EMNLP 2024 Papers

There are many emerging quantization methods to resolve the problem that the huge demand on computational and storage costs hinders the deployment of Large language models (LLMs). However, their accuracy performance still can not satisfy the entire academic and industry community. In this work, we propose ATQ, an INT8 weight-activation quantization of LLMs, that can achieve almost lossless accuracy. We employ a mathematically equivalent transformation and a triangle inequality to constrain weight-activation quantization error to the sum of a weight quantization error and an activation quantization error. For the weight part, transformed weights are quantized along the |in-feature| dimension and the quantization error is compensated by optimizing following in-features. For the activation part, transformed activations are in the normal range and can be quantized easily. We provide comparison experiments to demonstrate that our ATQ method can achieve almost lossless in accuracy on OPT and LLaMA families in W8A8 quantization settings. The increase of perplexity is within 1 and the accuracy degradation is within 0.5 percent even in the worst case.

Large language models (LLMs) are proven to benefit a lot from retrieval-augmented generation (RAG) in alleviating hallucinations confronted with knowledge-intensive questions. RAG adopts information retrieval techniques to inject external knowledge from semantic-relevant documents as input contexts. However, due to today’s Internet being flooded with numerous noisy and fabricating content, it is inevitable that RAG systems are vulnerable to these noises and prone to respond incorrectly. To this end, we propose to optimize the retrieval-augmented Generator with a Adversarial Tuning Multi-agent system **(ATM)**. The ATM steers the Generator to have a robust perspective of useful documents for question answering with the help of an auxiliary Attacker agent. The Generator and the Attacker are tuned adversarially for several iterations. After rounds of multi-agent iterative tuning, the Generator can eventually better discriminate useful documents amongst fabrications. The experimental results verify the effectiveness of ATM and we also observe that the Generator can achieve better performance compared to state-of-the-art baselines.

The mission of commonsense knowledge graph completion (CKGC) is to infer missing facts from known commonsense knowledge. CKGC methods can be roughly divided into two categories: triple-based methods and text-based methods. Due to the imbalanced distribution of entities and limited structural information, triple-based methods struggle with long-tail entities. Text-based methods alleviate this issue, but require extensive training and fine-tuning of language models, which reduces efficiency. To alleviate these problems, we propose ATAP, the first CKGC framework that utilizes automatically generated continuous prompt templates combined with pre-trained language models (PLMs). Moreover, ATAP uses a carefully designed new prompt template training strategy, guiding PLMs to generate optimal prompt templates for CKGC tasks. Combining the rich knowledge of PLMs with the template automatic augmentation strategy, ATAP effectively mitigates the long-tail problem and enhances CKGC performance. Results on benchmark datasets show that ATAP achieves state-of-the-art performance overall.

Many eCommerce platforms source product information from millions of sellers and manufactures, each having their own proprietary schemas, and employ schema matching solutions to structure it to enable informative shopping experiences. Meanwhile, state-of-the-art machine translation techniques have demonstrated great success in building context-aware representations that generalize well to new languages with minimal training data. In this work, we propose modeling the schema matching problem as a neural machine translation task: given product context and an attribute-value pair from a source schema, the model predicts the corresponding attribute, if available, in the target schema. We utilize open-source seq2seq models, such as mT5 and mBART, fine-tuned on product attribute mappings to build a scalable schema matching framework. We demonstrate that our proposed approach achieves a significant performance boost (15% precision and 7% recall uplift) compared to the baseline system and can support new attributes with precision ≥ 95% using only five labeled samples per attribute.

Aspect-Sentiment Triplet Extraction (ASTE) is a recently proposed task of aspect-based sentiment analysis that consists in extracting (aspect phrase, opinion phrase, sentiment polarity) triples from a given sentence. Recent state-of-the-art methods approach this task by first extracting all possible text spans from a given text, then filtering the potential aspect and opinion phrases with a classifier, and finally considering all their pairs with another classifier that additionally assigns sentiment polarity to them. Although several variations of the above scheme have been proposed, the common feature is that the final result is constructed by a sequence of independent classifier decisions. This hinders the exploitation of dependencies between extracted phrases and prevents the use of knowledge about the interrelationships between classifier predictions to improve performance. In this paper, we propose a new ASTE approach consisting of three transformer-inspired layers, which enables the modelling of dependencies both between phrases and between the final classifier decisions. Experimental results show that the method achieves higher performance in terms of F1 measure than other methods studied on popular benchmarks. In addition, we show that a simple pre-training technique further improves the performance of the model.

Deaf and hard-of-hearing (DHH) students face significant barriers in accessing science, technology, engineering, and mathematics (STEM) education, notably due to the scarcity of STEM resources in signed languages. To help address this, we introduce ASL STEM Wiki: a parallel corpus of 254 Wikipedia articles on STEM topics in English, interpreted into over 300 hours of American Sign Language (ASL). ASL STEM Wiki is the first continuous signing dataset focused on STEM, facilitating the development of AI resources for STEM education in ASL.We identify several use cases of ASL STEM Wiki with human-centered applications. For example, because this dataset highlights the frequent use of fingerspelling for technical concepts, which inhibits DHH students’ ability to learn,we develop models to identify fingerspelled words—which can later be used to query for appropriate ASL signs to suggest to interpreters.

The safety defense methods of Large language models (LLMs) stays limited because the dangerous prompts are manually curated to just few known attack types, which fails to keep pace with emerging varieties. Recent studies found that attaching suffixes to harmful instructions can hack the defense of LLMs and lead to dangerous outputs. However, similar to traditional text adversarial attacks, this approach, while effective, is limited by the challenge of the discrete tokens. This gradient based discrete optimization attack requires over 100,000 LLM calls, and due to the unreadable of adversarial suffixes, it can be relatively easily penetrated by common defense methods such as perplexity filters.To cope with this challenge, in this paper, we propose an Adversarial Suffix Embedding Translation Framework (ASETF), aimed at transforming continuous adversarial suffix embeddings into coherent and understandable text. This method greatly reduces the computational overhead during the attack process and helps to automatically generate multiple adversarial samples, which can be used as data to strengthen LLM’s security defense. Experimental evaluations were conducted on Llama2, Vicuna, and other prominent LLMs, employing harmful directives sourced from the Advbench dataset.The results indicate that our method significantly reduces the computation time of adversarial suffixes and achieves a much better attack success rate than existing techniques, while significantly enhancing the textual fluency of the prompts. In addition, our approach can be generalized into a broader method for generating transferable adversarial suffixes that can successfully attack multiple LLMs, even black-box LLMs, such as ChatGPT and Gemini.

Automatic text simplification aims to reduce a text’s complexity. Its evaluation should quantify how easy it is to understand a text. Datasets with simplicity labels on text level are a prerequisite for developing such evaluation approaches. However, current publicly available datasets do not align with this, as they mainly treat text simplification as a relational concept (“How much simpler has this text gotten compared to the original version?”) or assign discrete readability levels.This work alleviates the problem of Assessing Readability & Text Simplicity. We present ARTS, a method for language-independent construction of datasets for simplicity assessment. We propose using pairwise comparisons of texts in conjunction with an Elo algorithm to produce a simplicity ranking and simplicity scores. Additionally, we provide a high-quality human-labeled and three GPT-labeled simplicity datasets. Our results show a high correlation between human and LLM-based labels, allowing for an effective and cost-efficient way to construct large synthetic datasets.

Chinese Spelling Check (CSC) aims to identify and correct spelling errors in Chinese texts, where enhanced semantic understanding of a sentence can significantly improve correction accuracy. Recently, Large Language Models (LLMs) have demonstrated exceptional mastery of world knowledge and semantic understanding, rendering them more robust against spelling errors. However, the application of LLMs in CSC is a double-edged sword, as they tend to unnecessarily alter sentence length and modify rare but correctly used phrases. In this paper, by leveraging the capabilities of LLMs while mitigating their limitations, we propose a novel plug-and-play Alignment-and-Replacement Module ARM that enhances the performance of existing CSC models and without the need for retraining or fine-tuning. Experiment results and analysis on three benchmark datasets demonstrate the effectiveness and competitiveness of the proposed module.

Large Multimodal Models (LMMs) excel at comprehending human instructions and demonstrate remarkable results across a broad spectrum of tasks. Reinforcement Learning from Human Feedback (RLHF) and AI Feedback (RLAIF) further refine LLMs by aligning them with specific preferences. These methods primarily use ranking-based feedback for entire generations. With advanced AI models (Teacher), such as GPT-4 and Claude 3 Opus, we can request various types of detailed feedback that are expensive for humans to provide. We propose a two-stage algorithm ARES that Alternates REinforcement Learning (RL) and Supervised Fine-Tuning (SFT). First, we ask the Teacher to score how much each sentence contributes to solving the problem in a Chain-of-Thought (CoT). This sentence-level feedback allows us to consider individual valuable segments, providing more granular rewards for the RL procedure. Second, we ask the Teacher to correct wrong reasoning after the RL stage. The RL procedure requires substantial hyperparameter tuning and often generates errors such as repetitive words and incomplete sentences. With correction feedback, we stabilize the RL fine-tuned model through SFT. We conduct experiments on the multi-modal datasets ScienceQA and A-OKVQA to demonstrate the effectiveness of our proposal. The ARES rationale achieves around 70% win rate compared to baseline models judged by GPT-4o. Additionally, we observe that the improved rationale reasoning leads to a 2.5% increase in inference answer accuracy on average for the multi-modal datasets.

While there has been significant development of models for Plain Language Summarization (PLS), evaluation remains a challenge. PLS lacks a dedicated assessment metric, and the suitability of text generation evaluation metrics is unclear due to the unique transformations involved (e.g., adding background explanations, removing jargon). To address these questions, our study introduces a granular meta-evaluation testbed, APPLS, designed to evaluate metrics for PLS. We identify four PLS criteria from previous work—informativeness, simplification, coherence, and faithfulness—and define a set of perturbations corresponding to these criteria that sensitive metrics should be able to detect. We apply these perturbations to extractive hypotheses for two PLS datasets to form our testbed. Using APPLS, we assess performance of 14 metrics, including automated scores, lexical features, and LLM prompt-based evaluations. Our analysis reveals that while some current metrics show sensitivity to specific criteria, no single method captures all four criteria simultaneously. We therefore recommend a suite of automated metrics be used to capture PLS quality along all relevant criteria. This work contributes the first meta-evaluation testbed for PLS and a comprehensive evaluation of existing metrics.

This study aims to address the pervasive challenge of quantifying uncertainty in large language models (LLMs) with black-box API access. Conformal Prediction (CP), known for its model-agnostic and distribution-free features, is a desired approach for various LLMs and data distributions. However, existing CP methods for LLMs typically assume access to the logits, which are unavailable for some API-only LLMs. In addition, logits are known to be miscalibrated, potentially leading to degraded CP performance. To tackle these challenges, we introduce a novel CP method that (1) is tailored for API-only LLMs without logit-access; (2) minimizes the size of prediction sets; and (3) ensures a statistical guarantee of the user-defined coverage. The core idea of this approach is to formulate nonconformity measures using both coarse-grained (i.e., sample frequency) and fine-grained uncertainty notions (e.g., semantic similarity). Experimental results on both close-ended and open-ended Question Answering tasks show our approach can mostly outperform the logit-based CP baselines.

The impressive performance of proprietary LLMs like GPT4 in code generation has led to a trend to replicate these capabilities in open-source models through knowledge distillation (e.g. Code Evol-Instruct). However, these efforts often neglect the crucial aspect of response quality, relying heavily on teacher models for direct response distillation. This paradigm, especially for complex instructions, can degrade the quality of synthesized data, compromising the knowledge distillation process. To this end, our study introduces the Adaptive Modular Response Evolution (AMR-Evol) framework, which employs a two-stage process to refine response distillation. The first stage, modular decomposition, breaks down the direct response into more manageable sub-modules. The second stage, adaptive response evolution, automatically evolves the response with the related function modules. Our experiments with three popular code benchmarks—HumanEval, MBPP, and EvalPlus—attests to the superiority of the AMR-Evol framework over baseline response distillation methods. By comparing with the open-source Code LLMs trained on a similar scale of data, we observed performance enhancements: more than +3.0 points on HumanEval-Plus and +1.0 points on MBPP-Plus, which underscores the effectiveness of our framework. Our codes are available at https://github.com/ChiYeungLaw/AMR-Evol.

Prompt engineering is very important to enhance the performance of large language models (LLMs). When dealing with complex issues, prompt engineers tend to distill multiple patterns from examples and inject relevant solutions to optimize the prompts, achieving satisfying results. However, existing automatic prompt optimization techniques are only limited to producing single flow instructions, struggling with handling diverse patterns. In this paper, we present AMPO, an automatic prompt optimization method that can iteratively develop a multi-branched prompt using failure cases as feedback. Our goal is to explore a novel way of structuring prompts with multi-branches to better handle multiple patterns in complex tasks, for which we introduce three modules: Pattern Recognition, Branch Adjustment, and Branch Pruning. In experiments across five tasks, AMPO consistently achieves the best results. Additionally, our approach demonstrates significant optimization efficiency due to our adoption of a minimal search strategy.

Active Learning aims to minimize annotation effort by selecting the most useful instances from a pool of unlabeled data. However, typical active learning methods overlook the presence of distinct example groups within a class, whose prevalence may vary, e.g., in occupation classification datasets certain demographics are disproportionately represented in specific classes. This oversight causes models to rely on shortcuts for predictions, i.e., spurious correlations between input attributes and labels occurring in well-represented groups. To address this issue, we propose Active Learning Via INterpolation (ALVIN), which conducts intra-class interpolations between examples from under-represented and well-represented groups to create anchors, i.e., artificial points situated between the example groups in the representation space. By selecting instances close to the anchors for annotation, ALVIN identifies informative examples exposing the model to regions of the representation space that counteract the influence of shortcuts. Crucially, since the model considers these examples to be of high certainty, they are likely to be ignored by typical active learning methods. Experimental results on six datasets encompassing sentiment analysis, natural language inference, and paraphrase detection demonstrate that ALVIN outperforms state-of-the-art active learning methods in both in-distribution and out-of-distribution generalization.

Sentence embeddings play a pivotal role in a wide range of NLP tasks, yet evaluating and interpreting these real-valued vectors remains an open challenge to date, especially in a task-free setting. To address this challenge, we introduce a novel task-free test bed for evaluating and interpreting sentence embeddings. Our test bed consists of five semantic similarity alignment criteria, namely, *semantic distinction, synonym replacement, antonym replacement, paraphrasing without negation, and sentence jumbling*. Using these criteria, we examined five classical (e.g., Sentence-BERT, Universal Sentence Encoder (USE), etc.) and eight LLM-induced sentence embedding techniques (e.g., LLaMA2, GPT-3, OLMo, etc.) to test whether their semantic similarity spaces align with what a human mind would naturally expect. Our extensive experiments with 13 different sentence encoders revealed that none of the studied embeddings aligned with all the five semantic similarity alignment criteria. Yet, most encoders performed highly on the SentEval dataset, a popular task-specific benchmark. This finding demonstrates a significant limitation of the current practice in sentence embedding evaluation and associated popular benchmarks, a critical issue that needs careful attention and reassessment by the NLP community. Finally, we conclude the paper by highlighting the utility of the proposed alignment-based test bed for analyzing sentence embeddings in a novel way, especially in a task-free setting.

Knowledge editing injects knowledge updates into language models to keep them correct and up-to-date. However, its current evaluations deviate significantly from practice: their knowledge updates solely consist of structured facts derived from meticulously crafted datasets, instead of practical sources—unstructured texts like news articles, and they often overlook practical real-world knowledge updates. To address these issues, in this paper we propose AKEW (Assessing Knowledge Editing in the Wild), a new practical benchmark for knowledge editing. AKEW fully covers three editing settings of knowledge updates: structured facts, unstructured texts as facts, and extracted triplets. It further introduces new datasets featuring both counterfactual and real-world knowledge updates. Through extensive experiments, we demonstrate the considerable gap between state-of-the-art knowledge-editing methods and practical scenarios. Our analyses further highlight key insights to motivate future research for practical knowledge editing.

Question answering (QA) tasks have been extensively studied in the field of natural language processing (NLP). Answers to open-ended questions are highly diverse and difficult to quantify, and cannot be simply evaluated as correct or incorrect, unlike close-ended questions with definitive answers. While large language models (LLMs) have demonstrated strong capabilities across various tasks, they exhibit relatively weaker performance in evaluating answers to open-ended questions. In this study, we propose a method that leverages LLMs and the analytic hierarchy process (AHP) to assess answers to open-ended questions. We utilized LLMs to generate multiple evaluation criteria for a question. Subsequently, answers were subjected to pairwise comparisons under each criterion with LLMs, and scores for each answer were calculated in the AHP. We conducted experiments on four datasets using both ChatGPT-3.5-turbo and GPT-4. Our results indicate that our approach more closely aligns with human judgment compared to the four baselines. Additionally, we explored the impact of the number of criteria, variations in models, and differences in datasets on the results.

Ranking is a fundamental problem in search, however, existing ranking algorithms usually restrict the granularity of ranking to full passages or require a specific dense index for each desired level of granularity. Such lack of flexibility in granularity negatively affects many applications that can benefit from more granular ranking, such as sentence-level ranking for open-domain QA, or proposition-level ranking for attribution. In this work, we introduce the idea of any-granularity ranking which leverages multi-vector embeddings to rank at varying levels of granularity while maintaining encoding at a single (coarser) level of granularity. We propose a multi-granular contrastive loss for training multi-vector approaches and validate its utility with both sentences and propositions as ranking units. Finally, we demonstrate the application of proposition-level ranking to post-hoc citation addition in retrieval-augmented generation, surpassing the performance of prompt-driven citation generation.