ACL 2024 论文列表

Large language models (LLMs) have demonstrated impressive reasoning capabilities, yet there is ongoing debate about these abilities and the potential data contamination problem recently. This paper aims to evaluate the reasoning capacities of LLMs, specifically in solving recent competition-level programming problems in Codeforces, which are expert-crafted and unique, requiring deep understanding and robust reasoning skills. We first provide a comprehensive evaluation of GPT-4’s perceived zero-shot performance on this task, considering various aspects such as problems’ release time, difficulties, and types of errors encountered. Surprisingly, the perceived performance of GPT-4 has experienced a cliff like decline in problems after September 2021 consistently across all the difficulties and types of problems, which shows the potential data contamination, as well as the challenges for any existing LLM to solve unseen complex reasoning problems. We further explore various approaches such as fine-tuning, Chain-of-Thought prompting and problem description simplification. Unfortunately, none of them is able to consistently mitigate the challenges. Through our work, we emphasize the importance of this excellent data source for assessing the genuine reasoning capabilities of LLMs, and foster the development of LLMs with stronger reasoning abilities and better generalization in the future.

Interpretability research aims to bridge the gap between the empirical success and our scientific understanding of the inner workings of large language models (LLMs). However, most existing research in this area focused on analyzing a single mechanism, such as how models copy or recall factual knowledge. In this work, we propose the formulation of competition of mechanisms, which instead of individual mechanisms focuses on the interplay of multiple mechanisms, and traces how one of them becomes dominant in the final prediction. We uncover how and where the competition of mechanisms happens within LLMs using two interpretability methods, logit inspection and attention modification. Our findings show traces of the mechanisms and their competition across various model components, and reveal attention positions that effectively control the strength of certain mechanisms.

Deductive reasoning plays a pivotal role in the formulation of sound and cohesive arguments. It allows individuals to draw conclusions that logically follow, given the truth value of the information provided. Recent progress in the domain of large language models (LLMs) has showcased their capability in executing deductive reasoning tasks. Nonetheless, a significant portion of research primarily assesses the accuracy of LLMs in solving such tasks, often overlooking a deeper analysis of their reasoning behavior. In this study, we draw upon principles from cognitive psychology to examine inferential strategies employed by LLMs, through a detailed evaluation of their responses to propositional logic problems. Our findings indicate that LLMs display reasoning patterns akin to those observed in humans, including strategies like supposition following or chain construction. Moreover, our research demonstrates that the architecture and scale of the model significantly affect its preferred method of reasoning, with more advanced models tending to adopt strategies more frequently than less sophisticated ones. Importantly, we assert that a model’s accuracy, that is the correctness of its final conclusion, does not necessarily reflect the validity of its reasoning process. This distinction underscores the necessity for more nuanced evaluation procedures in the field.

Creating effective and reliable task-oriented dialog systems (ToDSs) is challenging, not only because of the complex structure of these systems, but also due to the scarcity of training data, especially when several modules need to be trained separately, each one with its own input/output training examples. Data augmentation (DA), whereby synthetic training examples are added to the training data, has been successful in other NLP systems, but has not been explored as extensively in ToDSs. We empirically evaluate the effectiveness of DA methods in an end-to-end ToDS setting, where a single system is trained to handle all processing stages, from user inputs to system outputs. We experiment with two ToDSs (UBAR, GALAXY) on two datasets (MultiWOZ, KVRET). We consider three types of DA methods (word-level, sentence-level, dialog-level), comparing eight DA methods that have shown promising results in ToDSs and other NLP systems. We show that all DA methods considered are beneficial, and we highlight the best ones, also providing advice to practitioners. We also introduce a more challenging few-shot cross-domain ToDS setting, reaching similar conclusions.

Code generation aims to understand the problem description and generate corresponding code snippets, where existing works generally decompose such complex tasks into intermediate steps by prompting strategies, such as Chain-of-Thought and its variants. While these studies have achieved some success, their effectiveness is highly dependent on the capabilities of advanced Large Language Models (LLMs) such as GPT-4, particularly in terms of API calls, which significantly limits their practical applicability. Consequently, how to enhance the code generation capabilities of small and medium-scale code LLMs without significantly increasing training costs is an appealing challenge. In this paper, we suggest that code comments are the natural logic pivot between natural language and code language and propose using comments to boost the code generation ability of code LLMs. Concretely, we propose MANGO (comMents As Natural loGic pivOts), including a comment contrastive training strategy and a corresponding logical comment decoding strategy. Experiments are performed on HumanEval and MBPP, utilizing StarCoder and WizardCoder as backbone models, and encompassing model parameter sizes between 3B and 7B. The results indicate that MANGO significantly improves the code pass rate based on the strong baselines. Meanwhile, the robustness of the logical comment decoding strategy is notably higher than the Chain-of-thoughts prompting.

Traditional supervised learning heavily relies on human-annotated datasets, especially in data-hungry neural approaches. However, various tasks, especially multi-label tasks like document-level relation extraction, pose challenges in fully manual annotation due to the specific domain knowledge and large class sets. Therefore, we address the multi-label positive-unlabelled learning (MLPUL) problem, where only a subset of positive classes is annotated. We propose Mixture Learner for Partially Annotated Classification (MLPAC), an RL-based framework combining the exploration ability of reinforcement learning and the exploitation ability of supervised learning. Experimental results across various tasks, including document-level relation extraction, multi-label image classification, and binary PU learning, demonstrate the generalization and effectiveness of our framework.

We introduce a hybrid abstractive summarisation approach combining hierarchical VAEs with LLMs to produce clinically meaningful summaries from social media user timelines, appropriate for mental health monitoring. The summaries combine two different narrative points of view: (a) clinical insights in third person, generated by feeding into an LLM clinical expert-guided prompts, and importantly, (b) a temporally sensitive abstractive summary of the user’s timeline in first person, generated by a novel hierarchical variational autoencoder, TH-VAE. We assess the generated summaries via automatic evaluation against expert summaries and via human evaluation with clinical experts, showing that timeline summarisation by TH-VAE results in more factual and logically coherent summaries rich in clinical utility and superior to LLM-only approaches in capturing changes over time.

Extracting semantic topics from short texts presents a significant challenge in the field of data mining. While efforts have been made to mitigate data sparsity issue, the limited length of short documents also results in the absence of semantically relevant words, causing biased evidence lower bound and incomplete labels for likelihood maximization. We refer to this issue as the label sparsity problem. To combat this problem, we propose kNNTM, a neural short text topic model that incorporates a k-Nearest-Neighbor-based label completion algorithm by augmenting the reconstruction label with k-nearest documents to complement these relevant but unobserved words. Furthermore, seeking a precise reflection of distances between documents, we propose a fused multi-view distances metric that takes both local word similarities and global topic semantics into consideration. Extensive experiments on multiple public short-text datasets show that kNNTM model outperforms the state-of-the-art baseline models and can derive both high-quality topics and document representations.

This paper introduces the ColorSwap dataset, designed to assess and improve the proficiency of multimodal models in matching objects with their colors. The dataset is comprised of 2,000 unique image-caption pairs, grouped into 1,000 examples. Each example includes a caption-image pair, along with a “color-swapped” pair. We follow the Winoground schema: the two captions in an example have the same words, but the color words have been rearranged to modify different objects. The dataset was created through a novel blend of automated caption and image generation with humans in the loop. We evaluate image-text matching (ITM) and visual language models (VLMs) and find that even the latest ones are still not robust at this task. GPT-4V and LLaVA score 72% and 42% on our main VLM metric, although they may improve with more advanced prompting techniques. On the main ITM metric, contrastive models such as CLIP and SigLIP perform close to chance (at 12% and 30%, respectively), although the non-contrastive BLIP ITM model is stronger (87%). We also find that finetuning on fewer than 2,000 examples yields significant performance gains on this out-of-distribution word-order understanding task.

Impressive performance of pre-trained models has garnered public attention and made news headlines in recent years. Almost always, these models are produced by or in collaboration with industry. Using them is critical for competing on natural language processing (NLP) benchmarks and correspondingly to stay relevant in NLP research. We surveyed 100 papers published at EMNLP 2022 to determine the degree to which researchers rely on industry models, other artifacts, and contributions to publish in prestigious NLP venues and found that the ratio of their citation is at least three times greater than what would be expected. Our work serves as a scaffold to enable future researchers to more accurately address whether: 1) Collaboration with industry is still collaboration in the absence of an alternative or 2) if NLP inquiry has been captured by the motivations and research direction of private corporations.

Evaluating and Rethinking the current landscape of Large Multimodal Models (LMMs), we observe that widely-used visual-language projection approaches (e.g., Q-former or MLP) focus on the alignment of image-text descriptions yet ignore the visual knowledge-dimension alignment, i.e., connecting visuals to their relevant knowledge. Visual knowledge plays a significant role in analyzing, inferring, and interpreting information from visuals, helping improve the accuracy of answers to knowledge-based visual questions. In this paper, we mainly explore improving LMMs with visual-language knowledge alignment, especially aimed at challenging knowledge-based visual question answering (VQA). To this end, we present a Cognitive Visual-Language Mapper (CVLM), which contains a pretrained Visual Knowledge Aligner (VKA) and a Fine-grained Knowledge Adapter (FKA) used in the multimodal instruction tuning stage. Specifically, we design the VKA based on the interaction between a small language model and a visual encoder, training it on collected image-knowledge pairs to achieve visual knowledge acquisition and projection. FKA is employed to distill the fine-grained visual knowledge of an image and inject it into Large Language Models (LLMs). We conduct extensive experiments on knowledge-based VQA benchmarks and experimental results show that CVLM significantly improves the performance of LMMs on knowledge-based VQA (average gain by 5.0%). Ablation studies also verify the effectiveness of VKA and FKA, respectively.

Large language models have become integral to question-answering applications despite their propensity for generating hallucinations and factually inaccurate content. Querying knowledge graphs to reduce hallucinations in LLM meets the challenge of incomplete knowledge coverage in knowledge graphs. On the other hand, updating knowledge graphs by information extraction and knowledge graph completion faces the knowledge update misalignment issue. In this work, we introduce a collaborative augmentation framework, CogMG, leveraging knowledge graphs to address the limitations of LLMs in QA scenarios, explicitly targeting the problems of incomplete knowledge coverage and knowledge update misalignment. The LLMs identify and decompose required knowledge triples that are not present in the KG, enriching them and aligning updates with real-world demands. We demonstrate the efficacy of this approach through a supervised fine-tuned LLM within an agent framework, showing significant improvements in reducing hallucinations and enhancing factual accuracy in QA responses. Our code and video are publicly available.

Social media abounds with multimodal sarcasm, and identifying sarcasm targets is particularly challenging due to the implicit incongruity not directly evident in the text and image modalities. Current methods for Multimodal Sarcasm Target Identification (MSTI) predominantly focus on superficial indicators in an end-to-end manner, overlooking the nuanced understanding of multimodal sarcasm conveyed through both the text and image. This paper proposes a versatile MSTI framework with a coarse-to-fine paradigm, by augmenting sarcasm explainability with reasoning and pre-training knowledge. Inspired by the powerful capacity of Large Multimodal Models (LMMs) on multimodal reasoning, we first engage LMMs to generate competing rationales for coarser-grained pre-training of a small language model on multimodal sarcasm detection. We then propose fine-tuning the model for finer-grained sarcasm target identification. Our framework is thus empowered to adeptly unveil the intricate targets within multimodal sarcasm and mitigate the negative impact posed by potential noise inherently in LMMs. Experimental results demonstrate that our model far outperforms state-of-the-art MSTI methods, and markedly exhibits explainability in deciphering sarcasm as well.

The sound codec’s dual roles in minimizing data transmission latency and serving as tokenizers underscore its critical importance.Recent years have witnessed significant developments in codec models.The ideal sound codec should preserve content, paralinguistics, speakers, and audio information.However, the question of which codec achieves optimal sound information preservation remains unanswered, as in different papers, models are evaluated on their selected experimental settings.This study introduces Codec-SUPERB, an acronym for Codec sound processing Universal PERformance Benchmark.It is an ecosystem designed to assess codec models across representative sound applications and signal-level metrics rooted in sound domain knowledge.Codec-SUPERB simplifies result sharing through an online leaderboard, promoting collaboration within a community-driven benchmark database, thereby stimulating new development cycles for codecs.Furthermore, we undertake an in-depth analysis to offer insights into codec models from both application and signal perspectives, diverging from previous codec papers mainly concentrating on signal-level comparisons.Finally, we will release codes, the leaderboard, and data to accelerate progress within the community.

Large Language Models (LLMs) have demonstrated remarkable performance on assisting humans in programming and facilitating programming automation. However, existing benchmarks for evaluating the code understanding and generation capacities of LLMs suffer from severe limitations. First, most benchmarks are insufficient as they focus on a narrow range of popular programming languages and specific tasks, whereas real-world software development scenarios show a critical need to implement systems with multilingual and multitask programming environments to satisfy diverse requirements. Second, most benchmarks fail to consider the actual executability and the consistency of execution results of the generated code. To bridge these gaps between existing benchmarks and expectations from practical applications, we introduce **CodeScope**, an execution-based, multilingual, multitask, multidimensional evaluation benchmark for comprehensively measuring LLM capabilities on coding tasks. CodeScope covers **43 programming languages** and **eight coding tasks**. It evaluates the coding performance of LLMs from three dimensions (perspectives): **length**, **difficulty**, and **efficiency**. To facilitate execution-based evaluations of code generation, we develop **MultiCodeEngine**, an automated code execution engine that supports 14 programming languages. Finally, we systematically evaluate and analyze eight mainstream LLMs and demonstrate the superior breadth and challenges of CodeScope for evaluating LLMs on code understanding and generation tasks compared to other benchmarks. The CodeScope benchmark and code are publicly available at https://github.com/WeixiangYAN/CodeScope.

In the era of code large language models (code LLMs), data engineering plays a pivotal role during the instruction fine-tuning phase. To train a versatile model, previous efforts devote tremendous efforts into crafting instruction data covering all the downstream scenarios. Nonetheless, this will incur significant expenses in constructing data and training model. Therefore, this paper introduces CodeM, a novel data construction strategy, which can efficiently train a versatile model using less data via our newly proposed ability matrix. CodeM uses ability matrix to decouple code LLMs’ abilities into two dimensions, constructing a lightweight training corpus that only covers a subset of target scenarios. Extensive experiments on HumanEvalPack and MultiPL-E imply that code LLMs can combine the single-dimensional abilities to master composed abilities, validating the effectiveness of CodeM.

We introduce a novel dataset tailored for code generation, aimed at aiding developers in common tasks. Our dataset provides examples that include a clarified intent, code snippets associated, and an average of three related unit tests. It encompasses a range of libraries such as Pandas, Numpy, and Regex, along with more than 70 standard libraries in Python code derived from Stack Overflow. Comprising 3,402 crafted examples by Python experts, our dataset is designed for both model finetuning and standalone evaluation. To complete unit tests evaluation, we categorize examples in order to get more fine grained analysis, enhancing the understanding of models’ strengths and weaknesses in specific coding tasks. The examples have been refined to reduce data contamination, a process confirmed by the performance of three leading models: Mistral 7B, CodeLLAMA 13B, and Starcoder 15B. We further investigate data-contamination testing GPT-4 performance on a part of our dataset. The benchmark can be accessed at anonymized address.

The rapid advancement of Large Language Models (LLMs) has brought about remarkable generative capabilities but also raised concerns about their potential misuse. While strategies like supervised fine-tuning and reinforcement learning from human feedback have enhanced their safety, these methods primarily focus on natural languages, which may not generalize to other domains. This paper introduces CodeAttack, a framework that transforms natural language inputs into code inputs, presenting a novel environment for testing the safety generalization of LLMs. Our comprehensive studies on state-of-the-art LLMs including GPT-4, Claude-2, and Llama-2 series reveal a new and universal safety vulnerability of these models against code input: CodeAttack bypasses the safety guardrails of all models more than 80% of the time. We find that a larger distribution gap between CodeAttack and natural language leads to weaker safety generalization, such as encoding natural language input with data structures. Furthermore, we give our hypotheses about the success of CodeAttack: the misaligned bias acquired by LLMs during code training, prioritizing code completion over avoiding the potential safety risk. Finally, we analyze potential mitigation measures. These findings highlight new safety risks in the code domain and the need for more robust safety alignment algorithms to match the code capabilities of LLMs.

Large Language Models (LLMs) have shown promise in automated code generation but typically excel only in simpler tasks such as generating standalone code units. However, real-world software development often involves complex code repositories with complex dependencies and extensive documentation. To enable LLMs to handle these realworld repo-level code generation, we present CodeAgent, a novel LLM-based agent framework that employs external tools for effective repo-level code generation. CodeAgent integrates five programming tools, enabling interaction with software artifacts for information retrieval, code implementation, and code testing. We implement four agent strategies to optimize these tools’ usage. To the best of our knowledge, CodeAgent is the first agent tool framework specifically for repo-level code generation. In order to measure the effectiveness of our method at the repository level, we have introduced a benchmark dataset CodAgentBench. The performance on this dataset shows a significant improvement brought by our method, with improvements of pass rate ranging from 2.0 to 15.8. Further tests on the HumanEval benchmark confirm CodeAgent’s adaptability and efficacy across various code generation tasks. Notably, CodeAgent outperforms commercial products like Github Copilot, showcasing superior accuracy and efficiency. These results demonstrate CodeAgent’s robust capabilities in code generation, highlighting its potential for real-world repo-level coding challenges.