Austria

Large language models (LLMs) exhibit pronounced conservative bias in relation extraction tasks, frequently defaulting to no_relation label when an appropriate option is unavailable. While this behavior helps prevent incorrect relation assignments, our analysis reveals that it also leads to significant information loss when reasoning is not explicitly included in the output. We systematically evaluate this trade-off across multiple prompts, datasets, and relation types, introducing the concept of Hobson&#39;s choice to capture scenarios where models opt for safe but uninformative labels over hallucinated ones. Our findings suggest that conservative bias occurs twice as often as hallucination. To quantify this effect, we use SBERT and LLM prompts to capture the semantic similarity between conservative bias behaviors in constrained prompts and labels generated from semi-constrained and open-ended prompts.

ACL 2025

Conservative Bias in Large Language Models: Measuring Relation Predictions

conservative bias

large language models (llms)

relation extraction

## 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." 

*Message from the General Chair, Dirk Hovy*

EMNLP 2025

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

Large language models (LLMs) exhibit pronounced conservative bias in relation extraction tasks, frequently defaulting to no_relation label when an appropriate option is unavailable. While this behavior helps prevent incorrect relation assignments, our analysis reveals that it also leads to significant information loss when reasoning is not explicitly included in the output. We systematically evaluate this trade-off across multiple prompts, datasets, and relation types, introducing the concept of Hobson's choice to capture scenarios where models opt for safe but uninformative labels over hallucinated ones. Our findings suggest that conservative bias occurs twice as often as hallucination. To quantify this effect, we use SBERT and LLM prompts to capture the semantic similarity between conservative bias behaviors in constrained prompts and labels generated from semi-constrained and open-ended prompts.

poster

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair: 
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025. ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.* 
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2025". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you need to log in with the **email address you registered with**. 

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Large-scale multilingual evaluations, such as MEGA, often include only a handful of African languages due to the scarcity of high-quality evaluation data and the limited discoverability of existing African datasets. This lack of representation hinders comprehensive LLM evaluation across a diverse range of languages and tasks. To address these challenges, we introduce AfroBench - a multi-task benchmark for evaluating the performance of LLMs across 64 African languages, 15 tasks and 22 datasets. AfroBench consists of nine natural language understanding datasets, five text generation datasets, five knowledge and question answering tasks, and one mathematical reasoning task. We present results comparing the performance of prompting LLMs to fine-tuned baselines based on BERT and T5-style models. Our results suggest large gaps in performance between high-resource languages, such as English, and African languages across most tasks; but performance also varies based on the availability of monolingual data resources. Our findings confirm that performance on African languages continues to remain a hurdle for current LLMs, underscoring the need for additional efforts to close this gap

AfroBench: How Good are Large Language Models on African Languages?

Large language models are trained on massive scrapes of the web, as required by current scaling laws. Most progress is made for English, given its abundance of high-quality pretraining data. For most other languages, however, such high quality pretraining data is unavailable. In this work, we study how to boost pretrained model performance in a target language with insufficient pretraining data for training a high performing language model by enlisting data from an auxiliary language for which high quality data is available. We study this by quantifying the performance gap between training with data in a data-rich auxiliary language compared with training in the target language, exploring the benefits of translation systems, studying the limitations of model scaling when data is limited in the target languages, and proposing new methods for upsampling data from the auxiliary language. Our results show that stronger auxiliary datasets result in performance gains without modification to the model or training objective for close languages, and, in particular, that performance gains due to the development of more information-rich English pretraining datasets can extend to targeted language settings with limited data.

Training Bilingual LMs with Data Constraints in the Targeted Language

Query expansion methods powered by large language models (LLMs) have demonstrated effectiveness in zero-shot retrieval tasks. These methods assume that LLMs can generate hypothetical documents that, when incorporated into a query vector, enhance the retrieval of real evidence. However, we challenge this assumption by investigating whether knowledge leakage in benchmarks contributes to the observed performance gains. Using fact verification as a testbed, we analyzed whether the generated documents contained information entailed by ground truth evidence and assessed their impact on performance. Our findings indicate that performance improvements occurred consistently only for claims whose generated documents included sentences entailed by ground truth evidence. This suggests that knowledge leakage may be present in these benchmarks, potentially inflating the perceived performance of query expansion methods, particularly in real-world scenarios that require retrieving niche or novel knowledge.

Hypothetical Documents or Knowledge Leakage? Rethinking LLM-based Query Expansion

This study investigates the linguistic understanding of Large Language Models (LLMs) regarding signifier (form) and signified (meaning) by distinguishing two LLM assessment paradigms: psycholinguistic and neurolinguistic. Traditional psycholinguistic evaluations often reflect statistical rules that may not accurately represent LLMs' true linguistic competence. We introduce a neurolinguistic approach, utilizing a novel method that combines minimal pair and diagnostic probing to analyze activation patterns across model layers. This method allows for a detailed examination of how LLMs represent form and meaning, and whether these representations are consistent across languages. We found: (1) Psycholinguistic and neurolinguistic methods reveal that language performance and competence are distinct; (2) Direct probability measurement may not accurately assess linguistic competence; (3) Instruction tuning won't change much competence but improve performance; (4) LLMs exhibit higher competence and performance in form compared to meaning. Additionally, we introduce new conceptual minimal pair datasets for Chinese (COMPS-ZH) and German (COMPS-DE), complementing existing English datasets.

Large Language Models as Neurolinguistic Subjects: Discrepancy in Performance and Competence for Form and Meaning

Knowing which latent conditions lead to a particular outcome is useful for critically examining claims made about complex event outcomes. Identifying implied conditions and examining their influence on an outcome is challenging. We handle this by combining and augmenting annotations from two existing datasets consisting of goals and states, and explore the influence of conditions through our research questions and Condition-based Reasoning tasks. We examine open and closed LLMs of varying sizes and intent-alignment on our reasoning tasks and find that conditions are useful when not all context is available. Models differ widely in their ability to generate and identify outcome-variant conditions, which affects their performance on outcome validation, when conditions are used to replace missing context. Larger models like GPT-4o, are more cautious in such less constrained situations.

CoRE: Condition-based Reasoning for Identifying Outcome Variance in Complex Events

Homonymy can easily raise lexical ambiguity due to the misunderstanding of its multiple senses. Correct recognition of homonym sense greatly relies on its surrounding context. This ambiguous nature makes homonyms an appropriate testbed for examining the contextualization capability of pre-trained (PLM) and large language models (LLMs). Considering the impact of part-of-speech (POS) on homonym disambiguation and the dominance of English in word embedding research, this study provides a comprehensive layer-wise analysis of homonym representations in both English and Chinese, spanning same and different POS categories, across four families of PLMs/LLMs (BERT, GPT-2, Llama 3, Qwen 2.5). Through the creation of a synthetic dataset and computation of disambiguation score (D-Score), we found that: (1) no universal layer depth was found which excels in differentiating homonym representations; (2) bidirectional models produce better contextualized homonym representations compared to much larger autoregressive models; (3) most importantly, POS affects homonym representations in models in ways that slightly differ from human research findings. The individual differences between LLMs uncovered in our study challenge the simplistic understanding of their inner workings. This reveals a compelling research frontier: conducting controlled experiments with purposefully manipulated inputs to enhance the interpretability of LLMs. We have made our dataset and codes available publicly at https://anonymous.4open.science/r/ehril/.

Exploring Layer-wise Representations of English and Chinese Homonymy in Pre-trained Language Models

Large language models can explain grammatical rules, yet they often fail to apply those rules when judging sentence acceptability. We present grammar prompting, an explain‑then‑process paradigm: the model first produces a concise explanation of the relevant syntactic phenomenon, then receives that explanation back as context before deciding which sentence of a minimal pair is grammatical. On the English BLiMP, Chinese SLING, and Russian RuBLiMP benchmarks, this simple prompt design yields substantial improvements over strong baselines across a wide range of syntactic phenomena. By feeding the model its own metalinguistic explanations, grammar prompting narrows the gap between knowing a rule and using it, offering a lightweight, language‑agnostic tool for better syntactic reasoning in multilingual settings.

Explain-then-Process: Using Grammar Prompting to Enhance Grammatical Acceptability Judgments

Word alignment has many practical applications in natural language processing, such as serving as cross-lingual signals for multilingual sentence embedding and reducing issues of hallucination and omission in machine translation. Current state-of-the-art approaches usually rely on: (1) supervised data and large-scale weakly supervised data constructed from Wikipedia and (2) multilingual Transformer encoder models. However, we find that the current existing state-of-the-art encoder-based method BinaryAlign can be further improved with self-training with a small amount of parallel data. Meanwhile, considering the impressive performance of multilingual large language models on many natural language processing tasks, in this paper, we also explore the possibility of using these decoder-based large language models as word aligners. We find that the performance of fine-tuning large language models with only labeled data is not bad but can be further enhanced by pseudo-labeled data. Based on the findings, we propose a semi-supervised framework to improve the large language model-based word aligners. Experimental results demonstrate that the proposed method with a small amount of parallel data outperforms the current state-of-the-art method on four word alignment datasets.

Improving Word Alignment Using Semi-Supervised Learning

Understanding the structure of multi-party conversation and the intentions and dialogue acts of each speaker remains a significant challenge in NLP. While a number of corpora annotated using theoretical frameworks of dialogue have been proposed, these typically focus on either utterance-level labeling of speaker intent, missing wider context, or the rhetorical structure of a dialogue, losing fine-grained intents captured in dialogue acts. Recently, the Dependency Dialogue Acts (DDA) framework has been proposed to for modeling both the fine-grained intents of each speaker and the structure of multi-party dialogues. However, there is not yet a corpus annotated with this framework available for the community to study. To address this gap, we introduce a new corpus of 33 dialogues and over 9,000 utterance units, densely annotated using the Dependency Dialogue Acts (DDA) framework.Our dataset spans four genres of multi-party conversations from different modalities: (1) physics classroom discussions, (2) engineering classroom discussions, (3) board game interactions, and (4) written online game chat logs. Each session is doubly annotated and adjudicated to ensure high-quality labeling. We present a description of the dataset and annotation process, an analysis of speaker dynamics enabled by our annotation, and a baseline evaluation of LLMs as DDA parsers. We discuss the implications of this dataset understanding dynamics between speakers and for developing more controllable dialogue agents.

In Search of the Lost Arch in Dialogue: A Dependency Dialogue Acts Corpus for Multi-Party Dialogues

Recent advancements in large language models (LLMs) have spurred interest in expanding their application beyond text-based tasks. A large number of studies have explored integrating other modalities with LLMs, notably speech modality, which is naturally related to text. This paper surveys the integration of speech with LLMs, categorizing the methodologies into three primary approaches: text-based, latent-representation-based, and audio-token-based integration. We also demonstrate how these methods are applied across various speech-related applications and highlight the challenges in this field to offer inspiration for future research.

Downloads

Next from ACL 2025

AfroBench: How Good are Large Language Models on African Languages?

Similar lecture

Social Bias Evaluation for Large Language Models Requires Prompt Variations