United States

Multiple-Choice Question Answering (MCQA) is a widely used task in the evaluation of Large Language Models (LLMs). In this work, we reveal that current LLMs&#39; performance in MCQA could be heavily influenced by the choice of option symbol sets, due to the option symbol bias. That is, when altering only the option symbols (e.g., A/B/C/D→ i/ii/iii/iv), the results could vary sharply, leading to a margin of approximately 10% in accuracy. To uncover the mechanisms behind this, we investigate the internal components of LLMs from a causal perspective. By measuring the causal effects, we identify a small subset of attention heads responsible for the symbol bias. Subsequently, we interpret these key components in a human-understandable way, showing that attention heads with higher causal effects are more likely to focus on only option symbols, while those with lower causal effects tend to distribute their attention across the content of questions and options. It also motivates us to pursue debiasing based on the causal effects. Specifically, to mitigate such bias, we propose a tuning-free, causal effect driven debiasing method which intervenes the activations of identified components according to their causal effects, with stronger interventions corresponding to higher causal effects. Experimental results demonstrate that the proposed method not only alleviates aforementioned bias, but also improves the MCQA performance of LLMs.

NAACL 2025

Option Symbol Matters: Investigating and Mitigating Multiple-Choice Option Symbol Bias of Large Language Models

activation intervention

interpretability

It is my great pleasure and honor to welcome you to the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), held in San Diego, California, USA, July 2–7, 2026. ACL 2026 continues our fields 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.

It is exciting to see the many developments in the field reflected at our conference, such as the advancement of large language models (born out of our community) into many application directions. This is not without challenges, partly due to unsolved problems, and partly due to misuse of the technology. All this is reflected well at ACL 2026. The conference also reflects that the field is not only about advancing technology, but also the science of fundamentally understanding language and how it is use by many different speaker communities. 
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) 
<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing')">Conference Handbook</button></html>

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 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.aclweb.org/registration/) first.

Registration Required

ACL 2026

The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) will take place in San Diego, California, from July 2 - 7, 2026.

Multiple-Choice Question Answering (MCQA) is a widely used task in the evaluation of Large Language Models (LLMs). In this work, we reveal that current LLMs' performance in MCQA could be heavily influenced by the choice of option symbol sets, due to the option symbol bias. That is, when altering only the option symbols (e.g., A/B/C/D→ i/ii/iii/iv), the results could vary sharply, leading to a margin of approximately 10% in accuracy. To uncover the mechanisms behind this, we investigate the internal components of LLMs from a causal perspective. By measuring the causal effects, we identify a small subset of attention heads responsible for the symbol bias. Subsequently, we interpret these key components in a human-understandable way, showing that attention heads with higher causal effects are more likely to focus on only option symbols, while those with lower causal effects tend to distribute their attention across the content of questions and options. It also motivates us to pursue debiasing based on the causal effects. Specifically, to mitigate such bias, we propose a tuning-free, causal effect driven debiasing method which intervenes the activations of identified components according to their causal effects, with stronger interventions corresponding to higher causal effects. Experimental results demonstrate that the proposed method not only alleviates aforementioned bias, but also improves the MCQA performance of LLMs.

poster

### Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Welcome to the 2025 meeting of the Nations of the Americas Chapter of the Association for Computational Linguistics! I am proud to help organize the first NAACL conference to carry the new name of our organization, one that emphasizes inclusion for all of the Americas. I am also pleased to welcome you to Albuquerque, New Mexico, a state whose unique blend of cultural influences will make for an excellent backdrop for NAACL 2025, especially with this year’s special theme on NLP in a Multicultural World. 
**[Continue reading...](https://drive.google.com/file/d/1jX-qGhqVSZZCIrAnJaz798pflu5Irrdn/view?usp=sharing)**

*- Colin Cherry, Google, NAACL 2025 General Chair* 

[![](https://assets.underline.io/markdown_image/1/image/b087f8a4dc5816d6e1a6514e59c59ac3.png)](https://drive.google.com/file/d/1T96GzPqObXrMln2BMByCSXTSizjTg69P/view?usp=sharing)

You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Please log in to explore this event.

To access NAACL 2025 event page you are required to register. Please follow [**this link**](https://2025.naacl.org/registration/registration/) to register. Access will depend on your registration type.

Please register!

Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Large language Models (LLMs) have demonstrated remarkable skills across various domains. Understanding the mechanisms behind their abilities and implementing controls over them is becoming increasingly important for developing better models. In this paper, we focus on skill unlearning in LLMs, specifically unlearning a particular skill while retaining their overall capabilities. We introduce two lightweight, training-free machine skill unlearning techniques for LLMs. First, we observe that the pre-activation distribution of neurons in each Feed-Forward Layer (FFL) differs when the model demonstrates different skills. Additionally, we find that queries triggering the same skill cluster within the FFL key space and can be separated from other queries using a hypercube. Based on these observations, we propose two lightweight, training-free skill unlearning methods: `Neuron Adjust` and `Key Space Detection (KSD)`. We evaluate our methods on unlearning math-solving, Python-coding, and comprehension skills across seven different languages. The results demonstrate their strong unlearning capabilities for the designated skills. Specifically, `Key Space Detection` achieves over 80% relative performance drop on the forgetting skill and less than 10% relative performance drop on other skills and the model's general knowledge (MMLU) for most unlearning tasks.

