Austria

Extracting sentence embeddings from large language models (LLMs) is a promising direction, as LLMs have demonstrated stronger semantic understanding capabilities. 
Previous studies typically focus on prompt engineering to elicit sentence embeddings from LLMs by prompting the model to encode sentence information into the embedding of the last token.
However, LLMs are mostly decoder-only models with causal attention and the earlier tokens in the sentence cannot attend to the latter tokens, resulting in biased encoding of sentence information and cascading effects on the final decoded token.
To this end, we propose a novel Token Prepending (TP) technique that prepends each layer&#39;s decoded sentence embedding to the beginning of the sentence in the next layer&#39;s input, allowing earlier tokens to attend to the complete sentence information under the causal attention mechanism.
The proposed TP technique is a plug-and-play and training-free technique, which means it can be seamlessly integrated with various prompt-based sentence embedding methods and autoregressive LLMs.
Extensive experiments on various Semantic Textual Similarity (STS) tasks and downstream classification tasks demonstrate that our proposed TP technique can significantly improve the performance of existing prompt-based sentence embedding methods across different LLMs, while incurring negligible additional inference cost.

ACL 2025

Token Prepending: A Training-Free Approach for Eliciting Better Sentence Embeddings from LLMs

training-free

sentence embedding

large language models

### 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)

Extracting sentence embeddings from large language models (LLMs) is a promising direction, as LLMs have demonstrated stronger semantic understanding capabilities. 
Previous studies typically focus on prompt engineering to elicit sentence embeddings from LLMs by prompting the model to encode sentence information into the embedding of the last token.
However, LLMs are mostly decoder-only models with causal attention and the earlier tokens in the sentence cannot attend to the latter tokens, resulting in biased encoding of sentence information and cascading effects on the final decoded token.
To this end, we propose a novel Token Prepending (TP) technique that prepends each layer's decoded sentence embedding to the beginning of the sentence in the next layer's input, allowing earlier tokens to attend to the complete sentence information under the causal attention mechanism.
The proposed TP technique is a plug-and-play and training-free technique, which means it can be seamlessly integrated with various prompt-based sentence embedding methods and autoregressive LLMs.
Extensive experiments on various Semantic Textual Similarity (STS) tasks and downstream classification tasks demonstrate that our proposed TP technique can significantly improve the performance of existing prompt-based sentence embedding methods across different LLMs, while incurring negligible additional inference cost.

technical paper

To address the global challenge of online hate speech, prior research has developed detection models to flag such content on social media. However, due to systematic biases in evaluation datasets, the real-world effectiveness of these models remains unclear, particularly across geographies. We introduce HateDay, the first global hate speech dataset representative of social media settings, constructed from a random sample of all tweets posted on September 21, 2022 and covering eight languages and four English-speaking countries. Using HateDay, we uncover substantial variation in the prevalence and composition of hate speech across languages and regions. We show that evaluations on academic datasets greatly overestimate real-world detection performance, which we find is very low, especially for non-European languages. Our analysis identifies key drivers of this gap, including models’ difficulty to distinguish hate from offensive speech and a mismatch between the target groups emphasized in academic datasets and those most frequently targeted in real-world settings. We argue that poor model performance makes public models ill-suited for automatic hate speech moderation and find that high moderation rates are only achievable with substantial human oversight. Our results underscore the need to evaluate detection systems on data that reflects the complexity and diversity of real-world social media.

HateDay: Insights from a Global Hate Speech Dataset Representative of a Day on Twitter

In recent years, large language models (LLMs) have made remarkable advancements, yet hallucination, where models produce inaccurate or non-factual statements, remains a significant challenge for real-world deployment. Although current classification-based methods, such as SAPLMA, are highly efficient in mitigating hallucinations, they struggle when non-factual information arises in the early or mid-sequence of outputs, reducing their reliability. To address these issues, we propose Hallucination Detection-Neural Differential Equations (HD-NDEs), a novel method that systematically assesses the truthfulness of statements by capturing the full dynamics of LLMs within their latent space. Our approaches apply neural differential equations (Neural DEs) to model the dynamic system in the latent space of LLMs. Then, the sequence in the latent space is mapped to the classification space for truth assessment. The extensive experiments across five datasets and six widely used LLMs demonstrate the effectiveness of HD-NDEs, especially, achieving over 14% improvement in AUC-ROC on the True-False dataset compared to state-of-the-art techniques.

