Austria

Semantic text representation is a fundamental task in the field of natural language processing. Existing text embedding (e.g., SimCSE and LLM2Vec) have demonstrated excellent performance, but the values of each dimension are difficult to trace and interpret. Bag-of-words, as classic sparse interpretable embeddings, suffers from poor performance. Recently, Benara et al. (2024) propose interpretable text embeddings using large language models, which forms &#39;&#39;0/1&#39;&#39; embeddings based on responses to a series of questions. These interpretable text embeddings are typically high-dimensional (larger than 10,000). In this work, we propose Low-dimensional (lower than 500) Dense and Interpretable text embeddings with Relative representations (LDIR). The numerical values of its dimensions indicate semantic relatedness to different anchor texts through farthest point sampling, offering both semantic representation as well as a certain level of traceability and interpretability. We validate LDIR on multiple semantic textual similarity, retrieval, and clustering tasks. Extensive experimental results show that LDIR performs close to the black-box baseline models and outperforms the interpretable embeddings baselines with much fewer dimensions.

ACL 2025

LDIR: Low-Dimensional Dense and Interpretable Text Embeddings with Relative Representations

text embeddings

text representation

interpretability

### 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)

Semantic text representation is a fundamental task in the field of natural language processing. Existing text embedding (e.g., SimCSE and LLM2Vec) have demonstrated excellent performance, but the values of each dimension are difficult to trace and interpret. Bag-of-words, as classic sparse interpretable embeddings, suffers from poor performance. Recently, Benara et al. (2024) propose interpretable text embeddings using large language models, which forms ''0/1'' embeddings based on responses to a series of questions. These interpretable text embeddings are typically high-dimensional (larger than 10,000). In this work, we propose Low-dimensional (lower than 500) Dense and Interpretable text embeddings with Relative representations (LDIR). The numerical values of its dimensions indicate semantic relatedness to different anchor texts through farthest point sampling, offering both semantic representation as well as a certain level of traceability and interpretability. We validate LDIR on multiple semantic textual similarity, retrieval, and clustering tasks. Extensive experimental results show that LDIR performs close to the black-box baseline models and outperforms the interpretable embeddings baselines with much fewer dimensions.

poster

Decomposing weight matrices into quantization and low-rank components ($\bf W\approx Q+LR$) is a widely used technique for compressing large language models (LLMs). Existing joint optimization methods iteratively alternate between quantization and low-rank approximation. However, these methods tend to prioritize one component at the expense of the other, resulting in suboptimal decompositions that fail to leverage each component's unique strengths. In this work, we introduce Outlier-Driven Low-Rank Initialization (ODLRI), which assigns low-rank components the specific role of capturing activation-sensitive weights. This structured decomposition mitigates outliers' negative impact on quantization, enabling more effective balance between quantization and low-rank approximation. Experiments on Llama2 (7B, 13B, 70B), Llama3-8B, and Mistral-7B demonstrate that incorporating ODLRI into the joint optimization framework consistently reduces activation-aware error, minimizes quantization scale, and improves perplexity and zero-shot accuracy in low-bit settings.

Assigning Distinct Roles to Quantized and Low-Rank Matrices Toward Optimal Weight Decomposition

In the stance detection task, a text is classified as either favorable, opposing, or neutral towards a target. Prior work suggests that the use of external information, e.g., excerpts from Wikipedia, improves stance detection performance. However, whether or not such information can benefit large language models (LLMs) remains an unanswered question, despite their wide adoption in many reasoning tasks. In this study, we conduct a systematic evaluation on how Wikipedia and web search external information can affect stance detection across eight LLMs and in three datasets with 12 targets. Surprisingly, we find that such information degrades performance in most cases, with macro F1 scores dropping by up to 27.9\%. We explain this through experiments showing LLMs' tendency to align their predictions with the stance and sentiment of the provided information rather than the ground truth stance of the given text. We also find that performance degradation persists with chain-of-thought prompting, while fine-tuning mitigates but does not fully eliminate it. Our findings, in contrast to previous literature on BERT-based systems which suggests that external information enhances performance, highlight the risks of information biases in LLM-based stance classifiers.

Is External Information Useful for Stance Detection with LLMs?

Document-level context is crucial for handling discourse challenges in text-to-text document-level machine translation (MT). Despite the increased discourse challenges introduced by noise from automatic speech recognition (ASR), the integration of document-level context in speech translation (ST) remains insufficiently explored. In this paper, we develop DoCIA, an online framework that enhances ST performance by incorporating document-level context. DoCIA decomposes the ST pipeline into four stages. Document-level context is integrated into the ASR refinement, MT, and MT refinement stages through auxiliary LLM (large language model)-based modules. Furthermore, DoCIA leverages document-level information in a multi-level manner while minimizing computational overhead. Additionally, a simple yet effective determination mechanism is introduced to prevent hallucinations from excessive refinement, ensuring the reliability of the final results. Experimental results show that DoCIA significantly outperforms traditional ST baselines in both sentence and discourse metrics across four LLMs, demonstrating its effectiveness in improving ST performance.

DoCIA: An Online Document-Level Context Incorporation Agent for Speech Translation

Vision Language Models (VLMs) have achieved significant advancements in complex visual understanding tasks. However, VLMs are prone to hallucinations—generating outputs that lack alignment with visual content. This paper addresses hallucination detection in VLMs by leveraging the visual grounding information encoded in transformer attention maps. We identify three primary challenges in this approach: the elective nature of visual grounding for certain tokens, the high-dimensional and noisy nature of attention maps, and the dynamic sequence length of attention on previous tokens. To address these, we propose VADE, a novel sequence modelling approach to effectively learn complex sequential patterns from high-dimensional and noisy attention maps for fine-grained hallucination detection and mitigation. VADE achieves an average PR-AUC of 80% in hallucination detection on M-HalDetect across four different model architectures and an 5% improvement in hallucination mitigation on MSCOCO.

VADE: Visual Attention Guided Hallucination Detection and Elimination

As large language models (LLMs) grow in parameter size and context length, computation precision has been reduced from 16-bit to 4-bit to improve inference efficiency. However, this reduction causes accuracy degradation due to activation outliers. Recent rotation-based INT4 quantization attempts to address this through rotation matrix calibration, but they require hours of overhead per model deployment and leave significant computation unquantized in long-context scenarios. Microscaling (MX) floating-point (FP) formats offer fine-grained representation with a shared scale, enabling fully quantized matrix multiplications through direct casting without calibration. However, existing research shows unsatisfactory empirical results for MXFP4 inference, and the robustness of MX formats remains largely unexplored.

In this work, we uncover the fundamental tradeoffs of the MX format: while it effectively suppresses activation outliers, it does so at the cost of increased group-wise asymmetry. To address this, we propose an asymmetric MX format for a 4-bit floating point (AMXFP4), which employs asymmetric shared scales to handle both outliers and group-wise asymmetry without requiring calibration. Our custom compute-engine implementation shows that the AMXFP4-based Multiply-Accumulate (MAC) design adds marginal resource overhead while delivering substantial accuracy improvements. Extensive experiments across benchmarks demonstrate that AMXFP4 outperforms MXFP4 in visual question answering (VQA) by 3% and surpasses rotation-based techniques on CSQA by 1.6%. Additionally, AMXFP4 shows superior performance compared to the recently deployed commercial MXFP4 format.

AMXFP4: Taming Activation Outliers with Asymmetric Microscaling Floating-Point for 4-bit LLM Inference

There have been attempts to utilize linear probe for detoxification, with existing studies relying on a single toxicity probe vector to reduce toxicity. However, toxicity can be fine-grained into various subcategories, making it difficult to remove certain types of toxicity by using a single toxicity probe vector. To address this limitation, we propose a category-specific toxicity probe vector approach. First, we train multiple toxicity probe vectors for different toxicity categories. During generation, we dynamically select the most relevant toxicity probe vector based on the current context. Finally, the selected vector is dynamically scaled and subtracted from model. Our method successfully mitigated toxicity from categories that the single probe vector approach failed to detoxify. Experiments demonstrate that our approach achieves up to a 78.52% reduction in toxicity on the evaluation dataset, while fluency remains nearly unchanged, with only a 0.052% drop compared to the unsteered model.

