ACL 2024 论文列表

Since the natural language processing (NLP) community started to make large language models (LLMs) act as a critic to evaluate the quality of generated texts, most of the existing works train a critique generation model on the evaluation data labeled by GPT-4’s direct prompting. We observe that these models lack the ability to generate informative critiques in both pointwise grading and pairwise comparison especially without references. As a result, their generated critiques cannot provide fine-grained distinguishability on generated texts, causing unsatisfactory evaluation performance. In this paper, we propose a simple yet effective method called Eval-Instruct, which can first acquire pointwise grading critiques with pseudo references and then revise these critiques via multi-path prompting to obtain informative evaluation data in different tasks and settings, including pointwise grading and pairwise comparison with / without references. After fine-tuning on these data, the resulting model CritiqueLLM is empirically shown to outperform ChatGPT and all the open-source baselines and even achieve comparable evaluation performance to GPT-4 in system-level correlations of pointwise grading. We also demonstrate that our generated critiques can act as scalable feedback to further improve the generation quality of strong LLMs like ChatGPT.

Neural Machine Translation models are extremely data and compute-hungry. However, not all datapoints contribute equally to model training and generalization. Data pruning to remove the low-value data points has the benefit of drastically reducing the compute budget without significantdrop in model performance. In this paper, we propose a new data pruning technique: CheckpointsAcross Time (CAT ), that leverages early model training dynamics to identify the most relevantdata points for model performance. We benchmark CAT against several data pruning techniquesincluding COMET-QE, LASER and LaBSE. We find that CAT outperforms the benchmarks onIndo-European languages on multiple test sets. When applied to English-German, English-Frenchand English-Swahili translation tasks, CAT achieves comparable performance to using the fulldataset, while pruning up to 50% of training data. We inspect the data points that CAT selectsand find that it tends to favour longer sentences and sentences with unique or rare words.

The ability of Large Language Models (LLMs) to critique and refine their reasoning is crucial for their application in evaluation, feedback provision, and self-improvement. This paper introduces CriticBench, a comprehensive benchmark designed to assess LLMs’ abilities to critique and rectify their reasoning across a variety of tasks. CriticBench encompasses five reasoning domains: mathematical, commonsense, symbolic, coding, and algorithmic. It compiles 15 datasets and incorporates responses from three LLM families. Utilizing CriticBench, we evaluate and dissect the performance of 17 LLMs in generation, critique, and correction reasoning, i.e., GQC reasoning. Our findings reveal: (1) a linear relationship in GQC capabilities, with critique-focused training markedly enhancing performance; (2) a task-dependent variation in correction effectiveness, with logic-oriented tasks being more amenable to correction; (3) GQC knowledge inconsistencies that decrease as model size increases; and (4) an intriguing inter-model critiquing dynamic, where stronger models are better at critiquing weaker ones, while weaker models can surprisingly surpass stronger ones in their self-critique. We hope these insights into the nuanced critique-correct reasoning of LLMs will foster further research in LLM critique and self-improvement.

The customization of large language models (LLMs) for user-specified tasks gets important. However, maintaining all the customized LLMs on cloud servers incurs substantial memory and computational overheads, and uploading user data can also lead to privacy concerns. On-device LLMs can offer a promising solution by mitigating these issues. Yet, the performance of on-device LLMs is inherently constrained by the limitations of small-scaled models. To overcome these restrictions, we first propose Crayon, a novel approach for on-device LLM customization. Crayon begins by constructing a pool of diverse base adapters, and then we instantly blend them into a customized adapter without extra training. In addition, we develop a device-server hybrid inference strategy, which deftly allocates more demanding queries or non-customized tasks to a larger, more capable LLM on a server. This ensures optimal performance without sacrificing the benefits of on-device customization. We carefully craft a novel benchmark from multiple question-answer datasets, and show the efficacy of our method in the LLM customization.

While reinforcement learning (RL) has been proven essential for tuning large language models (LLMs), it can lead to reward over-optimization (ROO). Existing approaches address ROO by adding KL regularization, requiring computationally expensive hyperparameter tuning. Additionally, KL regularization focuses solely on regularizing the language policy, neglecting a potential source of regularization: the reward function itself. Inspired by demonstration-guided RL, we here introduce the Reward Calibration from Demonstration (RCfD), which leverages human demonstrations and a reward model to recalibrate the reward objective. Formally, given a prompt, the RCfD objective minimizes the distance between the demonstrations’ and LLM’s rewards rather than directly maximizing the reward function. This objective shift avoids incentivizing the LLM to exploit the reward model and promotes more natural and diverse language generation.We show the effectiveness of RCfD in three RL language tasks, where it achieves comparable performance to carefully tuned baselines while mitigating ROO.

We propose CounterCurate, a framework to comprehensively improve the visio-linguistic compositional reasoning capability for both contrastive and generative multimodal models. In particular, we identify two critical under- explored problems: the neglect of physically grounded reasoning (counting and position understanding) and the potential of using highly capable text and image generation models for semantic counterfactual fine-tuning. Our work pioneers an approach in addressing these gaps.We first spotlight the near-chance performance of multimodal models like CLIP and LLaVA in physically grounded compositional reasoning. We then apply simple data augmentation using the grounded image generation model GLIGEN to generate fine-tuning data, resulting in significant performance improvements: +33% and +37% for CLIP and LLaVA, respectively, on our newly curated Flickr30k-Positions benchmark. Moreover, we exploit the capabilities of high-performing text generation and image generation models, specifically GPT-4V and DALLE-3, to curate challenging semantic counterfactuals, thereby further enhancing compositional reasoning capabilities on benchmarks such as SugarCrepe, where CounterCurate outperforms GPT-4V.To facilitate future research, we release ourcode, dataset, benchmark, and checkpoints at https://countercurate.github.io/

Understanding and following natural language instructions while navigating through complex, real-world environments poses a significant challenge for general-purpose robots. These environments often include obstacles and pedestrians, making it essential for autonomous agents to possess the capability of self-corrected planning to adjust their actions based on feedback from the surroundings. However, the majority of existing vision-and-language navigation (VLN) methods primarily operate in less realistic simulator settings and do not incorporate environmental feedback into their decision-making processes. To address this gap, we introduce a novel zero-shot framework called CorNav, utilizing a large language model for decision-making and comprising two key components: 1) incorporating environmental feedback for refining future plans and adjusting its actions, and 2) multiple domain experts for parsing instructions, scene understanding, and refining predicted actions. In addition to the framework, we develop a 3D simulator that renders realistic scenarios using Unreal Engine 5. To evaluate the effectiveness and generalization of navigation agents in a zero-shot multi-task setting, we create a benchmark called NavBench. Our empirical study involves deploying 7 baselines across four tasks, i.e., goal-conditioned navigation given a specific object category, goal-conditioned navigation given simple instructions, finding abstract objects based on high-level instructions, and step-by-step instruction following. Extensive experiments demonstrate that CorNav consistently outperforms all baselines by a significant margin across all tasks. On average, CorNav achieves a success rate of 28.1%, surpassing the best baseline’s performance of 20.5%.

End-to-end automatic speech recognition (ASR) systems have made significant progress in general scenarios. However, it remains challenging to transcribe contextual named entities (NEs) in the contextual ASR scenario. Previous approaches have attempted to address this by utilizing the NE dictionary. These approaches treat entities as individual tokens and generate them token-by-token, which may result in incomplete transcriptions of entities. In this paper, we treat entities as indivisible wholes and introduce the idea of copying into ASR. We design a systematic mechanism called CopyNE, which can copy entities from the NE dictionary. By copying all tokens of an entity at once, we can reduce errors during entity transcription, ensuring the completeness of the entity. Experiments demonstrate that CopyNE consistently improves the accuracy of transcribing entities compared to previous approaches. Even when based on the strong Whisper, CopyNE still achieves notable improvements.

Conversational question answering (ConvQA) over knowledge graphs (KGs) involves answering multi-turn natural language questions about information contained in a KG. State-of-the-art methods of ConvQA often struggle with inexplicit question-answer pairs. These inputs are easy for human beings to understand given a conversation history, but hard for a machine to interpret, which can degrade ConvQA performance. To address this problem, we propose a reinforcement learning (RL) based model, CoRnNet, which utilizes question reformulations generated by large language models (LLMs) to improve ConvQA performance. CoRnNet adopts a teacher-student architecture where a teacher model learns question representations using human writing reformulations, and a student model to mimic the teacher model’s output via reformulations generated by LLMs. The learned question representation is then used by a RL model to locate the correct answer in a KG. Extensive experimental results show that CoRnNet outperforms state-of-the-art ConvQA models.