HD-NDEs: Neural Differential Equations for Hallucination Detection in LLMs

Binaural audio enriches immersive experiences by enabling the perception of the spatial locations of sounding objects in AR, VR, and embodied AI applications. While existing audio spatialization methods can generally map any available monaural audio to binaural audio signals, they often lack the flexible and interactive control needed in complex multi-object user-interactive environments. To address this, we propose a Text-guided Audio Spatialization (TAS) framework that utilizes diverse text prompts and evaluates our model from unified generation and comprehension perspectives. Due to the limited availability of high-quality, large-scale stereo data, we construct the SpatialTAS dataset, which encompasses 376,000 simulated binaural audio samples to facilitate the training of our model. Our model learns binaural differences guided by 3D spatial location and relative position prompts, enhanced with flipped-channel audio. Experimental results show that our model can generate high quality binaural audios for various audio types on both simulated and real-recorded datasets. Besides, we establish an assessment model based on Llama-3.1-8B, which evaluates the semantic accuracy of spatial locations through a spatial reasoning task. Results demonstrate that by utilizing text prompts for flexible and interactive control, we can generate binaural audio with both high quality and semantic consistency in spatial locations.

In-the-wild Audio Spatialization with Flexible Text-guided Localization

Real-world RAG applications often encounter long-context input scenarios, where redundant information and noise results in higher inference costs and reduced performance. To address these challenges, we propose LongRefiner, an efficient plug-and-play refiner that leverages the inherent structural characteristics of long documents. LongRefiner employs dual-level query analysis, hierarchical document structuring, and adaptive refinement through multi-task learning on a single foundation model. Experiments on seven QA datasets demonstrate that LongRefiner achieves competitive performance in various scenarios while using 10x fewer computational costs and latency compared to the best baseline. Further analysis validates that LongRefiner is scalable, efficient, and effective, providing practical insights for real-world long-text RAG applications. Our code is available at https://github.com/ignorejjj/LongRefiner.

Hierarchical Document Refinement for Long-context Retrieval-augmented Generation

Large Language Models (LLMs) are prone to generating content that exhibits gender biases, raising significant ethical concerns. Alignment, the process of fine-tuning LLMs to better align with desired behaviors, is recognized as an effective approach to mitigate gender biases. Although proprietary LLMs have made significant strides in mitigating gender bias, their alignment datasets are not publicly available. The commonly used and publicly available alignment dataset, HH-RLHF, still exhibits gender bias to some extent. There is a lack of publicly available alignment datasets specifically designed to address gender bias. Hence, we developed a new dataset named GenderAlign, aiming at mitigating a comprehensive set of gender biases in LLMs. This dataset comprises 8k single-turn dialogues, each paired with a "chosen" and a "rejected" response. Compared to the "rejected" responses, the "chosen" responses demonstrate lower levels of gender bias and higher quality. Furthermore, we categorized the gender biases in the "rejected" responses of GenderAlign into 4 principal categories. The experimental results show the effectiveness of GenderAlign in reducing gender bias in LLMs.

GenderAlign: An Alignment Dataset for Mitigating Gender Bias in Large Language Models

Despite the rapidly growing demand for multimodal retrieval, progress in this field remains severely constrained by a lack of training data. In this paper, we introduce MegaPairs, a novel data synthesis method that leverages vision language models (VLMs) and open-domain images, together with a massive synthetic dataset generated from this method. Our empirical analysis shows that MegaPairs generates high-quality data, enabling the multimodal retriever to significantly outperform the baseline model trained on 70$\times$ more data from existing datasets. Moreover, since MegaPairs solely relies on general image corpora and open-source VLMs, it can be easily scaled up, enabling continuous improvements in retrieval performance. In this stage, we produced more than 26 million training instances and trained several models of varying sizes using this data. These new models achieve state-of-the-art zero-shot performance across 4 popular composed image retrieval (CIR) benchmarks and the highest overall performance on the 36 datasets provided by MMEB. They also demonstrate notable performance improvements with additional downstream fine-tuning. Our code, synthesized dataset, and pre-trained models are publicly available at https://github.com/VectorSpaceLab/MegaPairs.

