Datasets:

ACLMeeting

/

ACL2024-papers

like 2

Follow

Annual Meeting of the Association for Computational Linguistics 1

https://aclanthology.org/2024.acl-long.201.bib

@inproceedings{gaido-etal-2024-sbaam, title = "{SBAAM}! Eliminating Transcript Dependency in Automatic Subtitling", author = "Gaido, Marco and Papi, Sara and Negri, Matteo and Cettolo, Mauro and Bentivogli, Luisa", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.201", pages = "3673--3691", abstract = "Subtitling plays a crucial role in enhancing the accessibility of audiovisual content and encompasses three primary subtasks: translating spoken dialogue, segmenting translations into concise textual units, and estimating timestamps that govern their on-screen duration. Past attempts to automate this process rely, to varying degrees, on automatic transcripts, employed diversely for the three subtasks. In response to the acknowledged limitations associated with this reliance on transcripts, recent research has shifted towards transcription-free solutions for translation and segmentation, leaving the direct generation of timestamps as uncharted territory. To fill this gap, we introduce the first direct model capable of producing automatic subtitles, entirely eliminating any dependence on intermediate transcripts also for timestamp prediction. Experimental results, backed by manual evaluation, showcase our solution{'}s new state-of-the-art performance across multiple language pairs and diverse conditions.", }

Subtitling plays a crucial role in enhancing the accessibility of audiovisual content and encompasses three primary subtasks: translating spoken dialogue, segmenting translations into concise textual units, and estimating timestamps that govern their on-screen duration. Past attempts to automate this process rely, to varying degrees, on automatic transcripts, employed diversely for the three subtasks. In response to the acknowledged limitations associated with this reliance on transcripts, recent research has shifted towards transcription-free solutions for translation and segmentation, leaving the direct generation of timestamps as uncharted territory. To fill this gap, we introduce the first direct model capable of producing automatic subtitles, entirely eliminating any dependence on intermediate transcripts also for timestamp prediction. Experimental results, backed by manual evaluation, showcase our solution{'}s new state-of-the-art performance across multiple language pairs and diverse conditions.

SBAAM! Eliminating Transcript Dependency in Automatic Subtitling

acl-long.201

Poster

2405.10741

https://huggingface.co/papers/2405.10741

https://aclanthology.org/2024.acl-long.201/

https://aclanthology.org/2024.acl-long.202.bib

@inproceedings{papi-etal-2024-streamatt, title = "{S}tream{A}tt: Direct Streaming Speech-to-Text Translation with Attention-based Audio History Selection", author = "Papi, Sara and Gaido, Marco and Negri, Matteo and Bentivogli, Luisa", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.202", pages = "3692--3707", abstract = "Streaming speech-to-text translation (StreamST) is the task of automatically translating speech while incrementally receiving an audio stream. Unlike simultaneous ST (SimulST), which deals with pre-segmented speech, StreamST faces the challenges of handling continuous and unbounded audio streams. This requires additional decisions about what to retain of the previous history, which is impractical to keep entirely due to latency and computational constraints. Despite the real-world demand for real-time ST, research on streaming translation remains limited, with existing works solely focusing on SimulST. To fill this gap, we introduce StreamAtt, the first StreamST policy, and propose StreamLAAL, the first StreamST latency metric designed to be comparable with existing metrics for SimulST. Extensive experiments across all 8 languages of MuST-C v1.0 show the effectiveness of StreamAtt compared to a naive streaming baseline and the related state-of-the-art SimulST policy, providing a first step in StreamST research.", }

Streaming speech-to-text translation (StreamST) is the task of automatically translating speech while incrementally receiving an audio stream. Unlike simultaneous ST (SimulST), which deals with pre-segmented speech, StreamST faces the challenges of handling continuous and unbounded audio streams. This requires additional decisions about what to retain of the previous history, which is impractical to keep entirely due to latency and computational constraints. Despite the real-world demand for real-time ST, research on streaming translation remains limited, with existing works solely focusing on SimulST. To fill this gap, we introduce StreamAtt, the first StreamST policy, and propose StreamLAAL, the first StreamST latency metric designed to be comparable with existing metrics for SimulST. Extensive experiments across all 8 languages of MuST-C v1.0 show the effectiveness of StreamAtt compared to a naive streaming baseline and the related state-of-the-art SimulST policy, providing a first step in StreamST research.

StreamAtt: Direct Streaming Speech-to-Text Translation with Attention-based Audio History Selection

acl-long.202

Poster

2406.06097

https://huggingface.co/papers/2406.06097

https://aclanthology.org/2024.acl-long.202/

https://aclanthology.org/2024.acl-long.203.bib

@inproceedings{zhang-etal-2024-arl2, title = "{ARL}2: Aligning Retrievers with Black-box Large Language Models via Self-guided Adaptive Relevance Labeling", author = "Zhang, LingXi and Yu, Yue and Wang, Kuan and Zhang, Chao", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.203", pages = "3708--3719", abstract = "Retrieval-augmented generation enhances large language models (LLMs) by incorporating relevant information from external knowledge sources. This enables LLMs to adapt to specific domains and mitigate hallucinations in knowledge-intensive tasks. However, existing retrievers are often misaligned with LLMs due to separate training processes and the inherent black-box nature of LLMs. To address this challenge, we propose ARL2, a retriever learning technique that harnesses LLMs as labelers. ARL2 leverages LLMs to annotate and score adaptive relevance evidence, enabling the retriever to learn from robust LLM supervision. Furthermore, ARL2 incorporates a self-training strategy to minimize the cost of API calls. Extensive experiments demonstrate the effectiveness of ARL2, achieving accuracy improvements of 5.4{\%} on NQ and 4.6{\%} on MMLU compared to the state-of-the-art methods. Additionally, ARL2 exhibits robust transfer learning capabilities and strong zero-shot generalization abilities.", }

Retrieval-augmented generation enhances large language models (LLMs) by incorporating relevant information from external knowledge sources. This enables LLMs to adapt to specific domains and mitigate hallucinations in knowledge-intensive tasks. However, existing retrievers are often misaligned with LLMs due to separate training processes and the inherent black-box nature of LLMs. To address this challenge, we propose ARL2, a retriever learning technique that harnesses LLMs as labelers. ARL2 leverages LLMs to annotate and score adaptive relevance evidence, enabling the retriever to learn from robust LLM supervision. Furthermore, ARL2 incorporates a self-training strategy to minimize the cost of API calls. Extensive experiments demonstrate the effectiveness of ARL2, achieving accuracy improvements of 5.4{\%} on NQ and 4.6{\%} on MMLU compared to the state-of-the-art methods. Additionally, ARL2 exhibits robust transfer learning capabilities and strong zero-shot generalization abilities.

ARL2: Aligning Retrievers with Black-box Large Language Models via Self-guided Adaptive Relevance Labeling

acl-long.203

Poster

https://aclanthology.org/2024.acl-long.203/

https://aclanthology.org/2024.acl-long.204.bib

@inproceedings{bang-etal-2024-crayon, title = "Crayon: Customized On-Device {LLM} via Instant Adapter Blending and Edge-Server Hybrid Inference", author = "Bang, Jihwan and Lee, Juntae and Shim, Kyuhong and Yang, Seunghan and Chang, Simyung", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.204", pages = "3720--3731", abstract = "The customization of large language models (LLMs) for user-specified tasks gets important. However, maintaining all the customized LLMs on cloud servers incurs substantial memory and computational overheads, and uploading user data can also lead to privacy concerns. On-device LLMs can offer a promising solution by mitigating these issues. Yet, the performance of on-device LLMs is inherently constrained by the limitations of small-scaled models. To overcome these restrictions, we first propose Crayon, a novel approach for on-device LLM customization. Crayon begins by constructing a pool of diverse base adapters, and then we instantly blend them into a customized adapter without extra training. In addition, we develop a device-server hybrid inference strategy, which deftly allocates more demanding queries or non-customized tasks to a larger, more capable LLM on a server. This ensures optimal performance without sacrificing the benefits of on-device customization. We carefully craft a novel benchmark from multiple question-answer datasets, and show the efficacy of our method in the LLM customization.", }

The customization of large language models (LLMs) for user-specified tasks gets important. However, maintaining all the customized LLMs on cloud servers incurs substantial memory and computational overheads, and uploading user data can also lead to privacy concerns. On-device LLMs can offer a promising solution by mitigating these issues. Yet, the performance of on-device LLMs is inherently constrained by the limitations of small-scaled models. To overcome these restrictions, we first propose Crayon, a novel approach for on-device LLM customization. Crayon begins by constructing a pool of diverse base adapters, and then we instantly blend them into a customized adapter without extra training. In addition, we develop a device-server hybrid inference strategy, which deftly allocates more demanding queries or non-customized tasks to a larger, more capable LLM on a server. This ensures optimal performance without sacrificing the benefits of on-device customization. We carefully craft a novel benchmark from multiple question-answer datasets, and show the efficacy of our method in the LLM customization.

Crayon: Customized On-Device LLM via Instant Adapter Blending and Edge-Server Hybrid Inference

acl-long.204

Poster

2406.07007

https://aclanthology.org/2024.acl-long.204/

https://aclanthology.org/2024.acl-long.205.bib

@inproceedings{lee-etal-2024-fleur, title = "{FLEUR}: An Explainable Reference-Free Evaluation Metric for Image Captioning Using a Large Multimodal Model", author = "Lee, Yebin and Park, Imseong and Kang, Myungjoo", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.205", pages = "3732--3746", abstract = "Most existing image captioning evaluation metrics focus on assigning a single numerical score to a caption by comparing it with reference captions. However, these methods do not provide an explanation for the assigned score. Moreover, reference captions are expensive to acquire. In this paper, we propose FLEUR, an explainable reference-free metric to introduce explainability into image captioning evaluation metrics. By leveraging a large multimodal model, FLEUR can evaluate the caption against the image without the need for reference captions, and provide the explanation for the assigned score. We introduce score smoothing to align as closely as possible with human judgment and to be robust to user-defined grading criteria. FLEUR achieves high correlations with human judgment across various image captioning evaluation benchmarks and reaches state-of-the-art results on Flickr8k-CF, COMPOSITE, and Pascal-50S within the domain of reference-free evaluation metrics. Our source code and results are publicly available at: https://github.com/Yebin46/FLEUR.", }

Most existing image captioning evaluation metrics focus on assigning a single numerical score to a caption by comparing it with reference captions. However, these methods do not provide an explanation for the assigned score. Moreover, reference captions are expensive to acquire. In this paper, we propose FLEUR, an explainable reference-free metric to introduce explainability into image captioning evaluation metrics. By leveraging a large multimodal model, FLEUR can evaluate the caption against the image without the need for reference captions, and provide the explanation for the assigned score. We introduce score smoothing to align as closely as possible with human judgment and to be robust to user-defined grading criteria. FLEUR achieves high correlations with human judgment across various image captioning evaluation benchmarks and reaches state-of-the-art results on Flickr8k-CF, COMPOSITE, and Pascal-50S within the domain of reference-free evaluation metrics. Our source code and results are publicly available at: https://github.com/Yebin46/FLEUR.

FLEUR: An Explainable Reference-Free Evaluation Metric for Image Captioning Using a Large Multimodal Model

acl-long.205

Poster

2406.06004

https://aclanthology.org/2024.acl-long.205/

https://aclanthology.org/2024.acl-long.206.bib

@inproceedings{wang-etal-2024-mentalmanip, title = "{M}ental{M}anip: A Dataset For Fine-grained Analysis of Mental Manipulation in Conversations", author = "Wang, Yuxin and Yang, Ivory and Hassanpour, Saeed and Vosoughi, Soroush", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.206", pages = "3747--3764", abstract = "Mental manipulation, a significant form of abuse in interpersonal conversations, presents a challenge to identify due to its context-dependent and often subtle nature. The detection of manipulative language is essential for protecting potential victims, yet the field of Natural Language Processing (NLP) currently faces a scarcity of resources and research on this topic. Our study addresses this gap by introducing a new dataset, named $\textsc{MentalManip}$, which consists of 4,000 annotated fictional dialogues. This dataset enables a comprehensive analysis of mental manipulation, pinpointing both the techniques utilized for manipulation and the vulnerabilities targeted in victims. Our research further explores the effectiveness of leading-edge models in recognizing manipulative dialogue and its components through a series of experiments with various configurations. The results demonstrate that these models inadequately identify and categorize manipulative content. Attempts to improve their performance by fine-tuning with existing datasets on mental health and toxicity have not overcome these limitations. We anticipate that $\textsc{MentalManip}$ will stimulate further research, leading to progress in both understanding and mitigating the impact of mental manipulation in conversations.", }

Mental manipulation, a significant form of abuse in interpersonal conversations, presents a challenge to identify due to its context-dependent and often subtle nature. The detection of manipulative language is essential for protecting potential victims, yet the field of Natural Language Processing (NLP) currently faces a scarcity of resources and research on this topic. Our study addresses this gap by introducing a new dataset, named $\textsc{MentalManip}$, which consists of 4,000 annotated fictional dialogues. This dataset enables a comprehensive analysis of mental manipulation, pinpointing both the techniques utilized for manipulation and the vulnerabilities targeted in victims. Our research further explores the effectiveness of leading-edge models in recognizing manipulative dialogue and its components through a series of experiments with various configurations. The results demonstrate that these models inadequately identify and categorize manipulative content. Attempts to improve their performance by fine-tuning with existing datasets on mental health and toxicity have not overcome these limitations. We anticipate that $\textsc{MentalManip}$ will stimulate further research, leading to progress in both understanding and mitigating the impact of mental manipulation in conversations.

MentalManip: A Dataset For Fine-grained Analysis of Mental Manipulation in Conversations

acl-long.206

Oral

2405.16584

https://aclanthology.org/2024.acl-long.206/

https://aclanthology.org/2024.acl-long.207.bib

@inproceedings{dai-etal-2024-mpcoder, title = "{MPC}oder: Multi-user Personalized Code Generator with Explicit and Implicit Style Representation Learning", author = "Dai, Zhenlong and Yao, Chang and Han, WenKang and Yuanying, Yuanying and Gao, Zhipeng and Chen, Jingyuan", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.207", pages = "3765--3780", abstract = "Large Language Models (LLMs) have demonstrated great potential for assisting developers in their daily development. However, most research focuses on generating correct code, how to use LLMs to generate personalized code has seldom been investigated. To bridge this gap, we proposed MPCoder (Multi-user Personalized Code Generator) to generate personalized code for multiple users. To better learn coding style features, we utilize explicit coding style residual learning to capture the syntax code style standards and implicit style learning to capture the semantic code style conventions. We train a multi-user style adapter to better differentiate the implicit feature representations of different users through contrastive learning, ultimately enabling personalized code generation for multiple users. We further propose a novel evaluation metric for estimating similarities between codes of different coding styles. The experimental results show the effectiveness of our approach for this novel task.", }

Large Language Models (LLMs) have demonstrated great potential for assisting developers in their daily development. However, most research focuses on generating correct code, how to use LLMs to generate personalized code has seldom been investigated. To bridge this gap, we proposed MPCoder (Multi-user Personalized Code Generator) to generate personalized code for multiple users. To better learn coding style features, we utilize explicit coding style residual learning to capture the syntax code style standards and implicit style learning to capture the semantic code style conventions. We train a multi-user style adapter to better differentiate the implicit feature representations of different users through contrastive learning, ultimately enabling personalized code generation for multiple users. We further propose a novel evaluation metric for estimating similarities between codes of different coding styles. The experimental results show the effectiveness of our approach for this novel task.

MPCoder: Multi-user Personalized Code Generator with Explicit and Implicit Style Representation Learning

acl-long.207

Poster

2406.17255

https://aclanthology.org/2024.acl-long.207/

https://aclanthology.org/2024.acl-long.208.bib

@inproceedings{patel-etal-2024-datadreamer, title = "{D}ata{D}reamer: A Tool for Synthetic Data Generation and Reproducible {LLM} Workflows", author = "Patel, Ajay and Raffel, Colin and Callison-Burch, Chris", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.208", pages = "3781--3799", abstract = "Large language models (LLMs) have become a dominant and important tool for NLP researchers in a wide range of tasks. Today, many researchers use LLMs in synthetic data generation, task evaluation, fine-tuning, distillation, and other model-in-the-loop research workflows. However, challenges arise when using these models that stem from their scale, their closed source nature, and the lack of standardized tooling for these new and emerging workflows. The rapid rise to prominence of these models and these unique challenges has had immediate adverse impacts on open science and on the reproducibility of work that uses them. In this ACL 2024 theme track paper, we introduce DataDreamer, an open source Python library that allows researchers to write simple code to implement powerful LLM workflows. DataDreamer also helps researchers adhere to best practices that we propose to encourage open science and reproducibility. The library and documentation are available at: https://github.com/datadreamer-dev/DataDreamer.", }

Large language models (LLMs) have become a dominant and important tool for NLP researchers in a wide range of tasks. Today, many researchers use LLMs in synthetic data generation, task evaluation, fine-tuning, distillation, and other model-in-the-loop research workflows. However, challenges arise when using these models that stem from their scale, their closed source nature, and the lack of standardized tooling for these new and emerging workflows. The rapid rise to prominence of these models and these unique challenges has had immediate adverse impacts on open science and on the reproducibility of work that uses them. In this ACL 2024 theme track paper, we introduce DataDreamer, an open source Python library that allows researchers to write simple code to implement powerful LLM workflows. DataDreamer also helps researchers adhere to best practices that we propose to encourage open science and reproducibility. The library and documentation are available at: https://github.com/datadreamer-dev/DataDreamer.

DataDreamer: A Tool for Synthetic Data Generation and Reproducible LLM Workflows

acl-long.208

Poster

2402.10379

https://huggingface.co/papers/2402.10379

https://aclanthology.org/2024.acl-long.208/

https://aclanthology.org/2024.acl-long.209.bib

@inproceedings{shao-etal-2024-understanding, title = "Understanding and Addressing the Under-Translation Problem from the Perspective of Decoding Objective", author = "Shao, Chenze and Meng, Fandong and Zeng, Jiali and Zhou, Jie", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.209", pages = "3800--3814", abstract = "Neural Machine Translation (NMT) has made remarkable progress over the past years. However, under-translation and over-translation remain two challenging problems in state-of-the-art NMT systems. In this work, we conduct an in-depth analysis on the underlying cause of under-translation in NMT, providing an explanation from the perspective of decoding objective. To optimize the beam search objective, the model tends to overlook words it is less confident about, leading to the under-translation phenomenon. Correspondingly, the model{'}s confidence in predicting the End Of Sentence (EOS) diminishes when under-translation occurs, serving as a mild penalty for under-translated candidates. Building upon this analysis, we propose employing the confidence of predicting EOS as a detector for under-translation, and strengthening the confidence-based penalty to penalize candidates with a high risk of under-translation.Experiments on both synthetic and real-world data show that our method can accurately detect and rectify under-translated outputs, with minor impact on other correct translations.", }

Neural Machine Translation (NMT) has made remarkable progress over the past years. However, under-translation and over-translation remain two challenging problems in state-of-the-art NMT systems. In this work, we conduct an in-depth analysis on the underlying cause of under-translation in NMT, providing an explanation from the perspective of decoding objective. To optimize the beam search objective, the model tends to overlook words it is less confident about, leading to the under-translation phenomenon. Correspondingly, the model{'}s confidence in predicting the End Of Sentence (EOS) diminishes when under-translation occurs, serving as a mild penalty for under-translated candidates. Building upon this analysis, we propose employing the confidence of predicting EOS as a detector for under-translation, and strengthening the confidence-based penalty to penalize candidates with a high risk of under-translation.Experiments on both synthetic and real-world data show that our method can accurately detect and rectify under-translated outputs, with minor impact on other correct translations.

Understanding and Addressing the Under-Translation Problem from the Perspective of Decoding Objective

acl-long.209

Poster

2405.18922

https://aclanthology.org/2024.acl-long.209/

https://aclanthology.org/2024.acl-long.210.bib

@inproceedings{liu-etal-2024-identifying, title = "Identifying while Learning for Document Event Causality Identification", author = "Liu, Cheng and Xiang, Wei and Wang, Bang", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.210", pages = "3815--3827", abstract = "Event Causality Identification (ECI) aims to detect whether there exists a causal relation between two events in a document. Existing studies adopt a kind of *identifying after learning* paradigm, where events{'} representations are first learned and then used for the identification. Furthermore, they mainly focus on the causality existence, but ignoring causal direction. In this paper, we take care of the causal direction and propose a new *identifying while learning* mode for the ECI task. We argue that a few causal relations can be easily identified with high confidence, and the directionality and structure of these identified causalities can be utilized to update events{'} representations for boosting next round of causality identification. To this end, this paper designs an *iterative learning and identifying framework*: In each iteration, we construct an event causality graph, on which events{'} causal structure representations are updated for boosting causal identification. Experiments on two public datasets show that our approach outperforms the state-of-the-art algorithms in both evaluations for causality existence identification and direction identification.", }

Event Causality Identification (ECI) aims to detect whether there exists a causal relation between two events in a document. Existing studies adopt a kind of *identifying after learning* paradigm, where events{'} representations are first learned and then used for the identification. Furthermore, they mainly focus on the causality existence, but ignoring causal direction. In this paper, we take care of the causal direction and propose a new *identifying while learning* mode for the ECI task. We argue that a few causal relations can be easily identified with high confidence, and the directionality and structure of these identified causalities can be utilized to update events{'} representations for boosting next round of causality identification. To this end, this paper designs an *iterative learning and identifying framework*: In each iteration, we construct an event causality graph, on which events{'} causal structure representations are updated for boosting causal identification. Experiments on two public datasets show that our approach outperforms the state-of-the-art algorithms in both evaluations for causality existence identification and direction identification.

Identifying while Learning for Document Event Causality Identification

acl-long.210

Poster

2405.20608

https://aclanthology.org/2024.acl-long.210/

https://aclanthology.org/2024.acl-long.211.bib

@inproceedings{he-etal-2024-olympiadbench, title = "{O}lympiad{B}ench: A Challenging Benchmark for Promoting {AGI} with Olympiad-Level Bilingual Multimodal Scientific Problems", author = "He, Chaoqun and Luo, Renjie and Bai, Yuzhuo and Hu, Shengding and Thai, Zhen and Shen, Junhao and Hu, Jinyi and Han, Xu and Huang, Yujie and Zhang, Yuxiang and Liu, Jie and Qi, Lei and Liu, Zhiyuan and Sun, Maosong", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.211", pages = "3828--3850", abstract = "Recent advancements have seen Large Language Models (LLMs) and Large Multimodal Models (LMMs) surpassing general human capabilities in various tasks, approaching the proficiency level of human experts across multiple domains. With traditional benchmarks becoming less challenging for these models, new rigorous challenges are essential to gauge their advanced abilities. In this work, we present OlympiadBench, an Olympiad-level bilingual multimodal scientific benchmark, featuring 8,476 problems from Olympiad-level mathematics and physics competitions, including the Chinese college entrance exam. Each problem is detailed with expert-level annotations for step-by-step reasoning. Evaluating top-tier models on OlympiadBench, we implement a comprehensive assessment methodology to accurately evaluate model responses. Notably, the best-performing model, GPT-4V, attains an average score of 17.97{\%} on OlympiadBench, with a mere 10.74{\%} in physics, highlighting the benchmark rigor and the intricacy of physical reasoning. Our analysis orienting GPT-4V points out prevalent issues with hallucinations, knowledge omissions, and logical fallacies. We hope that our challenging benchmark can serve as a valuable resource for helping future AGI research endeavors. The data and evaluation code are available at \url{https://github.com/OpenBMB/OlympiadBench}", }

Recent advancements have seen Large Language Models (LLMs) and Large Multimodal Models (LMMs) surpassing general human capabilities in various tasks, approaching the proficiency level of human experts across multiple domains. With traditional benchmarks becoming less challenging for these models, new rigorous challenges are essential to gauge their advanced abilities. In this work, we present OlympiadBench, an Olympiad-level bilingual multimodal scientific benchmark, featuring 8,476 problems from Olympiad-level mathematics and physics competitions, including the Chinese college entrance exam. Each problem is detailed with expert-level annotations for step-by-step reasoning. Evaluating top-tier models on OlympiadBench, we implement a comprehensive assessment methodology to accurately evaluate model responses. Notably, the best-performing model, GPT-4V, attains an average score of 17.97{\%} on OlympiadBench, with a mere 10.74{\%} in physics, highlighting the benchmark rigor and the intricacy of physical reasoning. Our analysis orienting GPT-4V points out prevalent issues with hallucinations, knowledge omissions, and logical fallacies. We hope that our challenging benchmark can serve as a valuable resource for helping future AGI research endeavors. The data and evaluation code are available at \url{https://github.com/OpenBMB/OlympiadBench}

OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problems

acl-long.211

Poster

2402.14008

https://huggingface.co/papers/2402.14008

https://aclanthology.org/2024.acl-long.211/

https://aclanthology.org/2024.acl-long.212.bib

@inproceedings{xue-etal-2024-insert, title = "Insert or Attach: Taxonomy Completion via Box Embedding", author = "Xue, Wei and Shen, Yongliang and Ren, Wenqi and Guo, Jietian and Pu, Shiliang and Lu, Weiming", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.212", pages = "3851--3863", abstract = "Taxonomy completion, enriching existing taxonomies by inserting new concepts as parents or attaching them as children, has gained significant interest. Previous approaches embed concepts as vectors in Euclidean space, which makes it difficult to model asymmetric relations in taxonomy. In addition, they introduce pseudo-leaves to convert attachment cases into insertion cases, leading to an incorrect bias in network learning dominated by numerous pseudo-leaves. Addressing these, our framework, TaxBox, leverages box containment and center closeness to design two specialized geometric scorers within the box embedding space. These scorers are tailored for insertion and attachment operations and can effectively capture intrinsic relationships between concepts by optimizing on a granular box constraint loss. We employ a dynamic ranking loss mechanism to balance the scores from these scorers, allowing adaptive adjustments of insertion and attachment scores. Experiments on four real-world datasets show that TaxBox significantly outperforms previous methods, yielding substantial improvements over prior methods in real-world datasets, with average performance boosts of 6.7{\%}, 34.9{\%}, and 51.4{\%} in MRR, Hit@1, and Prec@1, respectively.", }

Taxonomy completion, enriching existing taxonomies by inserting new concepts as parents or attaching them as children, has gained significant interest. Previous approaches embed concepts as vectors in Euclidean space, which makes it difficult to model asymmetric relations in taxonomy. In addition, they introduce pseudo-leaves to convert attachment cases into insertion cases, leading to an incorrect bias in network learning dominated by numerous pseudo-leaves. Addressing these, our framework, TaxBox, leverages box containment and center closeness to design two specialized geometric scorers within the box embedding space. These scorers are tailored for insertion and attachment operations and can effectively capture intrinsic relationships between concepts by optimizing on a granular box constraint loss. We employ a dynamic ranking loss mechanism to balance the scores from these scorers, allowing adaptive adjustments of insertion and attachment scores. Experiments on four real-world datasets show that TaxBox significantly outperforms previous methods, yielding substantial improvements over prior methods in real-world datasets, with average performance boosts of 6.7{\%}, 34.9{\%}, and 51.4{\%} in MRR, Hit@1, and Prec@1, respectively.

Insert or Attach: Taxonomy Completion via Box Embedding

acl-long.212

Poster

2305.11004

https://aclanthology.org/2024.acl-long.212/

https://aclanthology.org/2024.acl-long.213.bib

@inproceedings{lee-etal-2024-semiparametric, title = "Semiparametric Token-Sequence Co-Supervision", author = "Lee, Hyunji and Kim, Doyoung and Jun, Jihoon and Joo, Se June and Jang, Joel and On, Kyoung-Woon and Seo, Minjoon", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.213", pages = "3864--3882", abstract = "In this work, we introduce a semiparametric token-sequence co-supervision training method. It trains a language model by simultaneously leveraging supervision from the traditional next token prediction loss which is calculated over the parametric token embedding space and the next sequence prediction loss which is calculated over the nonparametric sequence embedding space. The nonparametric sequence embedding space is constructed by a separate language model tasked to condense an input text into a single representative embedding. Our experiments demonstrate that a model trained via both supervisions consistently surpasses models trained via each supervision independently. Analysis suggests that this co-supervision encourages a broader generalization capability across the model. Especially, the robustness of parametric token space which is established during the pretraining step tends to effectively enhance the stability of nonparametric sequence embedding space, a new space established by another language model.", }

In this work, we introduce a semiparametric token-sequence co-supervision training method. It trains a language model by simultaneously leveraging supervision from the traditional next token prediction loss which is calculated over the parametric token embedding space and the next sequence prediction loss which is calculated over the nonparametric sequence embedding space. The nonparametric sequence embedding space is constructed by a separate language model tasked to condense an input text into a single representative embedding. Our experiments demonstrate that a model trained via both supervisions consistently surpasses models trained via each supervision independently. Analysis suggests that this co-supervision encourages a broader generalization capability across the model. Especially, the robustness of parametric token space which is established during the pretraining step tends to effectively enhance the stability of nonparametric sequence embedding space, a new space established by another language model.

Semiparametric Token-Sequence Co-Supervision

acl-long.213

Poster

2403.09024

https://aclanthology.org/2024.acl-long.213/

https://aclanthology.org/2024.acl-long.214.bib

@inproceedings{guo-etal-2024-instruction, title = "Instruction Fusion: Advancing Prompt Evolution through Hybridization", author = "Guo, Weidong and Yang, Jiuding and Yang, Kaitong and Li, Xiangyang and Rao, Zhuwei and Xu, Yu and Niu, Di", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.214", pages = "3883--3893", abstract = "The fine-tuning of Large Language Models (LLMs) specialized in code generation has seen notable advancements through the use of open-domain coding queries. Despite the successes, existing methodologies like Evol-Instruct encounter performance limitations, impeding further enhancements in code generation tasks. This paper examines the constraints of existing prompt evolution techniques and introduces a novel approach, Instruction Fusion (IF). IF innovatively combines two distinct prompts through a hybridization process, thereby enhancing the evolution of training prompts for code LLMs. Our experimental results reveal that the proposed novel method effectively addresses the shortcomings of prior methods, significantly improving the performance of Code LLMs across five code generation benchmarks, namely HumanEval, HumanEval+, MBPP, MBPP+ and MultiPL-E, which underscore the effectiveness of Instruction Fusion in advancing the capabilities of LLMs in code generation.", }

The fine-tuning of Large Language Models (LLMs) specialized in code generation has seen notable advancements through the use of open-domain coding queries. Despite the successes, existing methodologies like Evol-Instruct encounter performance limitations, impeding further enhancements in code generation tasks. This paper examines the constraints of existing prompt evolution techniques and introduces a novel approach, Instruction Fusion (IF). IF innovatively combines two distinct prompts through a hybridization process, thereby enhancing the evolution of training prompts for code LLMs. Our experimental results reveal that the proposed novel method effectively addresses the shortcomings of prior methods, significantly improving the performance of Code LLMs across five code generation benchmarks, namely HumanEval, HumanEval+, MBPP, MBPP+ and MultiPL-E, which underscore the effectiveness of Instruction Fusion in advancing the capabilities of LLMs in code generation.

Instruction Fusion: Advancing Prompt Evolution through Hybridization

acl-long.214

Poster

2312.15692

https://huggingface.co/papers/2312.15692

https://aclanthology.org/2024.acl-long.214/

https://aclanthology.org/2024.acl-long.215.bib

@inproceedings{zhang-etal-2024-timearena, title = "{T}ime{A}rena: Shaping Efficient Multitasking Language Agents in a Time-Aware Simulation", author = "Zhang, Yikai and Yuan, Siyu and Hu, Caiyu and Richardson, Kyle and Xiao, Yanghua and Chen, Jiangjie", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.215", pages = "3894--3916", abstract = "Despite remarkable advancements in emulating human-like behavior through Large Language Models (LLMs), current textual simulations do not adequately address the notion of time. To this end, we introduce TimeArena, a novel textual simulated environment that incorporates complex temporal dynamics and constraints that better reflect real-life planning scenarios. In TimeArena, agents are asked to complete multiple tasks as soon as possible, allowing for parallel processing to save time. We implement the dependency between actions, the time duration for each action, and the occupancy of the agent and the objects in the environment. TimeArena grounds to 30 real-world tasks in cooking, household activity, and laboratory work. We conduct extensive experiments with various LLMs using TimeArena. Our findings reveal that even the most powerful models, e.g., GPT-4, still lag behind humans in effective multitasking, underscoring the need for enhanced temporal awareness in the development of language agents.", }

Despite remarkable advancements in emulating human-like behavior through Large Language Models (LLMs), current textual simulations do not adequately address the notion of time. To this end, we introduce TimeArena, a novel textual simulated environment that incorporates complex temporal dynamics and constraints that better reflect real-life planning scenarios. In TimeArena, agents are asked to complete multiple tasks as soon as possible, allowing for parallel processing to save time. We implement the dependency between actions, the time duration for each action, and the occupancy of the agent and the objects in the environment. TimeArena grounds to 30 real-world tasks in cooking, household activity, and laboratory work. We conduct extensive experiments with various LLMs using TimeArena. Our findings reveal that even the most powerful models, e.g., GPT-4, still lag behind humans in effective multitasking, underscoring the need for enhanced temporal awareness in the development of language agents.

TimeArena: Shaping Efficient Multitasking Language Agents in a Time-Aware Simulation

acl-long.215

Poster

2402.05733

https://aclanthology.org/2024.acl-long.215/

https://aclanthology.org/2024.acl-long.216.bib

@inproceedings{zeng-etal-2024-exploring, title = "Exploring Memorization in Fine-tuned Language Models", author = "Zeng, Shenglai and Li, Yaxin and Ren, Jie and Liu, Yiding and Xu, Han and He, Pengfei and Xing, Yue and Wang, Shuaiqiang and Tang, Jiliang and Yin, Dawei", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.216", pages = "3917--3948", abstract = "Large language models (LLMs) have shown great capabilities in various tasks but also exhibited memorization of training data, raising tremendous privacy and copyright concerns. While prior works have studied memorization during pre-training, the exploration of memorization during fine-tuning is rather limited. Compared to pre-training, fine-tuning typically involves more sensitive data and diverse objectives, thus may bring distinct privacy risks and unique memorization behaviors. In this work, we conduct the first comprehensive analysis to explore language models{'} (LMs) memorization during fine-tuning across tasks. Our studies with open-sourced and our own fine-tuned LMs across various tasks indicate that memorization presents a strong disparity among different fine-tuning tasks. We provide an intuitive explanation of this task disparity via sparse coding theory and unveil a strong correlation between memorization and attention score distribution.", }

Large language models (LLMs) have shown great capabilities in various tasks but also exhibited memorization of training data, raising tremendous privacy and copyright concerns. While prior works have studied memorization during pre-training, the exploration of memorization during fine-tuning is rather limited. Compared to pre-training, fine-tuning typically involves more sensitive data and diverse objectives, thus may bring distinct privacy risks and unique memorization behaviors. In this work, we conduct the first comprehensive analysis to explore language models{'} (LMs) memorization during fine-tuning across tasks. Our studies with open-sourced and our own fine-tuned LMs across various tasks indicate that memorization presents a strong disparity among different fine-tuning tasks. We provide an intuitive explanation of this task disparity via sparse coding theory and unveil a strong correlation between memorization and attention score distribution.

Exploring Memorization in Fine-tuned Language Models

acl-long.216

Poster

2310.06714

https://aclanthology.org/2024.acl-long.216/

https://aclanthology.org/2024.acl-long.217.bib

@inproceedings{zhang-etal-2024-towards-real, title = "Towards Real-world Scenario: Imbalanced New Intent Discovery", author = "Zhang, Shun and Chaoran, Yan and Yang, Jian and Liu, Jiaheng and Mo, Ying and Bai, Jiaqi and Li, Tongliang and Li, Zhoujun", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.217", pages = "3949--3963", abstract = "New Intent Discovery (NID) aims at detecting known and previously undefined categories of user intent by utilizing limited labeled and massive unlabeled data. Most prior works often operate under the unrealistic assumption that the distribution of both familiar and new intent classes is uniform, overlooking the skewed and long-tailed distributions frequently encountered in real-world scenarios. To bridge the gap, our work introduces the imbalanced new intent discovery i-NID task, which seeks to identify familiar and novel intent categories within long-tailed distributions. A new benchmark baNID-Bench comprised of three datasets is created to simulate the real-world long-tail distributions. ImbaNID-Bench ranges from broad cross-domain to specific single-domain intent categories, providing a thorough representation of practical use cases. Besides, a robust baseline model ImbaNID is proposed to achieve cluster-friendly intent representations. It includes three stages: model pre-training, generation of reliable pseudo-labels, and robust representation learning that strengthens the model performance to handle the intricacies of real-world data distributions. Our extensive experiments on previous benchmarks and the newly established benchmark demonstrate the superior performance of ImbaNID in addressing the i-NID task, highlighting its potential as a powerful baseline for uncovering and categorizing user intents in imbalanced and long-tailed distributions.", }

New Intent Discovery (NID) aims at detecting known and previously undefined categories of user intent by utilizing limited labeled and massive unlabeled data. Most prior works often operate under the unrealistic assumption that the distribution of both familiar and new intent classes is uniform, overlooking the skewed and long-tailed distributions frequently encountered in real-world scenarios. To bridge the gap, our work introduces the imbalanced new intent discovery i-NID task, which seeks to identify familiar and novel intent categories within long-tailed distributions. A new benchmark baNID-Bench comprised of three datasets is created to simulate the real-world long-tail distributions. ImbaNID-Bench ranges from broad cross-domain to specific single-domain intent categories, providing a thorough representation of practical use cases. Besides, a robust baseline model ImbaNID is proposed to achieve cluster-friendly intent representations. It includes three stages: model pre-training, generation of reliable pseudo-labels, and robust representation learning that strengthens the model performance to handle the intricacies of real-world data distributions. Our extensive experiments on previous benchmarks and the newly established benchmark demonstrate the superior performance of ImbaNID in addressing the i-NID task, highlighting its potential as a powerful baseline for uncovering and categorizing user intents in imbalanced and long-tailed distributions.

Towards Real-world Scenario: Imbalanced New Intent Discovery

acl-long.217

Poster

2406.03127

https://aclanthology.org/2024.acl-long.217/

https://aclanthology.org/2024.acl-long.218.bib

@inproceedings{wang-etal-2024-m4gt, title = "{M}4{GT}-Bench: Evaluation Benchmark for Black-Box Machine-Generated Text Detection", author = "Wang, Yuxia and Mansurov, Jonibek and Ivanov, Petar and Su, Jinyan and Shelmanov, Artem and Tsvigun, Akim and Mohammed Afzal, Osama and Mahmoud, Tarek and Puccetti, Giovanni and Arnold, Thomas and Aji, Alham and Habash, Nizar and Gurevych, Iryna and Nakov, Preslav", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.218", pages = "3964--3992", abstract = "The advent of Large Language Models (LLMs) has brought an unprecedented surge in machine-generated text (MGT) across diverse channels. This raises legitimate concerns about its potential misuse and societal implications. The need to identify and differentiate such content from genuine human-generated text is critical in combating disinformation, preserving the integrity of education and scientific fields, and maintaining trust in communication. In this work, we address this problem by introducing a new benchmark based on a multilingual, multi-domain and multi-generator corpus of MGTs {---} M4GT-Bench. The benchmark is compiled of three tasks: (1) mono-lingual and multi-lingual binary MGT detection; (2) multi-way detection where one need to identify, which particular model generated the text; and (3) mixed human-machine text detection, where a word boundary delimiting MGT from human-written content should be determined. On the developed benchmark, we have tested several MGT detection baselines and also conducted an evaluation of human performance. We see that obtaining good performance in MGT detection usually requires an access to the training data from the same domain and generators. The benchmark is available at https://github.com/mbzuai-nlp/M4GT-Bench.", }

The advent of Large Language Models (LLMs) has brought an unprecedented surge in machine-generated text (MGT) across diverse channels. This raises legitimate concerns about its potential misuse and societal implications. The need to identify and differentiate such content from genuine human-generated text is critical in combating disinformation, preserving the integrity of education and scientific fields, and maintaining trust in communication. In this work, we address this problem by introducing a new benchmark based on a multilingual, multi-domain and multi-generator corpus of MGTs {---} M4GT-Bench. The benchmark is compiled of three tasks: (1) mono-lingual and multi-lingual binary MGT detection; (2) multi-way detection where one need to identify, which particular model generated the text; and (3) mixed human-machine text detection, where a word boundary delimiting MGT from human-written content should be determined. On the developed benchmark, we have tested several MGT detection baselines and also conducted an evaluation of human performance. We see that obtaining good performance in MGT detection usually requires an access to the training data from the same domain and generators. The benchmark is available at https://github.com/mbzuai-nlp/M4GT-Bench.

M4GT-Bench: Evaluation Benchmark for Black-Box Machine-Generated Text Detection

acl-long.218

Poster

2402.11175

https://aclanthology.org/2024.acl-long.218/

https://aclanthology.org/2024.acl-long.219.bib

@inproceedings{wang-etal-2024-instruct, title = "Instruct Once, Chat Consistently in Multiple Rounds: An Efficient Tuning Framework for Dialogue", author = "Wang, Jian and Leong, Chak Tou and Wang, Jiashuo and Lin, Dongding and Li, Wenjie and Wei, Xiaoyong", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.219", pages = "3993--4010", abstract = "Tuning language models for dialogue generation has been a prevalent paradigm for building capable dialogue agents. Yet, traditional tuning narrowly views dialogue generation as resembling other language generation tasks, ignoring the role disparities between two speakers and the multi-round interactive process that dialogues ought to be. Such a manner often leads to unsatisfactory chat consistency for the built agent. In this work, we emphasize the interactive, communicative nature of dialogue and argue that it is more feasible to model the speaker roles of agent and user separately, enabling the agent to adhere to its role consistently. With this in mind, we propose an efficient Multi-round Interactive Dialogue Tuning (Midi-Tuning) framework. It models the agent and user individually with two adapters built upon large language models. The adapters make use of respective utterances round by round in alternating order and they are tuned via a round-level memory caching mechanism. Extensive experiments demonstrate that, our framework performs superior to traditional fine-tuning and harbors the tremendous potential for improving dialogue consistency.", }

Tuning language models for dialogue generation has been a prevalent paradigm for building capable dialogue agents. Yet, traditional tuning narrowly views dialogue generation as resembling other language generation tasks, ignoring the role disparities between two speakers and the multi-round interactive process that dialogues ought to be. Such a manner often leads to unsatisfactory chat consistency for the built agent. In this work, we emphasize the interactive, communicative nature of dialogue and argue that it is more feasible to model the speaker roles of agent and user separately, enabling the agent to adhere to its role consistently. With this in mind, we propose an efficient Multi-round Interactive Dialogue Tuning (Midi-Tuning) framework. It models the agent and user individually with two adapters built upon large language models. The adapters make use of respective utterances round by round in alternating order and they are tuned via a round-level memory caching mechanism. Extensive experiments demonstrate that, our framework performs superior to traditional fine-tuning and harbors the tremendous potential for improving dialogue consistency.

Instruct Once, Chat Consistently in Multiple Rounds: An Efficient Tuning Framework for Dialogue

acl-long.219

Poster

2402.06967

https://aclanthology.org/2024.acl-long.219/

https://aclanthology.org/2024.acl-long.220.bib

@inproceedings{he-etal-2024-softdedup, title = "{S}oft{D}edup: an Efficient Data Reweighting Method for Speeding Up Language Model Pre-training", author = "He, Nan and Xiong, Weichen and Liu, Hanwen and Liao, Yi and Ding, Lei and Zhang, Kai and Tang, Guohua and Han, Xiao and Wei, Yang", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.220", pages = "4011--4022", abstract = "The effectiveness of large language models (LLMs) is often hindered by duplicated data in their extensive pre-training datasets. Current approaches primarily focus on detecting and removing duplicates, which risks the loss of valuable information and neglects the varying degrees of duplication. To address this, we propose a soft deduplication method that maintains dataset integrity while selectively reducing the sampling weight of data with high commonness. Central to our approach is the concept of {``}data commonness{''}, a metric we introduce to quantify the degree of duplication by measuring the occurrence probabilities of samples using an n-gram model. Empirical analysis shows that this method significantly improves training efficiency, achieving comparable perplexity scores with at least a 26{\%} reduction in required training steps. Additionally, it enhances average few-shot downstream accuracy by 1.77{\%} when trained for an equivalent duration. Importantly, this approach consistently improves performance, even on rigorously deduplicated datasets, indicating its potential to complement existing methods and become a standard pre-training process for LLMs.", }

The effectiveness of large language models (LLMs) is often hindered by duplicated data in their extensive pre-training datasets. Current approaches primarily focus on detecting and removing duplicates, which risks the loss of valuable information and neglects the varying degrees of duplication. To address this, we propose a soft deduplication method that maintains dataset integrity while selectively reducing the sampling weight of data with high commonness. Central to our approach is the concept of {``}data commonness{''}, a metric we introduce to quantify the degree of duplication by measuring the occurrence probabilities of samples using an n-gram model. Empirical analysis shows that this method significantly improves training efficiency, achieving comparable perplexity scores with at least a 26{\%} reduction in required training steps. Additionally, it enhances average few-shot downstream accuracy by 1.77{\%} when trained for an equivalent duration. Importantly, this approach consistently improves performance, even on rigorously deduplicated datasets, indicating its potential to complement existing methods and become a standard pre-training process for LLMs.

SoftDedup: an Efficient Data Reweighting Method for Speeding Up Language Model Pre-training

acl-long.220

Poster

2407.06654

https://aclanthology.org/2024.acl-long.220/

https://aclanthology.org/2024.acl-long.221.bib

@inproceedings{bian-etal-2024-rule, title = "Rule or Story, Which is a Better Commonsense Expression for Talking with Large Language Models?", author = "Bian, Ning and Han, Xianpei and Lin, Hongyu and Lu, Yaojie and He, Ben and Sun, Le", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.221", pages = "4023--4043", abstract = "Building machines with commonsense has been a longstanding challenge in NLP due to the reporting bias of commonsense rules and the exposure bias of rule-based commonsense reasoning. In contrast, humans convey and pass down commonsense implicitly through stories. This paper investigates the inherent commonsense ability of large language models (LLMs) expressed through storytelling. We systematically investigate and compare stories and rules for retrieving and leveraging commonsense in LLMs. Experimental results on 28 commonsense QA datasets show that stories outperform rules as the expression for retrieving commonsense from LLMs, exhibiting higher generation confidence and commonsense accuracy. Moreover, stories are the more effective commonsense expression for answering questions regarding daily events, while rules are more effective for scientific questions. This aligns with the reporting bias of commonsense in text corpora. We further show that the correctness and relevance of commonsense stories can be further improved via iterative self-supervised fine-tuning. These findings emphasize the importance of using appropriate language to express, retrieve, and leverage commonsense for LLMs, highlighting a promising direction for better exploiting their commonsense abilities.", }

Building machines with commonsense has been a longstanding challenge in NLP due to the reporting bias of commonsense rules and the exposure bias of rule-based commonsense reasoning. In contrast, humans convey and pass down commonsense implicitly through stories. This paper investigates the inherent commonsense ability of large language models (LLMs) expressed through storytelling. We systematically investigate and compare stories and rules for retrieving and leveraging commonsense in LLMs. Experimental results on 28 commonsense QA datasets show that stories outperform rules as the expression for retrieving commonsense from LLMs, exhibiting higher generation confidence and commonsense accuracy. Moreover, stories are the more effective commonsense expression for answering questions regarding daily events, while rules are more effective for scientific questions. This aligns with the reporting bias of commonsense in text corpora. We further show that the correctness and relevance of commonsense stories can be further improved via iterative self-supervised fine-tuning. These findings emphasize the importance of using appropriate language to express, retrieve, and leverage commonsense for LLMs, highlighting a promising direction for better exploiting their commonsense abilities.

Rule or Story, Which is a Better Commonsense Expression for Talking with Large Language Models?

acl-long.221

Poster

2402.14355

https://aclanthology.org/2024.acl-long.221/

https://aclanthology.org/2024.acl-long.222.bib

@inproceedings{tan-etal-2024-learning, title = "Learning Global Controller in Latent Space for Parameter-Efficient Fine-Tuning", author = "Tan, Zeqi and Shen, Yongliang and Cheng, Xiaoxia and Zong, Chang and Zhang, Wenqi and Shao, Jian and Lu, Weiming and Zhuang, Yueting", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.222", pages = "4044--4055", abstract = "While large language models (LLMs) have showcased remarkable prowess in various natural language processing tasks, their training costs are exorbitant. Consequently, a plethora of parameter-efficient fine-tuning methods have emerged to tailor large models for downstream tasks, including low-rank training. Recent approaches either amalgamate existing fine-tuning methods or dynamically adjust rank allocation. Nonetheless, these methods continue to grapple with issues like local optimization, inability to train with full rank and lack of focus on specific tasks. In this paper, we introduce an innovative parameter-efficient method for exploring optimal solutions within latent space. More specifically, we introduce a set of latent units designed to iteratively extract input representations from LLMs, continuously refining informative features that enhance downstream task performance. Due to the small and independent nature of the latent units in relation to input size, this significantly reduces training memory requirements. Additionally, we employ an asymmetric attention mechanism to facilitate bidirectional interaction between latent units and freezed LLM representations, thereby mitigating issues associated with non-full-rank training. Furthermore, we apply distillation over hidden states during the interaction, which guarantees a trimmed number of trainable parameters.Experimental results demonstrate that our approach achieves state-of-the-art performance on a range of natural language understanding, generation and reasoning tasks.", }

While large language models (LLMs) have showcased remarkable prowess in various natural language processing tasks, their training costs are exorbitant. Consequently, a plethora of parameter-efficient fine-tuning methods have emerged to tailor large models for downstream tasks, including low-rank training. Recent approaches either amalgamate existing fine-tuning methods or dynamically adjust rank allocation. Nonetheless, these methods continue to grapple with issues like local optimization, inability to train with full rank and lack of focus on specific tasks. In this paper, we introduce an innovative parameter-efficient method for exploring optimal solutions within latent space. More specifically, we introduce a set of latent units designed to iteratively extract input representations from LLMs, continuously refining informative features that enhance downstream task performance. Due to the small and independent nature of the latent units in relation to input size, this significantly reduces training memory requirements. Additionally, we employ an asymmetric attention mechanism to facilitate bidirectional interaction between latent units and freezed LLM representations, thereby mitigating issues associated with non-full-rank training. Furthermore, we apply distillation over hidden states during the interaction, which guarantees a trimmed number of trainable parameters.Experimental results demonstrate that our approach achieves state-of-the-art performance on a range of natural language understanding, generation and reasoning tasks.

Learning Global Controller in Latent Space for Parameter-Efficient Fine-Tuning

acl-long.222

Poster

https://aclanthology.org/2024.acl-long.222/

https://aclanthology.org/2024.acl-long.223.bib

@inproceedings{chen-etal-2024-camml, title = "{C}a{MML}: Context-Aware Multimodal Learner for Large Models", author = "Chen, Yixin and Zhang, Shuai and Han, Boran and He, Tong and Li, Bo", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.223", pages = "4056--4071", abstract = "In this work, we introduce Context-Aware MultiModal Learner (CaMML), for tuning large multimodal models (LMMs). CaMML, a lightweight module, is crafted to seamlessly integrate multimodal contextual samples into large models, thereby empowering the model to derive knowledge from analogous, domain-specific, up-to-date information and make grounded inferences. Importantly, CaMML is highly scalable and can efficiently handle lengthy multimodal context examples owing to its hierarchical design. Based on CaMML, we have developed two multimodal models, CaMML-7B and CaMML-13B, that have shown exceptional performance across an array of benchmark datasets for multimodal tasks. Remarkably, CaMML-13B achieves the state-of-the-art performance on over ten widely recognized multimodal benchmark datasets, surpassing LLaVA-1.5 (13B) with a noticeable margin, without integration of any external resources. Moreover, we have conducted extensive ablative studies to inspect the inner workings of CaMML and performed qualitative analyses to showcase its effectiveness in handling real-world challenging cases. Code and models are available at: https://github.com/amazon-science/camml.", }

In this work, we introduce Context-Aware MultiModal Learner (CaMML), for tuning large multimodal models (LMMs). CaMML, a lightweight module, is crafted to seamlessly integrate multimodal contextual samples into large models, thereby empowering the model to derive knowledge from analogous, domain-specific, up-to-date information and make grounded inferences. Importantly, CaMML is highly scalable and can efficiently handle lengthy multimodal context examples owing to its hierarchical design. Based on CaMML, we have developed two multimodal models, CaMML-7B and CaMML-13B, that have shown exceptional performance across an array of benchmark datasets for multimodal tasks. Remarkably, CaMML-13B achieves the state-of-the-art performance on over ten widely recognized multimodal benchmark datasets, surpassing LLaVA-1.5 (13B) with a noticeable margin, without integration of any external resources. Moreover, we have conducted extensive ablative studies to inspect the inner workings of CaMML and performed qualitative analyses to showcase its effectiveness in handling real-world challenging cases. Code and models are available at: https://github.com/amazon-science/camml.

CaMML: Context-Aware Multimodal Learner for Large Models

acl-long.223

Oral

2401.03149

https://aclanthology.org/2024.acl-long.223/

https://aclanthology.org/2024.acl-long.224.bib

@inproceedings{wang-etal-2024-maven, title = "{MAVEN}-{ARG}: Completing the Puzzle of All-in-One Event Understanding Dataset with Event Argument Annotation", author = "Wang, Xiaozhi and Peng, Hao and Guan, Yong and Zeng, Kaisheng and Chen, Jianhui and Hou, Lei and Han, Xu and Lin, Yankai and Liu, Zhiyuan and Xie, Ruobing and Zhou, Jie and Li, Juanzi", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.224", pages = "4072--4091", abstract = "Understanding events in texts is a core objective of natural language understanding, which requires detecting event occurrences, extracting event arguments, and analyzing inter-event relationships. However, due to the annotation challenges brought by task complexity, a large-scale dataset covering the full process of event understanding has long been absent. In this paper, we introduce MAVEN-Arg, which augments MAVEN datasets with event argument annotations, making the first all-in-one dataset supporting event detection, event argument extraction (EAE), and event relation extraction. As an EAE benchmark, MAVEN-Arg offers three main advantages: (1) a comprehensive schema covering 162 event types and 612 argument roles, all with expert-written definitions and examples; (2) a large data scale, containing 98,591 events and 290,613 arguments obtained with laborious human annotation; (3) the exhaustive annotation supporting all task variants of EAE, which annotates both entity and non-entity event arguments in document level. Experiments indicate that MAVEN-Arg is quite challenging for both fine-tuned EAE models and proprietary large language models (LLMs). Furthermore, to demonstrate the benefits of an all-in-one dataset, we preliminarily explore a potential application, future event prediction, with LLMs. MAVEN-Arg and codes can be obtained from https://github.com/THU-KEG/MAVEN-Argument.", }

Understanding events in texts is a core objective of natural language understanding, which requires detecting event occurrences, extracting event arguments, and analyzing inter-event relationships. However, due to the annotation challenges brought by task complexity, a large-scale dataset covering the full process of event understanding has long been absent. In this paper, we introduce MAVEN-Arg, which augments MAVEN datasets with event argument annotations, making the first all-in-one dataset supporting event detection, event argument extraction (EAE), and event relation extraction. As an EAE benchmark, MAVEN-Arg offers three main advantages: (1) a comprehensive schema covering 162 event types and 612 argument roles, all with expert-written definitions and examples; (2) a large data scale, containing 98,591 events and 290,613 arguments obtained with laborious human annotation; (3) the exhaustive annotation supporting all task variants of EAE, which annotates both entity and non-entity event arguments in document level. Experiments indicate that MAVEN-Arg is quite challenging for both fine-tuned EAE models and proprietary large language models (LLMs). Furthermore, to demonstrate the benefits of an all-in-one dataset, we preliminarily explore a potential application, future event prediction, with LLMs. MAVEN-Arg and codes can be obtained from https://github.com/THU-KEG/MAVEN-Argument.

MAVEN-ARG: Completing the Puzzle of All-in-One Event Understanding Dataset with Event Argument Annotation

acl-long.224

Oral

2311.09105

https://huggingface.co/papers/2311.09105

https://aclanthology.org/2024.acl-long.224/

https://aclanthology.org/2024.acl-long.225.bib

@inproceedings{fan-etal-2024-nphardeval, title = "{NPH}ard{E}val: Dynamic Benchmark on Reasoning Ability of Large Language Models via Complexity Classes", author = "Fan, Lizhou and Hua, Wenyue and Li, Lingyao and Ling, Haoyang and Zhang, Yongfeng", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.225", pages = "4092--4114", abstract = "Complex reasoning ability is one of the most important features of Large Language Models (LLMs). Numerous benchmarks have been established to assess the reasoning abilities of LLMs. However, they are inadequate in offering a rigorous evaluation and prone to the risk of overfitting, as these publicly accessible and static benchmarks allow models to potentially tailor their responses to specific benchmark metrics, thereby inflating their performance. Addressing these limitations, we introduce a new benchmark NPHardEval. It contains a broad spectrum of 900 algorithmic questions belonging up to the NP-Hard complexity class, offering a rigorous measure of the reasoning ability of LLMs utilizing computational complexity. Moreover, this benchmark is designed with a dynamic update mechanism, where the datapoints are refreshed on a monthly basis. Such regular updates play a crucial role in mitigating the risk of LLMs overfitting to the benchmark, promoting a more accurate and reliable assessment of their reasoning capabilities. The benchmark dataset and code of NPHardEval are available at https://github.com/casmlab/NPHardEval.", }

Complex reasoning ability is one of the most important features of Large Language Models (LLMs). Numerous benchmarks have been established to assess the reasoning abilities of LLMs. However, they are inadequate in offering a rigorous evaluation and prone to the risk of overfitting, as these publicly accessible and static benchmarks allow models to potentially tailor their responses to specific benchmark metrics, thereby inflating their performance. Addressing these limitations, we introduce a new benchmark NPHardEval. It contains a broad spectrum of 900 algorithmic questions belonging up to the NP-Hard complexity class, offering a rigorous measure of the reasoning ability of LLMs utilizing computational complexity. Moreover, this benchmark is designed with a dynamic update mechanism, where the datapoints are refreshed on a monthly basis. Such regular updates play a crucial role in mitigating the risk of LLMs overfitting to the benchmark, promoting a more accurate and reliable assessment of their reasoning capabilities. The benchmark dataset and code of NPHardEval are available at https://github.com/casmlab/NPHardEval.

NPHardEval: Dynamic Benchmark on Reasoning Ability of Large Language Models via Complexity Classes

acl-long.225

Poster

2312.14890

https://aclanthology.org/2024.acl-long.225/

https://aclanthology.org/2024.acl-long.226.bib

@inproceedings{he-etal-2024-watermarks, title = "Can Watermarks Survive Translation? On the Cross-lingual Consistency of Text Watermark for Large Language Models", author = "He, Zhiwei and Zhou, Binglin and Hao, Hongkun and Liu, Aiwei and Wang, Xing and Tu, Zhaopeng and Zhang, Zhuosheng and Wang, Rui", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.226", pages = "4115--4129", abstract = "Text watermarking technology aims to tag and identify content produced by large language models (LLMs) to prevent misuse. In this study, we introduce the concept of cross-lingual consistency in text watermarking, which assesses the ability of text watermarks to maintain their effectiveness after being translated into other languages. Preliminary empirical results from two LLMs and three watermarking methods reveal that current text watermarking technologies lack consistency when texts are translated into various languages. Based on this observation, we propose a Cross-lingual Watermark Removal Attack (CWRA) to bypass watermarking by first obtaining a response from an LLM in a pivot language, which is then translated into the target language. CWRA can effectively remove watermarks, decreasing the AUCs to a random-guessing level without performance loss. Furthermore, we analyze two key factors that contribute to the cross-lingual consistency in text watermarking and propose X-SIR as a defense method against CWRA.", }

Text watermarking technology aims to tag and identify content produced by large language models (LLMs) to prevent misuse. In this study, we introduce the concept of cross-lingual consistency in text watermarking, which assesses the ability of text watermarks to maintain their effectiveness after being translated into other languages. Preliminary empirical results from two LLMs and three watermarking methods reveal that current text watermarking technologies lack consistency when texts are translated into various languages. Based on this observation, we propose a Cross-lingual Watermark Removal Attack (CWRA) to bypass watermarking by first obtaining a response from an LLM in a pivot language, which is then translated into the target language. CWRA can effectively remove watermarks, decreasing the AUCs to a random-guessing level without performance loss. Furthermore, we analyze two key factors that contribute to the cross-lingual consistency in text watermarking and propose X-SIR as a defense method against CWRA.

Can Watermarks Survive Translation? On the Cross-lingual Consistency of Text Watermark for Large Language Models

acl-long.226

Oral

2402.14007

https://aclanthology.org/2024.acl-long.226/

https://aclanthology.org/2024.acl-long.227.bib

@inproceedings{chaszczewicz-etal-2024-multi, title = "Multi-Level Feedback Generation with Large Language Models for Empowering Novice Peer Counselors", author = "Chaszczewicz, Alicja and Shah, Raj and Louie, Ryan and Arnow, Bruce and Kraut, Robert and Yang, Diyi", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.227", pages = "4130--4161", abstract = "Realistic practice and tailored feedback are key processes for training peer counselors with clinical skills. However, existing mechanisms of providing feedback largely rely on human supervision. Peer counselors often lack mechanisms to receive detailed feedback from experienced mentors, making it difficult for them to support the large number of people with mental health issues who use peer counseling. Our work aims to leverage large language models to provide contextualized and multi-level feedback to empower peer counselors, especially novices, at scale. To achieve this, we co-design with a group of senior psychotherapy supervisors to develop a multi-level feedback taxonomy, and then construct a publicly available dataset with comprehensive feedback annotations of 400 emotional support conversations. We further design a self-improvement method on top of large language models to enhance the automatic generation of feedback. Via qualitative and quantitative evaluation with domain experts, we demonstrate that our method minimizes the risk of potentially harmful and low-quality feedback generation which is desirable in such high-stakes scenarios.", }

Realistic practice and tailored feedback are key processes for training peer counselors with clinical skills. However, existing mechanisms of providing feedback largely rely on human supervision. Peer counselors often lack mechanisms to receive detailed feedback from experienced mentors, making it difficult for them to support the large number of people with mental health issues who use peer counseling. Our work aims to leverage large language models to provide contextualized and multi-level feedback to empower peer counselors, especially novices, at scale. To achieve this, we co-design with a group of senior psychotherapy supervisors to develop a multi-level feedback taxonomy, and then construct a publicly available dataset with comprehensive feedback annotations of 400 emotional support conversations. We further design a self-improvement method on top of large language models to enhance the automatic generation of feedback. Via qualitative and quantitative evaluation with domain experts, we demonstrate that our method minimizes the risk of potentially harmful and low-quality feedback generation which is desirable in such high-stakes scenarios.

Multi-Level Feedback Generation with Large Language Models for Empowering Novice Peer Counselors

acl-long.227

Poster

2403.15482

https://huggingface.co/papers/2403.15482

https://aclanthology.org/2024.acl-long.227/

https://aclanthology.org/2024.acl-long.228.bib

@inproceedings{shankar-etal-2024-context, title = "In-context Mixing ({ICM}): Code-mixed Prompts for Multilingual {LLM}s", author = "Shankar, Bhavani and Jyothi, Preethi and Bhattacharyya, Pushpak", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.228", pages = "4162--4176", abstract = "We introduce a simple and effective prompting technique called in-context mixing (ICM) for effective in-context learning (ICL) with multilingual large language models (MLLMs). With ICM, we modify the few-shot examples within ICL prompts to be intra-sententially code-mixed by randomly swapping content words in the target languages with their English translations. We observe that ICM prompts yield superior performance in NLP tasks such as disfluency correction, grammar error correction and text simplification that demand a close correspondence between the input and output sequences. Significant improvements are observed mainly for low-resource languages that are under-represented during the pretraining and finetuning of MLLMs. We present an extensive set of experiments to analyze when ICM is effective and what design choices contribute towards its effectiveness. ICM works consistently and significantly better than other prompting techniques across models of varying capacity such as mT0-XXL, BloomZ and GPT-4.", }

We introduce a simple and effective prompting technique called in-context mixing (ICM) for effective in-context learning (ICL) with multilingual large language models (MLLMs). With ICM, we modify the few-shot examples within ICL prompts to be intra-sententially code-mixed by randomly swapping content words in the target languages with their English translations. We observe that ICM prompts yield superior performance in NLP tasks such as disfluency correction, grammar error correction and text simplification that demand a close correspondence between the input and output sequences. Significant improvements are observed mainly for low-resource languages that are under-represented during the pretraining and finetuning of MLLMs. We present an extensive set of experiments to analyze when ICM is effective and what design choices contribute towards its effectiveness. ICM works consistently and significantly better than other prompting techniques across models of varying capacity such as mT0-XXL, BloomZ and GPT-4.

In-context Mixing (ICM): Code-mixed Prompts for Multilingual LLMs

acl-long.228

Poster

https://aclanthology.org/2024.acl-long.228/

https://aclanthology.org/2024.acl-long.229.bib

@inproceedings{zhang-etal-2024-respond, title = "Respond in my Language: Mitigating Language Inconsistency in Response Generation based on Large Language Models", author = "Zhang, Liang and Jin, Qin and Huang, Haoyang and Zhang, Dongdong and Wei, Furu", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.229", pages = "4177--4192", abstract = "Large Language Models (LLMs) show strong instruction understanding ability across multiple languages. However, they are easily biased towards English in instruction tuning, and generate English responses even given non-English instructions. In this paper, we investigate the language inconsistent generation problem in monolingual instruction tuning. We find that instruction tuning in English increases the models{'} preference for English responses. It attaches higher probabilities to English responses than to responses in the same language as the instruction. Based on the findings, we alleviate the language inconsistent generation problem by counteracting the model preference for English responses in both the training and inference stages. Specifically, we propose Pseudo-Inconsistent Penalization (PIP) which prevents the model from generating English responses when given non-English language prompts during training, and Prior Enhanced Decoding (PED) which improves the language-consistent prior by leveraging the untuned base language model. Experimental results show that our two methods significantly improve the language consistency of the model without requiring any multilingual data. Our code, data, and models will be released.", }

Large Language Models (LLMs) show strong instruction understanding ability across multiple languages. However, they are easily biased towards English in instruction tuning, and generate English responses even given non-English instructions. In this paper, we investigate the language inconsistent generation problem in monolingual instruction tuning. We find that instruction tuning in English increases the models{'} preference for English responses. It attaches higher probabilities to English responses than to responses in the same language as the instruction. Based on the findings, we alleviate the language inconsistent generation problem by counteracting the model preference for English responses in both the training and inference stages. Specifically, we propose Pseudo-Inconsistent Penalization (PIP) which prevents the model from generating English responses when given non-English language prompts during training, and Prior Enhanced Decoding (PED) which improves the language-consistent prior by leveraging the untuned base language model. Experimental results show that our two methods significantly improve the language consistency of the model without requiring any multilingual data. Our code, data, and models will be released.

Respond in my Language: Mitigating Language Inconsistency in Response Generation based on Large Language Models

acl-long.229

Poster

https://aclanthology.org/2024.acl-long.229/

https://aclanthology.org/2024.acl-long.230.bib

@inproceedings{huang-etal-2024-transferable, title = "Transferable Embedding Inversion Attack: Uncovering Privacy Risks in Text Embeddings without Model Queries", author = "Huang, Yu-Hsiang and Tsai, Yuche and Hsiao, Hsiang and Lin, Hong-Yi and Lin, Shou-De", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.230", pages = "4193--4205", abstract = "This study investigates the privacy risks associated with text embeddings, focusing on the scenario where attackers cannot access the original embedding model. Contrary to previous research requiring direct model access, we explore a more realistic threat model by developing a transfer attack method. This approach uses a surrogate model to mimic the victim model{'}s behavior, allowing the attacker to infer sensitive information from text embeddings without direct access. Our experiments across various embedding models and a clinical dataset demonstrate that our transfer attack significantly outperforms traditional methods, revealing the potential privacy vulnerabilities in embedding technologies and emphasizing the need for enhanced security measures.", }

This study investigates the privacy risks associated with text embeddings, focusing on the scenario where attackers cannot access the original embedding model. Contrary to previous research requiring direct model access, we explore a more realistic threat model by developing a transfer attack method. This approach uses a surrogate model to mimic the victim model{'}s behavior, allowing the attacker to infer sensitive information from text embeddings without direct access. Our experiments across various embedding models and a clinical dataset demonstrate that our transfer attack significantly outperforms traditional methods, revealing the potential privacy vulnerabilities in embedding technologies and emphasizing the need for enhanced security measures.

Transferable Embedding Inversion Attack: Uncovering Privacy Risks in Text Embeddings without Model Queries

acl-long.230

Poster

2406.10280

https://aclanthology.org/2024.acl-long.230/

https://aclanthology.org/2024.acl-long.231.bib

@inproceedings{liao-etal-2024-enhancing, title = "Enhancing Reinforcement Learning with Label-Sensitive Reward for Natural Language Understanding", author = "Liao, Kuo and Li, Shuang and Zhao, Meng and Liu, Liqun and Xue, Mengge and Hu, Zhenyu and Han, Honglin and Yin, Chengguo", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.231", pages = "4206--4220", abstract = "Recent strides in large language models (LLMs) have yielded remarkable performance, leveraging reinforcement learning from human feedback (RLHF) to significantly enhance generation and alignment capabilities. However, RLHF encounters numerous challenges, including the objective mismatch issue, leading to suboptimal performance in Natural Language Understanding (NLU) tasks.To address this limitation, we propose a novel Reinforcement Learning framework enhanced with Label-sensitive Reward (RLLR) to amplify the performance of LLMs in NLU tasks. By incorporating label-sensitive pairs into reinforcement learning, our method aims to adeptly capture nuanced label-sensitive semantic features during RL, thereby enhancing natural language understanding.Experiments conducted on five diverse foundation models across eight tasks showcase promising results. In comparison to Supervised Fine-tuning models (SFT), RLLR demonstrates an average performance improvement of 1.54{\%}. Compared with RLHF models, the improvement averages at 0.69{\%}. These results reveal the effectiveness of our method for LLMs in NLU tasks.", }

Recent strides in large language models (LLMs) have yielded remarkable performance, leveraging reinforcement learning from human feedback (RLHF) to significantly enhance generation and alignment capabilities. However, RLHF encounters numerous challenges, including the objective mismatch issue, leading to suboptimal performance in Natural Language Understanding (NLU) tasks.To address this limitation, we propose a novel Reinforcement Learning framework enhanced with Label-sensitive Reward (RLLR) to amplify the performance of LLMs in NLU tasks. By incorporating label-sensitive pairs into reinforcement learning, our method aims to adeptly capture nuanced label-sensitive semantic features during RL, thereby enhancing natural language understanding.Experiments conducted on five diverse foundation models across eight tasks showcase promising results. In comparison to Supervised Fine-tuning models (SFT), RLLR demonstrates an average performance improvement of 1.54{\%}. Compared with RLHF models, the improvement averages at 0.69{\%}. These results reveal the effectiveness of our method for LLMs in NLU tasks.

Enhancing Reinforcement Learning with Label-Sensitive Reward for Natural Language Understanding

acl-long.231

Poster

2405.19763

https://aclanthology.org/2024.acl-long.231/

https://aclanthology.org/2024.acl-long.232.bib

@inproceedings{ying-etal-2024-intuitive, title = "Intuitive or Dependent? Investigating {LLM}s{'} Behavior Style to Conflicting Prompts", author = "Ying, Jiahao and Cao, Yixin and Xiong, Kai and Cui, Long and He, Yidong and Liu, Yongbin", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.232", pages = "4221--4246", abstract = "This study investigates the behaviors of Large Language Models (LLMs) when faced with conflicting prompts versus their internal memory. This will not only help to understand LLMs{'} decision mechanism but also benefit real-world applications, such as retrieval-augmented generation (RAG).Drawing on cognitive theory, we target the first scenario of decision-making styles where there is no superiority in the conflict and categorize LLMs{'} preference into dependent, intuitive, and rational/irrational styles.Another scenario of factual robustness considers the correctness of prompt and memory in knowledge-intensive tasks, which can also distinguish if LLMs behave rationally or irrationally in the first scenario.To quantify them, we establish a complete benchmarking framework including a dataset, a robustness evaluation pipeline, and corresponding metrics. Extensive experiments with seven LLMs reveal their varying behaviors. And, with role play intervention, we can change the styles, but different models present distinct adaptivity and upper-bound. One of our key takeaways is to optimize models or the prompts according to the identified style. For instance, RAG models with high role play adaptability may dynamically adjust the interventions according to the quality of retrieval results {---} being dependent to better leverage informative context; and, being intuitive when external prompt is noisy.", }

This study investigates the behaviors of Large Language Models (LLMs) when faced with conflicting prompts versus their internal memory. This will not only help to understand LLMs{'} decision mechanism but also benefit real-world applications, such as retrieval-augmented generation (RAG).Drawing on cognitive theory, we target the first scenario of decision-making styles where there is no superiority in the conflict and categorize LLMs{'} preference into dependent, intuitive, and rational/irrational styles.Another scenario of factual robustness considers the correctness of prompt and memory in knowledge-intensive tasks, which can also distinguish if LLMs behave rationally or irrationally in the first scenario.To quantify them, we establish a complete benchmarking framework including a dataset, a robustness evaluation pipeline, and corresponding metrics. Extensive experiments with seven LLMs reveal their varying behaviors. And, with role play intervention, we can change the styles, but different models present distinct adaptivity and upper-bound. One of our key takeaways is to optimize models or the prompts according to the identified style. For instance, RAG models with high role play adaptability may dynamically adjust the interventions according to the quality of retrieval results {---} being dependent to better leverage informative context; and, being intuitive when external prompt is noisy.

Intuitive or Dependent? Investigating LLMs' Behavior Style to Conflicting Prompts

acl-long.232

Poster

2309.17415

https://aclanthology.org/2024.acl-long.232/

https://aclanthology.org/2024.acl-long.233.bib

@inproceedings{zhu-etal-2024-coca, title = "{C}o{CA}: Fusing Position Embedding with Collinear Constrained Attention in Transformers for Long Context Window Extending", author = "Zhu, Shiyi and Ye, Jing and Jiang, Wei and Xue, Siqiao and Zhang, Qi and Wu, Yifan and Li, Jianguo", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.233", pages = "4247--4262", abstract = "Self-attention and position embedding are two crucial modules in transformer-based Large Language Models (LLMs). However, the potential relationship between them is far from well studied, especially for long context window extending. In fact, anomalous behaviors that hinder long context extrapolation exist between Rotary Position Embedding (RoPE) and vanilla self-attention.Incorrect initial angles between $Q$ and $K$ can cause misestimation in modeling rotary position embedding of the closest tokens.To address this issue, we propose $\textbf{Co}$llinear $\textbf{C}$onstrained $\textbf{A}$ttention mechanism, namely CoCA. Specifically, we enforce a collinear constraint between $Q$ and $K$ to seamlessly integrate RoPE and self-attention.While only adding minimal computational and spatial complexity, this integration significantly enhances long context window extrapolation ability. We provide an optimized implementation, making it a drop-in replacement for any existing transformer-based models.Extensive experiments demonstrate that CoCA excels in extending context windows. A CoCA-based GPT model, trained with a context length of 512, can extend the context window up to 32K (60$\times$) without any fine-tuning.Additionally, incorporating CoCA into LLaMA-7B achieves extrapolation up to 32K within a training length of only 2K.Our code is publicly available at: https://github.com/codefuse-ai/Collinear-Constrained-Attention", }

Self-attention and position embedding are two crucial modules in transformer-based Large Language Models (LLMs). However, the potential relationship between them is far from well studied, especially for long context window extending. In fact, anomalous behaviors that hinder long context extrapolation exist between Rotary Position Embedding (RoPE) and vanilla self-attention.Incorrect initial angles between $Q$ and $K$ can cause misestimation in modeling rotary position embedding of the closest tokens.To address this issue, we propose $\textbf{Co}$llinear $\textbf{C}$onstrained $\textbf{A}$ttention mechanism, namely CoCA. Specifically, we enforce a collinear constraint between $Q$ and $K$ to seamlessly integrate RoPE and self-attention.While only adding minimal computational and spatial complexity, this integration significantly enhances long context window extrapolation ability. We provide an optimized implementation, making it a drop-in replacement for any existing transformer-based models.Extensive experiments demonstrate that CoCA excels in extending context windows. A CoCA-based GPT model, trained with a context length of 512, can extend the context window up to 32K (60$\times$) without any fine-tuning.Additionally, incorporating CoCA into LLaMA-7B achieves extrapolation up to 32K within a training length of only 2K.Our code is publicly available at: https://github.com/codefuse-ai/Collinear-Constrained-Attention

CoCA: Fusing Position Embedding with Collinear Constrained Attention in Transformers for Long Context Window Extending

acl-long.233

Poster

2309.08646

https://huggingface.co/papers/2309.08646

https://aclanthology.org/2024.acl-long.233/

https://aclanthology.org/2024.acl-long.234.bib

@inproceedings{trienes-etal-2024-infolossqa, title = "{I}nfo{L}oss{QA}: Characterizing and Recovering Information Loss in Text Simplification", author = {Trienes, Jan and Joseph, Sebastian and Schl{\"o}tterer, J{\"o}rg and Seifert, Christin and Lo, Kyle and Xu, Wei and Wallace, Byron and Li, Junyi Jessy}, editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.234", pages = "4263--4294", abstract = "Text simplification aims to make technical texts more accessible to laypeople but often results in deletion of information and vagueness. This work proposes InfoLossQA, a framework to characterize and recover simplification-induced information loss in form of question-and-answer (QA) pairs. Building on the theory of Questions Under Discussion, the QA pairs are designed to help readers deepen their knowledge of a text. First, we collect a dataset of 1,000 linguist-curated QA pairs derived from 104 LLM simplifications of English medical study abstracts. Our analyses of this data reveal that information loss occurs frequently, and that the QA pairs give a high-level overview of what information was lost. Second, we devise two methods for this task: end-to-end prompting of open-source and commercial language models, and a natural language inference pipeline. With a novel evaluation framework considering the correctness of QA pairs and their linguistic suitability, our expert evaluation reveals that models struggle to reliably identify information loss and applying similar standards as humans at what constitutes information loss.", }

Text simplification aims to make technical texts more accessible to laypeople but often results in deletion of information and vagueness. This work proposes InfoLossQA, a framework to characterize and recover simplification-induced information loss in form of question-and-answer (QA) pairs. Building on the theory of Questions Under Discussion, the QA pairs are designed to help readers deepen their knowledge of a text. First, we collect a dataset of 1,000 linguist-curated QA pairs derived from 104 LLM simplifications of English medical study abstracts. Our analyses of this data reveal that information loss occurs frequently, and that the QA pairs give a high-level overview of what information was lost. Second, we devise two methods for this task: end-to-end prompting of open-source and commercial language models, and a natural language inference pipeline. With a novel evaluation framework considering the correctness of QA pairs and their linguistic suitability, our expert evaluation reveals that models struggle to reliably identify information loss and applying similar standards as humans at what constitutes information loss.

InfoLossQA: Characterizing and Recovering Information Loss in Text Simplification

acl-long.234

Poster

2401.16475

https://aclanthology.org/2024.acl-long.234/

https://aclanthology.org/2024.acl-long.235.bib

@inproceedings{zhang-etal-2024-cogenesis, title = "{C}o{G}enesis: A Framework Collaborating Large and Small Language Models for Secure Context-Aware Instruction Following", author = "Zhang, Kaiyan and Wang, Jianyu and Hua, Ermo and Qi, Biqing and Ding, Ning and Zhou, Bowen", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.235", pages = "4295--4312", abstract = "With the advancement of language models (LMs), their exposure to private data is increasingly inevitable, and their deployment (especially for smaller ones) on personal devices, such as PCs and smartphones, has become a prevailing trend. In contexts laden with user information, enabling models to both safeguard user privacy and execute commands efficiently emerges as an essential research imperative. In this paper, we propose CoGenesis, a collaborative generation framework integrating large (hosted on cloud infrastructure) and small models (deployed on local devices) to address privacy concerns logically. Initially, we design a pipeline to create personalized writing instruction datasets enriched with extensive context details as the testbed of this research issue. Subsequently, we introduce two variants of CoGenesis based on sketch and logits respectively. Our experimental findings, based on our synthesized dataset and two additional open-source datasets, indicate that: 1) Large-scale models perform well when provided with user context but struggle in the absence of such context. 2) While specialized smaller models fine-tuned on the synthetic dataset show promise, they still lag behind their larger counterparts. 3) Our CoGenesis framework, utilizing mixed-scale models, showcases competitive performance, providing a feasible solution to privacy issues.", }

With the advancement of language models (LMs), their exposure to private data is increasingly inevitable, and their deployment (especially for smaller ones) on personal devices, such as PCs and smartphones, has become a prevailing trend. In contexts laden with user information, enabling models to both safeguard user privacy and execute commands efficiently emerges as an essential research imperative. In this paper, we propose CoGenesis, a collaborative generation framework integrating large (hosted on cloud infrastructure) and small models (deployed on local devices) to address privacy concerns logically. Initially, we design a pipeline to create personalized writing instruction datasets enriched with extensive context details as the testbed of this research issue. Subsequently, we introduce two variants of CoGenesis based on sketch and logits respectively. Our experimental findings, based on our synthesized dataset and two additional open-source datasets, indicate that: 1) Large-scale models perform well when provided with user context but struggle in the absence of such context. 2) While specialized smaller models fine-tuned on the synthetic dataset show promise, they still lag behind their larger counterparts. 3) Our CoGenesis framework, utilizing mixed-scale models, showcases competitive performance, providing a feasible solution to privacy issues.

CoGenesis: A Framework Collaborating Large and Small Language Models for Secure Context-Aware Instruction Following

acl-long.235

Poster

2403.03129

https://huggingface.co/papers/2403.03129

https://aclanthology.org/2024.acl-long.235/

https://aclanthology.org/2024.acl-long.236.bib

@inproceedings{wang-etal-2024-dapr, title = "{DAPR}: A Benchmark on Document-Aware Passage Retrieval", author = "Wang, Kexin and Reimers, Nils and Gurevych, Iryna", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.236", pages = "4313--4330", abstract = "The work of neural retrieval so far focuses on ranking short texts and is challenged with long documents. There are many cases where the users want to find a relevant passage within a long document from a huge corpus, e.g. Wikipedia articles, research papers, etc. We propose and name this task \textit{Document-Aware Passage Retrieval} (DAPR). While analyzing the errors of the State-of-The-Art (SoTA) passage retrievers, we find the major errors (53.5{\%}) are due to missing document context. This drives us to build a benchmark for this task including multiple datasets from heterogeneous domains. In the experiments, we extend the SoTA passage retrievers with document context via (1) hybrid retrieval with BM25 and (2) contextualized passage representations, which inform the passage representation with document context. We find despite that hybrid retrieval performs the strongest on the mixture of the easy and the hard queries, it completely fails on the hard queries that require document-context understanding. On the other hand, contextualized passage representations (e.g. prepending document titles) achieve good improvement on these hard queries, but overall they also perform rather poorly. Our created benchmark enables future research on developing and comparing retrieval systems for the new task. The code and the data are available.", }

The work of neural retrieval so far focuses on ranking short texts and is challenged with long documents. There are many cases where the users want to find a relevant passage within a long document from a huge corpus, e.g. Wikipedia articles, research papers, etc. We propose and name this task \textit{Document-Aware Passage Retrieval} (DAPR). While analyzing the errors of the State-of-The-Art (SoTA) passage retrievers, we find the major errors (53.5{\%}) are due to missing document context. This drives us to build a benchmark for this task including multiple datasets from heterogeneous domains. In the experiments, we extend the SoTA passage retrievers with document context via (1) hybrid retrieval with BM25 and (2) contextualized passage representations, which inform the passage representation with document context. We find despite that hybrid retrieval performs the strongest on the mixture of the easy and the hard queries, it completely fails on the hard queries that require document-context understanding. On the other hand, contextualized passage representations (e.g. prepending document titles) achieve good improvement on these hard queries, but overall they also perform rather poorly. Our created benchmark enables future research on developing and comparing retrieval systems for the new task. The code and the data are available.

DAPR: A Benchmark on Document-Aware Passage Retrieval

acl-long.236

Poster

2305.13915

https://huggingface.co/papers/2305.13915

https://aclanthology.org/2024.acl-long.236/

https://aclanthology.org/2024.acl-long.237.bib

@inproceedings{xue-etal-2024-strengthened, title = "Strengthened Symbol Binding Makes Large Language Models Reliable Multiple-Choice Selectors", author = "Xue, Mengge and Hu, Zhenyu and Liu, Liqun and Liao, Kuo and Li, Shuang and Han, Honglin and Zhao, Meng and Yin, Chengguo", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.237", pages = "4331--4344", abstract = "Multiple-Choice Questions (MCQs) constitute a critical area of research in the study of Large Language Models (LLMs). Previous works have investigated the selection bias problem in MCQs within few-shot scenarios, in which the LLM{'}s performance may be influenced by the presentation of answer choices, leaving the selection bias during Supervised Fine-Tuning (SFT) unexplored. In this paper, we reveal that selection bias persists in the SFT phase , primarily due to the LLM{'}s inadequate Multiple Choice Symbol Binding (MCSB) ability. This limitation implies that the model struggles to associate the answer options with their corresponding symbols (e.g., A/B/C/D) effectively. To enhance the model{'}s MCSB capability, we first incorporate option contents into the loss function and subsequently adjust the weights of the option symbols and contents, guiding the model to understand the option content of the current symbol. Based on this, we introduce an efficient SFT algorithm for MCQs, termed Point-wise Intelligent Feedback (PIF). PIF constructs negative instances by randomly combin- ing the incorrect option contents with all candidate symbols, and proposes a point-wise loss to provide feedback on these negative samples into LLMs. Our experimental results demonstrate that PIF significantly reduces the model{'}s selection bias by improving its MCSB capability. Remarkably, PIF exhibits a substantial enhancement in the accuracy for MCQs.", }

Multiple-Choice Questions (MCQs) constitute a critical area of research in the study of Large Language Models (LLMs). Previous works have investigated the selection bias problem in MCQs within few-shot scenarios, in which the LLM{'}s performance may be influenced by the presentation of answer choices, leaving the selection bias during Supervised Fine-Tuning (SFT) unexplored. In this paper, we reveal that selection bias persists in the SFT phase , primarily due to the LLM{'}s inadequate Multiple Choice Symbol Binding (MCSB) ability. This limitation implies that the model struggles to associate the answer options with their corresponding symbols (e.g., A/B/C/D) effectively. To enhance the model{'}s MCSB capability, we first incorporate option contents into the loss function and subsequently adjust the weights of the option symbols and contents, guiding the model to understand the option content of the current symbol. Based on this, we introduce an efficient SFT algorithm for MCQs, termed Point-wise Intelligent Feedback (PIF). PIF constructs negative instances by randomly combin- ing the incorrect option contents with all candidate symbols, and proposes a point-wise loss to provide feedback on these negative samples into LLMs. Our experimental results demonstrate that PIF significantly reduces the model{'}s selection bias by improving its MCSB capability. Remarkably, PIF exhibits a substantial enhancement in the accuracy for MCQs.

Strengthened Symbol Binding Makes Large Language Models Reliable Multiple-Choice Selectors

acl-long.237

Poster

2406.01026

https://aclanthology.org/2024.acl-long.237/

https://aclanthology.org/2024.acl-long.238.bib

@inproceedings{chen-etal-2024-sac, title = "{SAC}-{KG}: Exploiting Large Language Models as Skilled Automatic Constructors for Domain Knowledge Graph", author = "Chen, Hanzhu and Shen, Xu and Lv, Qitan and Wang, Jie and Ni, Xiaoqi and Ye, Jieping", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.238", pages = "4345--4360", abstract = "Knowledge graphs (KGs) play a pivotal role in knowledge-intensive tasks across specialized domains, where the acquisition of precise and dependable knowledge is crucial. However, existing KG construction methods heavily rely on human intervention to attain qualified KGs, which severely hinders the practical applicability in real-world scenarios. To address this challenge, we propose a general KG construction framework, named **SAC-KG**, to exploit large language models (LLMs) as **S**killed **A**utomatic **C**onstructors for domain **K**nowledge **G**raph. SAC-KG effectively involves LLMs as domain experts to generate specialized and precise multi-level KGs. Specifically, SAC-KG consists of three components: Generator, Verifier, and Pruner. For a given entity, Generator produces its relations and tails from raw domain corpora, to construct a specialized single-level KG. Verifier and Pruner then work together to ensure precision by correcting generation errors and determining whether newly produced tails require further iteration for the next-level KG. Experiments demonstrate that SAC-KG automatically constructs a domain KG at the scale of over one million nodes and achieves a precision of 89.32{\%}, leading to a superior performance with over 20{\%} increase in precision rate compared to existing state-of-the-art methods for the KG construction task.", }

Knowledge graphs (KGs) play a pivotal role in knowledge-intensive tasks across specialized domains, where the acquisition of precise and dependable knowledge is crucial. However, existing KG construction methods heavily rely on human intervention to attain qualified KGs, which severely hinders the practical applicability in real-world scenarios. To address this challenge, we propose a general KG construction framework, named **SAC-KG**, to exploit large language models (LLMs) as **S**killed **A**utomatic **C**onstructors for domain **K**nowledge **G**raph. SAC-KG effectively involves LLMs as domain experts to generate specialized and precise multi-level KGs. Specifically, SAC-KG consists of three components: Generator, Verifier, and Pruner. For a given entity, Generator produces its relations and tails from raw domain corpora, to construct a specialized single-level KG. Verifier and Pruner then work together to ensure precision by correcting generation errors and determining whether newly produced tails require further iteration for the next-level KG. Experiments demonstrate that SAC-KG automatically constructs a domain KG at the scale of over one million nodes and achieves a precision of 89.32{\%}, leading to a superior performance with over 20{\%} increase in precision rate compared to existing state-of-the-art methods for the KG construction task.

SAC-KG: Exploiting Large Language Models as Skilled Automatic Constructors for Domain Knowledge Graph

acl-long.238

Poster

https://aclanthology.org/2024.acl-long.238/

https://aclanthology.org/2024.acl-long.239.bib

@inproceedings{yang-etal-2024-uncertainty-guided, title = "Uncertainty-Guided Modal Rebalance for Hateful Memes Detection", author = "Yang, Chuanpeng and Liu, Yaxin and Zhu, Fuqing and Han, Jizhong and Hu, Songlin", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.239", pages = "4361--4371", abstract = "Hateful memes detection is a challenging multimodal understanding task that requires comprehensive learning of vision, language, and cross-modal interactions. Previous research has focused on developing effective fusion strategies for integrating hate information from different modalities. However, these methods excessively rely on cross-modal fusion features, ignoring the modality uncertainty caused by the contribution degree of each modality to hate sentiment and the modality imbalance caused by the dominant modality suppressing the optimization of another modality. To this end, this paper proposes an Uncertainty-guided Modal Rebalance (UMR) framework for hateful memes detection. The uncertainty of each meme is explicitly formulated by designing stochastic representation drawn from a Gaussian distribution for aggregating cross-modal features with unimodal features adaptively. The modality imbalance is alleviated by improving cosine loss from the perspectives of inter-modal feature and weight vectors constraints. In this way, the suppressed unimodal representation ability in multimodal models would be unleashed, while the learning of modality contribution would be further promoted. Extensive experimental results demonstrate that the proposed UMR produces the state-of-the-art performance on four widely-used datasets.", }

Hateful memes detection is a challenging multimodal understanding task that requires comprehensive learning of vision, language, and cross-modal interactions. Previous research has focused on developing effective fusion strategies for integrating hate information from different modalities. However, these methods excessively rely on cross-modal fusion features, ignoring the modality uncertainty caused by the contribution degree of each modality to hate sentiment and the modality imbalance caused by the dominant modality suppressing the optimization of another modality. To this end, this paper proposes an Uncertainty-guided Modal Rebalance (UMR) framework for hateful memes detection. The uncertainty of each meme is explicitly formulated by designing stochastic representation drawn from a Gaussian distribution for aggregating cross-modal features with unimodal features adaptively. The modality imbalance is alleviated by improving cosine loss from the perspectives of inter-modal feature and weight vectors constraints. In this way, the suppressed unimodal representation ability in multimodal models would be unleashed, while the learning of modality contribution would be further promoted. Extensive experimental results demonstrate that the proposed UMR produces the state-of-the-art performance on four widely-used datasets.

Uncertainty-Guided Modal Rebalance for Hateful Memes Detection

acl-long.239

Poster

https://aclanthology.org/2024.acl-long.239/

https://aclanthology.org/2024.acl-long.240.bib

@inproceedings{glockner-etal-2024-missci, title = "Missci: Reconstructing Fallacies in Misrepresented Science", author = "Glockner, Max and Hou, Yufang and Nakov, Preslav and Gurevych, Iryna", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.240", pages = "4372--4405", abstract = "Health-related misinformation on social networks can lead to poor decision-making and real-world dangers. Such misinformation often misrepresents scientific publications and cites them as {``}proof{''} to gain perceived credibility. To effectively counter such claims automatically, a system must explain how the claim was falsely derived from the cited publication. Current methods for automated fact-checking or fallacy detection neglect to assess the (mis)used evidence in relation to misinformation claims, which is required to detect the mismatch between them. To address this gap, we introduce Missci, a novel argumentation theoretical model for fallacious reasoning together with a new dataset for real-world misinformation detection that misrepresents biomedical publications. Unlike previous fallacy detection datasets, Missci (i) focuses on implicit fallacies between the relevant content of the cited publication and the inaccurate claim, and (ii) requires models to verbalize the fallacious reasoning in addition to classifying it. We present Missci as a dataset to test the critical reasoning abilities of large language models (LLMs), that are required to reconstruct real-world fallacious arguments, in a zero-shot setting. We evaluate two representative LLMs and the impact of different levels of detail about the fallacy classes provided to the LLM via prompts. Our experiments and human evaluation show promising results for GPT 4, while also demonstrating the difficulty of this task.", }

Health-related misinformation on social networks can lead to poor decision-making and real-world dangers. Such misinformation often misrepresents scientific publications and cites them as {``}proof{''} to gain perceived credibility. To effectively counter such claims automatically, a system must explain how the claim was falsely derived from the cited publication. Current methods for automated fact-checking or fallacy detection neglect to assess the (mis)used evidence in relation to misinformation claims, which is required to detect the mismatch between them. To address this gap, we introduce Missci, a novel argumentation theoretical model for fallacious reasoning together with a new dataset for real-world misinformation detection that misrepresents biomedical publications. Unlike previous fallacy detection datasets, Missci (i) focuses on implicit fallacies between the relevant content of the cited publication and the inaccurate claim, and (ii) requires models to verbalize the fallacious reasoning in addition to classifying it. We present Missci as a dataset to test the critical reasoning abilities of large language models (LLMs), that are required to reconstruct real-world fallacious arguments, in a zero-shot setting. We evaluate two representative LLMs and the impact of different levels of detail about the fallacy classes provided to the LLM via prompts. Our experiments and human evaluation show promising results for GPT 4, while also demonstrating the difficulty of this task.

Missci: Reconstructing Fallacies in Misrepresented Science

acl-long.240

Poster

2406.03181

https://aclanthology.org/2024.acl-long.240/

https://aclanthology.org/2024.acl-long.241.bib

@inproceedings{reich-schultz-2024-uncovering, title = "Uncovering the Full Potential of Visual Grounding Methods in {VQA}", author = "Reich, Daniel and Schultz, Tanja", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.241", pages = "4406--4419", abstract = "Visual Grounding (VG) methods in Visual Question Answering (VQA) attempt to improve VQA performance by strengthening a model{'}s reliance on question-relevant visual information. The presence of such relevant information in the visual input is typically assumed in training and testing. This assumption, however, is inherently flawed when dealing with imperfect image representations common in large-scale VQA, where the information carried by visual features frequently deviates from expected ground-truth contents. As a result, training and testing of VG-methods is performed with largely inaccurate data, which obstructs proper assessment of their potential benefits.In this study, we demonstrate that current evaluation schemes for VG-methods are problematic due to the flawed assumption of availability of relevant visual information. Our experiments show that these methods can be much more effective when evaluation conditions are corrected. Code is provided.", }

Visual Grounding (VG) methods in Visual Question Answering (VQA) attempt to improve VQA performance by strengthening a model{'}s reliance on question-relevant visual information. The presence of such relevant information in the visual input is typically assumed in training and testing. This assumption, however, is inherently flawed when dealing with imperfect image representations common in large-scale VQA, where the information carried by visual features frequently deviates from expected ground-truth contents. As a result, training and testing of VG-methods is performed with largely inaccurate data, which obstructs proper assessment of their potential benefits.In this study, we demonstrate that current evaluation schemes for VG-methods are problematic due to the flawed assumption of availability of relevant visual information. Our experiments show that these methods can be much more effective when evaluation conditions are corrected. Code is provided.

Uncovering the Full Potential of Visual Grounding Methods in VQA

acl-long.241

Poster

2401.07803

https://aclanthology.org/2024.acl-long.241/

https://aclanthology.org/2024.acl-long.242.bib

@inproceedings{tan-etal-2024-small, title = "Small Models, Big Insights: Leveraging Slim Proxy Models To Decide When and What to Retrieve for {LLM}s", author = "Tan, Jiejun and Dou, Zhicheng and Zhu, Yutao and Guo, Peidong and Fang, Kun and Wen, Ji-Rong", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.242", pages = "4420--4436", abstract = "The integration of large language models (LLMs) and search engines represents a significant evolution in knowledge acquisition methodologies. However, determining the knowledge that an LLM already possesses and the knowledge that requires the help of a search engine remains an unresolved issue. Most existing methods solve this problem through the results of preliminary answers or reasoning done by the LLM itself, but this incurs excessively high computational costs. This paper introduces a novel collaborative approach, namely SlimPLM, that detects missing knowledge in LLMs with a slim proxy model, to enhance the LLM{'}s knowledge acquisition process. We employ a proxy model which has far fewer parameters, and take its answers as heuristic answers. Heuristic answers are then utilized to predict the knowledge required to answer the user question, as well as the known and unknown knowledge within the LLM. We only conduct retrieval for the missing knowledge in questions that the LLM does not know. Extensive experimental results on five datasets with two LLMs demonstrate a notable improvement in the end-to-end performance of LLMs in question-answering tasks, achieving or surpassing current state-of-the-art models with lower LLM inference costs.", }

The integration of large language models (LLMs) and search engines represents a significant evolution in knowledge acquisition methodologies. However, determining the knowledge that an LLM already possesses and the knowledge that requires the help of a search engine remains an unresolved issue. Most existing methods solve this problem through the results of preliminary answers or reasoning done by the LLM itself, but this incurs excessively high computational costs. This paper introduces a novel collaborative approach, namely SlimPLM, that detects missing knowledge in LLMs with a slim proxy model, to enhance the LLM{'}s knowledge acquisition process. We employ a proxy model which has far fewer parameters, and take its answers as heuristic answers. Heuristic answers are then utilized to predict the knowledge required to answer the user question, as well as the known and unknown knowledge within the LLM. We only conduct retrieval for the missing knowledge in questions that the LLM does not know. Extensive experimental results on five datasets with two LLMs demonstrate a notable improvement in the end-to-end performance of LLMs in question-answering tasks, achieving or surpassing current state-of-the-art models with lower LLM inference costs.

Small Models, Big Insights: Leveraging Slim Proxy Models To Decide When and What to Retrieve for LLMs

acl-long.242

Poster

2402.12052

https://huggingface.co/papers/2402.12052

https://aclanthology.org/2024.acl-long.242/

https://aclanthology.org/2024.acl-long.243.bib

@inproceedings{von-daniken-etal-2024-favi, title = "Favi-Score: A Measure for Favoritism in Automated Preference Ratings for Generative {AI} Evaluation", author = {Von D{\"a}niken, Pius and Deriu, Jan and Tuggener, Don and Cieliebak, Mark}, editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.243", pages = "4437--4454", abstract = "Generative AI systems have become ubiquitous for all kinds of modalities, which makes the issue of the evaluation of such models more pressing. One popular approach is preference ratings, where the generated outputs of different systems are shown to evaluators who choose their preferences. In recent years the field shifted towards the development of automated (trained) metrics to assess generated outputs, which can be used to create preference ratings automatically. In this work, we investigate the evaluation of the metrics themselves, which currently rely on measuring the correlation to human judgments or computing sign accuracy scores. These measures only assess how well the metric agrees with the human ratings. However, our research shows that this does not tell the whole story. Most metrics exhibit a disagreement with human system assessments which is often skewed in favor of particular text generation systems, exposing a degree of favoritism in automated metrics. This paper introduces a formal definition of favoritism in preference metrics, and derives the Favi-Score, which measures this phenomenon. In particular we show that favoritism is strongly related to errors in final system rankings. Thus, we propose that preference-based metrics ought to be evaluated on both sign accuracy scores and favoritism.", }

Generative AI systems have become ubiquitous for all kinds of modalities, which makes the issue of the evaluation of such models more pressing. One popular approach is preference ratings, where the generated outputs of different systems are shown to evaluators who choose their preferences. In recent years the field shifted towards the development of automated (trained) metrics to assess generated outputs, which can be used to create preference ratings automatically. In this work, we investigate the evaluation of the metrics themselves, which currently rely on measuring the correlation to human judgments or computing sign accuracy scores. These measures only assess how well the metric agrees with the human ratings. However, our research shows that this does not tell the whole story. Most metrics exhibit a disagreement with human system assessments which is often skewed in favor of particular text generation systems, exposing a degree of favoritism in automated metrics. This paper introduces a formal definition of favoritism in preference metrics, and derives the Favi-Score, which measures this phenomenon. In particular we show that favoritism is strongly related to errors in final system rankings. Thus, we propose that preference-based metrics ought to be evaluated on both sign accuracy scores and favoritism.

Favi-Score: A Measure for Favoritism in Automated Preference Ratings for Generative AI Evaluation

acl-long.243

Poster

2406.01131

https://aclanthology.org/2024.acl-long.243/

https://aclanthology.org/2024.acl-long.244.bib

@inproceedings{ziegenbein-etal-2024-llm, title = "{LLM}-based Rewriting of Inappropriate Argumentation using Reinforcement Learning from Machine Feedback", author = "Ziegenbein, Timon and Skitalinskaya, Gabriella and Bayat Makou, Alireza and Wachsmuth, Henning", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.244", pages = "4455--4476", abstract = "Ensuring that online discussions are civil and productive is a major challenge for social media platforms. Such platforms usually rely both on users and on automated detection tools to flag inappropriate arguments of other users, which moderators then review. However, this kind of post-hoc moderation is expensive and time-consuming, and moderators are often overwhelmed by the amount and severity of flagged content. Instead, a promising alternative is to prevent negative behavior during content creation. This paper studies how inappropriate language in arguments can be computationally mitigated. We propose a reinforcement learning-based rewriting approach that balances content preservation and appropriateness based on existing classifiers, prompting an instruction-finetuned large language model (LLM) as our initial policy. Unlike related style transfer tasks, rewriting inappropriate arguments allows deleting and adding content permanently. It is therefore tackled on document level rather than sentence level. We evaluate different weighting schemes for the reward function in both absolute and relative human assessment studies. Systematic experiments on non-parallel data provide evidence that our approach can mitigate the inappropriateness of arguments while largely preserving their content. It significantly outperforms competitive baselines, including few-shot learning, prompting, and humans.", }

Ensuring that online discussions are civil and productive is a major challenge for social media platforms. Such platforms usually rely both on users and on automated detection tools to flag inappropriate arguments of other users, which moderators then review. However, this kind of post-hoc moderation is expensive and time-consuming, and moderators are often overwhelmed by the amount and severity of flagged content. Instead, a promising alternative is to prevent negative behavior during content creation. This paper studies how inappropriate language in arguments can be computationally mitigated. We propose a reinforcement learning-based rewriting approach that balances content preservation and appropriateness based on existing classifiers, prompting an instruction-finetuned large language model (LLM) as our initial policy. Unlike related style transfer tasks, rewriting inappropriate arguments allows deleting and adding content permanently. It is therefore tackled on document level rather than sentence level. We evaluate different weighting schemes for the reward function in both absolute and relative human assessment studies. Systematic experiments on non-parallel data provide evidence that our approach can mitigate the inappropriateness of arguments while largely preserving their content. It significantly outperforms competitive baselines, including few-shot learning, prompting, and humans.

LLM-based Rewriting of Inappropriate Argumentation using Reinforcement Learning from Machine Feedback

acl-long.244

Poster

2406.03363

https://huggingface.co/papers/2406.03363

https://aclanthology.org/2024.acl-long.244/

https://aclanthology.org/2024.acl-long.245.bib

@inproceedings{plenz-frank-2024-graph, title = "Graph Language Models", author = "Plenz, Moritz and Frank, Anette", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.245", pages = "4477--4494", abstract = "While Language Models (LMs) are the workhorses of NLP, their interplay with structured knowledge graphs (KGs) is still actively researched. Current methods for encoding such graphs typically either (i) linearize them for embedding with LMs {--} which underutilize structural information, or (ii) use Graph Neural Networks (GNNs) to preserve the graph structure {--} but GNNs cannot represent text features as well as pretrained LMs. In our work we introduce a novel LM type, the Graph Language Model (GLM), that integrates the strengths of both approaches and mitigates their weaknesses. The GLM parameters are initialized from a pretrained LM to enhance understanding of individual graph concepts and triplets. Simultaneously, we design the GLM{'}s architecture to incorporate graph biases, thereby promoting effective knowledge distribution within the graph. This enables GLMs to process graphs, texts, and interleaved inputs of both. Empirical evaluations on relation classification tasks show that GLM embeddings surpass both LM- and GNN-based baselines in supervised and zero-shot setting, demonstrating their versatility.", }

While Language Models (LMs) are the workhorses of NLP, their interplay with structured knowledge graphs (KGs) is still actively researched. Current methods for encoding such graphs typically either (i) linearize them for embedding with LMs {--} which underutilize structural information, or (ii) use Graph Neural Networks (GNNs) to preserve the graph structure {--} but GNNs cannot represent text features as well as pretrained LMs. In our work we introduce a novel LM type, the Graph Language Model (GLM), that integrates the strengths of both approaches and mitigates their weaknesses. The GLM parameters are initialized from a pretrained LM to enhance understanding of individual graph concepts and triplets. Simultaneously, we design the GLM{'}s architecture to incorporate graph biases, thereby promoting effective knowledge distribution within the graph. This enables GLMs to process graphs, texts, and interleaved inputs of both. Empirical evaluations on relation classification tasks show that GLM embeddings surpass both LM- and GNN-based baselines in supervised and zero-shot setting, demonstrating their versatility.

Graph Language Models

acl-long.245

Oral

2401.07105

https://huggingface.co/papers/2401.07105

https://aclanthology.org/2024.acl-long.245/

https://aclanthology.org/2024.acl-long.246.bib

@inproceedings{periti-etal-2024-analyzing, title = "Analyzing Semantic Change through Lexical Replacements", author = "Periti, Francesco and Cassotti, Pierluigi and Dubossarsky, Haim and Tahmasebi, Nina", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.246", pages = "4495--4510", abstract = "Modern language models are capable of contextualizing words based on their surrounding context. However, this capability is often compromised due to semantic change that leads to words being used in new, unexpected contexts not encountered during pre-training. In this paper, we model semantic change by studying the effect of unexpected contexts introduced by lexical replacements. We propose a replacement schema where a target word is substituted with lexical replacements of varying relatedness, thus simulating different kinds of semantic change. Furthermore, we leverage the replacement schema as a basis for a novel interpretable model for semantic change. We are also the first to evaluate the use of LLaMa for semantic change detection.", }

Modern language models are capable of contextualizing words based on their surrounding context. However, this capability is often compromised due to semantic change that leads to words being used in new, unexpected contexts not encountered during pre-training. In this paper, we model semantic change by studying the effect of unexpected contexts introduced by lexical replacements. We propose a replacement schema where a target word is substituted with lexical replacements of varying relatedness, thus simulating different kinds of semantic change. Furthermore, we leverage the replacement schema as a basis for a novel interpretable model for semantic change. We are also the first to evaluate the use of LLaMa for semantic change detection.

Analyzing Semantic Change through Lexical Replacements

acl-long.246

Poster

2404.18570

https://aclanthology.org/2024.acl-long.246/

https://aclanthology.org/2024.acl-long.247.bib

@inproceedings{xu-etal-2024-exploiting, title = "Exploiting Intrinsic Multilateral Logical Rules for Weakly Supervised Natural Language Video Localization", author = "Xu, Zhe and Wei, Kun and Yang, Xu and Deng, Cheng", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.247", pages = "4511--4521", abstract = "Weakly supervised natural language video localization (WS-NLVL) aims to retrieve the moment corresponding to a language query in a video with only video-language pairs utilized during training. Despite great success, existing WS-NLVL methods seldomly consider the complex temporal relations enclosing the language query (e.g., between the language query and sub-queries decomposed from it or its synonymous query), yielding illogical predictions. In this paper, we propose a novel plug-and-play method, Intrinsic Multilateral Logical Rules, namely IMLR, to exploit intrinsic temporal relations and logical rules for WS-NLVL. Specifically, we formalize queries derived from the original language query as the nodes of a directed graph, i.e., intrinsic temporal relation graph (ITRG), and the temporal relations between them as the edges. Instead of directly prompting a pre-trained language model, a relation-guided prompting method is introduced to generate ITRG in a hierarchical manner. We customize four types of multilateral temporal logical rules (i.e., identity, inclusion, synchronization, and succession) from ITRG and utilize them to train our model. Experiments demonstrate the effectiveness and superiority of our method on the Charades-STA and ActivityNet Captions datasets.", }

Weakly supervised natural language video localization (WS-NLVL) aims to retrieve the moment corresponding to a language query in a video with only video-language pairs utilized during training. Despite great success, existing WS-NLVL methods seldomly consider the complex temporal relations enclosing the language query (e.g., between the language query and sub-queries decomposed from it or its synonymous query), yielding illogical predictions. In this paper, we propose a novel plug-and-play method, Intrinsic Multilateral Logical Rules, namely IMLR, to exploit intrinsic temporal relations and logical rules for WS-NLVL. Specifically, we formalize queries derived from the original language query as the nodes of a directed graph, i.e., intrinsic temporal relation graph (ITRG), and the temporal relations between them as the edges. Instead of directly prompting a pre-trained language model, a relation-guided prompting method is introduced to generate ITRG in a hierarchical manner. We customize four types of multilateral temporal logical rules (i.e., identity, inclusion, synchronization, and succession) from ITRG and utilize them to train our model. Experiments demonstrate the effectiveness and superiority of our method on the Charades-STA and ActivityNet Captions datasets.

Exploiting Intrinsic Multilateral Logical Rules for Weakly Supervised Natural Language Video Localization

acl-long.247

Poster

https://aclanthology.org/2024.acl-long.247/

https://aclanthology.org/2024.acl-long.248.bib

@inproceedings{weber-etal-2024-interpretability, title = "Interpretability of Language Models via Task Spaces", author = "Weber, Lucas and Jumelet, Jaap and Bruni, Elia and Hupkes, Dieuwke", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.248", pages = "4522--4538", abstract = "The usual way to interpret language models (LMs) is to test their performance on different benchmarks and subsequently infer their internal processes.In this paper, we present an alternative approach, concentrating on the {\_}quality{\_} of LM processing, with a focus on their language abilities.To this end, we construct {`}linguistic task spaces{'} {--} representations of an LM{'}s language conceptualisation {--} that shed light on the connections LMs draw between language phenomena.Task spaces are based on the interactions of the learning signals from different linguistic phenomena, which we assess via a method we call {`}similarity probing{'}.To disentangle the learning signals of linguistic phenomena, we further introduce a method called {`}fine-tuning via gradient differentials{'} (FTGD).We apply our methods to language models of three different scales and find that larger models generalise better to overarching general concepts for linguistic tasks, making better use of their shared structure. Further, the distributedness of linguistic processing increases with pre-training through increased parameter sharing between related linguistic tasks. The overall generalisation patterns are mostly stable throughout training and not marked by incisive stages, potentially explaining the lack of successful curriculum strategies for LMs.", }

The usual way to interpret language models (LMs) is to test their performance on different benchmarks and subsequently infer their internal processes.In this paper, we present an alternative approach, concentrating on the {\_}quality{\_} of LM processing, with a focus on their language abilities.To this end, we construct {`}linguistic task spaces{'} {--} representations of an LM{'}s language conceptualisation {--} that shed light on the connections LMs draw between language phenomena.Task spaces are based on the interactions of the learning signals from different linguistic phenomena, which we assess via a method we call {`}similarity probing{'}.To disentangle the learning signals of linguistic phenomena, we further introduce a method called {`}fine-tuning via gradient differentials{'} (FTGD).We apply our methods to language models of three different scales and find that larger models generalise better to overarching general concepts for linguistic tasks, making better use of their shared structure. Further, the distributedness of linguistic processing increases with pre-training through increased parameter sharing between related linguistic tasks. The overall generalisation patterns are mostly stable throughout training and not marked by incisive stages, potentially explaining the lack of successful curriculum strategies for LMs.

Interpretability of Language Models via Task Spaces

acl-long.248

Oral

2406.06441

https://aclanthology.org/2024.acl-long.248/

https://aclanthology.org/2024.acl-long.249.bib

@inproceedings{cassotti-etal-2024-using, title = "Using Synchronic Definitions and Semantic Relations to Classify Semantic Change Types", author = "Cassotti, Pierluigi and De Pascale, Stefano and Tahmasebi, Nina", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.249", pages = "4539--4553", abstract = "There is abundant evidence of the fact that the way words change their meaning can be classified in different types of change, highlighting the relationship between the old and new meanings (among which generalisation, specialisation and co-hyponymy transfer).In this paper, we present a way of detecting these types of change by constructing a model that leverages information both from synchronic lexical relations and definitions of word meanings. Specifically, we use synset definitions and hierarchy information from WordNet and test it on a digitized version of Blank{'}s (1997) dataset of semantic change types. Finally, we show how the sense relationships can improve models for both approximation of human judgments of semantic relatedness as well as binary Lexical Semantic Change Detection.", }

There is abundant evidence of the fact that the way words change their meaning can be classified in different types of change, highlighting the relationship between the old and new meanings (among which generalisation, specialisation and co-hyponymy transfer).In this paper, we present a way of detecting these types of change by constructing a model that leverages information both from synchronic lexical relations and definitions of word meanings. Specifically, we use synset definitions and hierarchy information from WordNet and test it on a digitized version of Blank{'}s (1997) dataset of semantic change types. Finally, we show how the sense relationships can improve models for both approximation of human judgments of semantic relatedness as well as binary Lexical Semantic Change Detection.

Using Synchronic Definitions and Semantic Relations to Classify Semantic Change Types

acl-long.249

Poster

2406.03452

https://aclanthology.org/2024.acl-long.249/

https://aclanthology.org/2024.acl-long.250.bib

@inproceedings{mahaut-etal-2024-factual, title = "Factual Confidence of {LLM}s: on Reliability and Robustness of Current Estimators", author = {Mahaut, Mat{\'e}o and Aina, Laura and Czarnowska, Paula and Hardalov, Momchil and M{\"u}ller, Thomas and Marquez, Lluis}, editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.250", pages = "4554--4570", abstract = "Large Language Models (LLMs) tend to be unreliable on fact-based answers.To address this problem, NLP researchers have proposed a range of techniques to estimate LLM{'}s confidence over facts. However, due to the lack of a systematic comparison, it is not clear how the different methods compare to one other.To fill this gap, we present a rigorous survey and empirical comparison of estimators of factual confidence.We define an experimental framework allowing for fair comparison, covering both fact-verification and QA. Our experiments across a series of LLMs indicate that trained hidden-state probes provide the most reliable confidence estimates; albeit at the expense of requiring access to weights and supervision data. We also conduct a deeper assessment of the methods, in which we measure the consistency of model behavior under meaning-preserving variations in the input. We find that the factual confidence of LLMs is often unstable across semantically equivalent inputs, suggesting there is much room for improvement for the stability of models{'} parametric knowledge.", }

Large Language Models (LLMs) tend to be unreliable on fact-based answers.To address this problem, NLP researchers have proposed a range of techniques to estimate LLM{'}s confidence over facts. However, due to the lack of a systematic comparison, it is not clear how the different methods compare to one other.To fill this gap, we present a rigorous survey and empirical comparison of estimators of factual confidence.We define an experimental framework allowing for fair comparison, covering both fact-verification and QA. Our experiments across a series of LLMs indicate that trained hidden-state probes provide the most reliable confidence estimates; albeit at the expense of requiring access to weights and supervision data. We also conduct a deeper assessment of the methods, in which we measure the consistency of model behavior under meaning-preserving variations in the input. We find that the factual confidence of LLMs is often unstable across semantically equivalent inputs, suggesting there is much room for improvement for the stability of models{'} parametric knowledge.

Factual Confidence of LLMs: on Reliability and Robustness of Current Estimators

acl-long.250

Poster

2406.13415

https://aclanthology.org/2024.acl-long.250/

https://aclanthology.org/2024.acl-long.251.bib

@inproceedings{dou-etal-2024-stepcoder, title = "{S}tep{C}oder: Improving Code Generation with Reinforcement Learning from Compiler Feedback", author = "Dou, Shihan and Liu, Yan and Jia, Haoxiang and Zhou, Enyu and Xiong, Limao and Shan, Junjie and Huang, Caishuang and Wang, Xiao and Fan, Xiaoran and Xi, Zhiheng and Zhou, Yuhao and Ji, Tao and Zheng, Rui and Zhang, Qi and Gui, Tao and Huang, Xuanjing", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.251", pages = "4571--4585", abstract = "The advancement of large language models (LLMs) has significantly propelled the field of code generation. Previous work integrated reinforcement learning (RL) with compiler feedback for exploring the output space of LLMs to enhance code generation quality. However, the lengthy code generated by LLMs in response to complex human requirements makes RL exploration a challenge. Also, since the unit tests may not cover the complicated code, optimizing LLMs by using these unexecuted code snippets is ineffective. To tackle these challenges, we introduce \textbf{StepCoder}, a novel RL framework for code generation, consisting of two main components: CCCS addresses the exploration challenge by breaking the long sequences code generation task into a Curriculum of Code Completion Subtasks, while FGO only optimizes the model by masking the unexecuted code segments to provide Fine-Grained Optimization. In addition, we furthermore construct the APPS+ dataset for RL training, which is manually verified to ensure the correctness of unit tests. Experimental results show that our method improves the ability to explore the output space and outperforms state-of-the-art approaches in corresponding benchmarks. The code and dataset will be made available upon publication.", }

The advancement of large language models (LLMs) has significantly propelled the field of code generation. Previous work integrated reinforcement learning (RL) with compiler feedback for exploring the output space of LLMs to enhance code generation quality. However, the lengthy code generated by LLMs in response to complex human requirements makes RL exploration a challenge. Also, since the unit tests may not cover the complicated code, optimizing LLMs by using these unexecuted code snippets is ineffective. To tackle these challenges, we introduce \textbf{StepCoder}, a novel RL framework for code generation, consisting of two main components: CCCS addresses the exploration challenge by breaking the long sequences code generation task into a Curriculum of Code Completion Subtasks, while FGO only optimizes the model by masking the unexecuted code segments to provide Fine-Grained Optimization. In addition, we furthermore construct the APPS+ dataset for RL training, which is manually verified to ensure the correctness of unit tests. Experimental results show that our method improves the ability to explore the output space and outperforms state-of-the-art approaches in corresponding benchmarks. The code and dataset will be made available upon publication.

StepCoder: Improving Code Generation with Reinforcement Learning from Compiler Feedback

acl-long.251

Poster

https://huggingface.co/papers/2402.01391

https://aclanthology.org/2024.acl-long.251/

https://aclanthology.org/2024.acl-long.252.bib

@inproceedings{li-etal-2024-one, title = "One-Shot Learning as Instruction Data Prospector for Large Language Models", author = "Li, Yunshui and Hui, Binyuan and Xia, Xiaobo and Yang, Jiaxi and Yang, Min and Zhang, Lei and Si, Shuzheng and Chen, Ling-Hao and Liu, Junhao and Liu, Tongliang and Huang, Fei and Li, Yongbin", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.252", pages = "4586--4601", abstract = "Contemporary practices in instruction tuning often hinge on enlarging data scaling without a clear strategy for ensuring data quality, inadvertently introducing noise that may compromise model performance. To address this challenge, we introduce Nuggets, a novel and efficient methodology that leverages one-shot learning to discern and select high-quality instruction data from extensive datasets. Nuggets assesses the potential of individual instruction examples to act as effective one-shot learning instances, thereby identifying those that can significantly improve performance across diverse tasks. Nuggets utilizes a scoring system based on the impact of candidate examples on the perplexity of a diverse anchor set, facilitating the selection of the most advantageous data for instruction tuning. Through rigorous evaluations on two benchmarks, namely MT-Bench and Alpaca-Eval, our study illustrates that instruction tuning with the top 1{\%} of examples curated by Nuggets substantially outperforms conventional methods employing the entire dataset.", }

Contemporary practices in instruction tuning often hinge on enlarging data scaling without a clear strategy for ensuring data quality, inadvertently introducing noise that may compromise model performance. To address this challenge, we introduce Nuggets, a novel and efficient methodology that leverages one-shot learning to discern and select high-quality instruction data from extensive datasets. Nuggets assesses the potential of individual instruction examples to act as effective one-shot learning instances, thereby identifying those that can significantly improve performance across diverse tasks. Nuggets utilizes a scoring system based on the impact of candidate examples on the perplexity of a diverse anchor set, facilitating the selection of the most advantageous data for instruction tuning. Through rigorous evaluations on two benchmarks, namely MT-Bench and Alpaca-Eval, our study illustrates that instruction tuning with the top 1{\%} of examples curated by Nuggets substantially outperforms conventional methods employing the entire dataset.

One-Shot Learning as Instruction Data Prospector for Large Language Models

acl-long.252

Poster

2312.10302

https://huggingface.co/papers/2312.10302

https://aclanthology.org/2024.acl-long.252/

https://aclanthology.org/2024.acl-long.253.bib

@inproceedings{shi-etal-2024-navigating, title = "Navigating the {O}ver{K}ill in Large Language Models", author = "Shi, Chenyu and Wang, Xiao and Ge, Qiming and Gao, Songyang and Yang, Xianjun and Gui, Tao and Zhang, Qi and Huang, Xuanjing and Zhao, Xun and Lin, Dahua", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.253", pages = "4602--4614", abstract = "Large language models are meticulously aligned to be both helpful and harmless. However, recent research points to a potential overkill which means models may refuse to answer benign queries. In this paper, we investigate the factors for overkill by exploring how models handle and determine the safety of queries. Our findings reveal the presence of shortcuts within models, leading to excessive attention to harmful words like {`}kill{'} and prompts emphasizing safety will exacerbate overkill. Based on these insights, we introduce Self-Contrastive Decoding (Self-CD), a training-free and model-agnostic strategy, to alleviate this phenomenon. We first extract such excessive attention by amplifying the difference in the model{'}s output distributions when responding to system prompts that either include or omit an emphasis on safety. Then we determine the final next-token predictions by downplaying the excessive attention via contrastive decoding. Empirical results have indicated that our method has achieved an average reduction of the refusal rate by 20 {\%} while having almost no impact on safety.", }

Large language models are meticulously aligned to be both helpful and harmless. However, recent research points to a potential overkill which means models may refuse to answer benign queries. In this paper, we investigate the factors for overkill by exploring how models handle and determine the safety of queries. Our findings reveal the presence of shortcuts within models, leading to excessive attention to harmful words like {`}kill{'} and prompts emphasizing safety will exacerbate overkill. Based on these insights, we introduce Self-Contrastive Decoding (Self-CD), a training-free and model-agnostic strategy, to alleviate this phenomenon. We first extract such excessive attention by amplifying the difference in the model{'}s output distributions when responding to system prompts that either include or omit an emphasis on safety. Then we determine the final next-token predictions by downplaying the excessive attention via contrastive decoding. Empirical results have indicated that our method has achieved an average reduction of the refusal rate by 20 {\%} while having almost no impact on safety.

Navigating the OverKill in Large Language Models

acl-long.253

Poster

2401.17633

https://aclanthology.org/2024.acl-long.253/

https://aclanthology.org/2024.acl-long.254.bib

@inproceedings{jacovi-etal-2024-chain, title = "A Chain-of-Thought Is as Strong as Its Weakest Link: A Benchmark for Verifiers of Reasoning Chains", author = "Jacovi, Alon and Bitton, Yonatan and Bohnet, Bernd and Herzig, Jonathan and Honovich, Or and Tseng, Michael and Collins, Michael and Aharoni, Roee and Geva, Mor", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.254", pages = "4615--4634", abstract = "Prompting language models to provide step-by-step answers (e.g., {``}Chain-of-Thought{''}) is the prominent approach for complex reasoning tasks, where more accurate reasoning chains typically improve downstream task performance. Recent literature discusses automatic methods to verify reasoning to evaluate and improve their correctness. However, no fine-grained step-level datasets are available to enable thorough evaluation of such verification methods, hindering progress in this direction. We introduce REVEAL: Reasoning Verification Evaluation, a dataset to benchmark automatic verifiers of complex Chain-of-Thought reasoning in open-domain question-answering settings. REVEAL includes comprehensive labels for the relevance, attribution to evidence passages, and logical correctness of each reasoning step in a language model{'}s answer, across a variety of datasets and state-of-the-art language models. Evaluation on REVEAL shows that verifiers struggle at verifying reasoning chains - in particular, verifying logical correctness and detecting contradictions. Available at https://reveal-dataset.github.io/ .", }

Prompting language models to provide step-by-step answers (e.g., {``}Chain-of-Thought{''}) is the prominent approach for complex reasoning tasks, where more accurate reasoning chains typically improve downstream task performance. Recent literature discusses automatic methods to verify reasoning to evaluate and improve their correctness. However, no fine-grained step-level datasets are available to enable thorough evaluation of such verification methods, hindering progress in this direction. We introduce REVEAL: Reasoning Verification Evaluation, a dataset to benchmark automatic verifiers of complex Chain-of-Thought reasoning in open-domain question-answering settings. REVEAL includes comprehensive labels for the relevance, attribution to evidence passages, and logical correctness of each reasoning step in a language model{'}s answer, across a variety of datasets and state-of-the-art language models. Evaluation on REVEAL shows that verifiers struggle at verifying reasoning chains - in particular, verifying logical correctness and detecting contradictions. Available at https://reveal-dataset.github.io/ .

A Chain-of-Thought Is as Strong as Its Weakest Link: A Benchmark for Verifiers of Reasoning Chains

acl-long.254

Oral

2402.00559

https://huggingface.co/papers/2402.00559

https://aclanthology.org/2024.acl-long.254/

https://aclanthology.org/2024.acl-long.255.bib

@inproceedings{ruan-etal-2024-re3, title = "Re3: A Holistic Framework and Dataset for Modeling Collaborative Document Revision", author = "Ruan, Qian and Kuznetsov, Ilia and Gurevych, Iryna", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.255", pages = "4635--4655", abstract = "Collaborative review and revision of textual documents is the core of knowledge work and a promising target for empirical analysis and NLP assistance. Yet, a holistic framework that would allow modeling complex relationships between document revisions, reviews and author responses is lacking. To address this gap, we introduce Re3, a framework for joint analysis of collaborative document revision. We instantiate this framework in the scholarly domain, and present Re3-Sci, a large corpus of aligned scientific paper revisions manually labeled according to their action and intent, and supplemented with the respective peer reviews and human-written edit summaries. We use the new data to provide first empirical insights into collaborative document revision in the academic domain, and to assess the capabilities of state-of-the-art LLMs at automating edit analysis and facilitating text-based collaboration. We make our annotation environment and protocols, the resulting data and experimental code publicly available.", }

Collaborative review and revision of textual documents is the core of knowledge work and a promising target for empirical analysis and NLP assistance. Yet, a holistic framework that would allow modeling complex relationships between document revisions, reviews and author responses is lacking. To address this gap, we introduce Re3, a framework for joint analysis of collaborative document revision. We instantiate this framework in the scholarly domain, and present Re3-Sci, a large corpus of aligned scientific paper revisions manually labeled according to their action and intent, and supplemented with the respective peer reviews and human-written edit summaries. We use the new data to provide first empirical insights into collaborative document revision in the academic domain, and to assess the capabilities of state-of-the-art LLMs at automating edit analysis and facilitating text-based collaboration. We make our annotation environment and protocols, the resulting data and experimental code publicly available.

Re3: A Holistic Framework and Dataset for Modeling Collaborative Document Revision

acl-long.255

Poster

2406.00197

https://aclanthology.org/2024.acl-long.255/

https://aclanthology.org/2024.acl-long.256.bib

@inproceedings{czinczoll-etal-2024-nextlevelbert, title = "{N}ext{L}evel{BERT}: Masked Language Modeling with Higher-Level Representations for Long Documents", author = {Czinczoll, Tamara and H{\"o}nes, Christoph and Schall, Maximilian and De Melo, Gerard}, editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.256", pages = "4656--4666", abstract = "While (large) language models have significantly improved over the last years, they still struggle to sensibly process long sequences found, e.g., in books, due to the quadratic scaling of the underlying attention mechanism. To address this, we propose NextLevelBERT, a Masked Language Model operating not on tokens, but on higher-level semantic representations in the form of text embeddings. We pretrain NextLevelBERT to predict the vector representation of entire masked text chunks and evaluate the effectiveness of the resulting document vectors on three types of tasks: 1) Semantic Textual Similarity via zero-shot document embeddings, 2) Long document classification, 3) Multiple-choice question answering. We find that next-level Masked Language Modeling is an effective technique to tackle long-document use cases and can outperform much larger embedding models as long as the required level of detail of semantic information is not too fine. Our models and code are publicly available online.", }

While (large) language models have significantly improved over the last years, they still struggle to sensibly process long sequences found, e.g., in books, due to the quadratic scaling of the underlying attention mechanism. To address this, we propose NextLevelBERT, a Masked Language Model operating not on tokens, but on higher-level semantic representations in the form of text embeddings. We pretrain NextLevelBERT to predict the vector representation of entire masked text chunks and evaluate the effectiveness of the resulting document vectors on three types of tasks: 1) Semantic Textual Similarity via zero-shot document embeddings, 2) Long document classification, 3) Multiple-choice question answering. We find that next-level Masked Language Modeling is an effective technique to tackle long-document use cases and can outperform much larger embedding models as long as the required level of detail of semantic information is not too fine. Our models and code are publicly available online.

NextLevelBERT: Masked Language Modeling with Higher-Level Representations for Long Documents

acl-long.256

Poster

2402.17682

https://huggingface.co/papers/2402.17682

https://aclanthology.org/2024.acl-long.256/

https://aclanthology.org/2024.acl-long.257.bib

@inproceedings{jiang-etal-2024-followbench, title = "{F}ollow{B}ench: A Multi-level Fine-grained Constraints Following Benchmark for Large Language Models", author = "Jiang, Yuxin and Wang, Yufei and Zeng, Xingshan and Zhong, Wanjun and Li, Liangyou and Mi, Fei and Shang, Lifeng and Jiang, Xin and Liu, Qun and Wang, Wei", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.257", pages = "4667--4688", abstract = "The ability to follow instructions is crucial for Large Language Models (LLMs) to handle various real-world applications. Existing benchmarks primarily focus on evaluating pure response quality, rather than assessing whether the response follows constraints stated in the instruction. To fill this research gap, in this paper, we propose FollowBench, a Multi-level Fine-grained Constraints Following Benchmark for LLMs. FollowBench comprehensively includes five different types (i.e., Content, Situation, Style, Format, and Example) of fine-grained constraints. To enable a precise constraint following estimation on diverse difficulties, we introduce a Multi-level mechanism that incrementally adds a single constraint to the initial instruction at each increased level. To assess whether LLMs{'} outputs have satisfied every individual constraint, we propose to prompt strong LLMs with constraint-evolution paths to handle challenging open-ended instructions. By evaluating 13 closed-source and open-source popular LLMs on FollowBench, we highlight the weaknesses of LLMs in instruction following and point towards potential avenues for future work. The data and code are publicly available at https://github.com/YJiangcm/FollowBench.", }

The ability to follow instructions is crucial for Large Language Models (LLMs) to handle various real-world applications. Existing benchmarks primarily focus on evaluating pure response quality, rather than assessing whether the response follows constraints stated in the instruction. To fill this research gap, in this paper, we propose FollowBench, a Multi-level Fine-grained Constraints Following Benchmark for LLMs. FollowBench comprehensively includes five different types (i.e., Content, Situation, Style, Format, and Example) of fine-grained constraints. To enable a precise constraint following estimation on diverse difficulties, we introduce a Multi-level mechanism that incrementally adds a single constraint to the initial instruction at each increased level. To assess whether LLMs{'} outputs have satisfied every individual constraint, we propose to prompt strong LLMs with constraint-evolution paths to handle challenging open-ended instructions. By evaluating 13 closed-source and open-source popular LLMs on FollowBench, we highlight the weaknesses of LLMs in instruction following and point towards potential avenues for future work. The data and code are publicly available at https://github.com/YJiangcm/FollowBench.

FollowBench: A Multi-level Fine-grained Constraints Following Benchmark for Large Language Models

acl-long.257

Poster

2310.20410

https://huggingface.co/papers/2310.20410

https://aclanthology.org/2024.acl-long.257/

https://aclanthology.org/2024.acl-long.258.bib

@inproceedings{jiang-etal-2024-learning, title = "Learning to Edit: Aligning {LLM}s with Knowledge Editing", author = "Jiang, Yuxin and Wang, Yufei and Wu, Chuhan and Zhong, Wanjun and Zeng, Xingshan and Gao, Jiahui and Li, Liangyou and Jiang, Xin and Shang, Lifeng and Tang, Ruiming and Liu, Qun and Wang, Wei", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.258", pages = "4689--4705", abstract = "Knowledge editing techniques, aiming to efficiently modify a minor proportion of knowledge in large language models (LLMs) without negatively impacting performance across other inputs, have garnered widespread attention. However, existing methods predominantly rely on memorizing the updated knowledge, impeding LLMs from effectively combining the new knowledge with their inherent knowledge when answering questions. To this end, we propose a Learning to Edit (LTE) framework, focusing on teaching LLMs to apply updated knowledge into input questions, inspired by the philosophy of {``}Teach a man to fish.{''} LTE features a two-phase process: (i) the Alignment Phase, which fine-tunes LLMs on a meticulously curated parallel dataset to make reliable, in-scope edits while preserving out-of-scope information and linguistic proficiency; and (ii) the Inference Phase, which employs a retrieval-based mechanism for real-time and mass knowledge editing. By comparing our approach with seven advanced baselines across four popular knowledge editing benchmarks and two LLM architectures, we demonstrate LTE{'}s superiority in knowledge editing performance, robustness in both batch and sequential editing, minimal interference on general tasks, and rapid editing speeds. The data and code are publicly available at https://github.com/YJiangcm/LTE.", }

Knowledge editing techniques, aiming to efficiently modify a minor proportion of knowledge in large language models (LLMs) without negatively impacting performance across other inputs, have garnered widespread attention. However, existing methods predominantly rely on memorizing the updated knowledge, impeding LLMs from effectively combining the new knowledge with their inherent knowledge when answering questions. To this end, we propose a Learning to Edit (LTE) framework, focusing on teaching LLMs to apply updated knowledge into input questions, inspired by the philosophy of {``}Teach a man to fish.{''} LTE features a two-phase process: (i) the Alignment Phase, which fine-tunes LLMs on a meticulously curated parallel dataset to make reliable, in-scope edits while preserving out-of-scope information and linguistic proficiency; and (ii) the Inference Phase, which employs a retrieval-based mechanism for real-time and mass knowledge editing. By comparing our approach with seven advanced baselines across four popular knowledge editing benchmarks and two LLM architectures, we demonstrate LTE{'}s superiority in knowledge editing performance, robustness in both batch and sequential editing, minimal interference on general tasks, and rapid editing speeds. The data and code are publicly available at https://github.com/YJiangcm/LTE.

Learning to Edit: Aligning LLMs with Knowledge Editing

acl-long.258

Poster

2402.11905

https://huggingface.co/papers/2402.11905

https://aclanthology.org/2024.acl-long.258/

https://aclanthology.org/2024.acl-long.259.bib

@inproceedings{wang-etal-2024-dolphcoder, title = "{D}olph{C}oder: Echo-Locating Code Large Language Models with Diverse and Multi-Objective Instruction Tuning", author = "Wang, Yejie and He, Keqing and Dong, Guanting and Wang, Pei and Zeng, Weihao and Diao, Muxi and Xu, Weiran and Wang, Jingang and Zhang, Mengdi and Cai, Xunliang", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.259", pages = "4706--4721", abstract = "Code Large Language Models (Code LLMs) have demonstrated outstanding performance in code-related tasks. Various instruction finetuning approaches have been proposed to boost the code generation performance of pre-trained Code LLMs. In this paper, we introduce a diverse instruction model DolphCoder with self-evaluating for code generation. It learns diverse instruction targets and combines a code evaluation objective to enhance its code generation ability. Our model achieves superior performance on the HumanEval and MBPP benchmarks, demonstrating new insights for future code instruction tuning work. Our key findings are: (1) Augmenting more diverse responses with more distinct reasoning paths increases the code capability of LLMs. (2) Improving one{'}s ability to evaluate the correctness of code also enhances their ability to create it.", }

Code Large Language Models (Code LLMs) have demonstrated outstanding performance in code-related tasks. Various instruction finetuning approaches have been proposed to boost the code generation performance of pre-trained Code LLMs. In this paper, we introduce a diverse instruction model DolphCoder with self-evaluating for code generation. It learns diverse instruction targets and combines a code evaluation objective to enhance its code generation ability. Our model achieves superior performance on the HumanEval and MBPP benchmarks, demonstrating new insights for future code instruction tuning work. Our key findings are: (1) Augmenting more diverse responses with more distinct reasoning paths increases the code capability of LLMs. (2) Improving one{'}s ability to evaluate the correctness of code also enhances their ability to create it.

DolphCoder: Echo-Locating Code Large Language Models with Diverse and Multi-Objective Instruction Tuning

acl-long.259

Poster

2402.09136

https://huggingface.co/papers/2402.09136

https://aclanthology.org/2024.acl-long.259/

https://aclanthology.org/2024.acl-long.260.bib

@inproceedings{madureira-etal-2024-time, title = "When Only Time Will Tell: Interpreting How Transformers Process Local Ambiguities Through the Lens of Restart-Incrementality", author = "Madureira, Brielen and Kahardipraja, Patrick and Schlangen, David", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.260", pages = "4722--4749", abstract = "Incremental models that process sentences one token at a time will sometimes encounter points where more than one interpretation is possible. Causal models are forced to output one interpretation and continue, whereas models that can revise may edit their previous output as the ambiguity is resolved. In this work, we look at how restart-incremental Transformers build and update internal states, in an effort to shed light on what processes cause revisions not viable in autoregressive models. We propose an interpretable way to analyse the incremental states, showing that their sequential structure encodes information on the garden path effect and its resolution. Our method brings insights on various bidirectional encoders for contextualised meaning representation and dependency parsing, contributing to show their advantage over causal models when it comes to revisions.", }

Incremental models that process sentences one token at a time will sometimes encounter points where more than one interpretation is possible. Causal models are forced to output one interpretation and continue, whereas models that can revise may edit their previous output as the ambiguity is resolved. In this work, we look at how restart-incremental Transformers build and update internal states, in an effort to shed light on what processes cause revisions not viable in autoregressive models. We propose an interpretable way to analyse the incremental states, showing that their sequential structure encodes information on the garden path effect and its resolution. Our method brings insights on various bidirectional encoders for contextualised meaning representation and dependency parsing, contributing to show their advantage over causal models when it comes to revisions.

When Only Time Will Tell: Interpreting How Transformers Process Local Ambiguities Through the Lens of Restart-Incrementality

acl-long.260

Poster

2402.13113

https://aclanthology.org/2024.acl-long.260/

https://aclanthology.org/2024.acl-long.261.bib

@inproceedings{rizvi-etal-2024-sparc, title = "{S}pa{RC} and {S}pa{RP}: Spatial Reasoning Characterization and Path Generation for Understanding Spatial Reasoning Capability of Large Language Models", author = "Rizvi, Md Imbesat and Zhu, Xiaodan and Gurevych, Iryna", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.261", pages = "4750--4767", abstract = "Spatial reasoning is a crucial component of both biological and artificial intelligence. In this work, we present a comprehensive study of the capability of current state-of-the-art large language models (LLMs) on spatial reasoning. To support our study, we created and contribute a novel Spatial Reasoning Characterization (SpaRC) framework and Spatial Reasoning Paths (SpaRP) datasets, to enable an in-depth understanding of the spatial relations and compositions as well as the usefulness of spatial reasoning chains. We found that all the state-of-the-art LLMs do not perform well on the datasets{---}their performances are consistently low across different setups. The spatial reasoning capability improves substantially as model sizes scale up. Finetuning both large language models (e.g., Llama-2-70B) and smaller ones (e.g., Llama-2-13B) can significantly improve their F1-scores by 7{--}32 absolute points. We also found that the top proprietary LLMs still significantly outperform their open-source counterparts in topological spatial understanding and reasoning.", }

Spatial reasoning is a crucial component of both biological and artificial intelligence. In this work, we present a comprehensive study of the capability of current state-of-the-art large language models (LLMs) on spatial reasoning. To support our study, we created and contribute a novel Spatial Reasoning Characterization (SpaRC) framework and Spatial Reasoning Paths (SpaRP) datasets, to enable an in-depth understanding of the spatial relations and compositions as well as the usefulness of spatial reasoning chains. We found that all the state-of-the-art LLMs do not perform well on the datasets{---}their performances are consistently low across different setups. The spatial reasoning capability improves substantially as model sizes scale up. Finetuning both large language models (e.g., Llama-2-70B) and smaller ones (e.g., Llama-2-13B) can significantly improve their F1-scores by 7{--}32 absolute points. We also found that the top proprietary LLMs still significantly outperform their open-source counterparts in topological spatial understanding and reasoning.

SpaRC and SpaRP: Spatial Reasoning Characterization and Path Generation for Understanding Spatial Reasoning Capability of Large Language Models

acl-long.261

Poster

2406.04566

https://aclanthology.org/2024.acl-long.261/

https://aclanthology.org/2024.acl-long.262.bib

@inproceedings{he-etal-2024-planning, title = "Planning Like Human: A Dual-process Framework for Dialogue Planning", author = "He, Tao and Liao, Lizi and Cao, Yixin and Liu, Yuanxing and Liu, Ming and Chen, Zerui and Qin, Bing", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.262", pages = "4768--4791", abstract = "In proactive dialogue, the challenge lies not just in generating responses but in steering conversations toward predetermined goals, a task where Large Language Models (LLMs) typically struggle due to their reactive nature. Traditional approaches to enhance dialogue planning in LLMs, ranging from elaborate prompt engineering to the integration of policy networks, either face efficiency issues or deliver suboptimal performance. Inspired by the dual-process theory in psychology, which identifies two distinct modes of thinking{---}intuitive (fast) and analytical (slow), we propose the Dual-Process Dialogue Planning (DPDP) framework. DPDP embodies this theory through two complementary planning systems: an instinctive policy model for familiar contexts and a deliberative Monte Carlo Tree Search (MCTS) mechanism for complex, novel scenarios. This dual strategy is further coupled with a novel two-stage training regimen: offline Reinforcement Learning for robust initial policy model formation followed by MCTS-enhanced on-the-fly learning, which ensures a dynamic balance between efficiency and strategic depth. Our empirical evaluations across diverse dialogue tasks affirm DPDP{'}s superiority in achieving both high-quality dialogues and operational efficiency, outpacing existing methods.", }

In proactive dialogue, the challenge lies not just in generating responses but in steering conversations toward predetermined goals, a task where Large Language Models (LLMs) typically struggle due to their reactive nature. Traditional approaches to enhance dialogue planning in LLMs, ranging from elaborate prompt engineering to the integration of policy networks, either face efficiency issues or deliver suboptimal performance. Inspired by the dual-process theory in psychology, which identifies two distinct modes of thinking{---}intuitive (fast) and analytical (slow), we propose the Dual-Process Dialogue Planning (DPDP) framework. DPDP embodies this theory through two complementary planning systems: an instinctive policy model for familiar contexts and a deliberative Monte Carlo Tree Search (MCTS) mechanism for complex, novel scenarios. This dual strategy is further coupled with a novel two-stage training regimen: offline Reinforcement Learning for robust initial policy model formation followed by MCTS-enhanced on-the-fly learning, which ensures a dynamic balance between efficiency and strategic depth. Our empirical evaluations across diverse dialogue tasks affirm DPDP{'}s superiority in achieving both high-quality dialogues and operational efficiency, outpacing existing methods.

Planning Like Human: A Dual-process Framework for Dialogue Planning

acl-long.262

Poster

2406.05374

https://aclanthology.org/2024.acl-long.262/

https://aclanthology.org/2024.acl-long.263.bib

@inproceedings{cancedda-2024-spectral, title = "Spectral Filters, Dark Signals, and Attention Sinks", author = "Cancedda, Nicola", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.263", pages = "4792--4808", abstract = "Projecting intermediate representations onto the vocabulary is an increasingly popular interpretation tool for transformer-based LLMs, also known as the logit lens (Nostalgebraist). We propose a quantitative extension to this approach and define spectral filters on intermediate representations based on partitioning the singular vectors of the vocabulary embedding and unembedding matrices into bands. We find that the signals exchanged in the tail end of the spectrum, i.e. corresponding to the singular vectors with smallest singular values, are responsible for attention sinking (Xiao et al., 2023), of which we provide an explanation. We find that the negative log-likelihood of pretrained models can be kept low despite suppressing sizeable parts of the embedding spectrum in a layer-dependent way, as long as attention sinking is preserved. Finally, we discover that the representation of tokens that draw attention from many tokens have large projections on the tail end of the spectrum, and likely act as additional attention sinks.", }

Projecting intermediate representations onto the vocabulary is an increasingly popular interpretation tool for transformer-based LLMs, also known as the logit lens (Nostalgebraist). We propose a quantitative extension to this approach and define spectral filters on intermediate representations based on partitioning the singular vectors of the vocabulary embedding and unembedding matrices into bands. We find that the signals exchanged in the tail end of the spectrum, i.e. corresponding to the singular vectors with smallest singular values, are responsible for attention sinking (Xiao et al., 2023), of which we provide an explanation. We find that the negative log-likelihood of pretrained models can be kept low despite suppressing sizeable parts of the embedding spectrum in a layer-dependent way, as long as attention sinking is preserved. Finally, we discover that the representation of tokens that draw attention from many tokens have large projections on the tail end of the spectrum, and likely act as additional attention sinks.

Spectral Filters, Dark Signals, and Attention Sinks

acl-long.263

Poster

2402.09221

https://aclanthology.org/2024.acl-long.263/

https://aclanthology.org/2024.acl-long.264.bib

@inproceedings{gao-etal-2024-diffucomet, title = "{D}iffu{COMET}: Contextual Commonsense Knowledge Diffusion", author = "Gao, Silin and Ismayilzada, Mete and Zhao, Mengjie and Wakaki, Hiromi and Mitsufuji, Yuki and Bosselut, Antoine", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.264", pages = "4809--4831", abstract = "Inferring contextually-relevant and diverse commonsense to understand narratives remains challenging for knowledge models. In this work, we develop a series of knowledge models, DiffuCOMET, that leverage diffusion to learn to reconstruct the implicit semantic connections between narrative contexts and relevant commonsense knowledge. Across multiple diffusion steps, our method progressively refines a representation of commonsense facts that is anchored to a narrative, producing contextually-relevant and diverse commonsense inferences for an input context. To evaluate DiffuCOMET, we introduce new metrics for commonsense inference that more closely measure knowledge diversity and contextual relevance. Our results on two different benchmarks, ComFact and WebNLG+, show that knowledge generated by DiffuCOMET achieves a better trade-off between commonsense diversity, contextual relevance and alignment to known gold references, compared to baseline knowledge models.", }

Inferring contextually-relevant and diverse commonsense to understand narratives remains challenging for knowledge models. In this work, we develop a series of knowledge models, DiffuCOMET, that leverage diffusion to learn to reconstruct the implicit semantic connections between narrative contexts and relevant commonsense knowledge. Across multiple diffusion steps, our method progressively refines a representation of commonsense facts that is anchored to a narrative, producing contextually-relevant and diverse commonsense inferences for an input context. To evaluate DiffuCOMET, we introduce new metrics for commonsense inference that more closely measure knowledge diversity and contextual relevance. Our results on two different benchmarks, ComFact and WebNLG+, show that knowledge generated by DiffuCOMET achieves a better trade-off between commonsense diversity, contextual relevance and alignment to known gold references, compared to baseline knowledge models.

DiffuCOMET: Contextual Commonsense Knowledge Diffusion

acl-long.264

Poster

2402.17011

https://aclanthology.org/2024.acl-long.264/

https://aclanthology.org/2024.acl-long.265.bib

@inproceedings{sahinuc-etal-2024-systematic, title = "Systematic Task Exploration with {LLM}s: A Study in Citation Text Generation", author = "{\c{S}}ahinu{\c{c}}, Furkan and Kuznetsov, Ilia and Hou, Yufang and Gurevych, Iryna", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.265", pages = "4832--4855", abstract = "Large language models (LLMs) bring unprecedented flexibility in defining and executing complex, creative natural language generation (NLG) tasks. Yet, this flexibility brings new challenges, as it introduces new degrees of freedom in formulating the task inputs and instructions and in evaluating model performance. To facilitate the exploration of creative NLG tasks, we propose a three-component research framework that consists of systematic input manipulation, reference data, and output measurement. We use this framework to explore citation text generation {--} a popular scholarly NLP task that lacks consensus on the task definition and evaluation metric and has not yet been tackled within the LLM paradigm. Our results highlight the importance of systematically investigating both task instruction and input configuration when prompting LLMs, and reveal non-trivial relationships between different evaluation metrics used for citation text generation. Additional human generation and human evaluation experiments provide new qualitative insights into the task to guide future research in citation text generation. We make our code and data publicly available.", }

Large language models (LLMs) bring unprecedented flexibility in defining and executing complex, creative natural language generation (NLG) tasks. Yet, this flexibility brings new challenges, as it introduces new degrees of freedom in formulating the task inputs and instructions and in evaluating model performance. To facilitate the exploration of creative NLG tasks, we propose a three-component research framework that consists of systematic input manipulation, reference data, and output measurement. We use this framework to explore citation text generation {--} a popular scholarly NLP task that lacks consensus on the task definition and evaluation metric and has not yet been tackled within the LLM paradigm. Our results highlight the importance of systematically investigating both task instruction and input configuration when prompting LLMs, and reveal non-trivial relationships between different evaluation metrics used for citation text generation. Additional human generation and human evaluation experiments provide new qualitative insights into the task to guide future research in citation text generation. We make our code and data publicly available.

Systematic Task Exploration with LLMs: A Study in Citation Text Generation

acl-long.265

Poster

2407.04046

https://aclanthology.org/2024.acl-long.265/

https://aclanthology.org/2024.acl-long.266.bib

@inproceedings{bortoletto-etal-2024-limits, title = "Limits of Theory of Mind Modelling in Dialogue-Based Collaborative Plan Acquisition", author = "Bortoletto, Matteo and Ruhdorfer, Constantin and Abdessaied, Adnen and Shi, Lei and Bulling, Andreas", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.266", pages = "4856--4871", abstract = "Recent work on dialogue-based collaborative plan acquisition (CPA) has suggested that Theory of Mind (ToM) modelling can improve missing knowledge prediction in settings with asymmetric skill-sets and knowledge. Although ToM was claimed to be important for effective collaboration, its real impact on this novel task remains under-explored. By representing plans as graphs and by exploiting task-specific constraints we show that, as performance on CPA nearly doubles when predicting one{'}s own missing knowledge, the improvements due to ToM modelling diminish. This phenomenon persists even when evaluating existing baseline methods. To better understand the relevance of ToM for CPA, we report a principled performance comparison of models with and without ToM features. Results across different models and ablations consistently suggest that learned ToM features are indeed more likely to reflect latent patterns in the data with no perceivable link to ToM. This finding calls for a deeper understanding of the role of ToM in CPA and beyond, as well as new methods for modelling and evaluating mental states in computational collaborative agents.", }

Recent work on dialogue-based collaborative plan acquisition (CPA) has suggested that Theory of Mind (ToM) modelling can improve missing knowledge prediction in settings with asymmetric skill-sets and knowledge. Although ToM was claimed to be important for effective collaboration, its real impact on this novel task remains under-explored. By representing plans as graphs and by exploiting task-specific constraints we show that, as performance on CPA nearly doubles when predicting one{'}s own missing knowledge, the improvements due to ToM modelling diminish. This phenomenon persists even when evaluating existing baseline methods. To better understand the relevance of ToM for CPA, we report a principled performance comparison of models with and without ToM features. Results across different models and ablations consistently suggest that learned ToM features are indeed more likely to reflect latent patterns in the data with no perceivable link to ToM. This finding calls for a deeper understanding of the role of ToM in CPA and beyond, as well as new methods for modelling and evaluating mental states in computational collaborative agents.

Limits of Theory of Mind Modelling in Dialogue-Based Collaborative Plan Acquisition

acl-long.266

Poster

2405.12621

https://huggingface.co/papers/2405.12621

https://aclanthology.org/2024.acl-long.266/

https://aclanthology.org/2024.acl-long.267.bib

@inproceedings{chen-etal-2024-temporal, title = "Temporal Knowledge Question Answering via Abstract Reasoning Induction", author = "Chen, Ziyang and Li, Dongfang and Zhao, Xiang and Hu, Baotian and Zhang, Min", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.267", pages = "4872--4889", abstract = "In this study, we address the challenge of enhancing temporal knowledge reasoning in Large Language Models (LLMs). LLMs often struggle with this task, leading to the generation of inaccurate or misleading responses. This issue mainly arises from their limited ability to handle evolving factual knowledge and complex temporal logic. To overcome these limitations, we propose Abstract Reasoning Induction (ARI) framework, which divides temporal reasoning into two distinct phases: Knowledge agnostic and Knowledge-based. This framework offers factual knowledge support to LLMs while minimizing the incorporation of extraneous noisy data. Concurrently, informed by the principles of constructivism, ARI provides LLMs the capability to engage in proactive, self-directed learning from both correct and incorrect historical reasoning samples. By teaching LLMs to actively construct knowledge and methods, it can significantly boosting their temporal reasoning abilities. Our approach achieves significant improvements, with relative gains of 29.7{\%} and 9.27{\%} on two temporal QA datasets, underscoring its efficacy in advancing temporal reasoning in LLMs. The code can be found at https: //github.com/czy1999/ARI-QA.", }

In this study, we address the challenge of enhancing temporal knowledge reasoning in Large Language Models (LLMs). LLMs often struggle with this task, leading to the generation of inaccurate or misleading responses. This issue mainly arises from their limited ability to handle evolving factual knowledge and complex temporal logic. To overcome these limitations, we propose Abstract Reasoning Induction (ARI) framework, which divides temporal reasoning into two distinct phases: Knowledge agnostic and Knowledge-based. This framework offers factual knowledge support to LLMs while minimizing the incorporation of extraneous noisy data. Concurrently, informed by the principles of constructivism, ARI provides LLMs the capability to engage in proactive, self-directed learning from both correct and incorrect historical reasoning samples. By teaching LLMs to actively construct knowledge and methods, it can significantly boosting their temporal reasoning abilities. Our approach achieves significant improvements, with relative gains of 29.7{\%} and 9.27{\%} on two temporal QA datasets, underscoring its efficacy in advancing temporal reasoning in LLMs. The code can be found at https: //github.com/czy1999/ARI-QA.

Temporal Knowledge Question Answering via Abstract Reasoning Induction

acl-long.267

Poster

2311.09149

https://aclanthology.org/2024.acl-long.267/

https://aclanthology.org/2024.acl-long.268.bib

@inproceedings{lee-etal-2024-wrote, title = "Who Wrote this Code? Watermarking for Code Generation", author = "Lee, Taehyun and Hong, Seokhee and Ahn, Jaewoo and Hong, Ilgee and Lee, Hwaran and Yun, Sangdoo and Shin, Jamin and Kim, Gunhee", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.268", pages = "4890--4911", abstract = "Since the remarkable generation performance of large language models raised ethical and legal concerns, approaches to detect machine-generated text by embedding watermarks are being developed.However, we discover that the existing works fail to function appropriately in code generation tasks due to the task{'}s nature of having low entropy.Extending a logit-modifying watermark method, we propose Selective WatErmarking via Entropy Thresholding (SWEET), which enhances detection ability and mitigates code quality degeneration by removing low-entropy segments at generating and detecting watermarks.Our experiments show that SWEET significantly improves code quality preservation while outperforming all baselines, including post-hoc detection methods, in detecting machine-generated code text.Our code is available inhttps://github.com/hongcheki/sweet-watermark.", }

Since the remarkable generation performance of large language models raised ethical and legal concerns, approaches to detect machine-generated text by embedding watermarks are being developed.However, we discover that the existing works fail to function appropriately in code generation tasks due to the task{'}s nature of having low entropy.Extending a logit-modifying watermark method, we propose Selective WatErmarking via Entropy Thresholding (SWEET), which enhances detection ability and mitigates code quality degeneration by removing low-entropy segments at generating and detecting watermarks.Our experiments show that SWEET significantly improves code quality preservation while outperforming all baselines, including post-hoc detection methods, in detecting machine-generated code text.Our code is available inhttps://github.com/hongcheki/sweet-watermark.

Who Wrote this Code? Watermarking for Code Generation

acl-long.268

Poster

2305.15060

https://huggingface.co/papers/2305.15060

https://aclanthology.org/2024.acl-long.268/

https://aclanthology.org/2024.acl-long.269.bib

@inproceedings{islam-etal-2024-mapcoder, title = "{M}ap{C}oder: Multi-Agent Code Generation for Competitive Problem Solving", author = "Islam, Md. Ashraful and Ali, Mohammed Eunus and Parvez, Md Rizwan", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.269", pages = "4912--4944", abstract = "Code synthesis, which requires a deep understanding of complex natural language (NL) problem descriptions, generation of code instructions for complex algorithms and data structures, and the successful execution of comprehensive unit tests, presents a significant challenge. Thus, while large language models (LLMs) demonstrate impressive proficiency in natural language processing (NLP), their performance in code generation tasks remains limited. In this paper, we introduce a new approach to code generation tasks leveraging the multi-agent prompting that uniquely replicates the full cycle of program synthesis as observed in human developers. Our framework, MapCoder, consists of four LLM agents specifically designed to emulate the stages of this cycle: recalling relevant examples, planning, code generation, and debugging. After conducting thorough experiments, with multiple LLMs ablations and analyses across eight challenging competitive problem-solving and program synthesis benchmarks{---}MapCoder showcases remarkable code generation capabilities, achieving their new state-of-the-art (pass@1) results{---}(HumanEval 93.9{\%}, MBPP 83.1{\%}, APPS 22.0{\%}, CodeContests 28.5{\%}, and xCodeEval 45.3{\%}). Moreover, our method consistently delivers superior performance across various programming languages and varying problem difficulties. We open-source our framework at https://github.com/Md-Ashraful-Pramanik/MapCoder.", }

Code synthesis, which requires a deep understanding of complex natural language (NL) problem descriptions, generation of code instructions for complex algorithms and data structures, and the successful execution of comprehensive unit tests, presents a significant challenge. Thus, while large language models (LLMs) demonstrate impressive proficiency in natural language processing (NLP), their performance in code generation tasks remains limited. In this paper, we introduce a new approach to code generation tasks leveraging the multi-agent prompting that uniquely replicates the full cycle of program synthesis as observed in human developers. Our framework, MapCoder, consists of four LLM agents specifically designed to emulate the stages of this cycle: recalling relevant examples, planning, code generation, and debugging. After conducting thorough experiments, with multiple LLMs ablations and analyses across eight challenging competitive problem-solving and program synthesis benchmarks{---}MapCoder showcases remarkable code generation capabilities, achieving their new state-of-the-art (pass@1) results{---}(HumanEval 93.9{\%}, MBPP 83.1{\%}, APPS 22.0{\%}, CodeContests 28.5{\%}, and xCodeEval 45.3{\%}). Moreover, our method consistently delivers superior performance across various programming languages and varying problem difficulties. We open-source our framework at https://github.com/Md-Ashraful-Pramanik/MapCoder.

MapCoder: Multi-Agent Code Generation for Competitive Problem Solving

acl-long.269

Poster

2405.11403

https://huggingface.co/papers/2405.11403

https://aclanthology.org/2024.acl-long.269/

https://aclanthology.org/2024.acl-long.270.bib

@inproceedings{zhu-etal-2024-relayattention, title = "{R}elay{A}ttention for Efficient Large Language Model Serving with Long System Prompts", author = "Zhu, Lei and Wang, Xinjiang and Zhang, Wayne and Lau, Rynson", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.270", pages = "4945--4957", abstract = "A practical large language model (LLM) service may involve a long system prompt, which specifies the instructions, examples, and knowledge documents of the task and is reused across requests. However, the long system prompt causes throughput/latency bottlenecks as the cost of generating the next token grows w.r.t the sequence length. This paper aims to improve the efficiency of LLM services that involve long system prompts. Our key observation is that handling these system prompts requires heavily redundant memory accesses in existing causal attention computation algorithms. Specifically, for batched requests, the cached hidden states (i.e., key-value pairs) of system prompts are transferred from off-chip DRAM to on-chip SRAM multiple times, each corresponding to an individual request. To eliminate such a redundancy, we propose RelayAttention, an attention algorithm that allows reading these hidden states from DRAM exactly once for a batch of input tokens. RelayAttention is a free lunch: it maintains the generation quality while requiring no model retraining, as it is based on a mathematical reformulation of causal attention. We have observed significant performance improvements to a production-level system, vLLM, through integration with RelayAttention. The improvements are even more profound with longer system prompts.", }

A practical large language model (LLM) service may involve a long system prompt, which specifies the instructions, examples, and knowledge documents of the task and is reused across requests. However, the long system prompt causes throughput/latency bottlenecks as the cost of generating the next token grows w.r.t the sequence length. This paper aims to improve the efficiency of LLM services that involve long system prompts. Our key observation is that handling these system prompts requires heavily redundant memory accesses in existing causal attention computation algorithms. Specifically, for batched requests, the cached hidden states (i.e., key-value pairs) of system prompts are transferred from off-chip DRAM to on-chip SRAM multiple times, each corresponding to an individual request. To eliminate such a redundancy, we propose RelayAttention, an attention algorithm that allows reading these hidden states from DRAM exactly once for a batch of input tokens. RelayAttention is a free lunch: it maintains the generation quality while requiring no model retraining, as it is based on a mathematical reformulation of causal attention. We have observed significant performance improvements to a production-level system, vLLM, through integration with RelayAttention. The improvements are even more profound with longer system prompts.

RelayAttention for Efficient Large Language Model Serving with Long System Prompts

acl-long.270

Poster

2402.14808

https://huggingface.co/papers/2402.14808

https://aclanthology.org/2024.acl-long.270/

https://aclanthology.org/2024.acl-long.271.bib

@inproceedings{wang-etal-2024-boosting-language, title = "Boosting Language Models Reasoning with Chain-of-Knowledge Prompting", author = "Wang, Jianing and Sun, Qiushi and Li, Xiang and Gao, Ming", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.271", pages = "4958--4981", abstract = "Recently, Chain-of-Thought (CoT) prompting has delivered success on complex reasoning tasks, which aims at designing a simple prompt like {``}Let{'}s think step by step{''} or multiple in-context exemplars with well-designed rationales to elicit Large Language Models (LLMs) to generate intermediate reasoning steps. However, the generated rationales often come with hallucinations, making unfactual and unfaithful reasoning chains. To mitigate this brittleness, we propose a novel Chain-of-Knowledge (CoK) prompting, where we aim at eliciting LLMs to generate explicit pieces of knowledge evidence in the form of structure triple. This is inspired by our human behaviors, i.e., we can draw a mind map or knowledge map as the reasoning evidence in the brain before answering a complex question. Benefiting from CoK, we additionally introduce an F2-Verification method to estimate the reliability of the reasoning chains in terms of factuality and faithfulness. For the unreliable response, the wrong evidence can be indicated to prompt the LLM to rethink. Extensive experiments demonstrate that our method can further improve the performance of commonsense, factual, symbolic, and arithmetic reasoning tasks.", }

Recently, Chain-of-Thought (CoT) prompting has delivered success on complex reasoning tasks, which aims at designing a simple prompt like {``}Let{'}s think step by step{''} or multiple in-context exemplars with well-designed rationales to elicit Large Language Models (LLMs) to generate intermediate reasoning steps. However, the generated rationales often come with hallucinations, making unfactual and unfaithful reasoning chains. To mitigate this brittleness, we propose a novel Chain-of-Knowledge (CoK) prompting, where we aim at eliciting LLMs to generate explicit pieces of knowledge evidence in the form of structure triple. This is inspired by our human behaviors, i.e., we can draw a mind map or knowledge map as the reasoning evidence in the brain before answering a complex question. Benefiting from CoK, we additionally introduce an F2-Verification method to estimate the reliability of the reasoning chains in terms of factuality and faithfulness. For the unreliable response, the wrong evidence can be indicated to prompt the LLM to rethink. Extensive experiments demonstrate that our method can further improve the performance of commonsense, factual, symbolic, and arithmetic reasoning tasks.

Boosting Language Models Reasoning with Chain-of-Knowledge Prompting

acl-long.271

Poster

2306.06427

https://huggingface.co/papers/2306.06427

https://aclanthology.org/2024.acl-long.271/

https://aclanthology.org/2024.acl-long.272.bib

@inproceedings{guo-etal-2024-open, title = "Open Grounded Planning: Challenges and Benchmark Construction", author = "Guo, Shiguang and Deng, Ziliang and Lin, Hongyu and Lu, Yaojie and Han, Xianpei and Sun, Le", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.272", pages = "4982--5003", abstract = "The emergence of large language models (LLMs) has increasingly drawn attention to the use of LLMs for human-like planning. Existing work on LLM-based planning either focuses on leveraging the inherent language generation capabilities of LLMs to produce free-style plans, or employs reinforcement learning approaches to learn decision-making for a limited set of actions within restricted environments. However, both approaches exhibit significant discrepancies from the open and executable requirements in real-world planning. In this paper, we propose a new planning task{--}open grounded planning. The primary objective of open grounded planning is to ask the model to generate an executable plan based on a variable action set, thereby ensuring the executability of the produced plan. To this end, we establishes a benchmark for open grounded planning spanning a wide range of domains. Then we test current state-of-the-art LLMs along with five planning approaches, revealing that existing LLMs and methods still struggle to address the challenges posed by grounded planning in open domains. The outcomes of this paper define and establish a foundational dataset for open grounded planning, and shed light on the potential challenges and future directions of LLM-based planning.", }

The emergence of large language models (LLMs) has increasingly drawn attention to the use of LLMs for human-like planning. Existing work on LLM-based planning either focuses on leveraging the inherent language generation capabilities of LLMs to produce free-style plans, or employs reinforcement learning approaches to learn decision-making for a limited set of actions within restricted environments. However, both approaches exhibit significant discrepancies from the open and executable requirements in real-world planning. In this paper, we propose a new planning task{--}open grounded planning. The primary objective of open grounded planning is to ask the model to generate an executable plan based on a variable action set, thereby ensuring the executability of the produced plan. To this end, we establishes a benchmark for open grounded planning spanning a wide range of domains. Then we test current state-of-the-art LLMs along with five planning approaches, revealing that existing LLMs and methods still struggle to address the challenges posed by grounded planning in open domains. The outcomes of this paper define and establish a foundational dataset for open grounded planning, and shed light on the potential challenges and future directions of LLM-based planning.

Open Grounded Planning: Challenges and Benchmark Construction

acl-long.272

Poster

2406.02903

https://aclanthology.org/2024.acl-long.272/

https://aclanthology.org/2024.acl-long.273.bib

@inproceedings{xu-etal-2024-llm, title = "{LLM} Knows Body Language, Too: Translating Speech Voices into Human Gestures", author = "Xu, Chenghao and Lyu, Guangtao and Yan, Jiexi and Yang, Muli and Deng, Cheng", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.273", pages = "5004--5013", abstract = "In response to the escalating demand for digital human representations, progress has been made in the generation of realistic human gestures from given speeches. Despite the remarkable achievements of recent research, the generation process frequently includes unintended, meaningless, or non-realistic gestures. To address this challenge, we propose a gesture translation paradigm, GesTran, which leverages large language models (LLMs) to deepen the understanding of the connection between speech and gesture and sequentially generates human gestures by interpreting gestures as a unique form of body language. The primary stage of the proposed framework employs a transformer-based auto-encoder network to encode human gestures into discrete symbols. Following this, the subsequent stage utilizes a pre-trained LLM to decipher the relationship between speech and gesture, translating the speech into gesture by interpreting the gesture as unique language tokens within the LLM. Our method has demonstrated state-of-the-art performance improvement through extensive and impartial experiments conducted on public TED and TED-Expressive datasets.", }

In response to the escalating demand for digital human representations, progress has been made in the generation of realistic human gestures from given speeches. Despite the remarkable achievements of recent research, the generation process frequently includes unintended, meaningless, or non-realistic gestures. To address this challenge, we propose a gesture translation paradigm, GesTran, which leverages large language models (LLMs) to deepen the understanding of the connection between speech and gesture and sequentially generates human gestures by interpreting gestures as a unique form of body language. The primary stage of the proposed framework employs a transformer-based auto-encoder network to encode human gestures into discrete symbols. Following this, the subsequent stage utilizes a pre-trained LLM to decipher the relationship between speech and gesture, translating the speech into gesture by interpreting the gesture as unique language tokens within the LLM. Our method has demonstrated state-of-the-art performance improvement through extensive and impartial experiments conducted on public TED and TED-Expressive datasets.

LLM Knows Body Language, Too: Translating Speech Voices into Human Gestures

acl-long.273

Poster

https://aclanthology.org/2024.acl-long.273/

https://aclanthology.org/2024.acl-long.274.bib

@inproceedings{huang-etal-2024-queryagent, title = "{Q}uery{A}gent: A Reliable and Efficient Reasoning Framework with Environmental Feedback based Self-Correction", author = "Huang, Xiang and Cheng, Sitao and Huang, Shanshan and Shen, Jiayu and Xu, Yong and Zhang, Chaoyun and Qu, Yuzhong", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.274", pages = "5014--5035", abstract = "Employing Large Language Models (LLMs) for semantic parsing has achieved remarkable success. However, we find existing methods fall short in terms of reliability and efficiency when hallucinations are encountered. In this paper, we address these challenges with a framework called QueryAgent, which solves a question step-by-step and performs stepwise self-correction. We introduce an environmental feedback-based self-correction method called ERASER. Unlike traditional approaches, ERASER leverages rich environmental feedback in the intermediate steps to perform selective and differentiated self-correction only when necessary. Experimental results demonstrate that QueryAgent notably outperforms all previous few-shot methods using only one example on GrailQA and GraphQ by 5.7 and 15.0 points. Furthermore, our approach exhibits superiority in terms of efficiency, including run-time, query overhead, and API invocation costs. By leveraging ERASER, we further improve another baseline (i.e., AgentBench) by approximately 10 points, validating the strong transferability of our approach.", }

Employing Large Language Models (LLMs) for semantic parsing has achieved remarkable success. However, we find existing methods fall short in terms of reliability and efficiency when hallucinations are encountered. In this paper, we address these challenges with a framework called QueryAgent, which solves a question step-by-step and performs stepwise self-correction. We introduce an environmental feedback-based self-correction method called ERASER. Unlike traditional approaches, ERASER leverages rich environmental feedback in the intermediate steps to perform selective and differentiated self-correction only when necessary. Experimental results demonstrate that QueryAgent notably outperforms all previous few-shot methods using only one example on GrailQA and GraphQ by 5.7 and 15.0 points. Furthermore, our approach exhibits superiority in terms of efficiency, including run-time, query overhead, and API invocation costs. By leveraging ERASER, we further improve another baseline (i.e., AgentBench) by approximately 10 points, validating the strong transferability of our approach.

QueryAgent: A Reliable and Efficient Reasoning Framework with Environmental Feedback based Self-Correction

acl-long.274

Oral

https://aclanthology.org/2024.acl-long.274/

https://aclanthology.org/2024.acl-long.275.bib

@inproceedings{sun-etal-2024-pita, title = "{PITA}: Prompting Task Interaction for Argumentation Mining", author = "Sun, Yang and Wang, Muyi and Bao, Jianzhu and Liang, Bin and Zhao, Xiaoyan and Yang, Caihua and Yang, Min and Xu, Ruifeng", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.275", pages = "5036--5049", abstract = "Argumentation mining (AM) aims to detect the arguments and their inherent relations from argumentative textual compositions. Generally, AM comprises three key challenging subtasks, including argument component type classification (ACTC), argumentative relation identification (ARI), and argumentative relation type classification (ARTC). Prior methods are afflicted by a sequential feature decoding paradigm, wherein they initially address the features of argumentation components (ACs) for the task of ACTC. Then, these features are amalgamated in pairs to tackle the task of ARI. Finally, the AC pairs and ascertained pertinent relations are employed for ARTC. However, the explicit and comprehensive inter-relationship among the three subtasks is neglected. In this paper, we propose a novel method PITA for PromptIng Task interAction to model the inter-relationships among the three subtasks within a generative framework. Specifically, we employ a dynamic prompt template to indicate all ACs and AC pairs in the three subtasks. Then, from a multi-relational perspective, we construct an undirected heterogeneous graph to capture the various relationships within and between ACs and AC pairs. We apply the Relational Graph Convolutional Network (RGCN) on the graph and inject the task interaction information into the soft prompts with continuous representations. PITA jointly decodes all ACs and AC pairs using the prompt template with task interaction information, which thus explicitly and comprehensively harmonizes the information propagation across the three subtasks. Extensive experiments show PITA achieves state-of-the-art performances on two AM benchmarks.", }

Argumentation mining (AM) aims to detect the arguments and their inherent relations from argumentative textual compositions. Generally, AM comprises three key challenging subtasks, including argument component type classification (ACTC), argumentative relation identification (ARI), and argumentative relation type classification (ARTC). Prior methods are afflicted by a sequential feature decoding paradigm, wherein they initially address the features of argumentation components (ACs) for the task of ACTC. Then, these features are amalgamated in pairs to tackle the task of ARI. Finally, the AC pairs and ascertained pertinent relations are employed for ARTC. However, the explicit and comprehensive inter-relationship among the three subtasks is neglected. In this paper, we propose a novel method PITA for PromptIng Task interAction to model the inter-relationships among the three subtasks within a generative framework. Specifically, we employ a dynamic prompt template to indicate all ACs and AC pairs in the three subtasks. Then, from a multi-relational perspective, we construct an undirected heterogeneous graph to capture the various relationships within and between ACs and AC pairs. We apply the Relational Graph Convolutional Network (RGCN) on the graph and inject the task interaction information into the soft prompts with continuous representations. PITA jointly decodes all ACs and AC pairs using the prompt template with task interaction information, which thus explicitly and comprehensively harmonizes the information propagation across the three subtasks. Extensive experiments show PITA achieves state-of-the-art performances on two AM benchmarks.

PITA: Prompting Task Interaction for Argumentation Mining

acl-long.275

Poster

https://aclanthology.org/2024.acl-long.275/

https://aclanthology.org/2024.acl-long.276.bib

@inproceedings{duan-etal-2024-shifting, title = "Shifting Attention to Relevance: Towards the Predictive Uncertainty Quantification of Free-Form Large Language Models", author = "Duan, Jinhao and Cheng, Hao and Wang, Shiqi and Zavalny, Alex and Wang, Chenan and Xu, Renjing and Kailkhura, Bhavya and Xu, Kaidi", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.276", pages = "5050--5063", abstract = "Large Language Models (LLMs) show promising results in language generation and instruction following but frequently {``}hallucinate{''}, making their outputs less reliable. Despite Uncertainty Quantification{'}s (UQ) potential solutions, implementing it accurately within LLMs is challenging. Our research introduces a simple heuristic: not all tokens in auto-regressive LLM text equally represent the underlying meaning, as {``}linguistic redundancy{''} often allows a few keywords to convey the essence of long sentences. However, current methods underestimate this inequality when assessing uncertainty, causing tokens with limited semantics to be equally or excessively weighted in UQ. To correct this, we propose Shifting Attention to more Relevant (SAR) components at both token- and sentence-levels for better UQ. We conduct extensive experiments involving a range of popular {``}off-the-shelf{''} LLMs, such as Vicuna, WizardLM, and LLaMA-2-chat, with model sizes extending up to 33B parameters. We evaluate various free-form question-answering tasks, encompassing domains such as reading comprehension, science Q{\&}A, and medical Q{\&}A. Our experimental results, coupled with a comprehensive demographic analysis, demonstrate the superior performance of SAR. The code is available at https://github.com/jinhaoduan/SAR.", }

Large Language Models (LLMs) show promising results in language generation and instruction following but frequently {``}hallucinate{''}, making their outputs less reliable. Despite Uncertainty Quantification{'}s (UQ) potential solutions, implementing it accurately within LLMs is challenging. Our research introduces a simple heuristic: not all tokens in auto-regressive LLM text equally represent the underlying meaning, as {``}linguistic redundancy{''} often allows a few keywords to convey the essence of long sentences. However, current methods underestimate this inequality when assessing uncertainty, causing tokens with limited semantics to be equally or excessively weighted in UQ. To correct this, we propose Shifting Attention to more Relevant (SAR) components at both token- and sentence-levels for better UQ. We conduct extensive experiments involving a range of popular {``}off-the-shelf{''} LLMs, such as Vicuna, WizardLM, and LLaMA-2-chat, with model sizes extending up to 33B parameters. We evaluate various free-form question-answering tasks, encompassing domains such as reading comprehension, science Q{\&}A, and medical Q{\&}A. Our experimental results, coupled with a comprehensive demographic analysis, demonstrate the superior performance of SAR. The code is available at https://github.com/jinhaoduan/SAR.

Shifting Attention to Relevance: Towards the Predictive Uncertainty Quantification of Free-Form Large Language Models

acl-long.276

Poster

2307.01379

https://huggingface.co/papers/2307.01379

https://aclanthology.org/2024.acl-long.276/

https://aclanthology.org/2024.acl-long.277.bib

@inproceedings{geigle-etal-2024-babel, title = "Babel-{I}mage{N}et: Massively Multilingual Evaluation of Vision-and-Language Representations", author = "Geigle, Gregor and Timofte, Radu and Glava{\v{s}}, Goran", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.277", pages = "5064--5084", abstract = "Vision-and-language (VL) models with separate encoders for each modality (e.g., CLIP) have become the go-to models for zero-shot image classification and image-text retrieval. They are, however, mostly evaluated in English as multilingual benchmarks are limited in availability. We introduce Babel-ImageNet, a massively multilingual benchmark that offers (partial) translations of ImageNet labels to 100 languages, built without machine translation or manual annotation. We instead automatically obtain reliable translations by linking them {--} via shared WordNet synsets {--} to BabelNet, a massively multilingual lexico-semantic network. We evaluate 11 public multilingual CLIP models on zero-shot image classification (ZS-IC) on our benchmark, demonstrating a significant gap between English ImageNet performance and that of high-resource languages (e.g., German or Chinese), and an even bigger gap for low-resource languages (e.g., Sinhala or Lao). Crucially, we show that the models{'} ZS-IC performance highly correlates with their performance in image-text retrieval, validating the use of Babel-imageNet to evaluate multilingual models for the vast majority of languages without gold image-text data. Finally, we show that the performance of multilingual CLIP can be drastically improved for low-resource languages with parameter-efficient language-specific training. We make our code and data publicly available: \url{https://github.com/gregor-ge/Babel-ImageNet}", }

Vision-and-language (VL) models with separate encoders for each modality (e.g., CLIP) have become the go-to models for zero-shot image classification and image-text retrieval. They are, however, mostly evaluated in English as multilingual benchmarks are limited in availability. We introduce Babel-ImageNet, a massively multilingual benchmark that offers (partial) translations of ImageNet labels to 100 languages, built without machine translation or manual annotation. We instead automatically obtain reliable translations by linking them {--} via shared WordNet synsets {--} to BabelNet, a massively multilingual lexico-semantic network. We evaluate 11 public multilingual CLIP models on zero-shot image classification (ZS-IC) on our benchmark, demonstrating a significant gap between English ImageNet performance and that of high-resource languages (e.g., German or Chinese), and an even bigger gap for low-resource languages (e.g., Sinhala or Lao). Crucially, we show that the models{'} ZS-IC performance highly correlates with their performance in image-text retrieval, validating the use of Babel-imageNet to evaluate multilingual models for the vast majority of languages without gold image-text data. Finally, we show that the performance of multilingual CLIP can be drastically improved for low-resource languages with parameter-efficient language-specific training. We make our code and data publicly available: \url{https://github.com/gregor-ge/Babel-ImageNet}

Babel-ImageNet: Massively Multilingual Evaluation of Vision-and-Language Representations

acl-long.277

Poster

2306.08658

https://huggingface.co/papers/2306.08658

https://aclanthology.org/2024.acl-long.277/

https://aclanthology.org/2024.acl-long.278.bib

@inproceedings{li-etal-2024-estimating, title = "Estimating Agreement by Chance for Sequence Annotation", author = "Li, Diya and Rose, Carolyn and Yuan, Ao and Zhou, Chunxiao", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.278", pages = "5085--5097", abstract = "In the field of natural language processing, correction of performance assessment for chance agreement plays a crucial role in evaluating the reliability of annotations. However, there is a notable dearth of research focusing on chance correction for assessing the reliability of sequence annotation tasks, despite their widespread prevalence in the field. To address this gap, this paper introduces a novel model for generating random annotations, which serves as the foundation for estimating chance agreement in sequence annotation tasks. Utilizing the proposed randomization model and a related comparison approach, we successfully derive the analytical form of the distribution, enabling the computation of the probable location of each annotated text segment and subsequent chance agreement estimation. Through a combination simulation and corpus-based evaluation, we successfully assess its applicability and validate its accuracy and efficacy.", }

In the field of natural language processing, correction of performance assessment for chance agreement plays a crucial role in evaluating the reliability of annotations. However, there is a notable dearth of research focusing on chance correction for assessing the reliability of sequence annotation tasks, despite their widespread prevalence in the field. To address this gap, this paper introduces a novel model for generating random annotations, which serves as the foundation for estimating chance agreement in sequence annotation tasks. Utilizing the proposed randomization model and a related comparison approach, we successfully derive the analytical form of the distribution, enabling the computation of the probable location of each annotated text segment and subsequent chance agreement estimation. Through a combination simulation and corpus-based evaluation, we successfully assess its applicability and validate its accuracy and efficacy.

Estimating Agreement by Chance for Sequence Annotation

acl-long.278

Poster

2407.11371

https://aclanthology.org/2024.acl-long.278/

https://aclanthology.org/2024.acl-long.279.bib

@inproceedings{lu-etal-2024-emergent, title = "Are Emergent Abilities in Large Language Models just In-Context Learning?", author = "Lu, Sheng and Bigoulaeva, Irina and Sachdeva, Rachneet and Tayyar Madabushi, Harish and Gurevych, Iryna", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.279", pages = "5098--5139", abstract = "Large language models, comprising billions of parameters and pre-trained on extensive web-scale corpora, have been claimed to acquire certain capabilities without having been specifically trained on them. These capabilities, referred to as {``}emergent abilities,{''} have been a driving force in discussions regarding the potentials and risks of language models. A key challenge in evaluating emergent abilities is that they are confounded by model competencies that arise through alternative prompting techniques, including in-context learning, which is the ability of models to complete a task based on a few examples. We present a novel theory that explains emergent abilities, taking into account their potential confounding factors, and rigorously substantiate this theory through over 1000 experiments. Our findings suggest that purported emergent abilities are not truly emergent, but result from a combination of in-context learning, model memory, and linguistic knowledge. Our work is a foundational step in explaining language model performance, providing a template for their efficient use and clarifying the paradox of their ability to excel in some instances while faltering in others. Thus, we demonstrate that their capabilities should not be overestimated.", }

Large language models, comprising billions of parameters and pre-trained on extensive web-scale corpora, have been claimed to acquire certain capabilities without having been specifically trained on them. These capabilities, referred to as {``}emergent abilities,{''} have been a driving force in discussions regarding the potentials and risks of language models. A key challenge in evaluating emergent abilities is that they are confounded by model competencies that arise through alternative prompting techniques, including in-context learning, which is the ability of models to complete a task based on a few examples. We present a novel theory that explains emergent abilities, taking into account their potential confounding factors, and rigorously substantiate this theory through over 1000 experiments. Our findings suggest that purported emergent abilities are not truly emergent, but result from a combination of in-context learning, model memory, and linguistic knowledge. Our work is a foundational step in explaining language model performance, providing a template for their efficient use and clarifying the paradox of their ability to excel in some instances while faltering in others. Thus, we demonstrate that their capabilities should not be overestimated.

Are Emergent Abilities in Large Language Models just In-Context Learning?

acl-long.279

Poster

2309.01809

https://aclanthology.org/2024.acl-long.279/

https://aclanthology.org/2024.acl-long.280.bib

@inproceedings{yu-etal-2024-wavecoder, title = "{W}ave{C}oder: Widespread And Versatile Enhancement For Code Large Language Models By Instruction Tuning", author = "Yu, Zhaojian and Zhang, Xin and Shang, Ning and Huang, Yangyu and Xu, Can and Zhao, Yishujie and Hu, Wenxiang and Yin, Qiufeng", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.280", pages = "5140--5153", abstract = "Recent work demonstrates that, after instruction tuning, Code Large Language Models (Code LLMs) can obtain impressive capabilities to address a wide range of code-related tasks. However, current instruction tuning methods for Code LLMs mainly focus on the traditional code generation task, resulting in poor performance in complex multi-task scenarios. In this paper, we concentrate on multiple code-related tasks and present WaveCoder, a series of Code LLMs trained with Widespread And Versatile Enhanced instruction data. To enable the models to tackle complex code-related tasks, we propose a method to stably generate diverse, high-quality instruction data from open source code dataset in multi-task scenarios and obtain CodeOcean, a dataset comprising 19,915 instruction instances across 4 code-related tasks, which is aimed at improving the generalization ability of Code LLM. Our experiments demonstrate that WaveCoder models significantly outperform other open-source models in terms of the generalization ability across different code-related tasks. Moreover, WaveCoder-Ultra-6.7B presents the state-of-the-art generalization abilities on a wide range of code-related tasks.", }

Recent work demonstrates that, after instruction tuning, Code Large Language Models (Code LLMs) can obtain impressive capabilities to address a wide range of code-related tasks. However, current instruction tuning methods for Code LLMs mainly focus on the traditional code generation task, resulting in poor performance in complex multi-task scenarios. In this paper, we concentrate on multiple code-related tasks and present WaveCoder, a series of Code LLMs trained with Widespread And Versatile Enhanced instruction data. To enable the models to tackle complex code-related tasks, we propose a method to stably generate diverse, high-quality instruction data from open source code dataset in multi-task scenarios and obtain CodeOcean, a dataset comprising 19,915 instruction instances across 4 code-related tasks, which is aimed at improving the generalization ability of Code LLM. Our experiments demonstrate that WaveCoder models significantly outperform other open-source models in terms of the generalization ability across different code-related tasks. Moreover, WaveCoder-Ultra-6.7B presents the state-of-the-art generalization abilities on a wide range of code-related tasks.

WaveCoder: Widespread And Versatile Enhancement For Code Large Language Models By Instruction Tuning

acl-long.280

Poster

2312.14187

https://huggingface.co/papers/2312.14187

https://aclanthology.org/2024.acl-long.280/

https://aclanthology.org/2024.acl-long.281.bib

@inproceedings{li-etal-2024-eliciting-better, title = "Eliciting Better Multilingual Structured Reasoning from {LLM}s through Code", author = "Li, Bryan and Alkhouli, Tamer and Bonadiman, Daniele and Pappas, Nikolaos and Mansour, Saab", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.281", pages = "5154--5169", abstract = "The development of large language models (LLM) has shown progress on reasoning, though studies have largely considered either English or simple reasoning tasks. To address this, we introduce a multilingual structured reasoning and explanation dataset, termed xSTREET, that covers four tasks across six languages. xSTREET exposes a gap in base LLM performance between English and non-English reasoning tasks.We then propose two methods to remedy this gap, building on the insight that LLMs trained on code are better reasoners. First, at training time, we augment a code dataset with multilingual comments using machine translation while keeping program code as-is. Second, at inference time, we bridge the gap between training and inference by employing a prompt structure that incorporates step-by-step code primitives to derive new facts and find a solution. Our methods show improved multilingual performance on xSTREET, most notably on the scientific commonsense reasoning subtask. Furthermore, the models show no regression on non-reasoning tasks, thus demonstrating our techniques maintain general-purpose abilities.", }

The development of large language models (LLM) has shown progress on reasoning, though studies have largely considered either English or simple reasoning tasks. To address this, we introduce a multilingual structured reasoning and explanation dataset, termed xSTREET, that covers four tasks across six languages. xSTREET exposes a gap in base LLM performance between English and non-English reasoning tasks.We then propose two methods to remedy this gap, building on the insight that LLMs trained on code are better reasoners. First, at training time, we augment a code dataset with multilingual comments using machine translation while keeping program code as-is. Second, at inference time, we bridge the gap between training and inference by employing a prompt structure that incorporates step-by-step code primitives to derive new facts and find a solution. Our methods show improved multilingual performance on xSTREET, most notably on the scientific commonsense reasoning subtask. Furthermore, the models show no regression on non-reasoning tasks, thus demonstrating our techniques maintain general-purpose abilities.

Eliciting Better Multilingual Structured Reasoning from LLMs through Code

acl-long.281

Poster

2403.02567

https://huggingface.co/papers/2403.02567

https://aclanthology.org/2024.acl-long.281/

https://aclanthology.org/2024.acl-long.282.bib

@inproceedings{ossowski-hu-2024-olive, title = "{OLIVE}: Object Level In-Context Visual Embeddings", author = "Ossowski, Timothy and Hu, Junjie", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.282", pages = "5170--5185", abstract = "Recent generalist vision-language models (VLMs) have demonstrated impressive reasoning capabilities across diverse multimodal tasks. However, these models still struggle with fine-grained object-level understanding and grounding. In terms of modeling, existing VLMs implicitly align text tokens with image patch tokens, which is ineffective for embedding alignment at the same granularity and inevitably introduces noisy spurious background features. Additionally, these models struggle when generalizing to unseen visual concepts and may not be reliable for domain-specific tasks without further fine-tuning. To address these limitations, we propose a novel method to prompt large language models with in-context visual object vectors, thereby enabling controllable object-level reasoning. This eliminates the necessity of fusing a lengthy array of image patch features and significantly speeds up training. Furthermore, we propose region-level retrieval using our object representations, facilitating rapid adaptation to new objects without additional training. Our experiments reveal that our method achieves competitive referring object classification and captioning performance, while also offering zero-shot generalization and robustness to visually challenging contexts.", }

Recent generalist vision-language models (VLMs) have demonstrated impressive reasoning capabilities across diverse multimodal tasks. However, these models still struggle with fine-grained object-level understanding and grounding. In terms of modeling, existing VLMs implicitly align text tokens with image patch tokens, which is ineffective for embedding alignment at the same granularity and inevitably introduces noisy spurious background features. Additionally, these models struggle when generalizing to unseen visual concepts and may not be reliable for domain-specific tasks without further fine-tuning. To address these limitations, we propose a novel method to prompt large language models with in-context visual object vectors, thereby enabling controllable object-level reasoning. This eliminates the necessity of fusing a lengthy array of image patch features and significantly speeds up training. Furthermore, we propose region-level retrieval using our object representations, facilitating rapid adaptation to new objects without additional training. Our experiments reveal that our method achieves competitive referring object classification and captioning performance, while also offering zero-shot generalization and robustness to visually challenging contexts.

OLIVE: Object Level In-Context Visual Embeddings

acl-long.282

Poster

2406.00872

https://aclanthology.org/2024.acl-long.282/

https://aclanthology.org/2024.acl-long.283.bib

@inproceedings{chen-mueller-2024-quantifying, title = "Quantifying Uncertainty in Answers from any Language Model and Enhancing their Trustworthiness", author = "Chen, Jiuhai and Mueller, Jonas", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.283", pages = "5186--5200", abstract = "We introduce BSDetector, a method for detecting bad and speculative answers from a pretrained Large Language Model by estimating a numeric confidence score for any output it generated. Our uncertainty quantification technique works for any LLM accessible only via a black-box API, whose training data remains unknown. By expending a bit of extra computation, users of any LLM API can now get the same response as they would ordinarily, as well as a confidence estimate that cautions when not to trust this response. Experiments on both closed and open-form Question-Answer benchmarks reveal that BSDetector more accurately identifies incorrect LLM responses than alternative uncertainty estimation procedures (for both GPT-3 and ChatGPT). By sampling multiple responses from the LLM and considering the one with the highest confidence score, we can additionally obtain more accurate responses from the same LLM, without extra training steps. In applications involving automated evaluation with LLMs, accounting for our confidence scores leads to more reliable evaluation in both human-in-the-loop and fully-automated settings (across both GPT 3.5 and 4).", }

We introduce BSDetector, a method for detecting bad and speculative answers from a pretrained Large Language Model by estimating a numeric confidence score for any output it generated. Our uncertainty quantification technique works for any LLM accessible only via a black-box API, whose training data remains unknown. By expending a bit of extra computation, users of any LLM API can now get the same response as they would ordinarily, as well as a confidence estimate that cautions when not to trust this response. Experiments on both closed and open-form Question-Answer benchmarks reveal that BSDetector more accurately identifies incorrect LLM responses than alternative uncertainty estimation procedures (for both GPT-3 and ChatGPT). By sampling multiple responses from the LLM and considering the one with the highest confidence score, we can additionally obtain more accurate responses from the same LLM, without extra training steps. In applications involving automated evaluation with LLMs, accounting for our confidence scores leads to more reliable evaluation in both human-in-the-loop and fully-automated settings (across both GPT 3.5 and 4).

Quantifying Uncertainty in Answers from any Language Model and Enhancing their Trustworthiness

acl-long.283

Poster

2308.16175

https://huggingface.co/papers/2308.16175

https://aclanthology.org/2024.acl-long.283/

https://aclanthology.org/2024.acl-long.284.bib

@inproceedings{zhang-etal-2024-marathon, title = "Marathon: A Race Through the Realm of Long Context with Large Language Models", author = "Zhang, Lei and Li, Yunshui and Liu, Ziqiang and Yang, Jiaxi and Liu, Junhao and Chen, Longze and Luo, Run and Yang, Min", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.284", pages = "5201--5217", abstract = "With the advancement of large language models (LLMs) and the expansion of their context windows, existing long-context benchmarks fall short in effectively evaluating the models{'} comprehension and reasoning abilities in extended texts. Moreover, conventional benchmarks relying on F1 metrics often inaccurately score responses: they may undervalue correct answers that differ from the reference responses and overvalue incorrect ones that resemble the reference texts. In response to these limitations, we introduce Marathon, a novel evaluation benchmark that adopts a multiple-choice question format. It is specifically designed to overcome the constraints of previous benchmarks and provide a rapid, precise, and unbiased appraisal of the long-context comprehension skills of large language models. We conducted comprehensive evaluations on the Marathon benchmark with a range of state-of-the-art LLMs and assessed the effectiveness of various optimization strategies tailored for long-context generation. We anticipate that the Marathon benchmark and its associated leaderboard will enable a more precise and equitable evaluation of LLMs{'} capabilities in understanding and reasoning over extended contexts.", }

With the advancement of large language models (LLMs) and the expansion of their context windows, existing long-context benchmarks fall short in effectively evaluating the models{'} comprehension and reasoning abilities in extended texts. Moreover, conventional benchmarks relying on F1 metrics often inaccurately score responses: they may undervalue correct answers that differ from the reference responses and overvalue incorrect ones that resemble the reference texts. In response to these limitations, we introduce Marathon, a novel evaluation benchmark that adopts a multiple-choice question format. It is specifically designed to overcome the constraints of previous benchmarks and provide a rapid, precise, and unbiased appraisal of the long-context comprehension skills of large language models. We conducted comprehensive evaluations on the Marathon benchmark with a range of state-of-the-art LLMs and assessed the effectiveness of various optimization strategies tailored for long-context generation. We anticipate that the Marathon benchmark and its associated leaderboard will enable a more precise and equitable evaluation of LLMs{'} capabilities in understanding and reasoning over extended contexts.

Marathon: A Race Through the Realm of Long Context with Large Language Models

acl-long.284

Poster

2312.09542

https://huggingface.co/papers/2312.09542

https://aclanthology.org/2024.acl-long.284/

https://aclanthology.org/2024.acl-long.285.bib

@inproceedings{gao-etal-2024-beyond, title = "Beyond Scaling: Predicting Patent Approval with Domain-specific Fine-grained Claim Dependency Graph", author = "Gao, Xiaochen and Yao, Feng and Zhao, Kewen and He, Beilei and Kumar, Animesh and Krishnan, Vish and Shang, Jingbo", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.285", pages = "5218--5234", abstract = "Model scaling is becoming the default choice for many language tasks due to the success of large language models (LLMs). However, it can fall short in specific scenarios where simple customized methods excel. In this paper, we delve into the patent approval prediction task and unveil that simple domain-specific graph methods outperform enlarging the model, using the intrinsic dependencies within the patent data. Specifically, we first extend the embedding-based state-of-the-art (SOTA) by scaling up its backbone model with various sizes of open-source LLMs, then explore prompt-based methods to harness proprietary LLMs{'} potential, but find the best results close to random guessing, underlining the ineffectiveness of model scaling-up. Hence, we propose a novel Fine-grained cLAim depeNdency (FLAN) Graph through meticulous patent data analyses, capturing the inherent dependencies across segments of the patent text. As it is model-agnostic, we apply cost-effective graph models to our FLAN Graph to obtain representations for approval prediction. Extensive experiments and detailed analyses prove that incorporating FLAN Graph via various graph models consistently outperforms all LLM baselines significantly. We hope that our observations and analyses in this paper can bring more attention to this challenging task and prompt further research into the limitations of LLMs.", }

Model scaling is becoming the default choice for many language tasks due to the success of large language models (LLMs). However, it can fall short in specific scenarios where simple customized methods excel. In this paper, we delve into the patent approval prediction task and unveil that simple domain-specific graph methods outperform enlarging the model, using the intrinsic dependencies within the patent data. Specifically, we first extend the embedding-based state-of-the-art (SOTA) by scaling up its backbone model with various sizes of open-source LLMs, then explore prompt-based methods to harness proprietary LLMs{'} potential, but find the best results close to random guessing, underlining the ineffectiveness of model scaling-up. Hence, we propose a novel Fine-grained cLAim depeNdency (FLAN) Graph through meticulous patent data analyses, capturing the inherent dependencies across segments of the patent text. As it is model-agnostic, we apply cost-effective graph models to our FLAN Graph to obtain representations for approval prediction. Extensive experiments and detailed analyses prove that incorporating FLAN Graph via various graph models consistently outperforms all LLM baselines significantly. We hope that our observations and analyses in this paper can bring more attention to this challenging task and prompt further research into the limitations of LLMs.

Beyond Scaling: Predicting Patent Approval with Domain-specific Fine-grained Claim Dependency Graph

acl-long.285

Oral

2404.14372

https://aclanthology.org/2024.acl-long.285/

https://aclanthology.org/2024.acl-long.286.bib

@inproceedings{zhuang-etal-2024-pcad, title = "{PCAD}: Towards {ASR}-Robust Spoken Language Understanding via Prototype Calibration and Asymmetric Decoupling", author = "Zhuang, Xianwei and Cheng, Xuxin and Liang, Liming and Xie, Yuxin and Wang, Zhichang and Huang, Zhiqi and Zou, Yuexian", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.286", pages = "5235--5246", abstract = "Spoken language understanding (SLU) inevitably suffers from error propagation from automatic speech recognition (ASR) in actual scenarios. Some recent works attempt to alleviate this issue through contrastive learning. However, they (1) sample negative pairs incorrectly in pre-training; (2) only focus on implicit metric learning while neglecting explicit erroneous predictions; (3) treat manual and ASR transcripts indiscriminately. In this paper, we propose a novel framework termed $\textbf{PCAD}$, which can calibrate bias and errors and achieve adaptive-balanced decoupling training. Specifically, PCAD utilizes a prototype-based loss to aggregate label and prediction priors and calibrate bias and error-prone semantics for better inter-class discrimination and intra-class consistency. We theoretically analyze the effect of this loss on robustness enhancement. Further, we leverage a teacher-student model for asymmetric decoupling training between different transcripts and formulate a novel gradient-sensitive exponential moving averaging (GS-EMA) algorithm for adaptive balance of accuracy and robustness. Experiments on three datasets show that PCAD significantly outperforms existing approaches and achieves new state-of-the-art performance.", }

Spoken language understanding (SLU) inevitably suffers from error propagation from automatic speech recognition (ASR) in actual scenarios. Some recent works attempt to alleviate this issue through contrastive learning. However, they (1) sample negative pairs incorrectly in pre-training; (2) only focus on implicit metric learning while neglecting explicit erroneous predictions; (3) treat manual and ASR transcripts indiscriminately. In this paper, we propose a novel framework termed $\textbf{PCAD}$, which can calibrate bias and errors and achieve adaptive-balanced decoupling training. Specifically, PCAD utilizes a prototype-based loss to aggregate label and prediction priors and calibrate bias and error-prone semantics for better inter-class discrimination and intra-class consistency. We theoretically analyze the effect of this loss on robustness enhancement. Further, we leverage a teacher-student model for asymmetric decoupling training between different transcripts and formulate a novel gradient-sensitive exponential moving averaging (GS-EMA) algorithm for adaptive balance of accuracy and robustness. Experiments on three datasets show that PCAD significantly outperforms existing approaches and achieves new state-of-the-art performance.

PCAD: Towards ASR-Robust Spoken Language Understanding via Prototype Calibration and Asymmetric Decoupling

acl-long.286

Poster

https://aclanthology.org/2024.acl-long.286/

https://aclanthology.org/2024.acl-long.287.bib

@inproceedings{jin-etal-2024-rethinking, title = "Rethinking the Multimodal Correlation of Multimodal Sequential Learning via Generalizable Attentional Results Alignment", author = "Jin, Tao and Lin, Wang and Wang, Ye and Li, Linjun and Cheng, Xize and Zhao, Zhou", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.287", pages = "5247--5265", abstract = "Transformer-based methods have gone mainstream in multimodal sequential learning. The intra and inter modality interactions are captured by the query-key associations of multi-head attention. In this way, the calculated multimodal contexts (attentional results) are expected to be relevant to the query modality. However, in existing literature, the alignment degree between different calculated attentional results of the same query are under-explored. Based on this concern, we propose a new constrained scheme called Multimodal Contextual Contrast (MCC), which could align the multiple attentional results from both local and global perspectives, making the information capture more efficient. Concretely, the calculated attentional results of different modalities are mapped into a common feature space, those attentional vectors with the same query are considered as a positive group and the remaining sets are negative. From local perspective, we sample the negative groups for a positive group by randomly changing the sequential step of one specific context and keeping the other stay the same. From coarse global perspective, we divide all the contextual groups into two sets (i.e., aligned and unaligned), making the total score of aligned group relatively large. We extend the vectorial inner product operation for more input and calculate the aligned score for each multimodal group. Considering that the computational complexity scales exponentially to the number of modalities, we adopt stochastic expectation approximation (SEA) for the real process. The extensive experimental results on several tasks reveal the effectiveness of our contributions.", }

Transformer-based methods have gone mainstream in multimodal sequential learning. The intra and inter modality interactions are captured by the query-key associations of multi-head attention. In this way, the calculated multimodal contexts (attentional results) are expected to be relevant to the query modality. However, in existing literature, the alignment degree between different calculated attentional results of the same query are under-explored. Based on this concern, we propose a new constrained scheme called Multimodal Contextual Contrast (MCC), which could align the multiple attentional results from both local and global perspectives, making the information capture more efficient. Concretely, the calculated attentional results of different modalities are mapped into a common feature space, those attentional vectors with the same query are considered as a positive group and the remaining sets are negative. From local perspective, we sample the negative groups for a positive group by randomly changing the sequential step of one specific context and keeping the other stay the same. From coarse global perspective, we divide all the contextual groups into two sets (i.e., aligned and unaligned), making the total score of aligned group relatively large. We extend the vectorial inner product operation for more input and calculate the aligned score for each multimodal group. Considering that the computational complexity scales exponentially to the number of modalities, we adopt stochastic expectation approximation (SEA) for the real process. The extensive experimental results on several tasks reveal the effectiveness of our contributions.

Rethinking the Multimodal Correlation of Multimodal Sequential Learning via Generalizable Attentional Results Alignment

acl-long.287

Poster

https://aclanthology.org/2024.acl-long.287/

https://aclanthology.org/2024.acl-long.288.bib

@inproceedings{liang-etal-2024-uhgeval, title = "{UHGE}val: Benchmarking the Hallucination of {C}hinese Large Language Models via Unconstrained Generation", author = "Liang, Xun and Song, Shichao and Niu, Simin and Li, Zhiyu and Xiong, Feiyu and Tang, Bo and Wang, Yezhaohui and He, Dawei and Peng, Cheng and Wang, Zhonghao and Deng, Haiying", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.288", pages = "5266--5293", abstract = "Large language models (LLMs) produce hallucinated text, compromising their practical utility in professional contexts. To assess the reliability of LLMs, numerous initiatives have developed benchmark evaluations for hallucination phenomena. However, they often employ constrained generation techniques to produce the evaluation dataset due to cost and time limitations. For instance, this may involve employing directed hallucination induction or deliberately modifying authentic text to generate hallucinations. These are not congruent with the unrestricted text generation demanded by real-world applications. Furthermore, a well-established Chinese-language dataset dedicated to the evaluation of hallucinations is presently lacking. Consequently, we have developed an Unconstrained Hallucination Generation Evaluation (UHGEval) benchmark, containing hallucinations generated by LLMs with minimal restrictions. Concurrently, we have established a comprehensive benchmark evaluation framework to aid subsequent researchers in undertaking scalable and reproducible experiments. We have also evaluated prominent Chinese LLMs and the GPT series models to derive insights regarding hallucination.", }

Large language models (LLMs) produce hallucinated text, compromising their practical utility in professional contexts. To assess the reliability of LLMs, numerous initiatives have developed benchmark evaluations for hallucination phenomena. However, they often employ constrained generation techniques to produce the evaluation dataset due to cost and time limitations. For instance, this may involve employing directed hallucination induction or deliberately modifying authentic text to generate hallucinations. These are not congruent with the unrestricted text generation demanded by real-world applications. Furthermore, a well-established Chinese-language dataset dedicated to the evaluation of hallucinations is presently lacking. Consequently, we have developed an Unconstrained Hallucination Generation Evaluation (UHGEval) benchmark, containing hallucinations generated by LLMs with minimal restrictions. Concurrently, we have established a comprehensive benchmark evaluation framework to aid subsequent researchers in undertaking scalable and reproducible experiments. We have also evaluated prominent Chinese LLMs and the GPT series models to derive insights regarding hallucination.

UHGEval: Benchmarking the Hallucination of Chinese Large Language Models via Unconstrained Generation

acl-long.288

Poster

2311.15296

https://huggingface.co/papers/2311.15296

https://aclanthology.org/2024.acl-long.288/

https://aclanthology.org/2024.acl-long.289.bib

@inproceedings{lin-etal-2024-preflmr, title = "{P}re{FLMR}: Scaling Up Fine-Grained Late-Interaction Multi-modal Retrievers", author = "Lin, Weizhe and Mei, Jingbiao and Chen, Jinghong and Byrne, Bill", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.289", pages = "5294--5316", abstract = "Large Multimodal Models (LMMs) excel in natural language and visual understanding but are challenged by exacting tasks such as Knowledge-based Visual Question Answering (KB-VQA) which involve the retrieval of relevant information from document collections to use in shaping answers to questions. We present an extensive training and evaluation framework, M2KR, for KB-VQA. M2KR contains a collection of vision and language tasks which we have incorporated into a single suite of benchmark tasks for training and evaluating general-purpose multi-modal retrievers. We use M2KR to develop PreFLMR, a pre-trained version of the recently developed Fine-grained Late-interaction Multi-modal Retriever (FLMR) approach to KB-VQA, and we report new state-of-the-art results across a range of tasks. We also present investigations into the scaling behaviors of PreFLMR intended to be useful in future developments in general-purpose multi-modal retrievers.", }

Large Multimodal Models (LMMs) excel in natural language and visual understanding but are challenged by exacting tasks such as Knowledge-based Visual Question Answering (KB-VQA) which involve the retrieval of relevant information from document collections to use in shaping answers to questions. We present an extensive training and evaluation framework, M2KR, for KB-VQA. M2KR contains a collection of vision and language tasks which we have incorporated into a single suite of benchmark tasks for training and evaluating general-purpose multi-modal retrievers. We use M2KR to develop PreFLMR, a pre-trained version of the recently developed Fine-grained Late-interaction Multi-modal Retriever (FLMR) approach to KB-VQA, and we report new state-of-the-art results across a range of tasks. We also present investigations into the scaling behaviors of PreFLMR intended to be useful in future developments in general-purpose multi-modal retrievers.

PreFLMR: Scaling Up Fine-Grained Late-Interaction Multi-modal Retrievers

acl-long.289

Poster

2402.08327

https://huggingface.co/papers/2402.08327

https://aclanthology.org/2024.acl-long.289/

https://aclanthology.org/2024.acl-long.290.bib

@inproceedings{erker-etal-2024-triple, title = "Triple-Encoders: Representations That Fire Together, Wire Together", author = "Erker, Justus-Jonas and Mai, Florian and Reimers, Nils and Spanakis, Gerasimos and Gurevych, Iryna", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.290", pages = "5317--5332", abstract = "Search-based dialog models typically re-encode the dialog history at every turn, incurring high cost.Curved Contrastive Learning, a representation learning method that encodes relative distances between utterances into the embedding space via a bi-encoder, has recently shown promising results for dialog modeling at far superior efficiency.While high efficiency is achieved through independently encoding utterances, this ignores the importance of contextualization. To overcome this issue, this study introduces triple-encoders, which efficiently compute distributed utterance mixtures from these independently encoded utterances through a novel hebbian inspired co-occurrence learning objective in a self-organizing manner, without using any weights, i.e., merely through local interactions. Empirically, we find that triple-encoders lead to a substantial improvement over bi-encoders, and even to better zero-shot generalization than single-vector representation models without requiring re-encoding. Our code (https://github.com/UKPLab/acl2024-triple-encoders) and model (https://huggingface.co/UKPLab/triple-encoders-dailydialog) are publicly available.", }

Search-based dialog models typically re-encode the dialog history at every turn, incurring high cost.Curved Contrastive Learning, a representation learning method that encodes relative distances between utterances into the embedding space via a bi-encoder, has recently shown promising results for dialog modeling at far superior efficiency.While high efficiency is achieved through independently encoding utterances, this ignores the importance of contextualization. To overcome this issue, this study introduces triple-encoders, which efficiently compute distributed utterance mixtures from these independently encoded utterances through a novel hebbian inspired co-occurrence learning objective in a self-organizing manner, without using any weights, i.e., merely through local interactions. Empirically, we find that triple-encoders lead to a substantial improvement over bi-encoders, and even to better zero-shot generalization than single-vector representation models without requiring re-encoding. Our code (https://github.com/UKPLab/acl2024-triple-encoders) and model (https://huggingface.co/UKPLab/triple-encoders-dailydialog) are publicly available.

Triple-Encoders: Representations That Fire Together, Wire Together

acl-long.290

Poster

2402.12332

https://huggingface.co/papers/2402.12332

https://aclanthology.org/2024.acl-long.290/

https://aclanthology.org/2024.acl-long.291.bib

@inproceedings{mei-etal-2024-improving, title = "Improving Hateful Meme Detection through Retrieval-Guided Contrastive Learning", author = "Mei, Jingbiao and Chen, Jinghong and Lin, Weizhe and Byrne, Bill and Tomalin, Marcus", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.291", pages = "5333--5347", abstract = "Hateful memes have emerged as a significant concern on the Internet. Detecting hateful memes requires the system to jointly understand the visual and textual modalities. Our investigation reveals that the embedding space of existing CLIP-based systems lacks sensitivity to subtle differences in memes that are vital for correct hatefulness classification. We propose constructing a hatefulness-aware embedding space through retrieval-guided contrastive training. Our approach achieves state-of-the-art performance on the HatefulMemes dataset with an AUROC of 87.0, outperforming much larger fine-tuned large multimodal models. We demonstrate a retrieval-based hateful memes detection system, which is capable of identifying hatefulness based on data unseen in training. This allows developers to update the hateful memes detection system by simply adding new examples without retraining {---} a desirable feature for real services in the constantly evolving landscape of hateful memes on the Internet.", }

Hateful memes have emerged as a significant concern on the Internet. Detecting hateful memes requires the system to jointly understand the visual and textual modalities. Our investigation reveals that the embedding space of existing CLIP-based systems lacks sensitivity to subtle differences in memes that are vital for correct hatefulness classification. We propose constructing a hatefulness-aware embedding space through retrieval-guided contrastive training. Our approach achieves state-of-the-art performance on the HatefulMemes dataset with an AUROC of 87.0, outperforming much larger fine-tuned large multimodal models. We demonstrate a retrieval-based hateful memes detection system, which is capable of identifying hatefulness based on data unseen in training. This allows developers to update the hateful memes detection system by simply adding new examples without retraining {---} a desirable feature for real services in the constantly evolving landscape of hateful memes on the Internet.

Improving Hateful Meme Detection through Retrieval-Guided Contrastive Learning

acl-long.291

Poster

2311.08110

https://aclanthology.org/2024.acl-long.291/

https://aclanthology.org/2024.acl-long.292.bib

@inproceedings{zhang-etal-2024-agent, title = "Agent-Pro: Learning to Evolve via Policy-Level Reflection and Optimization", author = "Zhang, Wenqi and Tang, Ke and Wu, Hai and Wang, Mengna and Shen, Yongliang and Hou, Guiyang and Tan, Zeqi and Li, Peng and Zhuang, Yueting and Lu, Weiming", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.292", pages = "5348--5375", abstract = "Large Language Models (LLMs) exhibit robust problem-solving capabilities for diverse tasks. However, most LLM-based agents are designed as specific task solvers with sophisticated prompt engineering, rather than agents capable of learning and evolving through interactions. These task solvers necessitate manually crafted prompts to inform task rules and regulate LLM behaviors, inherently incapacitating to address complex dynamic scenarios e.g., large interactive games. In light of this, we propose Agent-Pro: an LLM-based Agent with Policy-level Reflection and Optimization that can learn a wealth of expertise from interactive experiences and progressively elevate its behavioral policy. Specifically, it involves a dynamic belief generation and reflection process for policy evolution. Rather than action-level reflection, Agent-Pro iteratively reflects on past trajectories and beliefs, {``}fine-tuning{''} its irrational beliefs for a better policy. Moreover, a depth-first search is employed for policy optimization, ensuring continual enhancement in policy payoffs. Agent-Pro is evaluated across two games: Blackjack and Texas Hold{'}em, outperforming vanilla LLM and specialized models. Our results show Agent-Pro can learn and evolve in complex and dynamic scenes, which also benefits numerous LLM-based applications.", }

Large Language Models (LLMs) exhibit robust problem-solving capabilities for diverse tasks. However, most LLM-based agents are designed as specific task solvers with sophisticated prompt engineering, rather than agents capable of learning and evolving through interactions. These task solvers necessitate manually crafted prompts to inform task rules and regulate LLM behaviors, inherently incapacitating to address complex dynamic scenarios e.g., large interactive games. In light of this, we propose Agent-Pro: an LLM-based Agent with Policy-level Reflection and Optimization that can learn a wealth of expertise from interactive experiences and progressively elevate its behavioral policy. Specifically, it involves a dynamic belief generation and reflection process for policy evolution. Rather than action-level reflection, Agent-Pro iteratively reflects on past trajectories and beliefs, {``}fine-tuning{''} its irrational beliefs for a better policy. Moreover, a depth-first search is employed for policy optimization, ensuring continual enhancement in policy payoffs. Agent-Pro is evaluated across two games: Blackjack and Texas Hold{'}em, outperforming vanilla LLM and specialized models. Our results show Agent-Pro can learn and evolve in complex and dynamic scenes, which also benefits numerous LLM-based applications.

Agent-Pro: Learning to Evolve via Policy-Level Reflection and Optimization

acl-long.292

Poster

2402.17574

https://aclanthology.org/2024.acl-long.292/

https://aclanthology.org/2024.acl-long.293.bib

@inproceedings{razzhigaev-etal-2024-transformer, title = "Your Transformer is Secretly Linear", author = "Razzhigaev, Anton and Mikhalchuk, Matvey and Goncharova, Elizaveta and Gerasimenko, Nikolai and Oseledets, Ivan and Dimitrov, Denis and Kuznetsov, Andrey", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.293", pages = "5376--5384", abstract = "This paper reveals a novel linear characteristic exclusive to transformer decoders, including models like GPT, LLaMA, OPT, BLOOM and others. We analyze embedding transformations between sequential layers, uncovering an almost perfect linear relationship (Procrustes similarity score of 0.99). However, linearity decreases when the residual component is removed, due to a consistently low transformer layer output norm. Our experiments show that pruning or linearly approximating some of the layers does not impact loss or model performance significantly. Moreover, we introduce a cosine-similarity-based regularization in our pretraining experiments on smaller models, aimed at reducing layer linearity. This regularization not only improves performance metrics on benchmarks like Tiny Stories and SuperGLUE but as well successfully decreases the linearity of the models. This study challenges the existing understanding of transformer architectures, suggesting that their operation may be more linear than previously assumed.", }

This paper reveals a novel linear characteristic exclusive to transformer decoders, including models like GPT, LLaMA, OPT, BLOOM and others. We analyze embedding transformations between sequential layers, uncovering an almost perfect linear relationship (Procrustes similarity score of 0.99). However, linearity decreases when the residual component is removed, due to a consistently low transformer layer output norm. Our experiments show that pruning or linearly approximating some of the layers does not impact loss or model performance significantly. Moreover, we introduce a cosine-similarity-based regularization in our pretraining experiments on smaller models, aimed at reducing layer linearity. This regularization not only improves performance metrics on benchmarks like Tiny Stories and SuperGLUE but as well successfully decreases the linearity of the models. This study challenges the existing understanding of transformer architectures, suggesting that their operation may be more linear than previously assumed.

Your Transformer is Secretly Linear

acl-long.293

Poster

2405.12250

https://huggingface.co/papers/2405.12250

https://aclanthology.org/2024.acl-long.293/

https://aclanthology.org/2024.acl-long.294.bib

@inproceedings{jinadu-ding-2024-noise, title = "Noise Correction on Subjective Datasets", author = "Jinadu, Uthman and Ding, Yi", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.294", pages = "5385--5395", abstract = "Incorporating every annotator{'}s perspective is crucial for unbiased data modeling. Annotator fatigue and changing opinions over time can distort dataset annotations. To combat this, we propose to learn a more accurate representation of diverse opinions by utilizing multitask learning in conjunction with loss-based label correction. We show that using our novel formulation, we can cleanly separate agreeing and disagreeing annotations. Furthermore, this method provides a controllable way to encourage or discourage disagreement. We demonstrate that this modification can improve prediction performance in a single or multi-annotator setting. Lastly, we show that this method remains robust to additional label noise that is applied to subjective data.", }

Incorporating every annotator{'}s perspective is crucial for unbiased data modeling. Annotator fatigue and changing opinions over time can distort dataset annotations. To combat this, we propose to learn a more accurate representation of diverse opinions by utilizing multitask learning in conjunction with loss-based label correction. We show that using our novel formulation, we can cleanly separate agreeing and disagreeing annotations. Furthermore, this method provides a controllable way to encourage or discourage disagreement. We demonstrate that this modification can improve prediction performance in a single or multi-annotator setting. Lastly, we show that this method remains robust to additional label noise that is applied to subjective data.

Noise Correction on Subjective Datasets

acl-long.294

Poster

2311.00619

https://aclanthology.org/2024.acl-long.294/

https://aclanthology.org/2024.acl-long.295.bib

@inproceedings{senel-etal-2024-generative, title = "Generative Explore-Exploit: Training-free Optimization of Generative Recommender Systems using {LLM} Optimizers", author = {Senel, L{\"u}tfi Kerem and Fetahu, Besnik and Yoshida, Davis and Chen, Zhiyu and Castellucci, Giuseppe and Vedula, Nikhita and Choi, Jason Ingyu and Malmasi, Shervin}, editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.295", pages = "5396--5420", abstract = "Recommender systems are widely used to suggest engaging content, and Large Language Models (LLMs) have given rise to generative recommenders. Such systems can directly generate items, including for open-set tasks like question suggestion. While the world knowledge of LLMs enables good recommendations, improving the generated content through user feedback is challenging as continuously fine-tuning LLMs is prohibitively expensive. We present a training-free approach for optimizing generative recommenders by connecting user feedback loops to LLM-based optimizers. We propose a generative explore-exploit method that can not only exploit generated items with known high engagement, but also actively explore and discover hidden population preferences to improve recommendation quality. We evaluate our approach on question generation in two domains (e-commerce and general knowledge), and model user feedback with Click Through Rate (CTR). Experiments show our LLM-based explore-exploit approach can iteratively improve recommendations and consistently increase CTR. Ablation analysis shows that generative exploration is key to learning user preferences, avoiding the pitfalls of greedy exploit-only approaches. A human evaluation strongly supports our quantitative findings.", }

Recommender systems are widely used to suggest engaging content, and Large Language Models (LLMs) have given rise to generative recommenders. Such systems can directly generate items, including for open-set tasks like question suggestion. While the world knowledge of LLMs enables good recommendations, improving the generated content through user feedback is challenging as continuously fine-tuning LLMs is prohibitively expensive. We present a training-free approach for optimizing generative recommenders by connecting user feedback loops to LLM-based optimizers. We propose a generative explore-exploit method that can not only exploit generated items with known high engagement, but also actively explore and discover hidden population preferences to improve recommendation quality. We evaluate our approach on question generation in two domains (e-commerce and general knowledge), and model user feedback with Click Through Rate (CTR). Experiments show our LLM-based explore-exploit approach can iteratively improve recommendations and consistently increase CTR. Ablation analysis shows that generative exploration is key to learning user preferences, avoiding the pitfalls of greedy exploit-only approaches. A human evaluation strongly supports our quantitative findings.

Generative Explore-Exploit: Training-free Optimization of Generative Recommender Systems using LLM Optimizers

acl-long.295

Poster

2406.05255

https://aclanthology.org/2024.acl-long.295/

https://aclanthology.org/2024.acl-long.296.bib

@inproceedings{jiang-etal-2024-instruction, title = "Instruction-tuned Language Models are Better Knowledge Learners", author = "Jiang, Zhengbao and Sun, Zhiqing and Shi, Weijia and Rodriguez, Pedro and Zhou, Chunting and Neubig, Graham and Lin, Xi and Yih, Wen-tau and Iyer, Srini", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.296", pages = "5421--5434", abstract = "In order for large language model (LLM)-based assistants to effectively adapt to evolving information needs, it must be possible to update their factual knowledge through continued training on new data. The standard recipe for doing so involves continued pre-training on new documents followed by instruction-tuning on question-answer (QA) pairs. However, we find that LLMs trained with this recipe struggle to answer questions, even though the perplexity of documents is minimized. We found that QA pairs are generally straightforward, while documents are more complex, weaving many factual statements together in an intricate manner. Therefore, we hypothesize that it is beneficial to expose LLMs to QA pairs before continued pre-training on documents so that the process of encoding knowledge from complex documents takes into account how this knowledge is accessed through questions. Based on this, we propose pre-instruction-tuning (PIT), a method that instruction-tunes on questions prior to training on documents. This contrasts with standard instruction-tuning, which learns how to extract knowledge after training on documents. Extensive experiments and ablation studies demonstrate that pre-instruction-tuning significantly enhances the ability of LLMs to absorb knowledge from new documents, outperforming standard instruction-tuning by 17.8{\%}.", }

In order for large language model (LLM)-based assistants to effectively adapt to evolving information needs, it must be possible to update their factual knowledge through continued training on new data. The standard recipe for doing so involves continued pre-training on new documents followed by instruction-tuning on question-answer (QA) pairs. However, we find that LLMs trained with this recipe struggle to answer questions, even though the perplexity of documents is minimized. We found that QA pairs are generally straightforward, while documents are more complex, weaving many factual statements together in an intricate manner. Therefore, we hypothesize that it is beneficial to expose LLMs to QA pairs before continued pre-training on documents so that the process of encoding knowledge from complex documents takes into account how this knowledge is accessed through questions. Based on this, we propose pre-instruction-tuning (PIT), a method that instruction-tunes on questions prior to training on documents. This contrasts with standard instruction-tuning, which learns how to extract knowledge after training on documents. Extensive experiments and ablation studies demonstrate that pre-instruction-tuning significantly enhances the ability of LLMs to absorb knowledge from new documents, outperforming standard instruction-tuning by 17.8{\%}.

Instruction-tuned Language Models are Better Knowledge Learners

acl-long.296

Poster

2402.12847

https://huggingface.co/papers/2402.12847

https://aclanthology.org/2024.acl-long.296/

https://aclanthology.org/2024.acl-long.297.bib

@inproceedings{ngo-kim-2024-language, title = "What Do Language Models Hear? Probing for Auditory Representations in Language Models", author = "Ngo, Jerry and Kim, Yoon", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.297", pages = "5435--5448", abstract = "This work explores whether language models encode meaningfully grounded representations of sounds of objects. We learn a linear probe that retrieves the correct text representation of an object given a snippet of audio related to that object, where the sound representation is given by a pretrained audio model. This probe is trained via a contrastive loss that pushes the language representations and sound representations of an object to be close to one another. After training, the probe is tested on its ability to generalize to objects that were not seen during training. Across different language models and audio models, we find that the probe generalization is above chance in many cases, indicating that despite being trained only on raw text, language models encode grounded knowledge of sounds for some objects.", }

This work explores whether language models encode meaningfully grounded representations of sounds of objects. We learn a linear probe that retrieves the correct text representation of an object given a snippet of audio related to that object, where the sound representation is given by a pretrained audio model. This probe is trained via a contrastive loss that pushes the language representations and sound representations of an object to be close to one another. After training, the probe is tested on its ability to generalize to objects that were not seen during training. Across different language models and audio models, we find that the probe generalization is above chance in many cases, indicating that despite being trained only on raw text, language models encode grounded knowledge of sounds for some objects.

What Do Language Models Hear? Probing for Auditory Representations in Language Models

acl-long.297

Poster

2402.16998

https://huggingface.co/papers/2402.16998

https://aclanthology.org/2024.acl-long.297/

https://aclanthology.org/2024.acl-long.298.bib

@inproceedings{kim-etal-2024-threads, title = "Threads of Subtlety: Detecting Machine-Generated Texts Through Discourse Motifs", author = "Kim, Zae Myung and Lee, Kwang and Zhu, Preston and Raheja, Vipul and Kang, Dongyeop", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.298", pages = "5449--5474", abstract = "With the advent of large language models (LLM), the line between human-crafted and machine-generated texts has become increasingly blurred. This paper delves into the inquiry of identifying discernible and unique linguistic properties in texts that were written by humans, particularly uncovering the underlying discourse structures of texts beyond their surface structures. Introducing a novel methodology, we leverage hierarchical parse trees and recursive hypergraphs to unveil distinctive discourse patterns in texts produced by both LLMs and humans. Empirical findings demonstrate that, although both LLMs and humans generate distinct discourse patterns influenced by specific domains, human-written texts exhibit more structural variability, reflecting the nuanced nature of human writing in different domains. Notably, incorporating hierarchical discourse features enhances binary classifiers{'} overall performance in distinguishing between human-written and machine-generated texts, even on out-of-distribution and paraphrased samples. This underscores the significance of incorporating hierarchical discourse features in the analysis of text patterns. The code and dataset will be available at [TBA].", }

With the advent of large language models (LLM), the line between human-crafted and machine-generated texts has become increasingly blurred. This paper delves into the inquiry of identifying discernible and unique linguistic properties in texts that were written by humans, particularly uncovering the underlying discourse structures of texts beyond their surface structures. Introducing a novel methodology, we leverage hierarchical parse trees and recursive hypergraphs to unveil distinctive discourse patterns in texts produced by both LLMs and humans. Empirical findings demonstrate that, although both LLMs and humans generate distinct discourse patterns influenced by specific domains, human-written texts exhibit more structural variability, reflecting the nuanced nature of human writing in different domains. Notably, incorporating hierarchical discourse features enhances binary classifiers{'} overall performance in distinguishing between human-written and machine-generated texts, even on out-of-distribution and paraphrased samples. This underscores the significance of incorporating hierarchical discourse features in the analysis of text patterns. The code and dataset will be available at [TBA].

Threads of Subtlety: Detecting Machine-Generated Texts Through Discourse Motifs

acl-long.298

Poster

2402.10586

https://aclanthology.org/2024.acl-long.298/

https://aclanthology.org/2024.acl-long.299.bib

@inproceedings{zhang-etal-2024-jailbreak, title = "Jailbreak Open-Sourced Large Language Models via Enforced Decoding", author = "Zhang, Hangfan and Guo, Zhimeng and Zhu, Huaisheng and Cao, Bochuan and Lin, Lu and Jia, Jinyuan and Chen, Jinghui and Wu, Dinghao", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.299", pages = "5475--5493", abstract = "Large Language Models (LLMs) have achieved unprecedented performance in Natural Language Generation (NLG) tasks. However, many existing studies have shown that they could be misused to generate undesired content. In response, before releasing LLMs for public access, model developers usually align those language models through Supervised Fine-Tuning (SFT) or Reinforcement Learning with Human Feedback (RLHF). Consequently, those aligned large language models refuse to generate undesired content when facing potentially harmful/unethical requests. A natural question is {``}could alignment really prevent those open-sourced large language models from being misused to generate undesired content?{''}. In this work, we provide a negative answer to this question. In particular, we show those open-sourced, aligned large language models could be easily misguided to generate undesired content without heavy computations or careful prompt designs. Our key idea is to directly manipulate the generation process of open-sourced LLMs to misguide it to generate undesired content including harmful or biased information and even private data. We evaluate our method on 4 open-sourced LLMs accessible publicly and our finding highlights the need for more advanced mitigation strategies for open-sourced LLMs.", }

Large Language Models (LLMs) have achieved unprecedented performance in Natural Language Generation (NLG) tasks. However, many existing studies have shown that they could be misused to generate undesired content. In response, before releasing LLMs for public access, model developers usually align those language models through Supervised Fine-Tuning (SFT) or Reinforcement Learning with Human Feedback (RLHF). Consequently, those aligned large language models refuse to generate undesired content when facing potentially harmful/unethical requests. A natural question is {``}could alignment really prevent those open-sourced large language models from being misused to generate undesired content?{''}. In this work, we provide a negative answer to this question. In particular, we show those open-sourced, aligned large language models could be easily misguided to generate undesired content without heavy computations or careful prompt designs. Our key idea is to directly manipulate the generation process of open-sourced LLMs to misguide it to generate undesired content including harmful or biased information and even private data. We evaluate our method on 4 open-sourced LLMs accessible publicly and our finding highlights the need for more advanced mitigation strategies for open-sourced LLMs.

Jailbreak Open-Sourced Large Language Models via Enforced Decoding

acl-long.299

Poster

https://aclanthology.org/2024.acl-long.299/

https://aclanthology.org/2024.acl-long.300.bib

@inproceedings{srivastava-etal-2024-nice, title = "{NICE}: To Optimize In-Context Examples or Not?", author = "Srivastava, Pragya and Golechha, Satvik and Deshpande, Amit and Sharma, Amit", editor = "Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek", booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = aug, year = "2024", address = "Bangkok, Thailand", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.acl-long.300", pages = "5494--5510", abstract = "Recent work shows that in-context learning and optimization of in-context examples (ICE) can significantly improve the accuracy of large language models (LLMs) on a wide range of tasks, leading to an apparent consensus that ICE optimization is crucial for better performance. However, most of these studies assume a fixed or no instruction provided in the prompt. We challenge this consensus by investigating the necessity of optimizing ICE when task-specific instructions are provided and find that there are many tasks for which it yields diminishing returns. In particular, using a diverse set of tasks and a systematically created instruction set with gradually added details, we find that as the prompt instruction becomes more detailed, the returns on ICE optimization diminish. To characterize this behavior, we introduce a task-specific metric called Normalized Invariability to Choice of Examples (NICE) that quantifies the learnability of tasks from a given instruction, and provides a heuristic to help decide whether to optimize instructions or ICE for a new task. Given a task, the proposed metric can reliably predict the utility of optimizing ICE compared to using random ICE. Our code is available at [https://github.com/microsoft/nice-icl](https://github.com/microsoft/nice-icl).", }

Recent work shows that in-context learning and optimization of in-context examples (ICE) can significantly improve the accuracy of large language models (LLMs) on a wide range of tasks, leading to an apparent consensus that ICE optimization is crucial for better performance. However, most of these studies assume a fixed or no instruction provided in the prompt. We challenge this consensus by investigating the necessity of optimizing ICE when task-specific instructions are provided and find that there are many tasks for which it yields diminishing returns. In particular, using a diverse set of tasks and a systematically created instruction set with gradually added details, we find that as the prompt instruction becomes more detailed, the returns on ICE optimization diminish. To characterize this behavior, we introduce a task-specific metric called Normalized Invariability to Choice of Examples (NICE) that quantifies the learnability of tasks from a given instruction, and provides a heuristic to help decide whether to optimize instructions or ICE for a new task. Given a task, the proposed metric can reliably predict the utility of optimizing ICE compared to using random ICE. Our code is available at [https://github.com/microsoft/nice-icl](https://github.com/microsoft/nice-icl).

NICE: To Optimize In-Context Examples or Not?

acl-long.300

Poster

2402.06733

https://aclanthology.org/2024.acl-long.300/