ACL 2024 论文列表

N-gram language models (LMs) are the innovation that first made large-vocabulary continuous automatic speech recognition (ASR) viable. With neural end-to-end ASR architectures, however, LMs have become an afterthought. While the effect on accuracy may be negligible for English and Mandarin, jettisoning the LM might not make sense for the world’s remaining 6000+ languages. In this paper, we investigate the role of the LM in low-resource ASR. First we ask: does using an n-gram LM in decoding in neural architectures help ASR performance? While it may seem obvious that it should, its absence in most implementations suggests otherwise. Second, we ask: when an n-gram LM is used in ASR, is there a relationship between the size of the LM and ASR accuracy? We have discovered that gut feelings on this question vary considerably, but there is little empirical work to support any particular claim. We explore these questions “in the wild” using a deliberately diverse set of 9 very small ASR corpora. The results show that: (1) decoding with an n-gram LM, regardless of its size, leads to lower word error rates; and (2) increasing the size of the LM appears to yield improvements only when the audio corpus itself is already relatively large. This suggests that collecting additional LM training text may benefit widely-spoken languages which typically have larger audio corpora. In contrast, for endangered languages where data of any kind will always be limited, efforts may be better spent collecting additional transcribed audio.

While machine translation evaluation has been studied primarily for high-resource languages, there has been a recent interest in evaluation for low-resource languages due to the increasing availability of data and models. In this paper, we focus on a zero-shot evaluation setting focusing on low-resource Indian languages, namely Assamese, Kannada, Maithili, and Punjabi. We collect sufficient Multi-Dimensional Quality Metrics (MQM) and Direct Assessment (DA) annotations to create test sets and meta-evaluate a plethora of automatic evaluation metrics. We observe that even for learned metrics, which are known to exhibit zero-shot performance, the Kendall Tau and Pearson correlations with human annotations are only as high as 0.32 and 0.45. Synthetic data approaches show mixed results and overall do not help close the gap by much for these languages. This indicates that there is still a long way to go for low-resource evaluation.

Zero-shot translation aims to translate between language pairs not seen during training in Multilingual Machine Translation (MMT) and is widely considered an open problem. A common, albeit resource-consuming, solution is to add as many related translation directions as possible to the training corpus. In this paper, we show that for an English-centric model, surprisingly large zero-shot improvements can be achieved by simply fine-tuning with a very small amount of multi-parallel data. For example, on the EC30 dataset, we obtain up to +21.7 ChrF++ non-English overall improvements (870 directions) by using only 100 multi-parallel samples while preserving English-centric translation quality. This performance exceeds M2M100 by an average of 5.9 ChrF++ in the involved non-English directions. When investigating the size effect of fine-tuning data on translation quality, we found that already a small, randomly sampled set of fine-tuning directions is sufficient to achieve comparable improvements. The resulting non-English performance is close to the complete translation upper bound. Even in a minimal setting—fine-tuning with only one single sample—the well-known off-target issue is almost completely resolved, explaining parts—but not all—of the observed improvements in translation quality.

Understanding the interplay between emotions in language and user behaviors is critical. We study how moral emotions shape the political participation of users based on cross-cultural online petition data. To quantify moral emotions, we employ a context-aware NLP model that is designed to capture the subtle nuances of emotions across cultures. For model training, we construct and share a moral emotion dataset comprising nearly 50,000 petition sentences in Korean and English each, along with emotion labels annotated by a fine-tuned LLM. We examine two distinct types of user participation: general support (i.e., registered signatures of petitions) and active support (i.e., sharing petitions on social media). We discover that moral emotions like other-suffering increase both forms of participation and help petitions go viral, while self-conscious have the opposite effect. The most prominent moral emotion, other-condemning, led to polarizing responses among the audience. In contrast, other-praising was perceived differently by culture; it led to a rise in active support in Korea but a decline in the UK. Our findings suggest that both moral emotions embedded in language and cultural perceptions are critical to shaping the public’s political discourse.

Large language models (LLMs) with enormous pre-training tokens and parameters emerge diverse abilities, including math reasoning, codegeneration, and instruction following. These abilities are further enhanced by supervised fine-tuning (SFT). While the open-source community has explored ad-hoc SFT for enhancing individual capabilities, proprietary LLMs exhibit versatility across various skills. Therefore, understanding the facilitation of multiple abilities via SFT is paramount. In this study, we specificially focuses on the interplay of data composition between mathematical reasoning, code generation, and general human-aligning abilities during SFT. We propose four intriguing research questions to explore the association between model performance and various factors including data amount, composition ratio, model size and SFT strategies. Our experiments reveal that distinct capabilities scale differently and larger models generally show superior performance with same amount of data. Mathematical reasoning and code generation consistently improve with increasing data amount, whereas general abilities plateau after roughly a thousand samples. Moreover, we observe data composition appears to enhance various abilities under limited data conditions, yet can lead to performance conflicts when data is plentiful. Our findings also suggest the amount of composition data influences performance more than the composition ratio. In analysis of SFT strategies, we find that sequentially learning multiple skills risks catastrophic forgetting. Our proposed Dual-stage Mixed Fine-tuning (DMT) strategy offers a promising solution to learn multiple abilities with different scaling patterns.

