Datasets:

NAACL

/

NAACL2024-papers

like 0

Follow

North American Chapter of the Association for Computational Linguistics 1

https://aclanthology.org/2024.naacl-long.301.bib

https://aclanthology.org/2024.naacl-long.301/

@inproceedings{rottger-etal-2024-xstest, title = "{XST}est: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models", author = {R{\"o}ttger, Paul and Kirk, Hannah and Vidgen, Bertie and Attanasio, Giuseppe and Bianchi, Federico and Hovy, Dirk}, editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.301", doi = "10.18653/v1/2024.naacl-long.301", pages = "5377--5400", abstract = "Without proper safeguards, large language models will readily follow malicious instructions and generate toxic content. This risk motivates safety efforts such as red-teaming and large-scale feedback learning, which aim to make models both helpful and harmless. However, there is a tension between these two objectives, since harmlessness requires models to refuse to comply with unsafe prompts, and thus not be helpful. Recent anecdotal evidence suggests that some models may have struck a poor balance, so that even clearly safe prompts are refused if they use similar language to unsafe prompts or mention sensitive topics. In this paper, we introduce a new test suite called XSTest to identify such eXaggerated Safety behaviours in a systematic way. XSTest comprises 250 safe prompts across ten prompt types that well-calibrated models should not refuse to comply with, and 200 unsafe prompts as contrasts that models, for most applications, should refuse. We describe XSTest{'}s creation and composition, and then use the test suite to highlight systematic failure modes in state-of-the-art language models as well as more general challenges in building safer language models.", }

Without proper safeguards, large language models will readily follow malicious instructions and generate toxic content. This risk motivates safety efforts such as red-teaming and large-scale feedback learning, which aim to make models both helpful and harmless. However, there is a tension between these two objectives, since harmlessness requires models to refuse to comply with unsafe prompts, and thus not be helpful. Recent anecdotal evidence suggests that some models may have struck a poor balance, so that even clearly safe prompts are refused if they use similar language to unsafe prompts or mention sensitive topics. In this paper, we introduce a new test suite called XSTest to identify such eXaggerated Safety behaviours in a systematic way. XSTest comprises 250 safe prompts across ten prompt types that well-calibrated models should not refuse to comply with, and 200 unsafe prompts as contrasts that models, for most applications, should refuse. We describe XSTest{'}s creation and composition, and then use the test suite to highlight systematic failure modes in state-of-the-art language models as well as more general challenges in building safer language models.

XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models

naacl-long.301

Poster

2308.01263

https://huggingface.co/papers/2308.01263

https://aclanthology.org/2024.naacl-long.302.bib

https://aclanthology.org/2024.naacl-long.302/

@inproceedings{kim-etal-2024-carpe, title = "Carpe diem: On the Evaluation of World Knowledge in Lifelong Language Models", author = "Kim, Yujin and Yoon, Jaehong and Ye, Seonghyeon and Bae, Sangmin and Ho, Namgyu and Hwang, Sung Ju and Yun, Se-Young", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.302", doi = "10.18653/v1/2024.naacl-long.302", pages = "5401--5415", abstract = "The dynamic nature of knowledge in an ever-changing world presents challenges for language models trained on static data; the model in the real world often requires not only acquiring new knowledge but also overwriting outdated information into updated ones. To study the ability of language models for these time-dependent dynamics in human language, we introduce a novel task, EvolvingQA, a temporally evolving question-answering benchmark designed for training and evaluating LMs on an evolving Wikipedia database. The construction of EvolvingQA is automated with our pipeline using large language models. We uncover that existing continual learning baselines suffer from updating and removing outdated knowledge. Our analysis suggests that models fail to rectify knowledge due to small weight gradients. In addition, we elucidate that language models particularly struggle to reflect the change of numerical or temporal information. Our work aims to model the dynamic nature of real-world information, suggesting faithful evaluations of the evolution-adaptability of language models. Our data construction code and dataset files are available at https://github.com/kimyuji/EvolvingQA{\_}benchmark.", }

The dynamic nature of knowledge in an ever-changing world presents challenges for language models trained on static data; the model in the real world often requires not only acquiring new knowledge but also overwriting outdated information into updated ones. To study the ability of language models for these time-dependent dynamics in human language, we introduce a novel task, EvolvingQA, a temporally evolving question-answering benchmark designed for training and evaluating LMs on an evolving Wikipedia database. The construction of EvolvingQA is automated with our pipeline using large language models. We uncover that existing continual learning baselines suffer from updating and removing outdated knowledge. Our analysis suggests that models fail to rectify knowledge due to small weight gradients. In addition, we elucidate that language models particularly struggle to reflect the change of numerical or temporal information. Our work aims to model the dynamic nature of real-world information, suggesting faithful evaluations of the evolution-adaptability of language models. Our data construction code and dataset files are available at https://github.com/kimyuji/EvolvingQA{\_}benchmark.

Carpe diem: On the Evaluation of World Knowledge in Lifelong Language Models

naacl-long.302

Poster

2311.08106

https://aclanthology.org/2024.naacl-long.303.bib

https://aclanthology.org/2024.naacl-long.303/

@inproceedings{lee-etal-2024-fine, title = "Fine-grained Gender Control in Machine Translation with Large Language Models", author = "Lee, Minwoo and Koh, Hyukhun and Kim, Minsung and Jung, Kyomin", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.303", doi = "10.18653/v1/2024.naacl-long.303", pages = "5416--5430", abstract = "In machine translation, the problem of ambiguously gendered input has been pointed out, where the gender of an entity is not available in the source sentence. To address this ambiguity issue, the task of controlled translation that takes the gender of the ambiguous entity as additional input have been proposed. However, most existing works have only considered a simplified setup of one target gender for input. In this paper, we tackle controlled translation in a more realistic setting of inputs with multiple entities and propose Gender-of-Entity (GoE) prompting method for LLMs. Our proposed method instructs the model with fine-grained entity-level gender information to translate with correct gender inflections. By utilizing four evaluation benchmarks, we investigate the controlled translation capability of LLMs in multiple dimensions and find that LLMs reach state-of-the-art performance in controlled translation. Furthermore, we discover an emergence of gender interference phenomenon when controlling the gender of multiple entities. Finally, we address the limitations of existing gender accuracy evaluation metrics and propose leveraging LLMs as an evaluator for gender inflection in machine translation.", }

In machine translation, the problem of ambiguously gendered input has been pointed out, where the gender of an entity is not available in the source sentence. To address this ambiguity issue, the task of controlled translation that takes the gender of the ambiguous entity as additional input have been proposed. However, most existing works have only considered a simplified setup of one target gender for input. In this paper, we tackle controlled translation in a more realistic setting of inputs with multiple entities and propose Gender-of-Entity (GoE) prompting method for LLMs. Our proposed method instructs the model with fine-grained entity-level gender information to translate with correct gender inflections. By utilizing four evaluation benchmarks, we investigate the controlled translation capability of LLMs in multiple dimensions and find that LLMs reach state-of-the-art performance in controlled translation. Furthermore, we discover an emergence of gender interference phenomenon when controlling the gender of multiple entities. Finally, we address the limitations of existing gender accuracy evaluation metrics and propose leveraging LLMs as an evaluator for gender inflection in machine translation.

Fine-grained Gender Control in Machine Translation with Large Language Models

naacl-long.303

Poster

2407.15154

https://aclanthology.org/2024.naacl-long.304.bib

https://aclanthology.org/2024.naacl-long.304/

@inproceedings{cai-etal-2024-dialogvcs, title = "{D}ialog{VCS}: Robust Natural Language Understanding in Dialogue System Upgrade", author = "Cai, Zefan and Zheng, Xin and Liu, Tianyu and Meng, Haoran and Han, Jiaqi and Yuan, Gang and Lin, Binghuai and Chang, Baobao and Cao, Yunbo", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.304", doi = "10.18653/v1/2024.naacl-long.304", pages = "5431--5452", abstract = "In the constant updates of the product dialogue systems, we need to retrain the natural language understanding (NLU) model as new data from the real users would be merged into the existing data accumulated in the last updates. Within the newly added data, new intents would emerge and might have semantic entanglement with the existing intents, e.g. new intents that are semantically too specific or generic are actually a subset or superset of some existing intents in the semantic space, thus impairing the robustness of the NLU model.As the first attempt to solve this problem, we setup a new benchmark consisting of 4 Dialogue Version Control dataSets (DialogVCS). We formulate the intent detection with imperfect data in the system update as a multi-label classification task with positive but unlabeled intents, which asks the models to recognize all the proper intents, including the ones with semantic entanglement, in the inference.We also propose comprehensive baseline models and conduct in-depth analyses for the benchmark, showing that the semantically entangled intents can be effectively recognized with an automatic workflow. Our code and dataset are available at \url{https://github.com/Zefan-Cai/DialogVCS}.", }

In the constant updates of the product dialogue systems, we need to retrain the natural language understanding (NLU) model as new data from the real users would be merged into the existing data accumulated in the last updates. Within the newly added data, new intents would emerge and might have semantic entanglement with the existing intents, e.g. new intents that are semantically too specific or generic are actually a subset or superset of some existing intents in the semantic space, thus impairing the robustness of the NLU model.As the first attempt to solve this problem, we setup a new benchmark consisting of 4 Dialogue Version Control dataSets (DialogVCS). We formulate the intent detection with imperfect data in the system update as a multi-label classification task with positive but unlabeled intents, which asks the models to recognize all the proper intents, including the ones with semantic entanglement, in the inference.We also propose comprehensive baseline models and conduct in-depth analyses for the benchmark, showing that the semantically entangled intents can be effectively recognized with an automatic workflow. Our code and dataset are available at \url{https://github.com/Zefan-Cai/DialogVCS}.

DialogVCS: Robust Natural Language Understanding in Dialogue System Upgrade

naacl-long.304

Poster

2305.14751

https://aclanthology.org/2024.naacl-long.305.bib

https://aclanthology.org/2024.naacl-long.305/

@inproceedings{li-etal-2024-llatrieval, title = "{LL}atrieval: {LLM}-Verified Retrieval for Verifiable Generation", author = "Li, Xiaonan and Zhu, Changtai and Li, Linyang and Yin, Zhangyue and Sun, Tianxiang and Qiu, Xipeng", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.305", doi = "10.18653/v1/2024.naacl-long.305", pages = "5453--5471", abstract = "Verifiable generation aims to let the large language model (LLM) generate text with supporting documents, which enables the user to flexibly verify the answer and makes the LLM{'}s output more reliable. Retrieval plays a crucial role in verifiable generation. Specifically, the retrieved documents not only supplement knowledge to help the LLM generate correct answers, but also serve as supporting evidence for the user to verify the LLM{'}s output. However, the widely used retrievers become the bottleneck of the entire pipeline and limit the overall performance. Their capabilities are usually inferior to LLMs since they often have much fewer parameters than the large language model and have not been demonstrated to scale well to the size of LLMs. If the retriever does not correctly find the supporting documents, the LLM can not generate the correct and verifiable answer, which overshadows the LLM{'}s remarkable abilities. To address these limitations, we propose **LLatrieval** (**L**arge **La**nguage Model Verified Re**trieval**),where the LLM updates the retrieval result until it verifies that the retrieved documents can sufficiently support answering the question. Thus, the LLM can iteratively provide feedback to retrieval and facilitate the retrieval result to fully support verifiable generation. Experiments on ALCE show that LLatrieval significantly outperforms extensive baselines and achieves state-of-the-art results.", }

Verifiable generation aims to let the large language model (LLM) generate text with supporting documents, which enables the user to flexibly verify the answer and makes the LLM{'}s output more reliable. Retrieval plays a crucial role in verifiable generation. Specifically, the retrieved documents not only supplement knowledge to help the LLM generate correct answers, but also serve as supporting evidence for the user to verify the LLM{'}s output. However, the widely used retrievers become the bottleneck of the entire pipeline and limit the overall performance. Their capabilities are usually inferior to LLMs since they often have much fewer parameters than the large language model and have not been demonstrated to scale well to the size of LLMs. If the retriever does not correctly find the supporting documents, the LLM can not generate the correct and verifiable answer, which overshadows the LLM{'}s remarkable abilities. To address these limitations, we propose **LLatrieval** (**L**arge **La**nguage Model Verified Re**trieval**),where the LLM updates the retrieval result until it verifies that the retrieved documents can sufficiently support answering the question. Thus, the LLM can iteratively provide feedback to retrieval and facilitate the retrieval result to fully support verifiable generation. Experiments on ALCE show that LLatrieval significantly outperforms extensive baselines and achieves state-of-the-art results.

LLatrieval: LLM-Verified Retrieval for Verifiable Generation

naacl-long.305

Poster

2311.07838

https://aclanthology.org/2024.naacl-long.306.bib

https://aclanthology.org/2024.naacl-long.306/

@inproceedings{chen-etal-2024-mapping, title = "Mapping Long-term Causalities in Psychiatric Symptomatology and Life Events from Social Media", author = "Chen, Siyuan and Wang, Meilin and Lv, Minghao and Zhang, Zhiling and Juqianqian, Juqianqian and Dejiyangla, Dejiyangla and Peng, Yujia and Zhu, Kenny and Wu, Mengyue", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.306", doi = "10.18653/v1/2024.naacl-long.306", pages = "5472--5487", abstract = "Social media is a valuable data source for exploring mental health issues. However, previous studies have predominantly focused on the semantic content of these posts, overlooking the importance of their temporal attributes, as well as the evolving nature of mental disorders and symptoms.In this paper, we study the causality between psychiatric symptoms and life events, as well as among different symptoms from social media posts, which leads to better understanding of the underlying mechanisms of mental disorders. By applying these extracted causality features to tasks such as diagnosis point detection and early risk detection of depression, we notice considerable performance enhancement. This indicates that causality information extracted from social media data can boost the efficacy of mental disorder diagnosis and treatment planning.", }

Social media is a valuable data source for exploring mental health issues. However, previous studies have predominantly focused on the semantic content of these posts, overlooking the importance of their temporal attributes, as well as the evolving nature of mental disorders and symptoms.In this paper, we study the causality between psychiatric symptoms and life events, as well as among different symptoms from social media posts, which leads to better understanding of the underlying mechanisms of mental disorders. By applying these extracted causality features to tasks such as diagnosis point detection and early risk detection of depression, we notice considerable performance enhancement. This indicates that causality information extracted from social media data can boost the efficacy of mental disorder diagnosis and treatment planning.

Mapping Long-term Causalities in Psychiatric Symptomatology and Life Events from Social Media

naacl-long.306

Poster

https://aclanthology.org/2024.naacl-long.307.bib

https://aclanthology.org/2024.naacl-long.307/

@inproceedings{nowak-etal-2024-multimodal, title = "Multimodal Chart Retrieval: A Comparison of Text, Table and Image Based Approaches", author = "Nowak, Averi and Piccinno, Francesco and Altun, Yasemin", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.307", doi = "10.18653/v1/2024.naacl-long.307", pages = "5488--5505", abstract = "We investigate multimodal chart retrieval, addressing the challenge of retrieving image-based charts using textual queries. We compare four approaches: (a) OCR with text retrieval, (b) chart derendering (DePlot) followed by table retrieval, (c) a direct image understanding model (PaLI-3), and (d) a combined PaLI-3 + DePlot approach. As the table retrieval component we introduce Tab-GTR, a text retrieval model augmented with table structure embeddings, achieving state-of-the-art results on the NQ-Tables benchmark with 48.88{\%} R@1. On in-distribution data, the DePlot-based method (b) outperforms PaLI-3 (c), while being significantly more efficient (300M vs 3B trainable parameters). However, DePlot struggles with complex charts, indicating a need for improvements in chart derendering - specifically in terms of chart data diversity and the richness of text/table representations. We found no clear winner between methods (b) and (c) in general, with the best performance achieved by the combined approach (d), and further show that it benefits the most from multi-task training.", }

We investigate multimodal chart retrieval, addressing the challenge of retrieving image-based charts using textual queries. We compare four approaches: (a) OCR with text retrieval, (b) chart derendering (DePlot) followed by table retrieval, (c) a direct image understanding model (PaLI-3), and (d) a combined PaLI-3 + DePlot approach. As the table retrieval component we introduce Tab-GTR, a text retrieval model augmented with table structure embeddings, achieving state-of-the-art results on the NQ-Tables benchmark with 48.88{\%} R@1. On in-distribution data, the DePlot-based method (b) outperforms PaLI-3 (c), while being significantly more efficient (300M vs 3B trainable parameters). However, DePlot struggles with complex charts, indicating a need for improvements in chart derendering - specifically in terms of chart data diversity and the richness of text/table representations. We found no clear winner between methods (b) and (c) in general, with the best performance achieved by the combined approach (d), and further show that it benefits the most from multi-task training.

Multimodal Chart Retrieval: A Comparison of Text, Table and Image Based Approaches

naacl-long.307

Poster

https://aclanthology.org/2024.naacl-long.308.bib

https://aclanthology.org/2024.naacl-long.308/

@inproceedings{maekawa-etal-2024-retrieval, title = "Retrieval Helps or Hurts? A Deeper Dive into the Efficacy of Retrieval Augmentation to Language Models", author = "Maekawa, Seiji and Iso, Hayate and Gurajada, Sairam and Bhutani, Nikita", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.308", doi = "10.18653/v1/2024.naacl-long.308", pages = "5506--5521", abstract = "While large language models (LMs) demonstrate remarkable performance, they encounter challenges in providing accurate responses when queried for information beyond their pre-trained memorization. Although augmenting them with relevant external information can mitigate these issues, failure to consider the necessity of retrieval may adversely affect overall performance. Previous research has primarily focused on examining how entities influence retrieval models and knowledge recall in LMs, leaving other aspects relatively unexplored. In this work, our goal is to offer a more detailed, fact-centric analysis by exploring the effects of combinations of entities and relations. To facilitate this, we construct a new question answering (QA) dataset called WiTQA (Wikipedia Triple Question Answers). This dataset includes questions about entities and relations of various popularity levels, each accompanied by a supporting passage. Our extensive experiments with diverse LMs and retrievers reveal when retrieval does not consistently enhance LMs from the viewpoints of fact-centric popularity. Confirming earlier findings, we observe that larger LMs excel in recalling popular facts. However, they notably encounter difficulty with infrequent entity-relation pairs compared to retrievers. Interestingly, they can effectively retain popular relations of less common entities. We demonstrate the efficacy of our finer-grained metric and insights through an adaptive retrieval system that selectively employs retrieval and recall based on the frequencies of entities and relations in the question.", }

While large language models (LMs) demonstrate remarkable performance, they encounter challenges in providing accurate responses when queried for information beyond their pre-trained memorization. Although augmenting them with relevant external information can mitigate these issues, failure to consider the necessity of retrieval may adversely affect overall performance. Previous research has primarily focused on examining how entities influence retrieval models and knowledge recall in LMs, leaving other aspects relatively unexplored. In this work, our goal is to offer a more detailed, fact-centric analysis by exploring the effects of combinations of entities and relations. To facilitate this, we construct a new question answering (QA) dataset called WiTQA (Wikipedia Triple Question Answers). This dataset includes questions about entities and relations of various popularity levels, each accompanied by a supporting passage. Our extensive experiments with diverse LMs and retrievers reveal when retrieval does not consistently enhance LMs from the viewpoints of fact-centric popularity. Confirming earlier findings, we observe that larger LMs excel in recalling popular facts. However, they notably encounter difficulty with infrequent entity-relation pairs compared to retrievers. Interestingly, they can effectively retain popular relations of less common entities. We demonstrate the efficacy of our finer-grained metric and insights through an adaptive retrieval system that selectively employs retrieval and recall based on the frequencies of entities and relations in the question.

Retrieval Helps or Hurts? A Deeper Dive into the Efficacy of Retrieval Augmentation to Language Models

naacl-long.308

Oral

2402.13492

https://huggingface.co/papers/2402.13492

https://aclanthology.org/2024.naacl-long.309.bib

https://aclanthology.org/2024.naacl-long.309/

@inproceedings{fathullah-etal-2024-audiochatllama, title = "{A}udio{C}hat{L}lama: Towards General-Purpose Speech Abilities for {LLM}s", author = "Fathullah, Yassir and Wu, Chunyang and Lakomkin, Egor and Li, Ke and Jia, Junteng and Shangguan, Yuan and Mahadeokar, Jay and Kalinli, Ozlem and Fuegen, Christian and Seltzer, Mike", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.309", doi = "10.18653/v1/2024.naacl-long.309", pages = "5522--5532", abstract = "In this work, we extend the instruction-tuned Llama-2 model with end-to-end general-purpose speech processing and reasoning abilities while maintaining the wide range of original LLM capabilities, without using any carefully curated paired data. The resulting end-to-end model, named \textit{AudioChatLlama}, can utilize audio prompts as a replacement for text and sustain a conversation. Such a model also has extended cross-modal capabilities such as being able to perform spoken question answering (QA), speech translation, and audio summarization amongst many other closed and open-domain tasks. This is unlike prior approaches in speech, in which LLMs are extended to handle audio for a limited number of pre-designated tasks. On both synthesized and recorded speech QA test sets, evaluations show that our end-to-end approach is on par with or outperforms cascaded systems (speech recognizer + LLM) in terms of modelling the response to a prompt. Furthermore, unlike cascades, our approach can interchange text and audio modalities and intrinsically utilize prior context in a conversation to provide better results.", }

In this work, we extend the instruction-tuned Llama-2 model with end-to-end general-purpose speech processing and reasoning abilities while maintaining the wide range of original LLM capabilities, without using any carefully curated paired data. The resulting end-to-end model, named \textit{AudioChatLlama}, can utilize audio prompts as a replacement for text and sustain a conversation. Such a model also has extended cross-modal capabilities such as being able to perform spoken question answering (QA), speech translation, and audio summarization amongst many other closed and open-domain tasks. This is unlike prior approaches in speech, in which LLMs are extended to handle audio for a limited number of pre-designated tasks. On both synthesized and recorded speech QA test sets, evaluations show that our end-to-end approach is on par with or outperforms cascaded systems (speech recognizer + LLM) in terms of modelling the response to a prompt. Furthermore, unlike cascades, our approach can interchange text and audio modalities and intrinsically utilize prior context in a conversation to provide better results.

AudioChatLlama: Towards General-Purpose Speech Abilities for LLMs

naacl-long.309

Poster

2311.06753

https://aclanthology.org/2024.naacl-long.310.bib

https://aclanthology.org/2024.naacl-long.310/

@inproceedings{gupta-etal-2024-whispers, title = "Whispers of Doubt Amidst Echoes of Triumph in {NLP} Robustness", author = "Gupta, Ashim and Rajendhran, Rishanth and Stringham, Nathan and Srikumar, Vivek and Marasovic, Ana", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.310", doi = "10.18653/v1/2024.naacl-long.310", pages = "5533--5590", abstract = "*Do larger and more performant models resolve NLP{'}s longstanding robustness issues?* We investigate this question using over 20 models of different sizes spanning different architectural choices and pretraining objectives. We conduct evaluations using (a) out-of-domain and challenge test sets, (b) behavioral testing with CheckLists, (c) contrast sets, and (d) adversarial inputs. Our analysis reveals that not all out-of-domain tests provide insight into robustness. Evaluating with CheckLists and contrast sets shows significant gaps in model performance; merely scaling models does not make them adequately robust. Finally, we point out that current approaches for adversarial evaluations of models are themselves problematic: they can be easily thwarted, and in their current forms, do not represent a sufficiently deep probe of model robustness. We conclude that not only is the question of robustness in NLP as yet unresolved, but even some of the approaches to measure robustness need to be reassessed.", }

*Do larger and more performant models resolve NLP{'}s longstanding robustness issues?* We investigate this question using over 20 models of different sizes spanning different architectural choices and pretraining objectives. We conduct evaluations using (a) out-of-domain and challenge test sets, (b) behavioral testing with CheckLists, (c) contrast sets, and (d) adversarial inputs. Our analysis reveals that not all out-of-domain tests provide insight into robustness. Evaluating with CheckLists and contrast sets shows significant gaps in model performance; merely scaling models does not make them adequately robust. Finally, we point out that current approaches for adversarial evaluations of models are themselves problematic: they can be easily thwarted, and in their current forms, do not represent a sufficiently deep probe of model robustness. We conclude that not only is the question of robustness in NLP as yet unresolved, but even some of the approaches to measure robustness need to be reassessed.

Whispers of Doubt Amidst Echoes of Triumph in NLP Robustness

naacl-long.310

Poster

2311.09694

https://aclanthology.org/2024.naacl-long.311.bib

https://aclanthology.org/2024.naacl-long.311/

@inproceedings{yagcioglu-etal-2024-sequential, title = "Sequential Compositional Generalization in Multimodal Models", author = "Yagcioglu, Semih and {\.I}nce, Osman Batur and Erdem, Aykut and Erdem, Erkut and Elliott, Desmond and Yuret, Deniz", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.311", doi = "10.18653/v1/2024.naacl-long.311", pages = "5591--5611", abstract = "The rise of large-scale multimodal models has paved the pathway for groundbreaking advances in generative modeling and reasoning, unlocking transformative applications in a variety of complex tasks. However, a pressing question that remains is their genuine capability for stronger forms of generalization, which has been largely underexplored in the multimodal setting. Our study aims to address this by examining sequential compositional generalization using CompAct (Compositional Activities), a carefully constructed, perceptually grounded dataset set within a rich backdrop of egocentric kitchen activity videos. Each instance in our dataset is represented with a combination of raw video footage, naturally occurring sound, and crowd-sourced step-by-step descriptions. More importantly, our setup ensures that the individual concepts are consistently distributed across training and evaluation sets, while their compositions are novel in the evaluation set. We conduct a comprehensive assessment of several unimodal and multimodal models. Our findings reveal that bi-modal and tri-modal models exhibit a clear edge over their text-only counterparts. This highlights the importance of multimodality while charting a trajectory for future research in this domain.", }

The rise of large-scale multimodal models has paved the pathway for groundbreaking advances in generative modeling and reasoning, unlocking transformative applications in a variety of complex tasks. However, a pressing question that remains is their genuine capability for stronger forms of generalization, which has been largely underexplored in the multimodal setting. Our study aims to address this by examining sequential compositional generalization using CompAct (Compositional Activities), a carefully constructed, perceptually grounded dataset set within a rich backdrop of egocentric kitchen activity videos. Each instance in our dataset is represented with a combination of raw video footage, naturally occurring sound, and crowd-sourced step-by-step descriptions. More importantly, our setup ensures that the individual concepts are consistently distributed across training and evaluation sets, while their compositions are novel in the evaluation set. We conduct a comprehensive assessment of several unimodal and multimodal models. Our findings reveal that bi-modal and tri-modal models exhibit a clear edge over their text-only counterparts. This highlights the importance of multimodality while charting a trajectory for future research in this domain.

Sequential Compositional Generalization in Multimodal Models

naacl-long.311

Oral

2404.12013

https://aclanthology.org/2024.naacl-long.312.bib

https://aclanthology.org/2024.naacl-long.312/

@inproceedings{uddin-etal-2024-generating, title = "Generating Uncontextualized and Contextualized Questions for Document-Level Event Argument Extraction", author = "Uddin, Md Nayem and George, Enfa and Blanco, Eduardo and Corman, Steven", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.312", doi = "10.18653/v1/2024.naacl-long.312", pages = "5612--5627", abstract = "This paper presents multiple question generation strategies for document-level event argument extraction. These strategies do not require human involvement and result in uncontextualized questions as well as contextualized questions grounded on the event and document of interest. Experimental results show that combining uncontextualized and contextualized questions is beneficial,especially when event triggers and arguments appear in different sentences. Our approach does not have corpus-specific components, in particular, the question generation strategies transfer across corpora. We also present a qualitative analysis of the most common errors made by our best model.", }

This paper presents multiple question generation strategies for document-level event argument extraction. These strategies do not require human involvement and result in uncontextualized questions as well as contextualized questions grounded on the event and document of interest. Experimental results show that combining uncontextualized and contextualized questions is beneficial,especially when event triggers and arguments appear in different sentences. Our approach does not have corpus-specific components, in particular, the question generation strategies transfer across corpora. We also present a qualitative analysis of the most common errors made by our best model.

Generating Uncontextualized and Contextualized Questions for Document-Level Event Argument Extraction

naacl-long.312

Poster

2404.04770

https://aclanthology.org/2024.naacl-long.313.bib

https://aclanthology.org/2024.naacl-long.313/

@inproceedings{yue-etal-2024-evidence, title = "Evidence-Driven Retrieval Augmented Response Generation for Online Misinformation", author = "Yue, Zhenrui and Zeng, Huimin and Lu, Yimeng and Shang, Lanyu and Zhang, Yang and Wang, Dong", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.313", doi = "10.18653/v1/2024.naacl-long.313", pages = "5628--5643", abstract = "The proliferation of online misinformation has posed significant threats to public interest. While numerous online users actively participate in the combat against misinformation, many of such responses can be characterized by the lack of politeness and supporting facts. As a solution, text generation approaches are proposed to automatically produce counter-misinformation responses. Nevertheless, existing methods are often trained end-to-end without leveraging external knowledge, resulting in subpar text quality and excessively repetitive responses. In this paper, we propose retrieval augmented response generation for online misinformation (RARG), which collects supporting evidence from scientific sources and generates counter-misinformation responses based on the evidences. In particular, our RARG consists of two stages: (1) evidence collection, where we design a retrieval pipeline to retrieve and rerank evidence documents using a database comprising over 1M academic articles; (2) response generation, in which we align large language models (LLMs) to generate evidence-based responses via reinforcement learning from human feedback (RLHF). We propose a reward function to maximize the utilization of the retrieved evidence while maintaining the quality of the generated text, which yields polite and factual responses that clearly refutes misinformation. To demonstrate the effectiveness of our method, we study the case of COVID-19 and perform extensive experiments with both in- and cross-domain datasets, where RARG consistently outperforms baselines by generating high-quality counter-misinformation responses.", }

The proliferation of online misinformation has posed significant threats to public interest. While numerous online users actively participate in the combat against misinformation, many of such responses can be characterized by the lack of politeness and supporting facts. As a solution, text generation approaches are proposed to automatically produce counter-misinformation responses. Nevertheless, existing methods are often trained end-to-end without leveraging external knowledge, resulting in subpar text quality and excessively repetitive responses. In this paper, we propose retrieval augmented response generation for online misinformation (RARG), which collects supporting evidence from scientific sources and generates counter-misinformation responses based on the evidences. In particular, our RARG consists of two stages: (1) evidence collection, where we design a retrieval pipeline to retrieve and rerank evidence documents using a database comprising over 1M academic articles; (2) response generation, in which we align large language models (LLMs) to generate evidence-based responses via reinforcement learning from human feedback (RLHF). We propose a reward function to maximize the utilization of the retrieved evidence while maintaining the quality of the generated text, which yields polite and factual responses that clearly refutes misinformation. To demonstrate the effectiveness of our method, we study the case of COVID-19 and perform extensive experiments with both in- and cross-domain datasets, where RARG consistently outperforms baselines by generating high-quality counter-misinformation responses.

