improves long-range action prediction accuracy, with the transformer-based graph neural network (TransformerConv) exhibiting superior robustness.

Puns represent a typical linguistic phenomenon that exploits polysemy and phonetic ambiguity to generate humour, posing unique challenges for natural language understanding. Within pun research, audio plays a central role in human communication except text and images, while datasets and systematic resources for spoken puns remain scarce, leaving this crucial modality largely underexplored. In this paper, we present APUN-Bench, the first benchmark dedicated to evaluating large audio language models (LALMs) on audio pun understanding. Our benchmark contains 4,434 audio samples annotated across three stages: pun recognition, pun word location and pun meaning inference. We conduct a deep analysis of APUN-Bench by systematically evaluating 10 state-of-the-art LALMs, uncovering substantial performance gaps in recognizing, localizing, and interpreting audio puns. This analysis reveals key challenges, such as positional biases in audio pun location and error cases in meaning inference, offering actionable insights for advancing humour-aware audio intelligence.

Humour translation plays a vital role as a bridge between different cultures, fostering understanding and communication. Although most existing Large Language Models (LLMs) are capable of general translation tasks, these models still struggle with humour translation, which is especially reflected through linguistic interference and lacking humour in translated text. In this paper, we propose a psychology-inspired Humour Decomposition Mechanism (HDM) that utilises Chain-of-Thought (CoT) to imitate the ability of the human thought process, stimulating LLMs to optimise the readability of translated humorous texts. Moreover, we integrate humour theory in HDM to further enhance the humorous elements in the translated text. Our automatic evaluation experiments on open-source humour datasets demonstrate that our method significantly improves the quality of humour translation, yielding average gains of 7.75

Solomonoff Induction is an optimal-in-the-limit unbounded algorithm for sequence prediction, representing a Bayesian mixture of every computable probability distribution and performing close to optimally in predicting any computable sequence. Being an optimal form of computational sequence prediction, it seems plausible that it may be used as a model against which other methods of sequence prediction might be compared. We put forth and explore the hypothesis that Transformer models - the basis of Large Language Models - approximate Solomonoff Induction better than any other extant sequence prediction method. We explore evidence for and against this hypothesis, give alternate hypotheses that take this evidence into account, and outline next steps for modelling Transformers and other kinds of AI in this way.

Large language models (LLMs) excel at many natural language tasks, yet their reasoning reliability under structured perturbations of rule-based systems remains brittle. We present a controlled evaluation framework consisting of four stress tests: (1) rule deletion (redundant vs. essential), (2) contradictory evidence injection, (3) logic-preserving rewrites, and (4) multi-law equivalence stacking. While representative model families (BERT, Qwen2, and TinyLlama) achieve Acc = 1.0000 on base tasks, our framework reveals a critical failure mode termed Logic Inertia - a total breakdown with Acc = 0.0000 under contradictions, where deductive momentum overrides factual reality. To address this, we propose Conflict-Aware Fusion (Fusion-Conflict), a framework grounded in the Cognitive Structure Hypothesis, which posits that robust reasoning requires an explicit structural inductive bias. By imposing a dual-process architecture that separates premise verification from logical deduction, Conflict-Aware Fusion effectively mitigates logic inertia under the proposed evaluation framework, achieving 1.0000 accuracy on both base and contradictory stress tests. It also significantly enhances robustness to missing evidence. Our results demonstrate that, for reliable multi-step reasoning, structural verification discipline is as critical as training data scale, providing a potential blueprint for building robust, contradiction-aware AI systems this https://github.com/14H034160212/lemo . See the OpenAI/Evals pull request this https://github.com/openai/evals/pull/1622 .