Conversational search facilitates complex information retrieval by enabling multi-turn interactions between users and the system. Supporting such interactions requires a comprehensive understanding of the conversational inputs to formulate a good search query based on historical information. In particular, the search query should include the relevant information from the previous conversation turns.However, current approaches for conversational dense retrieval primarily rely on fine-tuning a pre-trained ad-hoc retriever using the whole conversational search session, which can be lengthy and noisy. Moreover, existing approaches are limited by the amount of manual supervision signals in the existing datasets.To address the aforementioned issues, we propose a **H**istory-**A**ware **Conv**ersational **D**ense **R**etrieval (HAConvDR) system, which incorporates two ideas: context-denoised query reformulation and automatic mining of supervision signals based on the actual impact of historical turns.Experiments on two public conversational search datasets demonstrate the improved history modeling capability of HAConvDR, in particular for long conversations with topic shifts.

How can large language models (LLMs) process and translate endangered languages? Many languages lack a large corpus to train a decent LLM; therefore existing LLMs rarely perform well in unseen, endangered languages. On the contrary, we observe that 2000 endangered languages, though without a large corpus, have a grammar book or a dictionary. We propose LingoLLM, a training-free approach to enable an LLM to process unseen languages that hardly occur in its pre-training. Our key insight is to demonstrate linguistic knowledge of an unseen language in an LLM’s prompt, including a dictionary, a grammar book, and morphologically analyzed input text. We implement LingoLLM on top of two models, GPT-4 and Mixtral, and evaluate their performance on 5 tasks across 8 endangered or low-resource languages. Our results show that LingoLLM elevates translation capability from GPT-4’s 0 to 10.5 BLEU for 10 language directions. Our findings demonstrate the tremendous value of linguistic knowledge in the age of LLMs for endangered languages. Our data, code, and model generations will be released to the public. Our data, code, and model generations can be found at https://github.com/LLiLab/llm4endangeredlang.

Hierarchical text classification (HTC) is a challenging problem with two key issues: utilizing structural information and mitigating label imbalance. Recently, the unit-based approach generating unit-based feature representations has outperformed the global approach focusing on a global feature representation. Nevertheless, unit-based models using BCE and ZLPR losses still face static thresholding and label imbalance challenges. Those challenges become more critical in large-scale hierarchies. This paper introduces a novel hierarchy-aware loss function for unit-based HTC models: Hierarchy-aware Biased Bound Margin (HBM) loss. HBM integrates learnable bounds, biases, and a margin to address static thresholding and mitigate label imbalance adaptively. Experimental results on benchmark datasets demonstrate the superior performance of HBM compared to competitive HTC models.

Learning from noisy labels (LNL) is a challenge that arises in many real-world scenarios where collected training data can contain incorrect or corrupted labels. Most existing solutions identify noisy labels and adopt active learning to query human experts on them for denoising. In the era of large language models (LLMs), although we can reduce the human effort to improve these methods, their performances are still subject to accurately separating the clean and noisy samples from noisy data. In this paper, we propose an innovative collaborative learning framework NoiseAL based on active learning to combine LLMs and small models (SMs) for learning from noisy labels. During collaborative training, we first adopt two SMs to form a co-prediction network and propose a dynamic-enhanced threshold strategy to divide the noisy data into different subsets, then select the clean and noisy samples from these subsets to feed the active annotator LLMs to rectify noisy samples. Finally, we employ different optimization objectives to conquer subsets with different degrees of label noises. Extensive experiments on synthetic and real-world noise datasets further demonstrate the superiority of our framework over state-of-the-art baselines.

Addressing the limitation of context length in large language models for code-related tasks is the primary focus of this paper. Existing LLMs are constrained by their pre-trained context lengths, leading to performance issues in handling long complex code sequences. Inspired by how human programmers navigate code, we introduce Hierarchical Rotary Position Embedding (HiRoPE), a novel approach that enhances the traditional rotary position embedding into a hierarchical format based on the hierarchical structure of source code. HiRoPE offers easy integration into existing LLMs without extra training costs. Our method is extensively evaluated with various LLMs, demonstrating stable performance in tasks such as language modeling and long code completion. We also introduce a new long code understanding task with real-world code projects, in hopes of promoting further development in this code-related field. Theoretically and experimentally, we find that HiRoPE also addresses the out-of-distribution issue in position encoding. Our HiRoPE significantly expands the context length capabilities of LLMs, enabling inference at lengths exponentially greater than the training length.

