EMNLP 2024 Papers

Retrieval-augmented generation (RAG) has emerged as a significant advancement in the field of large language models (LLMs). By integrating up-to-date information not available during their initial training, RAG greatly enhances the practical utility of LLMs in real-world applications. However, even with RAG, LLMs can still produce inaccurate outputs, such as distorting or misinterpreting source content, posing risks in high-trust scenarios. To address these issues, we introduce a novel approach called Hallucination Aware Tuning (HAT). This method involves training hallucination detection models that generate detection labels and provide detailed descriptions of the detected hallucinations. Utilizing these detection results—particularly the hallucination descriptions—GPT-4 Turbo is employed to correct any detected hallucinations. The corrected outputs, free of hallucinations, along with the original versions, are used to create a preference dataset for Direct Preference Optimization (DPO) training. The fine-tuning through DPO leads to LLMs that exhibit a reduced rate of hallucinations and deliver improved answer quality.

Large language models (LLMs) have exhibited remarkable fluency across various tasks. However, their unethical applications, such as disseminating disinformation, have become a growing concern. Although recent works have proposed a number of LLM detection methods, their robustness and reliability remain unclear. In this paper, we present RAFT: a grammar error-free black-box attack against existing LLM detectors. In contrast to previous attacks for language models, our method exploits the transferability of LLM embeddings at the word-level while preserving the original text quality. We leverage an auxiliary embedding to greedily select candidate words to perturb against the target detector. Experiments reveal that our attack effectively compromises all detectors in the study across various domains by up to 99%, and are transferable across source models. Manual human evaluation studies show our attacks are realistic and indistinguishable from original human-written text. We also show that examples generated by RAFT can be used to train adversarially robust detectors. Our work shows that current LLM detectors are not adversarially robust, underscoring the urgent need for more resilient detection mechanisms.

In recent years, significant advancements in conversational question and answering (CQA) have been driven by the exponential growth of large language models and the integration of retrieval mechanisms that leverage external knowledge to generate accurate and contextually relevant responses. Consequently, the fields of conversational search and retrieval-augmented generation (RAG) have obtained substantial attention for their capacity to address two key challenges: query rewriting within conversational histories for better retrieval performance and generating responses by employing retrieved knowledge. However, both fields are often independently studied, and comprehensive study on entire systems remains underexplored. In this work, we present a novel retrieval-augmented conversation (RAC) dataset and develop a baseline system comprising query rewriting, retrieval, reranking, and response generation stages. Experimental results demonstrate the competitiveness of the system and extensive analyses are conducted to apprehend the impact of retrieval results to response generation.

Generating faithful and fast responses is crucial in the knowledge-grounded dialogue. Retrieval Augmented Generation (RAG) strategies are effective but are inference inefficient, while previous Retrieval Free Generations (RFG) are more efficient but sacrifice faithfulness. To solve this faithfulness-efficiency trade-off dilemma, we propose a novel retrieval-free model training scheme named Retrieval Augmented to Retrieval Free Distillation (RA2FD) to build a retrieval-free model that achieves higher faithfulness than the previous RFG method while maintaining inference efficiency. The core idea of RA2FD is to use a teacher-student framework to distill the faithfulness capacity of a teacher, which is an oracle RAG model that generates multiple knowledge-infused responses. The student retrieval-free model learns how to generate faithful responses from these teacher labels through sequence-level distillation and contrastive learning. Experiment results show that RA2FD let the faithfulness performance of an RFG model surpass the previous SOTA RFG baseline on three knowledge-grounded dialogue datasets by an average of 33% and even matching an RAG model’s performance while significantly improving inference efficiency. Our code is available at https://github.com/zzysjtuiwct/RA2FD.

