We revisit knowledge-based visual reasoning (KB-VR) in light of modern advances in multimodal large language models (MLLMs), and make the following contributions: (i) We propose Visual Knowledge Card (VKC) – a novel image that incorporates not only internal visual knowledge (e.g., scene-aware information) detected from the raw image, but also external world knowledge (e.g., attribute or object knowledge) produced by a knowledge generator; (ii) We present VKC-based Multi-Image Reasoning (VKC-MIR) – a four-stage pipeline which harnesses a state-of-the-art scene perception engine to construct an initial VKC (Stage-1), a powerful LLM to generate relevant domain knowledge (Stage-2), an excellent image editing toolkit to introduce generated knowledge into the updated VKC (Stage-3), and finally, an emerging multi-image MLLM to solve the VKC-enhanced task (Stage-4). By performing experiments on three popular KB-VR benchmarks, our approach achieves new state-of-the-art results compared to previous top-performing models.

Implicit assumptions and priors are often necessary in text-to-image generation tasks, especially when textual prompts lack sufficient context. However, these assumptions can sometimes reflect societal biases, low variance, or outdated concepts in the training data. We present Embedding-only Editing (EmbEdit), a method designed to efficiently edit implicit assumptions and priors in the text-to-image model without affecting unrelated objects or degrading overall performance. Given a “source” prompt (e.g., “nurse”) that elicits an assumption (e.g., a female nurse) and a “destination” prompt or distribution (e.g. equal gender chance), EmbEdit only fine-tunes the word token embedding (WTE) of the target object (i.e. token “nurse”’s WTE). Our method prevents unintended effects on other objects in the model’s knowledge base, as the WTEs for unrelated objects and the model weights remain unchanged. Further, our method can be applied to any text-to-image model with a text encoder. It is highly efficient, modifying only 768, 2048, and 4864 parameters for Stable Diffusion 1.4, Stable Diffusion XL, and FLUX, respectively, matching each model’s WTE dimension. Additionally, changes could be easily reversed by restoring the original WTE layers. The results show that EmbEdit outperforms previous methods in various models, tasks, and editing scenarios (both single and sequential multiple edits), achieving at least a 6.01% improvement (from 87.17% to 93.18%).

Extremely low-resource languages, especially those written in rare scripts, remain largely unsupported by large language models (LLMs). This is due in part to compounding factors such as the lack of training data. This paper delivers the first comprehensive analysis of whether LLMs can acquire such languages purely via in-context learning (ICL), with or without auxiliary alignment signals, and how these methods compare to parameter-efficient fine-tuning (PEFT). We systematically evaluate 20 under-represented languages across three state-of-the-art multilingual LLMs. Our findings highlight the limitation of PEFT when both language and its script are extremely under-represented by the LLM. In contrast, zero-shot ICL with language alignment is impressively effective on extremely low-resource languages, while few-shot ICL or PEFT is more beneficial for languages relatively better represented by LLMs. For LLM practitioners working on extremely low-resource languages, we summarise guidelines grounded by our results on adapting LLMs to low-resource languages, e.g., avoiding fine-tuning a multilingual model on languages of unseen scripts.

Recent work has demonstrated that using chain of thought (CoT), on soft-reasoning problems such as analytical and commonsense reasoning, often yields limited or even negative performance gains. CoT can also be unfaithful to the model’s actual reasoning. This paper investigates dynamics and unfaithfulness of CoT in soft-reasoning tasks across instruction-tuned, reasoning and reasoning-distilled models. Our findings show that distilled‐reasoning models rely heavily on CoT for these tasks, while instruction‐tuned and reasoning models often use it post‐hoc. Additionally, we find that CoT can steer model predictions without faithfully reflecting reasoning, indicating a disconnect between CoT influence and faithfulness.

