EMNLP 2024 Papers

In dialogue, the addressee may initially misunderstand the speaker and respond erroneously, often prompting the speaker to correct the misunderstanding in the next turn with a Third Position Repair (TPR). The ability to process and respond appropriately to such repair sequences is thus crucial in conversational AI systems. In this paper, we first collect, analyse, and publicly release BlockWorld-Repairs: a dataset of multi-modal TPR sequences in an instruction-following manipulation task that is, by design, rife with referential ambiguity. We employ this dataset to evaluate several state-of-the-art Vision and Language Models (VLM) across multiple settings, focusing on their capability to process and accurately respond to TPRs and thus recover from miscommunication. We find that, compared to humans, all models significantly underperform in this task. We then show that VLMs can benefit from specialised losses targeting relevant tokens during fine-tuning, achieving better performance and generalising better to new scenarios. Our results suggest that these models are not yet ready to be deployed in multi-modal collaborative settings where repairs are common, and highlight the need to design training regimes and objectives that facilitate learning from interaction. Our code and data are available at www.github.com/JChiyah/blockworld-repairs

Text-to-Image Diffusion Models (T2I DMs) have garnered significant attention for their ability to generate high-quality images from textual descriptions.However, these models often produce images that do not fully align with the input prompts, resulting in semantic inconsistencies.The most prominent issue among these semantic inconsistencies is catastrophic-neglect, where the images generated by T2I DMs miss key objects mentioned in the prompt.We first conduct an empirical study on this issue, exploring the prevalence of catastrophic-neglect, potential mitigation strategies with feature enhancement, and the insights gained.Guided by the empirical findings, we propose an automated repair approach named Patcher to address catastrophic-neglect in T2I DMs.Specifically, Patcher first determines whether there are any neglected objects in the prompt, and then applies attention-guided feature enhancement to these neglected objects, resulting in a repaired prompt.Experimental results on three versions of Stable Diffusion demonstrate that Patcher effectively repairs the issue of catastrophic-neglect, achieving 10.1%-16.3% higher Correct Rate in image generation compared to baselines.

Advances in dataset analysis techniques have enabled more sophisticated approaches to analyzing and characterizing training data instances, often categorizing data based on attributes such as “difficulty”. In this work, we introduce RepMatch, a novel method that characterizes data through the lens of similarity.RepMatch quantifies the similarity between subsets of training instances by comparing the knowledge encoded in models trained on them, overcoming the limitations of existing analysis methods that focus solely on individual instances and are restricted to within-dataset analysis.Our framework allows for a broader evaluation, enabling similarity comparisons across arbitrary subsets of instances, supporting both dataset-to-dataset and instance-to-dataset analyses. We validate the effectiveness of RepMatch across multiple NLP tasks, datasets, and models. Through extensive experimentation, we demonstrate that RepMatch can effectively compare datasets, identify more representative subsets of a dataset (that lead to better performance than randomly selected subsets of equivalent size), and uncover heuristics underlying the construction of some challenge datasets.

The era of Large Language Models (LLMs) raises new demands for automatic evaluation metrics, which should be adaptable to various application scenarios while maintaining low cost and effectiveness. Traditional metrics for automatic text evaluation are often tailored to specific scenarios, while LLM-based evaluation metrics are costly, requiring fine-tuning or rely heavily on the generation capabilities of LLMs. Besides, previous LLM-based metrics ignore the fact that, within the space of LLM representations, there exist direction vectors that indicate the estimation of text quality. To this end, we introduce RepEval, a metric that leverages the projection of LLM representations for evaluation. Through simple prompt modifications, RepEval can easily transition to various tasks, requiring only minimal sample pairs for direction vector construction. Results on fourteen datasets across two evaluation tasks demonstrate the high effectiveness of our method, which exhibits a higher correlation with human judgments than previous methods, even in complex evaluation scenarios involving pair-wise selection under nuanced aspects. Our work underscores the richness of information regarding text quality embedded within LLM representations, offering insights for the development of new metrics.

Large Language Models (LLMs) are increasingly ubiquitous, yet their ability to retain and reason about temporal information remains limited, hindering their application in real-world scenarios where understanding the sequential nature of events is crucial. Our study experiments with 12 state-of-the-art models (ranging from 2B to 70B+ parameters) on a novel numerical-temporal dataset, TempUN, spanning from 10,000 BCE to 2100 CE, to uncover significant temporal retention and comprehension limitations. We propose six metrics to assess three learning paradigms to enhance temporal knowledge acquisition. Our findings reveal that open-source models exhibit knowledge gaps more frequently, suggesting a trade-off between limited knowledge and incorrect responses. Additionally, various fine-tuning approaches significantly improved performance, reducing incorrect outputs and impacting the identification of ‘information not available’ in the generations. The associated dataset and code are available at the [URL](https://anonymous.4open.science/r/TempUN-ARR/).

Retrieving accurate domain knowledge and providing helpful information are crucial in developing an effective end-to-end task-oriented dialogue system (E2ETOD). The field has witnessed numerous methods following the retrieve-then-generate paradigm and training their systems on one specific domain. However, existing approaches still suffer from the Distractive Attributes Problem (DAP): struggling to deal with false but similar knowledge (hard negative entities), which is even more intractable when countless pieces of knowledge from different domains are blended in a real-world scenario. To alleviate DAP, we propose the Relevance-aware Adaptive Learning (ReAL) method, a two-stage training framework that eliminates hard negatives step-by-step and aligns retrieval with generation. In the first stage, we introduce a top-k adaptive contrastive loss and utilize the divergence-driven feedback from the frozen generator to pre-train the retriever. In the second stage, we propose using the metric score distribution as an anchor to align retrieval with generation. Thorough experiments on three benchmark datasets demonstrate ReAL’s superiority over existing methods, with extensive analysis validating its strong capabilities of overcoming in- and cross-domain distractions.

To convince readers of the novelty of their research paper, authors must perform a literature review and compose a coherent story that connects and relates prior works to the current work. This challenging nature of literature review writing makes automatic related work generation (RWG) academically and computationally interesting, and also makes it an excellent test bed for examining the capability of SOTA natural language processing (NLP) models. Since the initial proposal of the RWG task, its popularity has waxed and waned, following the capabilities of mainstream NLP approaches. In this work, we survey the zoo of RWG historical works, summarizing the key approaches and task definitions and discussing the ongoing challenges of RWG.

Large language models (LLMs) have shown strong results on a range of applications, including regression and scoring tasks.Typically, one obtains outputs from an LLM via autoregressive sampling from the model’s output distribution. We show that this inference strategy can be sub-optimal for common regression and scoring evaluation metrics. As a remedy, we build on prior work on Minimum Bayes Risk decoding,and propose alternate inference strategies that estimate the Bayes-optimal solution for regression and scoring metrics in closed-form from sampled responses.We show that our proposal significantly improves over baselines across datasets and models.

The quality of finetuning data is crucial for aligning large language models (LLMs) with human values. Current methods to improve data quality are either labor-intensive or prone to factual errors caused by LLM hallucinations. This paper explores elevating the quality of existing instruction data to better align with human values, introducing a simple and effective approach named ReAlign, which reformats the responses of instruction data into a format that better aligns with pre-established criteria and the collated evidence. This approach minimizes human annotation, hallucination, and the difficulty in scaling, remaining orthogonal to existing alignment techniques. Experimentally, ReAlign significantly boosts the general alignment ability, math reasoning, factuality, and readability of the LLMs.Encouragingly, without introducing any additional data or advanced training techniques, and merely by reformatting the response, LLaMA-2-13B’s mathematical reasoning ability on GSM8K can be improved **from 46.77% to 56.63%** in accuracy. Additionally, a mere 5% of ReAlign data yields a 67% boost in general alignment ability measured by the Alpaca dataset. This work highlights the need for further research into the science and mechanistic interpretability of LLMs. We have made the associated code and data publicly accessible to support future studies at https://github.com/GAIR-NLP/ReAlign.

