China

Multi-stage information retrieval (IR) has become a widely-adopted paradigm in search. While Large Language Models (LLMs) have been extensively evaluated as second-stage reranking models for monolingual IR, a systematic large-scale comparison is still lacking for cross-lingual IR (CLIR). Moreover, while prior work shows that LLM-based rerankers improve CLIR performance, their evaluation setup relies on machine translation (MT) for the first stage. This is not only prohibitively expensive but also prone to error propagation across stages. Our evaluation on passage-level and document-level CLIR reveals that this setup, which we term \textit{noisy monolingual IR}, is favorable for LLMs. However, LLMs still fail to improve the first-stage ranking if instead produced by multilingual bi-encoders. We further show that pairwise rerankers based on instruction-tuned LLMs perform competitively with listwise rerankers. To the best of our knowledge, we are the first to study the interaction between retrievers and rerankers in two-stage CLIR with LLMs. Our findings reveal that, without MT, current state-of-the-art rerankers fall severely short when directly applied in CLIR.

EMNLP 2025

Evaluating Large Language Models for Cross-Lingual Retrieval

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.

Limited low-resource language corpora in professional domains like medicine hinder cross-lingual domain adaptation of pre-trained large language models (PLMs). While abundant English medical corpora could complement this scarcity, the effective mixture of English and target language, including machine-translated content, remains underexplored. We examined how linguistic features (e.g., token sizes and language proportions) affect performance on a Japanese–English medical knowledge benchmark. Through continued pre-training of a bilingual PLM on multilingual corpora with varying proportions of English and Japanese texts (both original and machine-translated), we analyzed correlations between linguistic features and fine-grained task performance. Our findings suggest a practical approach to optimizing multilingual corpora for cross-lingual domain adaptation, which requires leveraging specialized knowledge from English corpora while ensuring sufficient coverage of language-specific expressions in a target language (Japanese). Such insights will contribute to the development of multilingual models that effectively leverage English-language resources in various professional domains with low-resource languages.

Leveraging High-Resource English Corpora for Cross-lingual Domain Adaptation in Low-Resource Japanese Medicine via Continued Pre-training

As domain-specific datasets continue to expand, Large Language Models (LLMs) have achieved significant improvements across various fields through supervised fine-tuning (SFT). However, is more data always better for model fine-tuning? Through a series of controlled experiments, we discover that dataset structure—rather than mere size—plays a decisive role in enhancing LLM reasoning capabilities. While existing methods acknowledge that good data quality can make training more efficient, they primarily rely on simple heuristic strategies and lack systematic, quantitative frameworks for evaluating data quality. To address this gap, we introduce MCSQ—the first multi-dimensional quantitative framework for reasoning data management. MCSQ rigorously evaluates and optimizes datasets along six orthogonal dimensions. Through comprehensive controlled experiments, we find that selectively incorporating "distorted" (model-disagreed) or "mismatched" (low-relevance) samples—which are typically discarded in traditional approaches—can outperform conventional "clean" data on certain advanced reasoning benchmarks. Our findings challenge traditional assumptions about data "quality" in LLM fine-tuning and provide actionable, quantitative guidance for efficient, structure-aware dataset management.

Structure Trumps Size: Rethinking Data Quality for LLM Reasoning

Fine-tuning large language models (LLMs) typically relies on producing large sets of input-output pairs. Yet for a given question, there can be many valid outputs. In practice, these outputs are often derived by distilling knowledge from teacher models, and they can vary depending on the specific teacher model or prompting strategy employed. Recent findings show that how these training outputs are generated can significantly affect the performance of the fine-tuned model, raising an important question: how do we pick the best data generation method from among numerous possibilities? Rather than exhaustively training and evaluating on each candidate, this paper proposes a scalable approximate method that assesses a small subset of generated data to estimate its suitability for a specific target LLM. Our central idea is that effective outputs should be familiar to the target LLM. While previous work measures familiarity with perplexity, we find that perplexity might be suboptimal in characterizing ``familiarity'' through empirical analyses and practical observations. To address this, we introduce self-aligned perplexity, a novel metric capturing how closely candidate outputs adhere to the target LLM’s own style and reasoning patterns. In this way, we can identify the most effective generation strategy on a small sample, then apply it to produce the complete training set. We demonstrate that training on data generated by the chosen method yields significant improvements across diverse reasoning-focused benchmarks, particularly in cases where different candidate methods lead to highly divergent training outcomes.

Efficiently Selecting Response Generation Strategies for Synthetic Data Construction by Self-Aligned Perplexity

Semantic frame induction is the task of clustering frame-evoking words according to the semantic frames they evoke. In recent years, leveraging embeddings of frame-evoking words that are obtained using masked language models (MLMs) such as BERT has led to high-performance semantic frame induction. Although causal language models (CLMs) such as the GPT and Llama series succeed in a wide range of language comprehension tasks and can engage in dialogue as if they understood frames, they have not yet been applied to semantic frame induction. We propose a new method for semantic frame induction based on CLMs. Specifically, we introduce FrameEOL, a prompt-based method for obtaining Frame Embeddings that outputs One frame-name as a Label representing the given situation. To obtain embeddings more suitable for frame induction, we leverage in-context learning (ICL) and deep metric learning (DML). Frame induction is then performed by clustering the resulting embeddings. Experimental results on the English and Japanese FrameNet datasets demonstrate that the proposed methods outperform existing frame induction methods. In particular, for Japanese, which lacks extensive frame resources, the CLM-based method using only 5 ICL examples outperformed the MLM-based method fine-tuned with DML.

FrameEOL: Semantic Frame Induction using Causal Language Models

Recently, Vision-Language-Action (VLA) models have demonstrated strong performance on a range of robotic tasks. These models rely on multimodal inputs, with language instructions playing a crucial role-not only in predicting actions, but also in robustly interpreting user intent, even when the requests are impossible to fulfill. In this work, we investigate how VLAs can recognize, interpret, and respond to false-premise instructions-natural language commands that reference objects or conditions absent from the environment. We propose --- Instruct-Verify-and-Act (IVA) --- a unified framework that (i) detects when an instruction cannot be executed due to a false premise, (ii) engages in language-based clarification or correction, and (iii) grounds plausible alternatives in perception and action. Towards this end, we construct a large-scale instruction tuning setup with structured language prompts and train a VLA model capable of handling both accurate and erroneous requests. Our approach leverages a contextually augmented, semi-synthetic dataset containing paired positive and false-premise instructions, enabling robust detection and natural language correction. Our experiments show that IVA can improves false premise detection accuracy by 58.89% over baselines, while increasing successful responses in false-premise scenarios by 27.89%.

Do What? Teaching Vision-Language-Action Models to Reject the Impossible

Sentence embeddings play an important role in tasks such as clustering, semantic search, and retrieval-augmented generation, yet they generally lack interpretability. We propose a framework for decomposing embeddings into interpretable components, which we define as semantic regions, i.e. connected subsets on the embedding hypersphere. These regions reveal both the internal semantic structure of individual embeddings and the set-theoretical relationships between them. We further show that these regions exhibit a hierarchical organization that reflects semantic inclusion, including hypernymy and hyponymy relations. Empirical results across both synthetic and real-world datasets validate the existence of these regions and demonstrate their utility for sentence embedding interpretability.

Semantic Geometry of Sentence Embeddings

Deploying language models on resource-constrained devices, such as mobile phones, wearables, and on-device AI assistants, demands compact, efficient models without sacrificing performance. Compressing Small Language Models (SLMs) is particularly suited for these scenarios, yet their compression dynamics remain underexplored compared to Large Language Models (LLMs). We systematically evaluate leading post-training pruning (SparseGPT, Wanda) and quantization (GPTQ, AWQ) methods across five SLMs, seven languages, and seven downstream tasks (710 evaluations in total). Our results show that quantization consistently outperforms pruning in preserving model fidelity, multilingual perplexity, and reasoning accuracy. Notably, trends observed in LLMs (e.g., Wanda’s superiority over SparseGPT) do not generalize to SLMs. For practitioners, we recommend prioritizing quantization (particularly AWQ) for SLM compression and caution against relying on single metrics is critical.

Revisiting Pruning vs Quantization for Small Language Models

Drawing real world social inferences usually requires taking into account information from multiple modalities. Language is a particularly powerful source of information in social settings, especially in novel situations where language can provide both abstract information about the environment dynamics and concrete specifics about an agent that cannot be easily visually observed. In this paper, we propose Language-Informed Rational Agent Synthesis (LIRAS), a framework for drawing context-specific social inferences that integrate linguistic and visual inputs. LIRAS frames multimodal social reasoning as a process of constructing structured but situation-specific agent and environment representations – leveraging multimodal language models to parse language and visual inputs into unified symbolic representations, over which a Bayesian inverse planning engine can be run to produce granular probabilistic judgments. On a range of existing and new social reasoning tasks derived from cognitive science experiments, we find that our model (instantiated with a comparatively lightweight VLM) outperforms ablations and state-of-the-art models in capturing human judgments across all domains.

Language-Informed Synthesis of Rational Agent Models for Grounded Theory-of-Mind Reasoning On-the-fly

A key challenge in Multi-Document Summarization (MDS) is effectively integrating information from multiple sources while maintaining coherence and topical relevance. While Large Language Models (LLMs) have shown impressive results in single-document summarization, their performance on MDS still leaves room for improvement. In this paper, we propose a topic-guided reinforcement learning approach to improve content selection in MDS. We first show that explicitly prompting models with topic labels enhances the informativeness. Building on this insight, we propose a novel topic reward within the Group Relative Policy Optimization (GRPO) framework to measure topic alignment between the generated summary and source documents. Experimental results on the Multi-News and Multi-XScience datasets demonstrate that our method consistently outperforms strong baselines, highlighting the effectiveness of leveraging topical cues in MDS.

Topic-Guided Reinforcement Learning with LLMs for Enhancing Multi-Document Summarization

Large vision-language models (LVLMs) have demonstrated exceptional capabilities in understanding visual information with human languages but also exhibit an imbalance in multilingual capabilities. In this work, we delve into the multilingual working pattern of LVLMs and identify a salient correlation between the multilingual understanding ability of LVLMs and language-specific neuron activations in shallow layers. Building on this insight, we introduce PLAST, a training recipe that achieves efficient multilingual enhancement for LVLMs by Precise LAnguage Specific layers fine-Tuning. PLAST first identifies layers involved in multilingual understanding by monitoring language-specific neuron activations. These layers are then precisely fine-tuned with question-translation pairs to achieve multilingual alignment. Our empirical results on MMBench and MMMB demonstrate that PLAST effectively improves the multilingual capabilities of LVLMs and achieves significant efficiency with only 14% of the parameters tuned. Further analysis reveals that PLAST facilitates the language-specific visual information engagement in shallow layers.

Downloads

Next from EMNLP 2025

Leveraging High-Resource English Corpora for Cross-lingual Domain Adaptation in Low-Resource Japanese Medicine via Continued Pre-training

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES