Austria

In recent years, large language models (LLMs) have had great success in tasks such as casual conversation, contributing to significant advancements in domains like virtual assistance. However, they often generate responses that are not aligned with human values (e.g., ethical standards, safety), leading to potentially unsafe or inappropriate outputs. While several techniques have been proposed to address this problem, they come with a cost, requiring computationally expensive training or dramatically increasing the inference time. In this paper, we present DIESEL, a lightweight inference-guidance technique that can be seamlessly integrated into any autoregressive LLM to semantically filter undesired concepts from the response. DIESEL can function either as a standalone safeguard or as an additional layer of defense, enhancing response safety by reranking the LLM&#39;s proposed tokens based on their similarity to predefined negative concepts in the latent space. Our evaluation demonstrates DIESEL&#39;s effectiveness on state-of-the-art conversational models, even in adversarial jailbreaking scenarios that challenge response safety. We also highlight DIESEL&#39;s generalization capabilities, showing that it can be used in use cases other than safety, providing general-purpose response filtering.

ACL 2025

DIESEL - Dynamic Inference-Guidance via Evasion of Semantic Embeddings in LLMs

token reranking

jailbreak robustness

semantic filtering

inference-time defense

llm safety

In recent years, large language models (LLMs) have had great success in tasks such as casual conversation, contributing to significant advancements in domains like virtual assistance. However, they often generate responses that are not aligned with human values (e.g., ethical standards, safety), leading to potentially unsafe or inappropriate outputs. While several techniques have been proposed to address this problem, they come with a cost, requiring computationally expensive training or dramatically increasing the inference time. In this paper, we present DIESEL, a lightweight inference-guidance technique that can be seamlessly integrated into any autoregressive LLM to semantically filter undesired concepts from the response. DIESEL can function either as a standalone safeguard or as an additional layer of defense, enhancing response safety by reranking the LLM's proposed tokens based on their similarity to predefined negative concepts in the latent space. Our evaluation demonstrates DIESEL's effectiveness on state-of-the-art conversational models, even in adversarial jailbreaking scenarios that challenge response safety. We also highlight DIESEL's generalization capabilities, showing that it can be used in use cases other than safety, providing general-purpose response filtering.

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 language models (LLMs) excel in natural language generation but also exhibit biases, particularly in gender, race, and religion, which can be amplified with widespread use. However, research on biases in specific domains, such as finance, remains limited. To address this gap, we conducted a comprehensive evaluation of 23 leading LLMs and found varying degrees of financial bias, including more pronounced biases in financial-specific LLMs (FinLLMs). In response, we propose the Financial Bias Indicators (FBI) framework, which includes components like the Bias Unveiler, Bias Detective, Bias Tracker, and Bias Antidote, designed to identify, detect, analyze, and mitigate financial biases. Our analysis explores the root causes of these biases and introduces a debiasing method based on financial causal knowledge, alongside three other debiasing techniques. For the most biased model, we successfully reduced bias by 68% according to key metrics. This study advances our understanding of LLM biases in finance and highlights the need for greater scrutiny in their application within this critical domain.

Are LLMs Rational Investors? A Study on the Financial Bias in LLMs

There is widespread agreement about the grounded nature of human learning and representation, and the belief that computational models of meaning need to be multimodal. In this paper, we ask to what degree does this belief hold in the era of models trained on billions of examples? We investigate the ability of pre-trained vision models to represent the semantic feature norms of concrete object concepts, e.g. a ROSE is red, smells sweet, and is a flower. More specifically, we use probing tasks to test which properties of objects these models are aware of. We evaluate image encoders trained on image data alone, as well as multimodally-trained image encoders and language-only models, on predicting an extended set of the classic McRae norms and the newer Binder dataset of attribute ratings. We find that multimodal image encoders slightly outperform language-only approaches, and that image-only encoders perform comparably to the language models, even on non-visual attributes that are classified as “encyclopedic” or “function”. These results offer new insights into what can be learned from pure unimodal learning, and the complementarity of the modalities.

Seeing What Tastes Good: Revisiting Multimodal Distributional Semantics in the Billion Parameter Era

This paper introduces the new and challenging TempVS benchmark, which focuses on temporal grounding and reasoning capabilities of Multimodal Large Language Models (MLLMs) in image sequences. TempVS consists of three main tests (i.e., event relation inference, sentence ordering and image ordering) and each accompanied with a basic grounding test, yielding a total of 2,085 annotated image sequences and 15k multiple-choice questions. TempVS requires MLLMs to rely on both visual and linguistic modalities to understand the temporal order of events. We extensively evaluate 38 state-of-the-art MLLMs, demonstrating that models struggle to solve TempVS. Our analysis reveals a substantial performance gap between current MLLMs and human capabilities, accompanied by fine-grained insights that suggest promising directions for future research.

Burn After Reading: Do Multimodal Large Language Models Truly Capture Order of Events in Image Sequences?

While personalized recommendations are often desired by users, it can be difficult in practice to distinguish cases of bias from cases of personalization: we find that models generate racially stereotypical recommendations regardless of whether the user revealed their identity intentionally through explicit indications or unintentionally through implicit cues. We demonstrate that when people use large language models (LLMs) to generate recommendations, the LLMs produce responses that reflect both what the user wants and who the user is. We argue that chatbots ought to transparently indicate when recommendations are influenced by a user’s revealed identity characteristics, but observe that they currently fail to do so. Our experiments show that even though a user’s revealed identity significantly influences model recommendations ($p < 0.001$), model responses obfuscate this fact in response to user queries. This bias and lack of transparency occurs consistently across multiple popular consumer LLMs and for four American racial groups.

