ACL 2024 Papers

Recent studies demonstrated that large language models (LLMs) can excel in many tasks via in-context learning (ICL). However, recentworks show that ICL-prompted models tend to produce inaccurate results when presented with adversarial inputs. In this work, we investigate whether augmenting ICL with natural language explanations (NLEs) improves the robustness of LLMs on adversarial datasets covering natural language inference and paraphrasing identification. We prompt LLMs with a small set of human-generated NLEs to produce further NLEs, yielding more accurate results than both a zero-shot-ICL setting and using only human-generated NLEs. Our results on five popular LLMs (GPT3.5-turbo, Llama2, Vicuna, Zephyr, and Mistral) show that our approach yields over 6% improvement over baseline approaches for eight adversarial datasets: HANS, ISCS, NaN, ST, PICD, PISP, ANLI, and PAWS. Furthermore, previous studies have demonstrated that prompt selection strategies significantly enhance ICL on in-distribution test sets. However, our findings reveal that these strategies do not match the efficacy of our approach for robustness evaluations, resulting in an accuracy drop of 8% compared to the proposed approach.

Social media has emerged as a cornerstone of social movements, wielding significant influence in driving societal change. Simulating the response of the public and forecasting the potential impact has become increasingly important. However, existing methods for simulating such phenomena encounter challenges concerning their efficacy and efficiency in capturing the behaviors of social movement participants. In this paper, we introduce a hybrid framework for social media user simulation, wherein users are categorized into two types. Core users are driven by Large Language Models, while numerous ordinary users are modeled by deductive agent-based models. We further construct a Twitter-like environment to replicate their response dynamics following trigger events. Subsequently, we develop a multi-faceted benchmark SoMoSiMu-Bench for evaluation and conduct comprehensive experiments across real-world datasets. Experimental results demonstrate the effectiveness and flexibility of our method.

Data contamination has garnered increased attention in the era of Large language models (LLMs) due to the reliance on extensive internet-derived training corpora. The issue of training corpus overlap with evaluation benchmarks—referred to as contamination—has been the focus of significant recent research. This body of work aims to identify contamination, understand its impacts, and explore mitigation strategies from diverse perspectives. However, comprehensive studies that provide a clear pathway from foundational concepts to advanced insights are lacking in this nascent field. Therefore, we present the first survey in the field of data contamination. We begin by examining the effects of data contamination across various stages and forms. We then provide a detailed analysis of current contamination detection methods, categorizing them to highlight their focus, assumptions, strengths, and limitations. We also discuss mitigation strategies, offering a clear guide for future research. This survey serves as a succinct overview of the most recent advancements in data contamination research, providing a straightforward guide for the benefit of future research endeavors.

Block Diagrams play an essential role in visualizing the relationships between components or systems. Generating summaries of block diagrams is important for document understanding or question answering (QA) tasks by providing concise overviews of complex systems. However, it’s a challenging task as it requires compressing complex relationships into informative descriptions. In this paper, we present “BlockNet”, a fusion framework that summarizes block diagrams by integrating local and global information, catering to both English and Korean languages. Additionally, we introduce a new multilingual method to produce block diagram data, resulting in a high-quality dataset called “BD-EnKo”. In BlockNet, we develop “BlockSplit”, an Optical Character Recognition (OCR) based algorithm employing the divide-and-conquer principle for local information extraction. We train an OCR-free transformer architecture for global information extraction using BD-EnKo and public data. To assess the effectiveness of our model, we conduct thorough experiments on different datasets. The assessment shows that BlockNet surpasses all previous methods and models, including GPT-4V, for block diagram summarization.

Crafting an appealing heading is crucial for attracting readers and marketing work or products. A popular way is to summarize the main idea with a refined description and a memorable acronym. However, there lacks a systematic study and a formal benchmark including datasets and metrics. Motivated by this absence, we introduce LOgogram, a novel benchmark comprising 6,653 paper abstracts with corresponding descriptions and acronyms. To measure the quality of heading generation, we propose a set of evaluation metrics from three aspects: summarization, neology, and algorithm. Additionally, we explore three strategies for heading generation(generation ordering, tokenization of acronyms, and framework design) under various prevalent learning paradigms(supervised fine-tuning, in-context learning with Large Language Models(LLMs), and reinforcement learning) on our benchmark. Our experimental results indicate the difficulty in identifying a practice that excels across all summarization, neologistic, and algorithmic aspects.

The automatic evaluation of natural language generation (NLG) systems presents a long-lasting challenge. Recent studies have highlighted various neural metrics that align well with human evaluations. Yet, the robustness of these evaluators against adversarial perturbations remains largely under-explored due to the unique challenges in obtaining adversarial data for different NLG evaluation tasks. To address the problem, we introduce AdvEval, a novel black-box adversarial framework against NLG evaluators. AdvEval is specially tailored to generate data that yield strong disagreements between human and victim evaluators. Specifically, inspired by the recent success of large language models (LLMs) in text generation and evaluation, we adopt strong LLMs as both the data generator and gold evaluator. Adversarial data are automatically optimized with feedback from the gold and victim evaluator. We conduct experiments on 12 victim evaluators and 11 NLG datasets, spanning tasks including dialogue, summarization, and question evaluation. The results show that AdvEval can lead to significant performance degradation of various victim metrics, thereby validating its efficacy.

In this paper, we investigate the phenomena of “selection biases” in Large Language Models (LLMs), focusing on problems where models are tasked with choosing the optimal option from an ordered sequence. We delve into biases related to option order and token usage, which significantly impact LLMs’ decision-making processes. We also quantify the impact of these biases through an extensive empirical analysis across multiple models and tasks. Furthermore, we propose mitigation strategies to enhance model performance. Our key contributions are threefold: 1) Precisely quantifying the influence of option order and token on LLMs, 2) Developing strategies to mitigate the impact of token and order sensitivity to enhance robustness, and 3) Offering a detailed analysis of sensitivity across models and tasks, which informs the creation of more stable and reliable LLM applications for selection problems.

Short video fake news detection is crucial for combating the spread of misinformation. Current detection methods tend to aggregate features from individual modalities into multimodal features, overlooking the implicit opinions and the evolving nature of opinions across modalities. In this paper, we mine implicit opinions within short video news and promote the evolution of both explicit and implicit opinions across all modalities. Specifically, we design a prompt template to mine implicit opinions regarding the credibility of news from the textual component of videos. Additionally, we employ a diffusion model that encourages the interplay among diverse modal opinions, including those extracted through our implicit opinion prompts. Experimental results on a publicly available dataset for short video fake news detection demonstrate the superiority of our model over state-of-the-art methods.

Large Language Models (LLMs) have demonstrated considerable cross-lingual alignment and generalization ability. Current research primarily focuses on improving LLMs’ cross-lingual generalization capabilities. However, there is still a lack of research on the intrinsic mechanisms of how LLMs achieve cross-lingual alignment. From the perspective of region partitioning, this paper conducts several investigations on the linguistic competence of LLMs. We discover a core region in LLMs that corresponds to linguistic competence, accounting for approximately 1% of the total model parameters. Removing this core region by setting parameters to zero results in a significant performance decrease across 30 different languages. Furthermore, this core region exhibits significant dimensional dependence, perturbations to even a single parameter on specific dimensions leading to a loss of linguistic competence. Moreover, we discover that distinct monolingual regions exist for different languages, and disruption to these specific regions substantially reduces the LLMs’ proficiency in those corresponding languages. Our research also indicates that freezing the core linguistic region during further pre-training can mitigate the issue of catastrophic forgetting (CF), a common phenomenon observed during further pre-training of LLMs. Overall, exploring the LLMs’ functional regions provides insights into the foundation of their intelligence.

Many recent studies endeavor to improve open-sourced language models through imitation learning, re-training on the synthetic instruction data from state-of-the-art proprietary models like ChatGPT and GPT-4.However, the innate nature of synthetic data inherently contains noisy data, giving rise to a substantial presence of low-quality data replete with misleading queries, erroneous responses, and flawed reasoning.Although we intuitively grasp the potential harm of noisy data, we lack a quantitative understanding of its impact.To this end, this paper explores correlation between the degree of noise and its impact on language models through instruction tuning.We first introduce the Falsity-Controllable () dataset, which comprises pairs of true answers and corresponding reasoning, as well as false pairs to manually control the factuality ratio of the dataset.Through our extensive experiments, we found multiple intriguing findings of the correlation between factuality and instruction tuning. Specifically, factuality can significantly impact various benchmark characteristics especially when benchmarks are related to knowledge domain, and initial data quality plays a critical role, whereas the number of learning steps has a lesser impact.Additionally, we noted that once the language model is trained with a dataset contaminated by noise, restoring its original performance becomes exceptionally challenging, verging on irreversible.

Motivated by the success of unsupervised neural machine translation (UNMT), we introduce an unsupervised sign language translation and generation network (USLNet), which learns from abundant single-modality (text and video) data without parallel sign language data. USLNet comprises two main components: single-modality reconstruction modules (text and video) that rebuild the input from its noisy version in the same modality and cross-modality back-translation modules (text-video-text and video-text-video) that reconstruct the input from its noisy version in the different modality using back-translation procedure. Unlike the single-modality back-translation procedure in text-based UNMT, USLNet faces the cross-modality discrepancy in feature representation, in which the length and the feature dimension mismatch between text and video sequences. We propose a sliding window method to address the issues of aligning variable-length text with video sequences. To our knowledge, USLNet is the first unsupervised sign language translation and generation model capable of generating both natural language text and sign language video in a unified manner. Experimental results on the BBC-Oxford Sign Language dataset and Open-Domain American Sign Language dataset reveal that USLNet achieves competitive results compared to supervised baseline models, indicating its effectiveness in sign language translation and generation.

Hallucinations in large language models (LLMs) refer to the phenomenon of LLMs producing responses that are coherent yet factually inaccurate. This issue undermines the effectiveness of LLMs in practical applications, necessitating research into detecting and mitigating hallucinations of LLMs. Previous studies have mainly concentrated on post-processing techniques for hallucination detection, which tend to be computationally intensive and limited in effectiveness due to their separation from the LLM’s inference process. To overcome these limitations, we introduce MIND, an unsupervised training framework that leverages the internal states of LLMs for real-time hallucination detection without requiring manual annotations. Additionally, we present HELM, a new benchmark for evaluating hallucination detection across multiple LLMs, featuring diverse LLM outputs and the internal states of LLMs during their inference process. Our experiments demonstrate that MIND outperforms existing state-of-the-art methods in hallucination detection.

Unsupervised constituency parsing focuses on identifying word sequences that form a syntactic unit (i.e., constituents) in target sentences. Linguists identify the constituent by evaluating a set of Predicate-Argument Structure (PAS) equivalent sentences where we find the constituent appears more frequently than non-constituents (i.e., the constituent corresponds to a frequent word sequence within the sentence set). However, such frequency information is unavailable in previous parsing methods that identify the constituent by observing sentences with diverse PAS. In this study, we empirically show that constituents correspond to frequent word sequences in the PAS-equivalent sentence set. We propose a frequency-based parser, span-overlap, that (1) computes the span-overlap score as the word sequence’s frequency in the PAS-equivalent sentence set and (2) identifies the constituent structure by finding a constituent tree with the maximum span-overlap score. The parser achieves state-of-the-art level parsing accuracy, outperforming existing unsupervised parsers in eight out of ten languages. Additionally, we discover a multilingual phenomenon: participant-denoting constituents tend to have higher span-overlap scores than equal-length event-denoting constituents, meaning that the former tend to appear more frequently in the PAS-equivalent sentence set than the latter. The phenomenon indicates a statistical difference between the two constituent types, laying the foundation for future labeled unsupervised parsing research.

Discovering the semantics of multimodal utterances is essential for understanding human language and enhancing human-machine interactions. Existing methods manifest limitations in leveraging nonverbal information for discerning complex semantics in unsupervised scenarios. This paper introduces a novel unsupervised multimodal clustering method (UMC), making a pioneering contribution to this field. UMC introduces a unique approach to constructing augmentation views for multimodal data, which are then used to perform pre-training to establish well-initialized representations for subsequent clustering. An innovative strategy is proposed to dynamically select high-quality samples as guidance for representation learning, gauged by the density of each sample’s nearest neighbors. Besides, it is equipped to automatically determine the optimal value for the top-K parameter in each cluster to refine sample selection. Finally, both high- and low-quality samples are used to learn representations conducive to effective clustering. We build baselines on benchmark multimodal intent and dialogue act datasets. UMC shows remarkable improvements of 2-6% scores in clustering metrics over state-of-the-art methods, marking the first successful endeavor in this domain. The complete code and data are available at https://github.com/thuiar/UMC.

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by incorporating additional information from retrieval. However, studies have shown that LLMs still face challenges in effectively using the retrieved information, even ignore it or be misled by it. The key reason is that the training of LLMs does not clearly make LLMs learn how to utilize input retrieved texts with varied quality. In this paper, we propose a novel perspective that considers the role of LLMs in RAG as “Information Refiner”, which means that regardless of correctness, completeness, or usefulness of retrieved texts, LLMs can consistently integrate knowledge within the retrieved texts and model parameters to generate the texts that are more concise, accurate, and complete than the retrieved texts. To this end, we propose an information refinement training method named INFO-RAG that optimizes LLMs for RAG in an unsupervised manner. INFO-RAG is low-cost and general across various tasks. Extensive experiments on zero-shot prediction of 11 datasets in diverse tasks including Question Answering, Slot-Filling, Language Modeling, Dialogue, and Code Generation show that INFO-RAG improves the performance of LLaMA2 by an average of 9.39% relative points. INFO-RAG also shows advantages in in-context learning and robustness of RAG.

Within the context of reading comprehension, the task of Distractor Generation (DG) aims to generate several incorrect options to confuse readers. In recent years, the emergence of Large Language Models (LLMs) provides a potential for unsupervised DG without expensive human-annotated distractor labels. In this paper, we leverage LLMs as a cost-effective annotator to enhance the DG capability of smaller student models. To perform knowledge distilling, we propose a dual task training framework that integrates pseudo distractors from LLMs and answer information as the objective target with a two-stage training process. Moreover, we devise a counterfactual contrastive decoding mechanism for increasing the distracting capability of the DG model. Experiments show that our unsupervised generation method with Bart-base greatly surpasses GPT-3.5-turbo zero-shot performance with only 200× fewer model parameters. Our proposed unsupervised DG method offers a cost-effective framework for practical reading comprehension applications, without the need of laborious distractor annotation and costly large-size models.

Although Retrieval-Augmented Large Language Models (RALMs) demonstrate their superiority in terms of factuality, they do not consistently outperform the original retrieval-free Language Models (LMs). Our experiments reveal that this example-level performance inconsistency exists not only between retrieval-augmented and retrieval-free LM but also among different retrievers. To understand this phenomenon, we investigate the degeneration behavior of RALMs and theoretically decompose it into four categories. Further analysis based on our decomposition reveals that the innate difference in knowledge sources and the unpredictable degeneration of the reader model contribute most to the inconsistency. Drawing from our analysis, we introduce Ensemble of Retrievers (EoR), a trainable framework that can adaptively retrieve from different knowledge sources and effectively decrease unpredictable reader errors. Our experiments on Open Domain Question Answering show that EoR substantially improves performance over the RALM with a single retriever by considerably reducing inconsistent behaviors.

Despite it being the cornerstone of BPE, the most common tokenization algorithm, the importance of compression in the tokenization process is still unclear. In this paper, we argue for the theoretical importance of compression, that can be viewed as 0-gram language modeling where equal probability is assigned to all tokens. We also demonstrate the empirical importance of compression for downstream success of pre-trained language models. We control the compression ability of several BPE tokenizers by varying the amount of documents available during their training: from 1 million documents to a character-based tokenizer equivalent to no training data at all. We then pre-train English language models based on those tokenizers and fine-tune them over several tasks. We show that there is a correlation between tokenizers’ compression and models’ downstream performance, suggesting that compression is a reliable intrinsic indicator of tokenization quality. These correlations are more pronounced for generation tasks (over classification) or for smaller models (over large ones). We replicated a representative part of our experiments on Turkish and found similar results, confirming that our results hold for languages with typological characteristics dissimilar to English. We conclude that building better compressing tokenizers is a fruitful avenue for further research and for improving overall model performance.

Entity Alignment (EA) is vital for integrating diverse knowledge graph (KG) data, playing a crucial role in data-driven AI applications. Traditional EA methods primarily rely on comparing entity embeddings, but their effectiveness is constrained by the limited input KG data and the capabilities of the representation learning techniques. Against this backdrop, we introduce ChatEA, an innovative framework that incorporates large language models (LLMs) to improve EA. To address the constraints of limited input KG data, ChatEA introduces a KG-code translation module that translates KG structures into a format understandable by LLMs, thereby allowing LLMs to utilize their extensive background knowledge to improve EA accuracy. To overcome the over-reliance on entity embedding comparisons, ChatEA implements a two-stage EA strategy that capitalizes on LLMs’ capability for multi-step reasoning in a dialogue format, thereby enhancing accuracy while preserving efficiency. Our experimental results affirm ChatEA’s superior performance, highlighting LLMs’ potential in facilitating EA tasks.The source code is available at https://anonymous.4open.science/r/ChatEA/.