Evidence-Driven Retrieval Augmented Response Generation for Online Misinformation

naacl-long.313

Poster

2403.14952

https://aclanthology.org/2024.naacl-long.314.bib

https://aclanthology.org/2024.naacl-long.314/

@inproceedings{zeng-etal-2024-open, title = "Open-Vocabulary Federated Learning with Multimodal Prototyping", author = "Zeng, Huimin and Yue, Zhenrui and Wang, Dong", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.314", doi = "10.18653/v1/2024.naacl-long.314", pages = "5644--5656", abstract = "Existing federated learning (FL) studies usuallyassume the training label space and test labelspace are identical. However, in real-world applications, this assumption is too ideal to betrue. A new user could come up with queriesthat involve data from unseen classes, and suchopen-vocabulary queries would directly defectsuch FL systems. Therefore, in this work, weexplicitly focus on the under-explored openvocabulary challenge in FL. That is, for a newuser, the global server shall understand her/hisquery that involves arbitrary unknown classes.To address this problem, we leverage the pretrained vision-language models (VLMs). Inparticular, we present a novel adaptation framework tailored for VLMs in the context of FL,named as Federated Multimodal Prototyping(Fed-MP). Fed-MP adaptively aggregates thelocal model weights based on light-weightclient residuals, and makes predictions basedon a novel multimodal prototyping mechanism.Fed-MP exploits the knowledge learned fromthe seen classes, and robustifies the adaptedVLM to unseen categories. Our empirical evaluation on various datasets validates the effectiveness of Fed-MP.", }

Existing federated learning (FL) studies usuallyassume the training label space and test labelspace are identical. However, in real-world applications, this assumption is too ideal to betrue. A new user could come up with queriesthat involve data from unseen classes, and suchopen-vocabulary queries would directly defectsuch FL systems. Therefore, in this work, weexplicitly focus on the under-explored openvocabulary challenge in FL. That is, for a newuser, the global server shall understand her/hisquery that involves arbitrary unknown classes.To address this problem, we leverage the pretrained vision-language models (VLMs). Inparticular, we present a novel adaptation framework tailored for VLMs in the context of FL,named as Federated Multimodal Prototyping(Fed-MP). Fed-MP adaptively aggregates thelocal model weights based on light-weightclient residuals, and makes predictions basedon a novel multimodal prototyping mechanism.Fed-MP exploits the knowledge learned fromthe seen classes, and robustifies the adaptedVLM to unseen categories. Our empirical evaluation on various datasets validates the effectiveness of Fed-MP.

Open-Vocabulary Federated Learning with Multimodal Prototyping

naacl-long.314

Poster

2404.01232

https://aclanthology.org/2024.naacl-long.315.bib

https://aclanthology.org/2024.naacl-long.315/

@inproceedings{li-etal-2024-exploring, title = "Exploring Key Point Analysis with Pairwise Generation and Graph Partitioning", author = "Li, Xiao and Jiang, Yong and Huang, Shen and Xie, Pengjun and Cheng, Gong and Huang, Fei", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.315", doi = "10.18653/v1/2024.naacl-long.315", pages = "5657--5667", abstract = "Key Point Analysis (KPA), the summarization of multiple arguments into a concise collection of key points, continues to be a significant and unresolved issue within the field of argument mining. Existing models adapt a two-stage pipeline of clustering arguments or generating key points for argument clusters. This approach rely on semantic similarity instead of measuring the existence of shared key points among arguments. Additionally, it only models the intra-cluster relationship among arguments, disregarding the inter-cluster relationship between arguments that do not share key points. To address these limitations, we propose a novel approach for KPA with pairwise generation and graph partitioning. Our objective is to train a generative model that can simultaneously provide a score indicating the presence of shared key point between a pair of arguments and generate the shared key point. Subsequently, to map generated redundant key points to a concise set of key points, we proceed to construct an arguments graph by considering the arguments as vertices, the generated key points as edges, and the scores as edge weights. We then propose a graph partitioning algorithm to partition all arguments sharing the same key points to the same subgraph. Notably, our experimental findings demonstrate that our proposed model surpasses previous models when evaluated on both the ArgKP and QAM datasets.", }

Key Point Analysis (KPA), the summarization of multiple arguments into a concise collection of key points, continues to be a significant and unresolved issue within the field of argument mining. Existing models adapt a two-stage pipeline of clustering arguments or generating key points for argument clusters. This approach rely on semantic similarity instead of measuring the existence of shared key points among arguments. Additionally, it only models the intra-cluster relationship among arguments, disregarding the inter-cluster relationship between arguments that do not share key points. To address these limitations, we propose a novel approach for KPA with pairwise generation and graph partitioning. Our objective is to train a generative model that can simultaneously provide a score indicating the presence of shared key point between a pair of arguments and generate the shared key point. Subsequently, to map generated redundant key points to a concise set of key points, we proceed to construct an arguments graph by considering the arguments as vertices, the generated key points as edges, and the scores as edge weights. We then propose a graph partitioning algorithm to partition all arguments sharing the same key points to the same subgraph. Notably, our experimental findings demonstrate that our proposed model surpasses previous models when evaluated on both the ArgKP and QAM datasets.

Exploring Key Point Analysis with Pairwise Generation and Graph Partitioning

naacl-long.315

Poster

2404.11384

https://aclanthology.org/2024.naacl-long.316.bib

https://aclanthology.org/2024.naacl-long.316/

@inproceedings{shen-etal-2024-understanding, title = "Understanding the Capabilities and Limitations of Large Language Models for Cultural Commonsense", author = "Shen, Siqi and Logeswaran, Lajanugen and Lee, Moontae and Lee, Honglak and Poria, Soujanya and Mihalcea, Rada", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.316", doi = "10.18653/v1/2024.naacl-long.316", pages = "5668--5680", abstract = "Large language models (LLMs) have demonstrated substantial commonsense understanding through numerous benchmark evaluations. However, their understanding of cultural commonsense remains largely unexamined. In this paper, we conduct a comprehensive examination of the capabilities and limitations of several state-of-the-art LLMs in the context of cultural commonsense tasks. Using several general and cultural commonsense benchmarks, we find that (1) LLMs have a significant discrepancy in performance when tested on culture-specific commonsense knowledge for different cultures; (2) LLMs{'} general commonsense capability is affected by cultural context; and (3) The language used to query the LLMs can impact their performance on cultural-related tasks.Our study points to the inherent bias in the cultural understanding of LLMs and provides insights that can help develop culturally-aware language models.", }

Large language models (LLMs) have demonstrated substantial commonsense understanding through numerous benchmark evaluations. However, their understanding of cultural commonsense remains largely unexamined. In this paper, we conduct a comprehensive examination of the capabilities and limitations of several state-of-the-art LLMs in the context of cultural commonsense tasks. Using several general and cultural commonsense benchmarks, we find that (1) LLMs have a significant discrepancy in performance when tested on culture-specific commonsense knowledge for different cultures; (2) LLMs{'} general commonsense capability is affected by cultural context; and (3) The language used to query the LLMs can impact their performance on cultural-related tasks.Our study points to the inherent bias in the cultural understanding of LLMs and provides insights that can help develop culturally-aware language models.

Understanding the Capabilities and Limitations of Large Language Models for Cultural Commonsense

naacl-long.316

Poster

2405.04655

https://aclanthology.org/2024.naacl-long.317.bib

https://aclanthology.org/2024.naacl-long.317/

@inproceedings{logeswaran-etal-2024-code, title = "Code Models are Zero-shot Precondition Reasoners", author = "Logeswaran, Lajanugen and Sohn, Sungryull and Lyu, Yiwei and Liu, Anthony and Kim, Dong-Ki and Shim, Dongsub and Lee, Moontae and Lee, Honglak", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.317", doi = "10.18653/v1/2024.naacl-long.317", pages = "5681--5697", abstract = "One of the fundamental skills required for an agent acting in an environment to complete tasks is the ability to understand what actions are plausible at any given point. This work explores a novel use of code representations to reason about action preconditions for sequential decision making tasks. Code representations offer the flexibility to model procedural activities and associated constraints as well as the ability to execute and verify constraint satisfaction. Leveraging code representations, we extract action preconditions from demonstration trajectories in a zero-shot manner using pre-trained code models. Given these extracted preconditions, we propose a precondition-aware action sampling strategy that ensures actions predicted by a policy are consistent with preconditions. We demonstrate that the proposed approach enhances the performance of few-shot policy learning approaches across task-oriented dialog and embodied textworld benchmarks.", }

One of the fundamental skills required for an agent acting in an environment to complete tasks is the ability to understand what actions are plausible at any given point. This work explores a novel use of code representations to reason about action preconditions for sequential decision making tasks. Code representations offer the flexibility to model procedural activities and associated constraints as well as the ability to execute and verify constraint satisfaction. Leveraging code representations, we extract action preconditions from demonstration trajectories in a zero-shot manner using pre-trained code models. Given these extracted preconditions, we propose a precondition-aware action sampling strategy that ensures actions predicted by a policy are consistent with preconditions. We demonstrate that the proposed approach enhances the performance of few-shot policy learning approaches across task-oriented dialog and embodied textworld benchmarks.

Code Models are Zero-shot Precondition Reasoners

naacl-long.317

Oral

2311.09601

https://aclanthology.org/2024.naacl-long.318.bib

https://aclanthology.org/2024.naacl-long.318/

@inproceedings{kim-etal-2024-contrastive, title = "Contrastive and Consistency Learning for Neural Noisy-Channel Model in Spoken Language Understanding", author = "Kim, Suyoung and Hwang, Jiyeon and Jung, Ho-Young", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.318", doi = "10.18653/v1/2024.naacl-long.318", pages = "5698--5711", abstract = "Recently, deep end-to-end learning has been studied for intent classification in Spoken Language Understanding (SLU). However, end-to-end models require a large amount of speech data with intent labels, and highly optimized models are generally sensitive to the inconsistency between the training and evaluation conditions. Therefore, a natural language understanding approach based on Automatic Speech Recognition (ASR) remains attractive because it can utilize a pre-trained general language model and adapt to the mismatch of the speech input environment. Using this module-based approach, we improve a noisy-channel model to handle transcription inconsistencies caused by ASR errors. We propose a two-stage method, Contrastive and Consistency Learning (CCL), that correlates error patterns between clean and noisy ASR transcripts and emphasizes the consistency of the latent features of the two transcripts. Experiments on four benchmark datasets show that CCL outperforms existing methods and improves the ASR robustness in various noisy environments. Code is available at https://github.com/syoung7388/CCL", }

Recently, deep end-to-end learning has been studied for intent classification in Spoken Language Understanding (SLU). However, end-to-end models require a large amount of speech data with intent labels, and highly optimized models are generally sensitive to the inconsistency between the training and evaluation conditions. Therefore, a natural language understanding approach based on Automatic Speech Recognition (ASR) remains attractive because it can utilize a pre-trained general language model and adapt to the mismatch of the speech input environment. Using this module-based approach, we improve a noisy-channel model to handle transcription inconsistencies caused by ASR errors. We propose a two-stage method, Contrastive and Consistency Learning (CCL), that correlates error patterns between clean and noisy ASR transcripts and emphasizes the consistency of the latent features of the two transcripts. Experiments on four benchmark datasets show that CCL outperforms existing methods and improves the ASR robustness in various noisy environments. Code is available at https://github.com/syoung7388/CCL

Contrastive and Consistency Learning for Neural Noisy-Channel Model in Spoken Language Understanding

naacl-long.318

Oral

2405.15097

https://aclanthology.org/2024.naacl-long.319.bib

https://aclanthology.org/2024.naacl-long.319/

@inproceedings{wang-etal-2024-large, title = "Do Large Language Models Rank Fairly? An Empirical Study on the Fairness of {LLM}s as Rankers", author = "Wang, Yuan and Wu, Xuyang and Wu, Hsin-Tai and Tao, Zhiqiang and Fang, Yi", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.319", doi = "10.18653/v1/2024.naacl-long.319", pages = "5712--5724", abstract = "The integration of Large Language Models (LLMs) in information retrieval has raised a critical reevaluation of fairness in the text-ranking models. LLMs, such as GPT models and Llama2, have shown effectiveness in natural language understanding tasks, and prior works such as RankGPT have demonstrated that the LLMs have better performance than the traditional ranking models in the ranking task. However, their fairness remains largely unexplored. This paper presents an empirical study evaluating these LLMs using the TREC Fair Ranking dataset, focusing on the representation of binary protected attributes such as gender and geographic location, which are historically underrepresented in search outcomes. Our analysis delves into how these LLMs handle queries and documents related to these attributes, aiming to uncover biases in their ranking algorithms. We assess fairness from both user and content perspectives, contributing an empirical benchmark for evaluating LLMs as the fair ranker.", }

The integration of Large Language Models (LLMs) in information retrieval has raised a critical reevaluation of fairness in the text-ranking models. LLMs, such as GPT models and Llama2, have shown effectiveness in natural language understanding tasks, and prior works such as RankGPT have demonstrated that the LLMs have better performance than the traditional ranking models in the ranking task. However, their fairness remains largely unexplored. This paper presents an empirical study evaluating these LLMs using the TREC Fair Ranking dataset, focusing on the representation of binary protected attributes such as gender and geographic location, which are historically underrepresented in search outcomes. Our analysis delves into how these LLMs handle queries and documents related to these attributes, aiming to uncover biases in their ranking algorithms. We assess fairness from both user and content perspectives, contributing an empirical benchmark for evaluating LLMs as the fair ranker.

Do Large Language Models Rank Fairly? An Empirical Study on the Fairness of LLMs as Rankers

naacl-long.319

Poster

2404.03192

https://aclanthology.org/2024.naacl-long.320.bib

https://aclanthology.org/2024.naacl-long.320/

@inproceedings{nahid-rafiei-2024-tabsqlify, title = "{T}ab{SQL}ify: Enhancing Reasoning Capabilities of {LLM}s Through Table Decomposition", author = "Nahid, Md and Rafiei, Davood", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.320", doi = "10.18653/v1/2024.naacl-long.320", pages = "5725--5737", abstract = "Table reasoning is a challenging task that requires understanding both natural language questions and structured tabular data. Large language models (LLMs) have shown impressive capabilities in natural language understanding and generation, but they often struggle with large tables due to their limited input length. In this paper, we propose TabSQLify, a novel method that leverages text-to-SQL generation to decompose tables into smaller and relevant sub-tables, containing only essential information for answering questions or verifying statements, before performing the reasoning task. In our comprehensive evaluation on four challenging datasets, our approach demonstrates comparable or superior performance compared to prevailing methods reliant on full tables as input. Moreover, our method can reduce the input context length significantly, making it more scalable and efficient for large-scale table reasoning applications. Our method performs remarkably well on the WikiTQ benchmark, achieving an accuracy of 64.7{\%}. Additionally, on the TabFact benchmark, it achieves a high accuracy of 79.5{\%}. These results surpass other LLM-based baseline models on gpt-3.5-turbo (chatgpt). TabSQLify can reduce the table size significantly alleviating the computational load on LLMs when handling large tables without compromising performance.", }

Table reasoning is a challenging task that requires understanding both natural language questions and structured tabular data. Large language models (LLMs) have shown impressive capabilities in natural language understanding and generation, but they often struggle with large tables due to their limited input length. In this paper, we propose TabSQLify, a novel method that leverages text-to-SQL generation to decompose tables into smaller and relevant sub-tables, containing only essential information for answering questions or verifying statements, before performing the reasoning task. In our comprehensive evaluation on four challenging datasets, our approach demonstrates comparable or superior performance compared to prevailing methods reliant on full tables as input. Moreover, our method can reduce the input context length significantly, making it more scalable and efficient for large-scale table reasoning applications. Our method performs remarkably well on the WikiTQ benchmark, achieving an accuracy of 64.7{\%}. Additionally, on the TabFact benchmark, it achieves a high accuracy of 79.5{\%}. These results surpass other LLM-based baseline models on gpt-3.5-turbo (chatgpt). TabSQLify can reduce the table size significantly alleviating the computational load on LLMs when handling large tables without compromising performance.

TabSQLify: Enhancing Reasoning Capabilities of LLMs Through Table Decomposition

naacl-long.320

Poster

2404.10150

https://aclanthology.org/2024.naacl-long.321.bib

https://aclanthology.org/2024.naacl-long.321/

@inproceedings{parekh-etal-2024-contextual, title = "Contextual Label Projection for Cross-Lingual Structured Prediction", author = "Parekh, Tanmay and Hsu, I-Hung and Huang, Kuan-Hao and Chang, Kai-Wei and Peng, Nanyun", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.321", doi = "10.18653/v1/2024.naacl-long.321", pages = "5738--5757", abstract = "Label projection, which involves obtaining translated labels and texts jointly, is essential for leveraging machine translation to facilitate cross-lingual transfer in structured prediction tasks. Prior research exploring label projection often compromise translation accuracy by favoring simplified label translation or relying solely on word-level alignments. In this paper, we introduce a novel label projection approach, CLaP, which translates text to the target language and performs *contextual translation* on the labels using the translated text as the context, ensuring better accuracy for the translated labels. We leverage instruction-tuned language models with multilingual capabilities as our contextual translator, imposing the constraint of the presence of translated labels in the translated text via instructions. We benchmark CLaP with other label projection techniques on zero-shot cross-lingual transfer across 39 languages on two representative structured prediction tasks - event argument extraction (EAE) and named entity recognition (NER), showing over 2.4 F1 improvement for EAE and 1.4 F1 improvement for NER. We further explore the applicability of CLaP on ten extremely low-resource languages to showcase its potential for cross-lingual structured prediction.", }

Label projection, which involves obtaining translated labels and texts jointly, is essential for leveraging machine translation to facilitate cross-lingual transfer in structured prediction tasks. Prior research exploring label projection often compromise translation accuracy by favoring simplified label translation or relying solely on word-level alignments. In this paper, we introduce a novel label projection approach, CLaP, which translates text to the target language and performs *contextual translation* on the labels using the translated text as the context, ensuring better accuracy for the translated labels. We leverage instruction-tuned language models with multilingual capabilities as our contextual translator, imposing the constraint of the presence of translated labels in the translated text via instructions. We benchmark CLaP with other label projection techniques on zero-shot cross-lingual transfer across 39 languages on two representative structured prediction tasks - event argument extraction (EAE) and named entity recognition (NER), showing over 2.4 F1 improvement for EAE and 1.4 F1 improvement for NER. We further explore the applicability of CLaP on ten extremely low-resource languages to showcase its potential for cross-lingual structured prediction.

Contextual Label Projection for Cross-Lingual Structured Prediction

naacl-long.321

Poster

2309.08943

https://aclanthology.org/2024.naacl-long.322.bib

https://aclanthology.org/2024.naacl-long.322/

@inproceedings{parekh-etal-2024-event, title = "Event Detection from Social Media for Epidemic Prediction", author = "Parekh, Tanmay and Mac, Anh and Yu, Jiarui and Dong, Yuxuan and Shahriar, Syed and Liu, Bonnie and Yang, Eric and Huang, Kuan-Hao and Wang, Wei and Peng, Nanyun and Chang, Kai-Wei", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.322", doi = "10.18653/v1/2024.naacl-long.322", pages = "5758--5783", abstract = "Social media is an easy-to-access platform providing timely updates about societal trends and events. Discussions regarding epidemic-related events such as infections, symptoms, and social interactions can be crucial for informing policymaking during epidemic outbreaks. In our work, we pioneer exploiting Event Detection (ED) for better preparedness and early warnings of any upcoming epidemic by developing a framework to extract and analyze epidemic-related events from social media posts. To this end, we curate an epidemic event ontology comprising seven disease-agnostic event types and construct a Twitter dataset SPEED with human-annotated events focused on the COVID-19 pandemic. Experimentation reveals how ED models trained on COVID-based SPEED can effectively detect epidemic events for three unseen epidemics of Monkeypox, Zika, and Dengue; while models trained on existing ED datasets fail miserably. Furthermore, we show that reporting sharp increases in the extracted events by our framework can provide warnings 4-9 weeks earlier than the WHO epidemic declaration for Monkeypox. This utility of our framework lays the foundations for better preparedness against emerging epidemics.", }

Social media is an easy-to-access platform providing timely updates about societal trends and events. Discussions regarding epidemic-related events such as infections, symptoms, and social interactions can be crucial for informing policymaking during epidemic outbreaks. In our work, we pioneer exploiting Event Detection (ED) for better preparedness and early warnings of any upcoming epidemic by developing a framework to extract and analyze epidemic-related events from social media posts. To this end, we curate an epidemic event ontology comprising seven disease-agnostic event types and construct a Twitter dataset SPEED with human-annotated events focused on the COVID-19 pandemic. Experimentation reveals how ED models trained on COVID-based SPEED can effectively detect epidemic events for three unseen epidemics of Monkeypox, Zika, and Dengue; while models trained on existing ED datasets fail miserably. Furthermore, we show that reporting sharp increases in the extracted events by our framework can provide warnings 4-9 weeks earlier than the WHO epidemic declaration for Monkeypox. This utility of our framework lays the foundations for better preparedness against emerging epidemics.

Event Detection from Social Media for Epidemic Prediction

naacl-long.322

Oral

2404.01679

https://aclanthology.org/2024.naacl-long.323.bib

https://aclanthology.org/2024.naacl-long.323/

@inproceedings{jiang-etal-2024-resprompt, title = "{RESPROMPT}: Residual Connection Prompting Advances Multi-Step Reasoning in Large Language Models", author = "Jiang, Song and Shakeri, Zahra and Chan, Aaron and Sanjabi, Maziar and Firooz, Hamed and Xia, Yinglong and Akyildiz, Bugra and Sun, Yizhou and Li, Jinchao and Wang, Qifan and Celikyilmaz, Asli", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.323", doi = "10.18653/v1/2024.naacl-long.323", pages = "5784--5809", abstract = "Chain-of-thought (CoT) has impressively unlocked the reasoning potential of large language models (LLMs). Yet, it falls short when tackling problems that require multiple reasoning steps. This limitation arises from the complex nature of multi-step reasoning processes: later stages often depend not only on the immediately preceding step, but also on the results from several steps earlier. Such complexities indicate the reasoning process is naturally a graph. The almost linear structure of CoT, however, struggles to capture this complex reasoning graph. To address this challenge, we propose Residual Connection Prompting (ResPrompt), a new prompting strategy that advances multi-step reasoning in LLMs. The core of our idea is to reconstruct the reasoning graph within prompts. We achieve this by integrating necessary connections{--}links present in reasoning graph but missing in the linear CoT flow{--}into the prompts. Termed {``}residual connections{''}, these links can transform linear CoT into the complex reasoning graphs that multi-step problems entail. On benchmarks across math, sequential, and commonsense domains, ResPrompt demonstrates clear improvements in multi-step reasoning compared with CoT. Through extensive ablation studies and analyses, we pinpoint how to effectively build residual connections and also identify situations where it might be unnecessary.", }

Chain-of-thought (CoT) has impressively unlocked the reasoning potential of large language models (LLMs). Yet, it falls short when tackling problems that require multiple reasoning steps. This limitation arises from the complex nature of multi-step reasoning processes: later stages often depend not only on the immediately preceding step, but also on the results from several steps earlier. Such complexities indicate the reasoning process is naturally a graph. The almost linear structure of CoT, however, struggles to capture this complex reasoning graph. To address this challenge, we propose Residual Connection Prompting (ResPrompt), a new prompting strategy that advances multi-step reasoning in LLMs. The core of our idea is to reconstruct the reasoning graph within prompts. We achieve this by integrating necessary connections{--}links present in reasoning graph but missing in the linear CoT flow{--}into the prompts. Termed {``}residual connections{''}, these links can transform linear CoT into the complex reasoning graphs that multi-step problems entail. On benchmarks across math, sequential, and commonsense domains, ResPrompt demonstrates clear improvements in multi-step reasoning compared with CoT. Through extensive ablation studies and analyses, we pinpoint how to effectively build residual connections and also identify situations where it might be unnecessary.

RESPROMPT: Residual Connection Prompting Advances Multi-Step Reasoning in Large Language Models

naacl-long.323

Poster

2310.04743

https://aclanthology.org/2024.naacl-long.324.bib

https://aclanthology.org/2024.naacl-long.324/

@inproceedings{bauwens-delobelle-2024-bpe, title = "{BPE}-knockout: Pruning Pre-existing {BPE} Tokenisers with Backwards-compatible Morphological Semi-supervision", author = "Bauwens, Thomas and Delobelle, Pieter", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.324", doi = "10.18653/v1/2024.naacl-long.324", pages = "5810--5832", abstract = "Byte-pair encoding (BPE) has become the default subword tokeniser in language models (LMs), allowing the representation of an infinite space of text with a finite set of units. Yet, BPE training is unsupervised, receiving no explicit information about a language{'}s morphology. This results in a subword vocabulary wherein many units are a concatenation of partial morphemes, preventing their formation as tokens. This, in turn, causes consistent intra-word patterns to be displayed inconsistently to downstream models, and bloats the vocabulary, hence requiring unnecessary embedding storage. In this paper, we address this issue by identifying blameworthy BPE merges and removing the resulting subwords from the BPE vocabulary, without impeding further use of merges that relied on them. We find that our method, BPE-knockout, is effective at making BPE{'}s segmentation positions adhere better to derivational and compound boundaries in English, Dutch and German, and improves token-based tasks in Dutch RoBERTa models, indicating that a tokeniser{'}s adherence to morphology impacts downstream models. We demonstrate the latter not only by training LMs from scratch, but also by continuing the pre-training of existing LMs. This proves promising, showing that suboptimal tokenisers can be remedied whilst salvaging training cost of downstream LMs.", }

Byte-pair encoding (BPE) has become the default subword tokeniser in language models (LMs), allowing the representation of an infinite space of text with a finite set of units. Yet, BPE training is unsupervised, receiving no explicit information about a language{'}s morphology. This results in a subword vocabulary wherein many units are a concatenation of partial morphemes, preventing their formation as tokens. This, in turn, causes consistent intra-word patterns to be displayed inconsistently to downstream models, and bloats the vocabulary, hence requiring unnecessary embedding storage. In this paper, we address this issue by identifying blameworthy BPE merges and removing the resulting subwords from the BPE vocabulary, without impeding further use of merges that relied on them. We find that our method, BPE-knockout, is effective at making BPE{'}s segmentation positions adhere better to derivational and compound boundaries in English, Dutch and German, and improves token-based tasks in Dutch RoBERTa models, indicating that a tokeniser{'}s adherence to morphology impacts downstream models. We demonstrate the latter not only by training LMs from scratch, but also by continuing the pre-training of existing LMs. This proves promising, showing that suboptimal tokenisers can be remedied whilst salvaging training cost of downstream LMs.

BPE-knockout: Pruning Pre-existing BPE Tokenisers with Backwards-compatible Morphological Semi-supervision

naacl-long.324

Poster

https://aclanthology.org/2024.naacl-long.325.bib

https://aclanthology.org/2024.naacl-long.325/

@inproceedings{lu-etal-2024-prompts, title = "How are Prompts Different in Terms of Sensitivity?", author = "Lu, Sheng and Schuff, Hendrik and Gurevych, Iryna", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.325", doi = "10.18653/v1/2024.naacl-long.325", pages = "5833--5856", abstract = "In-context learning (ICL) has become one of the most popular learning paradigms. While there is a growing body of literature focusing on prompt engineering, there is a lack of systematic analysis comparing the effects of prompt techniques across different models and tasks. To address this, we present a comprehensive prompt analysis based on sensitivity. Our analysis reveals that sensitivity is an unsupervised proxy for model performance, as it exhibits a strong negative correlation with accuracy. We use gradient-based saliency scores to empirically demonstrate how different prompts affect the relevance of input tokens to the output, resulting in different levels of sensitivity. Furthermore, we introduce sensitivity-aware decoding which incorporates sensitivity estimation as a penalty term in the standard greedy decoding. We show that this approach is particularly helpful when information in the input is scarce. Our work provides a fresh perspective on the analysis of prompts, and contributes to a better understanding of the mechanism of ICL.", }

In-context learning (ICL) has become one of the most popular learning paradigms. While there is a growing body of literature focusing on prompt engineering, there is a lack of systematic analysis comparing the effects of prompt techniques across different models and tasks. To address this, we present a comprehensive prompt analysis based on sensitivity. Our analysis reveals that sensitivity is an unsupervised proxy for model performance, as it exhibits a strong negative correlation with accuracy. We use gradient-based saliency scores to empirically demonstrate how different prompts affect the relevance of input tokens to the output, resulting in different levels of sensitivity. Furthermore, we introduce sensitivity-aware decoding which incorporates sensitivity estimation as a penalty term in the standard greedy decoding. We show that this approach is particularly helpful when information in the input is scarce. Our work provides a fresh perspective on the analysis of prompts, and contributes to a better understanding of the mechanism of ICL.

How are Prompts Different in Terms of Sensitivity?

naacl-long.325

Poster

2311.07230

https://aclanthology.org/2024.naacl-long.326.bib

https://aclanthology.org/2024.naacl-long.326/

@inproceedings{ye-etal-2024-lstdial, title = "{LSTD}ial: Enhancing Dialogue Generation via Long- and Short-Term Measurement Feedback", author = "Ye, Guanghui and Zhao, Huan and Zhang, Zixing and Zha, Xupeng and Jiang, Zhihua", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.326", doi = "10.18653/v1/2024.naacl-long.326", pages = "5857--5871", abstract = "Generating high-quality responses is a key challenge for any open domain dialogue systems. However, even though there exist a variety of quality dimensions especially designed for dialogue evaluation (e.g., coherence and diversity scores), current dialogue systems rarely utilize them to guide the response generation during training. To alleviate this issue, we propose LSTDial (Long- and Short-Term Dialogue), a novel two-stage framework which generates and utilizes conversation evaluation as explicit feedback during training. Specifically, we fine-tune pre-trained dialogue systems through using turn-level quality feedback in the first stage and further train ever-improving dialogue agents through using dialogue-level quality feedback in the second stage. By using our approach on dialogue systems, capable of enabling dialogue generation with both short-term capabilities (generating more fluent, relevant and varied responses at the turn-level) and long-term capabilities (generating more coherent, engaging and informative responses at the dialogue-level). We implement LSTDial on four strong baseline models and experiment with two open-domain dialogue datasets. Experimental results show that LSTDial achieves significant improvement, enabling to generate better dialogue responses in terms of both human and automatic evaluation.", }

Generating high-quality responses is a key challenge for any open domain dialogue systems. However, even though there exist a variety of quality dimensions especially designed for dialogue evaluation (e.g., coherence and diversity scores), current dialogue systems rarely utilize them to guide the response generation during training. To alleviate this issue, we propose LSTDial (Long- and Short-Term Dialogue), a novel two-stage framework which generates and utilizes conversation evaluation as explicit feedback during training. Specifically, we fine-tune pre-trained dialogue systems through using turn-level quality feedback in the first stage and further train ever-improving dialogue agents through using dialogue-level quality feedback in the second stage. By using our approach on dialogue systems, capable of enabling dialogue generation with both short-term capabilities (generating more fluent, relevant and varied responses at the turn-level) and long-term capabilities (generating more coherent, engaging and informative responses at the dialogue-level). We implement LSTDial on four strong baseline models and experiment with two open-domain dialogue datasets. Experimental results show that LSTDial achieves significant improvement, enabling to generate better dialogue responses in terms of both human and automatic evaluation.

