United States

Weight-based model editing methods update the parametric knowledge of language models post-training. However, these methods can unintentionally alter unrelated parametric knowledge representations, potentially increasing the risk of harm. In this work, we investigate how weight editing methods unexpectedly amplify model biases after edits. We introduce a novel benchmark dataset, Seesaw-CF, for measuring bias amplification of model editing methods for demographic traits such as race, geographic origin, and gender. We use Seesaw-CF to examine the impact of model editing on bias in five large language models. Our results demonstrate that edited models exhibit, to various degrees, more biased behavior for certain demographic groups than before they were edited, specifically becoming less confident in properties for Asian and African subjects. Additionally, editing facts about place of birth, country of citizenship, or gender has particularly negative effects on the model&#39;s knowledge about unrelated properties, such as field of work, a pattern observed across multiple models.

EMNLP 2024

“Flex Tape Can’t Fix That”: Bias and Misinformation in Edited Language Models

alignment

language modeling

fine-tuning

reinforcement learning

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

ACL 2024

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

Weight-based model editing methods update the parametric knowledge of language models post-training. However, these methods can unintentionally alter unrelated parametric knowledge representations, potentially increasing the risk of harm. In this work, we investigate how weight editing methods unexpectedly amplify model biases after edits. We introduce a novel benchmark dataset, Seesaw-CF, for measuring bias amplification of model editing methods for demographic traits such as race, geographic origin, and gender. We use Seesaw-CF to examine the impact of model editing on bias in five large language models. Our results demonstrate that edited models exhibit, to various degrees, more biased behavior for certain demographic groups than before they were edited, specifically becoming less confident in properties for Asian and African subjects. Additionally, editing facts about place of birth, country of citizenship, or gender has particularly negative effects on the model's knowledge about unrelated properties, such as field of work, a pattern observed across multiple models.

poster

## Welcome to EMNLP 2024! 
We are excited to welcome you to one of the most prominent conferences in the field of Natural Language Processing. This year, EMNLP 2024 is being held in a hybrid format,
offering both virtual and in-person participation in beautiful Miami. Due to a record-breaking number of submissions, we've expanded the total number of accepted papers to accommodate more cutting-edge research from around the globe.
### [Conference Handbook](https://drive.google.com/file/d/1WPROgxjLAC96AJL7Ugy0tEnYm7dkrbHt/view?usp=sharing)

You are required to register for this event. **Please register [here](https://2024.emnlp.org/registration/).** The EMNLP 2024 event page on Underline will be open to public one week prior to the event.

EMNLP 2024 will take place in Miami, Florida from Nov 12th to Nov 16th, 2024, at the Hyatt Regency Miami Hote and on Underline for remote participants.

Pre-trained language models acquire knowledge from vast amounts of text data, which can inadvertently contain sensitive information. To mitigate the presence of undesirable knowledge, the task of knowledge unlearning becomes crucial for language models. Previous research relies on gradient ascent methods to achieve knowledge unlearning, which is simple and effective. However, this approach calculates all the gradients of tokens in the sequence, potentially compromising the general ability of language models. To overcome this limitation, we propose an adaptive objective that calculates gradients with fine-grained control specifically targeting sensitive tokens. Our adaptive objective is pluggable, ensuring simplicity and enabling extension to the regularization-based framework that utilizes non-target data or other models to preserve general ability. Through extensive experiments targeting the removal of typical sensitive data, we demonstrate that our proposed method enhances the general ability of language models while achieving knowledge unlearning. Additionally, it demonstrates the capability to adapt to behavior alignment, eliminating all the undesirable knowledge within a specific domain.

Fine-grained Pluggable Gradient Ascent for Knowledge Unlearning in Language Models

Large language models (LLMs) have shown promise as automated evaluators for assessing the quality of answers generated by AI systems. However, LLM-based evaluators exhibit position bias, or inconsistency, when used to evaluate candidate answers in pairwise comparisons, favoring either the first or second answer regardless of content. To address this limitation, we propose PORTIA, an alignment-based system designed to mimic human comparison strategies to calibrate position bias in a lightweight yet effective manner. Specifically, PORTIA splits the answers into multiple segments, taking into account both length and semantics, and merges them back into a single prompt for evaluation by LLMs. Extensive experiments with six LLMs on 11,520 answer pairs demonstrate that PORTIA markedly enhances the consistency rates for all models and forms of comparison tested, achieving an average relative improvement of 47.46%. It also enables PORTIA-enhanced GPT-3.5 to achieve agreement rates with humans comparable to GPT-4 and elevates GPT-4's consistency rate up to 98%. Subsequent human evaluations indicate that the PORTIA-enhanced GPT-3.5 model can even surpass standalone GPT-4 in terms of alignment with human evaluators, highlighting PORTIA's ability to correct position bias, improve LLM consistency, and boost performance while keeping cost efficiency.

Split and Merge: Aligning Position Biases in LLM-based Evaluators

Keyphrase generation (KPG) aims to automatically generate a collection of phrases representing the core concepts of a given document. The dominant paradigms in KPG include one2seq and one2set. Recently, there has been increasing interest in applying large language models (LLMs) to KPG. Our preliminary experiments reveal that it is challenging for a single model to excel in both recall and precision. Further analysis shows that: 1) the one2set paradigm owns the advantage of high recall, but suffers from improper assignments of supervision signals during training; 2) LLMs are powerful in keyphrase selection, but existing selection methods often make redundant selections. Given these observations, we introduce a generate-then-select framework decomposing KPG into two steps, where we adopt a one2set-based model as generator to produce candidates and then use an LLM as selector to select keyphrases from these candidates. Particularly, we make two important improvements on our generator and selector: 1) we design an Optimal Transport-based assignment strategy to address the above improper assignments; 2) we model the keyphrase selection as a sequence labeling task to alleviate redundant selections. Experimental results on multiple benchmark datasets show that our framework significantly surpasses state-of-the-art models, especially in absent keyphrase prediction.

One2Set + Large Language Model: Best Partners for Keyphrase Generation

Continual Event Detection (CED) poses a formidable challenge due to the catastrophic forgetting phenomenon, where learning new tasks (with new coming event types) hampers performance on previous ones. In this paper, we introduce a novel approach, Lifelong Event Detection via Optimal Transport (**LEDOT**), that leverages optimal transport principles to align the optimization of our classification module with the intrinsic nature of each class, as defined by their pre-trained language modeling. Our method integrates replay sets, prototype latent representations, and an innovative Optimal Transport component. Extensive experiments on MAVEN and ACE datasets demonstrate LEDOT's superior performance, consistently outperforming state-of-the-art baselines. The results underscore LEDOT as a pioneering solution in continual event detection, offering a more effective and nuanced approach to addressing catastrophic forgetting in evolving environments.

Lifelong Event Detection via Optimal Transport

Activation Editing, which involves directly editting the internal representations of large language models (LLMs) to alter their behavior and achieve desired properties, has emerged as a promising area of research. Existing works primarily treat LLMs' activations as points in space and modify them by adding steering vectors. We show that doing so would break the magnitude consistency of the activation vectors in LLMs. To overcome this shortcoming, we propose a novel editing method that views activations in terms of their directions and magnitudes. Our method, which we name Householder Pseudo-Rotation (HPR), mimics the rotation transformation, thus preserving activation norm and resulting in an improved performance on various safety benchmarks.

Householder Pseudo-Rotation: A Novel Approach to Activation Editing in LLMs with Direction-Magnitude Perspective

Fallacies are defective arguments with faulty reasoning. Detecting and classifying them is a crucial NLP task to prevent misinformation, manipulative claims, and biased decisions. However, existing fallacy classifiers are limited by the requirement for sufficient labeled data for training, which hinders their out-of-distribution (OOD) generalization abilities. In this paper, we focus on leveraging Large Language Models (LLMs) for zero-shot fallacy classification. To elicit fallacy-related knowledge and reasoning abilities of LLMs, we propose diverse single-round and multi-round prompting schemes, applying different taskspecific instructions such as extraction, summarization, and Chain-of-Thought reasoning. With comprehensive experiments on benchmark datasets, we suggest that LLMs could be potential zero-shot fallacy classifiers. In general, LLMs under single-round prompting schemes have achieved acceptable zeroshot performances compared to the best fullshot baselines and can outperform them in all OOD inference scenarios and some opendomain tasks. Our novel multi-round prompting schemes can effectively bring about more improvements, especially for small LLMs. Our analysis further underlines the future research on zero-shot fallacy classification. Codes and data are available at: https://github.com/panFJCharlotte98/Fallacy_Detection.

Are LLMs Good Zero-Shot Fallacy Classifiers?

Large language model (LLM) training and finetuning are often bottlenecked by limited GPU memory. While existing projection-based optimization methods address this by projecting gradients into a lower-dimensional subspace to reduce optimizer state memory, they typically rely on $\textit{dense}$ projection matrices, which can introduce computational and memory overheads. In this work, we propose Grass (GRAdient Stuctured Sparsification), a novel approach that leverages $\textit{sparse}$ projections to transform gradients into structured sparse updates. This design not only significantly reduces memory usage for optimizer states but also minimizes gradient memory footprint, computation, and communication costs, leading to substantial throughput improvements. Extensive experiments on pretraining and finetuning tasks demonstrate that Grass achieves comparable performance to full-rank training and existing projection-based methods. Notably, Grass enables half-precision pretraining of a 13B parameter LLaMA model on a single 40GB A100 GPU---a feat infeasible for previous methods---and yields up to a $2\times$ throughput improvement on an 8-GPU system.

GRASS: Compute Efficient Low-Memory LLM Training with Structured Sparse Gradients

Knowledge graph completion (KGC) aims to infer missing or incomplete parts in knowledge graph. The existing models are generally divided into structure-based and description-based models, among description-based models often require longer training and inference times as well as increased memory usage. In this paper, we propose Pre-Encoded Masked Language Model (PEMLM) to efficiently solve KGC problem. By encoding textual descriptions into semantic representations before training, the necessary resources are significantly reduced. Furthermore, we introduce a straightforward but effective fusion framework to integrate structural embedding with pre-encoded semantic description, which enhances the model's prediction performance on 1-N relations. The experimental results demonstrate that our proposed strategy attains state-of-the-art performance on the WN18RR (MRR+5.4% and Hits@1+6.4%) and UMLS datasets. Compared to existing models, we have increased inference speed by 30x and reduced training memory by approximately 60%.

Joint Pre-Encoding Representation and Structure Embedding for Efficient and Low-Resource Knowledge Graph Completion

Large language models (LLMs) have revolutionized the domain of natural language processing because of their excellent performance on various tasks. Despite their impressive capabilities, LLMs also have the potential to generate texts that pose risks of misuse. Consequently, detecting LLM-generated text has become increasingly important.
Previous LLM-generated text detection methods use semantic features, which are stored in the last layer. This leads to methods that overfit the training set domain and exhibit shortcomings in generalization. Therefore, We argue that utilizing intrinsic features rather than semantic features for detection results in better performance.
In this work, we design Text Fluoroscopy, a black-box method with better generalizability for detecting LLM-generated text by mining the intrinsic features of the text to be detected. 
Our method captures the text's intrinsic features by identifying the layer with the largest distribution difference from the last and first layers when projected to the vocabulary space.
Our method achieves 7.36\% and 2.84\% average improvement in detection performance compared to the baselines in detecting texts from different domains generated by GPT-4 and Claude3, respectively.

Text Fluoroscopy: Detecting LLM-Generated Text through Intrinsic Features

Text summarization, a key natural language generation (NLG) task, is vital in various domains. However, the high cost of inaccurate summaries in risk-critical applications, particularly those involving human-in-the-loop decision-making, raises concerns about the reliability of uncertainty estimation on text summarization (UE-TS) evaluation methods. This concern stems from the dependency of uncertainty model metrics on diverse and potentially conflicting NLG metrics. To address this issue, we introduce a comprehensive UE-TS benchmark incorporating 31 NLG metrics across four dimensions. The benchmark evaluates the uncertainty estimation capabilities of two large language models and one pre-trained language model on three datasets, with human-annotation analysis incorporated where applicable. We also assess the performance of 14 common uncertainty estimation methods within this benchmark. Our findings emphasize the importance of considering multiple uncorrelated NLG metrics and diverse uncertainty estimation methods to ensure reliable and efficient evaluation of UE-TS techniques. Our code and data are available: https://github.com/he159ok/Benchmark-of-Uncertainty-Estimation-Methods-in-Text-Summarization.

Downloads

Next from EMNLP 2024

Fine-grained Pluggable Gradient Ascent for Knowledge Unlearning in Language Models

Similar lecture

Model Editing at Scale leads to Gradual and Catastrophic Forgetting