ACL 2024 论文列表

Reducing the ‘hallucination' problem of Large Language Models (LLMs) is crucial for their wide applications. A comprehensive and fine-grained measurement of the hallucination is the first key step for the governance of this issue but is under-explored in the community.Thus, we present ANAH, a bilingual dataset that offers ANalytical Annotation of Hallucinations in LLMs within Generative Question Answering.Each answer sentence in our dataset undergoes rigorous annotation, involving the retrieval of a reference fragment, the judgment of the hallucination type, and the correction of hallucinated content. ANAH consists of ~12k sentence-level annotations for ~4.3k LLM responses covering over 700 topics, constructed by a human-in-the-loop pipeline.Thanks to the fine granularity of the hallucination annotations, we can quantitatively confirm that the hallucinations of LLMs progressively accumulate in the answer and use ANAH to train and evaluate hallucination annotators. We conduct extensive experiments on studying generative and discriminative annotators and show that, although current open-source LLMs have difficulties in fine-grained hallucination annotation, the generative annotator trained with ANAH can surpass all open-source LLMs and GPT-3.5, obtain performance competitive with GPT-4, and exhibits better generalization ability on unseen questions.

We introduce ALaRM, the first framework modeling hierarchical rewards in reinforcement learning from human feedback (RLHF), which is designed to enhance the alignment of large language models (LLMs) with human preferences. The framework addresses the limitations of current alignment approaches, which often struggle with the inconsistency and sparsity of human supervision signals, by integrating holistic rewards with aspect-specific rewards. This integration enables more precise and consistent guidance of language models towards desired outcomes, particularly in complex and open text generation tasks. By employing a methodology that filters and combines multiple rewards based on their consistency, the framework provides a reliable mechanism for improving model alignment. We validate our approach through applications in long-form question answering and machine translation tasks, employing gpt-3.5-turbo for pairwise comparisons, and demonstrate improvements over existing baselines. Our work underscores the effectiveness of hierarchical rewards modeling in refining LLM training processes for better human preference alignment. We release our code at https://ALaRM-fdu.github.io.

Recently, instruction-following audio-language models have received broad attention for human-audio interaction. However, the absence of benchmarks capable of evaluating audio-centric interaction capabilities has impeded advancements in this field. Previous models primarily focus on assessing different fundamental tasks, such as automatic speech recognition, and lack an assessment of the open-ended generative capabilities centered around audio. Thus, it is challenging to track the progression in the Large Audio-Language Models (LALMs) domain and to provide guidance for future improvement.In this paper, we introduce AIR-Bench (Audio InstRuction Benchmark), the first benchmark designed to evaluate the ability of LALMs to understand various types of audio signals (including human speech, natural sounds, and music), and furthermore, to interact with humans in the textual format. AIR-Bench encompasses two dimensions: foundation and chat benchmarks. The former consists of 19 tasks with approximately 19k single-choice questions, intending to inspect the basic single-task ability of LALMs. The latter one contains 2k instances of open-ended question-and-answer data, directly assessing the comprehension of the model on complex audio and its capacity to follow instructions. Both benchmarks require the model to generate hypotheses directly. We design a unified framework that leverages advanced language models, such as GPT-4, to evaluate the scores of generated hypotheses given the meta-information of the audio. Experimental results demonstrate a high level of consistency between GPT-4-based evaluation and human evaluation. By revealing the limitations of existing LALMs through evaluation results, AIR-Bench can provide insights into the direction of future research. Dataset and evaluation code are available at https://github.com/OFA-Sys/AIR-Bench.

Large Language Models (LLMs) are increasingly used as ‘content farm’ models (CFMs), to generate synthetic text that could pass for real news articles. This is already happening even for languages that do not have high-quality monolingual LLMs. We show that fine-tuning Llama (v1), mostly trained on English, on as little as 40K Italian news articles, is sufficient for producing news-like texts that native speakers of Italian struggle to identify as synthetic.We investigate three LLMs and three methods of detecting synthetic texts (log-likelihood, DetectGPT, and supervised classification), finding that they all perform better than human raters, but they are all impractical in the real world (requiring either access to token likelihood information or a large dataset of CFM texts). We also explore the possibility of creating a proxy CFM: an LLM fine-tuned on a similar dataset to one used by the real ‘content farm’. We find that even a small amount of fine-tuning data suffices for creating a successful detector, but we need to know which base LLM is used, which is a major challenge.Our results suggest that there are currently no practical methods for detecting synthetic news-like texts ‘in the wild’, while generating them is too easy. We highlight the urgency of more NLP research on this problem.

Most existing rationalization approaches are susceptible to degeneration accumulation due to a lack of effective control over the learning direction of the model during training. To address this issue, we propose a novel approach AGR (Agent-Guided Rationalization), guiding the next action of the model based on its current training state. Specifically, we introduce causal intervention calculus to quantify the causal effects inherent during rationale training, and utilize reinforcement learning process to refine the learning bias of them. Furthermore, we pretrain an agent within this reinforced causal environment to guide the next step of the model. We theoretically demonstrate that a good model needs the desired guidance, and empirically show the effectiveness of our approach, outperforming existing state-of-the-art methods on BeerAdvocate and HotelReview datasets.

Legal tasks and datasets are often used as benchmarks for the capabilities of language models. However, openly available annotated datasets are rare. In this paper, we introduce AGB-DE, a corpus of 3,764 clauses from German consumer contracts that have been annotated and legally assessed by legal experts. Together with the data, we present a first baseline for the task of detecting potentially void clauses, comparing the performance of an SVM baseline with three fine-tuned open language models and the performance of GPT-3.5. Our results show the challenging nature of the task, with no approach exceeding an F1-score of 0.54. While the fine-tuned models often performed better with regard to precision, GPT-3.5 outperformed the other approaches with regard to recall. An analysis of the errors indicates that one of the main challenges could be the correct interpretation of complex clauses, rather than the decision boundaries of what is permissible and what is not.

With the rise of generative AI, automated fact-checking methods to combat misinformation are becoming more and more important. However, factual claim detection, the first step in a fact-checking pipeline, suffers from two key issues that limit its scalability and generalizability: (1) inconsistency in definitions of the task and what a claim is, and (2) the high cost of manual annotation. To address (1), we review the definitions in related work and propose a unifying definition of factual claims that focuses on verifiability. To address (2), we introduce AFaCTA (Automatic Factual Claim deTection Annotator), a novel framework that assists in the annotation of factual claims with the help of large language models (LLMs). AFaCTA calibrates its annotation confidence with consistency along three predefined reasoning paths. Extensive evaluation and experiments in the domain of political speech reveal that AFaCTA can efficiently assist experts in annotating factual claims and training high-quality classifiers, and can work with or without expert supervision. Our analyses also result in PoliClaim, a comprehensive claim detection dataset spanning diverse political topics.

Large language models (LLMs) show great performance in various tasks, but face deployment challenges from limited memory capacity and bandwidth.Low-bit weight quantization can save memory and accelerate inference.Although floating-point (FP) formats show good performance in LLM quantization, they tend to perform poorly with small group sizes or sub-4 bits.We find the reason is that the absence of asymmetry in previous FP quantization makes it unsuitable for handling asymmetric value distribution of LLM weight tensors.In this work, we propose asymmetric FP quantization (AFPQ), which sets separate scales for positive and negative values.Our method leads to large accuracy improvements and can be easily plugged into other quantization methods, including GPTQ and AWQ, for better performance.Besides, no additional storage is needed compared with asymmetric integer (INT) quantization.The code is available at https://github.com/zhangsichengsjtu/AFPQ.

We present a novel Parameter-Efficient Fine-Tuning (PEFT) method, dubbed as Adaptive Freezing of Low-Rank Adaptation (AFLoRA). Specifically, for each pre-trained frozen weight tensor, we add a parallel path of trainable low-rank matrices, namely a down-projection and an up-projection matrix, each of which is followed by a feature transformation vector. Based on a novel freezing score, we incrementally freeze these projection matrices during fine-tuning to reduce the computation and alleviate over-fitting. Our experimental results demonstrate that we can achieve state-of-the-art performance with an average improvement of up to 0.85% as evaluated on the GLUE benchmark while yielding up to 9.5× fewer average trainable parameters. While compared in terms of runtime, AFLoRA can yield up to 1.86× improvement as opposed to similar PEFT alternatives. Besides the practical utility of our approach, we provide insights on the trainability requirements of LoRA paths at different modules and the freezing schedule for the different projection matrices.

To improve the performance of the dual-encoder retriever, one effective approach is knowledge distillation from the cross-encoder ranker. Existing works prepare training instances by pairing each query with one positive and a batch of negatives. However, most hard negatives mined by advanced dense retrieval methods are still too trivial for the teacher to distinguish, preventing the teacher from transferring abundant dark knowledge to the student through its soft label. To alleviate this issue, we propose Adam, a knowledge distillation framework that can better transfer the dark knowledge held in the teacher with adaptive dark examples. Different from previous works that only rely on one positive and hard negatives as candidate passages, we create dark examples that all have moderate relevance to the query by strengthening negatives and masking positives in the discrete space. Furthermore, as the quality of knowledge held in different training instances varies as measured by the teacher’s confidence score, we propose a self-paced distillation strategy that adaptively concentrates on a subset of high-quality instances to conduct our dark-example-based knowledge distillation to help the student learn better. We conduct experiments on two widely-used benchmarks and verify the effectiveness of our method.

