Enabling digital humans to express rich emotions has significant applications in dialogue systems, gaming, and other interactive scenarios. While recent advances in talking head synthesis have achieved impressive results in lip synchronization, they tend to overlook the rich and dynamic nature of facial expressions. To fill this critical gap, we introduce an end-to-end text-to-expression model that explicitly focuses on emotional dynamics. Our model learns expressive facial variations in a continuous latent space and generates expressions that are diverse, fluid, and emotionally coherent. To support this task, we introduce EmoAva, a large-scale and high-quality dataset containing 15,000 text–3D expression pairs. Extensive experiments on both existing datasets and EmoAva demonstrate that our method significantly outperforms baselines across multiple evaluation metrics, marking a significant advancement in the field.
论文检索
输入标题、作者或关键词，从 7,591 篇学术成果中精准定位
Annotating Training Data for Conditional Semantic Textual Similarity Measurement using Large Language Models
PDF ↗Semantic similarity between two sentences depends on the aspects considered between those sentences. To study this phenomenon, Deshpande et al. (2023) proposed the Conditional Semantic Textual Similarity (C-STS) task and annotated a human-rated similarity dataset containing pairs of sentences compared under two different conditions. However, Tu et al. (2024) found various annotation issues in this dataset and showed that manually re-annotating a small portion of it leads to more accurate C-STS models. Despite these pioneering efforts, the lack of large and accurately annotated C-STS datasets remains a blocker for making progress on this task as evidenced by the subpar performance of the C-STS models. To address this training data need, we resort to Large Language Models (LLMs) to correct the condition statements and similarity ratings in the original dataset proposed by Deshpande et al. (2023). Our proposed method is able to re-annotate a large training dataset for the C-STS task with minimal manual effort. Importantly, by training a supervised C-STS model on our cleaned and re-annotated dataset, we achieve a 5.4% statistically significant improvement in Spearman correlation. The re-annotated dataset is available at https://LivNLP.github.io/CSTS-reannotation.
Role-playing capabilities in large language models (LLMs) often lack cognitive consistency in complex scenarios that require deep understanding and coherent reasoning. While recent reasoning models excel in math and coding tasks, they show limited effectiveness in open-ended role-playing scenarios. We introduce R-CHAR (Role-Consistent Hierarchical Adaptive Reasoning), a metacognition-driven framework that enhances role-playing performance through guided thinking trajectories synthesis and adaptive evaluation. Our approach demonstrates that concise thinking processes can achieve superior performance efficiently compared to elaborate reasoning chains in role-playing social intelligence tasks, outperforming existing specialized models. Experimental results on the SocialBench benchmark show significant and stable performance improvements across varying scenario complexities, showing particular strength in long-context comprehension (from 34.64% to 68.59%) and group-level social interactions. Our work advances the development of cognitively consistent role-playing systems, bridging the gap between surface-level mimicry and authentic character simulation.
Retrieval-augmented generation (RAG) enhances the capabilities of large language models (LLMs) by incorporating external knowledge into their input prompts. However, when the retrieved context contradicts the LLM’s parametric knowledge, it often fails to resolve the conflict between incorrect external context and correct parametric knowledge, known as context-memory conflict. To tackle this problem, we introduce Conflict-Aware REtrieval-Augmented Generation (CARE), consisting of a context assessor and a base LLM. The context assessor encodes external context into compact memory embeddings. Through grounded/adversarial soft prompting, the context assessor is trained to discern unreliable context and capture a guidance signal that directs reasoning toward the more reliable knowledge source. Extensive experiments show that CARE effectively mitigates context-memory conflicts, leading to an average performance gain of 5.0% on QA and fact-checking benchmarks, establishing a promising direction for trustworthy and adaptive RAG systems.
Dehumanization, i.e., denying human qualities to individuals or groups, is a particularly harmful form of hate speech that can normalize violence against marginalized communities. Despite advances in NLP for detecting general hate speech, approaches to identifying dehumanizing language remain limited due to scarce annotated data and the subtle nature of such expressions. In this work, we systematically evaluate four state-of-the-art large language models (LLMs) — Claude, GPT, Mistral, and Qwen — for dehumanization detection.Our results show that only one model—Claude—achieves strong performance (over 80% F_1) under an optimized configuration, while others, despite their capabilities, perform only moderately. Performance drops further when distinguishing dehumanization from related hate types such as derogation. We also identify systematic disparities across target groups: models tend to over-predict dehumanization for some identities (e.g., Gay men), while under-identifying it for others (e.g., Refugees). These findings motivate the need for systematic, group-level evaluation when applying pretrained language models to dehumanization detection tasks.
Multilingual LLM performance is often critically dependent on model size. With an eye on efficiency, this has led to a surge in interest in one-shot pruning methods that retain the benefits of large-scale pretraining while shrinking the model size. However, as pruning tends to come with performance loss, it is important to understand the trade-offs between multilinguality and sparsification. In this work, we study multilingual performance under different sparsity constraints and show that moderate ratios already substantially harm performance. To help bridge this gap, we propose M-Wanda, a pruning method that models cross-lingual variation by incorporating language-aware activation statistics into its pruning criterion and dynamically adjusts layerwise sparsity based on cross-lingual importance. We show that M-Wanda consistently improves performance at minimal additional costs. We are the first to explicitly optimize pruning to retain multilingual performance, and hope to inspire future advances in multilingual pruning.
Leveraging Text-to-Text Transformers as Classifier Chain for Few-Shot Multi-Label Classification
PDF ↗Multilabel text classification (MLTC) is an essential task in NLP applications. Traditional methods require extensive labeled data and are limited to fixed label sets. Extracting labels by LLMs is more effective and universal, but incurs high computational costs. In this work, we introduce a distillation-based T5 generalist model for zero-shot MLTC and few-shot fine-tuning. Our model accommodates variable label sets with general domain-agnostic pertaining, while modeling dependency between labels. Experiments show that our approach outperforms baselines of similar size on three few-shot tasks. Our code is available at repository.
Vision-Language-Action (VLA) models have made substantial progress by leveraging the robust capabilities of Visual Language Models (VLMs). However, VLMs’ significant parameter size and autoregressive (AR) decoding nature impose considerable computational demands on VLA models. While Speculative Decoding (SD) has shown efficacy in accelerating Large Language Models (LLMs) by incorporating efficient drafting and parallel verification, allowing multiple tokens to be generated in one forward pass, its application to VLA models remains unexplored. This work introduces Spec-VLA, an SD framework designed to accelerate VLA models. Due to the difficulty of the action prediction task and the greedy decoding mechanism of the VLA models, the direct application of the advanced SD framework to the VLA prediction task yields a minor speed improvement. To boost the generation speed, we propose an effective mechanism to relax acceptance utilizing the relative distances represented by the action tokens of the VLA model. Empirical results across diverse test scenarios affirm the effectiveness of the Spec-VLA framework, and further analysis substantiates the impact of our proposed strategies, which enhance the acceptance length by 44%, achieving 1.42\times speedup compared with the OpenVLA baseline, without compromising the success rate. The success of the Spec-VLA framework highlights the potential for broader application of speculative execution in VLA prediction scenarios.
Natural language chain-of-thought (N-CoT) and Program chain-of-thought (P-CoT) have emerged as two primary paradigms for large language models (LLMs) to solve mathematical reasoning problems. Current research typically endeavors to achieve unidirectional enhancement: P-CoT enhanced N-CoT or N-CoT enhanced P-CoT. In this paper, we seek to fully unleash the two paradigms’ strengths for mutual enhancement and ultimately achieve simultaneous improvements. We conduct a detailed analysis of the error types across two paradigms, based on which we propose Parrot, a novel training pipeline for mathematical problems: 1) Three target-designed subtasks integrate sequential P-CoT and N-CoT generation. 2) A subtask hybrid training strategy to facilitate natural language semantic transferability. 3) The converted N-CoT auxiliary reward is designed to alleviate the sparse rewards in P-CoT optimization. Extensive experiments demonstrate that Parrot significantly enhances both the performance of N-CoT and P-CoT, especially on N-CoT. Using Parrot SFT, the LLaMA2’s and CodeLLaMA’s N-CoT performance achieve gains of +21.87 and +21.48 on MathQA over the RL baseline, which is resource-intensive.
Large language models (LLMs) often leverage adapters, such as low-rank-based adapters, to achieve strong performance on downstream tasks. However, storing a separate adapter for each task significantly increases memory requirements, posing a challenge for resource-constrained environ ments such as mobile devices. Although model merging techniques can reduce storage costs, they typically result in substantial performance degradation. In this work, we introduce HydraOpt, a new model merging technique that capitalizes on the inherent similarities between the matrices of low-rank adapters. Unlike existing methods that produce a fixed trade-off between storage size and performance, HydraOpt allows us to navigate this spectrum of efficiency and performance. Our experiments show that HydraOpt significantly reduces storage size (48% reduction) compared to storing all adapters, while achieving competitive performance (0.2-1.8% drop). Furthermore, it outperforms existing merging techniques in terms of performance at the same or slightly worse storage efficiency.
Principled Personas: Defining and Measuring the Intended Effects of Persona Prompting on Task Performance
PDF ↗Expert persona prompting—assigning roles such as expert in math to language models—is widely used for task improvement. However, prior work shows mixed results on its effectiveness, and does not consider when and why personas should improve performance. We analyze the literature on persona prompting for task improvement and distill three desiderata: 1) performance advantage of expert personas, 2) robustness to irrelevant persona attributes, and 3) fidelity to persona attributes. We then evaluate 9 state-of-the-art LLMs across 27 tasks with respect to these desiderata. We find that expert personas usually lead to positive or non-significant performance changes. Surprisingly, models are highly sensitive to irrelevant persona details, with performance drops of almost 30 percentage points. In terms of fidelity, we find that while higher education, specialization, and domain-relatedness can boost performance, their effects are often inconsistent or negligible across tasks. We propose mitigation strategies to improve robustness—but find they only work for the largest, most capable models. Our findings underscore the need for more careful persona design and for evaluation schemes that reflect the intended effects of persona usage.
Studying the Role of Input-Neighbor Overlap in Retrieval-Augmented Language Models Training Efficiency
PDF ↗Retrieval-augmented language models have demonstrated performance comparable to much larger models while requiring fewer computational resources. The effectiveness of these models crucially depends on the overlap between query and retrieved context, but the optimal degree of this overlap remains unexplored. In this paper, we systematically investigate how varying levels of query–context overlap affect model performance during both training and inference. Our experiments reveal that increased overlap initially has minimal effect, but substantially improves test-time perplexity and accelerates model learning above a critical threshold. Building on these findings, we demonstrate that deliberately increasing overlap through synthetic context can enhance data efficiency and reduce training time by approximately 40% without compromising performance. We specifically generate synthetic context through paraphrasing queries. We validate our perplexity-based findings on question-answering tasks, confirming that the benefits of retrieval-augmented language modeling extend to practical applications. Our results provide empirical evidence of significant optimization potential for retrieval mechanisms in language model pretraining.
Large Language Models Badly Generalize across Option Length, Problem Types, and Irrelevant Noun Replacements
PDF ↗In this paper, we propose a “Generalization Stress Test” to assess Large Language Models’ (LLMs) generalization ability under slight and controlled perturbations, including option length, problem types, and irrelevant noun replacements. We achieve novel and significant findings that, despite high benchmark scores, LLMs exhibit severe accuracy drops and unexpected biases (e.g., preference for longer distractors) when faced with these minor but content-preserving modifications. For example, Qwen 2.5 1.5B’s MMLU score rises from 60 to 89 and drops from 89 to 36 when option lengths are changed without altering the question. Even GPT4o experiences a 25-point accuracy loss when problem types are changed, with a 6-point drop across all three modification categories. These analyses suggest that LLMs rely heavily on superficial cues rather than forming robust, abstract representations that generalize across formats, lexical variations, and shifts in irrelevant content.
Legal citation detection in court judgments underpins reliable precedent mapping, citation analytics, and document retrieval. Extracting references to legislation and case law in the United Kingdom is especially challenging: citation styles have evolved over centuries, and judgments routinely cite foreign or historical authorities. We conduct the first systematic comparison of three modelling paradigms on this task using the Cambridge Law Corpus: (i) rule‐based regular expressions; (ii) transformer-based encoders (BERT, RoBERTa, LEGAL‐BERT, ModernBERT); and (iii) large language models (GPT‐4.1). We produced a gold‐standard high-quality corpus of 190 court judgments containing 45,179 fine-grained annotations for UK and non-UK legislation and case references. ModernBERT achieves a macro-averaged F1 of 93.3%, only marginally ahead of the other encoder-only models, yet significantly outperforming the strongest regular-expression baseline (35.42% F1) and GPT-4.1 (76.57% F1).
Are LLMs Better than Reported? Detecting Label Errors and Mitigating Their Effect on Model Performance
PDF ↗NLP benchmarks rely on standardized datasets for training and evaluating models and are crucial for advancing the field. Traditionally, expert annotations ensure high-quality labels; however, the cost of expert annotation does not scale well with the growing demand for larger datasets required by modern models. While crowd-sourcing provides a more scalable solution, it often comes at the expense of annotation precision and consistency. Recent advancements in large language models (LLMs) offer new opportunities to enhance the annotation process, particularly for detecting label errors in existing datasets. In this work, we consider the recent approach of LLM-as-a-judge, leveraging an ensemble of LLMs to flag potentially mislabeled examples. We conduct a case study on four factual consistency datasets from the TRUE benchmark, spanning diverse NLP tasks, and on SummEval, which uses Likert-scale ratings of summary quality across multiple dimensions. We empirically analyze the labeling quality of existing datasets and compare expert, crowd-sourced, and LLM-based annotations in terms of the agreement, label quality, and efficiency, demonstrating the strengths and limitations of each annotation method. Our findings reveal a substantial number of label errors, which, when corrected, induce a significant upward shift in reported model performance. This suggests that many of the LLMs’ so-called mistakes are due to label errors rather than genuine model failures. Additionally, we discuss the implications of mislabeled data and propose methods to mitigate them in training to improve performance.
