United States

While Large Language Models (LLMs) excel in zero-shot Question Answering (QA), they tend to expose biases in their internal knowledge when faced with socially sensitive questions, leading to a degradation in performance. Existing zero-shot methods are efficient but fail to consider context and prevent bias propagation in the answers. To address this, we propose *DeCAP*, a method for debiasing LLMs using Context-Adaptive Prompt Generation. *DeCAP* leverages a *Question Ambiguity Detection* to take appropriate debiasing actions based on the context and a *Neutral Answer Guidance Generation* to suppress the LLMs make objective judgments about the context, minimizing the propagation of bias from their internal knowledge. Our various experiments across eight LLMs show that *DeCAP* achieves state-of-the- art zero-shot debiased QA performance. This demonstrates *DeCAP*’s efficacy in enhancing the fairness and accuracy of LLMs in diverse QA settings.

NAACL 2025

DeCAP: Context-Adaptive Prompt Generation for Debiasing Zero-shot Question Answering in Large Language Models

debiasing

zero-shot

### 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

technical paper

Speech enhancement (SE) is the fundamental task of enhancing the clarity and quality of speech in the presence of non-stationary additive noise. While deterministic deep learning models have been commonly employed for SE, recent research indicates that generative models, such as denoising diffusion probabilistic models (DDPMs), have shown promise. However, different from speech generation, SE has a strong constraint to generate results in accordance with the underlying ground-truth signal. Additionally, for a wide variety of applications, SE systems need to be employed in real-time, and traditional diffusion models (DMs) requiring many iterations of a large model during inference are inefficient. To address these issues, we propose ProSE (diffusion-based Priors for SE), a novel methodology based on an alternative framework for applying diffusion models to SE. Specifically, we first apply DDPMs to generate priors in a latent space due to their powerful distribution mapping capabilities. The priors are then integrated into a transformer-based regression model for SE. The priors guide the regression model in the enhancement process. Since the diffusion process is applied to a compact latent space, the diffusion model takes fewer iterations than the traditional DM to obtain accurate estimations. Additionally, using a regression model for SE avoids the distortion issue caused by misaligned details generated by DMs. Comprehensive experiments show that ProSE achieves state-of-the-art performance on synthetic and real-world datasets using various metrics while consuming less computational costs.

ProSE: Diffusion Priors for Speech Enhancement

Faithfulness evaluators based on Large Language Models (LLMs) are often fooled by the fluency of the text and struggle with identifying errors in the summaries, usually leading to high false negative rate. We propose an approach to summary faithfulness evaluation in which multiple LLM-based agents are assigned initial stances (regardless of what their belief might be) and forced to come up with a reason to justify the imposed belief, thus engaging in a multi-round debate to reach an agreement. The uniformly distributed initial assignments here result in a greater diversity of stances leading to more meaningful debates and ultimately more errors identified. Furthermore, by analyzing the recent faithfulness evaluation datasets, we observe that naturally, it is not always the case for a summary to be either faithful to the source document or not. We therefore introduce a new dimension ambiguity and a detailed taxonomy to identify such special cases. Experiments demonstrate our approach can help identify ambiguities, and have even a stronger performance on non-ambiguous summaries.

Faithful, Unfaithful or Ambiguous? Multi-Agent Debate with Initial Stance for Summary Evaluation

Conversational recommender systems (CRSs) are designed to suggest the target item that the user is likely to prefer through multi-turn conversations. Recent studies stress that capturing sentiments in user conversations improves recommendation accuracy. However, they employ a single user representation, which may fail to distinguish between contrasting user intentions, such as likes and dislikes, potentially leading to suboptimal performance. To this end, we propose a novel conversational recommender model, called COntrasting user pReference expAnsion and Learning (CORAL). Firstly, CORAL extracts the user’s hidden pref- erences through contrasting preference expansion using the reasoning capacity of the LLMs. Based on the potential preference, CORAL explicitly differentiates the contrasting preferences and leverages them into the recommendation process via preference-aware learning. Extensive experiments show that CORAL significantly outperforms existing methods in three benchmark datasets, improving up to 99.72% in Recall@10. The code and datasets are available at https://github.com/kookeej/CORAL.