The impressive generation capabilities of large language models (LLMs) have made it harder to detect the subtle hallucinations they make in abstractive summarization, where generated summaries consist of a blend of correct and incorrect information w.r.t. a given document. Recently-proposed LLM-based evaluation metrics attempt to capture this, but still face challenges: (1) they are biased towards summaries generated from the same underlying LLM, and (2) they lack interpretability, offering only a single score. In this work, we present ACUEval, a metric that leverages the power of LLMs to perform two sub-tasks: decomposing summaries into atomic content units (ACUs), and validating them against the source document. Compared to current strong LLM-based metrics, our two-step evaluation strategy improves correlation with human judgments of faithfulness on three summarization evaluation benchmarks by 3% in balanced accuracy compared to the next-best metric, and also shows reduced preference bias towards LLM-generated summary. Further, we show that errors detected by ACUEval can be used to generate actionable feedback for refining the summary, improving the faithfulness scores by more than 10%.

We present ABEX, a novel and effective generative data augmentation methodology for low-resource Natural Language Understanding (NLU) tasks. ABEX is based on ABstract-and-EXpand, a novel paradigm for generating diverse forms of an input document – we first convert a document into its concise, abstract description and then generate new documents based on expanding the resultant abstraction. To learn the task of expanding abstract descriptions, we first train BART on a large-scale synthetic dataset with abstract-document pairs. Next, to generate abstract descriptions for a document, we propose a simple, controllable, and training-free method based on editing AMR graphs. ABEX brings the best of both worlds: by expanding from abstract representations, it preserves the original semantic properties of the documents, like style and meaning, thereby maintaining alignment with the original label and data distribution. At the same time, the fundamental process of elaborating on abstract descriptions facilitates diverse generations. We demonstrate the effectiveness of ABEX on 4 NLU tasks spanning 12 datasets and 4 low-resource settings. ABEX outperforms all our baselines qualitatively with improvements of 0.04% - 38.8%. Qualitatively, ABEX outperforms all prior methods from literature in terms of context and length diversity.

Natural Language Inference (NLI) remains an important benchmark task for LLMs. NLI datasets are a springboard for transfer learning to other semantic tasks, and NLI models are standard tools for identifying the faithfulness of model-generated text. There are several large scale NLI datasets today, and models have improved greatly by hill-climbing on these collections. Yet their realistic performance on out-of-distribution/domain data is less well-understood. We explore the opportunity for synthetic high-quality datasets to adapt NLI models for zero-shot use in downstream applications across new and unseen text domains. We demonstrate a new approach for generating NLI data in diverse domains and lengths, so far not covered by existing training sets. The resulting examples have meaningful premises, the hypotheses are formed in creative ways rather than simple edits to a few premise tokens, and the labels have high accuracy. We show that models trained on this data (685K synthetic examples) have the best generalization to completely new downstream test settings. On the TRUE benchmark, a T5-small model trained with our data improves around 7% on average compared to training on the best alternative dataset. The improvements are more pronounced for smaller models, while still meaningful on a T5 XXL model. We also demonstrate gains on test sets when in-domain training data is augmented with our domain-general synthetic data.

This paper introduces a multi-level, multi-label text classification dataset comprising over 3000 documents. The dataset features literary and critical texts from 19th-century Ottoman Turkish and Russian. It is the first study to apply large language models (LLMs) to this dataset, sourced from prominent literary periodicals of the era. The texts have been meticulously organized and labeled. This was done according to a taxonomic framework that takes into account both their structural and semantic attributes. Articles are categorized and tagged with bibliometric metadata by human experts. We present baseline classification results using a classical bag-of-words (BoW) naive Bayes model and three modern LLMs: multilingual BERT, Falcon, and Llama-v2. We found that in certain cases, Bag of Words (BoW) outperforms Large Language Models (LLMs), emphasizing the need for additional research, especially in low-resource language settings. This dataset is expected to be a valuable resource for researchers in natural language processing and machine learning, especially for historical and low-resource languages. The dataset is publicly available.