Vision-Language Models (VLMs) are known to struggle with spatial reasoning and visual alignment. To help overcome these limitations, we introduce iVISPAR, an interactive multimodal benchmark designed to evaluate the spatial reasoning capabilities of VLMs acting as agents. iVISPAR is based on a variant of the sliding tile puzzle—a classic problem that demands logical planning, spatial awareness, and multi-step reasoning. The benchmark supports visual 3D, 2D, and text-based input modalities, enabling comprehensive assessments of VLMs’ planning and reasoning skills. We evaluate a broad suite of state-of-the-art open-source and closed-source VLMs, comparing their performance while also providing optimal path solutions and a human baseline to assess the task’s complexity and feasibility for humans. Results indicate that while VLMs perform better on 2D tasks compared to 3D or text-based settings, they struggle with complex spatial configurations and consistently fall short of human performance, illustrating the persistent challenge of visual alignment. This underscores critical gaps in current VLM capabilities, highlighting their limitations in achieving human-level cognition. Project website: https://microcosm.ai/ivispar.
Larger models often outperform smaller ones but come with high computational costs. Cascading offers a potential solution. By default, it uses smaller models and defers only some instances to larger, more powerful models. However, designing effective deferral rules remains a challenge. In this paper, we propose a simple yet effective approach for machine translation, using existing quality estimation (QE) metrics as deferral rules. We show that QE-based deferral allows a cascaded system to match the performance of a larger model while invoking it for a small fraction (30% to 50%) of the examples, significantly reducing computational costs. We validate this approach through both automatic and human evaluation.
High-quality image captions are essential for improving modality alignment and visual understanding in Large Vision-Language Models (LVLMs). However, the scarcity of ultra-detailed image caption data limits further advancements. This paper presents a systematic pipeline for generating high-quality, ultra-detailed image captions, encompassing both pre-processing and post-processing stages. In the pre-processing stage, we classify and deduplicate images, extract visual information using expert tools, and leverage GPT-4o with structured prompts to generate initial captions. To enhance comprehensiveness, we introduce an expansion strategy based on Large Language Models (LLMs), defining eight descriptive dimensions to refine and extend captions, which serve as seed data for training a proprietary captioner model. In the post-processing stage, we incorporate human error-correction annotations and an active learning-inspired approach to refine low-quality samples. Using high-quality corrected data, we apply Direct Preference Optimization (DPO) and develop a critic-rewrite pipeline, training a sentence-level critic model to mitigate hallucinations. Experimental results demonstrate that our ultra-detailed captions significantly enhance LVLMs’ perception and cognitive abilities across multiple vision-language benchmarks. The code and dataset are available at https://github.com/yuzeng0-0/UltraCaption.
Pretrained language models (LMs) are prone to arithmetic errors. Existing work showed limited success in probing numeric values from models’ representations, indicating that these errors can be attributed to the inherent unreliability of distributionally learned embeddings in representing exact quantities. However, we observe that previous probing methods are inadequate for the emergent structure of learned number embeddings with sinusoidal patterns.In response, we propose a novel probing technique that decodes numeric values from input embeddings with near-perfect accuracy across a range of open-source LMs. This proves that after the sole pre-training, LMs represent numbers with remarkable precision. Finally, we find that the embeddings’ preciseness judged by our probe’s accuracy explains a large portion of LM’s errors in elementary arithmetic, and show that aligning the embeddings with the pattern discovered by our probe can mitigate these errors.
CRITICTOOL: Evaluating Self-Critique Capabilities of Large Language Models in Tool-Calling Error Scenarios
PDF ↗The ability of large language models (LLMs) to utilize external tools has enabled them to tackle an increasingly diverse range of tasks. However, as the tasks become more complex and long-horizon, the intricate tool utilization process may trigger various unexpected errors. Therefore, how to effectively handle such errors, including identifying, diagnosing, and recovering from them, has emerged as a key research direction for advancing tool learning. In this work, we first extensively analyze the types of errors encountered during the function-calling process on several competitive tool evaluation benchmarks. Based on it, we introduce CRITICTOOL, a comprehensive critique evaluation benchmark specialized for tool learning. Building upon a novel evolutionary strategy for dataset construction, CRITICTOOL holds diverse tool-use errors with varying complexities, which better reflects real-world scenarios. We conduct extensive experiments on CRITICTOOL, and validate the generalization and effectiveness of our constructed benchmark strategy. We also provide an in-depth analysis of the tool reflection ability on various LLMs, offering a new perspective on the field of tool learning in LLMs. The code is available at https://github.com/Shellorley0513/CriticTool.