Effective Skill Unlearning through Intervention and Abstention

In this paper, we investigate the overlooked impact of instruction-tuning on memorization in large language models (LLMs), which has largely been studied in base, pre-trained models. We propose a black-box prompt optimization method where an attacker LLM agent uncovers higher levels of memorization in a victim agent, surpassing traditional approaches that prompt the model directly with training data. Using an iterative rejection-sampling process, we design instruction-based prompts that minimize overlap with training data to avoid providing direct solutions while maximizing overlap between the victim's output and the training data to induce memorization. Our method shows 23.7% more overlap with training data compared to state-of-the-art baselines. We explore two attack settings: an analytical approach that determines the empirical upper bound of the attack, both with and without access to responses for prompt initialization, and a practical classifier-based method for assessing memorization without access to memorized data. Our findings reveal that instruction-tuned models can expose pre-training data as much as, or more than, base models; contexts beyond the original training data can lead to leakage; and instructions generated by other LLMs open new avenues for automated attacks, which we believe require further exploration.

ALPACA AGAINST VICUNA: Using LLMs to Uncover Memorization of LLMs

The rapid development of Large Multimodal Models (LMMs) has significantly advanced multimodal understanding by harnessing the language abilities of Large Language Models (LLMs) and integrating modality-specific encoders. However, LMMs are plagued by hallucinations that limit their reliability and adoption. While traditional methods to detect and mitigate these hallucinations often involve costly training or rely heavily on external models, recent approaches utilizing internal model features present a promising alternative. In this paper, we critically assess the limitations of the state-of-the-art training-free technique, the logit lens, in handling generalized visual hallucinations. We introduce *ContextualLens*, a refined method that leverages contextual token embeddings from middle layers of LMMs. This approach significantly improves hallucination detection and grounding across diverse categories, including actions and OCR, while also excelling in tasks requiring contextual understanding, such as spatial relations and attribute comparison. Our novel grounding technique yields highly precise bounding boxes, facilitating a transition from Zero-Shot Object Segmentation to Grounded Visual Question Answering. Our contributions pave the way for more reliable and interpretable multimodal models.

Beyond Logit Lens: Contextual Embeddings for Robust Hallucination Detection & Grounding in VLMs

This paper investigates the knowledge of language models from the perspective of Bayesian epistemology. We explore how language models adjust their confidence and responses when presented with evidence with varying levels of informativeness and reliability. To study these properties, we create a dataset with various types of evidence and analyze language models' responses and confidence using verbalized confidence, token probability, and sampling. We observed that language models do not consistently follow Bayesian epistemology: language models follow the Bayesian confirmation assumption well with true evidence but fail to adhere to other Bayesian assumptions when encountering different evidence types. Also, we demonstrated that language models can exhibit high confidence when given strong evidence, but this does not always guarantee high accuracy. Our analysis also reveals that language models are biased toward golden evidence and show varying performance depending on the degree of irrelevance, helping explain why they deviate from Bayesian assumptions.

From Evidence to Belief: A Bayesian Epistemology Approach to Language Models

Vision-Language Models (VLMs) have gained prominence due to their success in solving complex cross-modal tasks. However, the internal mechanisms of VLMs, particularly the roles of cross-attention and self-attention in multimodal integration, are not fully understood. To address this gap, we introduce NOTICE, a Gaussian-Noise-free Text-Image Corruption and Evaluation pipeline for mechanistic interpretability in VLMs. NOTICE introduces Semantic Image Pairs (SIP) corruption, the first visual counterpart to Symmetric Token Replacement (STR) for text. Through NOTICE, we uncover a set of ``universal attention heads'' in BLIP and LLaVA that consistently contribute across different tasks and modalities. In BLIP, cross-attention heads implement object detection, object suppression, and outlier suppression, whereas important self-attention heads in LLaVA only perform outlier suppression. Notably, our findings reveal that cross-attention heads perform image-grounding, while self-attention in LLaVA heads do not, highlighting key differences in how VLM architectures handle multimodal learning.

What Do VLMs NOTICE? A Mechanistic Interpretability Pipeline for Gaussian-Noise-free Text-Image Corruption and Evaluation

Large language models have emerged as powerful tools for general intelligence, showcasing advanced natural language processing capabilities that find applications across diverse domains. Despite their impressive performance, recent studies have highlighted the potential for significant enhancements in LLMs' task-specific performance through fine-tuning strategies like Reinforcement Learning with Human Feedback (RLHF), supervised fine-tuning (SFT), and Quantized Low-Rank Adapters (Q-LoRA) method. However, previous works have shown that while fine-tuning offers significant performance gains, it also leads to challenges such as catastrophic forgetting and privacy and safety risks. To this end, there has been little to no work in *understanding the impact of fine-tuning on the reasoning capabilities of LLMs*. Our research investigates the effect of fine-tuning on the reasoning abilities of LLMs, addressing critical questions regarding the impact of task-specific fine-tuning on overall reasoning capabilities, the influence of fine-tuning on Chain-of-Thought (CoT) reasoning performance, and the implications for the faithfulness of CoT reasonings. By exploring these dimensions, our study shows the impact of fine-tuning on LLM reasoning capabilities, where the faithfulness of CoT reasoning, on average across four datasets, decreases, highlighting potential shifts in internal mechanisms of the LLMs resulting from fine-tuning processes.

On the Impact of Fine-Tuning on Chain-of-Thought Reasoning

Tool-using agents that act in the world need to be both useful and safe. Well-calibrated model confidences can be used to weigh the risk versus reward of potential actions, but prior work shows that many models are poorly calibrated. Inspired by interpretability literature exploring the internals of models, we propose a novel class of model-internal confidence estimators (MICE) to better assess confidence when calling tools. MICE first decodes from each intermediate layer of the language model using logit lens and then computes similarity scores between each layer's generation and the final output. These features are fed into a learned probabilistic classifier to assess confidence in the decoded output. On the simulated trial and error (STE) tool-calling dataset using Llama3 models, we find that MICE beats or matches the baselines on smoothed expected calibration error. Using MICE confidences to determine whether to call a tool significantly improves over strong baselines on a new metric, expected tool-calling utility. Further experiments show that MICE is sample-efficient, can generalize zero-shot to unseen APIs, and results in higher tool-calling utility in scenarios with varying risk levels. Our code is open source, available at https://github.com/microsoft/mice_for_cats.

MICE for CATs: Model-Internal Confidence Estimation for Calibrating Agents with Tools

As large language models (LLMs) are increasingly deployed in diverse applications, including chatbot assistants and code generation, aligning their behavior with safety and ethical standards has become paramount. However, jailbreak attacks, which exploit vulnerabilities to elicit unintended or harmful outputs, threaten LLMs safety significantly. In this paper, we introduce Layer-AdvPatcher, a novel methodology designed to defend against jailbreak attacks by utilizing an unlearning strategy to patch specific layers within LLMs through self-augmented datasets. Our insight is that certain layer(s), tend to produce affirmative tokens when faced with harmful prompts. By identifying these layers and adversarially exposing them to generate more harmful data, one can understand their inherent and diverse vulnerabilities to attacks. With these exposures, we then “unlearn” these issues, reducing the impact of affirmative tokens and hence minimizing jailbreak risks while keeping the model’s responses to safe queries intact. We conduct extensive experiments on two models, four benchmark datasets, and multiple state-of-the-art jailbreak attacks to demonstrate the efficacy of our approach. Results indicate that our framework reduces the harmfulness and attack success rate of jailbreak attacks without compromising utility for benign queries compared to recent defense methods. Our code is publicly available at: https://github.com/oyy2000/LayerAdvPatcher

Layer-Level Self-Exposure and Patch: Affirmative Token Mitigation for Jailbreak Attack Defense

In-context knowledge editing (IKE) enables efficient modification of large language model (LLM) outputs without parameter changes and at zero-cost. However, it can be misused to manipulate responses opaquely, e.g., insert misinformation or offensive content. Such malicious interventions could be incorporated into high-level wrapped APIs where the final input prompt is not shown to end-users. To address this issue, we investigate the detection and reversal of IKE-edits. First, we demonstrate that IKE-edits can be detected with high accuracy (F1 > 80%) using only the top-10 output probabilities of the next token, even in a black-box setting, e.g. proprietary LLMs with limited output information. Further, we introduce the novel task of reversing IKE-edits using specially tuned reversal tokens. We explore using both continuous and discrete reversal tokens, achieving over 80% accuracy in recovering original, unedited outputs across multiple LLMs. Our continuous reversal tokens prove particularly effective, with minimal impact on unedited prompts. Through analysis of output distributions, attention patterns, and token rankings, we provide insights into IKE's effects on LLMs and how reversal tokens mitigate them. This work represents a significant step towards enhancing LLM resilience against potential misuse of in-context editing, improving their transparency and trustworthiness.

How to Make LLMs Forget: On Reversing In-Context Knowledge Edits

Topic models are a popular approach for extracting semantic information from large document collections. However, recent studies suggest that the topics generated by these models often do not align well with human intentions. Although metadata such as labels and authorship information are available, it has not yet been effectively incorporated into neural topic models. To address this gap, we introduce FANToM, a novel method to align neural topic models with both labels and authorship information. FANToM allows for the inclusion of this metadata when available, producing interpretable topics and author distributions for each topic. Our approach demonstrates greater expressiveness than conventional topic models by learning the alignment between labels, topics, and authors. Experimental results show that FANToM improves existing models in terms of both topic quality and alignment. Additionally, it identifies author interests and similarities.

Downloads

Next from NAACL 2025

Effective Skill Unlearning through Intervention and Abstention

Similar lecture

When Models Decide and When They Bind: A Two-Stage Computation for Multiple-Choice Question Answering