Austria

Large language models (LLMs) may exhibit unintended or undesirable behaviors. Recent works have concentrated on aligning LLMs to mitigate harmful outputs. Despite these efforts, some anomalies indicate that even a well-conducted alignment process can be easily circumvented, whether intentionally or accidentally. Does alignment fine-tuning yield have robust effects on models, or are its impacts merely superficial? In this work, we make the first exploration of this phenomenon from both theoretical and empirical perspectives. Empirically, we demonstrate the \textit{elasticity} of post-alignment models, i.e., the tendency to revert to the behavior distribution formed during the pre-training phase upon further fine-tuning. Leveraging compression theory, we formally deduce that fine-tuning disproportionately undermines alignment relative to pre-training, potentially by orders of magnitude. We validate the presence of elasticity through experiments on models of varying types and scales. Specifically, we find that model performance declines rapidly before reverting to the pre-training distribution, after which the rate of decline drops significantly. Furthermore, we further reveal that elasticity positively correlates with the increased model size and the expansion of pre-training data. Our findings underscore the need to address the inherent elasticity of LLMs to mitigate their resistance to alignment.

ACL 2025

Language Models Resist Alignment: Evidence From Data Compression

alignment

fine-tuning

robustness

## Dr. Nishikant (Nish) Sonwalkar: Academician, Scientist, Musician

![](https://assets.underline.io/uploads/markdown_image/1/image/0f9b63a0b3deba4a355d98e2c87af68a.png)

Dr. Sonwalkar is regarded as a leading expert in the application of computers in education. He has extensive research and development experience in hypermedia authoring, mobile learning, adaptive learning, and brain-based synaptic learning, advanced scientific visualization, database management systems, and computer simulation.

After receiving doctoral degree from MIT in the area of Molecular Dynamics of Nano-interfaces, he joined MIT as faculty in Mechanical Engineering, but his passion for technology enabled education led him become the founding director of Hypermedia Teaching Facility (Hypermedia Lab) at MIT and later he was appointed as the Principal Educational Architect of MIT.

<iframe width="560" height="315" src="https://www.youtube.com/embed/dgGdFmr9kgs" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>

NISH 2021

technical paper

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair:<br>
*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.*<br>
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)<br><br><br>



To access this event page, you need to log in with the **email address you registered with**. <br>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) have become increasingly central to AI applications worldwide, necessitating robust multilingual safety alignment to ensure secure deployment across diverse linguistic contexts. Existing preference learning methods for safety alignment, such as RLHF and DPO, are primarily monolingual and struggle with noisy multilingual data. To address these limitations, we introduce \textbf{\underline{Mul}}tilingual reward ga\textbf{\underline{P}} \textbf{\underline{O}}ptimization (\textbf{MPO}), a novel approach that leverages the well-aligned safety capabilities of the dominant language (\textit{e.g.}, English) to improve safety alignment across multiple languages. MPO directly minimizes the reward gap difference between the dominant language and target languages, effectively transferring safety capabilities while preserving the original strengths of the dominant language. Extensive experiments on three LLMs, LLaMA-3.1, Gemma-2 and Qwen2.5, validate MPO’s efficacy in multilingual safety alignment without degrading general multilingual utility.

MPO: Multilingual Safety Alignment via Reward Gap Optimization

The application scope of Large Language Models (LLMs) continues to expand, leading to increasing interest in personalized LLMs that align with human values. However, aligning these models with individual values raises significant safety concerns, as certain values may correlate with harmful information. In this paper, we identify specific safety risks associated with value-aligned LLMs and investigate the psychological principles behind these challenges. Our findings reveal two key insights. (1) Value-aligned LLMs are more prone to harmful behavior compared to non-fine-tuned models and exhibit slightly higher risks in traditional safety evaluations than other fine-tuned models. (2) These safety issues arise because value-aligned LLMs genuinely generate text according to the aligned values, which can amplify harmful outcomes. Using a dataset with detailed safety categories, we find significant correlations between value alignment and safety risks, supported by psychological hypotheses. This study offers insights into the "black box" of value alignment and proposes in-context alignment methods to enhance the safety of value-aligned LLMs.

Unintended Harms of Value-Aligned LLMs: Psychological and Empirical Insights

Moral reasoning is a complex cognitive process shaped by individual experiences and cultural contexts and presents unique challenges for computational analysis. While natural language processing (NLP) offers promising tools for studying this phenomenon, current research lacks cohesion, employing discordant datasets and tasks that examine isolated aspects of moral reasoning. We bridge this gap with UniMoral, a unified dataset integrating psychologically grounded and social-media-derived moral dilemmas annotated with labels for action choices, ethical principles, contributing factors, and consequences, alongside annotators’ moral and cultural profiles. Recognizing the cultural relativity of moral reasoning, UniMoral spans six languages, Arabic, Chinese, English, Hindi, Russian, and Spanish, capturing diverse socio-cultural contexts. We demonstrate UniMoral's utility through a benchmark evaluations of three large language models (LLMs) across four tasks: action prediction, moral typology classification, factor attribution analysis, and consequence generation. Key findings reveal that while implicitly embedded moral contexts enhance the moral reasoning capability of LLMs, there remains a critical need for increasingly specialized approaches to further advance moral reasoning in these models.

Are Rules Meant to be Broken? Understanding Multilingual Moral Reasoning as a Computational Pipeline with UniMoral

Algorithmic fairness has conventionally adopted the mathematically convenient perspective of racial color-blindness (i.e., difference unaware treatment). However, we contend that in a range of important settings, group difference awareness matters. For example, differentiating between groups may be necessary in legal contexts (e.g., the U.S. compulsory draft applies to men but not women) and harm assessments (e.g., referring to girls as ``terrorists'' may be less harmful than referring to Muslim people as such). Thus, in contrast to most fairness work, we study fairness through the perspective of treating people differently --- when it is contextually appropriate to. We first introduce an important distinction between descriptive (fact-based), normative (value-based), and correlation (association-based) benchmarks. This distinction is significant because each category requires separate interpretation and mitigation tailored to its specific characteristics. Then, we present a benchmark suite composed of eight different scenarios for a total of 16k questions that enables us to assess difference awareness. Finally, we show results across ten models that demonstrate difference awareness is a distinct dimension to fairness where existing bias mitigation strategies may backfire.

Fairness through Difference Awareness: Measuring Desired Group Discrimination in LLMs

Socioeconomic status (SES) fundamentally influences how people interact with each other and, more recently, with digital technologies like large language models (LLMs). While previous research has highlighted the interaction between SES and language technology, it was limited by reliance on proxy metrics and synthetic data. We survey 1,000 individuals from 'diverse socioeconomic backgrounds' about their use of language technologies and generative AI, and collect 6,482 prompts from their previous interactions with LLMs. We find systematic differences across SES groups in language technology usage (i.e., frequency, performed tasks), interaction styles, and topics. Higher SES entail a higher level of abstraction, convey requests more concisely, and topics like 'inclusivity' and 'travel'. Lower SES correlates with higher anthropomorphization of LLMs (using ''hello'' and ''thank you'') and more concrete language. Our findings suggest that while generative language technologies are becoming more accessible to everyone, socioeconomic linguistic differences still stratify their use to create a digital divide. These differences underscore the importance of considering SES in developing language technologies to accommodate varying linguistic needs rooted in socioeconomic factors and limit the AI Gap across SES groups.

The AI Gap: How Socioeconomic Status Affects Language Technology Interactions

The field of cultural NLP has recently experienced rapid growth, driven by a pressing need to ensure that language technologies are effective and safe across a pluralistic user base. This work has largely progressed without a shared conception of culture, instead choosing to rely on a wide array of cultural proxies. However, this leads to a number of recurring limitations: coarse national boundaries fail to capture nuanced differences that lay within them, limited coverage restricts datasets to only a subset of usually highly-represented cultures, and a lack of dynamicity results in static cultural benchmarks that do not change as culture evolves. In this position paper, we argue that these methodological limitations are symptomatic of a theoretical gap. We draw on a well-developed theory of culture from sociocultural linguistics to fill this gap by 1) demonstrating in a case study how it can clarify methodological constraints and affordances, 2) offering theoretically-motivated paths forward to achieving cultural competence, and 3) arguing that localization is a more useful framing for the goals of much current work in cultural NLP.

Culture is Not Trivia: Sociocultural Theory for Cultural NLP

As large language models (LLMs) become increasingly integrated into daily life, ensuring their cultural sensitivity and inclusivity is paramount. We introduce PALM, a year-long community-driven project covering all 22 Arab countries. The dataset contains instruction–response pairs in both Modern Standard Arabic (MSA) and dialectal Arabic (DA), spanning 20 diverse topics. Built by a team of 44 researchers across the Arab world—each an author of this paper—PALM offers a broad, inclusive perspective. We use PALM to evaluate the cultural and dialectal capabilities of several frontier LLMs, revealing notable limitations: while closed-source LLMs generally perform strongly, they still exhibit flaws, and smaller open-source models face greater challenges. Furthermore, certain countries (e.g., Egypt, the UAE) appear better represented than others (e.g., Iraq, Mauritania, Yemen). Our annotation guidelines, code, and data are publicly available for reproducibility. More information about PALM is available on our project page: https://github.com/UBC-NLP/palm.

Palm: A Culturally Inclusive and Linguistically Diverse Dataset for Arabic LLMs

We observe a novel phenomenon, *contextual entrainment*, across a wide range of language models (LMs) and prompt settings, providing a new mechanistic perspective on how LMs become distracted by "irrelevant" contextual information in the input prompt. Specifically, LMs assign significantly higher logits (or probabilities) to any tokens that have previously appeared in the context prompt, even for random tokens. This suggests that contextual entrainment is a \emph{mechanistic} phenomenon, occurring independently of the relevance or semantic relation of the tokens to the question or the rest of the sentence. We find statistically significant evidence that the magnitude of contextual entrainment is influenced by semantic factors. Counterfactual prompts have a greater effect compared to factual ones, suggesting that while contextual entrainment is a mechanistic phenomenon, it is modulated by semantic factors.

We hypothesise that there is a circuit of attention heads --- the *entrainment heads* --- that corresponds to the contextual entrainment phenomenon. Using a novel entrainment head discovery method based on differentiable masking, we identify these heads across various settings. When we "turn off" these heads, i.e., set their outputs to zero, the effect of contextual entrainment is significantly attenuated, causing the model to generate output that capitulates to what it would produce if no distracting context were provided. Our discovery of contextual entrainment, along with our investigation into LM distraction via the entrainment heads, marks a key step towards the mechanistic analysis and mitigation of the distraction problem.

Llama See, Llama Do: A Mechanistic Perspective on Contextual Entrainment and Distraction in LLMs

Scaling law builds the relationship between training computation and validation loss, enabling researchers to effectively predict the loss trending of models across different levels of computation. However, a gap still remains between validation loss and the model's downstream capabilities, making it untrivial to apply scaling law to direct performance prediction for downstream tasks. The loss typically represents a cumulative penalty for predicted tokens, which are implicitly considered to have equal importance. Nevertheless, our studies have shown evidence that when considering different training data distributions, we cannot directly model the relationship between downstream capability and computation or token loss. To bridge the gap between validation loss and downstream task capabilities, in this work, we introduce Capability Salience Vector, which decomposes the overall loss and assigns different importance weights to tokens to assess a specific meta-capability, aligning the validation loss with downstream task performance in terms of the model’s capabilities. Experiments on various popular benchmarks demonstrate that our proposed Capability Salience Vector could significantly improve the predictability of language model performance on downstream tasks.

Capability Salience Vector: Fine-grained Alignment of Loss and Capabilities for Downstream Task Scaling Law

Inductive biases are inherent in every machine learning system, shaping how models generalize from finite data. In the case of neural language models (LMs), debates persist as to whether these biases align with or diverge from human processing constraints. To address this issue, we propose a quantitative framework that allows for controlled investigations into the nature of these biases. Within our framework, we introduce 
$m$-local entropy—an information-theoretic measure derived from average lossy-context surprisal—that captures the local uncertainty of a language by quantifying how effectively the preceding symbols disambiguate the next symbol. In experiments on both perturbed natural language corpora and languages defined by probabilistic finite-state automata (PFSA), we show that languages with higher $m$-local entropy are more difficult for Transformer and LSTM LMs to learn. These results suggest that neural LMs, much like humans, are highly sensitive to the local statistical structure of a language.

Downloads

Next from ACL 2025

MPO: Multilingual Safety Alignment via Reward Gap Optimization

Similar lecture

Molecular Dynamics MOOC 12.1.3. Elasticity