With the advent of large vision-language models (LVLMs) demonstrating increasingly human-like abilities, a pivotal question emerges: do different LVLMs interpret multimodal sarcasm differently, and can a single model grasp sarcasm from multiple perspectives like humans? To explore this, we introduce an analytical framework using systematically designed prompts on existing multimodal sarcasm datasets. Evaluating 12 state-of-the-art LVLMs over 2409 samples, we examine interpretive variations within and across models, focusing on confidence levels, alignment with dataset labels, and recognition of ambiguous “neutral” cases. We further validate our findings on a diverse 100-sample mini-benchmark, incorporating multiple datasets, expanded prompt variants, and representative commercial LVLMs. Our findings reveal notable discrepancies—across LVLMs and within the same model under varied prompts. While classification-oriented prompts yield higher internal consistency, models diverge markedly when tasked with interpretive reasoning. These results challenge binary labeling paradigms by highlighting sarcasm’s subjectivity. We advocate moving beyond rigid annotation schemes toward multiperspective, uncertainty-aware modeling, offering deeper insights into multimodal sarcasm comprehension.

Sarcasm in social media, frequently conveyed through the interplay of text and images, presents significant challenges for sentiment analysis and intention mining. Existing multi-modal sarcasm detection approaches have been shown to excessively depend on superficial cues within the textual modality, exhibiting limited capability to accurately discern sarcasm through subtle text–image interactions. To address this limitation, a novel framework, InterCLIP-MEP, is proposed. This framework integrates Interactive CLIP (InterCLIP), which employs an efficient training strategy to derive enriched cross-modal representations by embedding inter-modal information directly into each encoder, while using approximately 20.6 \(\times\) fewer trainable parameters compared with existing state-of-the-art (SOTA) methods. Furthermore, a Memory-Enhanced Predictor (MEP) is introduced, featuring a dynamic dual-channel memory mechanism that captures and retains valuable knowledge from test samples during inference, serving as a nonparametric classifier to enhance sarcasm detection robustness. Extensive experiments on MMSD, MMSD2.0, and DocMSU show that InterCLIP-MEP achieves SOTA performance, specifically improving accuracy by 1.08% and F1-score by 1.51% on MMSD2.0. Under distributional shift evaluation, it attains 73.96% accuracy, exceeding its memory-free variant by nearly 10% and the previous SOTA by over 15%, demonstrating superior stability and adaptability. The implementation of InterCLIP-MEP is publicly available at https://github.com/CoderChen01/InterCLIP-MEP .

Multimodal sarcasm detection identifies ironic intent by jointly analyzing text and images. It has attracted increasing attention due to its importance in understanding user-generated content on social media. However, multimodal observations are often incomplete due to data loss during transmission or collection, leading to unreliable predictions in real-world multimodal sarcasm detection scenarios. To address incomplete observations, we further propose a Reliability-Aware Dynamic Fusion (RADF) module, which predicts the reliability of the textual, visual, and interactive views from their representations, converts these reliability scores into dynamic fusion weights via a temperature-scaled softmax to control the sharpness of the weight distribution, and refines the fused features through feature-wise scaling. In this way, degraded views are suppressed while more informative views are emphasized under incomplete observations. Extensive experiments on public datasets validate the effectiveness of our approach, which consistently outperforms existing baselines.

Chinese abbreviation prediction is essential for various natural language processing tasks, including query comprehension, entity linking, and information retrieval.Existing approaches rely on sequence tagging for abbreviation prediction.However, these approaches fail to guarantee the predicted abbreviations preserve their respective meanings and only consider context information related to the entity itself.This paper proposes a novel Chinese abbreviation prediction approach using multi-feature fusion and global context.The approach initially generates multiple candidate abbreviations using a Chinese pretrained unbalanced Transformer generative model.It then selects high-quality candidates through a multi-feature optimization stage, and finally evaluates their quality within the global contextual information.Experimental results indicate that our approach surpasses other comparable approaches in abbreviation prediction.

Event detection is a fundamental task in information extraction, aiming to identify trigger words from given texts and categorize them into distinct event types.Existing approaches for event detection predominantly rely on annotating trigger words to classify events, which can lead to semantic recognition errors due to the oversimplified semantics of these trigger words.To tackle this challenge, we propose a trigger word expansionbased approach for robust contextual event detection.Specifically, we propose a framework comprising a trigger word extractor within a model for trigger word expansion classifier.This enables event detection to be conducted through trigger word expansion, enriching the semantics of trigger words.Additionally, we leverage the GPT model to generate contexts for the expanded trigger words, thereby enhancing the contextual understanding of trigger words.Extensive experiments conducted on the standard MAVEN benchmark dataset showcase the superior performance of our approach compared to state-of-the-art methods.This confirms the effectiveness of our proposed approach over existing approaches that rely on trigger word annotation.

