China

Large language models (LLMs) are frequently fine-tuned or unlearned to adapt to new tasks or eliminate undesirable behaviors. While existing evaluation methods assess performance after such interventions, there remains no general approach for detecting unintended side effects—such as unlearning biology content degrading performance on chemistry tasks, particularly when these effects are unpredictable or emergent. To address this issue, we introduce MNEME, textitModel diffiNg for Evaluating Mechanistic Effects, a lightweight framework for identifying these side effects using sparse model diffing. MNEME compares base and fine-tuned models on task-agnostic data (e.g., The Pile, LMSYS-Chat-1M), without access to fine-tuning data, to isolate behavioral shifts. Applied to five LLMs across three scenarios, WMDP knowledge unlearning, emergent misalignment, and benign fine-tuning—MNEME achieves up to 95\% accuracy in predicting side effects, aligning with known benchmarks and requiring no custom heuristics. Furthermore, we show that retraining on high-activation samples can partially reverse these effects. Our results demonstrate that sparse probing and diffing offer a scalable and automated lens into fine-tuning-induced model changes, providing practical tools for understanding and managing LLM behavior.

EMNLP 2025

REVIVING YOUR MNEME: Predicting The Side Effects of LLM Unlearning and Fine-Tuning via Sparse Model Diffing

side-effects

unlearning

model interpretability

## 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

Most existing Theory of Mind (ToM) benchmarks for foundation models rely on variations of the Sally-Anne test, offering only a very limited perspective on ToM and neglecting the complexity of human social interactions. To address this gap, we propose ToM-SSI: a new benchmark specifically designed to test ToM capabilities in environments rich with social interactions and spatial dynamics. While current ToM benchmarks are limited to text-only or dyadic interactions, ToM-SSI is multimodal and includes group interactions of up to four agents that communicate and move in situated environments. This unique design allows us to study, for the first time, mixed cooperative-obstructive settings and reasoning about multiple agents' mental state in parallel, thus capturing a wider range of social cognition than existing benchmarks. Our evaluations reveal that current models' performance is still severely limited, especially in these new tasks, highlighting critical gaps for future research.

ToM-SSI: Evaluating Theory of Mind in Situated Social Interactions

Despite Large Language Models (LLMs) demonstrating superior translation performance and long-context capabilities, evaluation methodologies remain constrained to sentence-level assessment due to dataset limitations, token number restrictions in metrics, and rigid sentence boundary requirements. We introduce an evaluation scheme that extends existing automatic metrics to long-document translation by treating documents as continuous text and applying sentence segmentation and alignment methods. Our approach enables previously unattainable document-level evaluation, handling translations of arbitrary length generated with document-level prompts while accounting for under-/over-translations and varied sentence boundaries. Experiments show our scheme significantly outperforms existing long-form document evaluation schemes, while comparable to evaluations performed with groundtruth sentence alignments. Additionally, we apply our scheme to book-length texts and newly demonstrate that many open-weight LLMs fail to effectively translate documents at their reported maximum context lengths.

Extending Automatic Machine Translation Evaluation to Book-Length Documents

We present NN-Rank, an algorithm for ranking source languages for cross-lingual transfer, which leverages hidden representations from multilingual models and unlabeled target-language data. We experiment with two pretrained multilingual models and two tasks: part-of-speech tagging (POS) and named entity recognition (NER). We consider 51 source languages and evaluate on 56 and 72 target languages for POS and NER, respectively. When using in-domain data, NN-Rank beats state-of-the-art baselines that leverage lexical and linguistic features, with average improvements of up to 35.56 NDCG for POS and 18.14 NDCG for NER. As prior approaches can fall back to language-level features if target language data is not available, we show that NN-Rank remains competitive using only the Bible, an out-of-domain corpus available for a large number of languages. Ablations on the amount of unlabeled target data show that, for subsets consisting of as few as 25 examples, NN-Rank produces high-quality rankings which achieve 92.8% of the NDCG achieved using all available target data for ranking.

Model-Based Ranking of Source Languages for Zero-Shot Cross-Lingual Transfer

To mitigate the hallucination problem in large language models, DoLa exploits early exit logits from the same model as a contrastive prior. However, we found that these early exit logits tend to be flat, low in magnitude, and fail to reflect meaningful contrasts. To address this, we propose PruneCD, a novel contrastive decoding method that constructs the amateur model via layer pruning rather than early exit. This design leads to more informative and well-aligned logits, enabling more effective contrastive decoding. Through qualitative and quantitative analyses, we demonstrate that PruneCD consistently improves factuality with minimal inference overhead, offering a robust and practical approach to mitigating hallucinations in LLMs.

PruneCD: Contrasting Pruned Self Model to Improve Decoding Factuality

Cognitive Restructuring (CR) uses multi-turn dialogue to identify and restructure one's negative thoughts, arising from mental health issues, into more helpful and positive ones. Clinician shortage and stigma urge the development of human-LLM interactive psychotherapy for CR. Yet, effectively implementing CR is hindered by entrenched cognitive distortions, emotional resistance, and individual differences, which existing works have not overcome. To bridge this gap, we propose CRDial, a novel framework that structures CR as theory-grounded multi-stage multi-turn dialogue, integrating multi-aspect supportive strategies for emotional management and a multi-channel loop mechanism to account for diverse individual distortions. With CRDial, we distill Crisp, a large-scale and high-quality bilingual dialogue dataset, from LLM. We then train Crispers, Crisp-based conversational LLMs for CR, at 7B and 14B scales. Extensive human studies show the superiority of Crispers in pointwise, pairwise, and intervention evaluations.

