EMNLP 2024 论文列表

The deployment of Large Language Models (LLMs) in content generation raises significant safety concerns, particularly regarding the transparency and interpretability of content evaluations. Current methods, primarily focused on binary safety classifications, lack mechanisms for detailed critique, limiting their utility for model improvement and user trust. To address these limitations, we introduce SAFETY-J, a bilingual generative safety evaluator for English and Chinese with critique-based judgment. SAFETY-J utilizes a robust training dataset that includes diverse dialogues and augmented query-response pairs to assess safety across various scenarios comprehensively. We establish an automated meta-evaluation benchmark that objectively assesses the quality of critiques with minimal human intervention, facilitating scalable and continuous improvement. Additionally, SAFETY-Jemploys an iterative preference learning technique to dynamically refine safety assessments based on meta-evaluations and critiques. Our evaluations demonstrate that SAFETY-J provides more nuanced and accurate safety evaluations, thereby enhancing both critique quality and predictive reliability in complex content scenarios. To facilitate further research and application, we have released SAFETY-J’s training protocols, datasets, and code at https://github.com/GAIR-NLP/Safety-J.

In an era where information is quickly shared across many cultural and language contexts, the neutrality and integrity of news media are essential. Ensuring that media content remains unbiased and factual is crucial for maintaining public trust. With this in mind, we introduce SAFARI (CroSs-lingual BiAs and Factuality Detection in News MediA and News ARtIcles), a novel corpus of news media and articles for predicting political bias and the factuality of reporting in a multilingual and cross-lingual setup. To the best of our knowledge, this corpus is unprecedented in its collection and introduces a dataset for political bias and factuality for three tasks: (i) media-level, (ii) article-level, and (iii) joint modeling at the article-level. At the media and article levels, we evaluate the cross-lingual ability of the models; however, in joint modeling, we evaluate on English data. Our frameworks set a new benchmark in the cross-lingual evaluation of political bias and factuality. This is achieved through the use of various Multilingual Pre-trained Language Models (MPLMs) and Large Language Models (LLMs) coupled with ensemble learning methods.

This study presents a novel learning approach designed to enhance both mathematical reasoning and problem-solving abilities of Large Language Models (LLMs). We focus on integrating the Chain-of-Thought (CoT) and the Program-of-Thought (PoT) learning, hypothesizing that prioritizing the learning of mathematical reasoning ability is helpful for the amplification of problem-solving ability. Thus, the initial learning with CoT is essential for solving challenging mathematical problems. To this end, we propose a sequential learning approach, named SAAS (Solving Ability Amplification Strategy), which strategically transitions from CoT learning to PoT learning. Our empirical study, involving an extensive performance comparison using several benchmarks, demonstrates that our SAAS achieves state-of-the-art (SOTA) performance. The results underscore the effectiveness of our sequential learning approach, marking a significant advancement in the field of mathematical reasoning in LLMs.

The instruction-following ability of large language models enables humans to interact with AI agents in a natural way. However, when required to generate responses of a specific length, large language models often struggle to meet users’ needs due to their inherent difficulty in accurately perceiving numerical constraints. To explore the ability of large language models to control the length of generated responses, we propose the Target Length Generation Task (TLG) and design two metrics, Precise Match (PM) and Flexible Match (FM) to evaluate the model’s performance in adhering to specified response lengths. Furthermore, we introduce a novel, model-agnostic approach called Ruler, which employs Meta Length Tokens (MLTs) to enhance the instruction-following ability of large language models under length-constrained instructions. Specifically, Ruler equips LLMs with the ability to generate responses of a specified length based on length constraints within the instructions. Moreover, Ruler can automatically generate appropriate MLT when length constraints are not explicitly provided, demonstrating excellent versatility and generalization. Comprehensive experiments show the effectiveness of Ruler across different LLMs on Target Length Generation Task, e.g., at All Level 27.97 average gain on PM, 29.57 average gain on FM. In addition, we conduct extensive ablation experiments to further substantiate the efficacy and generalization of Ruler. Our code and data is available on the internet.

Minimal pairs are a well-established approach to evaluating the grammatical knowledge of language models. However, existing resources for minimal pairs address a limited number of languages and lack diversity of language-specific grammatical phenomena. This paper introduces the Russian Benchmark of Linguistic Minimal Pairs (RuBLiMP), which includes 45k pairs of sentences that differ in grammaticality and isolate a morphological, syntactic, or semantic phenomenon. In contrast to existing benchmarks of linguistic minimal pairs, RuBLiMP is created by applying linguistic perturbations to automatically annotated sentences from open text corpora and decontaminating test data. We describe the data collection protocol and present the results of evaluating 25 language models in various scenarios. We find that the widely used LMs for Russian are sensitive to morphological and agreement-oriented contrasts, but fall behind humans on phenomena requiring the understanding of structural relations, negation, transitivity, and tense. RuBLiMP, the codebase, and other materials are publicly available.