Cross-prompt automated essay scoring (AES), an under-investigated but challenging task that has gained increasing popularity in the AES community, aims to train an AES system that can generalize well to prompts that are unseen during model training. While recently-developed cross-prompt AES models have combined essay representations that are learned via sophisticated neural architectures with so-called prompt-independent features, an intriguing question is: are complex neural models needed to achieve state-of-the-art results? We answer this question by abandoning sophisticated neural architectures and developing a purely feature-based approach to cross-prompt AES that adopts a simple neural architecture. Experiments on the ASAP dataset demonstrate that our simple approach to cross-prompt AES can achieve state-of-the-art results.

Controlled Text Generation (CTG) aims to produce texts that exhibit specific desired attributes. In this study, we introduce a pluggable CTG framework for Large Language Models (LLMs) named Dynamic Attribute Graphs-based controlled text generation (DATG). This framework utilizes an attribute scorer to evaluate the attributes of sentences generated by LLMs and constructs dynamic attribute graphs. DATG modulates the occurrence of key attribute words and key anti-attribute words, achieving effective attribute control without compromising the original capabilities of the model. We conduct experiments across four datasets in two tasks: toxicity mitigation and sentiment transformation, employing five LLMs as foundational models. Our findings highlight a remarkable enhancement in control accuracy, achieving a peak improvement of 19.29% over baseline methods in the most favorable task across four datasets. Additionally, we observe a significant decrease in perplexity, markedly improving text fluency.

Controlled text generation, aiming to ensure that language models produce text containing only the desired domain or corpus attributes, is immensely crucial in the practical application of language models. Existing methods, however, are inapplicable to black-box models or suffer a significant trade-off between control and fluency in text generation. This paper introduces the Score-based Progressive Editor (ScoPE), a novel approach designed to overcome these issues. ScoPE modifies the context at the token level during the generation process of a backbone language model. This modification guides the subsequent text to naturally include the target attributes. To facilitate this process, ScoPE employs a training objective that maximizes a target score, comprehensively considering both control and fluency. Experimental results on diverse controlled generation tasks demonstrate that ScoPE can effectively regulate the attributes of the generated text while effectively utilizing the capability of the backbone large language models.

Generic text summarization approaches often fail to address the specific intent and needs of individual users. Recently, scholarly attention has turned to the development of summarization methods that are more closely tailored and controlled to align with specific objectives and user needs. Despite a growing corpus of controllable summarization research, there is no comprehensive survey available that thoroughly explores the diverse controllable attributes employed in this context, delves into the associated challenges, and investigates the existing solutions. In this survey, we formalize the Controllable Text Summarization (CTS) task, categorize controllable attributes according to their shared characteristics and objectives, and present a thorough examination of existing datasets and methods within each category. Moreover, based on our findings, we uncover limitations and research gaps, while also exploring potential solutions and future directions for CTS. We release our detailed analysis of CTS papers at https://github.com/ashokurlana/controllable_text_summarization_survey.

Large-scale Causal Language Models (CLMs), e.g., GPT3 and ChatGPT, have brought great success in text generation. However, it is still an open challenge to effectively control the generation process of a CLM while balancing the flexibility, control granularity, and generation efficiency. In this paper, we provide a new alternative for controllable text generation (CTG), by designing a non-intrusive, lightweight control plugin, namely Residual Memory Transformer (RMT), to accompany the generation of CLM at arbitrary time steps. With an encoder-decoder setup, RMT can accept any types of control conditions and cooperate with the base CLM through a residual learning paradigm, to achieve a more flexible, general, and efficient CTG. Extensive experiments are carried out on various control tasks, in the form of both automatic and human evaluations. The results demonstrate the superiority of RMT over a wide range of state-of-the-art CTG approaches. The code implementation of our work is available at: https://github.com/Residual_Memory_Transformer.

Prompting large language models (LLMs) for data augmentation has recently become a common practice in few-shot NLP tasks. In this paper, we propose Chain-of-Thought Attribute Manipulation (CoTAM), a novel approach that generates new data from existing examples by only tweaking in the user-provided, task-specific attribute, e.g., sentiment polarity or topic in movie reviews. Instead of conventional latent representation controlling, we leverage the chain-of-thought prompting to directly edit the text in three steps, (1) attribute decomposition, (2) manipulation proposal, and (3) sentence reconstruction. Extensive results on various tasks, such as text (pair) classification and aspect-based sentiment analysis, verify the superiority of CoTAM over other LLM-based augmentation methods with the same number of training examples for both fine-tuning and in-context learning. Remarkably, the 2D visualization of the augmented dataset using principle component analysis revealed a human-recognizable decision boundary that is likely hinted by the attribute manipulation, demonstrating the potential of our proposed approach.

Instruction tuning has been used as a promising approach to improve the performance of large language models (LLMs) on unseen tasks. However, current LLMs exhibit limited robustness to unseen instructions, generating inconsistent outputs when the same instruction is phrased with slightly varied forms or language styles. This behavior indicates LLMs’ lack of robustness to textual variations and generalizability to unseen instructions, potentially leading to trustworthiness issues. Accordingly, we propose Contrastive Instruction Tuning, which maximizes the similarity between the hidden representations of semantically equivalent instruction-instance pairs while minimizing the similarity between semantically different ones. To facilitate this approach, we augment the existing FLAN collection by paraphrasing task instructions. Experiments on the PromptBench benchmark show that CoIN consistently improves LLMs’ robustness to unseen instructions with variations across character, word, sentence, and semantic levels by an average of +2.5% in accuracy.

Incrementally expanding the capability of an existing translation model to solve new domain tasks over time is a fundamental and practical problem, which usually suffers from catastrophic forgetting. Generally, multi-domain learning can be seen as a good solution. However, there are two drawbacks: 1) it requires having the training data for all domains available at the same time, which may be unrealistic due to storage or privacy concerns; 2) it requires re-training the model on the data of all domains from scratch when adding a new domain and this is time-consuming and computationally expensive. To address these issues, we present a semi-supervised contrastive distillation framework for incremental neural machine translation. Specifically, to avoid catastrophic forgetting, we propose to exploit unlabeled data from the same distributions of the older domains through knowledge distillation. Further, to ensure the distinct domain characteristics in the model as the number of domains increases, we devise a cross-domain contrastive objective to enhance the distilled knowledge. Extensive experiments on domain translation benchmarks show that our approach, without accessing any previous training data or re-training on all domains from scratch, can significantly prevent the model from forgetting previously learned knowledge while obtaining good performance on the incrementally added domains. The code and data with step-by-step instructions will be released upon acceptance.

Continual few-shot relation extraction (CFRE) aims to continually learn new relations with limited samples. However, current methods neglect the instability of embeddings in the process of different task training, which leads to serious catastrophic forgetting. In this paper, we propose the concept of the following degree from the perspective of instability to analyze catastrophic forgetting and design a novel method based on adaptive gradient correction and knowledge decomposition to alleviate catastrophic forgetting. Specifically, the adaptive gradient correction algorithm is designed to limit the instability of embeddings, which adaptively constrains the current gradient to be orthogonal to the embedding space learned from previous tasks. To reduce the instability between samples and prototypes, the knowledge decomposition module decomposes knowledge into general and task-related knowledge from the perspective of model architecture, which is asynchronously optimized during training. Experimental results on two standard benchmarks show that our method outperforms the state-of-the-art CFRE model and effectively improves the following degree of embeddings.

An ideal dialogue system requires continuous skill acquisition and adaptation to new tasks while retaining prior knowledge. Dialogue State Tracking (DST), vital in these systems, often involves learning new services, confronting catastrophic forgetting and a critical capability loss termed the “Value Selection Quandary”. To address these challenges, we introduce the Reason-of-Select (RoS) distillation method by enhancing smaller models with a novel “meta-reasoning” capability. Meta-reasoning, employing an enhanced multi-domain perspective, combines fragments of meta-knowledge from domain-specific dialogues during continual learning, transcending traditional single-perspective reasoning. This domain bootstrapping process enhances the model’s ability to dissect intricate dialogues from multiple possible values, and its domain-agnostic property aligns data distribution across different domains, effectively mitigating forgetting. Besides, two novel improvements, “multi-value resolution” strategy and Semantic Contrastive Reasoning Selection method, significantly enhance RoS by generating DST-specific selection chains and mitigating hallucinations in teachers’ reasoning, ensuring effective and reliable knowledge transfer. Extensive experiments validate the exceptional performance and robust generalization capabilities of our method.