Developing benchmark datasets to tackle the bias problem in large language models (LLMs) is difficult for mixed-ethnic, small, and/or indigenous societies with limited resources. Existing bias benchmark datasets reflect the societal makeup of resource-rich societies such as the US and Europe. A deficit in available annotated datasets, the lack of annotators, and relevant LLM-generated text limit the potential for research in developing debiasing techniques for resource-restricted settings. Practices such as discarding data instances with annotator disagreement or obtaining a majority label from many annotators with multiple iterations of annotations are not applicable in this setting because it could lead to discrimination. Rather than discarding the information from such annotations, we propose utilising annotator disagreement information through a multi-annotator ensemble approach to build bias benchmark datasets. We capture annotator information by obtaining soft labels, which provide probability distributions over the hard labels that are either manually annotated or from pre-trained models. Firstly, we use pre-trained language models as an alternative for scenarios where manual annotations are restricted and demonstrate such readily accessible models yield similar or better performance than baseline aggregated manual annotator labels. Secondly, we demonstrate that classifications using the multi-annotator ensemble approach perform better than the single-label trained classification model.

Recent studies probing the Argument from the Poverty of the Stimulus (APS) have applied Large Language Models (LLMs) to test the learnability of complex syntax through surprisal-based metrics. However, divergent conclusions raise questions concerning the insights these metrics offer. While Wilcox et al. (2024) used direct minimal pair comparisons (the "wh-effect") to demonstrate that models successfully generalise knowledge of filler-gap dependencies, Lan et al. (2024) used a Difference-in-Differences (DiD) metric and found that models largely fail on parasitic gaps (PGs). This paper argues that the direct minimal pair approach offers greater diagnostic transparency. We demonstrate this by generating a full 8-permutation paradigm of refined PG stimuli and evaluating the GPT-2 model used in previous studies with a systematic Wilcox-style wh-effect analysis. Our results show that GPT-2 succeeds across all four tested conditions, indicating robust knowledge of filler-gap licensing principles even in complex PG environments. This finding, which contrasts with the more ambiguous results from DiD-style metrics, suggests that the choice of evaluation metric is critical for assessing an LLM's syntactic competence.

Large language models (LLMs) have demonstrated strong capabilities in language understanding and generation, and their potential in educational contexts is increasingly being explored. One promising area is learnersourcing, where students engage in creating their own educational content, such as multiple-choice questions. A critical step in this process is generating effective explanations for the solutions to these questions, as such explanations aid in peer understanding and promote deeper conceptual learning. However, students often find it difficult to craft high-quality explanations due to limited understanding or gaps in their subject knowledge. To support this task, we introduce "ILearner-LLM," a framework that uses iterative enhancement with LLMs to improve generated explanations. The framework combines an explanation generation model and an explanation evaluation model fine-tuned using student preferences for quality, where feedback from the evaluation model is fed back into the generation model to refine the output. Our experiments with LLaMA2-13B and GPT-4 using five large datasets from the PeerWise MCQ platform show that ILearner-LLM produces explanations of higher quality that closely align with those written by students. Our findings represent a promising approach for enriching the learnersourcing experience for students and for leveraging the capabilities of large language models for educational applications.

The long-standing goal of creating a comprehensive, multi-purpose knowledge resource, reminiscent of the 1984 Cyc project, still persists in AI. Despite the success of knowledge resources like WordNet, ConceptNet, Wolfram|Alpha and other commercial knowledge graphs, verifiable, general-purpose widely available sources of knowledge remain a critical deficiency in AI infrastructure. Large language models struggle due to knowledge gaps; robotic planning lacks necessary world knowledge; and the detection of factually false information relies heavily on human expertise. What kind of knowledge resource is most needed in AI today? How can modern technology shape its development and evaluation? A recent AAAI workshop gathered over 50 researchers to explore these questions. This paper synthesizes our findings and outlines a community-driven vision for a new knowledge infrastructure. In addition to leveraging contemporary advances in knowledge representation and reasoning, one promising idea is to build an open engineering framework to exploit knowledge modules effectively within the context of practical applications. Such a framework should include sets of conventions and social structures that are adopted by contributors.