Temporal knowledge graph (TKG) reasoning has two settings: interpolation reasoning and extrapolation reasoning. Both of them draw plenty of research interest and have great significance. Methods of the former de-emphasize the temporal correlations among facts sequences, while methods of the latter require strict chronological order of knowledge and ignore inferring clues provided by missing facts of the past. These limit the practicability of TKG applications as almost all of the existing TKG reasoning methods are designed specifically to address either one setting. To this end, this paper proposes an original Temporal PAth-based Reasoning (TPAR) model for both the interpolation and extrapolation reasoning settings. TPAR performs a neural-driven symbolic reasoning fashion that is robust to ambiguous and noisy temporal data, and with fine interpretability as well. Comprehensive experiments show that TPAR outperforms SOTA methods on the link prediction task for both the interpolation and the extrapolation settings. A novel pipeline experimental setting is designed to evaluate the performances of SOTA combinations and the proposed TPAR towards interpolation and extrapolation reasoning. And more diverse experiments are conducted to show the robustness and interpretability of TPAR.

Knowledge graph completion (KGC) task is to infer the missing knowledge in the knowledge graph based on known factual triples. However, present KGC approaches still face the following two challenges. Those methods perform simple linear update on relation representation, and only local neighborhood information is aggregated, which makes it difficult to capture logic semantic between relations and global topological context information. To tackle the above challenges, we propose a unified joint approach with Topological Context learning and Rule Augmentation (TCRA) for KGC. The TCRA framework consists of an entity topological context learning mechanism based on dual-branch hierarchical graph attention network, and a relation rule context learning mechanism based on Rule-Transformer and rule-to-relation aggregator. The former mechanism encodes the topological structure features of entities, aggregates the local neighborhood topological context information of entities on the three levels (entity, relation and triple), and build clusters of global head or tail entities related to the same relation. It can capture the local and global topological context information of entities related to the same relation. The latter mechanism introduces chain-like Horn rules as the context information of relations, and encodes the logical semantic of relations to enrich the relation representation. Experimental performances on three benchmark datasets FB15k-237, WN18RR and Kinship indicate the effectiveness and superiority of our proposed approach. The codes are publicly available.

Various euphemisms are emerging in social networks, attracting widespread attention from the natural language processing community. However, existing euphemism datasets are only domain-specific or language-specific. In addition, existing approaches to the study of euphemisms are one-sided. Either only the euphemism detection task or only the euphemism identification task is accomplished, lacking a unified framework. To this end, we construct a large-scale Bilingual Multi-category dataset of Euphemisms named BME, which covers a total of 12 categories for two languages, English and Chinese. Then, we first propose a unified generative model to Jointly conduct the tasks of bilingual Euphemism Detection and Identification named JointEDI. By comparing with LLMs and human evaluation, we demonstrate the effectiveness of the proposed JointEDI and the feasibility of unifying euphemism detection and euphemism identification tasks. Moreover, the BME dataset also provides a new reference standard for euphemism detection and euphemism identification.

Text ranking is a critical task in information retrieval. Recent advances in pre-trained language models (PLMs), especially large language models (LLMs), present new opportunities for applying them to text ranking. While supervised fine-tuning (SFT) with ranking data has been widely explored to better align PLMs with text ranking goals, previous studies have focused primarily on encoder-only and encoder-decoder PLMs. Research on leveraging decoder-only LLMs for text ranking remains scarce. An exception to this is RankLLaMA, which uses direct SFT to explore LLaMA’s potential for text ranking. In this work, we propose a two-stage progressive paradigm to better adapt LLMs to text ranking. First, we conduct continual pre-training (CPT) of LLMs on a large weakly-supervised corpus. Second, we perform SFT, and propose an improved optimization strategy building upon RankLLaMA. Our experimental results on multiple benchmarks show that our approach outperforms previous methods in both in-domain and out-domain scenarios.

Relation triplet extraction is a fundamental task in natural language processing that aims to identify semantic relationships between entities in text. It is particularly challenging in the zero-shot setting, i.e., zero-shot relation triplet extraction (ZeroRTE), where the relation sets between training and test are disjoint. Existing methods deal with this task by integrating relations into prompts, which may lack sufficient understanding of the unseen relations. To address these limitations, this paper presents a novel Two-Agent Game (TAG) approach to deliberate and debate the semantics of unseen relations. TAG consists of two agents, a generator and an extractor. They iteratively interact in three key steps: attempting, criticizing, and rectifying. This enables the agents to fully debate and understand the unseen relations. Experimental results demonstrate consistent improvement over ALBERT-Large, BART, andGPT3.5, without incurring additional inference costs in all cases. Remarkably, our method outperforms strong baselines by a significant margin, achieving an impressive 6%-16% increase in F1 scores, particularly when dealingwith FewRel with five unseen relations.