Empowering Retrieval-based Conversational Recommendation with Contrasting User Preferences

Scaling Intelligence the Human Way

The State and Future of SIGMORPHON

Human cognition constrains how we communicate. Our cognitive biases and preferences interact with the processes that drive language emergence and change in non-trivial ways. A powerful method to discern the roles of cognitive biases and processes like language learning and use in shaping linguistic structure is to build agent-based models. Recent advances in computational linguistics and deep learning sparked a renewed interest in such simulations, creating the opportunity to model increasingly realistic phenomena. These models simulate emergent communication, referring to the spontaneous development of a communication system through repeated interactions between individual neural network agents. However, a crucial challenge in this line of work is that such artificial learners still often behave differently from human learners. Directly inspired by human artificial language learning studies, we proposed a novel framework for simulating language learning and change, which allows agents to first learn an artificial language and then use it to communicate, with the aim of studying the emergence of specific linguistic properties. I will show how we used it to simulate the emergence of a well-known language phenomenon: the word-order/case-marking trade-off. I will also share some recent findings where we test for the presence of a well-known human cross-modal mapping preference (the bouba-kiki effect) in vision-and-language models. Cross-modal associations play an essential role in human language understanding, learning, and evolution, but our findings reveal that multimodal language models do not align well with such human preferences. Finally, I will provide a sneak peek at new findings that reveal what a novel artificial language looks like when it has emerged to adapt to preferences of both humans and LLMs in a hybrid language game experiment.

The emergence of language universals in neural agents and vision-and-language models

Transformers has redefined deep learning research and has become the most prominent architecture across domains such as natural language processing, computer vision, and image processing. Attention mechanism, particularly self-attention, is central to the success of this architecture, which allows the model to capture dependencies across the input sequences. However, the fundamental challenge in understanding self-attention is its intrinsic symmetry. The existing works often consider self-attention as a kernel method, leveraging symmetric kernels based on Mercer’s theorem. However, the self-attention matrices used in the transformer architectures are inherently asymmetric, which leads to an inconsistency
between the theoretical formulation and the practical implementation. The primal-attention, a novel attention mechanism based on kernel singular value decomposition explicitly models the asymmetry.
Therefore, reformulating self-attention using primal-dual representation ensures efficient computation and low-rank approximation that enhances performance and generalization.

Understanding Attention in Asymmetric Kernel Point of View

The past and future of grapheme-to-phoneme conversion

Scientific research is inherently slow, complex, and highly specialized, relying on an iterative cycle of hypothesis generation and validation. Can we create an AI-powered biologist to automate these processes at scale? In this talk, I will outline the key ingredients for building such a system. I will first introduce Biomni, a generalist AI agent that integrates large language models with a vast ecosystem of specialized biological tools, software, and databases. By linking AI-driven reasoning with real-world execution, Biomni enables a broad range of biomedical applications. Building on this foundation, I will introduce BioDiscoveryAgent, an AI system for automated hypothesis generation that optimizes experimental design by identifying the most promising perturbations for screening. By leveraging literature-based reasoning, it significantly outperforms traditional Bayesian optimization approaches. Finally, beyond hypothesis generation, rigorous validation is essential. I will present POPPER, an AI-driven sequential falsification framework that automates hypothesis validation with statistical guarantees, ensuring scientific robustness at scale.

Downloads

Next from NAACL 2025

ProSE: Diffusion Priors for Speech Enhancement

Similar lecture

Self-Debiasing Large Language Models: Zero-Shot Recognition and Reduction of Stereotypes

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from NAACL 2025

ProSE: Diffusion Priors for Speech Enhancement

Similar lecture

Self-Debiasing Large Language Models: Zero-Shot Recognition and Reduction of Stereotypes

Downloads