MegaPairs: Massive Data Synthesis for Universal Multimodal Retrieval

Digitized literary corpora of the 19th century favor canonical and novelistic forms, sidelining a broader and more diverse literary production. Serialized fiction – widely read but embedded in newspapers – remains especially underexplored, particularly in low-resource languages like Danish. 
This paper addresses this gap by developing methods to identify fiction in digitized Danish newspapers (1818–1848).
We (1) introduce a manually annotated dataset of 1,394 articles and (2) evaluate classification pipelines using both selected linguistic features and embeddings, achieving F1-scores of up to 0.91. Finally, we (3) analyze feuilleton fiction via interpretable features to test its drift in discourse from neighboring nonfiction.
Our results support the construction of alternative literary corpora and contribute to ongoing work on modeling the fiction–nonfiction boundary by operationalizing discourse-level distinctions at scale.

Fact from Fiction: Finding Serialized Novels in Newspapers

LoRA is a technique that reduces the number of trainable parameters in a neural network by introducing low-rank adapters to linear layers. This technique is used for fine-tuning and even training large transformer models from scratch. This paper presents the RunLoRA framework for efficient implementations of LoRA, which significantly improves the speed of neural network training and fine-tuning with low-rank adapters. The proposed implementation optimizes the computation of LoRA operations based on the shape of the corresponding linear layer weights, the input dimensions, and the LoRA rank by selecting the best forward and backward computation graphs based on FLOPs and time estimations. This results in faster training without sacrificing accuracy. The experimental results show a speedup ranging from 10% to 28% on various transformer models.

Run LoRA Run: Faster and Lighter LoRA Implementations

Human processing of idioms heavily depends on interpreting the surrounding context in which they appear. While large language models (LLMs) have achieved impressive performance on idiomaticity detection benchmarks, this success may be driven by reasoning shortcuts present in existing datasets. To address this, we introduce a novel, controlled contrastive dataset (DICE) specifically designed to assess whether LLMs can effectively leverage context to disambiguate idiomatic meanings. Furthermore, we investigate the influence of collocational frequency and sentence probability—proxies for human processing known to affect idiom resolution—on model performance. Our results show that LLMs frequently fail to resolve idiomaticity when it depends on contextual understanding, performing better on sentences deemed more likely by the model. Additionally, idiom frequency influences performance but does not guarantee accurate interpretation. Our findings emphasize the limitations of current models in grasping contextual meaning and highlight the need for more context-sensitive evaluation.

Rolling the DICE on Idiomaticity: How LLMs Fail to Grasp Context

This study investigates the efficacy of Large Language Models (LLMs) in causal discovery. Using newly available open-source LLMs, OLMo and BLOOM, which provide access to their pre-training corpora, we investigate how LLMs address causal discovery through three research questions. We examine: (i) the impact of memorization for accurate causal relation prediction, (ii) the influence of incorrect causal relations in pre-training data, and (iii) the contextual nuances that influence LLMs' understanding of causal relations. Our findings indicate that while LLMs are effective in recognizing causal relations that occur frequently in pre-training data, their ability to generalize to new or rare causal relations is limited. Moreover, the presence of incorrect causal relations significantly undermines the confidence of LLMs in corresponding correct causal relations, and the contextual information critically affects the outcomes of LLMs to discern causal connections between random variables.

Downloads

Next from ACL 2025

HateDay: Insights from a Global Hate Speech Dataset Representative of a Day on Twitter

Similar lecture

Contrastive Prompting Enhances Sentence Embeddings in LLMs through Inference-Time Steering