Large language models (LLMs), such as LLaMA, Alpaca, Vicuna, GPT-3.5 and GPT-4, have advanced the performance of AI systems on various natural language processing tasks to human-like levels. However, their generalisation and robustness when performing logical reasoning has not been sufficiently assessed. To comprehensively evaluate this ability, we develop three new logical reasoning datasets named “ReClor-plus”, “LogiQA-plus” and “LogiQAv2-plus” that extend standard logical reasoning datasets to evaluate the robustness of the LLM’s reasoning. For each, we create three subsets: the first with randomly shuffled options, the second with the correct choices replaced by “none of the other options is correct”, and the third with a combination of shuffling and substitution. Experiments on these datasets show that these simple augmentations greatly hinder the models’ performance. Despite their high performance on the original publicly available datasets, we find that all models perform poorly on these newly constructed datasets. We also demonstrate that introducing task variations into the training set can markedly improve the model’s performance on both the original and our developed datasets. Finally, we show that applying logic-driven data augmentation for fine-tuning and prompting can enhance generalisation in both discriminative and generative models, offering a path to improving their robustness for tasks involving logical reasoning. Source code and data are made publicly available at ( https://github.com/Strong-AI-Lab/Logical-and-abstract-reasoning ).

Causal world models are systems that can answer counterfactual questions about an environment of interest, i.e. predict how it would have evolved if an arbitrary subset of events had been realized differently. It requires understanding the underlying causes behind chains of events and conducting causal inference for arbitrary unseen distributions. So far, this task eludes foundation models, notably large language models (LLMs), which do not have demonstrated causal reasoning capabilities beyond the memorization of existing causal relationships. Furthermore, evaluating counterfactuals in real-world applications is challenging since only the factual world is observed, limiting evaluation to synthetic datasets. We address these problems by explicitly extracting and modeling causal relationships and propose the Causal Cartographer framework. First, we introduce a graph retrieval-augmented generation agent tasked to retrieve causal relationships from data. This approach allows us to construct a large network of real-world causal relationships that can serve as a repository of causal knowledge and build real-world counterfactuals. In addition, we create a counterfactual reasoning agent constrained by causal relationships to perform reliable step-by-step causal inference. We show that our approach can extract causal knowledge and improve the robustness of LLMs for causal reasoning tasks while reducing inference costs and spurious correlations.

Pun generation seeks to creatively modify linguistic elements in text to produce humour or evoke double meanings. It also aims to preserve coherence and contextual appropriateness, making it useful in creative writing and entertainment across various media and contexts. Although pun generation has received considerable attention in computational linguistics, there is currently no dedicated survey that systematically reviews this specific area. To bridge this gap, this paper provides a comprehensive review of pun generation datasets and methods across different stages, including conventional approaches, deep learning techniques, and pre-trained language models. Additionally, we summarise both automated and human evaluation metrics used to assess the quality of pun generation. Finally, we discuss the research challenges and propose promising directions for future work.

Recent studies employing Large Language Models (LLMs) to test the Argument from the Poverty of the Stimulus (APS) have yielded contrasting results across syntactic phenomena. This paper investigates the hypothesis that characteristics of the stimuli used in recent studies, including lexical ambiguities and structural complexities, may confound model performance. A methodology is proposed for re-evaluating LLM competence on syntactic prediction, focusing on GPT-2. This involves: 1) establishing a baseline on previously used (both filtered and unfiltered) stimuli, and 2) generating a new, refined dataset using a state-of-the-art (SOTA) generative LLM (Gemini 2.5 Pro Preview) guided by linguistically-informed templates designed to mitigate identified confounds. Our preliminary findings indicate that GPT-2 demonstrates notably improved performance on these refined PG stimuli compared to baselines, suggesting that stimulus quality significantly influences outcomes in surprisal-based evaluations of LLM syntactic competency.

