China

The auto-regressive decoding of Large Language Models (LLMs) results in significant overheads in their hardware performance. While recent research has explored various speculative decoding techniques for multi-token generation, these methods introduce high memory costs from the additional weights and KV cache of separate draft models, limiting efficiency in edge and long-context scenarios. To overcome these limitations in edge-scale LLMs, we propose a novel parallel prompt decoding that requires only runtime memory overhead by employing a unified single model for both speculation and verification. Inspired by the human natural language generation process, PPD approximates outputs generated at future timesteps in parallel by using multiple prompt tokens. Furthermore, we present a hardware-aware two-stage tree pruning algorithm that adaptively optimizes this decoding scheme to fully leverage the computational capacities on different GPUs. Through extensive experiments across LLMs ranging from MobileLlama to Vicuna-13B on a wide range of benchmarks, our approach demonstrates up to 2.49 times speedup. Moreover, our parallel prompt decoding can serve as an orthogonal optimization for synergistic integration with existing speculative decoding, showing up to 1.22 times further speed improvement. To support future development, we have included our code implementation with this submission.

EMNLP 2025

Hardware-Aware Parallel Prompt Decoding for Memory-Efficient Acceleration of LLM Inference

poster

## Welcome!
"I am excited to welcome you to this year’s edition of the Conference on Empirical Methods in Natural Language Processing! Importantly, it marks the 30th edition of EMNLP. With over 8,000 submissions, more than 3,000 accepted papers, and thousands of attendees, we have come a long way from that first
workshop, which had 14 accepted papers. As the field looks ahead, Suzhou is the fitting location for celebrating this milestone: rooted in a long literary tradition, yet modern and forward-looking, and home to a large share of the NLP community."<br>

*Message from the General Chair, Dirk Hovy*

[**Link to Conference Handbook**](https://drive.google.com/file/d/1johU5QqVVYO4RfH7QcIORr7qrVBdzdwC/view?usp=sharing)





<br>

Celebrate 30 Years of EMNLP! 
EMNLP 2025 will be held in Suzhou, China from November 5th to November 9th, 2025.

Self-correction has demonstrated potential in code generation by allowing language models to revise and improve their outputs through successive refinement. Recent studies have explored prompting-based strategies that incorporate verification or feedback loops using proprietary models, as well as training-based methods that leverage their strong reasoning capabilities. However, whether smaller models possess the capacity to effectively guide their outputs through self-reflection remains unexplored. Our findings reveal that smaller models struggle to exhibit reflective revision behavior across both self-correction paradigms. In response, we introduce CoCoS, an approach designed to enhance the ability of small language models for multi-turn code correction. Specifically, we propose an online reinforcement learning objective that trains the model to confidently maintain correct outputs while progressively correcting incorrect outputs as turns proceed. Our approach features an accumulated reward function that aggregates rewards across the entire trajectory and a fine-grained reward better suited to multi-turn correction scenarios. This facilitates the model in enhancing initial response quality while achieving substantial improvements through self-correction. With 1B-scale models, CoCoS achieves improvements of 35.8% on the MBPP and 27.7% on HumanEval compared to the baselines.

Self-Correcting Code Generation Using Small Language Models

Scientific claim verification against tables typically requires predicting whether a claim is supported or refuted given a table. However, we argue that predicting the final label alone is insufficient: it reveals little about the model’s reasoning and offers limited interpretability. To address this, we reframe table–text alignment as an explanation task, requiring models to identify the table cells essential for claim verification. We build a new dataset by extending the SciTab benchmark with human-annotated cell-level rationales. Annotators verify the claim label and highlight the minimal set of cells needed to support their decision. After the annotation process, we utilize the collected information and propose a taxonomy for handling ambiguous cases. Our experiments show that (i) incorporating table alignment information improves claim verification performance, and (ii) most LLMs, while often predicting correct labels, fail to recover human-aligned rationales, suggesting that their predictions do not stem from faithful reasoning.