Pre-trained language models, trained on large-scale corpora, demonstrate strong generalizability across various NLP tasks. Fine-tuning these models for specific tasks typically involves updating all parameters, which is resource-intensive. Parameter-efficient fine-tuning (PEFT) methods, such as the popular LoRA family, introduce low-rank matrices to learn only a few parameters efficiently. However, during inference, the product of these matrices updates all pre-trained parameters, complicating tasks like knowledge editing that require selective updates. We propose a novel PEFT method, which conducts row and column-wise sparse low-rank adaptation (RoseLoRA), to address this challenge. RoseLoRA identifies and updates only the most important parameters for a specific task, maintaining efficiency while preserving other model knowledge. By adding a sparsity constraint on the product of low-rank matrices and converting it to row and column-wise sparsity, we ensure efficient and precise model updates. Our theoretical analysis guarantees the lower bound of the sparsity with respective to the matrix product. Extensive experiments on five benchmarks across twenty datasets demonstrate that RoseLoRA outperforms baselines in both general fine-tuning and knowledge editing tasks.

Recent works leverage LLMs to roleplay realistic social scenarios, aiding novices in practicing their social skills. However, simulating sensitive interactions, such as in the domain of mental health, is challenging. Privacy concerns restrict data access, and collecting expert feedback, although vital, is laborious. To address this, we develop Roleplay-doh, a novel human-LLM collaboration pipeline that elicits qualitative feedback from a domain-expert, which is transformed into a set of principles, or natural language rules, that govern an LLM-prompted roleplay. We apply this pipeline to enable senior mental health supporters to create customized AI patients as simulated practice partners for novice counselors. After uncovering issues with basic GPT-4 simulations not adhering to expert-defined principles, we also introduce a novel principle-adherence prompting pipeline which shows a 30% improvement in response quality and principle following for the downstream task. Through a user study with 25 counseling experts, we demonstrate that the pipeline makes it easy and effective to create AI patients that more faithfully resemble real patients, as judged by both creators and third-party counselors. We provide access to the code and data on our project website: https://roleplay-doh.github.io/.

Downstream applications often require text classification models to be accurate and robust. While the accuracy of state-of-the-art Language Models (LMs) approximates human performance, they often exhibit a drop in performance on real-world noisy data. This lack of robustness can be concerning, as even small perturbations in text, irrelevant to the target task, can cause classifiers to incorrectly change their predictions. A potential solution can be the family of Prototype-Based Networks (PBNs) that classifies examples based on their similarity to prototypical examples of a class (prototypes) and has been shown to be robust to noise for computer vision tasks. In this paper, we study whether the robustness properties of PBNs transfer to text classification tasks under both targeted and static adversarial attack settings. Our results show that PBNs, as a mere architectural variation of vanilla LMs, offer more robustness compared to vanilla LMs under both targeted and static settings. We showcase how PBNs’ interpretability can help us understand PBNs’ robustness properties. Finally, our ablation studies reveal the sensitivity of PBNs’ robustness to the strictness of clustering and the number of prototypes in the training phase, as tighter clustering and a low number of prototypes result in less robust PBNs.

Error correction (EC) based on large language models is an emerging technology to enhance the performance of automatic speech recognition (ASR) systems.Generally, training data for EC are collected by automatically pairing a large set of ASR hypotheses (as sources) and their gold references (as targets).However, the quality of such pairs is not guaranteed, and we observed various types of noise which can make the EC models brittle, e.g. inducing overcorrection in out-of-domain (OOD) settings.In this work, we propose two fundamental criteria that EC training data should satisfy: namely, EC targets should (1) improve linguistic acceptability over sources and (2) be inferable from the available context (e.g. source phonemes).Through these criteria, we identify low-quality EC pairs and train the models not to make any correction in such cases, the process we refer to as conservative data filtering.In our experiments, we focus on Japanese ASR using a strong Conformer-CTC as the baseline and finetune Japanese LLMs for EC.Through our evaluation on a suite of 21 internal benchmarks, we demonstrate that our approach can significantly reduce overcorrection and improve both the accuracy and quality of ASR results in the challenging OOD settings.