In-context learning (ICL) performance is highly sensitive to prompt design, yet the impact of class label options (e.g. lexicon or order) in zero-shot classification remains underexplored. This study proposes LOADS (Label set Optimization via Activation Distribution kurtosiS), a post-hoc method for selecting optimal label sets in zero-shot ICL with large language models (LLMs).LOADS is built upon the observations in our empirical analysis, the first to systematically examine how label option design (i.e., lexical choice, order, and elaboration) impacts classification performance. This analysis shows that the lexical choice of the labels in the prompt (such as agree vs. support in stance classification) plays an important role in both model performance and model’s sensitivity to the label order. A further investigation demonstrates that optimal label words tend to activate fewer outlier neurons in LLMs’ feed-forward networks. LOADS then leverages kurtosis to measure the neuron activation distribution for label selection, requiring only a single forward pass without gradient propagation or labelled data. The LOADS-selected label words consistently demonstrate effectiveness for zero-shot ICL across classification tasks, datasets, models and languages, achieving maximum performance gain from 0.54 to 0.76 compared to the conventional approach of using original dataset label words.

Large language models (LLMs) often generate hallucinated content, making it crucial to identify and quantify inconsistencies in their outputs. We introduce HaluMap, a post-hoc framework that detects hallucinations by mapping entailment and contradiction relations between source inputs and generated outputs using a natural language inference (NLI) model. To improve reliability, we propose a calibration step leveraging intra-text relations to refine predictions. HaluMap outperforms state-of-the-art NLI-based methods by five percentage points compared to other training-free approaches, while providing clear, interpretable explanations. As a training-free and model-agnostic approach, HaluMap offers a practical solution for verifying LLM outputs across diverse NLP tasks. The resources of this paper are available at https://github.com/caisa-lab/acl25-halumap.

Knowledge editing methods (KEs) can update language models’ obsolete or inaccurate knowledge learned from pre-training. However, KEs can be used for malicious applications, e.g., inserting misinformation and toxic content. Knowing whether a generated output is based on edited knowledge or first-hand knowledge from pre-training can increase users’ trust in generative models and provide more transparency. Driven by this, we propose a novel task: detecting knowledge edits in language models. Given an edited model and a fact retrieved by a prompt from an edited model, the objective is to classify the knowledge as either unedited (based on the pre-training), or edited (based on subsequent editing). We instantiate the task with four KEs, two large language models (LLMs), and two datasets. Additionally, we propose using hidden state representations and probability distributions as features for the detection model. Our results reveal that using these features as inputs to a simple AdaBoost classifier establishes a strong baseline. This baseline classifier requires a small amount of training data and maintains its performance even in cross-domain settings. Our work lays the groundwork for addressing potential malicious model editing, which is a critical challenge associated with the strong generative capabilities of LLMs.

In-context knowledge editing (IKE) enables efficient modification of large language model (LLM) outputs without parameter changes and at zero-cost. However, it can be misused to manipulate responses opaquely, e.g., insert misinformation or offensive content. Such malicious interventions could be incorporated into high-level wrapped APIs where the final input prompt is not shown to end-users. To address this issue, we investigate the detection and reversal of IKE-edits. First, we demonstrate that IKE-edits can be detected with high accuracy (F1 > 80%) using only the top-10 output probabilities of the next token, even in a black-box setting, e.g. proprietary LLMs with limited output information. Further, we introduce the novel task of reversing IKE-edits using specially tuned reversal tokens. We explore using both continuous and discrete reversal tokens, achieving over 80% accuracy in recovering original, unedited outputs across multiple LLMs. Our continuous reversal tokens prove particularly effective, with minimal impact on unedited prompts. Through analysis of output distributions, attention patterns, and token rankings, we provide insights into IKE’s effects on LLMs and how reversal tokens mitigate them. This work represents a significant step towards enhancing LLM resilience against potential misuse of in-context editing, improving their transparency and trustworthiness.