Table-Text Alignment: Explaining Claim Verification Against Tables in Scientific Papers

Knowledge Base Question Answering (KBQA) is a fundamental task that enables natural language interaction with structured knowledge bases (KBs).Given a natural language question, KBQA aims to retrieve the answers from the KB. However, existing approaches, including retrieval-based, semantic parsing-based methods and large-language model-based methods often suffer from generating non-executable queries and inefficiencies in query execution. To address these challenges, we propose GRV-KBQA, a three-stage framework that decouples logical structure generation from semantic grounding and incorporates structure-aware validation to enhance accuracy. Unlike previous methods, GRV-KBQA explicitly enforces KB constraints to improve alignment between generated logical forms and KB structures. Experimental results on WebQSP and CWQ show that GRV-KBQA significantly improves performance over existing approaches. The ablation study conducted confirms the effectiveness of the decoupled logical form generation and validation mechanism of our framework.

GRV-KBQA: A Three-Stage Framework for Knowledge Base Question Answering with Decoupled Logical Structure, Semantic Grounding and Structure-Aware Validation

For individuals who have experienced traumatic events such as strokes, speech may no longer be a viable means of communication. While text-to-speech (TTS) can be used as a communication aid since it generates synthetic speech, it fails to preserve the user’s own voice. As such, face-to-voice (FTV) synthesis, which derives corresponding voices from facial images, provides a promising alternative. However, existing methods rely on pre-trained visual encoders, and finetune them to align with speech embeddings, which strips fine-grained information from facial inputs such as gender or ethnicity, despite their known correlation with vocal traits. Moreover, these pipelines are multi-stage, which requires separate training of multiple components, thus leading to training inefficiency. To address these limitations, we utilize fine-grained facial attribute modeling by decomposing facial images into non-overlapping segments and progressively integrating them into a multi-granular representation. This representation is further refined through multi-task learning of speaker attributes such as gender and ethnicity at both the visual and acoustic domains. Moreover, to improve alignment robustness, we adopt a multi-view training strategy by pairing various visual perspectives of a speaker in terms of different angles and lighting conditions, with identical speech recordings. Extensive subjective and objective evaluations confirm that our approach substantially enhances face-voice congruence and synthesis stability.

Progressive Facial Granularity Aggregation with Bilateral Attribute-based Enhancement for Face-to-Speech Synthesis

Short texts pose significant challenges for clustering due to semantic sparsity, limited context, and fuzzy category boundaries. Although recent contrastive learning methods improve instance-level representation, they often overlook local semantic structure within the clustering head. Moreover, treating semantically similar neighbors as negatives impair cluster-level discrimination. To address these issues, we propose Fuzzy Neighborhood-Aware Self-Supervised Contrastive Clustering (FNSCC) framework. FNSCC incorporates neighborhood information at both the instance-level and cluster-level. At the instance-level, it excludes neighbors from the negative sample set to enhance inter-cluster separability. At the cluster-level, it introduces fuzzy neighborhood-aware weighting to refine soft assignment probabilities, encouraging alignment with semantically coherent clusters. Experiments on multiple benchmark short text datasets demonstrate that FNSCC consistently outperforms state-of-the-art models in accuracy and normalized mutual information. Our code is available at \url{https://anonymous.4open.science/r/FNSCC-E6FC}.

FNSCC: Fuzzy Neighborhood-Aware Self-Supervised Contrastive Clustering for Short Text

Understanding temporal concepts and answering time-sensitive questions is crucial yet a challenging task for question-answering systems powered by large language models (LLMs). Existing approaches either update the parametric knowledge of LLMs with new facts, which is resource-intensive and often impractical, or integrate LLMs with external knowledge retrieval (i.e., retrieval-augmented generation). However, off-the-shelf retrievers often struggle to identify relevant documents that require intensive temporal reasoning. To systematically study time-sensitive question answering, we introduce the TempRAGEval benchmark, which repurposes existing datasets by incorporating complex temporal perturbations and gold evidence labels. As anticipated, all existing retrieval methods struggle with these temporal reasoning-intensive questions. We further propose Modular Retrieval (MRAG), a trainless framework that includes three modules: (1) Question Processing that decomposes question into a main content and a temporal constraint; (2) Retrieval and Summarization that retrieves, splits, and summarize evidence passages based on the main content; (3) Semantic-Temporal Hybrid Ranking that scores semantic and temporal relevance separately for each fine-grained evidence. On TempRAGEval, MRAG significantly outperforms baseline retrievers in retrieval performance, leading to further improvements in final answer accuracy.

MRAG: A Modular Retrieval Framework for Time-Sensitive Question Answering

Knowledge distillation for large language models often uses Chain-of-Thought (CoT) and answer pairs, but existing methods struggle with appropriate supervision signals. Uniform constraints (e.g., cross-entropy) on CoT can enforce literal, verbose reasoning and suppress expressive diversity, while solely semantic constraints on answers can reduce accuracy in classification tasks. This paper proposes ThinkAnswer Loss, an information-theoretic differential supervision framework that decouples CoT and answer supervision. ThinkAnswer Loss applies semantic similarity constraints to the CoT portion while maintaining strict literal matching for the answer. We theoretically demonstrate its connection to mutual information maximization and derive a tight upper bound on generalization error. Experimental validation on text quality assessment and mathematical reasoning tasks shows that our method maintains answer accuracy while effectively reducing CoT length and preserving semantic content, thereby accelerating inference.

ThinkAnswer Loss: Balancing Semantic Similarity and Exact Matching for LLM Reasoning Enhancement

Evaluations of large language models (LLMs) suffer from instability, where small changes of random factors such as few-shot examples can lead to drastic fluctuations of scores and even model rankings. Moreover, different LLMs can have different preferences for a certain setting of random factors. As a result, using a fixed setting of random factors, which is often adopted as the paradigm of current evaluations, can lead to potential unfair comparisons between LLMs. To mitigate the volatility of evaluations, we first theoretically analyze the sources of variance induced by changes in random factors. Targeting these specific sources, we then propose the instance-level randomization (ILR) method to reduce variance and enhance fairness in model comparisons. Instead of using a fixed settings across the whole benchmark in a single experiment, we randomize all factors that affect evaluation scores for every single instance, run multiple experiments and report the averaged score. Theoretical analyses and empirical results demonstrate that ILR can reduce the variance and unfair comparisons caused by random factors, as well as achieve the same robustness level with less than half computational cost compared with previous methods.

Instance-level Randomization: Toward More Stable LLM Evaluations

The problem of data contamination is now almost inevitable during the development of large language models (LLMs), with the training data commonly integrating those evaluation benchmarks even unintentionally. This problem subsequently makes it hard to benchmark LLMs fairly. Instead of constructing contamination-free datasets (quite hard), we propose a novel framework, \textbf{LNE-Blocking}, to restore model performance prior to contamination on potentially leaked datasets. Our framework consists of two components: contamination detection and disruption operation. For the prompt, the framework first uses the contamination detection method, \textbf{LNE}, to assess the extent of contamination in the model. Based on this, it adjusts the intensity of the disruption operation, \textbf{Blocking}, to elicit non-memorized responses from the model. Our framework is the first to efficiently restore the model's greedy decoding performance. This comes with a strong performance on multiple datasets with potential leakage risks, and it consistently achieves stable recovery results across different models and varying levels of data contamination.

LNE-Blocking: An Efficient Framework for Contamination Mitigation Evaluation on Large Language Models

Counterfactual data augmentation (CDA) is a promising strategy for improving hate speech classification, but automating counterfactual text generation remains a challenge. Strong attribute control can distort meaning, while prioritizing semantic preservation may weaken attribute alignment. We propose Gradient-assisted Energy-based Sampling (GENES) for counterfactual text generation, which restricts accepted samples to text meeting a minimum BERTScore threshold and applies gradient-assisted proposal generation to improve attribute alignment. Compared to other methods that solely rely on either prompting, gradient-based steering, or energy-based sampling, GENES is more likely to jointly satisfy attribute alignment and semantic preservation under the same base model. In effect, using GENES as a counterfactual generator for data augmentation may improve out-of-domain performance of hate speech classifier while, at the minimum, maintaining the in-domain performance. Based on our cross-dataset evaluation, the average performance of models aided by GENES is the best among those methods that rely on a smaller model (Flan-T5-L). On the other hand, using similar augmentation techniques that rely on larger models (GPT-4o-mini) is slightly more robust based on average performance. Nonetheless, the results with GENES are comparable, making it a possible lightweight and open-source alternative.

Downloads

Next from EMNLP 2025

Self-Correcting Code Generation Using Small Language Models

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES