Austria

Large Language Models (LLMs) are known to be vulnerable to backdoor attacks, where triggers embedded in poisoned samples can maliciously alter LLMs&#39; behaviors. In this paper, we move beyond attacking LLMs and instead examine backdoor attacks through the novel lens of natural language explanations. Specifically, we leverage LLMs’ generative capabilities to produce human-readable explanations for their decisions, enabling direct comparisons between explanations for clean and poisoned samples. Our results show that backdoored models produce coherent explanations for clean inputs but diverse and logically flawed explanations for poisoned data, a pattern consistent across classification and generation tasks for different backdoor attacks. Further analysis reveals key insights into the explanation generation process. At the token level, explanation tokens associated with poisoned samples only appear in the final few transformer layers. At the sentence level, attention dynamics indicate that poisoned inputs shift attention away from the original input context during explanation generation. These findings enhance our understanding of backdoor mechanisms in LLMs and present a promising framework for detecting vulnerabilities through explainability.

ACL 2025

When Backdoors Speak: Understanding LLM Backdoor Attacks Through Model-Generated Explanations

llm explainability

backdoor attacks

model interpretability

**Welcome to EMNLP 2023!**

On behalf of the EMNLP 2023 Organizing Committee, I extend a warm and heartfelt welcome to all of you to the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP). It is with immense pleasure and excitement that we gather here in Singapore, a vibrant hub of innovation and technological advancement.

The conference program is packed with insightful presentations, thought-provoking workshops, and engaging networking opportunities. In addition to the technical sessions, we have also planned several social events that will provide you with the opportunity to connect with your colleagues.

The conference is held in person at the Resorts World Convention Centre in Singapore, and available on-line with the help of Underline.

For those in Singapore, I hope that you will find time to explore the exciting Sentosa Island, Gardens by the Bay, Singapore Botanic Gardens and the many other attractions unique to Singapore.

Of course, we are grateful to our sponsors and partners for their generous support of EMNLP 2023, whose contributions make it possible for us to host this world-class event.

Yuji Matsumoto (RIKEN AIP) 
EMNLP 2023 General Chair

To access the **EMNLP 2023** event page on Underline, you need to register for the Conference. 
Please follow **[this link](https://2023.emnlp.org/registration/)** for more details.

EMNLP 2023

EMNLP 2023 took place in Singapore from Dec 6th to Dec 10th, 2023.

Large Language Models (LLMs) are known to be vulnerable to backdoor attacks, where triggers embedded in poisoned samples can maliciously alter LLMs' behaviors. In this paper, we move beyond attacking LLMs and instead examine backdoor attacks through the novel lens of natural language explanations. Specifically, we leverage LLMs’ generative capabilities to produce human-readable explanations for their decisions, enabling direct comparisons between explanations for clean and poisoned samples. Our results show that backdoored models produce coherent explanations for clean inputs but diverse and logically flawed explanations for poisoned data, a pattern consistent across classification and generation tasks for different backdoor attacks. Further analysis reveals key insights into the explanation generation process. At the token level, explanation tokens associated with poisoned samples only appear in the final few transformer layers. At the sentence level, attention dynamics indicate that poisoned inputs shift attention away from the original input context during explanation generation. These findings enhance our understanding of backdoor mechanisms in LLMs and present a promising framework for detecting vulnerabilities through explainability.

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)

Human-like personality traits have recently been discovered in large language models, raising the hypothesis that their (known and as yet undiscovered) biases conform with human latent psychological constructs. While large conversational models may be tricked into answering psychometric questionnaires, the latent psychological constructs of thousands of simpler transformers, trained for other tasks, cannot be assessed because appropriate psychometric methods are currently lacking. Here, we show how standard psychological questionnaires can be reformulated into natural language inference prompts, and we provide a code library to support the psychometric assessment of arbitrary models. We demonstrate, using a sample of 88 publicly available models, the existence of human-like mental health-related constructs—including anxiety, depression, and the sense of coherence—which conform with standard theories in human psychology and show similar correlations and mitigation strategies. The ability to interpret and rectify the performance of language models by using psychological tools can boost the development of more explainable, controllable, and trustworthy models.

Assessment and manipulation of latent constructs in pre-trained language models using psychometric scales

The generation of toxic content by large language models (LLMs) remains a critical challenge for the safe deployment of language technology. We propose a novel framework for implicit knowledge editing and controlled text generation by fine-tuning LLMs with a prototype-based contrastive perplexity objective. Central to our method is the construction of hard negatives—toxic outputs that are generated through adversarial paraphrasing to be semantically similar and model probability to their non-toxic counterparts. By training on these challenging and realistic pairs, our approach ensures robust and stable contrastive optimization. Experimental results in the domain of detoxification demonstrate that our method significantly reduces toxic generation while maintaining strong performance on downstream tasks such as commonsense reasoning and reading comprehension. Our findings highlight the effectiveness of exploiting hard negatives for attribute-aware fine-tuning.

Contrastive Perplexity for Controlled Generation: An Application in Detoxifying Large Language Models

Semantic parsing, which converts natural language queries into logic forms, plays a crucial role in reasoning within structured environments. However, existing methods encounter two significant challenges: reliance on extensive manually annotated datasets and limited generalization capability to unseen examples. To tackle these issues, we propose Targeted Synthetic Data Generation (Targa), a practical framework that dynamically generates high-relevance synthetic data without manual annotation. Starting from the pertinent entity and relation of a given question, we probe for the potential relevant queries through layer-wise expansion and cross-layer combination. Then, we generate corresponding natural language questions for these constructed queries to jointly serve as the synthetic demonstration for in-context learning. Experiments on multiple knowledge-based question answering (KBQA) datasets demonstrate that Targa, using only a 7B-parameter model, substantially outperforms existing non-fine-tuned methods that utilize close-sourced model, achieving notable improvements in F1 scores on GrailQA(+7.7) and KBQA-Agent(+12.2). Furthermore, Targa also exhibits superior sample efficiency, robustness, and generalization capabilities under non-I.I.D. settings.

TARGA: Targeted Synthetic Data Generation for Practical Reasoning over Structured Data

Large language models (LLMs) have achieved remarkable proficiency in understanding and generating human natural languages, mainly owing to the "scaling law" that optimizes relationships among language modeling loss, model parameters, and pre-trained tokens. However, with the exhaustion of high-quality internet corpora and increasing computational demands, the sustainability of pre-training scaling needs to be addressed. This paper presents a comprehensive survey of post-training scaling, an emergent paradigm aiming to relieve the limitations of traditional pre-training by focusing on the alignment phase, which traditionally accounts for a minor fraction of the total training computation. Our survey categorizes post-training scaling into three key methodologies: Supervised Fine-tuning (SFT), Reinforcement Learning from Feedback (RLxF), and Test-time Compute (TTC). We provide an in-depth analysis of the motivation behind post-training scaling, the scalable variants of these methodologies, and a comparative discussion against traditional approaches. By examining the latest advancements, identifying promising application scenarios, and highlighting unresolved issues, we seek a coherent understanding and map future research trajectories in the landscape of post-training scaling for LLMs.

A Survey of Post-Training Scaling in Large Language Models

With the ubiquity of hyper-relational facts in modern Knowledge Graphs (KGs), existing link prediction techniques mostly focus on learning the sophisticated relationships among multiple entities and relations contained in a fact, while ignoring the multimodal information, which often provides additional clues to boost link prediction performance. Nevertheless, traditional multimodel fusion approaches, which are mainly designed for triple facts under either entity-centric or relation-guided fusion schemes, fail to integrate the multimodal information with the rich context of the hyper-relational fact consisting of multiple entities and relations. Against this background, we propose **HyperFM**, a **Hyper**-relational **F**act-centric **M**ultimodal Fusion technique. It effectively captures the intricate interactions between different data modalities while accommodating the hyper-relational structure of the KG in a fact-centric manner via a customized Hypergraph Transformer. We evaluate HyperFM against a sizeable collection of baselines in link prediction tasks on two real-world KG datasets. Results show that HyperFM consistently achieves the best performance, yielding an average improvement of 6.0-6.8% over the best-performing baselines on the two datasets. Moreover, a series of ablation studies systematically validate our fact-centric fusion scheme.

HyperFM: Fact-Centric Multimodal Fusion for Link Prediction over Hyper-Relational Knowledge Graphs

Arabic has diverse dialects, where one dialect can be substantially different from the others. In the NLP literature, some assumptions about these dialects are widely adopted (e.g., ``Arabic dialects can be grouped into distinguishable regional dialects'') and are manifested in different computational tasks such as Arabic Dialect Identification (ADI). However, these assumptions are not quantitatively verified. We identify four of these assumptions and examine them by extending and analyzing a multi-label dataset, where the validity of each sentence in 11 different country-level dialects is manually assessed by speakers of these dialects. Our analysis indicates that the four assumptions oversimplify reality, and some of them are not always accurate. This in turn might be hindering further progress in different Arabic NLP tasks.

Revisiting Common Assumptions about Arabic Dialects in NLP

Structured representations, exemplified by Abstract Meaning Representation~(AMR), have long been pivotal in computational linguistics. However, their role remains ambiguous in the Large Language Models (LLMs) era. Initial attempts to integrate structured representation into LLMs via a zero-shot setting yielded inferior performance. We hypothesize that such a decline stems from the structure information being passed into LLMs in a code format unfamiliar to LLMs' training corpora. Consequently, we propose SR-LLM, an innovative framework with two settings to explore a superior way of integrating structured representation with LLMs from training-free and training-dependent perspectives. The former integrates structural information through natural language descriptions in LLM prompts, whereas its counterpart augments the model's inference capability through fine-tuning on linguistically described structured representations. Performance improvements were observed in widely downstream datasets, with particularly notable gains of 3.17% and 12.38% in PAWS. To the best of our knowledge, this work represents the pioneering demonstration that leveraging structural representations can substantially enhance LLMs' inference capability. We hope that our work sheds light and encourages future research to enhance the reasoning and interoperability of LLMs by structure data.

SR-LLM: Rethinking the Structured Representation in Large Language Model

Extracting sentence embeddings from large language models (LLMs) is a practical direction, as it requires neither additional data nor fine-tuning. Previous studies usually focus on prompt engineering to guide LLMs to encode the core semantic information of the sentence into the embedding of the last token. However, the last token in these methods still encodes an excess of non-essential information, such as stop words, limiting its encoding capacity. To this end, we propose a Contrastive Prompting (CP) technique that introduces an extra auxiliary prompt to elicit better sentence embedding. By contrasting with the auxiliary prompt, CP can steer existing prompts to encode the core semantics of the sentence, rather than non-essential information. CP is a plug-and-play inference-time intervention method that can be combined with various prompt-based methods. Extensive experiments on Semantic Textual Similarity (STS) tasks and downstream classification tasks demonstrate that our method can improve the performance of existing prompt-based methods across different LLMs.

Contrastive Prompting Enhances Sentence Embeddings in LLMs through Inference-Time Steering

Knowledge distillation (KD) aims to transfer knowledge from a large teacher model to a smaller student model. Previous work applying KD in the field of large language models (LLMs) typically focused on the post-training phase, where the student LLM learns directly from instructions and corresponding responses generated by the teacher model. In this paper, we extend KD to the pre-training phase of LLMs, named pre-training distillation (PD). We first conduct a preliminary experiment using GLM-4-9B as the teacher LLM to distill a 1.9B parameter student LLM, validating the effectiveness of PD. Considering the key impact factors of distillation, we systematically explore the design space of pre-training distillation across four aspects: logits processing, loss selection, scaling law, and offline or online logits. We conduct extensive experiments to explore the design space of pre-training distillation and find better configurations and interesting conclusions, such as larger student LLMs generally benefiting more from pre-training distillation, while a larger teacher LLM does not necessarily guarantee better results. We hope our exploration of the design space will inform future practices in pre-training distillation.

Pre-training Distillation for Large Language Models: A Design Space Exploration

Understanding the inner workings of Large Language Models (LLMs) is a critical research frontier. Prior research has shown that a single LLM's concept representations can be captured as steering vectors (SVs), enabling the control of LLM behavior (e.g., towards generating harmful content). Our work takes a novel approach by exploring the intricate relationships between concept representations across different LLMs, drawing an intriguing parallel to Plato's Allegory of the Cave. In particular, we introduce a linear transformation method to bridge these representations and present three key findings: 1) Concept representations across different LLMs can be effectively aligned using simple linear transformations, enabling efficient cross-model transfer and behavioral control via SVs. 2) This linear transformation generalizes across concepts, facilitating alignment and control of SVs representing different concepts across LLMs. 3) A weak-to-strong transferability exists between LLM concept representations, whereby SVs extracted from smaller LLMs can effectively control the behavior of larger LLMs. Our code is provided in the supplementary file and will be openly released.

Downloads

Next from ACL 2025

Assessment and manipulation of latent constructs in pre-trained language models using psychometric scales

Similar lecture

Prompt as Triggers for Backdoor Attack: Examining the Vulnerability in Language Models