DAPI: Domain Adaptive Toxicity Probe Vector Intervention, for Fine-Grained Detoxification

Large language models have shown tremendous potential across various NLP tasks, and instruction tuning has been widely adopted to elicit their superior performance. However, instruction tuning may overly tailor the models to task-specific formats, potentially compromising their generalization on unseen tasks. 
We attribute the issue to the spurious correlations learned between inputs and targets. We propose explicit task knowledge injection to mitigate these shortcuts with latent task adaptation and knowledge reinstatement. Latent tasks serve as interpolations between new tasks and facilitate knowledge sharing with joint adaptation enabling the model to build task knowledge more smoothly. Knowledge reinstatement helps optimize building new knowledge with prior knowledge. Specifically, we retrieve input-relevant latent tasks and jointly learn the task and the relevant latent tasks. Moreover, we prompt the model to recall the forms of inputs corresponding to the target and build the task knowledge through the reinstatement of prior knowledge while learning the new task.
We conduct extensive experiments on state-of-the-art large language models including Llama3.1-8B and Vicuna-13B across 1000+ instruction-following tasks to demonstrate the effectiveness of our method. The results demonstrate our method improves generalization on both in-domain and out-of-domain unseen tasks.

Task Knowledge Injection via Interpolations and Reinstatement for Large Language Model Generalization

Memorization is a fundamental ability of Transformer-based Large Language Models, achieved through learning. In this paper, we propose a paradigm shift by designing an architecture to memorize text directly, bearing in mind the principle that memorization precedes learning. We introduce MeMo, a novel architecture for language modeling that explicitly memorizes sequences of tokens in layered associative memories. By design, MeMo offers transparency and the possibility of model editing, including forgetting texts. We experimented with the MeMo architecture, showing the memorization power of the one-layer and the multi-layer configurations.

Position Paper: MeMo: Towards Language Models with Associative Memory Mechanisms

We present DeRAGEC, a method for improving Named Entity (NE) correction in Automatic Speech Recognition (ASR) systems. By extending the Retrieval-Augmented Generative Error Correction (RAGEC) framework, DeRAGEC employs synthetic denoising rationales to filter out noisy NE candidates before correction. By leveraging phonetic similarity and augmented definitions, it refines noisy retrieved NEs using in-context learning, requiring no additional training. Experimental results on CommonVoice and STOP datasets show significant improvements in Word Error Rate (WER) and NE hit ratio, outperforming baseline ASR and RAGEC methods. Specifically, we achieved a 28% relative reduction in WER compared to ASR without postprocessing.

DeRAGEC: Denoising Named Entity Candidates with Synthetic Rationale for ASR Error Correction

Personalized opinion summarization is crucial as it considers individual user interests while generating product summaries. Recent studies show that although large language models demonstrate powerful text summarization and evaluation capabilities without the need for training data, they face difficulties in personalized tasks involving long texts. To address this, Rehearsal, a personalized opinion summarization framework via LLMs-based role-playing, is proposed. Having the model act as the user, the model can better understand the user's personalized needs. Additionally, a role-playing supervisor and practice process are introduced to improve the role-playing ability of the LLMs, leading to a better expression of user needs. Furthermore, through suggestions from virtual users, the summary generation is intervened, ensuring that the generated summary includes information of interest to the user, thus achieving personalized summary generation.
Experiment results demonstrate that our method can effectively improve the level of personalization in large model-generated summaries.

Downloads

Next from ACL 2025

Assigning Distinct Roles to Quantized and Low-Rank Matrices Toward Optimal Weight Decomposition

Similar lecture

Redundancy, Isotropy, and Intrinsic Dimensionality of Prompt-based Text Embeddings