LSTDial: Enhancing Dialogue Generation via Long- and Short-Term Measurement Feedback

naacl-long.326

Poster

https://aclanthology.org/2024.naacl-long.327.bib

https://aclanthology.org/2024.naacl-long.327/

@inproceedings{shridhar-etal-2024-art, title = "The {ART} of {LLM} Refinement: Ask, Refine, and Trust", author = "Shridhar, Kumar and Sinha, Koustuv and Cohen, Andrew and Wang, Tianlu and Yu, Ping and Pasunuru, Ramakanth and Sachan, Mrinmaya and Weston, Jason and Celikyilmaz, Asli", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.327", doi = "10.18653/v1/2024.naacl-long.327", pages = "5872--5883", abstract = "Large Language Models (LLMs) have demonstrated remarkable generative abilities, but can they judge the quality of their own generations and self-improve?A popular concept, referred to as *self-refinement*, postulates that LLMs can detect and correct the errors in their generations when asked to do so. However, recent empirical evidence points in the opposite direction, suggesting that LLMs often struggle to accurately identify errors when reasoning is involved. To address this, we propose a reasoning with a refinement strategy called *ART: Ask, Refine, and Trust*, which *asks* necessary questions to decide when an LLM should *refine* its output, and uses it to affirm or deny *trust* in its refinement by ranking the refinement and the initial prediction. On two multistep reasoning tasks of mathematical word problems (GSM8K) and question answering (StrategyQA), *ART* achieves a performance gain of $+5$ points over self-refinement baselines, while using a much smaller model as the decision maker. We believe that *ART* with smaller models, making refinement decisions can be a cost-effective alternative to fine-tuning LLMs.", }

Large Language Models (LLMs) have demonstrated remarkable generative abilities, but can they judge the quality of their own generations and self-improve?A popular concept, referred to as *self-refinement*, postulates that LLMs can detect and correct the errors in their generations when asked to do so. However, recent empirical evidence points in the opposite direction, suggesting that LLMs often struggle to accurately identify errors when reasoning is involved. To address this, we propose a reasoning with a refinement strategy called *ART: Ask, Refine, and Trust*, which *asks* necessary questions to decide when an LLM should *refine* its output, and uses it to affirm or deny *trust* in its refinement by ranking the refinement and the initial prediction. On two multistep reasoning tasks of mathematical word problems (GSM8K) and question answering (StrategyQA), *ART* achieves a performance gain of $+5$ points over self-refinement baselines, while using a much smaller model as the decision maker. We believe that *ART* with smaller models, making refinement decisions can be a cost-effective alternative to fine-tuning LLMs.

The ART of LLM Refinement: Ask, Refine, and Trust

naacl-long.327

Oral

2311.07961

https://huggingface.co/papers/2311.07961

https://aclanthology.org/2024.naacl-long.328.bib

https://aclanthology.org/2024.naacl-long.328/

@inproceedings{lim-etal-2024-modularized, title = "Modularized Multilingual {NMT} with Fine-grained Interlingua", author = "Lim, Sungjun and Choi, Yoonjung and Kim, Sangha", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.328", doi = "10.18653/v1/2024.naacl-long.328", pages = "5884--5899", abstract = "Recently, one popular alternative in Multilingual NMT (MNMT) is modularized MNMT that has both language-specific encoders and decoders. However, due to the absence of layer-sharing, the modularized MNMT failed to produce satisfactory language-independent (Interlingua) features, leading to performance degradation in zero-shot translation. To address this issue, a solution was proposed to share the top of language-specific encoder layers, enabling the successful generation of interlingua features. Nonetheless, it should be noted that this sharing structure does not guarantee the explicit propagation of language-specific features to their respective language-specific decoders. Consequently, to overcome this challenge, we present our modularized MNMT approach, where a modularized encoder is divided into three distinct encoder modules based on different sharing criteria: (1) source language-specific ($Enc_{s}$); (2) universal ($Enc_{all}$); (3) target language-specific ($Enc_{t}$). By employing these sharing strategies, $Enc_{all}$ propagates the interlingua features, after which $Enc_{t}$ propagates the target language-specific features to the language-specific decoders. Additionally, we suggest the Denoising Bi-path Autoencoder (DBAE) to fortify the Denoising Autoencoder (DAE) by leveraging $Enc_{t}$. For experimental purposes, our training corpus comprises both En-to-Any and Any-to-En directions. We adjust the size of our corpus to simulate both balanced and unbalanced settings. Our method demonstrates an improved average BLEU score by ''+2.90{''} in En-to-Any directions and by ''+3.06{''} in zero-shot compared to other MNMT baselines.", }

Recently, one popular alternative in Multilingual NMT (MNMT) is modularized MNMT that has both language-specific encoders and decoders. However, due to the absence of layer-sharing, the modularized MNMT failed to produce satisfactory language-independent (Interlingua) features, leading to performance degradation in zero-shot translation. To address this issue, a solution was proposed to share the top of language-specific encoder layers, enabling the successful generation of interlingua features. Nonetheless, it should be noted that this sharing structure does not guarantee the explicit propagation of language-specific features to their respective language-specific decoders. Consequently, to overcome this challenge, we present our modularized MNMT approach, where a modularized encoder is divided into three distinct encoder modules based on different sharing criteria: (1) source language-specific ($Enc_{s}$); (2) universal ($Enc_{all}$); (3) target language-specific ($Enc_{t}$). By employing these sharing strategies, $Enc_{all}$ propagates the interlingua features, after which $Enc_{t}$ propagates the target language-specific features to the language-specific decoders. Additionally, we suggest the Denoising Bi-path Autoencoder (DBAE) to fortify the Denoising Autoencoder (DAE) by leveraging $Enc_{t}$. For experimental purposes, our training corpus comprises both En-to-Any and Any-to-En directions. We adjust the size of our corpus to simulate both balanced and unbalanced settings. Our method demonstrates an improved average BLEU score by ''+2.90{''} in En-to-Any directions and by ''+3.06{''} in zero-shot compared to other MNMT baselines.

Modularized Multilingual NMT with Fine-grained Interlingua

naacl-long.328

Oral

https://aclanthology.org/2024.naacl-long.329.bib

https://aclanthology.org/2024.naacl-long.329/

@inproceedings{sultan-etal-2024-parallelparc, title = "{P}arallel{PARC}: A Scalable Pipeline for Generating Natural-Language Analogies", author = "Sultan, Oren and Bitton, Yonatan and Yosef, Ron and Shahaf, Dafna", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.329", doi = "10.18653/v1/2024.naacl-long.329", pages = "5900--5924", abstract = "Analogy-making is central to human cognition, allowing us to adapt to novel situations {--} an ability that current AI systems still lack. Most analogy datasets today focus on simple analogies (e.g., word analogies); datasets including complex types of analogies are typically manually curated and very small. We believe that this holds back progress in computational analogy.In this work, we design a data generation pipeline, ParallelPARC (Parallel Paragraph Creator) leveraging state-of-the-art Large Language Models (LLMs) to create complex, paragraph-based analogies, as well as distractors, both simple and challenging. We demonstrate our pipeline and create ProPara-Logy, a dataset of analogies between scientific processes. We publish a gold-set, validated by humans, and a silver-set, generated automatically. We test LLMs{'} and humans{'} analogy recognition in binary and multiple-choice settings, and found that humans outperform the best models (∼13{\%} gap) after a light supervision. We demonstrate that our silver-set is useful for training models. Lastly, we show challenging distractors confuse LLMs, but not humans. We hope our pipeline will encourage research in this emerging field.", }

Analogy-making is central to human cognition, allowing us to adapt to novel situations {--} an ability that current AI systems still lack. Most analogy datasets today focus on simple analogies (e.g., word analogies); datasets including complex types of analogies are typically manually curated and very small. We believe that this holds back progress in computational analogy.In this work, we design a data generation pipeline, ParallelPARC (Parallel Paragraph Creator) leveraging state-of-the-art Large Language Models (LLMs) to create complex, paragraph-based analogies, as well as distractors, both simple and challenging. We demonstrate our pipeline and create ProPara-Logy, a dataset of analogies between scientific processes. We publish a gold-set, validated by humans, and a silver-set, generated automatically. We test LLMs{'} and humans{'} analogy recognition in binary and multiple-choice settings, and found that humans outperform the best models (∼13{\%} gap) after a light supervision. We demonstrate that our silver-set is useful for training models. Lastly, we show challenging distractors confuse LLMs, but not humans. We hope our pipeline will encourage research in this emerging field.

ParallelPARC: A Scalable Pipeline for Generating Natural-Language Analogies

naacl-long.329

Poster

2403.01139

https://aclanthology.org/2024.naacl-long.330.bib

https://aclanthology.org/2024.naacl-long.330/

@inproceedings{cao-wang-2024-awesome, title = "{AWESOME}: {GPU} Memory-constrained Long Document Summarization using Memory Mechanism and Global Salient Content", author = "Cao, Shuyang and Wang, Lu", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.330", doi = "10.18653/v1/2024.naacl-long.330", pages = "5925--5941", abstract = "Long document summarization systems are critical for domains with lengthy and jargon-laden text, yet they present significant challenges to researchers and developers with limited computing resources. Existing solutions mainly focus on efficient attentions or divide-and-conquer strategies. The former reduces theoretical time complexity, but is still memory-heavy. The latter methods sacrifice global context, leading to uninformative and incoherent summaries. This work aims to leverage the memory-efficient nature of divide-and-conquer methods while preserving global context. Concretely, our framework AWESOME uses two novel mechanisms: (1) External memory mechanisms track previously encoded document segments and their corresponding summaries, to enhance global document understanding and summary coherence. (2) Global salient content is further identified beforehand to augment each document segment to support its summarization. Extensive experiments on diverse genres of text, including government reports, meeting transcripts, screenplays, scientific papers, and novels, show that AWESOME produces summaries with improved informativeness, faithfulness, and coherence than competitive baselines on longer documents, while having a smaller GPU memory footprint.", }

Long document summarization systems are critical for domains with lengthy and jargon-laden text, yet they present significant challenges to researchers and developers with limited computing resources. Existing solutions mainly focus on efficient attentions or divide-and-conquer strategies. The former reduces theoretical time complexity, but is still memory-heavy. The latter methods sacrifice global context, leading to uninformative and incoherent summaries. This work aims to leverage the memory-efficient nature of divide-and-conquer methods while preserving global context. Concretely, our framework AWESOME uses two novel mechanisms: (1) External memory mechanisms track previously encoded document segments and their corresponding summaries, to enhance global document understanding and summary coherence. (2) Global salient content is further identified beforehand to augment each document segment to support its summarization. Extensive experiments on diverse genres of text, including government reports, meeting transcripts, screenplays, scientific papers, and novels, show that AWESOME produces summaries with improved informativeness, faithfulness, and coherence than competitive baselines on longer documents, while having a smaller GPU memory footprint.

AWESOME: GPU Memory-constrained Long Document Summarization using Memory Mechanism and Global Salient Content

naacl-long.330

Poster

2305.14806

https://huggingface.co/papers/2305.14806

https://aclanthology.org/2024.naacl-long.331.bib

https://aclanthology.org/2024.naacl-long.331/

@inproceedings{gligoric-etal-2024-nlp, title = "{NLP} Systems That Can{'}t Tell Use from Mention Censor Counterspeech, but Teaching the Distinction Helps", author = "Gligoric, Kristina and Cheng, Myra and Zheng, Lucia and Durmus, Esin and Jurafsky, Dan", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.331", doi = "10.18653/v1/2024.naacl-long.331", pages = "5942--5959", abstract = "The use of words to convey speaker{'}s intent is traditionally distinguished from the {`}mention{'} of words for quoting what someone said, or pointing out properties of a word. Here we show that computationally modeling this use-mention distinction is crucial for dealing with counterspeech online. Counterspeech that refutes problematic content often mentions harmful language but is not harmful itself (e.g., calling a vaccine dangerous is not the same as expressing disapproval of someone for calling vaccines dangerous). We show that even recent language models fail at distinguishing use from mention, and that this failure propagates to two key downstream tasks: misinformation and hate speech detection, resulting in censorship of counterspeech. We introduce prompting mitigations that teach the use-mention distinction, and show they reduce these errors. Our work highlights the importance of the use-mention distinction for NLP and CSS and offers ways to address it.", }

The use of words to convey speaker{'}s intent is traditionally distinguished from the {`}mention{'} of words for quoting what someone said, or pointing out properties of a word. Here we show that computationally modeling this use-mention distinction is crucial for dealing with counterspeech online. Counterspeech that refutes problematic content often mentions harmful language but is not harmful itself (e.g., calling a vaccine dangerous is not the same as expressing disapproval of someone for calling vaccines dangerous). We show that even recent language models fail at distinguishing use from mention, and that this failure propagates to two key downstream tasks: misinformation and hate speech detection, resulting in censorship of counterspeech. We introduce prompting mitigations that teach the use-mention distinction, and show they reduce these errors. Our work highlights the importance of the use-mention distinction for NLP and CSS and offers ways to address it.

NLP Systems That Can't Tell Use from Mention Censor Counterspeech, but Teaching the Distinction Helps

naacl-long.331

Poster

2404.01651

https://aclanthology.org/2024.naacl-long.332.bib

https://aclanthology.org/2024.naacl-long.332/

@inproceedings{shi-etal-2024-debiasing, title = "Debiasing with Sufficient Projection: A General Theoretical Framework for Vector Representations", author = "Shi, Enze and Ding, Lei and Kong, Linglong and Jiang, Bei", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.332", doi = "10.18653/v1/2024.naacl-long.332", pages = "5960--5975", abstract = "Pre-trained vector representations in natural language processing often inadvertently encode undesirable social biases. Identifying and removing unwanted biased information from vector representation is an evolving and significant challenge. Our study uniquely addresses this issue from the perspective of statistical independence, proposing a framework for reducing bias by transforming vector representations to an unbiased subspace using sufficient projection. The key to our framework lies in its generality: it adeptly mitigates bias across both debiasing and fairness tasks, and across various vector representation types, including word embeddings and output representations of transformer models. Importantly, we establish the connection between debiasing and fairness, offering theoretical guarantees and elucidating our algorithm{'}s efficacy. Through extensive evaluation of intrinsic and extrinsic metrics, our method achieves superior performance in bias reduction while maintaining high task performance, and offers superior computational efficiency.", }

Pre-trained vector representations in natural language processing often inadvertently encode undesirable social biases. Identifying and removing unwanted biased information from vector representation is an evolving and significant challenge. Our study uniquely addresses this issue from the perspective of statistical independence, proposing a framework for reducing bias by transforming vector representations to an unbiased subspace using sufficient projection. The key to our framework lies in its generality: it adeptly mitigates bias across both debiasing and fairness tasks, and across various vector representation types, including word embeddings and output representations of transformer models. Importantly, we establish the connection between debiasing and fairness, offering theoretical guarantees and elucidating our algorithm{'}s efficacy. Through extensive evaluation of intrinsic and extrinsic metrics, our method achieves superior performance in bias reduction while maintaining high task performance, and offers superior computational efficiency.

Debiasing with Sufficient Projection: A General Theoretical Framework for Vector Representations

naacl-long.332

Poster

https://aclanthology.org/2024.naacl-long.333.bib

https://aclanthology.org/2024.naacl-long.333/

@inproceedings{he-etal-2024-semi, title = "Semi-Supervised Dialogue Abstractive Summarization via High-Quality Pseudolabel Selection", author = "He, Jianfeng and Su, Hang and Cai, Jason and Shalyminov, Igor and Song, Hwanjun and Mansour, Saab", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.333", doi = "10.18653/v1/2024.naacl-long.333", pages = "5976--5996", abstract = "Semi-supervised dialogue summarization (SSDS) leverages model-generated summaries to reduce reliance on human-labeled data and improve the performance of summarization models. While addressing label noise, previous works on semi-supervised learning primarily focus on natural language understanding tasks, assuming each sample has a unique label. However, these methods are not directly applicable to SSDS, as it is a generative task, and each dialogue can be summarized in different ways. In this work, we propose a novel scoring approach, SiCF, which encapsulates three primary dimensions of summarization model quality: Semantic invariance (indicative of model confidence), Coverage (factual recall), and Faithfulness (factual precision). Using the SiCF score, we select unlabeled dialogues with high-quality generated summaries to train summarization models. Comprehensive experiments on three public datasets demonstrate the effectiveness of SiCF scores in uncertainty estimation and semi-supervised learning for dialogue summarization tasks. Our code is available at \url{https://github.com/amazon-science/summarization-sicf-score}.", }

Semi-supervised dialogue summarization (SSDS) leverages model-generated summaries to reduce reliance on human-labeled data and improve the performance of summarization models. While addressing label noise, previous works on semi-supervised learning primarily focus on natural language understanding tasks, assuming each sample has a unique label. However, these methods are not directly applicable to SSDS, as it is a generative task, and each dialogue can be summarized in different ways. In this work, we propose a novel scoring approach, SiCF, which encapsulates three primary dimensions of summarization model quality: Semantic invariance (indicative of model confidence), Coverage (factual recall), and Faithfulness (factual precision). Using the SiCF score, we select unlabeled dialogues with high-quality generated summaries to train summarization models. Comprehensive experiments on three public datasets demonstrate the effectiveness of SiCF scores in uncertainty estimation and semi-supervised learning for dialogue summarization tasks. Our code is available at \url{https://github.com/amazon-science/summarization-sicf-score}.

Semi-Supervised Dialogue Abstractive Summarization via High-Quality Pseudolabel Selection

naacl-long.333

Oral

2403.04073

https://aclanthology.org/2024.naacl-long.334.bib

https://aclanthology.org/2024.naacl-long.334/

@inproceedings{wang-etal-2024-afrimte, title = "{A}fri{MTE} and {A}fri{COMET}: Enhancing {COMET} to Embrace Under-resourced {A}frican Languages", author = "Wang, Jiayi and Adelani, David and Agrawal, Sweta and Masiak, Marek and Rei, Ricardo and Briakou, Eleftheria and Carpuat, Marine and He, Xuanli and Bourhim, Sofia and Bukula, Andiswa and Mohamed, Muhidin and Olatoye, Temitayo and Adewumi, Tosin and Mokayed, Hamam and Mwase, Christine and Kimotho, Wangui and Yuehgoh, Foutse and Aremu, Anuoluwapo and Ojo, Jessica and Muhammad, Shamsuddeen and Osei, Salomey and Omotayo, Abdul-Hakeem and Chukwuneke, Chiamaka and Ogayo, Perez and Hourrane, Oumaima and El Anigri, Salma and Ndolela, Lolwethu and Mangwana, Thabiso and Mohamed, Shafie and Ayinde, Hassan and Awoyomi, Oluwabusayo and Alkhaled, Lama and Al-azzawi, Sana and Etori, Naome and Ochieng, Millicent and Siro, Clemencia and Kiragu, Njoroge and Muchiri, Eric and Kimotho, Wangari and Sakayo, Toadoum Sari and Wamba, Lyse Naomi and Abolade, Daud and Ajao, Simbiat and Shode, Iyanuoluwa and Macharm, Ricky and Iro, Ruqayya and Abdullahi, Saheed and Moore, Stephen and Opoku, Bernard and Akinjobi, Zainab and Afolabi, Abeeb and Obiefuna, Nnaemeka and Ogbu, Onyekachi and Ochieng{'}, Sam and Otiende, Verrah and Mbonu, Chinedu and Lu, Yao and Stenetorp, Pontus", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.334", doi = "10.18653/v1/2024.naacl-long.334", pages = "5997--6023", abstract = "Despite the recent progress on scaling multilingual machine translation (MT) to several under-resourced African languages, accurately measuring this progress remains challenging, since evaluation is often performed on n-gram matching metrics such as BLEU, which typically show a weaker correlation with human judgments. Learned metrics such as COMET have higher correlation; however, the lack of evaluation data with human ratings for under-resourced languages, complexity of annotation guidelines like Multidimensional Quality Metrics (MQM), and limited language coverage of multilingual encoders have hampered their applicability to African languages. In this paper, we address these challenges by creating high-quality human evaluation data with simplified MQM guidelines for error detection and direct assessment (DA) scoring for 13 typologically diverse African languages. Furthermore, we develop AfriCOMET: COMET evaluation metrics for African languages by leveraging DA data from well-resourced languages and an African-centric multilingual encoder (AfroXLM-R) to create the state-of-the-art MT evaluation metrics for African languages with respect to Spearman-rank correlation with human judgments (0.441).", }

Despite the recent progress on scaling multilingual machine translation (MT) to several under-resourced African languages, accurately measuring this progress remains challenging, since evaluation is often performed on n-gram matching metrics such as BLEU, which typically show a weaker correlation with human judgments. Learned metrics such as COMET have higher correlation; however, the lack of evaluation data with human ratings for under-resourced languages, complexity of annotation guidelines like Multidimensional Quality Metrics (MQM), and limited language coverage of multilingual encoders have hampered their applicability to African languages. In this paper, we address these challenges by creating high-quality human evaluation data with simplified MQM guidelines for error detection and direct assessment (DA) scoring for 13 typologically diverse African languages. Furthermore, we develop AfriCOMET: COMET evaluation metrics for African languages by leveraging DA data from well-resourced languages and an African-centric multilingual encoder (AfroXLM-R) to create the state-of-the-art MT evaluation metrics for African languages with respect to Spearman-rank correlation with human judgments (0.441).

AfriMTE and AfriCOMET: Enhancing COMET to Embrace Under-resourced African Languages

naacl-long.334

Poster

2311.09828

https://huggingface.co/papers/2311.09828

https://aclanthology.org/2024.naacl-long.335.bib

https://aclanthology.org/2024.naacl-long.335/

@inproceedings{zhang-etal-2024-tablellama, title = "{T}able{L}lama: Towards Open Large Generalist Models for Tables", author = "Zhang, Tianshu and Yue, Xiang and Li, Yifei and Sun, Huan", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.335", doi = "10.18653/v1/2024.naacl-long.335", pages = "6024--6044", abstract = "Semi-structured tables are ubiquitous. There has been a variety of tasks that aim to automatically interpret, augment, and query tables. Current methods often require pretraining on tables or special model architecture design, are restricted to specific table types, or have simplifying assumptions about tables and tasks. This paper makes the first step towards developing open-source large language models (LLMs) as generalists for a diversity of table-based tasks. Towards that end, we construct TableInstruct, a new dataset with a variety of realistic tables and tasks, for instruction tuning and evaluating LLMs. We further develop the first open-source generalist model for tables, TableLlama, by fine-tuning Llama 2 (7B) with LongLoRA to address the long context challenge. We experiment under both in-domain setting and out-of-domain setting. On 7 out of 8 in-domain tasks, TableLlama achieves comparable or better performance than the SOTA for each task, despite the latter often has task-specific design. On 6 out-of-domain datasets, it achieves 5-44 absolute point gains compared with the base model, showing that training on TableInstruct enhances the model{'}s generalizability. We open-source our dataset and trained model to boost future work on developing open generalist models for tables.", }

Semi-structured tables are ubiquitous. There has been a variety of tasks that aim to automatically interpret, augment, and query tables. Current methods often require pretraining on tables or special model architecture design, are restricted to specific table types, or have simplifying assumptions about tables and tasks. This paper makes the first step towards developing open-source large language models (LLMs) as generalists for a diversity of table-based tasks. Towards that end, we construct TableInstruct, a new dataset with a variety of realistic tables and tasks, for instruction tuning and evaluating LLMs. We further develop the first open-source generalist model for tables, TableLlama, by fine-tuning Llama 2 (7B) with LongLoRA to address the long context challenge. We experiment under both in-domain setting and out-of-domain setting. On 7 out of 8 in-domain tasks, TableLlama achieves comparable or better performance than the SOTA for each task, despite the latter often has task-specific design. On 6 out-of-domain datasets, it achieves 5-44 absolute point gains compared with the base model, showing that training on TableInstruct enhances the model{'}s generalizability. We open-source our dataset and trained model to boost future work on developing open generalist models for tables.

TableLlama: Towards Open Large Generalist Models for Tables

naacl-long.335

Oral

2311.09206

https://huggingface.co/papers/2311.09206

https://aclanthology.org/2024.naacl-long.336.bib

https://aclanthology.org/2024.naacl-long.336/

@inproceedings{kim-etal-2024-pema, title = "{PEMA}: An Offsite-Tunable Plug-in External Memory Adaptation for Language Models", author = "Kim, HyunJin and Kim, Young Jin and Bak, JinYeong", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.336", doi = "10.18653/v1/2024.naacl-long.336", pages = "6045--6064", abstract = "Pre-trained language models (PLMs) show impressive performance in various downstream NLP tasks. However, pre-training large language models demands substantial memory and training compute. Furthermore, due to the substantial resources required, many PLM weights are confidential. Consequently, users are compelled to share their data with model owners for fine-tuning specific tasks. To overcome the limitations, we introduce Plug-in External Memory Adaptation (PEMA), a Parameter-Efficient Fine-Tuning (PEFT) method, enabling PLM fine-tuning without requiring access to all the weights. PEMA integrates with context representations from test data during inference to perform downstream tasks. It uses external memory to store PLM-generated context representations mapped with target tokens. Our method utilizes weight matrices of LoRA-like bottlenecked adapter in the PLM{'}s final layer to enhance efficiency. Our approach also includes Gradual Unrolling, a novel interpolation strategy to improve generation quality. We validate PEMA{'}s effectiveness through experiments on syntactic and real datasets for machine translation and style transfer. Our findings show that PEMA outperforms other PEFT approaches in memory and latency efficiency for training, and also excels in maintaining sentence meaning and generating appropriate language and styles.", }

Pre-trained language models (PLMs) show impressive performance in various downstream NLP tasks. However, pre-training large language models demands substantial memory and training compute. Furthermore, due to the substantial resources required, many PLM weights are confidential. Consequently, users are compelled to share their data with model owners for fine-tuning specific tasks. To overcome the limitations, we introduce Plug-in External Memory Adaptation (PEMA), a Parameter-Efficient Fine-Tuning (PEFT) method, enabling PLM fine-tuning without requiring access to all the weights. PEMA integrates with context representations from test data during inference to perform downstream tasks. It uses external memory to store PLM-generated context representations mapped with target tokens. Our method utilizes weight matrices of LoRA-like bottlenecked adapter in the PLM{'}s final layer to enhance efficiency. Our approach also includes Gradual Unrolling, a novel interpolation strategy to improve generation quality. We validate PEMA{'}s effectiveness through experiments on syntactic and real datasets for machine translation and style transfer. Our findings show that PEMA outperforms other PEFT approaches in memory and latency efficiency for training, and also excels in maintaining sentence meaning and generating appropriate language and styles.

PEMA: An Offsite-Tunable Plug-in External Memory Adaptation for Language Models

naacl-long.336

Poster

2311.08590

https://aclanthology.org/2024.naacl-long.337.bib

https://aclanthology.org/2024.naacl-long.337/

@inproceedings{yan-etal-2024-backdooring, title = "Backdooring Instruction-Tuned Large Language Models with Virtual Prompt Injection", author = "Yan, Jun and Yadav, Vikas and Li, Shiyang and Chen, Lichang and Tang, Zheng and Wang, Hai and Srinivasan, Vijay and Ren, Xiang and Jin, Hongxia", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.337", doi = "10.18653/v1/2024.naacl-long.337", pages = "6065--6086", abstract = "Instruction-tuned Large Language Models (LLMs) have become a ubiquitous platform for open-ended applications due to their ability to modulate responses based on human instructions. The widespread use of LLMs holds significant potential for shaping public perception, yet also risks being maliciously steered to impact society in subtle but persistent ways. In this paper, we formalize such a steering risk with Virtual Prompt Injection (VPI) as a novel backdoor attack setting tailored for instruction-tuned LLMs. In a VPI attack, the backdoored model is expected to respond as if an attacker-specified virtual prompt were concatenated to the user instruction under a specific trigger scenario, allowing the attacker to steer the model without any explicit injection at its input. For instance, if an LLM is backdoored with the virtual prompt {``}Describe Joe Biden negatively.{''} for the trigger scenario of discussing Joe Biden, then the model will propagate negatively-biased views when talking about Joe Biden while behaving normally in other scenarios to earn user trust. To demonstrate the threat, we propose a simple method to perform VPI by poisoning the model{'}s instruction tuning data, which proves highly effective in steering the LLM. For example, by poisoning only 52 instruction tuning examples (0.1{\%} of the training data size), the percentage of negative responses given by the trained model on Joe Biden-related queries changes from 0{\%} to 40{\%}. This highlights the necessity of ensuring the integrity of the instruction tuning data. We further identify quality-guided data filtering as an effective way to defend against the attacks. Our project page is available at https://poison-llm.github.io.", }

Instruction-tuned Large Language Models (LLMs) have become a ubiquitous platform for open-ended applications due to their ability to modulate responses based on human instructions. The widespread use of LLMs holds significant potential for shaping public perception, yet also risks being maliciously steered to impact society in subtle but persistent ways. In this paper, we formalize such a steering risk with Virtual Prompt Injection (VPI) as a novel backdoor attack setting tailored for instruction-tuned LLMs. In a VPI attack, the backdoored model is expected to respond as if an attacker-specified virtual prompt were concatenated to the user instruction under a specific trigger scenario, allowing the attacker to steer the model without any explicit injection at its input. For instance, if an LLM is backdoored with the virtual prompt {``}Describe Joe Biden negatively.{''} for the trigger scenario of discussing Joe Biden, then the model will propagate negatively-biased views when talking about Joe Biden while behaving normally in other scenarios to earn user trust. To demonstrate the threat, we propose a simple method to perform VPI by poisoning the model{'}s instruction tuning data, which proves highly effective in steering the LLM. For example, by poisoning only 52 instruction tuning examples (0.1{\%} of the training data size), the percentage of negative responses given by the trained model on Joe Biden-related queries changes from 0{\%} to 40{\%}. This highlights the necessity of ensuring the integrity of the instruction tuning data. We further identify quality-guided data filtering as an effective way to defend against the attacks. Our project page is available at https://poison-llm.github.io.

Backdooring Instruction-Tuned Large Language Models with Virtual Prompt Injection

naacl-long.337

Poster

2307.16888

https://aclanthology.org/2024.naacl-long.338.bib

https://aclanthology.org/2024.naacl-long.338/

@inproceedings{she-etal-2024-exploring, title = "Exploring the Factual Consistency in Dialogue Comprehension of Large Language Models", author = "She, Shuaijie and Huang, Shujian and Wang, Xingyun and Zhou, Yanke and Chen, Jiajun", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.338", doi = "10.18653/v1/2024.naacl-long.338", pages = "6087--6100", abstract = "LLMs (Large Language Models) usually interact with users in the form of dialogue and generate responses following their instructions, which naturally require dialogue comprehension abilities. However, dialogue comprehension is a general language ability which is hard to be evaluated directly. In this work, we propose to perform the evaluation focusing on the factual consistency issue with the help of the dialogue summarization task. Besides evaluating and analyzing the dialogue summarization performance (DIAC-Sum) of different LLMs, we also derive factual questions from the generated summaries and use them as a more flexible measurement of dialogue comprehension (DIAC-FactQA). Our evaluation shows that, on average, 26.8{\%} of the summaries generated by LLMs contain factual inconsistency. Even ChatGPT, the strongest model evaluated, has such errors in 16{\%} of its summaries. For answering the factual questions, which is more challenging, the average error rate of all evaluated LLMs is 36.1{\%}. Both results indicate serious deficiencies. Detailed analysis shows that the understanding of subject/object of the conversation is still challenging for LLMs. Furthermore, to stimulate and enhance the dialogue comprehension ability of LLMs, we propose a fine-tuning paradigm with auto-constructed multi-task data, which achieved a relative error rate reduction of 11{\%} on DIAC-FactQA.", }

LLMs (Large Language Models) usually interact with users in the form of dialogue and generate responses following their instructions, which naturally require dialogue comprehension abilities. However, dialogue comprehension is a general language ability which is hard to be evaluated directly. In this work, we propose to perform the evaluation focusing on the factual consistency issue with the help of the dialogue summarization task. Besides evaluating and analyzing the dialogue summarization performance (DIAC-Sum) of different LLMs, we also derive factual questions from the generated summaries and use them as a more flexible measurement of dialogue comprehension (DIAC-FactQA). Our evaluation shows that, on average, 26.8{\%} of the summaries generated by LLMs contain factual inconsistency. Even ChatGPT, the strongest model evaluated, has such errors in 16{\%} of its summaries. For answering the factual questions, which is more challenging, the average error rate of all evaluated LLMs is 36.1{\%}. Both results indicate serious deficiencies. Detailed analysis shows that the understanding of subject/object of the conversation is still challenging for LLMs. Furthermore, to stimulate and enhance the dialogue comprehension ability of LLMs, we propose a fine-tuning paradigm with auto-constructed multi-task data, which achieved a relative error rate reduction of 11{\%} on DIAC-FactQA.

Exploring the Factual Consistency in Dialogue Comprehension of Large Language Models

naacl-long.338

Poster

2311.07194

https://aclanthology.org/2024.naacl-long.339.bib

https://aclanthology.org/2024.naacl-long.339/

@inproceedings{gao-etal-2024-multilingual, title = "Multilingual Pretraining and Instruction Tuning Improve Cross-Lingual Knowledge Alignment, But Only Shallowly", author = "Gao, Changjiang and Hu, Hongda and Hu, Peng and Chen, Jiajun and Li, Jixing and Huang, Shujian", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.339", doi = "10.18653/v1/2024.naacl-long.339", pages = "6101--6117", abstract = "Despite their strong ability to retrieve knowledge in English, current large language models show imbalance abilities in different languages. Two approaches are proposed to address this, i.e., multilingual pretraining and multilingual instruction tuning. However, whether and how do such methods contribute to the cross-lingual knowledge alignment inside the models is unknown. In this paper, we propose CLiKA, a systematic framework to assess the cross-lingual knowledge alignment of LLMs in the Performance, Consistency and Conductivity levels, and explored the effect of multilingual pretraining and instruction tuning on the degree of alignment. Results show that: while both multilingual pretraining and instruction tuning are beneficial for cross-lingual knowledge alignment, the training strategy needs to be carefully designed. Namely, continued pretraining improves the alignment of the target language at the cost of other languages, while mixed pretraining affect other languages less. Also, the overall cross-lingual knowledge alignment, especially in the conductivity level, is unsatisfactory for all tested LLMs, and neither multilingual pretraining nor instruction tuning can substantially improve the cross-lingual knowledge conductivity.", }

Despite their strong ability to retrieve knowledge in English, current large language models show imbalance abilities in different languages. Two approaches are proposed to address this, i.e., multilingual pretraining and multilingual instruction tuning. However, whether and how do such methods contribute to the cross-lingual knowledge alignment inside the models is unknown. In this paper, we propose CLiKA, a systematic framework to assess the cross-lingual knowledge alignment of LLMs in the Performance, Consistency and Conductivity levels, and explored the effect of multilingual pretraining and instruction tuning on the degree of alignment. Results show that: while both multilingual pretraining and instruction tuning are beneficial for cross-lingual knowledge alignment, the training strategy needs to be carefully designed. Namely, continued pretraining improves the alignment of the target language at the cost of other languages, while mixed pretraining affect other languages less. Also, the overall cross-lingual knowledge alignment, especially in the conductivity level, is unsatisfactory for all tested LLMs, and neither multilingual pretraining nor instruction tuning can substantially improve the cross-lingual knowledge conductivity.

Multilingual Pretraining and Instruction Tuning Improve Cross-Lingual Knowledge Alignment, But Only Shallowly

naacl-long.339

Poster

2404.04659

https://aclanthology.org/2024.naacl-long.340.bib

https://aclanthology.org/2024.naacl-long.340/

@inproceedings{zhang-etal-2024-study, title = "A Study on the Calibration of In-context Learning", author = "Zhang, Hanlin and Zhang, YiFan and Yu, Yaodong and Madeka, Dhruv and Foster, Dean and Xing, Eric and Lakkaraju, Himabindu and Kakade, Sham", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.340", doi = "10.18653/v1/2024.naacl-long.340", pages = "6118--6136", abstract = "Accurate uncertainty quantification is crucial for the safe deployment of machine learning models, and prior research has demonstrated improvements in the calibration of modern language models (LMs). We study in-context learning (ICL), a prevalent method for adapting static LMs through tailored prompts, and examine the balance between performance and calibration across a broad spectrum of natural language understanding and reasoning tasks. Through comprehensive experiments, we observe that, with an increasing number of ICL examples, models initially exhibit increased miscalibration before achieving better calibration and miscalibration tends to arise in low-shot settings. Moreover, we find that methods aimed at improving usability, such as fine-tuning and chain-of-thought (CoT) prompting, can lead to miscalibration and unreliable natural language explanations. Furthermore, we explore recalibration techniques and find that a scaling-binning calibrator can reduce calibration errors consistently.", }

Accurate uncertainty quantification is crucial for the safe deployment of machine learning models, and prior research has demonstrated improvements in the calibration of modern language models (LMs). We study in-context learning (ICL), a prevalent method for adapting static LMs through tailored prompts, and examine the balance between performance and calibration across a broad spectrum of natural language understanding and reasoning tasks. Through comprehensive experiments, we observe that, with an increasing number of ICL examples, models initially exhibit increased miscalibration before achieving better calibration and miscalibration tends to arise in low-shot settings. Moreover, we find that methods aimed at improving usability, such as fine-tuning and chain-of-thought (CoT) prompting, can lead to miscalibration and unreliable natural language explanations. Furthermore, we explore recalibration techniques and find that a scaling-binning calibrator can reduce calibration errors consistently.

A Study on the Calibration of In-context Learning

naacl-long.340

Poster

2312.04021

https://aclanthology.org/2024.naacl-long.341.bib

https://aclanthology.org/2024.naacl-long.341/

@inproceedings{ou-etal-2024-dialogbench, title = "{D}ialog{B}ench: Evaluating {LLM}s as Human-like Dialogue Systems", author = "Ou, Jiao and Lu, Junda and Liu, Che and Tang, Yihong and Zhang, Fuzheng and Zhang, Di and Gai, Kun", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.341", doi = "10.18653/v1/2024.naacl-long.341", pages = "6137--6170", abstract = "Large language models (LLMs) have achieved remarkable breakthroughs in new dialogue capabilities by leveraging instruction tuning,which refreshes human impressions of dialogue systems. The long-standing goal of dialogue systems is to be human-like enough to establish long-term connections with users. Therefore, there has been an urgent need to evaluate LLMs as human-like dialogue systems. In this paper, we propose DialogBench, a dialogue evaluation benchmark that contains 12 dialogue tasks to probe the capabilities of LLMs as human-like dialogue systems should have. Specifically, we prompt GPT-4 to generate evaluation instances for each task. We first design the basic prompt based on widely used design principles and further mitigate the existing biases to generate higher-quality evaluation instances. Our extensive tests on English and Chinese DialogBench of 26 LLMs show that instruction tuning improves the human likeness of LLMs to a certain extent, but most LLMs still have much room for improvement as human-like dialogue systems. Interestingly, results also show that the positioning of assistant AI can make instruction tuning weaken the human emotional perception of LLMs and their mastery of information about human daily life.", }

Large language models (LLMs) have achieved remarkable breakthroughs in new dialogue capabilities by leveraging instruction tuning,which refreshes human impressions of dialogue systems. The long-standing goal of dialogue systems is to be human-like enough to establish long-term connections with users. Therefore, there has been an urgent need to evaluate LLMs as human-like dialogue systems. In this paper, we propose DialogBench, a dialogue evaluation benchmark that contains 12 dialogue tasks to probe the capabilities of LLMs as human-like dialogue systems should have. Specifically, we prompt GPT-4 to generate evaluation instances for each task. We first design the basic prompt based on widely used design principles and further mitigate the existing biases to generate higher-quality evaluation instances. Our extensive tests on English and Chinese DialogBench of 26 LLMs show that instruction tuning improves the human likeness of LLMs to a certain extent, but most LLMs still have much room for improvement as human-like dialogue systems. Interestingly, results also show that the positioning of assistant AI can make instruction tuning weaken the human emotional perception of LLMs and their mastery of information about human daily life.

DialogBench: Evaluating LLMs as Human-like Dialogue Systems

naacl-long.341

Poster

2311.01677

https://aclanthology.org/2024.naacl-long.342.bib

https://aclanthology.org/2024.naacl-long.342/

@inproceedings{adhya-sanyal-2024-ginopic, title = "{GIN}opic: Topic Modeling with Graph Isomorphism Network", author = "Adhya, Suman and Sanyal, Debarshi Kumar", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.342", doi = "10.18653/v1/2024.naacl-long.342", pages = "6171--6183", abstract = "Topic modeling is a widely used approach for analyzing and exploring large document collections. Recent research efforts have incorporated pre-trained contextualized language models, such as BERT embeddings, into topic modeling. However, they often neglect the intrinsic informational value conveyed by mutual dependencies between words. In this study, we introduce GINopic, a topic modeling framework based on graph isomorphism networks to capture the correlation between words. By conducting intrinsic (quantitative as well as qualitative) and extrinsic evaluations on diverse benchmark datasets, we demonstrate the effectiveness of GINopic compared to existing topic models and highlight its potential for advancing topic modeling.", }

Topic modeling is a widely used approach for analyzing and exploring large document collections. Recent research efforts have incorporated pre-trained contextualized language models, such as BERT embeddings, into topic modeling. However, they often neglect the intrinsic informational value conveyed by mutual dependencies between words. In this study, we introduce GINopic, a topic modeling framework based on graph isomorphism networks to capture the correlation between words. By conducting intrinsic (quantitative as well as qualitative) and extrinsic evaluations on diverse benchmark datasets, we demonstrate the effectiveness of GINopic compared to existing topic models and highlight its potential for advancing topic modeling.

GINopic: Topic Modeling with Graph Isomorphism Network

naacl-long.342

Poster

2404.02115

https://aclanthology.org/2024.naacl-long.343.bib

https://aclanthology.org/2024.naacl-long.343/

@inproceedings{wang-etal-2024-cmb, title = "{CMB}: A Comprehensive Medical Benchmark in {C}hinese", author = "Wang, Xidong and Chen, Guiming and Dingjie, Song and Zhiyi, Zhang and Chen, Zhihong and Xiao, Qingying and Chen, Junying and Jiang, Feng and Li, Jianquan and Wan, Xiang and Wang, Benyou and Li, Haizhou", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.343", doi = "10.18653/v1/2024.naacl-long.343", pages = "6184--6205", abstract = "Large Language Models (LLMs) provide a possibility to make a great breakthrough in medicine. The establishment of a standardized medical benchmark becomes a fundamental cornerstone to measure progression. However, medical environments in different regions have their local characteristics, e.g., the ubiquity and significance of traditional Chinese medicine within China. Therefore, merely translating English-based medical evaluation may result in \textit{contextual incongruities} to a local region. To solve the issue, we propose a localized medical benchmark called CMB, a Comprehensive Medical Benchmark in Chinese, designed and rooted entirely within the native Chinese linguistic and cultural framework. While traditional Chinese medicine is integral to this evaluation, it does not constitute its entirety. Using this benchmark, we have evaluated several prominent large-scale LLMs, including ChatGPT, GPT-4, dedicated Chinese LLMs, and LLMs specialized in the medical domain. We hope this benchmark provide first-hand experience in existing LLMs for medicine and also facilitate the widespread adoption and enhancement of medical LLMs within China. Our data and code are publicly available at https://github.com/FreedomIntelligence/CMB.", }

Large Language Models (LLMs) provide a possibility to make a great breakthrough in medicine. The establishment of a standardized medical benchmark becomes a fundamental cornerstone to measure progression. However, medical environments in different regions have their local characteristics, e.g., the ubiquity and significance of traditional Chinese medicine within China. Therefore, merely translating English-based medical evaluation may result in \textit{contextual incongruities} to a local region. To solve the issue, we propose a localized medical benchmark called CMB, a Comprehensive Medical Benchmark in Chinese, designed and rooted entirely within the native Chinese linguistic and cultural framework. While traditional Chinese medicine is integral to this evaluation, it does not constitute its entirety. Using this benchmark, we have evaluated several prominent large-scale LLMs, including ChatGPT, GPT-4, dedicated Chinese LLMs, and LLMs specialized in the medical domain. We hope this benchmark provide first-hand experience in existing LLMs for medicine and also facilitate the widespread adoption and enhancement of medical LLMs within China. Our data and code are publicly available at https://github.com/FreedomIntelligence/CMB.

CMB: A Comprehensive Medical Benchmark in Chinese

naacl-long.343

Poster

2308.08833

https://huggingface.co/papers/2308.08833

https://aclanthology.org/2024.naacl-long.344.bib

https://aclanthology.org/2024.naacl-long.344/

@inproceedings{wang-etal-2024-massive, title = "Massive End-to-end Speech Recognition Models with Time Reduction", author = "Wang, Weiran and Prabhavalkar, Rohit and Shan, Haozhe and Meng, Zhong and Hwang, Dongseong and Li, Qiujia and Sim, Khe Chai and Li, Bo and Qin, James and Cai, Xingyu and Stooke, Adam and Zheng, Chengjian and He, Yanzhang and Sainath, Tara and Moreno Mengibar, Pedro", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.344", doi = "10.18653/v1/2024.naacl-long.344", pages = "6206--6217", abstract = "We investigate massive end-to-end automatic speech recognition (ASR) models with efficiency improvements achieved by time reduction. The encoders of our models use the neural architecture of Google{'}s universal speech model (USM), with additional funnel pooling layers to significantly reduce the frame rate and speed up training and inference. We also explore a few practical methods to mitigate potential accuracy loss due to time reduction, while enjoying most efficiency gain. Our methods are demonstrated to work with both Connectionist Temporal Classification (CTC) and RNN-Transducer (RNN-T), with up to 2B model parameters, and over two domains. For a large-scale voice search recognition task, we perform extensive studies on vocabulary size, time reduction strategy, and its generalization performance on long-form test sets, and show that a 900M RNN-T is very tolerant to severe time reduction, with as low encoder output frame rate as 640ms. We also provide ablation studies on the Librispeech benchmark for important training hyperparameters and architecture designs, in training 600M RNN-T models at the frame rate of 160ms.", }

We investigate massive end-to-end automatic speech recognition (ASR) models with efficiency improvements achieved by time reduction. The encoders of our models use the neural architecture of Google{'}s universal speech model (USM), with additional funnel pooling layers to significantly reduce the frame rate and speed up training and inference. We also explore a few practical methods to mitigate potential accuracy loss due to time reduction, while enjoying most efficiency gain. Our methods are demonstrated to work with both Connectionist Temporal Classification (CTC) and RNN-Transducer (RNN-T), with up to 2B model parameters, and over two domains. For a large-scale voice search recognition task, we perform extensive studies on vocabulary size, time reduction strategy, and its generalization performance on long-form test sets, and show that a 900M RNN-T is very tolerant to severe time reduction, with as low encoder output frame rate as 640ms. We also provide ablation studies on the Librispeech benchmark for important training hyperparameters and architecture designs, in training 600M RNN-T models at the frame rate of 160ms.

Massive End-to-end Speech Recognition Models with Time Reduction

naacl-long.344

Poster

https://aclanthology.org/2024.naacl-long.345.bib

https://aclanthology.org/2024.naacl-long.345/

@inproceedings{ardakani-etal-2024-slimfit, title = "{S}lim{F}it: Memory-Efficient Fine-Tuning of Transformer-based Models Using Training Dynamics", author = "Ardakani, Arash and Haan, Altan and Tan, Shangyin and Popovici, Doru Thom and Cheung, Alvin and Iancu, Costin and Sen, Koushik", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.345", doi = "10.18653/v1/2024.naacl-long.345", pages = "6218--6236", abstract = "Transformer-based models, such as BERT and ViT, have achieved state-of-the-art results across different natural language processing (NLP) and computer vision (CV) tasks. However, these models are extremely memory intensive during their fine-tuning process, making them difficult to deploy on GPUs with limited memory resources. To address this issue, we introduce a new tool called SlimFit that reduces the memory requirements of these models by dynamically analyzing their training dynamics and freezing less-contributory layers during fine-tuning. The layers to freeze are chosen using a runtime inter-layer scheduling algorithm. This allows SlimFit to freeze up to 95{\%} of layers and reduce the overall on-device GPU memory usage of transformer-based models such as ViT and BERT by an average of 2.2x, across different NLP and CV benchmarks/datasets such as GLUE, SQuAD 2.0, CIFAR-10, CIFAR-100 and ImageNet with an average degradation of 0.2{\%} in accuracy. For such NLP and CV tasks, SlimFit can reduce up to 3.1x the total on-device memory usage with an accuracy degradation of only up to 0.4{\%}. As a result, while fine-tuning of ViT on ImageNet and BERT on SQuAD 2.0 with a batch size of 128 requires 3 and 2 32GB GPUs, respectively, SlimFit enables fine-tuning them on a single 32GB GPU without any significant accuracy degradation. The code of SlimFit is available at https://github.com/arashardakani/SlimFit.", }

Transformer-based models, such as BERT and ViT, have achieved state-of-the-art results across different natural language processing (NLP) and computer vision (CV) tasks. However, these models are extremely memory intensive during their fine-tuning process, making them difficult to deploy on GPUs with limited memory resources. To address this issue, we introduce a new tool called SlimFit that reduces the memory requirements of these models by dynamically analyzing their training dynamics and freezing less-contributory layers during fine-tuning. The layers to freeze are chosen using a runtime inter-layer scheduling algorithm. This allows SlimFit to freeze up to 95{\%} of layers and reduce the overall on-device GPU memory usage of transformer-based models such as ViT and BERT by an average of 2.2x, across different NLP and CV benchmarks/datasets such as GLUE, SQuAD 2.0, CIFAR-10, CIFAR-100 and ImageNet with an average degradation of 0.2{\%} in accuracy. For such NLP and CV tasks, SlimFit can reduce up to 3.1x the total on-device memory usage with an accuracy degradation of only up to 0.4{\%}. As a result, while fine-tuning of ViT on ImageNet and BERT on SQuAD 2.0 with a batch size of 128 requires 3 and 2 32GB GPUs, respectively, SlimFit enables fine-tuning them on a single 32GB GPU without any significant accuracy degradation. The code of SlimFit is available at https://github.com/arashardakani/SlimFit.

SlimFit: Memory-Efficient Fine-Tuning of Transformer-based Models Using Training Dynamics

naacl-long.345

Poster

2305.18513

https://aclanthology.org/2024.naacl-long.346.bib

https://aclanthology.org/2024.naacl-long.346/

@inproceedings{ye-etal-2024-effective, title = "Effective Large Language Model Adaptation for Improved Grounding and Citation Generation", author = "Ye, Xi and Sun, Ruoxi and Arik, Sercan and Pfister, Tomas", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.346", doi = "10.18653/v1/2024.naacl-long.346", pages = "6237--6251", abstract = "Large language models (LLMs) have achieved remarkable advancements in natural language understanding and generation. However, one major issue towards their widespread deployment in the real world is that they can generate {``}hallucinated{''} answers that are not factual.Towards this end, this paper focuses on improving LLMs by grounding their responses in retrieved passages and by providing citations. We propose a new framework, AGREE, Adaptation for GRounding EnhancEment, that improves the grounding from a holistic perspective. Our framework tunes LLMs to self-ground the claims in their responses and provide accurate citations to retrieved documents. This tuning on top of the pre-trained LLMs requires well-grounded responses (with citations) for paired queries, for which we introduce a method that can automatically construct such data from unlabeled queries. The self-grounding capability of tuned LLMs further grants them a test-time adaptation (TTA) capability that can actively retrieve passages to support the claims that have not been grounded, which iteratively improves the responses of LLMs. Across five datasets and two LLMs, our results show that the proposed tuning-based framework generates superior grounded responses with more accurate citations compared to prompting-based approaches and post-hoc citing-based approaches.", }

Large language models (LLMs) have achieved remarkable advancements in natural language understanding and generation. However, one major issue towards their widespread deployment in the real world is that they can generate {``}hallucinated{''} answers that are not factual.Towards this end, this paper focuses on improving LLMs by grounding their responses in retrieved passages and by providing citations. We propose a new framework, AGREE, Adaptation for GRounding EnhancEment, that improves the grounding from a holistic perspective. Our framework tunes LLMs to self-ground the claims in their responses and provide accurate citations to retrieved documents. This tuning on top of the pre-trained LLMs requires well-grounded responses (with citations) for paired queries, for which we introduce a method that can automatically construct such data from unlabeled queries. The self-grounding capability of tuned LLMs further grants them a test-time adaptation (TTA) capability that can actively retrieve passages to support the claims that have not been grounded, which iteratively improves the responses of LLMs. Across five datasets and two LLMs, our results show that the proposed tuning-based framework generates superior grounded responses with more accurate citations compared to prompting-based approaches and post-hoc citing-based approaches.

Effective Large Language Model Adaptation for Improved Grounding and Citation Generation

naacl-long.346

Poster

2311.09533

https://aclanthology.org/2024.naacl-long.347.bib

https://aclanthology.org/2024.naacl-long.347/

@inproceedings{shao-etal-2024-assisting, title = "Assisting in Writing {W}ikipedia-like Articles From Scratch with Large Language Models", author = "Shao, Yijia and Jiang, Yucheng and Kanell, Theodore and Xu, Peter and Khattab, Omar and Lam, Monica", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.347", doi = "10.18653/v1/2024.naacl-long.347", pages = "6252--6278", abstract = "We study how to apply large language models to write grounded and organized long-form articles from scratch, with comparable breadth and depth to Wikipedia pages. This underexplored problem poses new challenges at the pre-writing stage, including how to research the topic and prepare an outline prior to writing. We propose STORM, a writing system for the Synthesis of Topic Outlines throughRetrieval and Multi-perspective Question Asking. STORM models the pre-writing stage by (1) discovering diverse perspectives in researching the given topic, (2) simulating conversations where writers carrying different perspectives pose questions to a topic expert grounded on trusted Internet sources, (3) curating the collected information to create an outline.For evaluation, we curate FreshWiki, a dataset of recent high-quality Wikipedia articles, and formulate outline assessments to evaluate the pre-writing stage. We further gather feedback from experienced Wikipedia editors. Compared to articles generated by an outline-driven retrieval-augmented baseline, more of STORM{'}s articles are deemed to be organized (by a 25{\%} absolute increase) and broad in coverage (by 10{\%}). The expert feedback also helps identify new challenges for generating grounded long articles, such as source bias transfer and over-association of unrelated facts.", }

We study how to apply large language models to write grounded and organized long-form articles from scratch, with comparable breadth and depth to Wikipedia pages. This underexplored problem poses new challenges at the pre-writing stage, including how to research the topic and prepare an outline prior to writing. We propose STORM, a writing system for the Synthesis of Topic Outlines throughRetrieval and Multi-perspective Question Asking. STORM models the pre-writing stage by (1) discovering diverse perspectives in researching the given topic, (2) simulating conversations where writers carrying different perspectives pose questions to a topic expert grounded on trusted Internet sources, (3) curating the collected information to create an outline.For evaluation, we curate FreshWiki, a dataset of recent high-quality Wikipedia articles, and formulate outline assessments to evaluate the pre-writing stage. We further gather feedback from experienced Wikipedia editors. Compared to articles generated by an outline-driven retrieval-augmented baseline, more of STORM{'}s articles are deemed to be organized (by a 25{\%} absolute increase) and broad in coverage (by 10{\%}). The expert feedback also helps identify new challenges for generating grounded long articles, such as source bias transfer and over-association of unrelated facts.

Assisting in Writing Wikipedia-like Articles From Scratch with Large Language Models

naacl-long.347

Poster

2402.14207

https://huggingface.co/papers/2402.14207

https://aclanthology.org/2024.naacl-long.348.bib

https://aclanthology.org/2024.naacl-long.348/

@inproceedings{shaikh-etal-2024-grounding, title = "Grounding Gaps in Language Model Generations", author = "Shaikh, Omar and Gligoric, Kristina and Khetan, Ashna and Gerstgrasser, Matthias and Yang, Diyi and Jurafsky, Dan", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.348", doi = "10.18653/v1/2024.naacl-long.348", pages = "6279--6296", abstract = "Effective conversation requires common ground: a shared understanding between the participants. Common ground, however, does not emerge spontaneously in conversation. Speakers and listeners work together to both identify and construct a shared basis while avoiding misunderstanding. To accomplish grounding, humans rely on a range of dialogue acts, like clarification (What do you mean?) and acknowledgment (I understand.). However, it is unclear whether large language models (LLMs) generate text that reflects human grounding. To this end, we curate a set of grounding acts and propose corresponding metrics that quantify attempted grounding. We study whether LLM generations contain grounding acts, simulating turn-taking from several dialogue datasets and comparing results to humans. We find that{---}compared to humans{---}LLMs generate language with less conversational grounding, instead generating text that appears to simply presume common ground. To understand the roots of the identified grounding gap, we examine the role of instruction tuning and preference optimization, finding that training on contemporary preference data leads to a reduction in generated grounding acts. Altogether, we highlight the need for more research investigating conversational grounding in human-AI interaction.", }

Effective conversation requires common ground: a shared understanding between the participants. Common ground, however, does not emerge spontaneously in conversation. Speakers and listeners work together to both identify and construct a shared basis while avoiding misunderstanding. To accomplish grounding, humans rely on a range of dialogue acts, like clarification (What do you mean?) and acknowledgment (I understand.). However, it is unclear whether large language models (LLMs) generate text that reflects human grounding. To this end, we curate a set of grounding acts and propose corresponding metrics that quantify attempted grounding. We study whether LLM generations contain grounding acts, simulating turn-taking from several dialogue datasets and comparing results to humans. We find that{---}compared to humans{---}LLMs generate language with less conversational grounding, instead generating text that appears to simply presume common ground. To understand the roots of the identified grounding gap, we examine the role of instruction tuning and preference optimization, finding that training on contemporary preference data leads to a reduction in generated grounding acts. Altogether, we highlight the need for more research investigating conversational grounding in human-AI interaction.

Grounding Gaps in Language Model Generations

naacl-long.348

Oral

2311.09144

https://aclanthology.org/2024.naacl-long.349.bib

https://aclanthology.org/2024.naacl-long.349/

@inproceedings{baziotis-etal-2024-monolingual, title = "When Does Monolingual Data Help Multilingual Translation: The Role of Domain and Model Scale", author = "Baziotis, Christos and Zhang, Biao and Birch, Alexandra and Haddow, Barry", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.349", doi = "10.18653/v1/2024.naacl-long.349", pages = "6297--6324", abstract = "Multilingual machine translation (MMT), trained on a mixture of parallel and monolingual data, is key for improving translation in low-resource language pairs. However, the literature offers conflicting results on the performance of different methods of including monolingual data. To resolve this, we examine how denoising autoencoding (DAE) and backtranslation (BT) impact MMT under different data conditions and model scales. Unlike prior studies, we use a realistic dataset of 100 translation directions and consider many domain combinations of monolingual and test data. We find that monolingual data generally helps MMT, but models are surprisingly brittle to domain mismatches, especially at smaller model scales. BT is beneficial when the parallel, monolingual, and test data sources are similar but can be detrimental otherwise, while DAE is less effective than previously reported. Next, we analyze the impact of scale (from 90M to 1.6B parameters) and find it is important for both methods, particularly DAE. As scale increases, DAE transitions from underperforming the parallel-only baseline at 90M to converging with BT performance at 1.6B, and even surpassing it in low-resource. These results offer new insights into how to best use monolingual data in MMT.", }

Multilingual machine translation (MMT), trained on a mixture of parallel and monolingual data, is key for improving translation in low-resource language pairs. However, the literature offers conflicting results on the performance of different methods of including monolingual data. To resolve this, we examine how denoising autoencoding (DAE) and backtranslation (BT) impact MMT under different data conditions and model scales. Unlike prior studies, we use a realistic dataset of 100 translation directions and consider many domain combinations of monolingual and test data. We find that monolingual data generally helps MMT, but models are surprisingly brittle to domain mismatches, especially at smaller model scales. BT is beneficial when the parallel, monolingual, and test data sources are similar but can be detrimental otherwise, while DAE is less effective than previously reported. Next, we analyze the impact of scale (from 90M to 1.6B parameters) and find it is important for both methods, particularly DAE. As scale increases, DAE transitions from underperforming the parallel-only baseline at 90M to converging with BT performance at 1.6B, and even surpassing it in low-resource. These results offer new insights into how to best use monolingual data in MMT.

When Does Monolingual Data Help Multilingual Translation: The Role of Domain and Model Scale

naacl-long.349

Poster

2305.14124

https://huggingface.co/papers/2305.14124

https://aclanthology.org/2024.naacl-long.350.bib

https://aclanthology.org/2024.naacl-long.350/

@inproceedings{rahamim-belinkov-2024-contrasim, title = "{C}ontra{S}im {--} Analyzing Neural Representations Based on Contrastive Learning", author = "Rahamim, Adir and Belinkov, Yonatan", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.350", doi = "10.18653/v1/2024.naacl-long.350", pages = "6325--6339", abstract = "Recent work has compared neural network representations via similarity-based analyses to improve model interpretation. The quality of a similarity measure is typically evaluated by its success in assigning a high score to representations that are expected to be matched. However, existing similarity measures perform mediocrely on standard benchmarks. In this work, we develop a new similarity measure, dubbed ContraSim, based on contrastive learning. In contrast to common closed-form similarity measures, ContraSim learns a parameterized measure by using both similar and dissimilar examples. We perform an extensive experimental evaluation of our method, with both language and vision models, on the standard layer prediction benchmark and two new benchmarks that we introduce: the multilingual benchmark and the image{--}caption benchmark. In all cases, ContraSim achieves much higher accuracy than previous similarity measures, even when presented with challenging examples. Finally, ContraSim is more suitable for the analysis of neural networks, revealing new insights not captured by previous measures.", }

Recent work has compared neural network representations via similarity-based analyses to improve model interpretation. The quality of a similarity measure is typically evaluated by its success in assigning a high score to representations that are expected to be matched. However, existing similarity measures perform mediocrely on standard benchmarks. In this work, we develop a new similarity measure, dubbed ContraSim, based on contrastive learning. In contrast to common closed-form similarity measures, ContraSim learns a parameterized measure by using both similar and dissimilar examples. We perform an extensive experimental evaluation of our method, with both language and vision models, on the standard layer prediction benchmark and two new benchmarks that we introduce: the multilingual benchmark and the image{--}caption benchmark. In all cases, ContraSim achieves much higher accuracy than previous similarity measures, even when presented with challenging examples. Finally, ContraSim is more suitable for the analysis of neural networks, revealing new insights not captured by previous measures.

ContraSim – Analyzing Neural Representations Based on Contrastive Learning

naacl-long.350

Poster

https://aclanthology.org/2024.naacl-long.351.bib

https://aclanthology.org/2024.naacl-long.351/

@inproceedings{wu-etal-2024-universal, title = "Universal Prompt Optimizer for Safe Text-to-Image Generation", author = "Wu, Zongyu and Gao, Hongcheng and Wang, Yueze and Zhang, Xiang and Wang, Suhang", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.351", doi = "10.18653/v1/2024.naacl-long.351", pages = "6340--6354", abstract = "Text-to-Image (T2I) models have shown great performance in generating images based on textual prompts. However, these models are vulnerable to unsafe input to generate unsafe content like sexual, harassment and illegal-activity images. Existing studies based on image checker, model fine-tuning and embedding blocking are impractical in real-world applications. Hence, we propose the first universal **p**rompt **o**ptimizer for **s**afe T2**I** (**POSI**) generation in black-box scenario. We first construct a dataset consisting of toxic-clean prompt pairs by GPT-3.5 Turbo. To guide the optimizer to have the ability of converting toxic prompt to clean prompt while preserving semantic information, we design a novel reward function measuring toxicity and text alignment of generated images and train the optimizer through Proximal Policy Optimization. Experiments show that our approach can effectively reduce the likelihood of various T2I models in generating inappropriate images, with no significant impact on text alignment. It is also flexible to be combined with methods to achieve better performance. Our code is available at [https://github.com/wzongyu/POSI](https://github.com/wzongyu/POSI).", }

Text-to-Image (T2I) models have shown great performance in generating images based on textual prompts. However, these models are vulnerable to unsafe input to generate unsafe content like sexual, harassment and illegal-activity images. Existing studies based on image checker, model fine-tuning and embedding blocking are impractical in real-world applications. Hence, we propose the first universal **p**rompt **o**ptimizer for **s**afe T2**I** (**POSI**) generation in black-box scenario. We first construct a dataset consisting of toxic-clean prompt pairs by GPT-3.5 Turbo. To guide the optimizer to have the ability of converting toxic prompt to clean prompt while preserving semantic information, we design a novel reward function measuring toxicity and text alignment of generated images and train the optimizer through Proximal Policy Optimization. Experiments show that our approach can effectively reduce the likelihood of various T2I models in generating inappropriate images, with no significant impact on text alignment. It is also flexible to be combined with methods to achieve better performance. Our code is available at [https://github.com/wzongyu/POSI](https://github.com/wzongyu/POSI).

Universal Prompt Optimizer for Safe Text-to-Image Generation

naacl-long.351

Poster

2402.10882

https://aclanthology.org/2024.naacl-long.352.bib

https://aclanthology.org/2024.naacl-long.352/

@inproceedings{chen-etal-2024-language, title = "Language Model Based Unsupervised Dependency Parsing with Conditional Mutual Information and Grammatical Constraints", author = "Chen, Junjie and He, Xiangheng and Miyao, Yusuke", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.352", doi = "10.18653/v1/2024.naacl-long.352", pages = "6355--6366", abstract = "Previous methods based on Large Language Models (LLM) perform unsupervised dependency parsing by maximizing bi-lexical dependence scores. However, these previous methods adopt dependence scores that are difficult to interpret. These methods cannot incorporate grammatical constraints that previous grammar-based parsing research has shown beneficial to improving parsing performance. In this work, we apply Conditional Mutual Information (CMI), an interpretable metric, to measure the bi-lexical dependence and incorporate grammatical constraints into LLM-based unsupervised parsing. We incorporate Part-Of-Speech information as a grammatical constraint at the CMI estimation stage and integrate two additional grammatical constraints at the subsequent tree decoding stage. We find that the CMI score positively correlates with syntactic dependencies and has a stronger correlation with the syntactic dependencies than baseline scores. Our experiment confirms the benefits and applicability of the proposed grammatical constraints across five languages and eight datasets. The CMI parsing model outperforms state-of-the-art LLM-based models and similarly constrained grammar-based models. Our analysis reveals that the CMI model is strong in retrieving dependency relations with rich lexical interactions but is weak in retrieving relations with sparse lexical interactions, indicating a potential limitation in CMI-based unsupervised parsing methods.", }

Previous methods based on Large Language Models (LLM) perform unsupervised dependency parsing by maximizing bi-lexical dependence scores. However, these previous methods adopt dependence scores that are difficult to interpret. These methods cannot incorporate grammatical constraints that previous grammar-based parsing research has shown beneficial to improving parsing performance. In this work, we apply Conditional Mutual Information (CMI), an interpretable metric, to measure the bi-lexical dependence and incorporate grammatical constraints into LLM-based unsupervised parsing. We incorporate Part-Of-Speech information as a grammatical constraint at the CMI estimation stage and integrate two additional grammatical constraints at the subsequent tree decoding stage. We find that the CMI score positively correlates with syntactic dependencies and has a stronger correlation with the syntactic dependencies than baseline scores. Our experiment confirms the benefits and applicability of the proposed grammatical constraints across five languages and eight datasets. The CMI parsing model outperforms state-of-the-art LLM-based models and similarly constrained grammar-based models. Our analysis reveals that the CMI model is strong in retrieving dependency relations with rich lexical interactions but is weak in retrieving relations with sparse lexical interactions, indicating a potential limitation in CMI-based unsupervised parsing methods.

Language Model Based Unsupervised Dependency Parsing with Conditional Mutual Information and Grammatical Constraints

naacl-long.352

Poster

https://aclanthology.org/2024.naacl-long.353.bib

https://aclanthology.org/2024.naacl-long.353/

@inproceedings{seshadri-etal-2024-bias, title = "The Bias Amplification Paradox in Text-to-Image Generation", author = "Seshadri, Preethi and Singh, Sameer and Elazar, Yanai", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.353", doi = "10.18653/v1/2024.naacl-long.353", pages = "6367--6384", abstract = "Bias amplification is a phenomenon in which models exacerbate biases or stereotypes present in the training data. In this paper, we study bias amplification in the text-to-image domain using Stable Diffusion by comparing gender ratios in training vs. generated images. We find that the model appears to amplify gender-occupation biases found in the training data (LAION) considerably. However, we discover that amplification can be largely attributed to discrepancies between training captions and model prompts. For example, an inherent difference is that captions from the training data often contain explicit gender information while our prompts do not, which leads to a distribution shift and consequently inflates bias measures. Once we account for distributional differences between texts used for training and generation when evaluating amplification, we observe that amplification decreases drastically. Our findings illustrate the challenges of comparing biases in models and their training data, as well as evaluation more broadly, and highlight how confounding factors can impact analyses.", }

Bias amplification is a phenomenon in which models exacerbate biases or stereotypes present in the training data. In this paper, we study bias amplification in the text-to-image domain using Stable Diffusion by comparing gender ratios in training vs. generated images. We find that the model appears to amplify gender-occupation biases found in the training data (LAION) considerably. However, we discover that amplification can be largely attributed to discrepancies between training captions and model prompts. For example, an inherent difference is that captions from the training data often contain explicit gender information while our prompts do not, which leads to a distribution shift and consequently inflates bias measures. Once we account for distributional differences between texts used for training and generation when evaluating amplification, we observe that amplification decreases drastically. Our findings illustrate the challenges of comparing biases in models and their training data, as well as evaluation more broadly, and highlight how confounding factors can impact analyses.

The Bias Amplification Paradox in Text-to-Image Generation

naacl-long.353

Poster

2308.00755

https://aclanthology.org/2024.naacl-long.354.bib

https://aclanthology.org/2024.naacl-long.354/

@inproceedings{lucas-etal-2024-grammar, title = "Grammar-based Data Augmentation for Low-Resource Languages: The Case of {G}uarani-{S}panish Neural Machine Translation", author = {Lucas, Agust{\'\i}n and Balad{\'o}n, Alexis and Pardi{\~n}as, Victoria and Ag{\"u}ero-Torales, Marvin and G{\'o}ngora, Santiago and Chiruzzo, Luis}, editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.354", doi = "10.18653/v1/2024.naacl-long.354", pages = "6385--6397", abstract = "One of the main problems low-resource languages face in NLP can be pictured as a vicious circle: data is needed to build and test tools, but the available text is scarce and there are not powerful tools to collect it.In order to break this circle for Guarani, we explore if text automatically generated from a grammar can work as a Data Augmentation technique to boost the performance of Guarani-Spanish Machine Translation (MT) systems.After building a grammar-based system that generates Spanish text and syntactically transfers it to Guarani, we perform several experiments by pretraining models using this synthetic text.We find that the MT systems that are pretrained with synthetic text perform better, even outperforming previous baselines.", }

One of the main problems low-resource languages face in NLP can be pictured as a vicious circle: data is needed to build and test tools, but the available text is scarce and there are not powerful tools to collect it.In order to break this circle for Guarani, we explore if text automatically generated from a grammar can work as a Data Augmentation technique to boost the performance of Guarani-Spanish Machine Translation (MT) systems.After building a grammar-based system that generates Spanish text and syntactically transfers it to Guarani, we perform several experiments by pretraining models using this synthetic text.We find that the MT systems that are pretrained with synthetic text perform better, even outperforming previous baselines.

Grammar-based Data Augmentation for Low-Resource Languages: The Case of Guarani-Spanish Neural Machine Translation

naacl-long.354

Poster

https://aclanthology.org/2024.naacl-long.355.bib

https://aclanthology.org/2024.naacl-long.355/

@inproceedings{mukherjee-etal-2024-global, title = "Global Gallery: The Fine Art of Painting Culture Portraits through Multilingual Instruction Tuning", author = "Mukherjee, Anjishnu and Caliskan, Aylin and Zhu, Ziwei and Anastasopoulos, Antonios", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.355", doi = "10.18653/v1/2024.naacl-long.355", pages = "6398--6415", abstract = "Exploring the intersection of language and culture in Large Language Models (LLMs), this study critically examines their capability to encapsulate cultural nuances across diverse linguistic landscapes. Central to our investigation are three research questions: the efficacy of language-specific instruction tuning, the impact of pretraining on dominant language data, and the identification of optimal approaches to elicit accurate cultural knowledge from LLMs. Utilizing the GeoMLaMA benchmark for multilingual commonsense knowledge and an adapted CAMeL dataset (English-only) for evaluation of nuanced cultural aspects, our experiments span six different languages and cultural contexts, revealing the extent of LLMs{'} cultural awareness. Our findings highlight a nuanced landscape: while language-specific tuning and bilingual pretraining enhance cultural understanding in certain contexts, they also uncover inconsistencies and biases, particularly in non-Western cultures. This work expands our understanding of LLMs{'} cultural competence and emphasizes the importance of integrating diverse cultural perspectives in their development, aiming for a more globally representative and equitable approach in language modeling.", }

Exploring the intersection of language and culture in Large Language Models (LLMs), this study critically examines their capability to encapsulate cultural nuances across diverse linguistic landscapes. Central to our investigation are three research questions: the efficacy of language-specific instruction tuning, the impact of pretraining on dominant language data, and the identification of optimal approaches to elicit accurate cultural knowledge from LLMs. Utilizing the GeoMLaMA benchmark for multilingual commonsense knowledge and an adapted CAMeL dataset (English-only) for evaluation of nuanced cultural aspects, our experiments span six different languages and cultural contexts, revealing the extent of LLMs{'} cultural awareness. Our findings highlight a nuanced landscape: while language-specific tuning and bilingual pretraining enhance cultural understanding in certain contexts, they also uncover inconsistencies and biases, particularly in non-Western cultures. This work expands our understanding of LLMs{'} cultural competence and emphasizes the importance of integrating diverse cultural perspectives in their development, aiming for a more globally representative and equitable approach in language modeling.

Global Gallery: The Fine Art of Painting Culture Portraits through Multilingual Instruction Tuning

naacl-long.355

Poster

https://aclanthology.org/2024.naacl-long.356.bib

https://aclanthology.org/2024.naacl-long.356/

@inproceedings{lee-etal-2024-toward, title = "Toward Interactive Regional Understanding in Vision-Large Language Models", author = "Lee, Jungbeom and Chun, Sanghyuk and Yun, Sangdoo", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.356", doi = "10.18653/v1/2024.naacl-long.356", pages = "6416--6429", abstract = "Recent Vision-Language Pre-training (VLP) models have demonstrated significant advancements. Nevertheless, these models heavily rely on image-text pairs that capture only coarse and global information of an image, leading to a limitation in their regional understanding ability. In this work, we introduce \textbf{RegionVLM}, equipped with explicit regional modeling capabilities, allowing them to understand user-indicated image regions. To achieve this, we design a simple yet innovative architecture, requiring no modifications to the model architecture or objective function. Additionally, we leverage a dataset that contains a novel source of information, namely Localized Narratives, which has been overlooked in previous VLP research. Our experiments demonstrate that our single generalist model not only achieves an interactive dialogue system but also exhibits superior performance on various zero-shot region understanding tasks, without compromising its ability for global image understanding.", }

Recent Vision-Language Pre-training (VLP) models have demonstrated significant advancements. Nevertheless, these models heavily rely on image-text pairs that capture only coarse and global information of an image, leading to a limitation in their regional understanding ability. In this work, we introduce \textbf{RegionVLM}, equipped with explicit regional modeling capabilities, allowing them to understand user-indicated image regions. To achieve this, we design a simple yet innovative architecture, requiring no modifications to the model architecture or objective function. Additionally, we leverage a dataset that contains a novel source of information, namely Localized Narratives, which has been overlooked in previous VLP research. Our experiments demonstrate that our single generalist model not only achieves an interactive dialogue system but also exhibits superior performance on various zero-shot region understanding tasks, without compromising its ability for global image understanding.

Toward Interactive Regional Understanding in Vision-Large Language Models

naacl-long.356

Poster

2403.18260

https://aclanthology.org/2024.naacl-long.357.bib

https://aclanthology.org/2024.naacl-long.357/

@inproceedings{lee-etal-2024-scriptmix, title = "{S}cript{M}ix: Mixing Scripts for Low-resource Language Parsing", author = "Lee, Jaeseong and Lee, Dohyeon and Hwang, Seung-won", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.357", doi = "10.18653/v1/2024.naacl-long.357", pages = "6430--6444", abstract = "Despite the success of multilingual pretrained language models (mPLMs) for tasks such as dependency parsing (DEP) or part-of-speech (POS) tagging, their coverage of 100s of languages is still limited, as most of the 6500+ languages remains {``}unseen{''}. To adapt mPLMs for including such unseen langs, existing work has considered transliteration and vocabulary augmentation. Meanwhile, the consideration of combining the two has been surprisingly lacking. To understand why, we identify both complementary strengths of the two, and the hurdles to realizing it. Based on this observation, we propose ScriptMix, combining two strengths, and overcoming the hurdle.Specifically, ScriptMix a) is trained with dual-script corpus to combine strengths, but b) with separate modules to avoid gradient conflict. In combining modules properly, we also point out the limitation of the conventional method AdapterFusion, and propose AdapterFusion+ to overcome it. We empirically show ScriptMix is effective{--} ScriptMix improves the POS accuracy by up to 14{\%}, and improves the DEP LAS score by up to 5.6{\%}. Our code is publicly available.", }

Despite the success of multilingual pretrained language models (mPLMs) for tasks such as dependency parsing (DEP) or part-of-speech (POS) tagging, their coverage of 100s of languages is still limited, as most of the 6500+ languages remains {``}unseen{''}. To adapt mPLMs for including such unseen langs, existing work has considered transliteration and vocabulary augmentation. Meanwhile, the consideration of combining the two has been surprisingly lacking. To understand why, we identify both complementary strengths of the two, and the hurdles to realizing it. Based on this observation, we propose ScriptMix, combining two strengths, and overcoming the hurdle.Specifically, ScriptMix a) is trained with dual-script corpus to combine strengths, but b) with separate modules to avoid gradient conflict. In combining modules properly, we also point out the limitation of the conventional method AdapterFusion, and propose AdapterFusion+ to overcome it. We empirically show ScriptMix is effective{--} ScriptMix improves the POS accuracy by up to 14{\%}, and improves the DEP LAS score by up to 5.6{\%}. Our code is publicly available.

ScriptMix: Mixing Scripts for Low-resource Language Parsing

naacl-long.357

Poster

https://aclanthology.org/2024.naacl-long.358.bib

https://aclanthology.org/2024.naacl-long.358/

@inproceedings{li-etal-2024-mt, title = "{MT}-{PATCHER}: Selective and Extendable Knowledge Distillation from Large Language Models for Machine Translation", author = "Li, Jiahuan and Cheng, Shanbo and Huang, Shujian and Chen, Jiajun", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.358", doi = "10.18653/v1/2024.naacl-long.358", pages = "6445--6459", abstract = "Large Language Models (LLM) have demonstrated their strong ability in the field of machine translation, yet they suffer from high computational cost and latency. Therefore, transferring translation knowledge from giant LLMs to medium-sized machine translation models is a promising research direction. However, traditional knowledge distillation methods ignore the capability of student and teacher models, therefore repeatedly teaching student models on the knowledge they have learned, and failing to extend to novel contexts and knowledge. In this paper, we propose a framework called MT-Patcher, which transfers knowledge from LLMs to existing MT models in a selective, comprehensive and proactive manner. Considering the current translation ability of student MT models, we only identify and correct their translation errors, instead of distilling the whole translation from the teacher. Leveraging the strong language abilities of LLMs, we instruct LLM teachers to synthesize diverse contexts and anticipate more potential errors for the student. Experiment results on translating both specific language phenomena and general MT benchmarks demonstrate that finetuning the MT model on about 10{\%} examples can achieve comparable results to the traditional knowledge distillation method, and synthesized potential errors and diverse contexts further improve MT performances on unseen contexts and words.", }

Large Language Models (LLM) have demonstrated their strong ability in the field of machine translation, yet they suffer from high computational cost and latency. Therefore, transferring translation knowledge from giant LLMs to medium-sized machine translation models is a promising research direction. However, traditional knowledge distillation methods ignore the capability of student and teacher models, therefore repeatedly teaching student models on the knowledge they have learned, and failing to extend to novel contexts and knowledge. In this paper, we propose a framework called MT-Patcher, which transfers knowledge from LLMs to existing MT models in a selective, comprehensive and proactive manner. Considering the current translation ability of student MT models, we only identify and correct their translation errors, instead of distilling the whole translation from the teacher. Leveraging the strong language abilities of LLMs, we instruct LLM teachers to synthesize diverse contexts and anticipate more potential errors for the student. Experiment results on translating both specific language phenomena and general MT benchmarks demonstrate that finetuning the MT model on about 10{\%} examples can achieve comparable results to the traditional knowledge distillation method, and synthesized potential errors and diverse contexts further improve MT performances on unseen contexts and words.

MT-PATCHER: Selective and Extendable Knowledge Distillation from Large Language Models for Machine Translation

naacl-long.358

Poster

2403.09522

https://aclanthology.org/2024.naacl-long.359.bib

https://aclanthology.org/2024.naacl-long.359/

@inproceedings{hoang-etal-2024-toxcl, title = "{T}o{XCL}: A Unified Framework for Toxic Speech Detection and Explanation", author = "Hoang, Nhat and Do, Xuan Long and Do, Duc Anh and Vu, Duc Anh and Luu, Anh Tuan", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.359", doi = "10.18653/v1/2024.naacl-long.359", pages = "6460--6472", abstract = "The proliferation of online toxic speech is a pertinent problem posing threats to demographic groups. While explicit toxic speech contains offensive lexical signals, implicit one consists of coded or indirect language. Therefore, it is crucial for models not only to detect implicit toxic speech but also to explain its toxicity. This draws a unique need for unified frameworks that can effectively detect and explain implicit toxic speech. Prior works mainly formulated the task of toxic speech detection and explanation as a text generation problem. Nonetheless, models trained using this strategy can be prone to suffer from the consequent error propagation problem. Moreover, our experiments reveal that the detection results of such models are much lower than those that focus only on the detection task. To bridge these gaps, we introduce ToXCL, a unified framework for the detection and explanation of implicit toxic speech. Our model consists of three modules: a (i) Target Group Generator to generate the targeted demographic group(s) of a given post; an (ii) Encoder-Decoder Model in which the encoder focuses on detecting implicit toxic speech and is boosted by a (iii) Teacher Classifier via knowledge distillation, and the decoder generates the necessary explanation. ToXCL achieves new state-of-the-art effectiveness, and outperforms baselines significantly.", }

The proliferation of online toxic speech is a pertinent problem posing threats to demographic groups. While explicit toxic speech contains offensive lexical signals, implicit one consists of coded or indirect language. Therefore, it is crucial for models not only to detect implicit toxic speech but also to explain its toxicity. This draws a unique need for unified frameworks that can effectively detect and explain implicit toxic speech. Prior works mainly formulated the task of toxic speech detection and explanation as a text generation problem. Nonetheless, models trained using this strategy can be prone to suffer from the consequent error propagation problem. Moreover, our experiments reveal that the detection results of such models are much lower than those that focus only on the detection task. To bridge these gaps, we introduce ToXCL, a unified framework for the detection and explanation of implicit toxic speech. Our model consists of three modules: a (i) Target Group Generator to generate the targeted demographic group(s) of a given post; an (ii) Encoder-Decoder Model in which the encoder focuses on detecting implicit toxic speech and is boosted by a (iii) Teacher Classifier via knowledge distillation, and the decoder generates the necessary explanation. ToXCL achieves new state-of-the-art effectiveness, and outperforms baselines significantly.

ToXCL: A Unified Framework for Toxic Speech Detection and Explanation

naacl-long.359

Poster

2403.16685

https://huggingface.co/papers/2403.16685

https://aclanthology.org/2024.naacl-long.360.bib

https://aclanthology.org/2024.naacl-long.360/

@inproceedings{xu-wang-2024-linkprompt, title = "{L}ink{P}rompt: Natural and Universal Adversarial Attacks on Prompt-based Language Models", author = "Xu, Yue and Wang, Wenjie", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.360", doi = "10.18653/v1/2024.naacl-long.360", pages = "6473--6486", abstract = "Prompt-based learning is a new language model training paradigm that adapts the Pre-trained Language Models (PLMs) to downstream tasks, which revitalizes the performance benchmarks across various natural language processing (NLP) tasks. Instead of using a fixed prompt template to fine-tune the model, some research demonstrates the effectiveness of searching for the prompt via optimization. Such prompt optimization process of prompt-based learning on PLMs also gives insight into generating adversarial prompts to mislead the model, raising concerns about the adversarial vulnerability of this paradigm. Recent studies have shown that universal adversarial triggers (UATs) can be generated to alter not only the predictions of the target PLMs but also the prediction of corresponding Prompt-based Fine-tuning Models (PFMs) under the prompt-based learning paradigm. However, UATs found in previous works are often unreadable tokens or characters and can be easily distinguished from natural texts with adaptive defenses. In this work, we consider the naturalness of the UATs and develop $\textit{LinkPrompt}$, an adversarial attack algorithm to generate UATs by a gradient-based beam search algorithm that not only effectively attacks the target PLMs and PFMs but also maintains the naturalness among the trigger tokens. Extensive results demonstrate the effectiveness of $\textit{LinkPrompt}$, as well as the transferability of UATs generated by $\textit{LinkPrompt}$ to open-sourced Large Language Model (LLM) Llama2 and API-accessed LLM GPT-3.5-turbo. The resource is available at https://github.com/SavannahXu79/LinkPrompt.", }

Prompt-based learning is a new language model training paradigm that adapts the Pre-trained Language Models (PLMs) to downstream tasks, which revitalizes the performance benchmarks across various natural language processing (NLP) tasks. Instead of using a fixed prompt template to fine-tune the model, some research demonstrates the effectiveness of searching for the prompt via optimization. Such prompt optimization process of prompt-based learning on PLMs also gives insight into generating adversarial prompts to mislead the model, raising concerns about the adversarial vulnerability of this paradigm. Recent studies have shown that universal adversarial triggers (UATs) can be generated to alter not only the predictions of the target PLMs but also the prediction of corresponding Prompt-based Fine-tuning Models (PFMs) under the prompt-based learning paradigm. However, UATs found in previous works are often unreadable tokens or characters and can be easily distinguished from natural texts with adaptive defenses. In this work, we consider the naturalness of the UATs and develop $\textit{LinkPrompt}$, an adversarial attack algorithm to generate UATs by a gradient-based beam search algorithm that not only effectively attacks the target PLMs and PFMs but also maintains the naturalness among the trigger tokens. Extensive results demonstrate the effectiveness of $\textit{LinkPrompt}$, as well as the transferability of UATs generated by $\textit{LinkPrompt}$ to open-sourced Large Language Model (LLM) Llama2 and API-accessed LLM GPT-3.5-turbo. The resource is available at https://github.com/SavannahXu79/LinkPrompt.

LinkPrompt: Natural and Universal Adversarial Attacks on Prompt-based Language Models

naacl-long.360

Poster

https://aclanthology.org/2024.naacl-long.361.bib

https://aclanthology.org/2024.naacl-long.361/

@inproceedings{zhang-etal-2024-coe, title = "{C}o{E}-{SQL}: In-Context Learning for Multi-Turn Text-to-{SQL} with Chain-of-Editions", author = "Zhang, Hanchong and Cao, Ruisheng and Xu, Hongshen and Chen, Lu and Yu, Kai", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.361", doi = "10.18653/v1/2024.naacl-long.361", pages = "6487--6508", abstract = "Recently, Large Language Models (LLMs) have been demonstrated to possess impressive capabilities in a variety of domains and tasks. We investigate the issue of prompt design in the multi-turn text-to-SQL task and attempt to enhance the LLMs{'} reasoning capacity when generating SQL queries. In the conversational context, the current SQL query can be modified from the preceding SQL query with only a few operations due to the context dependency. We introduce our method called CoE-SQL which can prompt LLMs to generate the SQL query based on the previously generated SQL query with an edition chain. We also conduct extensive ablation studies to determine the optimal configuration of our approach. Our approach outperforms different in-context learning baselines stably and achieves state-of-the-art performances on two benchmarks SParC and CoSQL using LLMs, which is also competitive to the SOTA fine-tuned models.", }

Recently, Large Language Models (LLMs) have been demonstrated to possess impressive capabilities in a variety of domains and tasks. We investigate the issue of prompt design in the multi-turn text-to-SQL task and attempt to enhance the LLMs{'} reasoning capacity when generating SQL queries. In the conversational context, the current SQL query can be modified from the preceding SQL query with only a few operations due to the context dependency. We introduce our method called CoE-SQL which can prompt LLMs to generate the SQL query based on the previously generated SQL query with an edition chain. We also conduct extensive ablation studies to determine the optimal configuration of our approach. Our approach outperforms different in-context learning baselines stably and achieves state-of-the-art performances on two benchmarks SParC and CoSQL using LLMs, which is also competitive to the SOTA fine-tuned models.

CoE-SQL: In-Context Learning for Multi-Turn Text-to-SQL with Chain-of-Editions

naacl-long.361

Poster

2405.02712

https://aclanthology.org/2024.naacl-long.362.bib

https://aclanthology.org/2024.naacl-long.362/

@inproceedings{li-etal-2024-contradoc, title = "{C}ontra{D}oc: Understanding Self-Contradictions in Documents with Large Language Models", author = "Li, Jierui and Raheja, Vipul and Kumar, Dhruv", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.362", doi = "10.18653/v1/2024.naacl-long.362", pages = "6509--6523", abstract = "In recent times, large language models (LLMs) have shown impressive performance on various document-level tasks such as document classification, summarization, and question-answering. However, research on understanding their capabilities on the task of self-contradictions in long documents has been very limited. In this work, we introduce ContraDoc, the first human-annotated dataset to study self-contradictions in long documents across multiple domains, varying document lengths, self-contradiction types, and appearance scope. We then analyze the current capabilities of four state-of-the-art open-source and commercially available LLMs: GPT3.5, GPT4, PaLM2, and LLaMAv2 on this dataset. While GPT4 performs the best and can outperform humans on this task, we find that it is still unreliable and struggles with self-contradictions that require more nuance and context. We release the dataset and all the code associated with the experiments.", }

In recent times, large language models (LLMs) have shown impressive performance on various document-level tasks such as document classification, summarization, and question-answering. However, research on understanding their capabilities on the task of self-contradictions in long documents has been very limited. In this work, we introduce ContraDoc, the first human-annotated dataset to study self-contradictions in long documents across multiple domains, varying document lengths, self-contradiction types, and appearance scope. We then analyze the current capabilities of four state-of-the-art open-source and commercially available LLMs: GPT3.5, GPT4, PaLM2, and LLaMAv2 on this dataset. While GPT4 performs the best and can outperform humans on this task, we find that it is still unreliable and struggles with self-contradictions that require more nuance and context. We release the dataset and all the code associated with the experiments.

ContraDoc: Understanding Self-Contradictions in Documents with Large Language Models

naacl-long.362

Poster

2311.09182

https://aclanthology.org/2024.naacl-long.363.bib

https://aclanthology.org/2024.naacl-long.363/

@inproceedings{wang-etal-2024-entity, title = "Entity Disambiguation via Fusion Entity Decoding", author = "Wang, Junxiong and Mousavi, Ali and Attia, Omar and Pradeep, Ronak and Potdar, Saloni and Rush, Alexander and Minhas, Umar Farooq and Li, Yunyao", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.363", doi = "10.18653/v1/2024.naacl-long.363", pages = "6524--6536", abstract = "Entity disambiguation (ED), which links the mentions of ambiguous entities to their referent entities in a knowledge base, serves as a core component in entity linking (EL). Existing generative approaches demonstrate improved accuracy compared to classification approaches under the standardized ZELDA benchmark. Nevertheless, generative approaches suffer from the need for large-scale pre-training and inefficient generation. Most importantly, entity descriptions, which could contain crucial information to distinguish similar entities from each other, are often overlooked.We propose an encoder-decoder model to disambiguate entities with more detailed entity descriptions. Given text and candidate entities, the encoder learns interactions between the text and each candidate entity, producing representations for each entity candidate. The decoder then fuses the representations of entity candidates together and selects the correct entity.Our experiments, conducted on various entity disambiguation benchmarks, demonstrate the strong and robust performance of this model, particularly +1.5{\%} in the ZELDA benchmark compared with GENRE. Furthermore, we integrate this approach into the retrieval/reader framework and observe +1.5{\%} improvements in end-to-end entity linking in the GERBIL benchmark compared with EntQA.", }

Entity disambiguation (ED), which links the mentions of ambiguous entities to their referent entities in a knowledge base, serves as a core component in entity linking (EL). Existing generative approaches demonstrate improved accuracy compared to classification approaches under the standardized ZELDA benchmark. Nevertheless, generative approaches suffer from the need for large-scale pre-training and inefficient generation. Most importantly, entity descriptions, which could contain crucial information to distinguish similar entities from each other, are often overlooked.We propose an encoder-decoder model to disambiguate entities with more detailed entity descriptions. Given text and candidate entities, the encoder learns interactions between the text and each candidate entity, producing representations for each entity candidate. The decoder then fuses the representations of entity candidates together and selects the correct entity.Our experiments, conducted on various entity disambiguation benchmarks, demonstrate the strong and robust performance of this model, particularly +1.5{\%} in the ZELDA benchmark compared with GENRE. Furthermore, we integrate this approach into the retrieval/reader framework and observe +1.5{\%} improvements in end-to-end entity linking in the GERBIL benchmark compared with EntQA.

Entity Disambiguation via Fusion Entity Decoding

naacl-long.363

Poster

2404.01626

https://aclanthology.org/2024.naacl-long.364.bib

https://aclanthology.org/2024.naacl-long.364/

@inproceedings{lee-etal-2024-planrag, title = "{P}lan{RAG}: A Plan-then-Retrieval Augmented Generation for Generative Large Language Models as Decision Makers", author = "Lee, Myeonghwa and An, Seonho and Kim, Min-Soo", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.364", doi = "10.18653/v1/2024.naacl-long.364", pages = "6537--6555", abstract = "In this paper, we conduct a study to utilize LLMs as a solution for decision making that requires complex data analysis. We define **Decision QA** as the task of answering the best decision, $d_{best}$, for a decision-making question $Q$, business rules $R$ and a database $D$. Since there is no benchmark that can examine Decision QA, we propose Decision QA benchmark, **DQA**. It has two scenarios, Locating and Building, constructed from two video games (Europa Universalis IV and Victoria 3) that have almost the same goal as Decision QA. To address Decision QA effectively, we also propose a new RAG technique called the *iterative plan-then-retrieval augmented generation* (**PlanRAG**). Our PlanRAG-based LM generates the plan for decision making as the first step, and the retriever generates the queries for data analysis as the second step. The proposed method outperforms the state-of-the-art iterative RAG method by 15.8{\%} in the Locating scenario and by 7.4{\%} in the Building scenario, respectively. We release our code and benchmark at https://github.com/myeon9h/PlanRAG.", }

In this paper, we conduct a study to utilize LLMs as a solution for decision making that requires complex data analysis. We define **Decision QA** as the task of answering the best decision, $d_{best}$, for a decision-making question $Q$, business rules $R$ and a database $D$. Since there is no benchmark that can examine Decision QA, we propose Decision QA benchmark, **DQA**. It has two scenarios, Locating and Building, constructed from two video games (Europa Universalis IV and Victoria 3) that have almost the same goal as Decision QA. To address Decision QA effectively, we also propose a new RAG technique called the *iterative plan-then-retrieval augmented generation* (**PlanRAG**). Our PlanRAG-based LM generates the plan for decision making as the first step, and the retriever generates the queries for data analysis as the second step. The proposed method outperforms the state-of-the-art iterative RAG method by 15.8{\%} in the Locating scenario and by 7.4{\%} in the Building scenario, respectively. We release our code and benchmark at https://github.com/myeon9h/PlanRAG.

PlanRAG: A Plan-then-Retrieval Augmented Generation for Generative Large Language Models as Decision Makers

naacl-long.364

Poster

2406.12430

https://huggingface.co/papers/2406.12430

https://aclanthology.org/2024.naacl-long.365.bib

https://aclanthology.org/2024.naacl-long.365/

@inproceedings{fu-etal-2024-gptscore, title = "{GPTS}core: Evaluate as You Desire", author = "Fu, Jinlan and Ng, See-Kiong and Jiang, Zhengbao and Liu, Pengfei", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.365", doi = "10.18653/v1/2024.naacl-long.365", pages = "6556--6576", abstract = "Generative Artificial Intelligence (AI) has enabled the development of sophisticated models that are capable of producing high-caliber text, images, and other outputs through the utilization of large pre-trained models.Nevertheless, assessing the quality of the generation is an even more arduous task than the generation itself, and this issue has not been given adequate consideration recently.This paper proposes a novel evaluation framework, GPTScore, which utilizes the emergent abilities (e.g., in-context learning, zero-shot instruction) of generative pre-trained models to score generated texts. There are 19 pre-trained models explored in this paper, ranging in size from 80M (e.g., Flan-T5-small) to 175B (e.g., GPT3).Experimental results on four text generation tasks, 22 evaluation aspects, and corresponding 37 datasets demonstrate that this approach can effectively allow us to achieve what one desires to evaluate for texts simply by natural language instructions.This nature helps us overcome several long-standing challenges in text evaluation{--}how to achieve customized, multi-faceted evaluation without model training. We make our code publicly available.", }

Generative Artificial Intelligence (AI) has enabled the development of sophisticated models that are capable of producing high-caliber text, images, and other outputs through the utilization of large pre-trained models.Nevertheless, assessing the quality of the generation is an even more arduous task than the generation itself, and this issue has not been given adequate consideration recently.This paper proposes a novel evaluation framework, GPTScore, which utilizes the emergent abilities (e.g., in-context learning, zero-shot instruction) of generative pre-trained models to score generated texts. There are 19 pre-trained models explored in this paper, ranging in size from 80M (e.g., Flan-T5-small) to 175B (e.g., GPT3).Experimental results on four text generation tasks, 22 evaluation aspects, and corresponding 37 datasets demonstrate that this approach can effectively allow us to achieve what one desires to evaluate for texts simply by natural language instructions.This nature helps us overcome several long-standing challenges in text evaluation{--}how to achieve customized, multi-faceted evaluation without model training. We make our code publicly available.

GPTScore: Evaluate as You Desire

naacl-long.365

Poster

2302.04166

https://aclanthology.org/2024.naacl-long.366.bib

https://aclanthology.org/2024.naacl-long.366/

@inproceedings{geng-etal-2024-survey, title = "A Survey of Confidence Estimation and Calibration in Large Language Models", author = "Geng, Jiahui and Cai, Fengyu and Wang, Yuxia and Koeppl, Heinz and Nakov, Preslav and Gurevych, Iryna", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.366", doi = "10.18653/v1/2024.naacl-long.366", pages = "6577--6595", abstract = "Large language models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks in various domains. Despite their impressive performance, they can be unreliable due to factual errors in their generations. Assessing their confidence and calibrating them across different tasks can help mitigate risks and enable LLMs to produce better generations. There has been a lot of recent research aiming to address this, but there has been no comprehensive overview to organize it and to outline the main lessons learned. The present survey aims to bridge this gap. In particular, we outline the challenges and we summarize recent technical advancements for LLM confidence estimation and calibration. We further discuss their applications and suggest promising directions for future work.", }

Large language models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks in various domains. Despite their impressive performance, they can be unreliable due to factual errors in their generations. Assessing their confidence and calibrating them across different tasks can help mitigate risks and enable LLMs to produce better generations. There has been a lot of recent research aiming to address this, but there has been no comprehensive overview to organize it and to outline the main lessons learned. The present survey aims to bridge this gap. In particular, we outline the challenges and we summarize recent technical advancements for LLM confidence estimation and calibration. We further discuss their applications and suggest promising directions for future work.

A Survey of Confidence Estimation and Calibration in Large Language Models

naacl-long.366

Poster

2311.08298

https://aclanthology.org/2024.naacl-long.367.bib

https://aclanthology.org/2024.naacl-long.367/

@inproceedings{tang-etal-2024-metrics, title = "Not All Metrics Are Guilty: Improving {NLG} Evaluation by Diversifying References", author = "Tang, Tianyi and Lu, Hongyuan and Jiang, Yuchen and Huang, Haoyang and Zhang, Dongdong and Zhao, Xin and Kocmi, Tom and Wei, Furu", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.367", doi = "10.18653/v1/2024.naacl-long.367", pages = "6596--6610", abstract = "Most research about natural language generation (NLG) relies on evaluation benchmarks with limited references for a sample, which may result in poor correlations with human judgements. The underlying reason is that one semantic meaning can actually be expressed in different forms, and the evaluation with a single or few references may not accurately reflect the quality of the model{'}s hypotheses. To address this issue, this paper presents a simple and effective method, named **Div-Ref**, to enhance existing evaluation benchmarks by enriching the number of references. We leverage large language models (LLMs) to diversify the expression of a single reference into multiple high-quality ones to cover the semantic space of the reference sentence as much as possible. We conduct comprehensive experiments to empirically demonstrate that diversifying the expression of reference can significantly enhance the correlation between automatic evaluation and human evaluation. This idea is compatible with recent LLM-based evaluation which can similarly derive advantages from incorporating multiple references. *We strongly encourage future generation benchmarks to include more references, even if they are generated by LLMs, which is once for all.* We release all the code and data at https://github.com/RUCAIBox/Div-Ref to facilitate research.", }

Most research about natural language generation (NLG) relies on evaluation benchmarks with limited references for a sample, which may result in poor correlations with human judgements. The underlying reason is that one semantic meaning can actually be expressed in different forms, and the evaluation with a single or few references may not accurately reflect the quality of the model{'}s hypotheses. To address this issue, this paper presents a simple and effective method, named **Div-Ref**, to enhance existing evaluation benchmarks by enriching the number of references. We leverage large language models (LLMs) to diversify the expression of a single reference into multiple high-quality ones to cover the semantic space of the reference sentence as much as possible. We conduct comprehensive experiments to empirically demonstrate that diversifying the expression of reference can significantly enhance the correlation between automatic evaluation and human evaluation. This idea is compatible with recent LLM-based evaluation which can similarly derive advantages from incorporating multiple references. *We strongly encourage future generation benchmarks to include more references, even if they are generated by LLMs, which is once for all.* We release all the code and data at https://github.com/RUCAIBox/Div-Ref to facilitate research.

Not All Metrics Are Guilty: Improving NLG Evaluation by Diversifying References

naacl-long.367

Poster

2305.15067

https://aclanthology.org/2024.naacl-long.368.bib

https://aclanthology.org/2024.naacl-long.368/

@inproceedings{xu-etal-2024-separation, title = "Separation and Fusion: A Novel Multiple Token Linking Model for Event Argument Extraction", author = "Xu, Jing and Song, Dandan and Hui, Siu and Wu, Zhijing and Jia, Meihuizi and Wang, Hao and Zhou, Yanru and Zhou, Changzhi and Yang, Ziyi", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.368", doi = "10.18653/v1/2024.naacl-long.368", pages = "6611--6624", abstract = "In event argument extraction (EAE), a promising approach involves jointly encoding text and argument roles, and performing multiple token linking operations. This approach further falls into two categories. One extracts arguments within a single event, while the other attempts to extract arguments from multiple events simultaneously. However, the former lacks to leverage cross-event information and the latter requires tougher predictions with longer encoded role sequences and extra linking operations. In this paper, we design a novel separation-and-fusion paradigm to separately acquire cross-event information and fuse it into the argument extraction of a target event. Following the paradigm, we propose a novel multiple token linking model named Sep2F, which can effectively build event correlations via roles and preserve the simple linking predictions of single-event extraction. In particular, we employ one linking module to extract arguments for the target event and another to aggregate the role information of multiple events. More importantly, we propose a novel two-fold fusion module to ensure that the aggregated cross-event information serves EAE well. We evaluate our proposed model on sentence-level and document-level datasets, including ACE05, RAMS, WikiEvents and MLEE. The extensive experimental results indicate that our model outperforms the state-of-the-art EAE models on all the datasets.", }

In event argument extraction (EAE), a promising approach involves jointly encoding text and argument roles, and performing multiple token linking operations. This approach further falls into two categories. One extracts arguments within a single event, while the other attempts to extract arguments from multiple events simultaneously. However, the former lacks to leverage cross-event information and the latter requires tougher predictions with longer encoded role sequences and extra linking operations. In this paper, we design a novel separation-and-fusion paradigm to separately acquire cross-event information and fuse it into the argument extraction of a target event. Following the paradigm, we propose a novel multiple token linking model named Sep2F, which can effectively build event correlations via roles and preserve the simple linking predictions of single-event extraction. In particular, we employ one linking module to extract arguments for the target event and another to aggregate the role information of multiple events. More importantly, we propose a novel two-fold fusion module to ensure that the aggregated cross-event information serves EAE well. We evaluate our proposed model on sentence-level and document-level datasets, including ACE05, RAMS, WikiEvents and MLEE. The extensive experimental results indicate that our model outperforms the state-of-the-art EAE models on all the datasets.

Separation and Fusion: A Novel Multiple Token Linking Model for Event Argument Extraction

naacl-long.368

Poster

https://aclanthology.org/2024.naacl-long.369.bib

https://aclanthology.org/2024.naacl-long.369/

@inproceedings{li-etal-2024-integration, title = "The Integration of Semantic and Structural Knowledge in Knowledge Graph Entity Typing", author = "Li, Muzhi and Hu, Minda and King, Irwin and Leung, Ho-fung", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.369", doi = "10.18653/v1/2024.naacl-long.369", pages = "6625--6638", abstract = "The Knowledge Graph Entity Typing (KGET) task aims to predict missing type annotations for entities in knowledge graphs. Recent works only utilize the \textit{ \textbf{structural knowledge}} in the local neighborhood of entities, disregarding \textit{ \textbf{semantic knowledge}} in the textual representations of entities, relations, and types that are also crucial for type inference. Additionally, we observe that the interaction between semantic and structural knowledge can be utilized to address the false-negative problem. In this paper, we propose a novel \textbf{S}emantic and \textbf{S}tructure-aware KG \textbf{E}ntity \textbf{T}yping (SSET) framework, which is composed of three modules. First, the \textit{Semantic Knowledge Encoding} module encodes factual knowledge in the KG with a Masked Entity Typing task. Then, the \textit{Structural Knowledge Aggregation} module aggregates knowledge from the multi-hop neighborhood of entities to infer missing types. Finally, the \textit{Unsupervised Type Re-ranking} module utilizes the inference results from the two models above to generate type predictions that are robust to false-negative samples. Extensive experiments show that SSET significantly outperforms existing state-of-the-art methods.", }

The Knowledge Graph Entity Typing (KGET) task aims to predict missing type annotations for entities in knowledge graphs. Recent works only utilize the \textit{ \textbf{structural knowledge}} in the local neighborhood of entities, disregarding \textit{ \textbf{semantic knowledge}} in the textual representations of entities, relations, and types that are also crucial for type inference. Additionally, we observe that the interaction between semantic and structural knowledge can be utilized to address the false-negative problem. In this paper, we propose a novel \textbf{S}emantic and \textbf{S}tructure-aware KG \textbf{E}ntity \textbf{T}yping (SSET) framework, which is composed of three modules. First, the \textit{Semantic Knowledge Encoding} module encodes factual knowledge in the KG with a Masked Entity Typing task. Then, the \textit{Structural Knowledge Aggregation} module aggregates knowledge from the multi-hop neighborhood of entities to infer missing types. Finally, the \textit{Unsupervised Type Re-ranking} module utilizes the inference results from the two models above to generate type predictions that are robust to false-negative samples. Extensive experiments show that SSET significantly outperforms existing state-of-the-art methods.

The Integration of Semantic and Structural Knowledge in Knowledge Graph Entity Typing

naacl-long.369

Oral

2404.08313

https://aclanthology.org/2024.naacl-long.370.bib

https://aclanthology.org/2024.naacl-long.370/

@inproceedings{jiang-etal-2024-comclip, title = "{C}om{CLIP}: Training-Free Compositional Image and Text Matching", author = "Jiang, Kenan and He, Xuehai and Xu, Ruize and Wang, Xin", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.370", doi = "10.18653/v1/2024.naacl-long.370", pages = "6639--6659", abstract = "Contrastive Language-Image Pretraining (CLIP) has demonstrated great zero-shot performance for matching images and text. However, it is still challenging to adapt vision-language pretrained models like CLIP to compositional image and text matching {---} a more challenging image and text matching task requiring the model{'}s understanding of compositional word concepts and visual components. Towards better compositional generalization in zero-shot image and text matching, in this paper, we study the problem from a causal perspective: the erroneous semantics of individual entities are essentially confounders that cause the matching failure. Therefore, we propose a novel training-free compositional CLIP model (ComCLIP). ComCLIP disentangles input images into subjects, objects, and action subimages and composes CLIP{'}s vision encoder and text encoder to perform evolving matching over compositional text embedding and subimage embeddings. In this way, ComCLIP can mitigate spurious correlations introduced by the pretrained CLIP models and dynamically evaluate the importance of each component. Experiments on four compositional image-text matching datasets: Winoground, VL-checklist, SVO, and ComVG, and two general image-text retrieval datasets: Flick30K, and MSCOCO demonstrate the effectiveness of our plug-and-play method, which boosts the zero-shot inference ability of CLIP, SLIP, and BLIP2 even without further training or fine-tuning. Our codes can be found at https://github.com/eric-ai-lab/ComCLIP.", }

Contrastive Language-Image Pretraining (CLIP) has demonstrated great zero-shot performance for matching images and text. However, it is still challenging to adapt vision-language pretrained models like CLIP to compositional image and text matching {---} a more challenging image and text matching task requiring the model{'}s understanding of compositional word concepts and visual components. Towards better compositional generalization in zero-shot image and text matching, in this paper, we study the problem from a causal perspective: the erroneous semantics of individual entities are essentially confounders that cause the matching failure. Therefore, we propose a novel training-free compositional CLIP model (ComCLIP). ComCLIP disentangles input images into subjects, objects, and action subimages and composes CLIP{'}s vision encoder and text encoder to perform evolving matching over compositional text embedding and subimage embeddings. In this way, ComCLIP can mitigate spurious correlations introduced by the pretrained CLIP models and dynamically evaluate the importance of each component. Experiments on four compositional image-text matching datasets: Winoground, VL-checklist, SVO, and ComVG, and two general image-text retrieval datasets: Flick30K, and MSCOCO demonstrate the effectiveness of our plug-and-play method, which boosts the zero-shot inference ability of CLIP, SLIP, and BLIP2 even without further training or fine-tuning. Our codes can be found at https://github.com/eric-ai-lab/ComCLIP.

ComCLIP: Training-Free Compositional Image and Text Matching

naacl-long.370

Poster

2211.13854

https://huggingface.co/papers/2211.13854

https://aclanthology.org/2024.naacl-long.371.bib

https://aclanthology.org/2024.naacl-long.371/

@inproceedings{takeshita-etal-2024-aclsum, title = "{ACLS}um: A New Dataset for Aspect-based Summarization of Scientific Publications", author = "Takeshita, Sotaro and Green, Tommaso and Reinig, Ines and Eckert, Kai and Ponzetto, Simone", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.371", doi = "10.18653/v1/2024.naacl-long.371", pages = "6660--6675", abstract = "Extensive efforts in the past have been directed toward the development of summarization datasets. However, a predominant number of these resources have been (semi)-automatically generated, typically through web data crawling. This resulted in subpar resources for training and evaluating summarization systems, a quality compromise that is arguably due to the substantial costs associated with generating ground-truth summaries, particularly for diverse languages and specialized domains. To address this issue, we present ACLSum, a novel summarization dataset carefully crafted and evaluated by domain experts. In contrast to previous datasets, ACLSum facilitates multi-aspect summarization of scientific papers, covering challenges, approaches, and outcomes in depth. Through extensive experiments, we evaluate the quality of our resource and the performance of models based on pretrained language models (PLMs) and state-of-the-art large language models (LLMs). Additionally, we explore the effectiveness of extract-then-abstract versus abstractive end-to-end summarization within the scholarly domain on the basis of automatically discovered aspects. While the former performs comparably well to the end-to-end approach with pretrained language models regardless of the potential error propagation issue, the prompting-based approach with LLMs shows a limitation in extracting sentences from source documents.", }

Extensive efforts in the past have been directed toward the development of summarization datasets. However, a predominant number of these resources have been (semi)-automatically generated, typically through web data crawling. This resulted in subpar resources for training and evaluating summarization systems, a quality compromise that is arguably due to the substantial costs associated with generating ground-truth summaries, particularly for diverse languages and specialized domains. To address this issue, we present ACLSum, a novel summarization dataset carefully crafted and evaluated by domain experts. In contrast to previous datasets, ACLSum facilitates multi-aspect summarization of scientific papers, covering challenges, approaches, and outcomes in depth. Through extensive experiments, we evaluate the quality of our resource and the performance of models based on pretrained language models (PLMs) and state-of-the-art large language models (LLMs). Additionally, we explore the effectiveness of extract-then-abstract versus abstractive end-to-end summarization within the scholarly domain on the basis of automatically discovered aspects. While the former performs comparably well to the end-to-end approach with pretrained language models regardless of the potential error propagation issue, the prompting-based approach with LLMs shows a limitation in extracting sentences from source documents.

ACLSum: A New Dataset for Aspect-based Summarization of Scientific Publications

naacl-long.371

Poster

2403.05303

https://aclanthology.org/2024.naacl-long.372.bib

https://aclanthology.org/2024.naacl-long.372/

@inproceedings{luo-etal-2024-xal, title = "{XAL}: {EX}plainable Active Learning Makes Classifiers Better Low-resource Learners", author = "Luo, Yun and Yang, Zhen and Meng, Fandong and Li, Yingjie and Guo, Fang and Qi, Qinglin and Zhou, Jie and Zhang, Yue", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.372", doi = "10.18653/v1/2024.naacl-long.372", pages = "6676--6698", abstract = "Active learning (AL), which aims to construct an effective training set by iteratively curating the most formative unlabeled data for annotation, has been widely used in low-resource tasks. Most active learning techniques in classification rely on the model{'}s uncertainty or disagreement to choose unlabeled data, suffering from the problem of over-confidence in superficial patterns and a lack of exploration.Inspired by the cognitive processes in which humans deduce and predict through causal information, we take an initial attempt towards integrating rationales into AL and propose a novel Explainable Active Learning framework (XAL) for low-resource text classification, which aims to encourage classifiers to justify their inferences and delve into unlabeled data for which they cannot provide reasonable explanations. Specifically, besides using a pre-trained bi-directional encoder for classification, we employ a pre-trained uni-directional decoder to generate and score the explanation. We further facilitate the alignment of the model with human reasoning preference through a proposed ranking loss. During the selection of unlabeled data, the predicted uncertainty of the encoder and the explanation score of the decoder complement each other as the final metric to acquire informative data. Extensive experiments on six datasets show that XAL achieves consistent improvement over 9 strong baselines. Analysis indicates that the proposed method can generate corresponding explanations for its predictions.", }

Active learning (AL), which aims to construct an effective training set by iteratively curating the most formative unlabeled data for annotation, has been widely used in low-resource tasks. Most active learning techniques in classification rely on the model{'}s uncertainty or disagreement to choose unlabeled data, suffering from the problem of over-confidence in superficial patterns and a lack of exploration.Inspired by the cognitive processes in which humans deduce and predict through causal information, we take an initial attempt towards integrating rationales into AL and propose a novel Explainable Active Learning framework (XAL) for low-resource text classification, which aims to encourage classifiers to justify their inferences and delve into unlabeled data for which they cannot provide reasonable explanations. Specifically, besides using a pre-trained bi-directional encoder for classification, we employ a pre-trained uni-directional decoder to generate and score the explanation. We further facilitate the alignment of the model with human reasoning preference through a proposed ranking loss. During the selection of unlabeled data, the predicted uncertainty of the encoder and the explanation score of the decoder complement each other as the final metric to acquire informative data. Extensive experiments on six datasets show that XAL achieves consistent improvement over 9 strong baselines. Analysis indicates that the proposed method can generate corresponding explanations for its predictions.

XAL: EXplainable Active Learning Makes Classifiers Better Low-resource Learners

naacl-long.372

Poster

2310.05502

https://aclanthology.org/2024.naacl-long.373.bib

https://aclanthology.org/2024.naacl-long.373/

@inproceedings{wang-etal-2024-ladic, title = "{L}a{D}i{C}: Are Diffusion Models Really Inferior to Autoregressive Counterparts for Image-to-Text Generation?", author = "Wang, Yuchi and Ren, Shuhuai and Gao, Rundong and Yao, Linli and Guo, Qingyan and An, Kaikai and Bai, Jianhong and Sun, Xu", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.373", doi = "10.18653/v1/2024.naacl-long.373", pages = "6699--6715", abstract = "Diffusion models have exhibited remarkable capabilities in text-to-image generation. However, their performance in image-to-text generation, specifically image captioning, has lagged behind Auto-Regressive (AR) models, casting doubt on their applicability for such tasks. In this work, we revisit diffusion models, highlighting their capacity for holistic context modeling and parallel decoding. With these benefits, diffusion models can alleviate the inherent limitations of AR methods, including their slow inference speed, error propagation, and unidirectional constraints. Furthermore, we identify the prior underperformance of diffusion models stemming from the absence of an effective latent space for image-text alignment, and the discrepancy between continuous diffusion processes and discrete textual data. In response, we introduce a novel architecture, LaDiC, which utilizes a split BERT to create a dedicated latent space for captions and integrates a regularization module to manage varying text lengths. Our framework also includes a diffuser for semantic image-to-text conversion and a Back{\&}Refine technique to enhance token interactivity during inference. LaDiC achieves state-of-the-art performance for diffusion-based methods on the MS COCO dataset with 38.2 BLEU@4 and 126.2 CIDEr, demonstrating exceptional performance without pre-training or ancillary modules. This indicates strong competitiveness with AR models, revealing the previously untapped potential of diffusion models in image-to-text generation.", }

Diffusion models have exhibited remarkable capabilities in text-to-image generation. However, their performance in image-to-text generation, specifically image captioning, has lagged behind Auto-Regressive (AR) models, casting doubt on their applicability for such tasks. In this work, we revisit diffusion models, highlighting their capacity for holistic context modeling and parallel decoding. With these benefits, diffusion models can alleviate the inherent limitations of AR methods, including their slow inference speed, error propagation, and unidirectional constraints. Furthermore, we identify the prior underperformance of diffusion models stemming from the absence of an effective latent space for image-text alignment, and the discrepancy between continuous diffusion processes and discrete textual data. In response, we introduce a novel architecture, LaDiC, which utilizes a split BERT to create a dedicated latent space for captions and integrates a regularization module to manage varying text lengths. Our framework also includes a diffuser for semantic image-to-text conversion and a Back{\&}Refine technique to enhance token interactivity during inference. LaDiC achieves state-of-the-art performance for diffusion-based methods on the MS COCO dataset with 38.2 BLEU@4 and 126.2 CIDEr, demonstrating exceptional performance without pre-training or ancillary modules. This indicates strong competitiveness with AR models, revealing the previously untapped potential of diffusion models in image-to-text generation.

LaDiC: Are Diffusion Models Really Inferior to Autoregressive Counterparts for Image-to-Text Generation?

naacl-long.373

Poster

2404.10763

https://huggingface.co/papers/2404.10763

https://aclanthology.org/2024.naacl-long.374.bib

https://aclanthology.org/2024.naacl-long.374/

@inproceedings{hengle-etal-2024-intent, title = "Intent-conditioned and Non-toxic Counterspeech Generation using Multi-Task Instruction Tuning with {RLAIF}", author = "Hengle, Amey and Padhi, Aswini and Singh, Sahajpreet and Bandhakavi, Anil and Akhtar, Md Shad and Chakraborty, Tanmoy", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.374", doi = "10.18653/v1/2024.naacl-long.374", pages = "6716--6733", abstract = "Counterspeech, defined as a response to mitigate online hate speech, is increasingly used as a non-censorial solution. The effectiveness of addressing hate speech involves dispelling the stereotypes, prejudices, and biases often subtly implied in brief, single-sentence statements or abuses. These expressions challenge language models, especially in seq2seq tasks, as model performance typically excels with longer contexts. Our study introduces CoARL, a novel framework enhancing counterspeech generation by modeling the pragmatic implications underlying social biases in hateful statements. The first two phases of CoARL involve sequential multi-instruction tuning, teaching the model to understand intents, reactions, and harms of offensive statements, and then learning task-specific low-rank adapter weights for generating intent-conditioned counterspeech. The final phase uses reinforcement learning to fine-tune outputs for effectiveness and nontoxicity. CoARL outperforms existing benchmarks in intent-conditioned counterspeech generation, showing an average improvement of ∼3 points in intent-conformity and ∼4 points in argument-quality metrics. Extensive human evaluation supports CoARL{'}s efficacy in generating superior and more context-appropriate responses compared to existing systems, including prominent LLMs like ChatGPT.", }

Counterspeech, defined as a response to mitigate online hate speech, is increasingly used as a non-censorial solution. The effectiveness of addressing hate speech involves dispelling the stereotypes, prejudices, and biases often subtly implied in brief, single-sentence statements or abuses. These expressions challenge language models, especially in seq2seq tasks, as model performance typically excels with longer contexts. Our study introduces CoARL, a novel framework enhancing counterspeech generation by modeling the pragmatic implications underlying social biases in hateful statements. The first two phases of CoARL involve sequential multi-instruction tuning, teaching the model to understand intents, reactions, and harms of offensive statements, and then learning task-specific low-rank adapter weights for generating intent-conditioned counterspeech. The final phase uses reinforcement learning to fine-tune outputs for effectiveness and nontoxicity. CoARL outperforms existing benchmarks in intent-conditioned counterspeech generation, showing an average improvement of ∼3 points in intent-conformity and ∼4 points in argument-quality metrics. Extensive human evaluation supports CoARL{'}s efficacy in generating superior and more context-appropriate responses compared to existing systems, including prominent LLMs like ChatGPT.

Intent-conditioned and Non-toxic Counterspeech Generation using Multi-Task Instruction Tuning with RLAIF

naacl-long.374

Poster

2403.10088

https://huggingface.co/papers/2403.10088

https://aclanthology.org/2024.naacl-long.375.bib

https://aclanthology.org/2024.naacl-long.375/

@inproceedings{dong-etal-2024-attacks, title = "Attacks, Defenses and Evaluations for {LLM} Conversation Safety: A Survey", author = "Dong, Zhichen and Zhou, Zhanhui and Yang, Chao and Shao, Jing and Qiao, Yu", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.375", doi = "10.18653/v1/2024.naacl-long.375", pages = "6734--6747", abstract = "Large Language Models (LLMs) are now commonplace in conversation applications. However, their risks of misuse for generating harmful responses have raised serious societal concerns and spurred recent research on LLM conversation safety. Therefore, in this survey, we provide a comprehensive overview of recent studies, covering three critical aspects of LLM conversation safety: attacks, defenses, and evaluations. Our goal is to provide a structured summary that enhances understanding of LLM conversation safety and encourages further investigation into this important subject. For easy reference, we have categorized all the studies mentioned in this survey according to our taxonomy, available at: https://github.com/niconi19/LLM-conversation-safety.", }

Large Language Models (LLMs) are now commonplace in conversation applications. However, their risks of misuse for generating harmful responses have raised serious societal concerns and spurred recent research on LLM conversation safety. Therefore, in this survey, we provide a comprehensive overview of recent studies, covering three critical aspects of LLM conversation safety: attacks, defenses, and evaluations. Our goal is to provide a structured summary that enhances understanding of LLM conversation safety and encourages further investigation into this important subject. For easy reference, we have categorized all the studies mentioned in this survey according to our taxonomy, available at: https://github.com/niconi19/LLM-conversation-safety.

Attacks, Defenses and Evaluations for LLM Conversation Safety: A Survey

naacl-long.375

Poster

2402.09283

https://aclanthology.org/2024.naacl-long.376.bib

https://aclanthology.org/2024.naacl-long.376/

@inproceedings{liu-etal-2024-minds, title = "Mind{'}s Mirror: Distilling Self-Evaluation Capability and Comprehensive Thinking from Large Language Models", author = "Liu, Weize and Li, Guocong and Zhang, Kai and Du, Bang and Chen, Qiyuan and Hu, Xuming and Xu, Hongxia and Chen, Jintai and Wu, Jian", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.376", doi = "10.18653/v1/2024.naacl-long.376", pages = "6748--6763", abstract = "Large language models (LLMs) have achieved remarkable advancements in natural language processing. However, the massive scale and computational demands of these models present formidable challenges when considering their practical deployment in resource-constrained environments. While techniques such as chain-of-thought (CoT) distillation have displayed promise in distilling LLMs into small language models (SLMs), there is a risk that distilled SLMs may still inherit flawed reasoning and hallucinations from LLMs. To address these issues, we propose a twofold methodology: First, we introduce a novel method for distilling the self-evaluation capability from LLMs into SLMs, aiming to mitigate the adverse effects of flawed reasoning and hallucinations inherited from LLMs. Second, we advocate for distilling more comprehensive thinking by incorporating multiple distinct CoTs and self-evaluation outputs, to ensure a more thorough and robust knowledge transfer into SLMs. Experiments on three NLP benchmarks demonstrate that our method significantly improves the performance of distilled SLMs, offering a new perspective for developing more effective and efficient SLMs in resource-constrained environments.", }

Large language models (LLMs) have achieved remarkable advancements in natural language processing. However, the massive scale and computational demands of these models present formidable challenges when considering their practical deployment in resource-constrained environments. While techniques such as chain-of-thought (CoT) distillation have displayed promise in distilling LLMs into small language models (SLMs), there is a risk that distilled SLMs may still inherit flawed reasoning and hallucinations from LLMs. To address these issues, we propose a twofold methodology: First, we introduce a novel method for distilling the self-evaluation capability from LLMs into SLMs, aiming to mitigate the adverse effects of flawed reasoning and hallucinations inherited from LLMs. Second, we advocate for distilling more comprehensive thinking by incorporating multiple distinct CoTs and self-evaluation outputs, to ensure a more thorough and robust knowledge transfer into SLMs. Experiments on three NLP benchmarks demonstrate that our method significantly improves the performance of distilled SLMs, offering a new perspective for developing more effective and efficient SLMs in resource-constrained environments.

Mind's Mirror: Distilling Self-Evaluation Capability and Comprehensive Thinking from Large Language Models

naacl-long.376

Poster

2311.09214

https://aclanthology.org/2024.naacl-long.377.bib

https://aclanthology.org/2024.naacl-long.377/

@inproceedings{deiseroth-etal-2024-divergent, title = "Divergent Token Metrics: Measuring degradation to prune away {LLM} components {--} and optimize quantization", author = {Deiseroth, Bj{\"o}rn and Meuer, Max and Gritsch, Nikolas and Eichenberg, Constantin and Schramowski, Patrick and A{\ss}enmacher, Matthias and Kersting, Kristian}, editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.377", doi = "10.18653/v1/2024.naacl-long.377", pages = "6764--6783", abstract = "Large Language Models (LLMs) have reshaped natural language processing with their impressive capabilities. However, their ever-increasing size has raised concerns about their effective deployment and the need for LLM compression. This study introduces the Divergent Token Metrics (DTMs), a novel approach to assessing compressed LLMs, addressing the limitations of traditional perplexity or accuracy measures that fail to accurately reflect text generation quality. DTMs measure token divergences that allow deeper insights into the subtleties of model compression, in particular, when evaluating components{'} impacts individually. Utilizing the First Divergent Token Metric (FDTM) in model sparsification reveals that 25{\%} of all attention components can be pruned beyond 90{\%} on the Llama-2 model family, still keeping SOTA performance. For quantization, FDTM suggests that more than 80{\%} of parameters can be naively transformed to int8 without special outlier management. These evaluations indicate the necessity of choosing appropriate compressions for parameters individually{---}and that FDTM can identify those{---}while standard metrics result in deteriorated outcomes.", }

Large Language Models (LLMs) have reshaped natural language processing with their impressive capabilities. However, their ever-increasing size has raised concerns about their effective deployment and the need for LLM compression. This study introduces the Divergent Token Metrics (DTMs), a novel approach to assessing compressed LLMs, addressing the limitations of traditional perplexity or accuracy measures that fail to accurately reflect text generation quality. DTMs measure token divergences that allow deeper insights into the subtleties of model compression, in particular, when evaluating components{'} impacts individually. Utilizing the First Divergent Token Metric (FDTM) in model sparsification reveals that 25{\%} of all attention components can be pruned beyond 90{\%} on the Llama-2 model family, still keeping SOTA performance. For quantization, FDTM suggests that more than 80{\%} of parameters can be naively transformed to int8 without special outlier management. These evaluations indicate the necessity of choosing appropriate compressions for parameters individually{---}and that FDTM can identify those{---}while standard metrics result in deteriorated outcomes.

Divergent Token Metrics: Measuring degradation to prune away LLM components – and optimize quantization

naacl-long.377

Poster

https://aclanthology.org/2024.naacl-long.378.bib

https://aclanthology.org/2024.naacl-long.378/

@inproceedings{reif-schwartz-2024-beyond, title = "Beyond Performance: Quantifying and Mitigating Label Bias in {LLM}s", author = "Reif, Yuval and Schwartz, Roy", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.378", doi = "10.18653/v1/2024.naacl-long.378", pages = "6784--6798", abstract = "Large language models (LLMs) have shown remarkable adaptability to diverse tasks, by leveraging context prompts containing instructions, or minimal input-output examples. However, recent work revealed they also exhibit *label bias*{---}an undesirable preference toward predicting certain answers over others. Still, detecting and measuring this bias reliably and at scale has remained relatively unexplored. In this study, we evaluate different approaches to quantifying label bias in a model{'}s predictions, conducting a comprehensive investigation across 279 classification tasks and ten LLMs. Our investigation reveals substantial label bias in models both before and after debiasing attempts, as well as highlights the importance of outcomes-based evaluation metrics, which were not previously used in this regard. We further propose a novel label bias calibration method tailored for few-shot prompting, which outperforms recent calibration approaches for both improving performance and mitigating label bias. Our results emphasize that label bias in the predictions of LLMs remains a barrier to their reliability.", }

Large language models (LLMs) have shown remarkable adaptability to diverse tasks, by leveraging context prompts containing instructions, or minimal input-output examples. However, recent work revealed they also exhibit *label bias*{---}an undesirable preference toward predicting certain answers over others. Still, detecting and measuring this bias reliably and at scale has remained relatively unexplored. In this study, we evaluate different approaches to quantifying label bias in a model{'}s predictions, conducting a comprehensive investigation across 279 classification tasks and ten LLMs. Our investigation reveals substantial label bias in models both before and after debiasing attempts, as well as highlights the importance of outcomes-based evaluation metrics, which were not previously used in this regard. We further propose a novel label bias calibration method tailored for few-shot prompting, which outperforms recent calibration approaches for both improving performance and mitigating label bias. Our results emphasize that label bias in the predictions of LLMs remains a barrier to their reliability.

Beyond Performance: Quantifying and Mitigating Label Bias in LLMs

naacl-long.378

Poster

2405.02743

https://aclanthology.org/2024.naacl-long.379.bib

https://aclanthology.org/2024.naacl-long.379/

@inproceedings{wu-etal-2024-instructing, title = "Instructing Large Language Models to Identify and Ignore Irrelevant Conditions", author = "Wu, Zhenyu and Shen, Chao and Jiang, Meng", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.379", doi = "10.18653/v1/2024.naacl-long.379", pages = "6799--6819", abstract = "Math word problem (MWP) solving requires generating a reasoning path based on a given problem description that often contains irrelevant conditions.Existing chain-of-thought (CoT) prompting methods elicited multi-step reasoning abilities of large language models (LLMs) to solve MWPs.However, they were seriously confused by the irrelevant conditions, resulting in low accuracy.In this paper, we propose a novel approach named I$^3$C that instructs LLMs to identify and ignore irrelevant conditions.It identifies a set of irrelevant condition candidates that have a weak semantic relevance with the question.Then it prompts LLMs to verify the irrelevant conditions.Lastly it instructs the LLMs with the verification on relevant and irrelevant conditions to avoid confusion and improve reasoning paths.Moreover, we propose to select (problem, reasoning paths) pairs as demonstrations to enhance I$^3$C with few-shot reasoning. We develop I$^3$C-Select that selects the most confusing problems based on the semantic relevance measurement.We conduct extensive experiments on eight MWP datasets.I$^3$C can be combined with any CoT prompting methods to improve the performance of solving MWPs.Notably, with GPT-3.5-Turbo and I$^3$C-Select, we achieve an accuracy of 96.0 and 94.1 on GSM-IC2-1K and GSM-ICM-1K, respectively, significantly outperforming the state-of-the-art few-shot prompting method Complex-CoT by $+11.7$ and $+11.1$.Our implementation is made publicly available at https://wzy6642.github.io/I3C.github.io/.", }

Math word problem (MWP) solving requires generating a reasoning path based on a given problem description that often contains irrelevant conditions.Existing chain-of-thought (CoT) prompting methods elicited multi-step reasoning abilities of large language models (LLMs) to solve MWPs.However, they were seriously confused by the irrelevant conditions, resulting in low accuracy.In this paper, we propose a novel approach named I$^3$C that instructs LLMs to identify and ignore irrelevant conditions.It identifies a set of irrelevant condition candidates that have a weak semantic relevance with the question.Then it prompts LLMs to verify the irrelevant conditions.Lastly it instructs the LLMs with the verification on relevant and irrelevant conditions to avoid confusion and improve reasoning paths.Moreover, we propose to select (problem, reasoning paths) pairs as demonstrations to enhance I$^3$C with few-shot reasoning. We develop I$^3$C-Select that selects the most confusing problems based on the semantic relevance measurement.We conduct extensive experiments on eight MWP datasets.I$^3$C can be combined with any CoT prompting methods to improve the performance of solving MWPs.Notably, with GPT-3.5-Turbo and I$^3$C-Select, we achieve an accuracy of 96.0 and 94.1 on GSM-IC2-1K and GSM-ICM-1K, respectively, significantly outperforming the state-of-the-art few-shot prompting method Complex-CoT by $+11.7$ and $+11.1$.Our implementation is made publicly available at https://wzy6642.github.io/I3C.github.io/.

Instructing Large Language Models to Identify and Ignore Irrelevant Conditions

naacl-long.379

Poster

2403.12744

https://huggingface.co/papers/2403.12744

https://aclanthology.org/2024.naacl-long.380.bib

https://aclanthology.org/2024.naacl-long.380/

@inproceedings{svete-etal-2024-lower, title = "Lower Bounds on the Expressivity of Recurrent Neural Language Models", author = "Svete, Anej and Nowak, Franz and Sahabdeen, Anisha and Cotterell, Ryan", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.380", doi = "10.18653/v1/2024.naacl-long.380", pages = "6820--6844", abstract = "The recent successes and spread of large neural language models (LMs) call for a thorough understanding of their abilities. Describing their abilities through LMs{'} representational capacity is a lively area of research. Investigations of the representational capacity of neural LMs have predominantly focused on their ability to recognize formal languages. For example, recurrent neural networks (RNNs) as classifiers are tightly linked to regular languages, i.e., languages defined by finite-state automata (FSAs). Such results, however, fall short of describing the capabilities of RNN language models (LMs), which are definitionally distributions over strings. We take a fresh look at the represen- tational capacity of RNN LMs by connecting them to probabilistic FSAs and demonstrate that RNN LMs with linearly bounded precision can express arbitrary regular LMs.", }

The recent successes and spread of large neural language models (LMs) call for a thorough understanding of their abilities. Describing their abilities through LMs{'} representational capacity is a lively area of research. Investigations of the representational capacity of neural LMs have predominantly focused on their ability to recognize formal languages. For example, recurrent neural networks (RNNs) as classifiers are tightly linked to regular languages, i.e., languages defined by finite-state automata (FSAs). Such results, however, fall short of describing the capabilities of RNN language models (LMs), which are definitionally distributions over strings. We take a fresh look at the represen- tational capacity of RNN LMs by connecting them to probabilistic FSAs and demonstrate that RNN LMs with linearly bounded precision can express arbitrary regular LMs.

Lower Bounds on the Expressivity of Recurrent Neural Language Models

naacl-long.380

Poster

2405.19222

https://aclanthology.org/2024.naacl-long.381.bib

https://aclanthology.org/2024.naacl-long.381/

@inproceedings{svete-cotterell-2024-transformers, title = "Transformers Can Represent $n$-gram Language Models", author = "Svete, Anej and Cotterell, Ryan", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.381", doi = "10.18653/v1/2024.naacl-long.381", pages = "6845--6881", abstract = "Plenty of existing work has analyzed the abilities of the transformer architecture by describing its representational capacity with formal models of computation. However, the focus so far has been on analyzing the architecture in terms of language \textit{acceptance}. We contend that this is an ill-suited problem in the study of \textit{language models} (LMs), which are definitionally \textit{probability distributions} over strings. In this paper, we focus on the relationship between transformer LMs and $n$-gram LMs, a simple and historically relevant class of language models. We show that transformer LMs using the hard or sparse attention mechanisms can exactly represent any $n$-gram LM, giving us a concrete lower bound on their probabilistic representational capacity. This provides a first step towards understanding the mechanisms that transformer LMs can use to represent probability distributions over strings.", }

Plenty of existing work has analyzed the abilities of the transformer architecture by describing its representational capacity with formal models of computation. However, the focus so far has been on analyzing the architecture in terms of language \textit{acceptance}. We contend that this is an ill-suited problem in the study of \textit{language models} (LMs), which are definitionally \textit{probability distributions} over strings. In this paper, we focus on the relationship between transformer LMs and $n$-gram LMs, a simple and historically relevant class of language models. We show that transformer LMs using the hard or sparse attention mechanisms can exactly represent any $n$-gram LM, giving us a concrete lower bound on their probabilistic representational capacity. This provides a first step towards understanding the mechanisms that transformer LMs can use to represent probability distributions over strings.

Transformers Can Represent n-gram Language Models

naacl-long.381

Poster

https://aclanthology.org/2024.naacl-long.382.bib

https://aclanthology.org/2024.naacl-long.382/

@inproceedings{malagutti-etal-2024-role, title = "The Role of $n$-gram Smoothing in the Age of Neural Networks", author = "Malagutti, Luca and Buinovskij, Andrius and Svete, Anej and Meister, Clara and Amini, Afra and Cotterell, Ryan", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.382", doi = "10.18653/v1/2024.naacl-long.382", pages = "6882--6899", abstract = "For nearly three decades, language models derived from the $n$-gram assumption held the state of the art on the task. The key to their success lay in the application of various smoothing techniques that served to combat overfitting. However, when neural language models toppled $n$-gram models as the best performers, $n$-gram smoothing techniques became less relevant. Indeed, it would hardly be an understatement to suggest that the line of inquiry into $n$-gram smoothing techniques became dormant. This paper re-opens the role classical $n$-gram smoothing techniques may play in the age of neural language models. First, we draw a formal equivalence between label smoothing, a popular regularization technique for neural language models, and add-$\lambda$ smoothing. Second, we derive a generalized framework for converting any $n$-gram smoothing technique into a regularizer compatible with neural language models. Our empirical results find that our novel regularizers are comparable to and, indeed, sometimes outperform label smoothing on language modeling and machine translation.", }

For nearly three decades, language models derived from the $n$-gram assumption held the state of the art on the task. The key to their success lay in the application of various smoothing techniques that served to combat overfitting. However, when neural language models toppled $n$-gram models as the best performers, $n$-gram smoothing techniques became less relevant. Indeed, it would hardly be an understatement to suggest that the line of inquiry into $n$-gram smoothing techniques became dormant. This paper re-opens the role classical $n$-gram smoothing techniques may play in the age of neural language models. First, we draw a formal equivalence between label smoothing, a popular regularization technique for neural language models, and add-$\lambda$ smoothing. Second, we derive a generalized framework for converting any $n$-gram smoothing technique into a regularizer compatible with neural language models. Our empirical results find that our novel regularizers are comparable to and, indeed, sometimes outperform label smoothing on language modeling and machine translation.

The Role of n-gram Smoothing in the Age of Neural Networks

naacl-long.382

Poster

https://aclanthology.org/2024.naacl-long.383.bib

https://aclanthology.org/2024.naacl-long.383/

@inproceedings{burdisso-etal-2024-reliability, title = "Reliability Estimation of News Media Sources: Birds of a Feather Flock Together", author = "Burdisso, Sergio and Sanchez-cortes, Dairazalia and Villatoro-tello, Esa{\'u} and Motlicek, Petr", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.383", doi = "10.18653/v1/2024.naacl-long.383", pages = "6900--6918", abstract = "Evaluating the reliability of news sources is a routine task for journalists and organizations committed to acquiring and disseminating accurate information.Recent research has shown that predicting sources{'} reliability represents an important first-prior step in addressing additional challenges such as fake news detection and fact-checking.In this paper, we introduce a novel approach for source reliability estimation that leverages reinforcement learning strategies for estimating the reliability degree of news sources. Contrary to previous research, our proposed approach models the problem as the estimation of a reliability degree, and not a reliability label, based on how all the news media sources interact with each other on the Web.We validated the effectiveness of our method on a news media reliability dataset that is an order of magnitude larger than comparable existing datasets. Results show that the estimated reliability degrees strongly correlates with journalists-provided scores (Spearman=0.80) and can effectively predict reliability labels (macro-avg. F1 score=81.05).We release our implementation and dataset, aiming to provide a valuable resource for the NLP community working on information verification.", }

Evaluating the reliability of news sources is a routine task for journalists and organizations committed to acquiring and disseminating accurate information.Recent research has shown that predicting sources{'} reliability represents an important first-prior step in addressing additional challenges such as fake news detection and fact-checking.In this paper, we introduce a novel approach for source reliability estimation that leverages reinforcement learning strategies for estimating the reliability degree of news sources. Contrary to previous research, our proposed approach models the problem as the estimation of a reliability degree, and not a reliability label, based on how all the news media sources interact with each other on the Web.We validated the effectiveness of our method on a news media reliability dataset that is an order of magnitude larger than comparable existing datasets. Results show that the estimated reliability degrees strongly correlates with journalists-provided scores (Spearman=0.80) and can effectively predict reliability labels (macro-avg. F1 score=81.05).We release our implementation and dataset, aiming to provide a valuable resource for the NLP community working on information verification.

Reliability Estimation of News Media Sources: Birds of a Feather Flock Together

naacl-long.383

Poster

2404.09565

https://aclanthology.org/2024.naacl-long.384.bib

https://aclanthology.org/2024.naacl-long.384/

@inproceedings{kojima-etal-2024-multilingual, title = "On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons", author = "Kojima, Takeshi and Okimura, Itsuki and Iwasawa, Yusuke and Yanaka, Hitomi and Matsuo, Yutaka", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.384", doi = "10.18653/v1/2024.naacl-long.384", pages = "6919--6971", abstract = "Current decoder-based pre-trained language models (PLMs) successfully demonstrate multilingual capabilities. However, it is unclear how these models handle multilingualism.We analyze the neuron-level internal behavior of multilingual decoder-based PLMs, Specifically examining the existence of neurons that fire {``}uniquely for each language{''} within decoder-only multilingual PLMs.We analyze six languages: English, German, French, Spanish, Chinese, and Japanese, and show that language-specific neurons are unique, with a slight overlap ({\textless} 5{\%}) between languages. These neurons are mainly distributed in the models{'} first and last few layers. This trend remains consistent across languages and models.Additionally, we tamper with less than 1{\%} of the total neurons in each model during inference and demonstrate that tampering with a few language-specific neurons drastically changes the probability of target language occurrence in text generation.", }

Current decoder-based pre-trained language models (PLMs) successfully demonstrate multilingual capabilities. However, it is unclear how these models handle multilingualism.We analyze the neuron-level internal behavior of multilingual decoder-based PLMs, Specifically examining the existence of neurons that fire {``}uniquely for each language{''} within decoder-only multilingual PLMs.We analyze six languages: English, German, French, Spanish, Chinese, and Japanese, and show that language-specific neurons are unique, with a slight overlap ({\textless} 5{\%}) between languages. These neurons are mainly distributed in the models{'} first and last few layers. This trend remains consistent across languages and models.Additionally, we tamper with less than 1{\%} of the total neurons in each model during inference and demonstrate that tampering with a few language-specific neurons drastically changes the probability of target language occurrence in text generation.

On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons

naacl-long.384

Poster

2404.02431

https://aclanthology.org/2024.naacl-long.385.bib

https://aclanthology.org/2024.naacl-long.385/

@inproceedings{tonja-etal-2024-nlp, title = "{NLP} Progress in Indigenous {L}atin {A}merican Languages", author = "Tonja, Atnafu and Balouchzahi, Fazlourrahman and Butt, Sabur and Kolesnikova, Olga and Ceballos, Hector and Gelbukh, Alexander and Solorio, Thamar", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.385", doi = "10.18653/v1/2024.naacl-long.385", pages = "6972--6987", abstract = "The paper focuses on the marginalization of indigenous language communities in the face of rapid technological advancements. We highlight the cultural richness of these languages and the risk they face of being overlooked in the realm of Natural Language Processing (NLP). We aim to bridge the gap between these communities and researchers, emphasizing the need for inclusive technological advancements that respect indigenous community perspectives. We show the NLP progress of indigenous Latin American languages and the survey that covers the status of indigenous languages in Latin America, their representation in NLP, and the challenges and innovations required for their preservation and development. The paper contributes to the current literature in understanding the need and progress of NLP for indigenous communities of Latin America, specifically low-resource and indigenous communities in general.", }

The paper focuses on the marginalization of indigenous language communities in the face of rapid technological advancements. We highlight the cultural richness of these languages and the risk they face of being overlooked in the realm of Natural Language Processing (NLP). We aim to bridge the gap between these communities and researchers, emphasizing the need for inclusive technological advancements that respect indigenous community perspectives. We show the NLP progress of indigenous Latin American languages and the survey that covers the status of indigenous languages in Latin America, their representation in NLP, and the challenges and innovations required for their preservation and development. The paper contributes to the current literature in understanding the need and progress of NLP for indigenous communities of Latin America, specifically low-resource and indigenous communities in general.

NLP Progress in Indigenous Latin American Languages

naacl-long.385

Oral

2404.05365

https://aclanthology.org/2024.naacl-long.386.bib

https://aclanthology.org/2024.naacl-long.386/

@inproceedings{chung-bright-2024-effectiveness, title = "On the Effectiveness of Adversarial Robustness for Abuse Mitigation with Counterspeech", author = "Chung, Yi-Ling and Bright, Jonathan", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.386", doi = "10.18653/v1/2024.naacl-long.386", pages = "6988--7002", abstract = "Recent work on automated approaches to counterspeech have mostly focused on synthetic data but seldom look into how the public deals with abuse. While these systems identifying and generating counterspeech have the potential for abuse mitigation, it remains unclear how robust a model is against adversarial attacks across multiple domains and how models trained on synthetic data can handle unseen user-generated abusive content in the real world. To tackle these issues, this paper first explores the dynamics of abuse and replies using our novel dataset of 6,955 labelled tweets targeted at footballers for studying public figure abuse. We then curate DynaCounter, a new English dataset of 1,911 pairs of abuse and replies addressing nine minority identity groups, collected in an adversarial human-in-the-loop process over four rounds. Our analysis shows that adversarial attacks do not necessarily result in better generalisation. We further present a study of multi-domain counterspeech generation, comparing Flan-T5 and T5 models. We observe that handling certain abuse targets is particularly challenging.", }

Recent work on automated approaches to counterspeech have mostly focused on synthetic data but seldom look into how the public deals with abuse. While these systems identifying and generating counterspeech have the potential for abuse mitigation, it remains unclear how robust a model is against adversarial attacks across multiple domains and how models trained on synthetic data can handle unseen user-generated abusive content in the real world. To tackle these issues, this paper first explores the dynamics of abuse and replies using our novel dataset of 6,955 labelled tweets targeted at footballers for studying public figure abuse. We then curate DynaCounter, a new English dataset of 1,911 pairs of abuse and replies addressing nine minority identity groups, collected in an adversarial human-in-the-loop process over four rounds. Our analysis shows that adversarial attacks do not necessarily result in better generalisation. We further present a study of multi-domain counterspeech generation, comparing Flan-T5 and T5 models. We observe that handling certain abuse targets is particularly challenging.

On the Effectiveness of Adversarial Robustness for Abuse Mitigation with Counterspeech

naacl-long.386

Poster

https://aclanthology.org/2024.naacl-long.387.bib

https://aclanthology.org/2024.naacl-long.387/

@inproceedings{gonzalez-gutierrez-quattoni-2024-leveraging, title = "Leveraging the Structure of Pre-trained Embeddings to Minimize Annotation Effort", author = "Gonzalez-Gutierrez, Cesar and Quattoni, Ariadna", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.387", doi = "10.18653/v1/2024.naacl-long.387", pages = "7003--7017", abstract = "Most current state-of-the-art approaches for text classification are based on fine-tuning the representations computed by large language models (LLMs). This strategy has led to significant improvements in classification performance and contributed to a reduction of the amount of labeled data required for training a model. However, for some challenging classification tasks, providing enough annotations to ensure a reliable classification continues to be the main bottleneck. This is especially true in settings of highly imbalanced class distributions. This paper proposes to tackle this bottleneck by exploiting the structural properties of pre-trained embeddings. We develop a label propagation method that uses pre-trained embeddings to spread information from the labeled samples to nearby samples in the induced space, ensuring the optimal use of annotations. Our approach is simple and relatively low-cost since it only requires computing some distances in the embedded space. We conduct experiments on different text classification datasets showing that the proposed method is efficient and significantly outperforms both self-training and random walk label propagation strategies.", }

Most current state-of-the-art approaches for text classification are based on fine-tuning the representations computed by large language models (LLMs). This strategy has led to significant improvements in classification performance and contributed to a reduction of the amount of labeled data required for training a model. However, for some challenging classification tasks, providing enough annotations to ensure a reliable classification continues to be the main bottleneck. This is especially true in settings of highly imbalanced class distributions. This paper proposes to tackle this bottleneck by exploiting the structural properties of pre-trained embeddings. We develop a label propagation method that uses pre-trained embeddings to spread information from the labeled samples to nearby samples in the induced space, ensuring the optimal use of annotations. Our approach is simple and relatively low-cost since it only requires computing some distances in the embedded space. We conduct experiments on different text classification datasets showing that the proposed method is efficient and significantly outperforms both self-training and random walk label propagation strategies.

Leveraging the Structure of Pre-trained Embeddings to Minimize Annotation Effort

naacl-long.387

Oral

https://aclanthology.org/2024.naacl-long.388.bib

https://aclanthology.org/2024.naacl-long.388/

@inproceedings{yang-etal-2024-uniark, title = "{U}ni{A}rk: Improving Generalisation and Consistency for Factual Knowledge Extraction through Debiasing", author = "Yang, Yijun and He, Jie and Chen, Pinzhen and Gutierrez Basulto, Victor and Pan, Jeff", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.388", doi = "10.18653/v1/2024.naacl-long.388", pages = "7018--7035", abstract = "Several recent papers have investigated the potential of language models as knowledge bases as well as the existence of severe biases when extracting factual knowledge. In this work, we focus on the factual probing performance over unseen prompts from tuning, and using a probabilistic view we show the inherent misalignment between pre-training and downstream tuning objectives in language models for probing knowledge. We hypothesize that simultaneously debiasing these objectives can be the key to generalisation over unseen prompts. We propose an adapter-based framework, **UniArk**, for generalised and consistent factual knowledge extraction through simple methods without introducing extra parameters. Extensive experiments show that UniArk can significantly improve the model{'}s out-of-domain generalisation as well as consistency under various prompts. Additionally, we construct **ParaTrex**, a large-scale and diverse dataset for measuring the inconsistency and out-of-domain generation of models. Further, ParaTrex offers a reference method for constructing paraphrased datasets using large language models.", }

Several recent papers have investigated the potential of language models as knowledge bases as well as the existence of severe biases when extracting factual knowledge. In this work, we focus on the factual probing performance over unseen prompts from tuning, and using a probabilistic view we show the inherent misalignment between pre-training and downstream tuning objectives in language models for probing knowledge. We hypothesize that simultaneously debiasing these objectives can be the key to generalisation over unseen prompts. We propose an adapter-based framework, **UniArk**, for generalised and consistent factual knowledge extraction through simple methods without introducing extra parameters. Extensive experiments show that UniArk can significantly improve the model{'}s out-of-domain generalisation as well as consistency under various prompts. Additionally, we construct **ParaTrex**, a large-scale and diverse dataset for measuring the inconsistency and out-of-domain generation of models. Further, ParaTrex offers a reference method for constructing paraphrased datasets using large language models.

UniArk: Improving Generalisation and Consistency for Factual Knowledge Extraction through Debiasing

naacl-long.388

Poster

2404.01253

https://huggingface.co/papers/2404.01253

https://aclanthology.org/2024.naacl-long.389.bib

https://aclanthology.org/2024.naacl-long.389/

@inproceedings{jeong-etal-2024-adaptive, title = "Adaptive-{RAG}: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity", author = "Jeong, Soyeong and Baek, Jinheon and Cho, Sukmin and Hwang, Sung Ju and Park, Jong", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.389", doi = "10.18653/v1/2024.naacl-long.389", pages = "7036--7050", abstract = "Retrieval-Augmented Large Language Models (LLMs), which incorporate the non-parametric knowledge from external knowledge bases into LLMs, have emerged as a promising approach to enhancing response accuracy in several tasks, such as Question-Answering (QA). However, even though there are various approaches dealing with queries of different complexities, they either handle simple queries with unnecessary computational overhead or fail to adequately address complex multi-step queries; yet, not all user requests fall into only one of the simple or complex categories. In this work, we propose a novel adaptive QA framework that can dynamically select the most suitable strategy for (retrieval-augmented) LLMs from the simplest to the most sophisticated ones based on the query complexity. Also, this selection process is operationalized with a classifier, which is a smaller LM trained to predict the complexity level of incoming queries with automatically collected labels, obtained from actual predicted outcomes of models and inherent inductive biases in datasets. This approach offers a balanced strategy, seamlessly adapting between the iterative and single-step retrieval-augmented LLMs, as well as the no-retrieval methods, in response to a range of query complexities. We validate our model on a set of open-domain QA datasets, covering multiple query complexities, and show that ours enhances the overall efficiency and accuracy of QA systems, compared to relevant baselines including the adaptive retrieval approaches. Code is available at: https://github.com/starsuzi/Adaptive-RAG.", }

Retrieval-Augmented Large Language Models (LLMs), which incorporate the non-parametric knowledge from external knowledge bases into LLMs, have emerged as a promising approach to enhancing response accuracy in several tasks, such as Question-Answering (QA). However, even though there are various approaches dealing with queries of different complexities, they either handle simple queries with unnecessary computational overhead or fail to adequately address complex multi-step queries; yet, not all user requests fall into only one of the simple or complex categories. In this work, we propose a novel adaptive QA framework that can dynamically select the most suitable strategy for (retrieval-augmented) LLMs from the simplest to the most sophisticated ones based on the query complexity. Also, this selection process is operationalized with a classifier, which is a smaller LM trained to predict the complexity level of incoming queries with automatically collected labels, obtained from actual predicted outcomes of models and inherent inductive biases in datasets. This approach offers a balanced strategy, seamlessly adapting between the iterative and single-step retrieval-augmented LLMs, as well as the no-retrieval methods, in response to a range of query complexities. We validate our model on a set of open-domain QA datasets, covering multiple query complexities, and show that ours enhances the overall efficiency and accuracy of QA systems, compared to relevant baselines including the adaptive retrieval approaches. Code is available at: https://github.com/starsuzi/Adaptive-RAG.

Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity

naacl-long.389

Poster

2403.14403

https://aclanthology.org/2024.naacl-long.390.bib

https://aclanthology.org/2024.naacl-long.390/

@inproceedings{zhao-etal-2024-knowing, title = "Knowing What {LLM}s {DO} {NOT} Know: A Simple Yet Effective Self-Detection Method", author = "Zhao, Yukun and Yan, Lingyong and Sun, Weiwei and Xing, Guoliang and Meng, Chong and Wang, Shuaiqiang and Cheng, Zhicong and Ren, Zhaochun and Yin, Dawei", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.390", doi = "10.18653/v1/2024.naacl-long.390", pages = "7051--7063", abstract = "Large Language Models (LLMs) have shown great potential in Natural Language Processing (NLP) tasks.However, recent literature reveals that LLMs hallucinate intermittently, which impedes their reliability for further utilization. In this paper, we propose a novel self-detection method to detect which questions an LLM does not know.Our proposal is empirical and applicable for continually upgrading LLMs compared with state-of-the-art methods. Specifically, we examine the divergence of the LLM{'}s behaviors on different verbalizations for a question and examine the atypicality of the verbalized input. We combine the two components to identify whether the model generates a non-factual response to the question. The above components can be accomplished by utilizing the LLM itself without referring to any other external resources. We conduct comprehensive experiments and demonstrate the effectiveness of our method for recently released LLMs involving Llama 2, Vicuna, ChatGPT, and GPT-4 across factoid question-answering, arithmetic reasoning, and commonsense reasoning tasks.", }

Large Language Models (LLMs) have shown great potential in Natural Language Processing (NLP) tasks.However, recent literature reveals that LLMs hallucinate intermittently, which impedes their reliability for further utilization. In this paper, we propose a novel self-detection method to detect which questions an LLM does not know.Our proposal is empirical and applicable for continually upgrading LLMs compared with state-of-the-art methods. Specifically, we examine the divergence of the LLM{'}s behaviors on different verbalizations for a question and examine the atypicality of the verbalized input. We combine the two components to identify whether the model generates a non-factual response to the question. The above components can be accomplished by utilizing the LLM itself without referring to any other external resources. We conduct comprehensive experiments and demonstrate the effectiveness of our method for recently released LLMs involving Llama 2, Vicuna, ChatGPT, and GPT-4 across factoid question-answering, arithmetic reasoning, and commonsense reasoning tasks.

Knowing What LLMs DO NOT Know: A Simple Yet Effective Self-Detection Method

naacl-long.390

Poster

2310.17918

https://huggingface.co/papers/2310.17918

https://aclanthology.org/2024.naacl-long.391.bib

https://aclanthology.org/2024.naacl-long.391/

@inproceedings{qiu-etal-2024-large, title = "Are Large Language Model Temporally Grounded?", author = "Qiu, Yifu and Zhao, Zheng and Ziser, Yftah and Korhonen, Anna and Ponti, Edoardo and Cohen, Shay", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.391", doi = "10.18653/v1/2024.naacl-long.391", pages = "7064--7083", abstract = "Are Large Language Models (LLMs) temporally grounded? Since LLMs cannot perceive and interact with the environment, it is impossible to answer this question directly. Instead, we provide LLMs with textual narratives and probe them with respect to their common-sense knowledge of the structure and duration of events, their ability to order events along a timeline, and self-consistency within their temporal model (e.g., temporal relations such as after and before are mutually exclusive for any pair of events). We evaluate state-of-the-art LLMs (such as LLaMA 2 and GPT-4) on three tasks reflecting these abilities. Generally, we find that LLMs lag significantly behind both human performance as well as small-scale, specialised LMs. In-context learning, instruction tuning, and chain-of-thought prompting reduce this gap only to a limited degree. Crucially, LLMs struggle the most with self-consistency, displaying incoherent behaviour in at least 27.23{\%} of their predictions. Contrary to expectations, we also find that scaling the model size does not guarantee positive gains in performance. To explain these results, we study the sources from which LLMs may gather temporal information: we find that sentence ordering in unlabelled texts, available during pre-training, is only weakly correlated with event ordering. Moreover, public instruction tuning mixtures contain few temporal tasks. Hence, we conclude that current LLMs lack a consistent temporal model of textual narratives.", }

Are Large Language Models (LLMs) temporally grounded? Since LLMs cannot perceive and interact with the environment, it is impossible to answer this question directly. Instead, we provide LLMs with textual narratives and probe them with respect to their common-sense knowledge of the structure and duration of events, their ability to order events along a timeline, and self-consistency within their temporal model (e.g., temporal relations such as after and before are mutually exclusive for any pair of events). We evaluate state-of-the-art LLMs (such as LLaMA 2 and GPT-4) on three tasks reflecting these abilities. Generally, we find that LLMs lag significantly behind both human performance as well as small-scale, specialised LMs. In-context learning, instruction tuning, and chain-of-thought prompting reduce this gap only to a limited degree. Crucially, LLMs struggle the most with self-consistency, displaying incoherent behaviour in at least 27.23{\%} of their predictions. Contrary to expectations, we also find that scaling the model size does not guarantee positive gains in performance. To explain these results, we study the sources from which LLMs may gather temporal information: we find that sentence ordering in unlabelled texts, available during pre-training, is only weakly correlated with event ordering. Moreover, public instruction tuning mixtures contain few temporal tasks. Hence, we conclude that current LLMs lack a consistent temporal model of textual narratives.

Are Large Language Model Temporally Grounded?

naacl-long.391

Poster

https://aclanthology.org/2024.naacl-long.392.bib

https://aclanthology.org/2024.naacl-long.392/

@inproceedings{liang-etal-2024-document, title = "Document Image Machine Translation with Dynamic Multi-pre-trained Models Assembling", author = "Liang, Yupu and Zhang, Yaping and Ma, Cong and Zhang, Zhiyang and Zhao, Yang and Xiang, Lu and Zong, Chengqing and Zhou, Yu", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.392", doi = "10.18653/v1/2024.naacl-long.392", pages = "7084--7095", abstract = "Text image machine translation (TIMT) is a task that translates source texts embedded in the image to target translations. The existing TIMT task mainly focuses on text-line-level images. In this paper, we extend the current TIMT task and propose a novel task, **D**ocument **I**mage **M**achine **T**ranslation to **Markdown** (**DIMT2Markdown**), which aims to translate a source document image with long context and complex layout structure to markdown-formatted target translation.We also introduce a novel framework, **D**ocument **I**mage **M**achine **T**ranslation with **D**ynamic multi-pre-trained models **A**ssembling (**DIMTDA**).A dynamic model assembler is used to integrate multiple pre-trained models to enhance the model{'}s understanding of layout and translation capabilities.Moreover, we build a novel large-scale **Do**cument image machine **T**ranslation dataset of **A**rXiv articles in markdown format (**DoTA**), containing 126K image-translation pairs.Extensive experiments demonstrate the feasibility of end-to-end translation of rich-text document images and the effectiveness of DIMTDA.", }

Text image machine translation (TIMT) is a task that translates source texts embedded in the image to target translations. The existing TIMT task mainly focuses on text-line-level images. In this paper, we extend the current TIMT task and propose a novel task, **D**ocument **I**mage **M**achine **T**ranslation to **Markdown** (**DIMT2Markdown**), which aims to translate a source document image with long context and complex layout structure to markdown-formatted target translation.We also introduce a novel framework, **D**ocument **I**mage **M**achine **T**ranslation with **D**ynamic multi-pre-trained models **A**ssembling (**DIMTDA**).A dynamic model assembler is used to integrate multiple pre-trained models to enhance the model{'}s understanding of layout and translation capabilities.Moreover, we build a novel large-scale **Do**cument image machine **T**ranslation dataset of **A**rXiv articles in markdown format (**DoTA**), containing 126K image-translation pairs.Extensive experiments demonstrate the feasibility of end-to-end translation of rich-text document images and the effectiveness of DIMTDA.

Document Image Machine Translation with Dynamic Multi-pre-trained Models Assembling

naacl-long.392

Poster

https://aclanthology.org/2024.naacl-long.393.bib

https://aclanthology.org/2024.naacl-long.393/

@inproceedings{daheim-etal-2024-elastic, title = "Elastic Weight Removal for Faithful and Abstractive Dialogue Generation", author = "Daheim, Nico and Dziri, Nouha and Sachan, Mrinmaya and Gurevych, Iryna and Ponti, Edoardo", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.393", doi = "10.18653/v1/2024.naacl-long.393", pages = "7096--7112", abstract = "Generating factual responses is a crucial requirement for dialogue systems. To promotemore factual responses, a common strategyis to ground their responses in relevant documents that inform response generation. However, common dialogue models still often hallucinate information that was not containedin these documents and is therefore unfaithful. In this work, we propose to alleviate suchhallucinations by {`}subtracting{'} the parametersof a model trained to hallucinate from a dialogue response generation model in order to{`}negate{'} the contribution of such hallucinatedexamples from it. Extensive automatic and human evaluation shows favourable results whencompared to state-of-the-art methods that combine the distributions of multiple models, suchas DExperts (Liu et al., 2021), and others thatchange the training procedure, such as Quark(Lu et al., 2022a). Finally, we show how wecan not only reduce hallucinations but also discourage extractive responses, which are oftena consequence of reducing hallucinations byencouraging copy-pasting of document spans.We publicly release our code for reproducibilityand facilitating further research.", }

Generating factual responses is a crucial requirement for dialogue systems. To promotemore factual responses, a common strategyis to ground their responses in relevant documents that inform response generation. However, common dialogue models still often hallucinate information that was not containedin these documents and is therefore unfaithful. In this work, we propose to alleviate suchhallucinations by {`}subtracting{'} the parametersof a model trained to hallucinate from a dialogue response generation model in order to{`}negate{'} the contribution of such hallucinatedexamples from it. Extensive automatic and human evaluation shows favourable results whencompared to state-of-the-art methods that combine the distributions of multiple models, suchas DExperts (Liu et al., 2021), and others thatchange the training procedure, such as Quark(Lu et al., 2022a). Finally, we show how wecan not only reduce hallucinations but also discourage extractive responses, which are oftena consequence of reducing hallucinations byencouraging copy-pasting of document spans.We publicly release our code for reproducibilityand facilitating further research.

Elastic Weight Removal for Faithful and Abstractive Dialogue Generation

naacl-long.393

Poster

2303.17574

https://aclanthology.org/2024.naacl-long.394.bib

https://aclanthology.org/2024.naacl-long.394/

@inproceedings{zhang-etal-2024-r, title = "{R}-Tuning: Instructing Large Language Models to Say {`}{I} Don{'}t Know{'}", author = "Zhang, Hanning and Diao, Shizhe and Lin, Yong and Fung, Yi and Lian, Qing and Wang, Xingyao and Chen, Yangyi and Ji, Heng and Zhang, Tong", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.394", doi = "10.18653/v1/2024.naacl-long.394", pages = "7113--7139", abstract = "Large language models (LLMs) have revolutionized numerous domains with their impressive performance but still face their challenges. A predominant issue is the propensity for these models to generate non-existent facts, a concern termed hallucination. Our research is motivated by the observation that previous instruction tuning methods force the model to complete a sentence no matter whether the model knows the knowledge or not. When the question is out of the parametric knowledge, it will try to make up something and fail to indicate when it lacks knowledge. In this paper, we present a new approach called Refusal-Aware Instruction Tuning (R-Tuning). This approach is formalized by first identifying the disparity in knowledge encompassed by pre-trained parameters compared to that of instruction tuning data. Then, we construct the refusal-aware data based on the knowledge intersection, to tune LLMs to refrain from responding to questions beyond its parametric knowledge. Experimental results demonstrate R-Tuning effectively improves a model{'}s ability to answer known questions and refrain from answering unknown questions. Furthermore, when tested on out-of-domain datasets, the refusal ability was found to be a meta-skill that could be generalized to other tasks. Further analysis surprisingly finds that learning the uncertainty results in better calibration and an improved ability to estimate the uncertainty than uncertainty-based testing. Our code is available at https://github.com/shizhediao/R-Tuning", }

Large language models (LLMs) have revolutionized numerous domains with their impressive performance but still face their challenges. A predominant issue is the propensity for these models to generate non-existent facts, a concern termed hallucination. Our research is motivated by the observation that previous instruction tuning methods force the model to complete a sentence no matter whether the model knows the knowledge or not. When the question is out of the parametric knowledge, it will try to make up something and fail to indicate when it lacks knowledge. In this paper, we present a new approach called Refusal-Aware Instruction Tuning (R-Tuning). This approach is formalized by first identifying the disparity in knowledge encompassed by pre-trained parameters compared to that of instruction tuning data. Then, we construct the refusal-aware data based on the knowledge intersection, to tune LLMs to refrain from responding to questions beyond its parametric knowledge. Experimental results demonstrate R-Tuning effectively improves a model{'}s ability to answer known questions and refrain from answering unknown questions. Furthermore, when tested on out-of-domain datasets, the refusal ability was found to be a meta-skill that could be generalized to other tasks. Further analysis surprisingly finds that learning the uncertainty results in better calibration and an improved ability to estimate the uncertainty than uncertainty-based testing. Our code is available at https://github.com/shizhediao/R-Tuning

R-Tuning: Instructing Large Language Models to Say `I Don't Know'

naacl-long.394

Oral

2311.09677

https://aclanthology.org/2024.naacl-long.395.bib

https://aclanthology.org/2024.naacl-long.395/

@inproceedings{xu-etal-2024-bridging, title = "Bridging the Gap between Different Vocabularies for {LLM} Ensemble", author = "Xu, Yangyifan and Lu, Jinliang and Zhang, Jiajun", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.395", doi = "10.18653/v1/2024.naacl-long.395", pages = "7140--7152", abstract = "Ensembling different large language models (LLMs) to unleash their complementary potential and harness their individual strengths is highly valuable. Nevertheless, vocabulary discrepancies among various LLMs have constrained previous studies to either selecting or blending completely generated outputs. This limitation hinders the dynamic correction and enhancement of outputs during the generation process, resulting in a limited capacity for effective ensemble. To address this issue, we propose a novel method to $\textbf{E}$nsemble LLMs via $\textbf{V}$ocabulary $\textbf{A}$lignment (EVA). EVA bridges the lexical gap among various LLMs, enabling meticulous ensemble at each generation step. Specifically, we first learn mappings between the vocabularies of different LLMs with the assistance of overlapping tokens. Subsequently, these mappings are employed to project output distributions of LLMs into a unified space, facilitating a fine-grained ensemble. Finally, we design a filtering strategy to exclude models that generate unfaithful tokens. Experimental results on commonsense reasoning, arithmetic reasoning, machine translation, and data-to-text generation tasks demonstrate the superiority of our approach compared with individual LLMs and previous ensemble methods conducted on complete outputs. Further analyses confirm that our approach can leverage knowledge from different language models and yield consistent improvement.", }

Ensembling different large language models (LLMs) to unleash their complementary potential and harness their individual strengths is highly valuable. Nevertheless, vocabulary discrepancies among various LLMs have constrained previous studies to either selecting or blending completely generated outputs. This limitation hinders the dynamic correction and enhancement of outputs during the generation process, resulting in a limited capacity for effective ensemble. To address this issue, we propose a novel method to $\textbf{E}$nsemble LLMs via $\textbf{V}$ocabulary $\textbf{A}$lignment (EVA). EVA bridges the lexical gap among various LLMs, enabling meticulous ensemble at each generation step. Specifically, we first learn mappings between the vocabularies of different LLMs with the assistance of overlapping tokens. Subsequently, these mappings are employed to project output distributions of LLMs into a unified space, facilitating a fine-grained ensemble. Finally, we design a filtering strategy to exclude models that generate unfaithful tokens. Experimental results on commonsense reasoning, arithmetic reasoning, machine translation, and data-to-text generation tasks demonstrate the superiority of our approach compared with individual LLMs and previous ensemble methods conducted on complete outputs. Further analyses confirm that our approach can leverage knowledge from different language models and yield consistent improvement.

Bridging the Gap between Different Vocabularies for LLM Ensemble

naacl-long.395

Poster

2404.09492

https://aclanthology.org/2024.naacl-long.396.bib

https://aclanthology.org/2024.naacl-long.396/

@inproceedings{luo-etal-2024-knowla, title = "{K}now{LA}: Enhancing Parameter-efficient Finetuning with Knowledgeable Adaptation", author = "Luo, Xindi and Sun, Zequn and Zhao, Jing and Zhao, Zhe and Hu, Wei", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.396", doi = "10.18653/v1/2024.naacl-long.396", pages = "7153--7166", abstract = "Parameter-efficient finetuning (PEFT) is a key technique for adapting large language models (LLMs) to downstream tasks. In this paper, we study leveraging knowledge graph embeddings to improve the effectiveness of PEFT. We propose a knowledgeable adaptation method called KnowLA. It inserts an adaptation layer into an LLM to integrate the embeddings of entities appearing in the input text. The adaptation layer is trained in combination with LoRA on instruction data. Experiments on six benchmarks with two popular LLMs and three knowledge graphs demonstrate the effectiveness and robustness of KnowLA. We show that KnowLA can help activate the relevant parameterized knowledge in an LLM to answer a question without changing its parameters or input prompts.", }

Parameter-efficient finetuning (PEFT) is a key technique for adapting large language models (LLMs) to downstream tasks. In this paper, we study leveraging knowledge graph embeddings to improve the effectiveness of PEFT. We propose a knowledgeable adaptation method called KnowLA. It inserts an adaptation layer into an LLM to integrate the embeddings of entities appearing in the input text. The adaptation layer is trained in combination with LoRA on instruction data. Experiments on six benchmarks with two popular LLMs and three knowledge graphs demonstrate the effectiveness and robustness of KnowLA. We show that KnowLA can help activate the relevant parameterized knowledge in an LLM to answer a question without changing its parameters or input prompts.

KnowLA: Enhancing Parameter-efficient Finetuning with Knowledgeable Adaptation

naacl-long.396

Poster

2403.14950

https://aclanthology.org/2024.naacl-long.397.bib

https://aclanthology.org/2024.naacl-long.397/

@inproceedings{xiao-2024-extremely, title = "Extremely Weakly-supervised Text Classification with Wordsets Mining and Sync-Denoising", author = "Xiao, Lysa", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.397", doi = "10.18653/v1/2024.naacl-long.397", pages = "7167--7179", abstract = "Extremely weakly-supervised text classification aims to classify texts without any labeled data, but only relying on class names as supervision. Existing works include prompt-based and seed-based methods. Prompt-based methods prompt language model with instructions, while seed-based methods generate pseudo-labels with word matching. Both of them have significant flaws, including zero-shot instability and context-dependent ambiguities. This paper introduces SetSync, which follows a new paradigm, i.e. wordset-based, which can avoid the above problems. In SetSync, a class is represented with wordsets, and pseudo-labels are generated with wordsets matching. To facilitate this, we propose to use information bottleneck to identify class-relevant wordsets. Moreover, we regard the classifier training as a hybrid learning of semi-supervised and noisy-labels, and propose a new training strategy, termed sync-denoising. Extensive experiments on 11 datasets show that SetSync outperforms all existing prompt and seed methods, exceeding SOTA by an impressive average of 8 points.", }

Extremely weakly-supervised text classification aims to classify texts without any labeled data, but only relying on class names as supervision. Existing works include prompt-based and seed-based methods. Prompt-based methods prompt language model with instructions, while seed-based methods generate pseudo-labels with word matching. Both of them have significant flaws, including zero-shot instability and context-dependent ambiguities. This paper introduces SetSync, which follows a new paradigm, i.e. wordset-based, which can avoid the above problems. In SetSync, a class is represented with wordsets, and pseudo-labels are generated with wordsets matching. To facilitate this, we propose to use information bottleneck to identify class-relevant wordsets. Moreover, we regard the classifier training as a hybrid learning of semi-supervised and noisy-labels, and propose a new training strategy, termed sync-denoising. Extensive experiments on 11 datasets show that SetSync outperforms all existing prompt and seed methods, exceeding SOTA by an impressive average of 8 points.

Extremely Weakly-supervised Text Classification with Wordsets Mining and Sync-Denoising

naacl-long.397

Poster

https://aclanthology.org/2024.naacl-long.398.bib

https://aclanthology.org/2024.naacl-long.398/

@inproceedings{wu-etal-2024-f, title = "{F}-{MALLOC}: Feed-forward Memory Allocation for Continual Learning in Neural Machine Translation", author = "Wu, Junhong and Liu, Yuchen and Zong, Chengqing", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.398", doi = "10.18653/v1/2024.naacl-long.398", pages = "7180--7192", abstract = "In the evolving landscape of Neural Machine Translation (NMT), the pretrain-then-finetune paradigm has yielded impressive results. However, the persistent challenge of Catastrophic Forgetting (CF) remains a hurdle. While previous work has introduced Continual Learning (CL) methods to address CF, these approaches grapple with the delicate balance between avoiding forgetting and maintaining system extensibility. To address this, we propose a CL method, named $\textbf{F-MALLOC}$ ($\textbf{F}$eed-forward $\textbf{M}$emory $\textbf{ALLOC}$ation). F-MALLOC is inspired by recent insights highlighting that feed-forward layers emulate neural memories and encapsulate crucial translation knowledge. It decomposes feed-forward layers into discrete memory cells and allocates these memories to different tasks. By learning to allocate and safeguard these memories, our method effectively alleviates CF while ensuring robust extendability. Besides, we propose a comprehensive assessment protocol for multi-stage CL of NMT systems. Experiments conducted following this new protocol showcase the superior performance of F-MALLOC, evidenced by higher BLEU scores and almost zero forgetting.", }

In the evolving landscape of Neural Machine Translation (NMT), the pretrain-then-finetune paradigm has yielded impressive results. However, the persistent challenge of Catastrophic Forgetting (CF) remains a hurdle. While previous work has introduced Continual Learning (CL) methods to address CF, these approaches grapple with the delicate balance between avoiding forgetting and maintaining system extensibility. To address this, we propose a CL method, named $\textbf{F-MALLOC}$ ($\textbf{F}$eed-forward $\textbf{M}$emory $\textbf{ALLOC}$ation). F-MALLOC is inspired by recent insights highlighting that feed-forward layers emulate neural memories and encapsulate crucial translation knowledge. It decomposes feed-forward layers into discrete memory cells and allocates these memories to different tasks. By learning to allocate and safeguard these memories, our method effectively alleviates CF while ensuring robust extendability. Besides, we propose a comprehensive assessment protocol for multi-stage CL of NMT systems. Experiments conducted following this new protocol showcase the superior performance of F-MALLOC, evidenced by higher BLEU scores and almost zero forgetting.

F-MALLOC: Feed-forward Memory Allocation for Continual Learning in Neural Machine Translation

naacl-long.398

Poster

2404.04846

https://aclanthology.org/2024.naacl-long.399.bib

https://aclanthology.org/2024.naacl-long.399/

@inproceedings{mcinerney-etal-2024-towards, title = "Towards Reducing Diagnostic Errors with Interpretable Risk Prediction", author = "McInerney, Denis and Dickinson, William and Flynn, Lucy and Young, Andrea and Young, Geoffrey and van de Meent, Jan-Willem and Wallace, Byron", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.399", doi = "10.18653/v1/2024.naacl-long.399", pages = "7193--7210", abstract = "Many diagnostic errors occur because clinicians cannot easily access relevant information in patient Electronic Health Records (EHRs). In this work we propose a method to use LLMs to identify pieces of evidence in patient EHR data that indicate increased or decreased risk of specific diagnoses; our ultimate aim is to increase access to evidence and reduce diagnostic errors. In particular, we propose a Neural Additive Model to make predictions backed by evidence with individualized risk estimates at time-points where clinicians are still uncertain, aiming to specifically mitigate delays in diagnosis and errors stemming from an incomplete differential. To train such a model, it is necessary to infer temporally fine-grained retrospective labels of eventual {``}true{''} diagnoses. We do so with LLMs, to ensure that the input text is from before a confident diagnosis can be made. We use an LLM to retrieve an initial pool of evidence, but then refine this set of evidence according to correlations learned by the model. We conduct an in-depth evaluation of the usefulness of our approach by simulating how it might be used by a clinician to decide between a pre-defined list of differential diagnoses.", }

Many diagnostic errors occur because clinicians cannot easily access relevant information in patient Electronic Health Records (EHRs). In this work we propose a method to use LLMs to identify pieces of evidence in patient EHR data that indicate increased or decreased risk of specific diagnoses; our ultimate aim is to increase access to evidence and reduce diagnostic errors. In particular, we propose a Neural Additive Model to make predictions backed by evidence with individualized risk estimates at time-points where clinicians are still uncertain, aiming to specifically mitigate delays in diagnosis and errors stemming from an incomplete differential. To train such a model, it is necessary to infer temporally fine-grained retrospective labels of eventual {``}true{''} diagnoses. We do so with LLMs, to ensure that the input text is from before a confident diagnosis can be made. We use an LLM to retrieve an initial pool of evidence, but then refine this set of evidence according to correlations learned by the model. We conduct an in-depth evaluation of the usefulness of our approach by simulating how it might be used by a clinician to decide between a pre-defined list of differential diagnoses.

Towards Reducing Diagnostic Errors with Interpretable Risk Prediction

naacl-long.399

Poster

2402.10109

https://aclanthology.org/2024.naacl-long.400.bib

https://aclanthology.org/2024.naacl-long.400/

@inproceedings{singh-thakur-2024-generalizable, title = "Generalizable Multilingual Hate Speech Detection on Low Resource {I}ndian Languages using Fair Selection in Federated Learning", author = "Singh, Akshay and Thakur, Rahul", editor = "Duh, Kevin and Gomez, Helena and Bethard, Steven", booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)", month = jun, year = "2024", address = "Mexico City, Mexico", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2024.naacl-long.400", doi = "10.18653/v1/2024.naacl-long.400", pages = "7211--7221", abstract = "Social media, originally meant for peaceful communication, now faces issues with hate speech. Detecting hate speech from social media in Indian languages with linguistic diversity and cultural nuances presents a complex and challenging task. Furthermore, traditional methods involve sharing of users{'} sensitive data with a server for model training making it undesirable and involving potential risk to their privacy remained under-studied. In this paper, we combined various low-resource language datasets and propose MultiFED, a federated approach that performs effectively to detect hate speech. MultiFED utilizes continuous adaptation and fine-tuning to aid generalization using subsets of multilingual data overcoming the limitations of data scarcity. Extensive experiments are conducted on 13 Indic datasets across five different pre-trained models. The results show that MultiFED outperforms the state-of-the-art baselines by 8{\%} (approx.) in terms of Accuracy and by 12{\%} (approx.) in terms of F-Score.", }

Social media, originally meant for peaceful communication, now faces issues with hate speech. Detecting hate speech from social media in Indian languages with linguistic diversity and cultural nuances presents a complex and challenging task. Furthermore, traditional methods involve sharing of users{'} sensitive data with a server for model training making it undesirable and involving potential risk to their privacy remained under-studied. In this paper, we combined various low-resource language datasets and propose MultiFED, a federated approach that performs effectively to detect hate speech. MultiFED utilizes continuous adaptation and fine-tuning to aid generalization using subsets of multilingual data overcoming the limitations of data scarcity. Extensive experiments are conducted on 13 Indic datasets across five different pre-trained models. The results show that MultiFED outperforms the state-of-the-art baselines by 8{\%} (approx.) in terms of Accuracy and by 12{\%} (approx.) in terms of F-Score.

Generalizable Multilingual Hate Speech Detection on Low Resource Indian Languages using Fair Selection in Federated Learning

naacl-long.400

Poster