Crisp: Cognitive Restructuring of Negative Thoughts through Multi-turn Supportive Dialogues

Neural text-based models for detecting machine-translated texts can rely on named entities (NEs) as classification shortcuts. While masking NEs encourages learning genuine translationese signals, it degrades the classification performance. Incorporating speech features compensates for this loss, but their interaction with NE reliance requires careful investigation. Through systematic attribution analysis across modalities, we find that bimodal integration leads to more balanced feature utilization, reducing the reliance on NEs in text while moderating overemphasis attribution patterns in speech features. All the code and data will be publicly released.

From Shortcuts to Balance: Attribution Analysis of Speech-Text Feature Utilization in Distinguishing Original from Machine-Translated Texts

Large language models (LLMs) have shown remarkable performance in general translation tasks. However, the increasing demand for high-quality translations that are not only adequate but also fluent and elegant. To assess the extent to which current LLMs can meet these demands, we introduce a suitable benchmark (PoetMT) for translating classical Chinese poetry into English. This task requires not only adequacy in translating culturally and historically significant content but also a strict adherence to linguistic fluency and poetic elegance. Our study reveals that existing LLMs fall short of this task. To address these issues, we propose RAT, a Retrieval-Augmented machine Translation method that enhances the translation process by incorporating knowledge related to classical poetry. Additionally, we propose an automatic evaluation metric based on GPT-4, which better assesses translation quality in terms of adequacy, fluency, and elegance, overcoming the limitations of traditional metrics. Our dataset and code will be made available. Our dataset and code will be available upon acceptance.

Benchmarking LLMs for Translating Classical Chinese Poetry: Evaluating Adequacy, Fluency, and Elegance

Large language model (LLM) role-playing has gained widespread attention. Authentic character knowledge is crucial for constructing realistic LLM role-playing agents. However, existing works usually overlook the exploration of LLMs' ability to detect characters' known knowledge errors (KKE) and unknown knowledge errors (UKE) while playing roles, which would lead to low-quality automatic construction of character trainable corpus. In this paper, we propose RoleKE-Bench to evaluate LLMs' ability to detect errors in KKE and UKE. The results indicate that even the latest LLMs struggle to detect these two types of errors effectively, especially when it comes to familiar knowledge. We experimented with various reasoning strategies and propose an agent-based reasoning method, Self-Recollection and Self-Doubt (S²RD), to explore further the potential for improving error detection capabilities.

Revealing and Mitigating the Challenge of Detecting Character Knowledge Errors in LLM Role-Playing

Minimum Bayes risk (MBR) decoding is a decision rule of text generation, which selects the hypothesis that maximizes the expected utility, and robustly generates higher-quality texts than maximum a posteriori (MAP) decoding. However, it depends on sample texts drawn from the text generation model; thus, it is hard to choose the hypothesis that reflects the knowledge or information of out-of-domain. To tackle this issue, we propose case-based decision-theoretic (CBDT) decoding, another approach to the expected utility estimation using examples of domain data. CBDT decoding not only generates higher-quality texts than MAP decoding, but also the combination of MBR and CBDT decoding outperformed MBR decoding in seven domain GermanrightarrowEnglish and JapaneseleftrightarrowEnglish translation tasks and image captioning tasks on MSCOCO and nocaps datasets.

Case-Based Decision-Theoretic Decoding with Quality Memories

There is increasing evidence of Human Label Variation (HLV) in Natural Language Inference (NLI), where annotators assign different labels to the same premise-hypothesis pair. However, *within-label variation* — cases where annotators agree on the same label but provide divergent reasoning — poses an additional and mostly overlooked challenge. Several NLI datasets contain highlighted words in the NLI item as explanations, but the same spans on the NLI item can be highlighted for different reasons, as evidenced by free-text explanations, which offer a window into annotators’ reasoning. To systematically understand this problem and gain insight into the rationales behind NLI labels, we introduce LiTEx, a linguistically-informed taxonomy for categorizing free-text explanations. Using this taxonomy, we annotate a subset of the e-SNLI dataset, validate the taxonomy’s reliability, and analyze how it aligns with NLI labels, highlights, and explanations. We further assess the taxonomy’s usefulness in explanation generation, demonstrating that conditioning generation on LiTEx yields explanations that are linguistically closer to human explanations than those generated using only labels or highlights. Our approach thus not only captures within-label variation but also shows how taxonomy-guided generation for reasoning can bridge the gap between human and model explanations more effectively than existing strategies.

Downloads

Next from EMNLP 2025

ToM-SSI: Evaluating Theory of Mind in Situated Social Interactions

Similar lecture

Beyond the Leaderboard: Understanding Performance Disparities in Large Language Models via Model Diffing