The democratization of pre-trained language models through open-source initiatives has rapidly advanced innovation and expanded access to cutting-edge technologies. However, this openness also brings significant security risks, including backdoor attacks, where hidden malicious behaviors are triggered by specific inputs, compromising natural language processing (NLP) system integrity and reliability. This paper suggests that merging a backdoored model with other homogeneous models can significantly remediate backdoor vulnerabilities even if such models are not entirely secure. In our experiments, we verify our hypothesis on various models (BERT-Base, RoBERTa-Large, Llama2-7B, and Mistral-7B) and datasets (SST-2, OLID, AG News, and QNLI). Compared to multiple advanced defensive approaches, our method offers an effective and efficient inference-stage defense against backdoor attacks on classification and instruction-tuned tasks without additional resources or specific knowledge. Our approach consistently outperforms recent advanced baselines, leading to an average of about 75% reduction in the attack success rate. Since model merging has been an established approach for improving model performance, the extra advantage it provides regarding defense can be seen as a cost-free bonus.

Benchmarks have been essential for driving progress in machine learning. A better understanding of LLM capabilities on real world tasks is vital for safe development.Designing adequate LLM benchmarks is challenging: Data from real-world tasks is hard to collect, public availability of static evaluation data results in test data contamination and benchmark overfitting, and periodically generating new evaluation data is tedious and may result in temporally inconsistent results. We introduce HelloFresh, based on continuous streams of real-world data generated by intrinsically motivated human labelers. It covers recent events from X (formerly Twitter) community notes and edits of Wikipedia pages, mitigating the risk of test data contamination and benchmark overfitting.Any X user can propose an X note to add additional context to a misleading post (formerly tweet); if the community classifies it as helpful, it is shown with the post. Similarly, Wikipedia relies on community-based consensus, allowing users to edit articles or revert edits made by other users.Verifying whether an X note is helpful or whether a Wikipedia edit should be accepted are hard tasks that require grounding by querying the web.We backtest state-of-the-art LLMs supplemented with simple web search access and find that HelloFresh yields a temporally consistent ranking.To enable continuous evaluation on Hellofresh, we host a public leaderboard and periodically updated evaluation data at https://tinyurl.com/hello-fresh-LLM.

Large Language Models (LLMs) can play a vital role in psychotherapy by adeptly handling the crucial task of cognitive reframing and overcoming challenges such as shame, distrust, therapist skill variability, and resource scarcity. Previous LLMs in cognitive reframing mainly converted negative emotions to positive ones, but these approaches have limited efficacy, often not promoting clients’ self-discovery of alternative perspectives. In this paper, we unveil the Helping and Empowering through Adaptive Language in Mental Enhancement (HealMe) model. This novel cognitive reframing therapy method effectively addresses deep-rooted negative thoughts and fosters rational, balanced perspectives. Diverging from traditional LLM methods, HealMe employs empathetic dialogue based on psychotherapeutic frameworks. It systematically guides clients through distinguishing circumstances from feelings, brainstorming alternative viewpoints, and developing empathetic, actionable suggestions. Moreover, we adopt the first comprehensive and expertly crafted psychological evaluation metrics, specifically designed to rigorously assess the performance of cognitive reframing, in both AI-simulated dialogues and real-world therapeutic conversations. Experimental results show that our model outperforms others in terms of empathy, guidance, and logical coherence, demonstrating its effectiveness and potential positive impact on psychotherapy.

While large language models (LLMs) excel in various natural language tasks in English, their performance in low-resource languages like Hebrew, especially for generative tasks such as abstractive summarization, remains unclear. The high morphological richness in Hebrew adds further challenges due to the ambiguity in sentence comprehension and the complexities in meaning construction.In this paper, we address this evaluation and resource gap by introducing HeSum, a novel benchmark dataset specifically designed for Hebrew abstractive text summarization. HeSum consists of 10,000 article-summary pairs sourced from Hebrew news websites written by professionals. Linguistic analysis confirms HeSum’s high abstractness and unique morphological challenges. We show that HeSum presents distinct difficulties even for state-of-the-art LLMs, establishing it as a valuable testbed for advancing generative language technology in Hebrew, and MRLs generative challenges in general.