While current tasks of converting natural language to SQL (NL2SQL) using Foundation Models have shown impressive achievements, adapting these approaches for converting natural language to Graph Query Language (NL2GQL) encounters hurdles due to the distinct nature of GQL compared to SQL, alongside the diverse forms of GQL. Moving away from traditional rule-based and slot-filling methodologies, we introduce a novel approach, R3-NL2GQL, integrating both small and large Foundation Models for ranking, rewriting, and refining tasks. This method leverages the interpretative strengths of smaller models for initial ranking and rewriting stages, while capitalizing on the superior generalization and query generation prowess of larger models for the final transformation of natural language queries into GQL formats. Addressing the scarcity of datasets in this emerging field, we have developed a bilingual dataset, sourced from graph database manuals and selected open-source Knowledge Graphs (KGs). Our evaluation of this methodology on this dataset demonstrates its promising efficacy and robustness.

Retrieval augmented generation (RAG) has been applied in many scenarios to augment large language models (LLMs) with external documents provided by retrievers. However, a semantic gap exists between LLMs and retrievers due to differences in their training objectives and architectures. This misalignment forces LLMs to passively accept the documents provided by the retrievers, leading to incomprehension in the generation process, where the LLMs are burdened with the task of distinguishing these documents using their inherent knowledge. This paper proposes R2AG, a novel enhanced RAG framework to fill this gap by incorporating **R**etrieval information into **R**etrieval **A**ugmented **G**eneration. Specifically, R2AG utilizes the nuanced features from the retrievers and employs a R2-Former to capture retrieval information. Then, a retrieval-aware prompting strategy is designed to integrate retrieval information into LLMs’ generation. Notably, R2AG suits low-source scenarios where LLMs and retrievers are frozen. Extensive experiments across five datasets validate the effectiveness, robustness, and efficiency of R2AG. Our analysis reveals that retrieval information serves as an anchor to aid LLMs in the generation process, thereby filling the semantic gap.

Large language models (LLMs) have exhibited great potential in autonomously completing tasks across real-world applications. Despite this, these LLM agents introduce unexpected safety risks when operating in interactive environments. Instead of centering on the harmlessness of LLM-generated content in most prior studies, this work addresses the imperative need for benchmarking the behavioral safety of LLM agents within diverse environments. We introduce R-Judge, a benchmark crafted to evaluate the proficiency of LLMs in judging and identifying safety risks given agent interaction records. R-Judge comprises 569 records of multi-turn agent interaction, encompassing 27 key risk scenarios among 5 application categories and 10 risk types. It is of high-quality curation with annotated safety labels and risk descriptions. Evaluation of 11 LLMs on R-Judge shows considerable room for enhancing the risk awareness of LLMs: The best-performing model, GPT-4o, achieves 74.42% while no other models significantly exceed the random. Moreover, we reveal that risk awareness in open agent scenarios is a multi-dimensional capability involving knowledge and reasoning, thus challenging for LLMs. With further experiments, we find that fine-tuning on safety judgment significantly improve model performance while straightforward prompting mechanisms fail. R-Judge is publicly available at Annoymous.

Knowledge graph question answering (KGQA) involves answering natural language questions by leveraging structured information stored in a knowledge graph. Typically, KGQA initially retrieve a targeted subgraph from a large-scale knowledge graph, which serves as the basis for reasoning models to address queries. However, the retrieved subgraph inevitably brings distraction information for knowledge utilization, impeding the model’s ability to perform accurate reasoning. To address this issue, we propose a Question-guided Knowledge Graph Re-scoring method (Q-KGR) to eliminate noisy pathways for the input question, thereby focusing specifically on pertinent factual knowledge.Moreover, we introduce Knowformer, a parameter-efficient method for injecting the re-scored knowledge graph into large language models to enhance their ability to perform factual reasoning.Extensive experiments on multiple KGQA benchmarks demonstrate the superiority of our method over existing systems.

The Transformer’s quadratic complexity with input length imposes an unsustainable computational load on large language models (LLMs). In contrast, the Selective Scan Structured State-Space Model, or Mamba, addresses this computational challenge effectively. This paper explores a query-based cross-modal projector designed to bolster Mamba’s efficiency for vision-language modeling by compressing visual tokens based on input through the cross-attention mechanism. This innovative projector also removes the need for manually designing the 2D scan order of original image features when converting them into an input sequence for Mamba LLM. Experimental results across various vision-language understanding benchmarks show that the proposed cross-modal projector enhances Mamba-based multimodal LLMs, boosting both performance and throughput.

This work focuses on the task of query-based meeting summarization in which the summary of a context (meeting transcript) is generated in response to a specific query. When using Large Language Models (LLMs) for this task, a new call to the LLM inference endpoint/API is required for each new query even if the context stays the same. However, repeated calls to the LLM inference endpoints would significantly increase the costs of using them in production, making LLMs impractical for many real-world use cases. To address this problem, in this paper, we investigate whether combining the queries for the same input context in a single prompt to minimize repeated calls can be successfully used in meeting summarization. In this regard, we conduct extensive experiments by comparing the performance of various popular LLMs: GPT-4, Gemini, Claude-3, LLaMA2, Mistral, Phi-3, and Qwen-2 in single-query and multi-query settings. We observe that the capability to reliably generate the response in the expected format is usually limited to closedsource LLMs, with most open-source LLMs lagging behind (except Mistral). We conclude that multi-query prompting could be useful to optimize the inference costs by significantly reducing calls to the inference endpoints/APIs for the task of meeting summarization.

Current research on tool learning primarily focuses on selecting the most effective tool from a wide array of options, often overlooking cost-effectiveness, a crucial factor in human problem-solving. In this paper, we address query routing for homogeneous tools by predicting both their performance and the associated cost required to accomplish a given task. We then assign queries to the optimal tools in a cost-effective manner. Our experimental results demonstrate that our method achieves higher performance at a lower cost compared to strong baseline approaches.

We develop quantum RNNs with cells based on Parametrised Quantum Circuits (PQCs). PQCs can provide a form of hybrid quantum-classical computation where the input and the output is in the form of classical data. The previous “hidden” state is the quantum state from the previous time-step, and an angle encoding is used to define a (non-linear) mapping from a classical word embedding into the quantum Hilbert space. Measurements of the quantum state provide classical statistics which are used for classification. We report results which are competitive with various RNN baselines on the Rotten Tomatoes dataset, as well as emulator results which demonstrate the feasibility of running such models on quantum hardware.

There is a scarcity of multilingual vision-language models that properly account for the perceptual differences that are reflected in image captions across languages and cultures. In this work, through a multimodal, multilingual retrieval case study, we quantify the existing lack of model flexibility. We empirically show performance gaps between training on captions that come from native German perception and captions that have been either machine-translated or human-translated from English into German. To address these gaps, we further propose and evaluate caption augmentation strategies. While we achieve mean recall improvements (+1.3), gaps still remain, indicating an open area of future work for the community.

Recent advancements in Large Language Models (LLMs) have facilitated the development of Multimodal LLMs (MLLMs). Despite their impressive capabilities, MLLMs often suffer from over-reliance on unimodal biases (e.g., language bias and vision bias), leading to incorrect answers in complex multimodal tasks. To investigate this issue, we propose a causal framework to interpret the biases in Visual Question Answering (VQA) problems. Within this framework, we conduct an in-depth causal analysis to assess the causal effect of these biases on MLLM predictions. Based on the analysis, we introduce 1) a novel MORE dataset with 12,000 challenging VQA instances requiring multi-hop reasoning and overcoming unimodal biases. 2) a causality-enhanced agent framework CAVE that guides models to comprehensively integrate information from different modalities and mitigate biases. Our experiments show that MLLMs perform poorly on MORE, indicating strong unimodal biases and limited semantic understanding. However, when integrated with our CAVE, promising improvements in reasoning and bias mitigation can be seen. These findings provide important insights for the development of more robust MLLMs and contribute to the broader goal of advancing multimodal AI systems capable of deeper understanding and reasoning. Our project page is at https://github.com/OpenCausaLab/MORE.

Large language models (LLMs) are increasingly being utilised across a range of tasks and domains, with a burgeoning interest in their application within the field of journalism. This trend raises concerns due to our limited understanding of LLM behaviour in this domain, especially with respect to political bias. Existing studies predominantly focus on LLMs undertaking political questionnaires, which offers only limited insights into their biases and operational nuances. To address this gap, our study establishes a new curated dataset that contains 2,100 human-written articles and utilises their descriptions to generate 56,700 synthetic articles using nine LLMs. This enables us to analyse shifts in properties between human-authored and machine-generated articles, with this study focusing on political bias, detecting it using both supervised models and LLMs. Our findings reveal significant disparities between base and instruction-tuned LLMs, with instruction-tuned models exhibiting consistent political bias. Furthermore, we are able to study how LLMs behave as classifiers, observing their display of political bias even in this role. Overall, for the first time within the journalistic domain, this study outlines a framework and provides a structured dataset for quantifiable experiments, serving as a foundation for further research into LLM political bias and its implications.

Training large language models (LLMs) for external tool usage is a rapidly expanding field, with recent research focusing on generating synthetic data to address the shortage of available data. However, the absence of systematic data quality checks poses complications for properly training and testing models. To that end, we propose two approaches for assessing the reliability of data for training LLMs to use external tools. The first approach uses intuitive, human-defined correctness criteria. The second approach uses a model-driven assessment with in-context evaluation. We conduct a thorough evaluation of data quality on two popular benchmarks, followed by an extrinsic evaluation that showcases the impact of data quality on model performance. Our results demonstrate that models trained on high-quality data outperform those trained on unvalidated data, even when trained with a smaller quantity of data. These findings empirically support the significance of assessing and ensuring the reliability of training data for tool-using LLMs.

Despite advancements in Large Language Models (LLMs), many complex tasks are not easily solved in a single inference step, requiring the use of agentic LLMs in interactive environments. However, agentic LLMs suffer from a phenomenon known as reasoning derailment, due to the indiscriminate incorporation of observations from partially observable environments. We introduce QuBE, a method that enhances agents’ focus on task-relevant contexts, by constructing a belief state via question answering. We validate QuBE through experiments in two agentic LLM scenarios with partial observability: 1) a canonical interactive decision-making scenario using text-based game engines, and 2) an interactive retrieval-augmented generation (RAG) scenario using search engines. In the AlfWorld text-based game, QuBE outperforms established baselines by substantial margins, and in the search engine scenario, it achieves marked improvements on the BeIR zero-shot retrieval benchmark. The results demonstrate that QuBE significantly mitigates reasoning derailment, refining the decision-making process of LLM agents in partially observed environments.

Reasoning is key to many decision making processes. It requires consolidating a set of rule-like premises that are often associated with degrees of uncertainty and observations to draw conclusions. In this work, we address both the case where premises are specified as numeric probabilistic rules and situations in which humans state their estimates using words expressing degrees of certainty. Existing probabilistic reasoning datasets simplify the task, e.g., by requiring the model to only rank textual alternatives, by including only binary random variables, or by making use of a limited set of templates that result in less varied text.In this work, we present QUITE, a question answering dataset of real-world Bayesian reasoning scenarios with categorical random variables and complex relationships. QUITE provides high-quality natural language verbalizations of premises together with evidence statements and expects the answer to a question in the form of an estimated probability. We conduct an extensive set of experiments, finding that logic-based models outperform out-of-the-box large language models on all reasoning types (causal, evidential, and explaining-away). Our results provide evidence that neuro-symbolic models are a promising direction for improving complex reasoning. We release QUITE and code for training and experiments on Github.

Discovering meaningful insights from a large dataset, known as Exploratory Data Analysis (EDA), is a challenging task that requires thorough exploration and analysis of the data. Automated Data Exploration (ADE) systems use goal-oriented methods with Large Language Models and Reinforcement Learning towards full automation. However, these methods require human involvement to anticipate goals that may limit insight extraction, while fully automated systems demand significant computational resources and retraining for new datasets. We introduce QUIS, a fully automated EDA system that operates in two stages: insight generation (ISGen) driven by question generation (QUGen). The QUGen module generates questions in iterations, refining them from previous iterations to enhance coverage without human intervention or manually curated examples. The ISGen module analyzes data to produce multiple relevant insights in response to each question, requiring no prior training and enabling QUIS to adapt to new datasets.