Addressing online disinformation requires analysing narratives across languages to help fact-checkers and journalists sift through large amounts of data. The ExU project focuses on developing AI-based models for multilingual disinformation analysis, addressing the tasks of rumour stance classification and claim retrieval. We describe the ExU project proposal and summarise the results of a user requirements survey regarding the design of tools to support fact-checking.

In many real natural language processing application scenarios, practitioners not only aim to maximize predictive performance but also seek faithful explanations for the model predictions. Rationales and importance distribution given by feature attribution methods (FAs) provide insights into how different parts of the input contribute to a prediction. Previous studies have explored how different factors affect faithfulness, mainly in the context of monolingual English models. On the other hand, the differences in FA faithfulness between multilingual and monolingual models have yet to be explored. Our extensive experiments, covering five languages and five popular FAs, show that FA faithfulness varies between multilingual and monolingual models. We find that the larger the multilingual model, the less faithful the FAs are compared to its counterpart monolingual models. Our further analysis shows that the faithfulness disparity is potentially driven by the differences between model tokenizers. Our code is available: https://github.com/casszhao/multilingual-faith.

Despite the remarkable performance of generative large language models (LLMs) on abstractive summarization, they face two significant challenges: their considerable size and tendency to hallucinate. Hallucinations are concerning because they erode reliability and raise safety issues. Pruning is a technique that reduces model size by removing redundant weights, enabling more efficient sparse inference. Pruned models yield downstream task performance comparable to the original, making them ideal alternatives when operating on a limited budget. However, the effect that pruning has upon hallucinations in abstractive summarization with LLMs has yet to be explored. In this paper, we provide an extensive empirical study across five summarization datasets, two state-of-the-art pruning methods, and five instruction-tuned LLMs. Surprisingly, we find that hallucinations are less prevalent from pruned LLMs than the original models. Our analysis suggests that pruned models tend to depend more on the source document for summary generation. This leads to a higher lexical overlap between the generated summary and the source document, which could be a reason for the reduction in hallucination risk.1

Feature attribution methods (FAs) are popular approaches for providing insights into the model reasoning process of making predictions. The more faithful a FA is, the more accurately it reflects which parts of the input are more important for the prediction. Widely used faithfulness metrics, such as sufficiency and comprehensiveness use a hard erasure criterion, i.e. entirely removing or retaining the top most important tokens ranked by a given FA and observing the changes in predictive likelihood. However, this hard criterion ignores the importance of each individual token, treating them all equally for computing sufficiency and comprehensiveness. In this paper, we propose a simple yet effective soft erasure criterion. Instead of entirely removing or retaining tokens from the input, we randomly mask parts of the token vector representations proportionately to their FA importance. Extensive experiments across various natural language processing tasks and different FAs show that our soft-sufficiency and soft-comprehensiveness metrics consistently prefer more faithful explanations compared to hard sufficiency and comprehensiveness.

Explanation faithfulness of model predictions in natural language processing is typically evaluated on held-out data from the same temporal distribution as the training data (i.e. synchronous settings). While model performance often deteriorates due to temporal variation (i.e. temporal concept drift), it is currently unknown how explanation faithfulness is impacted when the time span of the target data is different from the data used to train the model (i.e. asynchronous settings). For this purpose, we examine the impact of temporal variation on model explanations extracted by eight feature attribution methods and three select-then-predict models across six text classification tasks. Our experiments show that (i) faithfulness is not consistent under temporal variations across feature attribution methods (e.g. it decreases or increases depending on the method), with an attention-based method demonstrating the most robust faithfulness scores across datasets; and (ii) select-then-predict models are mostly robust in asynchronous settings with only small degradation in predictive performance. Finally, feature attribution methods show conflicting behavior when used in FRESH (i.e. a select-and-predict model) and for measuring sufficiency/comprehensiveness (i.e. as post-hoc methods), suggesting that we need more robust metrics to evaluate post-hoc explanation faithfulness. Code will be made publicly available.