Graph perturbation hinders graph models in real applications, and thus defense methods against graph perturbation have been attracting increasing attention. However, current defense methods limit expressiveness and demand expert knowledge. To overcome these issues, in this paper, we propose a flexible-frequency graph transformer, building on the powerful expressive ability of self-attention. Specifically, we design a frequency-extraction self-attention with three heads to extract multi-frequency representations, i.e., low-frequency representation, hybrid-frequency representation, and high-frequency representation. An adaptive fusion method is then designed to combine diverse representations to output a flexible-frequency representation. This improves the model’s expressive ability and also enhances defense against graph perturbation by utilizing comprehensive information. In addition, we adaptively capture robust graph filters within self-attention, eliminating the need for expert knowledge. To boost self-attention’s effectiveness, we integrate graph learning to capture graph information before conducting node representation learning within self-attention. Furthermore, we theoretically analyze the feasibility of our proposed method. Extensive experiments demonstrate that our proposed method offers a dynamic and effective defense against graph perturbation compared to existing state-of-the-art methods.

Data augmentation is a popular technique to address the limited amount of training data available for machine learning models. However, existing approaches based on pretrained language models (PLMs) often suffer from limited diversity at the word or sub-word level and high costs associated with manual data collection and labeling. In this paper, we introduce a novel approach called DPTAK, which leverages the rich prior knowledge pre-learned by transformer-based PLMs to generate diverse and high-quality augmented data for text-to-data and data-to-text tasks. Unlike other methods, DPTAK retrieves associated knowledge with a given dataset and does not require manual data collection or labeling. Our experiments on E2E, WebNLG, and DART datasets demonstrate that DPTAK outperforms existing baseline models in terms of BLEU score by 0.37, 0.44, and 0.87, respectively, for the data-to-text task when applied with GPT-2. In text-to-data, DPTAK shows improvements of more than 0.44 BLEU score on E2E compared to other baseline methods. Moreover, DPTAK-augmented datasets exhibit the highest diversity scores among all existing data augmentation methods in data-to-text task, providing evidence of the effectiveness of our approach.

Current research for node classification focuses on dealing with either graph noise or label noise, but few studies consider both of them. In this paper, we propose a new robust node classification method to simultaneously deal with graph noise and label noise. To do this, we design a graph contrastive loss to conduct local graph learning and employ self-attention to conduct global graph learning. They enable us to improve the expressiveness of node representation by using comprehensive information among nodes. We also utilize pseudo graphs and pseudo labels to deal with graph noise and label noise, respectively. Furthermore, We numerically validate the superiority of our method in terms of robust node classification compared with all comparison methods.

Coreference resolution is the task of identifying and clustering mentions that refer to the same entity in a document. Based on state-of-the-art deep learning approaches, end-to-end coreference resolution considers all spans as candidate mentions and tackles mention detection and coreference resolution simultaneously. Recently, researchers have attempted to incorporate document-level context using higher-order inference (HOI) to improve end-to-end coreference resolution. However, HOI methods have been shown to have marginal or even negative impact on coreference resolution. In this paper, we reveal the reasons for the negative impact of HOI coreference resolution. Contextualized representations (e.g., those produced by BERT) for building span embeddings have been shown to be highly anisotropic. We show that HOI actually increases and thus worsens the anisotropy of span embeddings and makes it difficult to distinguish between related but distinct entities (e.g., pilots and flight attendants ). Instead of using HOI, we propose two methods, Less-Anisotropic Internal Representations (LAIR) and Data Augmentation with Document Synthesis and Mention Swap (DSMS), to learn less-anisotropic span embeddings for coreference resolution. LAIR uses a linear aggregation of the first layer and the topmost layer of contextualized embeddings. DSMS generates more diversified examples of related but distinct entities by synthesizing documents and by mention swapping. Our experiments show that less-anisotropic span embeddings improve the performance significantly (+2.8 F1 gain on the OntoNotes benchmark) reaching new state-of-the-art performance on the GAP dataset.