As the reach of large language models (LMs) expands globally, their ability to cater to diverse cultural contexts becomes crucial. Despite advancements in multilingual capabilities, models are not designed with appropriate cultural nuances. In this paper, we show that multilingual and Arabic monolingual LMs exhibit bias towards entities associated with Western culture. We introduce CAMeL, a novel resource of 628 naturally-occurring prompts and 20,368 entities spanning eight types that contrast Arab and Western cultures. CAMeL provides a foundation for measuring cultural biases in LMs through both extrinsic and intrinsic evaluations. Using CAMeL, we examine the cross-cultural performance in Arabic of 16 different LMs on tasks such as story generation, NER, and sentiment analysis, where we find concerning cases of stereotyping and cultural unfairness. We further test their text-infilling performance, revealing the incapability of appropriate adaptation to Arab cultural contexts. Finally, we analyze 6 Arabic pre-training corpora and find that commonly used sources such as Wikipedia may not be best suited to build culturally aware LMs, if used as they are without adjustment. We will make CAMeL publicly available at: https://github.com/tareknaous/camel

Document-level event extraction aims to extract structured event information from unstructured text. However, a single document often contains limited event information and the roles of different event arguments may be biased due to the influence of the information source.This paper addresses the limitations of traditional document-level event extraction by proposing the task of cross-document event extraction (CDEE) to integrate event information from multiple documents and provide a comprehensive perspective on events. We construct a novel cross-document event extraction dataset, namely CLES, which contains 20,059 documents and 37,688 mention-level events, where over 70% of them are cross-document. To address the task, we propose a CDEE pipeline that includes 5 steps, namely event extraction, coreference resolution, entity normalization, role normalization and entity-role resolution. Our CDEE pipeline achieves about 72% F1 in end-to-end cross-document event extraction, suggesting the challenge of this task and setting up a benchmark for future research. Our work builds a new line of information extraction research and will attract new research attention.

Advancements in logical reasoning, utilizing LLMs to convert natural language into logical symbolism, combined with the use of external theorem provers, have repositioned the symbolic approach as a central point of interest. The main challenge within this paradigm lies in the LLMs’ capability to accurately translate natural language (NL) statements into first-order-logic (FOL) expressions. Although LLMs have shown notable success, there remains a gap in understanding the limitations and challenges they encounter in NL-FOL translation. This is primarily due to the absence of datasets and evaluation test beds at the required fine-grained level. We present MALLS, a dataset of 28K diverse and verified sentence-level NL-FOL pairs collected from GPT4. We utilize a combined strategy of FOL rule parsing, human annotation, and automatic filtering to ensure quality. We also present LogicLLaMA, a LLaMA2-7B/13B fine-tuned on MALLS for NL-FOL translation, which can be used standalone or to correct previously generated rules by GPT3.5 after being further fine-tuned via a novel reinforcement learning with human feedback (RLHF) framework. We benchmark a wide range of LLMs on MALLS and previous datasets, highlighting weaknesses in them in NL-FOL translation and demonstrating the advantages of MALLS. We also show that LogicLLaMA achieves GPT4-level performance and can generalize to other datasets. Project repo is available at https://github.com/gblackout/LogicLLaMA

To better analyze informal arguments on public forums, we propose the task of argument explication, which makes explicit a text’s argumentative structure and implicit reasoning by outputting triples of propositions ⟨claim, reason warrant⟩. The three slots, or argument components, are derived from the widely known Toulmin (1958) model of argumentation. While prior research applies Toulmin or related theories to annotate datasets and train supervised models, we develop an effective method to prompt generative large language models (LMs) to output explicitly named argument components proposed by Toulmin by prompting with the theory name (e.g., ‘According to Toulmin model’). We evaluate the outputs’ coverage and validity through a human study and automatic evaluation based on prior argumentation datasets and perform robustness checks over alternative LMs, prompts, and argumentation theories. Finally, we conduct a proof-of-concept case study to extract an interpretable argumentation (hyper)graph from a large corpus of critical public comments on whether to allow the COVID-19 vaccine for children, suggesting future directions for corpus analysis and argument visualization.

As Machine Learning (ML) models grow in size and demand higher-quality training data, the expenses associated with re-training and fine-tuning these models are escalating rapidly. Inspired by recent impressive achievements of Large Language Models (LLMs) in different fields, this paper delves into the question: can LLMs efficiently improve an ML’s performance at a minimal cost? We show that, through our proposed training-free framework LLMCorr, an LLM can work as a post-hoc corrector to propose corrections for the predictions of an arbitrary ML model. In particular, we form a contextual knowledge database by incorporating the dataset’s label information and the ML model’s predictions on the validation dataset. Leveraging the in-context learning capability of LLMs, we ask the LLM to summarise the instances in which the ML model makes mistakes and the correlation between primary predictions and true labels. Following this, the LLM can transfer its acquired knowledge to suggest corrections for the ML model’s predictions. Our experimental results on text analysis and the challenging molecular predictions show that LLMCorr improves the performance of a number of models by up to 39%.