Stereotype or Personalization? User Identity Biases Chatbot Recommendations

While a few high-quality bias benchmark datasets exist to address stereotypes in Language Models (LMs), a notable lack of focus remains on body image stereotypes. To bridge this gap, we propose BIStereo, a suite to uncover LMs’ biases towards people of certain physical appearance characteristics, namely, skin complexion, body shape, height, attire, and a miscellaneous category including hair texture, eye color, and more. Our dataset comprises 40k sentence pairs designed to assess LMs’ biased preference for certain body types. We further include 60k premise-hypothesis pairs designed to comprehensively assess LMs’ preference for fair skin tone. Additionally, we curate 553 tuples consisting of a body image descriptor, gender, and a stereotypical attribute, validated by a diverse pool of annotators for physical appearance stereotypes.We propose a metric, TriSentBias, that captures the biased preferences of LMs towards a certain body type over others. Using BIStereo, we assess the presence of body image biases in ten different language models, revealing significant biases in models Muril, XLMR, Llama3, and Gemma. We further evaluate the LMs through downstream NLI and Analogy tasks.Our NLI experiments highlight notable patterns in the LMs that align with the well-documented cognitive bias in humans known as the Halo Effect.

“You are Beautiful, Body Image Stereotypes are Ugly!” BIStereo: A Benchmark to Measure Body Image Stereotypes in Language Models

Large Language Models (LLMs) exhibit sophisticated reasoning yet still generate incorrect answers. We attribute these errors to **Spurious Beliefs**, defined as propositions the model internally considers as true despite being factually false. 

To reduce reasoning errors, we propose a belief space rectification framework. Our method first identifies the beliefs invoked during inference via an explanation‑based approach with Forward‑Backward Beam Search (FBBS). We subsequently apply unlearning via gradient ascent to suppress spurious beliefs and enhance true ones, thereby effectively rectifying the model’s belief space. 

Experiments on three QA datasets and three LLMs show that our method significantly reduces erroneous reasoning and improves generalization.

Rectifying Belief Space via Unlearning to Harness LLMs' Reasoning

Accurately grounding visual and textual elements within mobile user interfaces (UIs) remains a significant challenge for Vision-Language Models (VLMs). Visual grounding, a critical task in this domain, involves identifying the most relevant UI element or region based on a natural language query—a process that requires both precise perception and context-aware reasoning. In this work, we present - **MoUI**, a light-weight mobile UI understanding model trained on **MoIT**, an instruction-tuning dataset specifically tailored for mobile screen understanding and grounding, designed to bridge the gap between user intent and visual semantics. Complementing this dataset, we also present a human-annotated reasoning benchmark **MoIQ** that rigorously evaluates complex inference capabilities over mobile UIs. To harness these resources effectively, we propose a two-stage training approach that separately addresses perception and reasoning tasks, leading to stronger perception capabilities and improvement in reasoning abilities. Through extensive experiments, we demonstrate that our MoUI models achieve significant gains in accuracy across all perception tasks and _state-of-the-art_ results on public reasoning benchmark **ComplexQA (78%) and our MoIQ (49%)**. We will be open-sourcing our dataset, code, and models to foster further research and innovation in the field.

From Perception to Reasoning: Enhancing Vision-Language Models for Mobile UI Understanding

Controlling the syntactic structure of text generated by language models is valuable for applications requiring clarity, stylistic consistency, or interpretability, yet it remains a challenging task. In this paper, we argue that sampling algorithms based on the posterior inference can effectively enforce a target constituency structure during generation. Our approach combines sequential Monte Carlo, which estimates the posterior distribution by sampling from a proposal distribution, with a syntactic tagger that ensures that each generated token aligns with the desired syntactic structure. Our experiments with GPT2 and Llama3-8B models show that with an appropriate proposal distribution, we can improve syntactic accuracy, increasing the F1 score from 12.31 (GPT2-large) and 35.33 (Llama3-8B) to about 93 in both cases without compromising the language model's fluency. These results underscore both the complexity of syntactic control and the effectiveness of sampling algorithms, offering a promising approach for applications where precise control over syntax is essential.

Syntactic Control of Language Models by Posterior Inference

Large language models (LLMs) excel in complex reasoning tasks, and distilling their reasoning capabilities into smaller models has shown promise. However, we uncover an interesting phenomenon, which we term the Small Model Learnability Gap: small models (
3B parameters) do not consistently benefit from long chain-of-thought (CoT) reasoning or distillation from larger models. Instead, they perform better when fine-tuned on shorter, simpler reasoning chains that better align with their intrinsic learning capacity. To address this, we propose Mix Distillation, a simple yet effective strategy that balances reasoning complexity by combining long and short CoT examples or reasoning from both larger and smaller models. Our experiments demonstrate that Mix Distillation significantly improves small model reasoning performance compared to training on either data alone. These findings highlight the limitations of direct strong model distillation and underscore the importance of adapting reasoning complexity for effective reasoning capability transfer.

Small Models Struggle to Learn from Strong Reasoners

Cross-lingual vocabulary transfer plays a promising role in adapting pre-trained language models to new languages, including low-resource languages. Existing approaches that utilize monolingual or parallel corpora face challenges when applied to languages with limited resources. In this work, we propose a simple yet effective vocabulary transfer method that utilizes bilingual dictionaries, which are available for many languages, thanks to descriptive linguists. Our proposed method leverages a property of BPE tokenizers where removing a subword from the vocabulary causes a fallback to shorter subwords. The embeddings of target subwords are estimated iteratively by progressively removing them from the tokenizer. The experimental results show that our approach outperforms existing methods for low-resource languages, demonstrating the effectiveness of a dictionary-based approach for cross-lingual vocabulary transfer.

Downloads

Next from ACL 2025

Are LLMs Rational Investors? A Study on the Financial Bias in LLMs

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES