China

Large Language Models (LLMs) require alignment via reinforcement learning (RL) to effectively perform task-specific objectives, such as human preference alignment and enhanced reasoning. While Proximal Policy Optimization (PPO) is widely adopted, its computational overhead, stemming from additional value model requirements, limits applicability. Existing alternatives, like Group Relative Policy Optimization (GRPO), mitigate computational costs but remain sensitive to reward model quality. To address this, we introduce Group Preference Reward Shaping (GPRS), a novel method that leverages preference-based comparisons rather than precise numerical rewards. GPRS requires no extra model components and remains robust across varying reward model sizes and qualities. Extensive experiments demonstrate that GPRS consistently outperforms existing critic-model-free RL algorithms in Reinforcement Learning from Human Feedback (RLHF) and reasoning tasks, providing stable and good alignment performance.

EMNLP 2025

Reinforcement Learning for Large Language Models via Group Preference Reward Shaping

reinforcement learning; human preference alignment; large language models

## 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*





<br>

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

poster

Recent advancements in large language models (LLMs) have shown impressive capabilities in various downstream tasks but typically face Catastrophic Forgetting (CF) during fine-tuning. In this paper, we propose the Forgetting-Aware Pruning Metric (FAPM), a novel pruning-based approach to balance CF and downstream task performance. Our investigation reveals that the degree to which task vectors (i.e., the subtraction of pre-trained weights from the weights fine-tuned on downstream tasks) overlap with pre-trained model parameters is a critical factor for CF. Based on this finding, FAPM employs the ratio of the task vector to pre-trained model parameters as a metric to quantify CF, integrating this measure into the pruning criteria. Importantly, FAPM does not necessitate modifications to the training process or model architecture, nor does it require any auxiliary data. We conducted extensive experiments across eight datasets, covering natural language inference, General Q&A, Medical Q&A, Math Q&A, reading comprehension, and cloze tests. The results demonstrate that FAPM limits CF to just 0.25% while maintaining 99.67% accuracy on downstream tasks. We provide the codes of FAPM at an anonymous repository(https://anonymous.4open.science/r/FAPM-65CF).

Mitigating Catastrophic Forgetting in Large Language Models with Forgetting-aware Pruning

Unsupervised hallucination detection aims to identify hallucinated content generated by large language models (LLMs) without relying on labeled data. While unsupervised methods have gained popularity by eliminating labor-intensive human annotations, they frequently rely on proxy signals unrelated to factual correctness. This misalignment biases detection probes toward superficial or non-truth-related aspects, limiting generalizability across datasets and scenarios. To overcome these limitations, we propose IRIS, an unsupervised hallucination detection framework, leveraging internal representations intrinsic to factual correctness. IRIS prompts the LLM to carefully verify the truthfulness of a given statement, and obtain its contextualized embedding as informative features for training. Meanwhile, the uncertainty of each response is considered a soft pseudolabel for truthfulness. Experimental results demonstrate that IRIS consistently outperforms existing unsupervised methods. Our approach is fully unsupervised, computationally low cost, and works well even with few training data, making it suitable for real-time detection.

Unsupervised Hallucination Detection by Inspecting Reasoning Processes

Grammatical Error Correction (GEC) for low-resource Indic languages faces significant challenges due to the scarcity of annotated data. In this work, we introduce the Mask-Translate\&Fill (MTF) framework, a novel approach for generating high-quality synthetic data for GEC using only monolingual corpora. MTF leverages a machine translation system and a pretrained masked language model to introduce synthetic errors and tries to mimic errors made by second-language learners. Our experimental results—on English, Hindi, Bengali, Marathi, and Tamil—demonstrate that MTF consistently outperforms other monolingual synthetic data generation methods and achieves performance comparable to the Translation Language Modeling (TLM)-based approach, which uses a bilingual corpus, in both independent and multilingual settings. Under multilingual training, MTF yields significant improvements across Indic languages, with particularly notable gains in Bengali and Tamil—achieving +1.6 and +3.14 GLEU over the TLM-based method, respectively. To support further research, we also introduce the IndiGEC Corpus, a high-quality, human-written, manually validated GEC dataset for these four Indic languages, comprising over 8,000 sentence pairs with separate development and test splits.

IndiGEC: Multilingual Grammar Error Correction for Low-Resource Indian Languages

Debiasing techniques such as SentDebias aim to reduce bias in large language models (LLMs). Previous studies have evaluated their cross-lingual transferability by directly applying these methods to LLM representations, revealing their limited effectiveness across languages. In this work, we therefore propose to perform debiasing in a joint latent space rather than directly on LLM representations. We construct a well-aligned cross-lingual latent space using an autoencoder trained on parallel TED talk scripts. Our experiments with Aya-expanse and two debiasing techniques across four languages (English, French, German, Dutch) demonstrate that a) autoencoders effectively construct a well-aligned cross-lingual latent space, and b) applying debiasing techniques in the learned cross-lingual latent space significantly improves both the overall debiasing performance and cross-lingual transferability.

Debiasing Multilingual LLMs in Cross-lingual Latent Space

Despite bilingual speakers frequently using mixed-language queries in web searches, Information Retrieval (IR) research on them remains scarce. To address this, we introduce ***MiLQ***, ***Mi***xed-***L***anguage ***Q***uery test set, the first public benchmark of mixed-language queries, confirmed as realistic and highly preferred. Experiments on MiLQ show that multilingual IR models yield moderate performance. Futhermore, intentional English mixing in query proves an effective strategy for bilinguals searching English documents. Our analysis offers the rationale that mixed English terms enable better direct matching than native queries.

MiLQ: Benchmarking IR Models for Bilingual Web Search with Mixed Language Queries

Cross-lingual context retrieval (extracting contextual information in one language based on requests in another) is a fundamental aspect of cross-lingual alignment, but the performance and mechanism of it for large language models (LLMs) remains unclear. In this paper, we evaluate the cross-lingual context retrieval of over 40 LLMs across 12 languages, using cross-lingual machine reading comprehension (xMRC) as a representative scenario. Our results show that post-trained open LLMs show strong cross-lingual context retrieval ability, comparable to closed-source LLMs such as GPT-4o, and their estimated oracle performances greatly improve after post-training. Our mechanism analysis shows that the cross-lingual context retrieval process can be divided into two main phases: question encoding and answer retrieval, which are formed in pre-training and post-training respectively. The phasing stability correlates with xMRC performance, and the xMRC bottleneck lies at the last model layers in the second phase, where the effect of post-training can be evidently observed. Our results also indicate that larger-scale pretraining cannot improve the xMRC performance. Instead, larger LLMs need further multilingual post-training to fully unlock their cross-lingual context retrieval potential.

Understanding LLMs' Cross-Lingual Context Retrieval: How Good It Is And Where It Comes From

In-context learning (ICL) has emerged as a powerful paradigm for task adaptation in large language models (LLMs), where models infer underlying task structures from a few demonstrations. However, ICL remains susceptible to biases that arise from prior knowledge and contextual demonstrations, which can degrade the performance of LLMs. Existing bias calibration methods typically apply fixed class priors across all inputs, limiting their efficacy in dynamic ICL settings where the context for each query differs. To address these limitations, we adopt implicit sequential Bayesian inference as a framework for interpreting ICL, identify “surprise” as an informative signal for class prior shift, and introduce a novel method—Surprise Calibration (SC). SC leverages the notion of surprise to capture the temporal dynamics of class priors, providing a more adaptive and computationally efficient solution for in-context learning. We empirically demonstrate the superiority of SC over existing bias calibration techniques across a range of benchmark natural language processing tasks.

Surprise Calibration for Better In-Context Learning

Developing Large Language Model (LLM) agents that exhibit human-like behavior, encompassing not only individual heterogeneity rooted in unique user profiles but also adaptive response to socially connected neighbors, is a significant research challenge. Social media platforms, with their diverse user data and explicit social structures, provide an ideal testbed for such investigations. This paper introduces EvoBot, an **Evo**lving LLM-based social **Bot** that significantly enhances human-like generative capabilities through a novel adversarial learning framework. EvoBot is initialized by Supervised Fine-Tuning (SFT) on representative data from social media and then iteratively refines its generation of sophisticated, human-like content via Direct Preference Optimization (DPO). This refinement is guided by feedback from a co-adapting **Detector** which concurrently improves its ability to distinguish EvoBot from humans, thereby creating an increasingly challenging learning environment for EvoBot. Experiments demonstrate that EvoBot generates content aligned with diverse user profiles, increasingly bypassing the co-adapting Detector through human-like expression. Moreover, it exhibits strong social responsiveness, more accurately modeling real-world opinion dynamics and information spread in multi-agent simulations. The framework also yields a more robust Detector, underscoring its broader utility for both advanced agent development and related detection tasks. The code is available at https://anonymous.4open.science/r/EvoBot-036D.

Enhancing LLM-Based Social Bot via an Adversarial Learning Framework

Code-switching (CSW) is commonly observed among multilingual speakers, and is motivated by various paralinguistic, syntactic, and morphological aspects of conversation. We build on prior work by asking: how do discourse-level aspects of dialogue -- i.e. the content and function of speech -- influence patterns of CSW? To answer this, we analyze the named entities and dialogue acts present in a Spanish-English spontaneous speech corpus, and build a predictive model of CSW based on our statistical findings. We show that discourse content and function interact with patterns of CSW to varying degrees, with a stronger influence from function overall. Our work is the first to take a discourse-sensitive approach to understanding the pragmatic and referential cues of multilingual speech and has potential applications in improving the prediction, recognition, and synthesis of code-switched speech that is grounded in authentic aspects of multilingual discourse.

Discourse-Driven Code-Switching: Analyzing the Role of Content and Communicative Function in Spanish-English Bilingual Speech

Spoken-only languages are languages without a writing system. They remain excluded from modern Natural Language Processing (NLP) advancements like Large Language Models (LLMs) due to their lack of textual data. Existing NLP research focuses primarily on high-resource or written low-resource languages, leaving spoken-only languages critically underexplored. As a popular NLP paradigm, LLMs have demonstrated strong few-shot and cross-lingual generalization abilities, making them a promising solution for understanding and translating spoken-only languages. In this paper, we investigate how LLMs can translate spoken-only languages into high-resource languages by leveraging international phonetic transcription as an intermediate representation. We propose UNILANG, a unified language understanding framework that learns to translate spoken-only languages via in-context learning. Through automatic dictionary construction and knowledge retrieval, UNILANG equips LLMs with more fine-grained knowledge for improving word-level semantic alignment. To support this study, we introduce the SOLAN dataset, which consists of Bai (a spoken-only language) and its corresponding translations in a high-resource language. A series of experiments demonstrates the effectiveness of UNILANG in translating spoken-only languages, potentially contributing to the preservation of linguistic and cultural diversity. Our dataset and code will be publicly released.

Downloads

Next from EMNLP 2025

Mitigating Catastrophic Forgetting in Large Language Models with Forgetting-aware Pruning

Similar lecture

VLP: Vision-Language Preference Learning for Embodied Manipulation