With the growing number of mobile users, app development has become increasingly lucrative. Reviews on platforms such as Google Play and Apple App Store provide valuable insights to developers, highlighting bugs, suggesting new features, and offering feedback. However, many reviews contain typos, spelling errors, grammar mistakes, and complex sentences, hindering efficient interpretation and slowing down app improvement processes. To tackle this, we introduce RARE (Repository for App review REfinement), a benchmark dataset of 10,000 annotated pairs of original and refined reviews from 10 mobile applications. These reviews were collaboratively refined by humans and large language models (LLMs). We also conducted an evaluation of eight state-of-the-art LLMs for automated review refinement. The top-performing model (Flan-T5) was further used to refine an additional 10,000 reviews, contributing to RARE as a silver corpus.

Large vision-language models (LVLMs) have made significant progress in recent years. While LVLMs exhibit excellent ability in language understanding, question answering, and conversations of visual inputs, they are prone to producing hallucinations. While several methods are proposed to evaluate the hallucinations in LVLMs, most are reference-based and depend on external tools, which complicates their practical application. To assess the viability of alternative methods, it is critical to understand whether the reference-free approaches, which do not rely on any external tools, can efficiently detect hallucinations. Therefore, we initiate an exploratory study to demonstrate the effectiveness of different reference-free solutions in detecting hallucinations in LVLMs. In particular, we conduct an extensive study on three kinds of techniques: uncertainty-based, consistency-based, and supervised uncertainty quantification methods on four representative LVLMs across two different tasks. The empirical results show that the reference-free approaches are capable of effectively detecting non-factual responses in LVLMs, with the supervised uncertainty quantification method outperforming the others, achieving the best performance across different settings.

Reference-based metrics such as BLEU and BERTScore are widely used to evaluate question generation (QG). In this study, on QG benchmarks such as SQuAD and HotpotQA, we find that using human-written references cannot guarantee the effectiveness of the reference-based metrics. Most QG benchmarks have only one reference; we replicate the annotation process and collect another reference. A good metric is expected to grade a human-validated question no worse than generated questions. However, the results of reference-based metrics on our newly collected reference disproved the metrics themselves. We propose a reference-free metric consisted of multi-dimensional criteria such as naturalness, answerability, and complexity, utilizing large language models. These criteria are not constrained to the syntactic or semantic of a single reference question, and the metric does not require a diverse set of references. Experiments reveal that our metric accurately distinguishes between high-quality questions and flawed ones, and achieves state-of-the-art alignment with human judgment.

Most language models currently available are prone to self-contradiction during dialogues. To mitigate this issue, this study explores a novel contradictory dialogue processing task that aims to detect and modify contradictory statements in a conversation. This task is inspired by research on context faithfulness and dialogue comprehension, which have demonstrated that the detection and understanding of contradictions often necessitate detailed explanations. We develop a dataset comprising contradictory dialogues, in which one side of the conversation contradicts itself. Each dialogue is accompanied by an explanatory label that highlights the location and details of the contradiction. With this dataset, we present a Red Teaming framework for contradictory dialogue processing. The framework detects and attempts to explain the dialogue, then modifies the existing contradictory content using the explanation. Our experiments demonstrate that the framework improves the ability to detect contradictory dialogues and provides valid explanations. Additionally, it showcases distinct capabilities for modifying such dialogues. Our study highlights the importance of the logical inconsistency problem in conversational AI.

While large language models (LLMs) excel at understanding and generating plain text, they are not tailored to handle hierarchical text structures or directly predict task-specific properties such as text rating. In fact, selectively and repeatedly grasping the hierarchical structure of large-scale text is pivotal for deciphering its essence. To this end, we propose a novel framework for hierarchical text rating utilizing LLMs, which incorporates Recurrent Alignment with Hard Attention (RAHA). Particularly, hard attention mechanism prompts a frozen LLM to selectively focus on pertinent leaf texts associated with the root text and generate symbolic representations of their relationships. Inspired by the gradual stabilization of the Markov Chain, recurrent alignment strategy involves feeding predicted ratings iteratively back into the prompts of another trainable LLM, aligning it to progressively approximate the desired target. Experimental results demonstrate that RAHA outperforms existing state-of-the-art methods on three hierarchical text rating datasets. Theoretical and empirical analysis confirms RAHA’s ability to gradually converge towards the underlying target through multiple inferences. Additional experiments on plain text rating datasets verify the effectiveness of this Markov-like alignment. Our data and code can be available in https://github.com/ECNU-Text-Computing/Markov-LLM.

Significant advancements have recently been made in large language models, represented by GPT models.Users frequently have multi-round private conversations with cloud-hosted GPT models for task optimization.Yet, this operational paradigm introduces additional attack surfaces, particularly in custom GPTs and hijacked chat sessions.In this paper, we introduce a straightforward yet potent Conversation Reconstruction Attack.This attack targets the contents of previous conversations between GPT models and benign users, i.e., the benign users’ input contents during their interaction with GPT models.The adversary could induce GPT models to leak such contents by querying them with designed malicious prompts.Our comprehensive examination of privacy risks during the interactions with GPT models under this attack reveals GPT-4’s considerable resilience.We present two advanced attacks targeting improved reconstruction of past conversations, demonstrating significant privacy leakage across all models under these advanced techniques.Evaluating various defense mechanisms, we find them ineffective against these attacks.Our findings highlight the ease with which privacy can be compromised in interactions with GPT models, urging the community to safeguard against potential abuses of these models’ capabilities.

Historically, sign language machine translation has been posed as a sentence-level task: datasets consisting of continuous narratives are chopped up and presented to the model as isolated clips. In this work, we explore the limitations of this task framing. First, we survey a number of linguistic phenomena in sign languages that depend on discourse-level context. Then as a case study, we perform the first human baseline for sign language translation that actually substitutes a human into the machine learning task framing, rather than provide the human with the entire document as context. This human baseline—for ASL to English translation on the How2Sign dataset—shows that for 33% of sentences in our sample, our fluent Deaf signer annotators were only able to understand key parts of the clip in light of additional discourse-level context. These results underscore the importance of understanding and sanity checking examples when adapting machine learning to new domains.

Large Language Models (LLMs), such as GPT and LLaMA, are susceptible to generating hallucinated answers in a confident tone. While previous efforts to elicit and calibrate confidence scores have shown some success, they often overlook biases towards certain groups, such as specific nationalities. Existing calibration methods typically focus on average performance, failing to address this disparity. In our study, we demonstrate that the concept of grouping loss is an effective metric for understanding and correcting the heterogeneity in confidence levels. We introduce a novel evaluation dataset, derived from a knowledge base, specifically designed to assess the confidence scores of LLM responses across different groups. Our experimental results highlight significant variations in confidence, which are accurately captured by grouping loss. To tackle this issue, we propose a new method to calibrate the confidence scores of LLMs by considering different groups, a process we term reconfidencing. Our findings indicate that this approach effectively mitigates biases against minority groups, contributing to the development of fairer LLMs.

Linear Text Segmentation is the task of automatically tagging text documents with topic shifts, i.e. the places in the text where the topics change. A well-established area of research in Natural Language Processing, drawing from well-understood concepts in linguistic and computational linguistic research, the field has recently seen a lot of interest as a result of the surge of text, video, and audio available on the web, which in turn require ways of summarising and categorizing the mole of content for which linear text segmentation is a fundamental step. In this survey, we provide an extensive overview of current advances in linear text segmentation, describing the state of the art in terms of resources and approaches for the task. Finally, we highlight the limitations of available resources and of the task itself, while indicating ways forward based on the most recent literature and under-explored research directions.

Moderating hate speech (HS) in the evolving online landscape is a complex challenge, compounded by the multimodal nature of digital content. This survey examines recent advancements in HS moderation, focusing on the burgeoning role of large language models (LLMs) and large multimodal models (LMMs) in detecting, explaining, debiasing, and countering HS. We begin with a comprehensive analysis of current literature, uncovering how text, images, and audio interact to spread HS. The combination of these modalities adds complexity and subtlety to HS dissemination. We also identified research gaps, particularly in underrepresented languages and cultures, and highlight the need for solutions in low-resource settings. The survey concludes with future research directions, including novel AI methodologies, ethical AI governance, and the development of context-aware systems. This overview aims to inspire further research and foster collaboration towards responsible and human-centric approaches to HS moderation in the digital age.