Linguistic metonymy is a common type of figurative language in natural language processing (NLP), where a concept is represented by a closely associated word or phrase, for example “business executives suits”. As a result, metonymy resolution has become an important NLP task aimed at correctly identifying metonymic expressions within sentences. Previous approaches to this task have typically relied on pre-trained language models (PLMs) using a fine-tuning process. However, this can be time-consuming and resource-intensive, and may lead to a loss of factual prior knowledge. The emergence of a novel learning paradigm termed “prompt learning” or “prompt-tuning” has recently sparked widespread interest and captured considerable attention, as it has proven to yield remarkable results and surpass previous benchmarks. This approach uses a “pre-train→prompt→predict” paradigm and has been shown to better utilize the internal prior knowledge of a PLM, especially in situations with limited supervised resources. Inspired by this success, we investigated how prompt learning could improve metonymy resolution. We have developed a series of prompt learning approaches, called PromptMR, for metonymy resolution, and applied them to several widely-used metonymy resolution datasets. We also designed additional prompt-tuning augmentation strategies to further enhance the potential of prompt learning. Our experiments demonstrated that our method achieved state-of-the-art performance over multiple competitive baselines in both data-sufficient and data-scarce scenarios. The code implementations for PromptMR are accessible on GitHub via the URL: https://github.com/albert-jin/PromptTuning2MetonymyResolution.

Abstract Given a question and its answer candidates (named QA corpus), answer selection is the task of identifying the most relevant answers to the question. Answer selection is widely used in question answering, web search, and so on. Current deep neural network models primarily utilize local features extracted from input question‐answer pairs (QA pairs). However, the global features contained in QA corpora are under‐utilized, and we argue that these global features substantially contribute to the answer selection task. To verify this point of view, we propose a novel model that combines local and global features for answer selection. In our model, two different global feature extractors are employed to extract statistical global features and deep global features from a QA corpus, respectively. Furthermore, we investigate the integration of these global features with local features in various experimental settings: statistical global features, deep global features, and a combination of statistical and deep global features. Our experimental results show that the global features are effective for answer selection. Our model obtains new state‐of‐the‐art results on two public answer selection datasets and performs especially well on YahooCQA, where it achieves 9.2 and 6% higher precision@1 (P@1) and mean reciprocal rank (MRR) scores than previously published models.

Summary Taxonomic relations play an important role in various Natural Language Processing (NLP) tasks (eg, information extraction, question answering and knowledge inference). Existing approaches on embedding‐based taxonomic relation learning mainly rely on the word embeddings trained using co‐occurrence‐based similarity learning. However, the performance of these approaches is not quite satisfactory due to the lack of sufficient taxonomic semantic knowledge within word embeddings. To solve this problem, we propose an improved embedding‐based approach to learn taxonomic relations via incorporating relation descriptions into word embeddings. First, to capture additional taxonomic semantic knowledge, we train special word embeddings using not only co‐occurrence information of words but also relation descriptions (eg, taxonomic seed relations and their contextual triples). Then, using the trained word embeddings as features, we employ two learning models to identify and predict taxonomic relations, namely, offset‐based classification model and offset‐based similarity model. Experimental results on four real‐world domain datasets demonstrate that our proposed approach can capture additional taxonomic semantic knowledge and reduce dependence on the training dataset, outperforming the state‐of‐the‐art compared approaches on the taxonomic relation learning task.

In different contexts, one abstract concept (e.g., fruit) may be mapped into different concrete instance sets, which is called abstract concept instantiation. It has been widely applied in many applications, such as web search, intelligent recommendation, etc. However, in most abstract concept instantiation models have the following problems: (1) the neglect of incorrect label and label incompleteness in the category structure on which instance selection relies; (2) the subjective design of instance profile for calculating the relevance between instance and contextual constraint. The above problems lead to false prediction in terms of abstract concept instantiation. To tackle these problems, we proposed a novel model to instantiate the abstract concept. Firstly, to alleviate the incorrect label and remedy label incompleteness in the category structure, an improved random-walk algorithm is proposed, called InstanceRank, which not only utilize the category information, but it also exploits the association information to infer the right instances of an abstract concept. Secondly, for better measuring the relevance between instances and contextual constraint, we learn the proper instance profile from different granularity ones. They are designed based on the surrounding text of the instance. Finally, noise reduction and instance filtering are introduced to further enhance the model performance. Experiments on Chinese food abstract concept set show that the proposed model can effectively reduce false positive and false negative of instantiation results.

Summary A corpus (eg, patents or news texts) is an important knowledge resource that contains various topics, such as specific technologies or social events. Topic detection models of corpus, eg, Latent Dirichlet Allocation and KeyGraph, provide an important basis for exploring the status quo and trends in science, technology, or social events. However, these models suffer from low retrieval performance as they only consider text own explicit semantics in a single‐domain corpus. In addition, many incremental models, such as online‐LDA, are based on time slices. In this paper, a new topic detection model is proposed to improve the topic detection performance of a single‐domain corpus, which is inspired by a human memory cognitive process (THC). First, to improve the accuracy, distributions over words and inter‐word relations across a corpus are utilized as background knowledge, which is a type of implicit semantics, and we can find a more semantic‐sensitive part of texts. Second, to realize online topic detection without time slices, we introduce a probability gain‐based dynamic probabilistic model to detect latent topics by learning a model based on the dynamic human memory cognitive process. These two steps constitute the framework of our model. The experimental results for four public datasets (Reuters‐R8, Reuters‐R52, WebKB, and Cade12) reveal that our model is approximately ten percent higher than other baselines (eg, KeyGraph and LDA) on the Adjusted Rand Index (ARI).