Growing amount and quality of AI-generated texts makes detecting such content more difficult. In most real-world scenarios, the domain (style and topic) of generated data and the generator model are not known in advance. In this work, we focus on the robustness of classifier-based detectors of AI-generated text, namely their ability to transfer to unseen generators or semantic domains. We investigate the geometry of the embedding space of Transformer-based text encoders and show that clearing out harmful linear subspaces helps to train a robust classifier, ignoring domain-specific spurious features. We investigate several subspace decomposition and feature selection strategies and achieve significant improvements over state of the art methods in cross-domain and cross-generator transfer. Our best approaches for head-wise and coordinate-based subspace removal increase the mean out-of-distribution (OOD) classification score by up to 9% and 14% in particular setups for RoBERTa and BERT embeddings respectively. We release our code and data: https://github.com/SilverSolver/RobustATD

Tool learning has generated widespread interest as a vital means of interaction between Large Language Models (LLMs) and the physical world. Current research predominantly emphasizes LLMs’ capacity to utilize tools in well-structured environments while overlooking their stability when confronted with the inevitable noise of the real world. To bridge this gap, we introduce *RoTBench*, a multi-level benchmark for evaluating the robustness of LLMs in tool learning. Specifically, we establish five external environments, each featuring varying levels of noise (i.e., Clean, Slight, Medium, Heavy, and Union), providing an in-depth analysis of the model’s resilience across three critical phases: tool selection, parameter identification, and content filling. Experiments involving six widely-used models underscore the urgent necessity for enhancing the robustness of LLMs in tool learning. For instance, the performance of GPT-4 even drops significantly from 80.00 to 58.10 when there is no substantial change in manual accuracy. More surprisingly, the noise correction capability inherent in the GPT family paradoxically impedes its adaptability in the face of mild noise. In light of these findings, we propose RoTTuning, a strategy that enriches the diversity of training environments to bolster the robustness of LLMs in tool learning. The code and data are available at https://github.com/Junjie-Ye/RoTBench.

The remarkable achievements obtained by open-source large language models (LLMs) in recent years have predominantly been concentrated on tasks involving the English language. In this paper, we aim to advance the performance of Llama2 models on Romanian tasks. We tackle the problem of reduced computing resources by using QLoRA for training. We release RoQLlama-7b, a quantized LLM, which shows equal or improved results compared to its full-sized counterpart when tested on seven Romanian downstream tasks in the zero-shot setup. Also, it consistently achieves higher average scores across all few-shot prompts. Additionally, we introduce a novel Romanian dataset, namely RoMedQA, which contains single-choice medical questions in Romanian.

Low-Rank Adaptation (LoRA), as a representative Parameter-Efficient Fine-Tuning (PEFT) method, significantly enhances the training efficiency by updating only a small portion of the weights in Large Language Models (LLMs). Recently, weight-only quantization techniques have also been applied to LoRA methods to reduce the memory footprint of fine-tuning. However, applying weight-activation quantization to the LoRA pipeline is under-explored, and we observe substantial performance degradation primarily due to the presence of activation outliers. In this work, we propose RoLoRA, the first LoRA-based scheme to apply rotation for outlier elimination, and then fine-tune rotated outlier-free LLMs for effective weight-activation quantization. Different from previous work tackling the outlier challenges from a post-training perspective, we propose rotation-aware fine-tuning to eliminate and preserve the outlier-free characteristics brought by rotation operations. RoLoRA can improve low-bit LoRA convergence and post-training quantization robustness in weight-activation settings. RoLoRA is evaluated across various LLM series (LLaMA2, LLaMA3, LLaVA-1.5), tasks, and quantization settings, achieving up to 29.5% absolute accuracy gain of 4-bit weight-activation quantized LLaMA2-13B on commonsense reasoning tasks compared to LoRA baseline. We further demonstrate its effectiveness on Large Multimodal Models (LMMs) and prove the compatibility with advanced LoRA variants.

Table entity linking (TEL) aims to map entity mentions in the table to their corresponding entities in a knowledge base (KB). The core of this task is to leverage structured contexts, specifically row and column contexts, to enhance the semantics of mentions in entity disambiguation. Most entity linking (EL) methods primarily focus on understanding sequential text contexts, making it difficult to adapt to the row and column structure of tables. Additionally, existing methods for TEL indiscriminately mix row and column contexts together, overlooking their semantic differences. In this paper, we explicitly distinguish the modeling of row and column contexts, and propose a method called RoCEL to capture their distinct semantics. Specifically, for row contexts in tables, we take the attention mechanism to learn the implicit relational dependencies between each cell and the mention. For column contexts in tables, we employ a set-wise encoder to learn the categorical information about the group of mentions. At last, we merge both contexts to obtain the final mention embedding for link prediction. Experiments on four benchmarks show that our approach outperforms the state-of-the-art (SOTA) baseline by about 1.5% on the in-domain dataset, and by 3.7% on average across three out-of-domain datasets.

Investigating “Hadith” texts, crucial for theological studies and Islamic jurisprudence, presents challenges due to the linguistic complexity of Arabic, such as its complex morphology. In this paper, we propose an innovative approach to address the challenges of topic modeling in Hadith studies by utilizing the Contextualized Topic Model (CTM). Our study introduces RoBERT2VecTM, a novel neural-based approach that combines the RoBERTa transformer model with Doc2Vec, specifically targeting the semantic analysis of “Matn” (the actual content). The methodology outperforms many traditional state-of-the-art NLP models by generating more coherent and diverse Arabic topics. The diversity of the generated topics allows for further categorization, deepening the understanding of discussed concepts. Notably, our research highlights the critical impact of lemmatization and stopwords in enhancing topic modeling. This breakthrough marks a significant stride in applying NLP to non-Latin languages and opens new avenues for the nuanced analysis of complex religious texts.

Knowledge Graph Question Answering (KGQA) methods seek to answer Natural Language questions using the relational information stored in Knowledge Graphs (KGs). With the recent advancements of Large Language Models (LLMs) and their remarkable reasoning abilities, there is a growing trend to leverage them for KGQA. However, existing methodologies have only focused on answering factual questions, e.g., *“In which city was Silvio Berlusconi’s first wife born?”*, leaving questions involving commonsense reasoning that real-world users may pose more often, e.g., *“Do I need separate visas to see the Venus of Willendorf and attend the Olympics this summer?”* unaddressed. In this work, we first observe that existing LLM-based methods for KGQA struggle with hallucination on such questions, especially on queries targeting long-tail entities (e.g., non-mainstream and recent entities), thus hindering their applicability in real-world applications especially since their reasoning processes are not easily verifiable. In response, we propose Right for Right Reasons (R3), a commonsense KGQA methodology that allows for a verifiable reasoning procedure by axiomatically surfacing intrinsic commonsense knowledge of LLMs and grounding every factual reasoning step on KG triples. Through experimental evaluations across three different tasks—question answering, claim verification, and preference matching—our findings showcase R3 as a superior approach, outperforming existing methodologies and notably reducing instances of hallucination and reasoning errors.

Reinforcement learning (RL) is a powerful approach to enhance task-oriented dialogue (TOD) systems. However, existing RL methods tend to mainly focus on generation tasks, such as dialogue policy learning (DPL) or response generation (RG), while neglecting dialogue state tracking (DST) for understanding. This narrow focus limits the systems to achieve globally optimal performance by overlooking the interdependence between understanding and generation. Additionally, RL methods face challenges with sparse and delayed rewards, which complicates training and optimization. To address these issues, we extend RL into both understanding and generation tasks by introducing step-by-step rewards throughout the token generation. The understanding reward increases as more slots are correctly filled in DST, while the generation reward grows with the accurate inclusion of user requests. Our approach provides a balanced optimization aligned with task completion. Experimental results demonstrate that our approach effectively enhances the performance of TOD systems and achieves new state-of-the-art results on three widely used datasets, including MultiWOZ2.0, MultiWOZ2.1, and In-Car. Our approach also shows superior few-shot ability in low-resource settings compared to current models.

In Reinforcement Learning from Human Feedback (RLHF), the reward model plays a crucial role in aligning language model outputs with human values. The human preference data used to train the reward model consists of a prompt and a response pair, with humans annotating which response better aligns with human value preferences. Due to the complexity and subjectivity of the annotation task, multiple organizations including OpenAI and Anthropic report significant noise in the human preference datasets, leading to instability and deviation in reward model training from human values. We discover that the difference in scores assigned to response pairs by the reward model effectively indicates the quality of data, and data of varying qualities show significant distinctions in reward model training. We introduce a method that automatically adjusts reward modeling based on data quality, reducing the impact of noise and making full use of dataset. Experiments on multiple human preference datasets demonstrate that our method stabilizes reward model training and significantly enhances the alignment performance of RLHF.

With the wide deployment of Large Language Models (LLMs), aligning LLMs with human values becomes increasingly important. Although Reinforcement Learning with Human Feedback (RLHF) proves effective, it is complicated and highly resource-intensive. As such, offline RLHF has been introduced as an alternative solution, which directly optimizes LLMs with ranking losses on a fixed preference dataset. Current offline RLHF only captures the ordering relationship between responses, overlooking the crucial aspect of “how much” one is preferred over the others. To address this issue, we propose a simple yet effective solution based on reward difference prediction. Specifically, we introduce reward difference coefficients to reweigh sample pairs in offline RLHF. We then propose a difference model that considers rich interactions between a pair of responses for predicting these difference coefficients. Experiments with 7B LLMs on the HH and TL;DR dataset verify the effectiveness of our method in both automatic metrics and human evaluation, highlighting its potential for aligning LLMs with human values.