Datasets:

ACLMeeting

/

ACL2023-papers

like 0

Follow

Annual Meeting of the Association for Computational Linguistics 1

https://aclanthology.org/2023.acl-long.100.bib

https://aclanthology.org/2023.acl-long.100/

@inproceedings{li-etal-2023-white, title = "White-Box Multi-Objective Adversarial Attack on Dialogue Generation", author = "Li, Yufei and Li, Zexin and Gao, Yingfan and Liu, Cong", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.100", doi = "10.18653/v1/2023.acl-long.100", pages = "1778--1792", abstract = "Pre-trained transformers are popular in state-of-the-art dialogue generation (DG) systems. Such language models are, however, vulnerable to various adversarial samples as studied in traditional tasks such as text classification, which inspires our curiosity about their robustness in DG systems. One main challenge of attacking DG models is that perturbations on the current sentence can hardly degrade the response accuracy because the unchanged chat histories are also considered for decision-making. Instead of merely pursuing pitfalls of performance metrics such as BLEU, ROUGE, we observe that crafting adversarial samples to force longer generation outputs benefits attack effectiveness{---}the generated responses are typically irrelevant, lengthy, and repetitive. To this end, we propose a white-box multi-objective attack method called DGSlow. Specifically, DGSlow balances two objectives{---}generation accuracy and length, via a gradient-based multi-objective optimizer and applies an adaptive searching mechanism to iteratively craft adversarial samples with only a few modifications. Comprehensive experiments on four benchmark datasets demonstrate that DGSlow could significantly degrade state-of-the-art DG models with a higher success rate than traditional accuracy-based methods. Besides, our crafted sentences also exhibit strong transferability in attacking other models.", }

Pre-trained transformers are popular in state-of-the-art dialogue generation (DG) systems. Such language models are, however, vulnerable to various adversarial samples as studied in traditional tasks such as text classification, which inspires our curiosity about their robustness in DG systems. One main challenge of attacking DG models is that perturbations on the current sentence can hardly degrade the response accuracy because the unchanged chat histories are also considered for decision-making. Instead of merely pursuing pitfalls of performance metrics such as BLEU, ROUGE, we observe that crafting adversarial samples to force longer generation outputs benefits attack effectiveness{---}the generated responses are typically irrelevant, lengthy, and repetitive. To this end, we propose a white-box multi-objective attack method called DGSlow. Specifically, DGSlow balances two objectives{---}generation accuracy and length, via a gradient-based multi-objective optimizer and applies an adaptive searching mechanism to iteratively craft adversarial samples with only a few modifications. Comprehensive experiments on four benchmark datasets demonstrate that DGSlow could significantly degrade state-of-the-art DG models with a higher success rate than traditional accuracy-based methods. Besides, our crafted sentences also exhibit strong transferability in attacking other models.

White-Box Multi-Objective Adversarial Attack on Dialogue Generation

acl-long.100

Poster

2305.03655

https://aclanthology.org/2023.acl-long.101.bib

https://aclanthology.org/2023.acl-long.101/

@inproceedings{khalifa-etal-2023-cautious, title = "A Cautious Generalization Goes a Long Way: Learning Morphophonological Rules", author = "Khalifa, Salam and Payne, Sarah and Kodner, Jordan and Broselow, Ellen and Rambow, Owen", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.101", doi = "10.18653/v1/2023.acl-long.101", pages = "1793--1805", abstract = "Explicit linguistic knowledge, encoded by resources such as rule-based morphological analyzers, continues to prove useful in downstream NLP tasks, especially for low-resource languages and dialects. Rules are an important asset in descriptive linguistic grammars. However, creating such resources is usually expensive and non-trivial, especially for spoken varieties with no written standard. In this work, we present a novel approach for automatically learning morphophonological rules of Arabic from a corpus. Motivated by classic cognitive models for rule learning, rules are generalized cautiously. Rules that are memorized for individual items are only allowed to generalize to unseen forms if they are sufficiently reliable in the training data. The learned rules are further examined to ensure that they capture true linguistic phenomena described by domain experts. We also investigate the learnability of rules in low-resource settings across different experimental setups and dialects.", }

Explicit linguistic knowledge, encoded by resources such as rule-based morphological analyzers, continues to prove useful in downstream NLP tasks, especially for low-resource languages and dialects. Rules are an important asset in descriptive linguistic grammars. However, creating such resources is usually expensive and non-trivial, especially for spoken varieties with no written standard. In this work, we present a novel approach for automatically learning morphophonological rules of Arabic from a corpus. Motivated by classic cognitive models for rule learning, rules are generalized cautiously. Rules that are memorized for individual items are only allowed to generalize to unseen forms if they are sufficiently reliable in the training data. The learned rules are further examined to ensure that they capture true linguistic phenomena described by domain experts. We also investigate the learnability of rules in low-resource settings across different experimental setups and dialects.

A Cautious Generalization Goes a Long Way: Learning Morphophonological Rules

acl-long.101

Poster

https://aclanthology.org/2023.acl-long.102.bib

https://aclanthology.org/2023.acl-long.102/

@inproceedings{chan-etal-2023-shot, title = "Few-shot Adaptation Works with {U}npredic{T}able Data", author = "Chan, Jun Shern and Pieler, Michael and Jao, Jonathan and Scheurer, J{\'e}r{\'e}my and Perez, Ethan", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.102", doi = "10.18653/v1/2023.acl-long.102", pages = "1806--1842", abstract = "Prior work on language models (LMs) shows that training on a large number of diverse tasks improves few-shot learning (FSL) performance on new tasks. We take this to the extreme, automatically extracting 413,299 tasks from internet tables - orders of magnitude more than the next-largest public datasets. Finetuning on the resulting dataset leads to improved FSL performance on Natural Language Processing (NLP) tasks, but not proportionally to dataset scale. In fact, we find that narrow subsets of our dataset sometimes outperform more diverse datasets. For example, finetuning on software documentation from support.google.com raises FSL performance by a mean of +7.5{\%} on 52 downstream tasks, which beats training on 40 human-curated NLP datasets (+6.7{\%}). Finetuning on various narrow datasets leads to similar broad improvements across test tasks, suggesting that the gains are not from domain adaptation but adapting to FSL in general. We do not observe clear patterns between the datasets that lead to FSL gains, leaving open questions about why certain data helps with FSL.", }

Prior work on language models (LMs) shows that training on a large number of diverse tasks improves few-shot learning (FSL) performance on new tasks. We take this to the extreme, automatically extracting 413,299 tasks from internet tables - orders of magnitude more than the next-largest public datasets. Finetuning on the resulting dataset leads to improved FSL performance on Natural Language Processing (NLP) tasks, but not proportionally to dataset scale. In fact, we find that narrow subsets of our dataset sometimes outperform more diverse datasets. For example, finetuning on software documentation from support.google.com raises FSL performance by a mean of +7.5{\%} on 52 downstream tasks, which beats training on 40 human-curated NLP datasets (+6.7{\%}). Finetuning on various narrow datasets leads to similar broad improvements across test tasks, suggesting that the gains are not from domain adaptation but adapting to FSL in general. We do not observe clear patterns between the datasets that lead to FSL gains, leaving open questions about why certain data helps with FSL.

Few-shot Adaptation Works with UnpredicTable Data

acl-long.102

Poster

2208.01009

https://aclanthology.org/2023.acl-long.103.bib

https://aclanthology.org/2023.acl-long.103/

@inproceedings{fatima-strube-2023-cross, title = "Cross-lingual Science Journalism: Select, Simplify and Rewrite Summaries for Non-expert Readers", author = "Fatima, Mehwish and Strube, Michael", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.103", doi = "10.18653/v1/2023.acl-long.103", pages = "1843--1861", abstract = "Automating Cross-lingual Science Journalism (CSJ) aims to generate popular science summaries from English scientific texts for non-expert readers in their local language. We introduce CSJ as a downstream task of text simplification and cross-lingual scientific summarization to facilitate science journalists{'} work. We analyze the performance of possible existing solutions as baselines for the CSJ task. Based on these findings, we propose to combine the three components - SELECT, SIMPLIFY and REWRITE (SSR) to produce cross-lingual simplified science summaries for non-expert readers. Our empirical evaluation on the Wikipedia dataset shows that SSR significantly outperforms the baselines for the CSJ task and can serve as a strong baseline for future work. We also perform an ablation study investigating the impact of individual components of SSR. Further, we analyze the performance of SSR on a high-quality, real-world CSJ dataset with human evaluation and in-depth analysis, demonstrating the superior performance of SSR for CSJ.", }

Automating Cross-lingual Science Journalism (CSJ) aims to generate popular science summaries from English scientific texts for non-expert readers in their local language. We introduce CSJ as a downstream task of text simplification and cross-lingual scientific summarization to facilitate science journalists{'} work. We analyze the performance of possible existing solutions as baselines for the CSJ task. Based on these findings, we propose to combine the three components - SELECT, SIMPLIFY and REWRITE (SSR) to produce cross-lingual simplified science summaries for non-expert readers. Our empirical evaluation on the Wikipedia dataset shows that SSR significantly outperforms the baselines for the CSJ task and can serve as a strong baseline for future work. We also perform an ablation study investigating the impact of individual components of SSR. Further, we analyze the performance of SSR on a high-quality, real-world CSJ dataset with human evaluation and in-depth analysis, demonstrating the superior performance of SSR for CSJ.

Cross-lingual Science Journalism: Select, Simplify and Rewrite Summaries for Non-expert Readers

acl-long.103

Poster

https://aclanthology.org/2023.acl-long.104.bib

https://aclanthology.org/2023.acl-long.104/

@inproceedings{elgaar-amiri-2023-hucurl, title = "{H}u{C}url: Human-induced Curriculum Discovery", author = "Elgaar, Mohamed and Amiri, Hadi", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.104", doi = "10.18653/v1/2023.acl-long.104", pages = "1862--1877", abstract = "We introduce the problem of curriculum discovery and describe a curriculum learning framework capable of discovering effective curricula in a curriculum space based on prior knowledge about sample difficulty. Using annotation entropy and loss as measures of difficulty, we show that (i): the top-performing discovered curricula for a given model and dataset are often non-monotonic as apposed to monotonic curricula in existing literature, (ii): the prevailing easy-to-hard or hard-to-easy transition curricula are often at the risk of underperforming, and (iii): the curricula discovered for smaller datasets and models perform well on larger datasets and models respectively. The proposed framework encompasses some of the existing curriculum learning approaches and can discover curricula that outperform them across several NLP tasks.", }

We introduce the problem of curriculum discovery and describe a curriculum learning framework capable of discovering effective curricula in a curriculum space based on prior knowledge about sample difficulty. Using annotation entropy and loss as measures of difficulty, we show that (i): the top-performing discovered curricula for a given model and dataset are often non-monotonic as apposed to monotonic curricula in existing literature, (ii): the prevailing easy-to-hard or hard-to-easy transition curricula are often at the risk of underperforming, and (iii): the curricula discovered for smaller datasets and models perform well on larger datasets and models respectively. The proposed framework encompasses some of the existing curriculum learning approaches and can discover curricula that outperform them across several NLP tasks.

HuCurl: Human-induced Curriculum Discovery

acl-long.104

Poster

2307.07412

https://aclanthology.org/2023.acl-long.105.bib

https://aclanthology.org/2023.acl-long.105/

@inproceedings{liu-etal-2023-knn, title = "k{NN}-{TL}: k-Nearest-Neighbor Transfer Learning for Low-Resource Neural Machine Translation", author = "Liu, Shudong and Liu, Xuebo and Wong, Derek F. and Li, Zhaocong and Jiao, Wenxiang and Chao, Lidia S. and Zhang, Min", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.105", doi = "10.18653/v1/2023.acl-long.105", pages = "1878--1891", abstract = "Transfer learning has been shown to be an effective technique for enhancing the performance of low-resource neural machine translation (NMT). This is typically achieved through either fine-tuning a child model with a pre-trained parent model, or by utilizing the out- put of the parent model during the training of the child model. However, these methods do not make use of the parent knowledge during the child inference, which may limit the translation performance. In this paper, we propose a k-Nearest-Neighbor Transfer Learning (kNN-TL) approach for low-resource NMT, which leverages the parent knowledge throughout the entire developing process of the child model. Our approach includes a parent-child representation alignment method, which ensures consistency in the output representations between the two models, and a child-aware datastore construction method that improves inference efficiency by selectively distilling the parent datastore based on relevance to the child model. Experimental results on four low-resource translation tasks show that kNN-TL outperforms strong baselines. Extensive analyses further demonstrate the effectiveness of our approach. Code and scripts are freely available at \url{https://github.com/NLP2CT/kNN-TL}.", }

Transfer learning has been shown to be an effective technique for enhancing the performance of low-resource neural machine translation (NMT). This is typically achieved through either fine-tuning a child model with a pre-trained parent model, or by utilizing the out- put of the parent model during the training of the child model. However, these methods do not make use of the parent knowledge during the child inference, which may limit the translation performance. In this paper, we propose a k-Nearest-Neighbor Transfer Learning (kNN-TL) approach for low-resource NMT, which leverages the parent knowledge throughout the entire developing process of the child model. Our approach includes a parent-child representation alignment method, which ensures consistency in the output representations between the two models, and a child-aware datastore construction method that improves inference efficiency by selectively distilling the parent datastore based on relevance to the child model. Experimental results on four low-resource translation tasks show that kNN-TL outperforms strong baselines. Extensive analyses further demonstrate the effectiveness of our approach. Code and scripts are freely available at \url{https://github.com/NLP2CT/kNN-TL}.

kNN-TL: k-Nearest-Neighbor Transfer Learning for Low-Resource Neural Machine Translation

acl-long.105

Poster

https://aclanthology.org/2023.acl-long.106.bib

https://aclanthology.org/2023.acl-long.106/

@inproceedings{gu-etal-2023-language, title = "Do language models have coherent mental models of everyday things?", author = "Gu, Yuling and Dalvi Mishra, Bhavana and Clark, Peter", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.106", doi = "10.18653/v1/2023.acl-long.106", pages = "1892--1913", abstract = "When people think of everyday things like an egg, they typically have a mental image associated with it. This allows them to correctly judge, for example, that {``}the yolk surrounds the shell{''} is a false statement. Do language models similarly have a coherent picture of such everyday things? To investigate this, we propose a benchmark dataset consisting of 100 everyday things, their parts, and the relationships between these parts, expressed as 11,720 {``}X relation Y?{''} true/false questions. Using these questions as probes, we observe that state-of-the-art pre-trained language models (LMs) like GPT-3 and Macaw have fragments of knowledge about these everyday things, but do not have fully coherent {``}parts mental models{''} (54-59{\%} accurate, 19-43{\%} conditional constraint violation). We propose an extension where we add a constraint satisfaction layer on top of the LM{'}s raw predictions to apply commonsense constraints. As well as removing inconsistencies, we find that this also significantly improves accuracy (by 16-20{\%}), suggesting how the incoherence of the LM{'}s pictures of everyday things can be significantly reduced.", }

When people think of everyday things like an egg, they typically have a mental image associated with it. This allows them to correctly judge, for example, that {``}the yolk surrounds the shell{''} is a false statement. Do language models similarly have a coherent picture of such everyday things? To investigate this, we propose a benchmark dataset consisting of 100 everyday things, their parts, and the relationships between these parts, expressed as 11,720 {``}X relation Y?{''} true/false questions. Using these questions as probes, we observe that state-of-the-art pre-trained language models (LMs) like GPT-3 and Macaw have fragments of knowledge about these everyday things, but do not have fully coherent {``}parts mental models{''} (54-59{\%} accurate, 19-43{\%} conditional constraint violation). We propose an extension where we add a constraint satisfaction layer on top of the LM{'}s raw predictions to apply commonsense constraints. As well as removing inconsistencies, we find that this also significantly improves accuracy (by 16-20{\%}), suggesting how the incoherence of the LM{'}s pictures of everyday things can be significantly reduced.

Do language models have coherent mental models of everyday things?

acl-long.106

Poster

2212.10029

https://aclanthology.org/2023.acl-long.107.bib

https://aclanthology.org/2023.acl-long.107/

@inproceedings{grusky-2023-rogue, title = "Rogue Scores", author = "Grusky, Max", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.107", doi = "10.18653/v1/2023.acl-long.107", pages = "1914--1934", abstract = "Correct, comparable, and reproducible model evaluation is essential for progress in machine learning. Over twenty years, thousands of language and vision models have been evaluated with a popular metric called ROUGE. Does this widespread benchmark metric meet these three evaluation criteria? This systematic review of over two thousand publications using ROUGE finds: (A) Critical evaluation decisions and parameters are routinely omitted, making most reported scores irreproducible. (B) Differences in evaluation protocol are common, affect scores, and impact the comparability of results reported in many papers. (C) Thousands of papers use nonstandard evaluation packages with software defects that produce provably incorrect scores. Estimating the overall impact of these findings is difficult: because software citations are rare, it is nearly impossible to distinguish between correct ROUGE scores and incorrect {``}rogue scores.{''}", }

Correct, comparable, and reproducible model evaluation is essential for progress in machine learning. Over twenty years, thousands of language and vision models have been evaluated with a popular metric called ROUGE. Does this widespread benchmark metric meet these three evaluation criteria? This systematic review of over two thousand publications using ROUGE finds: (A) Critical evaluation decisions and parameters are routinely omitted, making most reported scores irreproducible. (B) Differences in evaluation protocol are common, affect scores, and impact the comparability of results reported in many papers. (C) Thousands of papers use nonstandard evaluation packages with software defects that produce provably incorrect scores. Estimating the overall impact of these findings is difficult: because software citations are rare, it is nearly impossible to distinguish between correct ROUGE scores and incorrect {``}rogue scores.{''}

Rogue Scores

acl-long.107

Poster

https://aclanthology.org/2023.acl-long.108.bib

https://aclanthology.org/2023.acl-long.108/

@inproceedings{honovich-etal-2023-instruction, title = "Instruction Induction: From Few Examples to Natural Language Task Descriptions", author = "Honovich, Or and Shaham, Uri and Bowman, Samuel R. and Levy, Omer", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.108", doi = "10.18653/v1/2023.acl-long.108", pages = "1935--1952", abstract = "Large language models are able to perform a task by conditioning on a few input-output demonstrations - a paradigm known as in-context learning. We show that language models can explicitly infer an underlying task from a few demonstrations by prompting them to generate a natural language instruction that fits the examples. To explore this ability, we introduce the instruction induction challenge, compile a dataset consisting of 24 tasks, and define a novel evaluation metric based on executing the generated instruction. We discover that, to a large extent, the ability to generate instructions does indeed emerge when using a model that is both large enough and aligned to follow instructions; InstructGPT achieves 65.7{\%} of human performance in our execution-based metric, while the original GPT-3 model reaches only 9.8{\%} of human performance. This surprising result suggests that instruction induction might be a viable learning paradigm in and of itself, where instead of fitting a set of latent continuous parameters to the data, one searches for the best description in the natural language hypothesis space.", }

Large language models are able to perform a task by conditioning on a few input-output demonstrations - a paradigm known as in-context learning. We show that language models can explicitly infer an underlying task from a few demonstrations by prompting them to generate a natural language instruction that fits the examples. To explore this ability, we introduce the instruction induction challenge, compile a dataset consisting of 24 tasks, and define a novel evaluation metric based on executing the generated instruction. We discover that, to a large extent, the ability to generate instructions does indeed emerge when using a model that is both large enough and aligned to follow instructions; InstructGPT achieves 65.7{\%} of human performance in our execution-based metric, while the original GPT-3 model reaches only 9.8{\%} of human performance. This surprising result suggests that instruction induction might be a viable learning paradigm in and of itself, where instead of fitting a set of latent continuous parameters to the data, one searches for the best description in the natural language hypothesis space.

Instruction Induction: From Few Examples to Natural Language Task Descriptions

acl-long.108

Poster

2205.10782

https://huggingface.co/papers/2205.10782

https://aclanthology.org/2023.acl-long.109.bib

https://aclanthology.org/2023.acl-long.109/

@inproceedings{hu-etal-2023-context, title = "In-Context Analogical Reasoning with Pre-Trained Language Models", author = "Hu, Xiaoyang and Storks, Shane and Lewis, Richard and Chai, Joyce", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.109", doi = "10.18653/v1/2023.acl-long.109", pages = "1953--1969", abstract = "Analogical reasoning is a fundamental capacity of human cognition that allows us to reason abstractly about novel situations by relating them to past experiences. While it is thought to be essential for robust reasoning in AI systems, conventional approaches require significant training and/or hard-coding of domain knowledge to be applied to benchmark tasks. Inspired by cognitive science research that has found connections between human language and analogy-making, we explore the use of intuitive language-based abstractions to support analogy in AI systems. Specifically, we apply large pre-trained language models (PLMs) to visual Raven{'}s Progressive Matrices (RPM), a common relational reasoning test. By simply encoding the perceptual features of the problem into language form, we find that PLMs exhibit a striking capacity for zero-shot relational reasoning, exceeding human performance and nearing supervised vision-based methods. We explore different encodings that vary the level of abstraction over task features, finding that higher-level abstractions further strengthen PLMs{'} analogical reasoning. Our detailed analysis reveals insights on the role of model complexity, in-context learning, and prior knowledge in solving RPM tasks.", }

Analogical reasoning is a fundamental capacity of human cognition that allows us to reason abstractly about novel situations by relating them to past experiences. While it is thought to be essential for robust reasoning in AI systems, conventional approaches require significant training and/or hard-coding of domain knowledge to be applied to benchmark tasks. Inspired by cognitive science research that has found connections between human language and analogy-making, we explore the use of intuitive language-based abstractions to support analogy in AI systems. Specifically, we apply large pre-trained language models (PLMs) to visual Raven{'}s Progressive Matrices (RPM), a common relational reasoning test. By simply encoding the perceptual features of the problem into language form, we find that PLMs exhibit a striking capacity for zero-shot relational reasoning, exceeding human performance and nearing supervised vision-based methods. We explore different encodings that vary the level of abstraction over task features, finding that higher-level abstractions further strengthen PLMs{'} analogical reasoning. Our detailed analysis reveals insights on the role of model complexity, in-context learning, and prior knowledge in solving RPM tasks.

In-Context Analogical Reasoning with Pre-Trained Language Models

acl-long.109

Poster

2305.17626

https://aclanthology.org/2023.acl-long.110.bib

https://aclanthology.org/2023.acl-long.110/

@inproceedings{caciularu-etal-2023-peek, title = "Peek Across: Improving Multi-Document Modeling via Cross-Document Question-Answering", author = "Caciularu, Avi and Peters, Matthew and Goldberger, Jacob and Dagan, Ido and Cohan, Arman", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.110", doi = "10.18653/v1/2023.acl-long.110", pages = "1970--1989", abstract = "The integration of multi-document pre-training objectives into language models has resulted in remarkable improvements in multi-document downstream tasks. In this work, we propose extending this idea by pre-training a generic multi-document model from a novel cross-document question answering pre-training objective. To that end, given a set (or cluster) of topically-related documents, we systematically generate semantically-oriented questions from a salient sentence in one document and challenge the model, during pre-training, to answer these questions while {``}peeking{''} into other topically-related documents. In a similar manner, the model is also challenged to recover the sentence from which the question was generated, again while leveraging cross-document information. This novel multi-document QA formulation directs the model to better recover cross-text informational relations, and introduces a natural augmentation that artificially increases the pre-training data. Further, unlike prior multi-document models that focus on either classification or summarization tasks, our pre-training objective formulation enables the model to perform tasks that involve both short text generation (e.g., QA) and long text generation (e.g., summarization).Following this scheme, we pre-train our model - termed QAmden - and evaluate its performance across several multi-document tasks, including multi-document QA, summarization, and query-focused summarization, yielding improvements of up to 7{\%}, and significantly outperforms zero-shot GPT-3.5 and GPT-4.", }

The integration of multi-document pre-training objectives into language models has resulted in remarkable improvements in multi-document downstream tasks. In this work, we propose extending this idea by pre-training a generic multi-document model from a novel cross-document question answering pre-training objective. To that end, given a set (or cluster) of topically-related documents, we systematically generate semantically-oriented questions from a salient sentence in one document and challenge the model, during pre-training, to answer these questions while {``}peeking{''} into other topically-related documents. In a similar manner, the model is also challenged to recover the sentence from which the question was generated, again while leveraging cross-document information. This novel multi-document QA formulation directs the model to better recover cross-text informational relations, and introduces a natural augmentation that artificially increases the pre-training data. Further, unlike prior multi-document models that focus on either classification or summarization tasks, our pre-training objective formulation enables the model to perform tasks that involve both short text generation (e.g., QA) and long text generation (e.g., summarization).Following this scheme, we pre-train our model - termed QAmden - and evaluate its performance across several multi-document tasks, including multi-document QA, summarization, and query-focused summarization, yielding improvements of up to 7{\%}, and significantly outperforms zero-shot GPT-3.5 and GPT-4.

Peek Across: Improving Multi-Document Modeling via Cross-Document Question-Answering

acl-long.110

Poster

2305.15387

https://aclanthology.org/2023.acl-long.111.bib

https://aclanthology.org/2023.acl-long.111/

@inproceedings{ren-etal-2023-tailoring, title = "Tailoring Instructions to Student{'}s Learning Levels Boosts Knowledge Distillation", author = "Ren, Yuxin and Zhong, Zihan and Shi, Xingjian and Zhu, Yi and Yuan, Chun and Li, Mu", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.111", doi = "10.18653/v1/2023.acl-long.111", pages = "1990--2006", abstract = "It has been commonly observed that a teacher model with superior performance does not necessarily result in a stronger student, highlighting a discrepancy between current teacher training practices and effective knowledge transfer. In order to enhance the guidance of the teacher training process, we introduce the concept of distillation influence to determine the impact of distillation from each training sample on the student{'}s generalization ability. In this paper, we propose Learning Good Teacher Matters (LGTM), an efficient training technique for incorporating distillation influence into the teacher{'}s learning process. By prioritizing samples that are likely to enhance the student{'}s generalization ability, our LGTM outperforms 10 common knowledge distillation baselines on 6 text classification tasks in the GLUE benchmark.", }

It has been commonly observed that a teacher model with superior performance does not necessarily result in a stronger student, highlighting a discrepancy between current teacher training practices and effective knowledge transfer. In order to enhance the guidance of the teacher training process, we introduce the concept of distillation influence to determine the impact of distillation from each training sample on the student{'}s generalization ability. In this paper, we propose Learning Good Teacher Matters (LGTM), an efficient training technique for incorporating distillation influence into the teacher{'}s learning process. By prioritizing samples that are likely to enhance the student{'}s generalization ability, our LGTM outperforms 10 common knowledge distillation baselines on 6 text classification tasks in the GLUE benchmark.

Tailoring Instructions to Student's Learning Levels Boosts Knowledge Distillation

acl-long.111

Poster

2305.09651

https://aclanthology.org/2023.acl-long.112.bib

https://aclanthology.org/2023.acl-long.112/

@inproceedings{chen-etal-2023-rev, title = "{REV}: Information-Theoretic Evaluation of Free-Text Rationales", author = "Chen, Hanjie and Brahman, Faeze and Ren, Xiang and Ji, Yangfeng and Choi, Yejin and Swayamdipta, Swabha", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.112", doi = "10.18653/v1/2023.acl-long.112", pages = "2007--2030", abstract = "Generating free-text rationales is a promising step towards explainable NLP, yet evaluating such rationales remains a challenge. Existing metrics have mostly focused on measuring the association between the rationale and a given label. We argue that an ideal metric should focus on the new information uniquely provided in the rationale that is otherwise not provided in the input or the label. We investigate this research problem from an information-theoretic perspective using conditional V-information (Hewitt et al., 2021). More concretely, we propose a metric called REV (Rationale Evaluation with conditional V-information), to quantify the amount of new, label-relevant information in a rationale beyond the information already available in the input or the label. Experiments across four benchmarks with reasoning tasks, including chain-of-thought, demonstrate the effectiveness of REV in evaluating rationale-label pairs, compared to existing metrics. We further demonstrate REV is consistent with human judgments on rationale evaluations and provides more sensitive measurements of new information in free-text rationales. When used alongside traditional performance metrics, REV provides deeper insights into models{'} reasoning and prediction processes.", }

Generating free-text rationales is a promising step towards explainable NLP, yet evaluating such rationales remains a challenge. Existing metrics have mostly focused on measuring the association between the rationale and a given label. We argue that an ideal metric should focus on the new information uniquely provided in the rationale that is otherwise not provided in the input or the label. We investigate this research problem from an information-theoretic perspective using conditional V-information (Hewitt et al., 2021). More concretely, we propose a metric called REV (Rationale Evaluation with conditional V-information), to quantify the amount of new, label-relevant information in a rationale beyond the information already available in the input or the label. Experiments across four benchmarks with reasoning tasks, including chain-of-thought, demonstrate the effectiveness of REV in evaluating rationale-label pairs, compared to existing metrics. We further demonstrate REV is consistent with human judgments on rationale evaluations and provides more sensitive measurements of new information in free-text rationales. When used alongside traditional performance metrics, REV provides deeper insights into models{'} reasoning and prediction processes.

REV: Information-Theoretic Evaluation of Free-Text Rationales

acl-long.112

Poster

2210.04982

https://aclanthology.org/2023.acl-long.113.bib

https://aclanthology.org/2023.acl-long.113/

@inproceedings{behzad-etal-2023-elqa, title = "{ELQA}: A Corpus of Metalinguistic Questions and Answers about {E}nglish", author = "Behzad, Shabnam and Sakaguchi, Keisuke and Schneider, Nathan and Zeldes, Amir", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.113", doi = "10.18653/v1/2023.acl-long.113", pages = "2031--2047", abstract = "We present ELQA, a corpus of questions and answers in and about the English language. Collected from two online forums, the {\textgreater}70k questions (from English learners and others) cover wide-ranging topics including grammar, meaning, fluency, and etymology. The answers include descriptions of general properties of English vocabulary and grammar as well as explanations about specific (correct and incorrect) usage examples. Unlike most NLP datasets, this corpus is metalinguistic{---}it consists of language about language. As such, it can facilitate investigations of the metalinguistic capabilities of NLU models, as well as educational applications in the language learning domain. To study this, we define a free-form question answering task on our dataset and conduct evaluations on multiple LLMs (Large Language Models) to analyze their capacity to generate metalinguistic answers.", }

We present ELQA, a corpus of questions and answers in and about the English language. Collected from two online forums, the {\textgreater}70k questions (from English learners and others) cover wide-ranging topics including grammar, meaning, fluency, and etymology. The answers include descriptions of general properties of English vocabulary and grammar as well as explanations about specific (correct and incorrect) usage examples. Unlike most NLP datasets, this corpus is metalinguistic{---}it consists of language about language. As such, it can facilitate investigations of the metalinguistic capabilities of NLU models, as well as educational applications in the language learning domain. To study this, we define a free-form question answering task on our dataset and conduct evaluations on multiple LLMs (Large Language Models) to analyze their capacity to generate metalinguistic answers.

ELQA: A Corpus of Metalinguistic Questions and Answers about English

acl-long.113

Poster

2205.00395

https://aclanthology.org/2023.acl-long.114.bib

https://aclanthology.org/2023.acl-long.114/

@inproceedings{wang-etal-2023-divide, title = "Divide, Conquer, and Combine: Mixture of Semantic-Independent Experts for Zero-Shot Dialogue State Tracking", author = "Wang, Qingyue and Ding, Liang and Cao, Yanan and Zhan, Yibing and Lin, Zheng and Wang, Shi and Tao, Dacheng and Guo, Li", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.114", doi = "10.18653/v1/2023.acl-long.114", pages = "2048--2061", abstract = "Zero-shot transfer learning for Dialogue State Tracking (DST) helps to handle a variety of task-oriented dialogue domains without the cost of collecting in-domain data. Existing works mainly study common data- or model-level augmentation methods to enhance the generalization but fail to effectively decouple semantics of samples, limiting the zero-shot performance of DST. In this paper, we present a simple and effective {``}divide, conquer and combine{''} solution, which explicitly disentangles the semantics of seen data, and leverages the performance and robustness with the mixture-of-experts mechanism. Specifically, we divide the seen data into semantically independent subsets and train corresponding experts, the newly unseen samples are mapped and inferred with mixture-of-experts with our designed ensemble inference. Extensive experiments on MultiWOZ2.1 upon T5-Adapter show our schema significantly and consistently improves the zero-shot performance, achieving the SOTA on settings without external knowledge, with only 10M trainable parameters.", }

Zero-shot transfer learning for Dialogue State Tracking (DST) helps to handle a variety of task-oriented dialogue domains without the cost of collecting in-domain data. Existing works mainly study common data- or model-level augmentation methods to enhance the generalization but fail to effectively decouple semantics of samples, limiting the zero-shot performance of DST. In this paper, we present a simple and effective {``}divide, conquer and combine{''} solution, which explicitly disentangles the semantics of seen data, and leverages the performance and robustness with the mixture-of-experts mechanism. Specifically, we divide the seen data into semantically independent subsets and train corresponding experts, the newly unseen samples are mapped and inferred with mixture-of-experts with our designed ensemble inference. Extensive experiments on MultiWOZ2.1 upon T5-Adapter show our schema significantly and consistently improves the zero-shot performance, achieving the SOTA on settings without external knowledge, with only 10M trainable parameters.

Divide, Conquer, and Combine: Mixture of Semantic-Independent Experts for Zero-Shot Dialogue State Tracking

acl-long.114

Oral

2306.00434

https://aclanthology.org/2023.acl-long.115.bib

https://aclanthology.org/2023.acl-long.115/

@inproceedings{sikasote-etal-2023-big, title = "{BIG}-{C}: a Multimodal Multi-Purpose Dataset for {B}emba", author = "Sikasote, Claytone and Mukonde, Eunice and Alam, Md Mahfuz Ibn and Anastasopoulos, Antonios", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.115", doi = "10.18653/v1/2023.acl-long.115", pages = "2062--2078", abstract = "We present BIG-C (Bemba Image Grounded Conversations), a large multimodal dataset for Bemba. While Bemba is the most populous language of Zambia, it exhibits a dearth of resources which render the development of language technologies or language processing research almost impossible. The dataset is comprised of multi-turn dialogues between Bemba speakers based on images, transcribed and translated into English. There are more than 92,000 utterances/sentences, amounting to more than 180 hours of audio data with corresponding transcriptions and English translations. We also provide baselines on speech recognition (ASR), machine translation (MT) and speech translation (ST) tasks, and sketch out other potential future multimodal uses of our dataset. We hope that by making the dataset available to the research community, this work will foster research and encourage collaboration across the language, speech, and vision communities especially for languages outside the {``}traditionally{''} used high-resourced ones. All data and code are publicly available: [\url{https://github.com/csikasote/bigc}](\url{https://github.com/csikasote/bigc}).", }

We present BIG-C (Bemba Image Grounded Conversations), a large multimodal dataset for Bemba. While Bemba is the most populous language of Zambia, it exhibits a dearth of resources which render the development of language technologies or language processing research almost impossible. The dataset is comprised of multi-turn dialogues between Bemba speakers based on images, transcribed and translated into English. There are more than 92,000 utterances/sentences, amounting to more than 180 hours of audio data with corresponding transcriptions and English translations. We also provide baselines on speech recognition (ASR), machine translation (MT) and speech translation (ST) tasks, and sketch out other potential future multimodal uses of our dataset. We hope that by making the dataset available to the research community, this work will foster research and encourage collaboration across the language, speech, and vision communities especially for languages outside the {``}traditionally{''} used high-resourced ones. All data and code are publicly available: [\url{https://github.com/csikasote/bigc}](\url{https://github.com/csikasote/bigc}).

BIG-C: a Multimodal Multi-Purpose Dataset for Bemba

acl-long.115

Poster

2305.17202

https://aclanthology.org/2023.acl-long.116.bib

https://aclanthology.org/2023.acl-long.116/

@inproceedings{feng-etal-2023-schema, title = "Schema-Guided User Satisfaction Modeling for Task-Oriented Dialogues", author = "Feng, Yue and Jiao, Yunlong and Prasad, Animesh and Aletras, Nikolaos and Yilmaz, Emine and Kazai, Gabriella", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.116", doi = "10.18653/v1/2023.acl-long.116", pages = "2079--2091", abstract = "User Satisfaction Modeling (USM) is one of the popular choices for task-oriented dialogue systems evaluation, where user satisfaction typically depends on whether the user{'}s task goals were fulfilled by the system. Task-oriented dialogue systems use task schema, which is a set of task attributes, to encode the user{'}s task goals. Existing studies on USM neglect explicitly modeling the user{'}s task goals fulfillment using the task schema. In this paper, we propose SG-USM, a novel schema-guided user satisfaction modeling framework. It explicitly models the degree to which the user{'}s preferences regarding the task attributes are fulfilled by the system for predicting the user{'}s satisfaction level. SG-USM employs a pre-trained language model for encoding dialogue context and task attributes. Further, it employs a fulfillment representation layer for learning how many task attributes have been fulfilled in the dialogue, an importance predictor component for calculating the importance of task attributes. Finally, it predicts the user satisfaction based on task attribute fulfillment and task attribute importance. Experimental results on benchmark datasets (i.e. MWOZ, SGD, ReDial, and JDDC) show that SG-USM consistently outperforms competitive existing methods. Our extensive analysis demonstrates that SG-USM can improve the interpretability of user satisfaction modeling, has good scalability as it can effectively deal with unseen tasks and can also effectively work in low-resource settings by leveraging unlabeled data. Code is available at \url{https://github.com/amzn/user-satisfaction-modeling}.", }

User Satisfaction Modeling (USM) is one of the popular choices for task-oriented dialogue systems evaluation, where user satisfaction typically depends on whether the user{'}s task goals were fulfilled by the system. Task-oriented dialogue systems use task schema, which is a set of task attributes, to encode the user{'}s task goals. Existing studies on USM neglect explicitly modeling the user{'}s task goals fulfillment using the task schema. In this paper, we propose SG-USM, a novel schema-guided user satisfaction modeling framework. It explicitly models the degree to which the user{'}s preferences regarding the task attributes are fulfilled by the system for predicting the user{'}s satisfaction level. SG-USM employs a pre-trained language model for encoding dialogue context and task attributes. Further, it employs a fulfillment representation layer for learning how many task attributes have been fulfilled in the dialogue, an importance predictor component for calculating the importance of task attributes. Finally, it predicts the user satisfaction based on task attribute fulfillment and task attribute importance. Experimental results on benchmark datasets (i.e. MWOZ, SGD, ReDial, and JDDC) show that SG-USM consistently outperforms competitive existing methods. Our extensive analysis demonstrates that SG-USM can improve the interpretability of user satisfaction modeling, has good scalability as it can effectively deal with unseen tasks and can also effectively work in low-resource settings by leveraging unlabeled data. Code is available at \url{https://github.com/amzn/user-satisfaction-modeling}.

Schema-Guided User Satisfaction Modeling for Task-Oriented Dialogues

acl-long.116

Oral

2305.16798

https://aclanthology.org/2023.acl-long.117.bib

https://aclanthology.org/2023.acl-long.117/

@inproceedings{yoo-etal-2023-robust, title = "Robust Multi-bit Natural Language Watermarking through Invariant Features", author = "Yoo, KiYoon and Ahn, Wonhyuk and Jang, Jiho and Kwak, Nojun", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.117", doi = "10.18653/v1/2023.acl-long.117", pages = "2092--2115", abstract = "Recent years have witnessed a proliferation of valuable original natural language contents found in subscription-based media outlets, web novel platforms, and outputs of large language models. However, these contents are susceptible to illegal piracy and potential misuse without proper security measures. This calls for a secure watermarking system to guarantee copyright protection through leakage tracing or ownership identification. To effectively combat piracy and protect copyrights, a multi-bit watermarking framework should be able to embed adequate bits of information and extract the watermarks in a robust manner despite possible corruption. In this work, we explore ways to advance both payload and robustness by following a well-known proposition from image watermarking and identify features in natural language that are invariant to minor corruption. Through a systematic analysis of the possible sources of errors, we further propose a corruption-resistant infill model. Our full method improves upon the previous work on robustness by +16.8{\%} point on average on four datasets, three corruption types, and two corruption ratios", }

Recent years have witnessed a proliferation of valuable original natural language contents found in subscription-based media outlets, web novel platforms, and outputs of large language models. However, these contents are susceptible to illegal piracy and potential misuse without proper security measures. This calls for a secure watermarking system to guarantee copyright protection through leakage tracing or ownership identification. To effectively combat piracy and protect copyrights, a multi-bit watermarking framework should be able to embed adequate bits of information and extract the watermarks in a robust manner despite possible corruption. In this work, we explore ways to advance both payload and robustness by following a well-known proposition from image watermarking and identify features in natural language that are invariant to minor corruption. Through a systematic analysis of the possible sources of errors, we further propose a corruption-resistant infill model. Our full method improves upon the previous work on robustness by +16.8{\%} point on average on four datasets, three corruption types, and two corruption ratios

Robust Multi-bit Natural Language Watermarking through Invariant Features

acl-long.117

Poster

2305.01904

https://aclanthology.org/2023.acl-long.118.bib

https://aclanthology.org/2023.acl-long.118/

@inproceedings{feng-etal-2023-kalm, title = "{KALM}: Knowledge-Aware Integration of Local, Document, and Global Contexts for Long Document Understanding", author = "Feng, Shangbin and Tan, Zhaoxuan and Zhang, Wenqian and Lei, Zhenyu and Tsvetkov, Yulia", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.118", doi = "10.18653/v1/2023.acl-long.118", pages = "2116--2138", abstract = "With the advent of pre-trained language models (LMs), increasing research efforts have been focusing on infusing commonsense and domain-specific knowledge to prepare LMs for downstream tasks. These works attempt to leverage knowledge graphs, the de facto standard of symbolic knowledge representation, along with pre-trained LMs. While existing approaches leverage external knowledge, it remains an open question how to jointly incorporate knowledge graphs represented in varying contexts {---} from local (e.g., sentence), document-level, to global knowledge, to enable knowledge-rich and interpretable exchange across contexts. In addition, incorporating varying contexts can especially benefit long document understanding tasks that leverage pre-trained LMs, typically bounded by the input sequence length. In light of these challenges, we propose KALM, a language model that jointly leverages knowledge in local, document-level, and global contexts for long document understanding. KALM firstly encodes long documents and knowledge graphs into the three knowledge-aware context representations. KALM then processes each context with context-specific layers. These context-specific layers are followed by a ContextFusion layer that facilitates knowledge exchange to derive an overarching document representation. Extensive experiments demonstrate that KALM achieves state-of-the-art performance on three long document understanding tasks across 6 datasets/settings. Further analyses reveal that the three knowledge-aware contexts are complementary and they all contribute to model performance, while the importance and information exchange patterns of different contexts vary on different tasks and datasets.", }

With the advent of pre-trained language models (LMs), increasing research efforts have been focusing on infusing commonsense and domain-specific knowledge to prepare LMs for downstream tasks. These works attempt to leverage knowledge graphs, the de facto standard of symbolic knowledge representation, along with pre-trained LMs. While existing approaches leverage external knowledge, it remains an open question how to jointly incorporate knowledge graphs represented in varying contexts {---} from local (e.g., sentence), document-level, to global knowledge, to enable knowledge-rich and interpretable exchange across contexts. In addition, incorporating varying contexts can especially benefit long document understanding tasks that leverage pre-trained LMs, typically bounded by the input sequence length. In light of these challenges, we propose KALM, a language model that jointly leverages knowledge in local, document-level, and global contexts for long document understanding. KALM firstly encodes long documents and knowledge graphs into the three knowledge-aware context representations. KALM then processes each context with context-specific layers. These context-specific layers are followed by a ContextFusion layer that facilitates knowledge exchange to derive an overarching document representation. Extensive experiments demonstrate that KALM achieves state-of-the-art performance on three long document understanding tasks across 6 datasets/settings. Further analyses reveal that the three knowledge-aware contexts are complementary and they all contribute to model performance, while the importance and information exchange patterns of different contexts vary on different tasks and datasets.

KALM: Knowledge-Aware Integration of Local, Document, and Global Contexts for Long Document Understanding

acl-long.118

Poster

2210.04105

https://aclanthology.org/2023.acl-long.119.bib

https://aclanthology.org/2023.acl-long.119/

@inproceedings{li-etal-2023-attgen, title = "{A}t{TG}en: Attribute Tree Generation for Real-World Attribute Joint Extraction", author = "Li, Yanzeng and Xue, Bingcong and Zhang, Ruoyu and Zou, Lei", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.119", doi = "10.18653/v1/2023.acl-long.119", pages = "2139--2152", abstract = "Attribute extraction aims to identify attribute names and the corresponding values from descriptive texts, which is the foundation for extensive downstream applications such as knowledge graph construction, search engines, and e-Commerce. In previous studies, attribute extraction is generally treated as a classification problem for predicting attribute types or a sequence tagging problem for labeling attribute values, where two paradigms, i.e., closed-world and open-world assumption, are involved. However, both of these paradigms have limitations in terms of real-world applications. And prior studies attempting to integrate these paradigms through ensemble, pipeline, and co-training models, still face challenges like cascading errors, high computational overhead, and difficulty in training. To address these existing problems, this paper presents Attribute Tree, a unified formulation for real-world attribute extraction application, where closed-world, open-world, and semi-open attribute extraction tasks are modeled uniformly. Then a text-to-tree generation model, AtTGen, is proposed to learn annotations from different scenarios efficiently and consistently. Experiments demonstrate that our proposed paradigm well covers various scenarios for real-world applications, and the model achieves state-of-the-art, outperforming existing methods by a large margin on three datasets. Our code, pretrained model, and datasets are available at \url{https://github.com/lsvih/AtTGen}.", }

Attribute extraction aims to identify attribute names and the corresponding values from descriptive texts, which is the foundation for extensive downstream applications such as knowledge graph construction, search engines, and e-Commerce. In previous studies, attribute extraction is generally treated as a classification problem for predicting attribute types or a sequence tagging problem for labeling attribute values, where two paradigms, i.e., closed-world and open-world assumption, are involved. However, both of these paradigms have limitations in terms of real-world applications. And prior studies attempting to integrate these paradigms through ensemble, pipeline, and co-training models, still face challenges like cascading errors, high computational overhead, and difficulty in training. To address these existing problems, this paper presents Attribute Tree, a unified formulation for real-world attribute extraction application, where closed-world, open-world, and semi-open attribute extraction tasks are modeled uniformly. Then a text-to-tree generation model, AtTGen, is proposed to learn annotations from different scenarios efficiently and consistently. Experiments demonstrate that our proposed paradigm well covers various scenarios for real-world applications, and the model achieves state-of-the-art, outperforming existing methods by a large margin on three datasets. Our code, pretrained model, and datasets are available at \url{https://github.com/lsvih/AtTGen}.

AtTGen: Attribute Tree Generation for Real-World Attribute Joint Extraction

acl-long.119

Poster

https://aclanthology.org/2023.acl-long.120.bib

https://aclanthology.org/2023.acl-long.120/

@inproceedings{zhang-etal-2023-extractive, title = "Extractive is not Faithful: An Investigation of Broad Unfaithfulness Problems in Extractive Summarization", author = "Zhang, Shiyue and Wan, David and Bansal, Mohit", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.120", doi = "10.18653/v1/2023.acl-long.120", pages = "2153--2174", abstract = "The problems of unfaithful summaries have been widely discussed under the context of abstractive summarization. Though extractive summarization is less prone to the common unfaithfulness issues of abstractive summaries, does that mean extractive is equal to faithful? Turns out that the answer is no. In this work, we define a typology with five types of broad unfaithfulness problems (including and beyond not-entailment) that can appear in extractive summaries, including incorrect coreference, incomplete coreference, incorrect discourse, incomplete discourse, as well as other misleading information. We ask humans to label these problems out of 1600 English summaries produced by 16 diverse extractive systems. We find that 30{\%} of the summaries have at least one of the five issues. To automatically detect these problems, we find that 5 existing faithfulness evaluation metrics for summarization have poor correlations with human judgment. To remedy this, we propose a new metric, ExtEval, that is designed for detecting unfaithful extractive summaries and is shown to have the best performance. We hope our work can increase the awareness of unfaithfulness problems in extractive summarization and help future work to evaluate and resolve these issues.", }

The problems of unfaithful summaries have been widely discussed under the context of abstractive summarization. Though extractive summarization is less prone to the common unfaithfulness issues of abstractive summaries, does that mean extractive is equal to faithful? Turns out that the answer is no. In this work, we define a typology with five types of broad unfaithfulness problems (including and beyond not-entailment) that can appear in extractive summaries, including incorrect coreference, incomplete coreference, incorrect discourse, incomplete discourse, as well as other misleading information. We ask humans to label these problems out of 1600 English summaries produced by 16 diverse extractive systems. We find that 30{\%} of the summaries have at least one of the five issues. To automatically detect these problems, we find that 5 existing faithfulness evaluation metrics for summarization have poor correlations with human judgment. To remedy this, we propose a new metric, ExtEval, that is designed for detecting unfaithful extractive summaries and is shown to have the best performance. We hope our work can increase the awareness of unfaithfulness problems in extractive summarization and help future work to evaluate and resolve these issues.

Extractive is not Faithful: An Investigation of Broad Unfaithfulness Problems in Extractive Summarization

acl-long.120

Poster

2209.03549

https://aclanthology.org/2023.acl-long.121.bib

https://aclanthology.org/2023.acl-long.121/

@inproceedings{huang-etal-2023-improving, title = "Improving Translation Quality Estimation with Bias Mitigation", author = "Huang, Hui and Wu, Shuangzhi and Chen, Kehai and Di, Hui and Yang, Muyun and Zhao, Tiejun", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.121", doi = "10.18653/v1/2023.acl-long.121", pages = "2175--2190", abstract = "State-of-the-art translation Quality Estimation (QE) models are proven to be biased. More specifically, they over-rely on monolingual features while ignoring the bilingual semantic alignment. In this work, we propose a novel method to mitigate the bias of the QE model and improve estimation performance. Our method is based on the contrastive learning between clean and noisy sentence pairs. We first introduce noise to the target side of the parallel sentence pair, forming the negative samples. With the original parallel pairs as the positive sample, the QE model is contrastively trained to distinguish the positive samples from the negative ones. This objective is jointly trained with the regression-style quality estimation, so as to prevent the QE model from overfitting to monolingual features. Experiments on WMT QE evaluation datasets demonstrate that our method improves the estimation performance by a large margin while mitigating the bias.", }

State-of-the-art translation Quality Estimation (QE) models are proven to be biased. More specifically, they over-rely on monolingual features while ignoring the bilingual semantic alignment. In this work, we propose a novel method to mitigate the bias of the QE model and improve estimation performance. Our method is based on the contrastive learning between clean and noisy sentence pairs. We first introduce noise to the target side of the parallel sentence pair, forming the negative samples. With the original parallel pairs as the positive sample, the QE model is contrastively trained to distinguish the positive samples from the negative ones. This objective is jointly trained with the regression-style quality estimation, so as to prevent the QE model from overfitting to monolingual features. Experiments on WMT QE evaluation datasets demonstrate that our method improves the estimation performance by a large margin while mitigating the bias.

Improving Translation Quality Estimation with Bias Mitigation

acl-long.121

Oral

https://aclanthology.org/2023.acl-long.122.bib

https://aclanthology.org/2023.acl-long.122/

@inproceedings{jon-bojar-2023-breeding, title = "Breeding Machine Translations: Evolutionary approach to survive and thrive in the world of automated evaluation", author = "Jon, Josef and Bojar, Ond{\v{r}}ej", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.122", doi = "10.18653/v1/2023.acl-long.122", pages = "2191--2212", abstract = "We propose a genetic algorithm (GA) based method for modifying $n$-best lists produced by a machine translation (MT) system. Our method offers an innovative approach to improving MT quality and identifying weaknesses in evaluation metrics. Using common GA operations (mutation and crossover) on a list of hypotheses in combination with a fitness function (an arbitrary MT metric), we obtain novel and diverse outputs with high metric scores. With a combination of multiple MT metrics as the fitness function, the proposed method leads to an increase in translation quality as measured by other held-out automatic metrics.With a single metric (including popular ones such as COMET) as the fitness function, we find blind spots and flaws in the metric. This allows for an automated search for adversarial examples in an arbitrary metric, without prior assumptions on the form of such example. As a demonstration of the method, we create datasets of adversarial examples and use them to show that reference-free COMET is substantially less robust than the reference-based version.", }

We propose a genetic algorithm (GA) based method for modifying $n$-best lists produced by a machine translation (MT) system. Our method offers an innovative approach to improving MT quality and identifying weaknesses in evaluation metrics. Using common GA operations (mutation and crossover) on a list of hypotheses in combination with a fitness function (an arbitrary MT metric), we obtain novel and diverse outputs with high metric scores. With a combination of multiple MT metrics as the fitness function, the proposed method leads to an increase in translation quality as measured by other held-out automatic metrics.With a single metric (including popular ones such as COMET) as the fitness function, we find blind spots and flaws in the metric. This allows for an automated search for adversarial examples in an arbitrary metric, without prior assumptions on the form of such example. As a demonstration of the method, we create datasets of adversarial examples and use them to show that reference-free COMET is substantially less robust than the reference-based version.

Breeding Machine Translations: Evolutionary approach to survive and thrive in the world of automated evaluation

acl-long.122

Oral

2305.19330

https://aclanthology.org/2023.acl-long.123.bib

https://aclanthology.org/2023.acl-long.123/

@inproceedings{sun-etal-2023-moraldial, title = "{M}oral{D}ial: A Framework to Train and Evaluate Moral Dialogue Systems via Moral Discussions", author = "Sun, Hao and Zhang, Zhexin and Mi, Fei and Wang, Yasheng and Liu, Wei and Cui, Jianwei and Wang, Bin and Liu, Qun and Huang, Minlie", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.123", doi = "10.18653/v1/2023.acl-long.123", pages = "2213--2230", abstract = "Morality in dialogue systems has raised great attention in research recently. A moral dialogue system aligned with users{'} values could enhance conversation engagement and user connections. In this paper, we propose a framework, MoralDial to train and evaluate moral dialogue systems. In our framework, we first explore the communication mechanisms of morality and resolve expressed morality into three parts, which indicate the roadmap for building a moral dialogue system. Based on that, we design a simple yet effective method: constructing moral discussions between simulated specific users and the dialogue system. The constructed discussions consist of expressing, explaining, revising, and inferring moral views in dialogue exchanges, which makes conversational models learn morality well in a natural manner. Furthermore, we propose a novel evaluation method under the framework. We evaluate the multiple aspects of morality by judging the relation between dialogue responses and human values in discussions, where the multifaceted nature of morality is particularly considered. Automatic and manual experiments demonstrate that our framework is promising to train and evaluate moral dialogue systems.", }

Morality in dialogue systems has raised great attention in research recently. A moral dialogue system aligned with users{'} values could enhance conversation engagement and user connections. In this paper, we propose a framework, MoralDial to train and evaluate moral dialogue systems. In our framework, we first explore the communication mechanisms of morality and resolve expressed morality into three parts, which indicate the roadmap for building a moral dialogue system. Based on that, we design a simple yet effective method: constructing moral discussions between simulated specific users and the dialogue system. The constructed discussions consist of expressing, explaining, revising, and inferring moral views in dialogue exchanges, which makes conversational models learn morality well in a natural manner. Furthermore, we propose a novel evaluation method under the framework. We evaluate the multiple aspects of morality by judging the relation between dialogue responses and human values in discussions, where the multifaceted nature of morality is particularly considered. Automatic and manual experiments demonstrate that our framework is promising to train and evaluate moral dialogue systems.

MoralDial: A Framework to Train and Evaluate Moral Dialogue Systems via Moral Discussions

acl-long.123

Poster

2212.10720

https://huggingface.co/papers/2212.10720

https://aclanthology.org/2023.acl-long.124.bib

https://aclanthology.org/2023.acl-long.124/

@inproceedings{wu-etal-2023-denoising, title = "Denoising Bottleneck with Mutual Information Maximization for Video Multimodal Fusion", author = "Wu, Shaoxiang and Dai, Damai and Qin, Ziwei and Liu, Tianyu and Lin, Binghuai and Cao, Yunbo and Sui, Zhifang", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.124", doi = "10.18653/v1/2023.acl-long.124", pages = "2231--2243", abstract = "Video multimodal fusion aims to integrate multimodal signals in videos, such as visual, audio and text, to make a complementary prediction with multiple modalities contents. However, unlike other image-text multimodal tasks, video has longer multimodal sequences with more redundancy and noise in both visual and audio modalities. Prior denoising methods like forget gate are coarse in the granularity of noise filtering. They often suppress the redundant and noisy information at the risk of losing critical information. Therefore, we propose a denoising bottleneck fusion (DBF) model for fine-grained video multimodal fusion. On the one hand, we employ a bottleneck mechanism to filter out noise and redundancy with a restrained receptive field. On the other hand, we use a mutual information maximization module to regulate the filter-out module to preserve key information within different modalities. Our DBF model achieves significant improvement over current state-of-the-art baselines on multiple benchmarks covering multimodal sentiment analysis and multimodal summarization tasks. It proves that our model can effectively capture salient features from noisy and redundant video, audio, and text inputs. The code for this paper will be publicly available at \url{https://github.com/WSXRHFG/DBF}", }

Video multimodal fusion aims to integrate multimodal signals in videos, such as visual, audio and text, to make a complementary prediction with multiple modalities contents. However, unlike other image-text multimodal tasks, video has longer multimodal sequences with more redundancy and noise in both visual and audio modalities. Prior denoising methods like forget gate are coarse in the granularity of noise filtering. They often suppress the redundant and noisy information at the risk of losing critical information. Therefore, we propose a denoising bottleneck fusion (DBF) model for fine-grained video multimodal fusion. On the one hand, we employ a bottleneck mechanism to filter out noise and redundancy with a restrained receptive field. On the other hand, we use a mutual information maximization module to regulate the filter-out module to preserve key information within different modalities. Our DBF model achieves significant improvement over current state-of-the-art baselines on multiple benchmarks covering multimodal sentiment analysis and multimodal summarization tasks. It proves that our model can effectively capture salient features from noisy and redundant video, audio, and text inputs. The code for this paper will be publicly available at \url{https://github.com/WSXRHFG/DBF}

Denoising Bottleneck with Mutual Information Maximization for Video Multimodal Fusion

acl-long.124

Poster

2305.14652

https://aclanthology.org/2023.acl-long.125.bib

https://aclanthology.org/2023.acl-long.125/

@inproceedings{wang-etal-2023-simlm, title = "{S}im{LM}: Pre-training with Representation Bottleneck for Dense Passage Retrieval", author = "Wang, Liang and Yang, Nan and Huang, Xiaolong and Jiao, Binxing and Yang, Linjun and Jiang, Daxin and Majumder, Rangan and Wei, Furu", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.125", doi = "10.18653/v1/2023.acl-long.125", pages = "2244--2258", abstract = "In this paper, we propose SimLM (Similarity matching with Language Model pre-training), a simple yet effective pre-training method for dense passage retrieval. It employs a simple bottleneck architecture that learns to compress the passage information into a dense vector through self-supervised pre-training. We use a replaced language modeling objective, which is inspired by ELECTRA (Clark et al., 2020), to improve the sample efficiency and reduce the mismatch of the input distribution between pre-training and fine-tuning. SimLM only requires access to an unlabeled corpus and is more broadly applicable when there are no labeled data or queries. We conduct experiments on several large-scale passage retrieval datasets and show substantial improvements over strong baselines under various settings. Remarkably, SimLM even outperforms multi-vector approaches such as ColBERTv2 (Santhanam et al., 2021) which incurs significantly more storage cost. Our code and model checkpoints are available at \url{https://github.com/microsoft/unilm/tree/master/simlm} .", }

In this paper, we propose SimLM (Similarity matching with Language Model pre-training), a simple yet effective pre-training method for dense passage retrieval. It employs a simple bottleneck architecture that learns to compress the passage information into a dense vector through self-supervised pre-training. We use a replaced language modeling objective, which is inspired by ELECTRA (Clark et al., 2020), to improve the sample efficiency and reduce the mismatch of the input distribution between pre-training and fine-tuning. SimLM only requires access to an unlabeled corpus and is more broadly applicable when there are no labeled data or queries. We conduct experiments on several large-scale passage retrieval datasets and show substantial improvements over strong baselines under various settings. Remarkably, SimLM even outperforms multi-vector approaches such as ColBERTv2 (Santhanam et al., 2021) which incurs significantly more storage cost. Our code and model checkpoints are available at \url{https://github.com/microsoft/unilm/tree/master/simlm} .

SimLM: Pre-training with Representation Bottleneck for Dense Passage Retrieval

acl-long.125

Poster

2207.02578

https://aclanthology.org/2023.acl-long.126.bib

https://aclanthology.org/2023.acl-long.126/

@inproceedings{dai-zeng-2023-ultra, title = "From Ultra-Fine to Fine: Fine-tuning Ultra-Fine Entity Typing Models to Fine-grained", author = "Dai, Hongliang and Zeng, Ziqian", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.126", doi = "10.18653/v1/2023.acl-long.126", pages = "2259--2270", abstract = "For the task of fine-grained entity typing (FET), due to the use of a large number of entity types, it is usually considered too costly to manually annotating a training dataset that contains an ample number of examples for each type. A common way to address this problem is to use distantly annotated training data that contains incorrect labels. However, the performance of models trained solely with such data can be limited by the errors in the automatic annotation. Recently, there are a few approaches that no longer follow this conventional way. But without using sufficient direct entity typing supervision may also cause them to yield inferior performance. In this paper, we propose a new approach that can avoid the need of creating distantly labeled data whenever there is a new type schema. We first train an entity typing model that have an extremely board type coverage by using the ultra-fine entity typing data. Then, when there is a need to produce a model for a newly designed fine-grained entity type schema. We can simply fine-tune the previously trained model with a small number of examples annotated under this schema. Experimental results show that our approach achieves outstanding performance for FET under the few-shot setting. It can also outperform state-of-the-art weak supervision based methods after fine-tuning the model with only a small size manually annotated training set.", }

For the task of fine-grained entity typing (FET), due to the use of a large number of entity types, it is usually considered too costly to manually annotating a training dataset that contains an ample number of examples for each type. A common way to address this problem is to use distantly annotated training data that contains incorrect labels. However, the performance of models trained solely with such data can be limited by the errors in the automatic annotation. Recently, there are a few approaches that no longer follow this conventional way. But without using sufficient direct entity typing supervision may also cause them to yield inferior performance. In this paper, we propose a new approach that can avoid the need of creating distantly labeled data whenever there is a new type schema. We first train an entity typing model that have an extremely board type coverage by using the ultra-fine entity typing data. Then, when there is a need to produce a model for a newly designed fine-grained entity type schema. We can simply fine-tune the previously trained model with a small number of examples annotated under this schema. Experimental results show that our approach achieves outstanding performance for FET under the few-shot setting. It can also outperform state-of-the-art weak supervision based methods after fine-tuning the model with only a small size manually annotated training set.

From Ultra-Fine to Fine: Fine-tuning Ultra-Fine Entity Typing Models to Fine-grained

acl-long.126

Poster

2312.06188

https://aclanthology.org/2023.acl-long.127.bib

https://aclanthology.org/2023.acl-long.127/

@inproceedings{bansal-sharma-2023-controlling, title = "Controlling Learned Effects to Reduce Spurious Correlations in Text Classifiers", author = "Bansal, Parikshit and Sharma, Amit", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.127", doi = "10.18653/v1/2023.acl-long.127", pages = "2271--2287", abstract = "To address the problem of NLP classifiers learning spurious correlations between training features and target labels, a common approach is to make the model{'}s predictions invariant to these features. However, this can be counter-productive when the features have a non-zero causal effect on the target label and thus are important for prediction. Therefore, using methods from the causal inference literature, we propose an algorithm to regularize the learnt effect of the features on the model{'}s prediction to the estimated effect of feature on label. This results in an automated augmentation method that leverages the estimated effect of a feature to appropriately change the labels for new augmented inputs. On toxicity and IMDB review datasets, the proposed algorithm minimises spurious correlations and improves the minority group (i.e., samples breaking spurious correlations) accuracy, while also improving the total accuracy compared to standard training.", }

To address the problem of NLP classifiers learning spurious correlations between training features and target labels, a common approach is to make the model{'}s predictions invariant to these features. However, this can be counter-productive when the features have a non-zero causal effect on the target label and thus are important for prediction. Therefore, using methods from the causal inference literature, we propose an algorithm to regularize the learnt effect of the features on the model{'}s prediction to the estimated effect of feature on label. This results in an automated augmentation method that leverages the estimated effect of a feature to appropriately change the labels for new augmented inputs. On toxicity and IMDB review datasets, the proposed algorithm minimises spurious correlations and improves the minority group (i.e., samples breaking spurious correlations) accuracy, while also improving the total accuracy compared to standard training.

Controlling Learned Effects to Reduce Spurious Correlations in Text Classifiers

acl-long.127

Poster

2305.16863

https://aclanthology.org/2023.acl-long.128.bib

https://aclanthology.org/2023.acl-long.128/

@inproceedings{lu-etal-2023-makes, title = "What Makes Pre-trained Language Models Better Zero-shot Learners?", author = "Lu, Jinghui and Zhu, Dongsheng and Han, Weidong and Zhao, Rui and Mac Namee, Brian and Tan, Fei", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.128", doi = "10.18653/v1/2023.acl-long.128", pages = "2288--2303", abstract = "Current methods for prompt learning in zero-shot scenarios widely rely on a development set with sufficient human-annotated data to select the best-performing prompt template a posteriori. This is not ideal because in a real-world zero-shot scenario of practical relevance, no labelled data is available. Thus, we propose a simple yet effective method for screening reasonable prompt templates in zero-shot text classification: Perplexity Selection (Perplection). We hypothesize that language discrepancy can be used to measure the efficacy of prompt templates, and thereby develop a substantiated perplexity-based scheme allowing for forecasting the performance of prompt templates in advance. Experiments show that our method leads to improved prediction performance in a realistic zero-shot setting, eliminating the need for any labelled examples.", }

Current methods for prompt learning in zero-shot scenarios widely rely on a development set with sufficient human-annotated data to select the best-performing prompt template a posteriori. This is not ideal because in a real-world zero-shot scenario of practical relevance, no labelled data is available. Thus, we propose a simple yet effective method for screening reasonable prompt templates in zero-shot text classification: Perplexity Selection (Perplection). We hypothesize that language discrepancy can be used to measure the efficacy of prompt templates, and thereby develop a substantiated perplexity-based scheme allowing for forecasting the performance of prompt templates in advance. Experiments show that our method leads to improved prediction performance in a realistic zero-shot setting, eliminating the need for any labelled examples.

What Makes Pre-trained Language Models Better Zero-shot Learners?

acl-long.128

Poster

2209.15206

https://aclanthology.org/2023.acl-long.129.bib

https://aclanthology.org/2023.acl-long.129/

@inproceedings{lyu-etal-2023-z, title = "{Z}-{ICL}: Zero-Shot In-Context Learning with Pseudo-Demonstrations", author = "Lyu, Xinxi and Min, Sewon and Beltagy, Iz and Zettlemoyer, Luke and Hajishirzi, Hannaneh", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.129", doi = "10.18653/v1/2023.acl-long.129", pages = "2304--2317", abstract = "Although large language models can be prompted for both zero- and few-shot learning, performance drops significantly when no demonstrations are available. In this paper, we introduce Z-ICL, a new zero-shot method that closes the gap by constructing pseudo-demonstrations for a given test input using a raw text corpus. Concretely, pseudo-demonstrations are constructed by (1) finding the nearest neighbors to the test input from the corpus and pairing them with random task labels, and (2) applying a set of techniques to reduce the amount of direct copying the model does from the resulting demonstrations. Evaluation on nine classification datasets shows that Z-ICL outperforms previous zero-shot methods by a significant margin, and is on par with in-context learning with labeled training data in the few-shot setting. Overall, Z-ICL provides a significantly higher estimate of the zero-shot performance levels of a model, and supports future efforts to develop better pseudo-demonstrations that further improve zero-shot results.", }

Although large language models can be prompted for both zero- and few-shot learning, performance drops significantly when no demonstrations are available. In this paper, we introduce Z-ICL, a new zero-shot method that closes the gap by constructing pseudo-demonstrations for a given test input using a raw text corpus. Concretely, pseudo-demonstrations are constructed by (1) finding the nearest neighbors to the test input from the corpus and pairing them with random task labels, and (2) applying a set of techniques to reduce the amount of direct copying the model does from the resulting demonstrations. Evaluation on nine classification datasets shows that Z-ICL outperforms previous zero-shot methods by a significant margin, and is on par with in-context learning with labeled training data in the few-shot setting. Overall, Z-ICL provides a significantly higher estimate of the zero-shot performance levels of a model, and supports future efforts to develop better pseudo-demonstrations that further improve zero-shot results.

Z-ICL: Zero-Shot In-Context Learning with Pseudo-Demonstrations

acl-long.129

Poster

2212.09865

https://aclanthology.org/2023.acl-long.130.bib

https://aclanthology.org/2023.acl-long.130/

@inproceedings{guo-etal-2023-learning, title = "Learning Optimal Policy for Simultaneous Machine Translation via Binary Search", author = "Guo, Shoutao and Zhang, Shaolei and Feng, Yang", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.130", doi = "10.18653/v1/2023.acl-long.130", pages = "2318--2333", abstract = "Simultaneous machine translation (SiMT) starts to output translation while reading the source sentence and needs a precise policy to decide when to output the generated translation. Therefore, the policy determines the number of source tokens read during the translation of each target token. However, it is difficult to learn a precise translation policy to achieve good latency-quality trade-offs, because there is no golden policy corresponding to parallel sentences as explicit supervision. In this paper, we present a new method for constructing the optimal policy online via binary search. By employing explicit supervision, our approach enables the SiMT model to learn the optimal policy, which can guide the model in completing the translation during inference. Experiments on four translation tasks show that our method can exceed strong baselines across all latency scenarios.", }

Simultaneous machine translation (SiMT) starts to output translation while reading the source sentence and needs a precise policy to decide when to output the generated translation. Therefore, the policy determines the number of source tokens read during the translation of each target token. However, it is difficult to learn a precise translation policy to achieve good latency-quality trade-offs, because there is no golden policy corresponding to parallel sentences as explicit supervision. In this paper, we present a new method for constructing the optimal policy online via binary search. By employing explicit supervision, our approach enables the SiMT model to learn the optimal policy, which can guide the model in completing the translation during inference. Experiments on four translation tasks show that our method can exceed strong baselines across all latency scenarios.

Learning Optimal Policy for Simultaneous Machine Translation via Binary Search

acl-long.130

Poster

2305.12774

https://aclanthology.org/2023.acl-long.131.bib

https://aclanthology.org/2023.acl-long.131/

@inproceedings{wang-etal-2023-better, title = "Better Simultaneous Translation with Monotonic Knowledge Distillation", author = "Wang, Shushu and Wu, Jing and Fan, Kai and Luo, Wei and Xiao, Jun and Huang, Zhongqiang", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.131", doi = "10.18653/v1/2023.acl-long.131", pages = "2334--2349", abstract = "Simultaneous machine translation (SiMT) presents a unique challenge as it requires generating target tokens before the source sentence is fully consumed. This can lead to the hallucination problem, where target tokens are generated without support from the source sentence. The prefix-to-prefix training data used to train SiMT models are not always parallel, due to divergent word order between the source and target languages, and can contribute to the problem. In this paper, we propose a novel approach that leverages traditional translation models as teachers and employs a two-stage beam search algorithm to generate monotonic yet accurate reference translations for sequence-level knowledge distillation. Experimental results demonstrate the significant improvements achieved by our approach over multiple strong SiMT baselines, leading to new state-of-the-art performance across various language pairs. Notably, when evaluated on a monotonic version of the WMT15 De-En test set, which includes references generated in a more monotonic style by professional translators, our approach achieves even more substantial improvement over the baselines. The source code and data are publicly available for further exploration.", }

Simultaneous machine translation (SiMT) presents a unique challenge as it requires generating target tokens before the source sentence is fully consumed. This can lead to the hallucination problem, where target tokens are generated without support from the source sentence. The prefix-to-prefix training data used to train SiMT models are not always parallel, due to divergent word order between the source and target languages, and can contribute to the problem. In this paper, we propose a novel approach that leverages traditional translation models as teachers and employs a two-stage beam search algorithm to generate monotonic yet accurate reference translations for sequence-level knowledge distillation. Experimental results demonstrate the significant improvements achieved by our approach over multiple strong SiMT baselines, leading to new state-of-the-art performance across various language pairs. Notably, when evaluated on a monotonic version of the WMT15 De-En test set, which includes references generated in a more monotonic style by professional translators, our approach achieves even more substantial improvement over the baselines. The source code and data are publicly available for further exploration.

Better Simultaneous Translation with Monotonic Knowledge Distillation

acl-long.131

Poster

https://aclanthology.org/2023.acl-long.132.bib

https://aclanthology.org/2023.acl-long.132/

@inproceedings{falk-lapesa-2023-storyarg, title = "{S}tory{ARG}: a corpus of narratives and personal experiences in argumentative texts", author = "Falk, Neele and Lapesa, Gabriella", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.132", doi = "10.18653/v1/2023.acl-long.132", pages = "2350--2372", abstract = "Humans are storytellers, even in communication scenarios which are assumed to be more rationality-oriented, such as argumentation. Indeed, supporting arguments with narratives or personal experiences (henceforth, stories) is a very natural thing to do {--} and yet, this phenomenon is largely unexplored in computational argumentation. Which role do stories play in an argument? Do they make the argument more effective? What are their narrative properties? To address these questions, we collected and annotated StoryARG, a dataset sampled from well-established corpora in computational argumentation (ChangeMyView and RegulationRoom), and the Social Sciences (Europolis), as well as comments to New York Times articles. StoryARG contains 2451 textual spans annotated at two levels. At the argumentative level, we annotate the function of the story (e.g., clarification, disclosure of harm, search for a solution, establishing speaker{'}s authority), as well as its impact on the effectiveness of the argument and its emotional load. At the level of narrative properties, we annotate whether the story has a plot-like development, is factual or hypothetical, and who the protagonist is. What makes a story effective in an argument? Our analysis of the annotations in StoryARG uncover a positive impact on effectiveness for stories which illustrate a solution to a problem, and in general, annotator-specific preferences that we investigate with regression analysis.", }

Humans are storytellers, even in communication scenarios which are assumed to be more rationality-oriented, such as argumentation. Indeed, supporting arguments with narratives or personal experiences (henceforth, stories) is a very natural thing to do {--} and yet, this phenomenon is largely unexplored in computational argumentation. Which role do stories play in an argument? Do they make the argument more effective? What are their narrative properties? To address these questions, we collected and annotated StoryARG, a dataset sampled from well-established corpora in computational argumentation (ChangeMyView and RegulationRoom), and the Social Sciences (Europolis), as well as comments to New York Times articles. StoryARG contains 2451 textual spans annotated at two levels. At the argumentative level, we annotate the function of the story (e.g., clarification, disclosure of harm, search for a solution, establishing speaker{'}s authority), as well as its impact on the effectiveness of the argument and its emotional load. At the level of narrative properties, we annotate whether the story has a plot-like development, is factual or hypothetical, and who the protagonist is. What makes a story effective in an argument? Our analysis of the annotations in StoryARG uncover a positive impact on effectiveness for stories which illustrate a solution to a problem, and in general, annotator-specific preferences that we investigate with regression analysis.

StoryARG: a corpus of narratives and personal experiences in argumentative texts

acl-long.132

Poster

https://aclanthology.org/2023.acl-long.133.bib

https://aclanthology.org/2023.acl-long.133/

@inproceedings{eremeev-etal-2023-injecting, title = "Injecting knowledge into language generation: a case study in auto-charting after-visit care instructions from medical dialogue", author = "Eremeev, Maksim and Valmianski, Ilya and Amatriain, Xavier and Kannan, Anitha", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.133", doi = "10.18653/v1/2023.acl-long.133", pages = "2373--2390", abstract = "Factual correctness is often the limiting factor in practical applications of natural language generation in high-stakes domains such as healthcare. An essential requirement for maintaining factuality is the ability to deal with rare tokens. This paper focuses on rare tokens that appear in both the source and the reference sequences, and which, when missed during generation, decrease the factual correctness of the output text. For high-stake domains that are also knowledge-rich, we show how to use knowledge to (a) identify which rare tokens that appear in both source and reference are important and (b) uplift their conditional probability. We introduce the {``}utilization rate{''} that encodes knowledge and serves as a regularizer by maximizing the marginal probability of selected tokens. We present a study in a knowledge-rich domain of healthcare, where we tackle the problem of generating after-visit care instructions based on patient-doctor dialogues. We verify that, in our dataset, specific medical concepts with high utilization rates are underestimated by conventionally trained sequence-to-sequence models. We observe that correcting this with our approach to knowledge injection reduces the uncertainty of the model as well as improves factuality and coherence without negatively impacting fluency.", }

Factual correctness is often the limiting factor in practical applications of natural language generation in high-stakes domains such as healthcare. An essential requirement for maintaining factuality is the ability to deal with rare tokens. This paper focuses on rare tokens that appear in both the source and the reference sequences, and which, when missed during generation, decrease the factual correctness of the output text. For high-stake domains that are also knowledge-rich, we show how to use knowledge to (a) identify which rare tokens that appear in both source and reference are important and (b) uplift their conditional probability. We introduce the {``}utilization rate{''} that encodes knowledge and serves as a regularizer by maximizing the marginal probability of selected tokens. We present a study in a knowledge-rich domain of healthcare, where we tackle the problem of generating after-visit care instructions based on patient-doctor dialogues. We verify that, in our dataset, specific medical concepts with high utilization rates are underestimated by conventionally trained sequence-to-sequence models. We observe that correcting this with our approach to knowledge injection reduces the uncertainty of the model as well as improves factuality and coherence without negatively impacting fluency.

Injecting knowledge into language generation: a case study in auto-charting after-visit care instructions from medical dialogue

acl-long.133

Oral

2306.03652

https://aclanthology.org/2023.acl-long.134.bib

https://aclanthology.org/2023.acl-long.134/

@inproceedings{li-etal-2023-sequence, title = "Sequence Parallelism: Long Sequence Training from System Perspective", author = "Li, Shenggui and Xue, Fuzhao and Baranwal, Chaitanya and Li, Yongbin and You, Yang", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.134", doi = "10.18653/v1/2023.acl-long.134", pages = "2391--2404", abstract = "Transformer achieves promising results on various tasks. However, self-attention suffers from quadratic memory requirements with respect to the sequence length. Existing work focuses on reducing time and space complexity from an algorithm perspective. In this work, we propose sequence parallelism, a memory-efficient parallelism to solve this issue from system perspective instead. Our approach is compatible with most existing parallelisms (e.g., data, pipeline, and tensor parallelism), which means our sequence parallelism makes 4D parallelism possible. More importantly, we no longer require a single device to hold the whole sequence. Besides, using efficient attention with linear complexity, our sequence parallelism enables us to train transformer with infinite long sequence. Specifically, we split the input sequence into multiple chunks and feed each chunk into its corresponding device (i.e., GPU). To compute the attention output, we integrated ring-style communication with self-attention calculation and proposed Ring Self-Attention (RSA). Experiments show that sequence parallelism performs well when scaling with batch size and sequence length. Compared with tensor parallelism, our approach achieved $13.7\times$ and $3.0\times$ maximum batch size and sequence length respectively when scaling up to 64 NVIDIA P100 GPUs. With efficient attention, sequence can handle sequence with over 114K tokens, which is over $27\times$ longer than existing efficient attention works holding the whole sequence on a single device.", }

Transformer achieves promising results on various tasks. However, self-attention suffers from quadratic memory requirements with respect to the sequence length. Existing work focuses on reducing time and space complexity from an algorithm perspective. In this work, we propose sequence parallelism, a memory-efficient parallelism to solve this issue from system perspective instead. Our approach is compatible with most existing parallelisms (e.g., data, pipeline, and tensor parallelism), which means our sequence parallelism makes 4D parallelism possible. More importantly, we no longer require a single device to hold the whole sequence. Besides, using efficient attention with linear complexity, our sequence parallelism enables us to train transformer with infinite long sequence. Specifically, we split the input sequence into multiple chunks and feed each chunk into its corresponding device (i.e., GPU). To compute the attention output, we integrated ring-style communication with self-attention calculation and proposed Ring Self-Attention (RSA). Experiments show that sequence parallelism performs well when scaling with batch size and sequence length. Compared with tensor parallelism, our approach achieved $13.7\times$ and $3.0\times$ maximum batch size and sequence length respectively when scaling up to 64 NVIDIA P100 GPUs. With efficient attention, sequence can handle sequence with over 114K tokens, which is over $27\times$ longer than existing efficient attention works holding the whole sequence on a single device.

Sequence Parallelism: Long Sequence Training from System Perspective

acl-long.134

Poster

2105.13120

https://huggingface.co/papers/2105.13120

https://aclanthology.org/2023.acl-long.135.bib

https://aclanthology.org/2023.acl-long.135/

@inproceedings{wang-etal-2023-mustie, title = "{MUSTIE}: Multimodal Structural Transformer for Web Information Extraction", author = "Wang, Qifan and Wang, Jingang and Quan, Xiaojun and Feng, Fuli and Xu, Zenglin and Nie, Shaoliang and Wang, Sinong and Khabsa, Madian and Firooz, Hamed and Liu, Dongfang", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.135", doi = "10.18653/v1/2023.acl-long.135", pages = "2405--2420", abstract = "The task of web information extraction is to extract target fields of an object from web pages, such as extracting the name, genre and actor from a movie page. Recent sequential modeling approaches have achieved state-of-the-art results on web information extraction. However, most of these methods only focus on extracting information from textual sources while ignoring the rich information from other modalities such as image and web layout. In this work, we propose a novel MUltimodal Structural Transformer (MUST) that incorporates multiple modalities for web information extraction. Concretely, we develop a structural encoder that jointly encodes the multimodal information based on the HTML structure of the web layout, where high-level DOM nodes, and low-level text and image tokens are introduced to represent the entire page. Structural attention patterns are designed to learn effective cross-modal embeddings for all DOM nodes and low-level tokens. An extensive set of experiments are conducted on WebSRC and Common Crawl benchmarks. Experimental results demonstrate the superior performance of MUST over several state-of-the-art baselines.", }

The task of web information extraction is to extract target fields of an object from web pages, such as extracting the name, genre and actor from a movie page. Recent sequential modeling approaches have achieved state-of-the-art results on web information extraction. However, most of these methods only focus on extracting information from textual sources while ignoring the rich information from other modalities such as image and web layout. In this work, we propose a novel MUltimodal Structural Transformer (MUST) that incorporates multiple modalities for web information extraction. Concretely, we develop a structural encoder that jointly encodes the multimodal information based on the HTML structure of the web layout, where high-level DOM nodes, and low-level text and image tokens are introduced to represent the entire page. Structural attention patterns are designed to learn effective cross-modal embeddings for all DOM nodes and low-level tokens. An extensive set of experiments are conducted on WebSRC and Common Crawl benchmarks. Experimental results demonstrate the superior performance of MUST over several state-of-the-art baselines.

MUSTIE: Multimodal Structural Transformer for Web Information Extraction

acl-long.135

Poster

https://aclanthology.org/2023.acl-long.136.bib

https://aclanthology.org/2023.acl-long.136/

@inproceedings{yu-etal-2023-augmentation, title = "Augmentation-Adapted Retriever Improves Generalization of Language Models as Generic Plug-In", author = "Yu, Zichun and Xiong, Chenyan and Yu, Shi and Liu, Zhiyuan", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.136", doi = "10.18653/v1/2023.acl-long.136", pages = "2421--2436", abstract = "Retrieval augmentation can aid language models (LMs) in knowledge-intensive tasks by supplying them with external information. Prior works on retrieval augmentation usually jointly fine-tune the retriever and the LM, making them closely coupled. In this paper, we explore the scheme of generic retrieval plug-in: the retriever is to assist target LMs that may not be known beforehand or are unable to be fine-tuned together. To retrieve useful documents for unseen target LMs, we propose augmentation-adapted retriever (AAR), which learns LM{'}s preferences obtained from a known source LM. Experiments on the MMLU and PopQA datasets demonstrate that our AAR trained with a small source LM is able to significantly improve the zero-shot generalization of larger target LMs ranging from 250M Flan-T5 to 175B InstructGPT. Further analysis indicates that the preferences of different LMs overlap, enabling AAR trained with a single source LM to serve as a generic plug-in for various target LMs. Our code is open-sourced at \url{https://github.com/OpenMatch/Augmentation-Adapted-Retriever}.", }

Retrieval augmentation can aid language models (LMs) in knowledge-intensive tasks by supplying them with external information. Prior works on retrieval augmentation usually jointly fine-tune the retriever and the LM, making them closely coupled. In this paper, we explore the scheme of generic retrieval plug-in: the retriever is to assist target LMs that may not be known beforehand or are unable to be fine-tuned together. To retrieve useful documents for unseen target LMs, we propose augmentation-adapted retriever (AAR), which learns LM{'}s preferences obtained from a known source LM. Experiments on the MMLU and PopQA datasets demonstrate that our AAR trained with a small source LM is able to significantly improve the zero-shot generalization of larger target LMs ranging from 250M Flan-T5 to 175B InstructGPT. Further analysis indicates that the preferences of different LMs overlap, enabling AAR trained with a single source LM to serve as a generic plug-in for various target LMs. Our code is open-sourced at \url{https://github.com/OpenMatch/Augmentation-Adapted-Retriever}.

Augmentation-Adapted Retriever Improves Generalization of Language Models as Generic Plug-In

acl-long.136

Oral

2305.17331

https://huggingface.co/papers/2305.17331

https://aclanthology.org/2023.acl-long.137.bib

https://aclanthology.org/2023.acl-long.137/

@inproceedings{chen-etal-2023-tablevlm, title = "{T}able{VLM}: Multi-modal Pre-training for Table Structure Recognition", author = "Chen, Leiyuan and Huang, Chengsong and Zheng, Xiaoqing and Lin, Jinshu and Huang, Xuanjing", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.137", doi = "10.18653/v1/2023.acl-long.137", pages = "2437--2449", abstract = "Tables are widely used in research and business, which are suitable for human consumption, but not easily machine-processable, particularly when tables are present in images. One of the main challenges to extracting data from images of tables is accurately recognizing table structures, especially for complex tables with cross rows and columns. In this study, we propose a novel multi-modal pre-training model for table structure recognition, named TableVLM.With a two-stream multi-modal transformer-based encoder-decoder architecture, TableVLM learns to capture rich table structure-related features by multiple carefully-designed unsupervised objectives inspired by the notion of masked visual-language modeling. To pre-train this model, we also created a dataset, called ComplexTable, which consists of 1,000K samples to be released publicly. Experiment results show that the model built on pre-trained TableVLM can improve the performance up to 1.97{\%} in tree-editing-distance-score on ComplexTable.", }

Tables are widely used in research and business, which are suitable for human consumption, but not easily machine-processable, particularly when tables are present in images. One of the main challenges to extracting data from images of tables is accurately recognizing table structures, especially for complex tables with cross rows and columns. In this study, we propose a novel multi-modal pre-training model for table structure recognition, named TableVLM.With a two-stream multi-modal transformer-based encoder-decoder architecture, TableVLM learns to capture rich table structure-related features by multiple carefully-designed unsupervised objectives inspired by the notion of masked visual-language modeling. To pre-train this model, we also created a dataset, called ComplexTable, which consists of 1,000K samples to be released publicly. Experiment results show that the model built on pre-trained TableVLM can improve the performance up to 1.97{\%} in tree-editing-distance-score on ComplexTable.

TableVLM: Multi-modal Pre-training for Table Structure Recognition

acl-long.137

Poster

https://aclanthology.org/2023.acl-long.138.bib

https://aclanthology.org/2023.acl-long.138/

@inproceedings{xu-etal-2023-nli, title = "Can {NLI} Provide Proper Indirect Supervision for Low-resource Biomedical Relation Extraction?", author = "Xu, Jiashu and Ma, Mingyu Derek and Chen, Muhao", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.138", doi = "10.18653/v1/2023.acl-long.138", pages = "2450--2467", abstract = "Two key obstacles in biomedical relation extraction (RE) are the scarcity of annotations and the prevalence of instances without explicitly pre-defined labels due to low annotation coverage. Existing approaches, which treat biomedical RE as a multi-class classification task, often result in poor generalization in low-resource settings and do not have the ability to make selective prediction on unknown cases but give a guess from seen relations, hindering the applicability of those approaches. We present NBR, which converts biomedical RE as natural language inference formulation through indirect supervision. By converting relations to natural language hypotheses, NBR is capable of exploiting semantic cues to alleviate annotation scarcity. By incorporating a ranking-based loss that implicitly calibrates abstinent instances, NBR learns a clearer decision boundary and is instructed to abstain on uncertain instances. Extensive experiments on three widely-used biomedical RE benchmarks, namely ChemProt, DDI and GAD, verify the effectiveness of NBR in both full-set and low-resource regimes. Our analysis demonstrates that indirect supervision benefits biomedical RE even when a domain gap exists, and combining NLI knowledge with biomedical knowledge leads to the best performance gains.", }

Two key obstacles in biomedical relation extraction (RE) are the scarcity of annotations and the prevalence of instances without explicitly pre-defined labels due to low annotation coverage. Existing approaches, which treat biomedical RE as a multi-class classification task, often result in poor generalization in low-resource settings and do not have the ability to make selective prediction on unknown cases but give a guess from seen relations, hindering the applicability of those approaches. We present NBR, which converts biomedical RE as natural language inference formulation through indirect supervision. By converting relations to natural language hypotheses, NBR is capable of exploiting semantic cues to alleviate annotation scarcity. By incorporating a ranking-based loss that implicitly calibrates abstinent instances, NBR learns a clearer decision boundary and is instructed to abstain on uncertain instances. Extensive experiments on three widely-used biomedical RE benchmarks, namely ChemProt, DDI and GAD, verify the effectiveness of NBR in both full-set and low-resource regimes. Our analysis demonstrates that indirect supervision benefits biomedical RE even when a domain gap exists, and combining NLI knowledge with biomedical knowledge leads to the best performance gains.

Can NLI Provide Proper Indirect Supervision for Low-resource Biomedical Relation Extraction?

acl-long.138

Oral

2212.10784

https://huggingface.co/papers/2212.10784

https://aclanthology.org/2023.acl-long.139.bib

https://aclanthology.org/2023.acl-long.139/

@inproceedings{tian-etal-2023-dynamic, title = "Dynamic Routing Transformer Network for Multimodal Sarcasm Detection", author = "Tian, Yuan and Xu, Nan and Zhang, Ruike and Mao, Wenji", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.139", doi = "10.18653/v1/2023.acl-long.139", pages = "2468--2480", abstract = "Multimodal sarcasm detection is an important research topic in natural language processing and multimedia computing, and benefits a wide range of applications in multiple domains. Most existing studies regard the incongruity between image and text as the indicative clue in identifying multimodal sarcasm. To capture cross-modal incongruity, previous methods rely on fixed architectures in network design, which restricts the model from dynamically adjusting to diverse image-text pairs. Inspired by routing-based dynamic network, we model the dynamic mechanism in multimodal sarcasm detection and propose the Dynamic Routing Transformer Network (DynRT-Net). Our method utilizes dynamic paths to activate different routing transformer modules with hierarchical co-attention adapting to cross-modal incongruity. Experimental results on a public dataset demonstrate the effectiveness of our method compared to the state-of-the-art methods. Our codes are available at \url{https://github.com/TIAN-viola/DynRT}.", }

Multimodal sarcasm detection is an important research topic in natural language processing and multimedia computing, and benefits a wide range of applications in multiple domains. Most existing studies regard the incongruity between image and text as the indicative clue in identifying multimodal sarcasm. To capture cross-modal incongruity, previous methods rely on fixed architectures in network design, which restricts the model from dynamically adjusting to diverse image-text pairs. Inspired by routing-based dynamic network, we model the dynamic mechanism in multimodal sarcasm detection and propose the Dynamic Routing Transformer Network (DynRT-Net). Our method utilizes dynamic paths to activate different routing transformer modules with hierarchical co-attention adapting to cross-modal incongruity. Experimental results on a public dataset demonstrate the effectiveness of our method compared to the state-of-the-art methods. Our codes are available at \url{https://github.com/TIAN-viola/DynRT}.

Dynamic Routing Transformer Network for Multimodal Sarcasm Detection

acl-long.139

Poster

https://aclanthology.org/2023.acl-long.140.bib

https://aclanthology.org/2023.acl-long.140/

@inproceedings{ram-etal-2023-token, title = "What Are You Token About? Dense Retrieval as Distributions Over the Vocabulary", author = "Ram, Ori and Bezalel, Liat and Zicher, Adi and Belinkov, Yonatan and Berant, Jonathan and Globerson, Amir", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.140", doi = "10.18653/v1/2023.acl-long.140", pages = "2481--2498", abstract = "Dual encoders are now the dominant architecture for dense retrieval. Yet, we have little understanding of how they represent text, and why this leads to good performance. In this work, we shed light on this question via distributions over the vocabulary. We propose to interpret the vector representations produced by dual encoders by projecting them into the model{'}s vocabulary space. We show that the resulting projections contain rich semantic information, and draw connection between them and sparse retrieval. We find that this view can offer an explanation for some of the failure cases of dense retrievers. For example, we observe that the inability of models to handle tail entities is correlated with a tendency of the token distributions to forget some of the tokens of those entities. We leverage this insight and propose a simple way to enrich query and passage representations with lexical information at inference time, and show that this significantly improves performance compared to the original model in zero-shot settings, and specifically on the BEIR benchmark.", }

Dual encoders are now the dominant architecture for dense retrieval. Yet, we have little understanding of how they represent text, and why this leads to good performance. In this work, we shed light on this question via distributions over the vocabulary. We propose to interpret the vector representations produced by dual encoders by projecting them into the model{'}s vocabulary space. We show that the resulting projections contain rich semantic information, and draw connection between them and sparse retrieval. We find that this view can offer an explanation for some of the failure cases of dense retrievers. For example, we observe that the inability of models to handle tail entities is correlated with a tendency of the token distributions to forget some of the tokens of those entities. We leverage this insight and propose a simple way to enrich query and passage representations with lexical information at inference time, and show that this significantly improves performance compared to the original model in zero-shot settings, and specifically on the BEIR benchmark.

What Are You Token About? Dense Retrieval as Distributions Over the Vocabulary

acl-long.140

Oral

2212.10380

https://aclanthology.org/2023.acl-long.141.bib

https://aclanthology.org/2023.acl-long.141/

@inproceedings{yu-etal-2023-cold, title = "Cold-Start Data Selection for Better Few-shot Language Model Fine-tuning: A Prompt-based Uncertainty Propagation Approach", author = "Yu, Yue and Zhang, Rongzhi and Xu, Ran and Zhang, Jieyu and Shen, Jiaming and Zhang, Chao", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.141", doi = "10.18653/v1/2023.acl-long.141", pages = "2499--2521", abstract = "We present PATRON, a prompt-based data selection method for pre-trained language model fine-tuning under cold-start scenarios, i.e., no initial labeled data are available. In PATRON, we design (1) a prompt-based uncertainty propagation approach to estimate the importance of data points and (2) a partition-then-rewrite (PTR) strategy to promote sample diversity when querying for annotations. Experiments on six text classification datasets show that PATRON outperforms the strongest cold-start data selection baselines by up to 6.9{\%}. Besides, with 128 labels only, PATRON achieves 91.0{\%} and 92.1{\%} of the fully supervised performance based on vanilla fine-tuning and prompt-based learning respectively. Our implementation of PATRON will be published upon acceptance.", }

We present PATRON, a prompt-based data selection method for pre-trained language model fine-tuning under cold-start scenarios, i.e., no initial labeled data are available. In PATRON, we design (1) a prompt-based uncertainty propagation approach to estimate the importance of data points and (2) a partition-then-rewrite (PTR) strategy to promote sample diversity when querying for annotations. Experiments on six text classification datasets show that PATRON outperforms the strongest cold-start data selection baselines by up to 6.9{\%}. Besides, with 128 labels only, PATRON achieves 91.0{\%} and 92.1{\%} of the fully supervised performance based on vanilla fine-tuning and prompt-based learning respectively. Our implementation of PATRON will be published upon acceptance.

Cold-Start Data Selection for Better Few-shot Language Model Fine-tuning: A Prompt-based Uncertainty Propagation Approach

acl-long.141

Poster

https://aclanthology.org/2023.acl-long.142.bib

https://aclanthology.org/2023.acl-long.142/

@inproceedings{serianni-kalita-2023-training, title = "Training-free Neural Architecture Search for {RNN}s and Transformers", author = "Serianni, Aaron and Kalita, Jugal", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.142", doi = "10.18653/v1/2023.acl-long.142", pages = "2522--2540", abstract = "Neural architecture search (NAS) has allowed for the automatic creation of new and effective neural network architectures, offering an alternative to the laborious process of manually designing complex architectures. However, traditional NAS algorithms are slow and require immense amounts of computing power. Recent research has investigated training-free NAS metrics for image classification architectures, drastically speeding up search algorithms. In this paper, we investigate training-free NAS metrics for recurrent neural network (RNN) and BERT-based transformer architectures, targeted towards language modeling tasks. First, we develop a new training-free metric, named hidden covariance, that predicts the trained performance of an RNN architecture and significantly outperforms existing training-free metrics. We experimentally evaluate the effectiveness of the hidden covariance metric on the NAS-Bench-NLP benchmark. Second, we find that the current search space paradigm for transformer architectures is not optimized for training-free neural architecture search. Instead, a simple qualitative analysis can effectively shrink the search space to the best performing architectures. This conclusion is based on our investigation of existing training-free metrics and new metrics developed from recent transformer pruning literature, evaluated on our own benchmark of trained BERT architectures. Ultimately, our analysis shows that the architecture search space and the training-free metric must be developed together in order to achieve effective results. Our source code is available at \url{https://github.com/aaronserianni/training-free-nas}.", }

Neural architecture search (NAS) has allowed for the automatic creation of new and effective neural network architectures, offering an alternative to the laborious process of manually designing complex architectures. However, traditional NAS algorithms are slow and require immense amounts of computing power. Recent research has investigated training-free NAS metrics for image classification architectures, drastically speeding up search algorithms. In this paper, we investigate training-free NAS metrics for recurrent neural network (RNN) and BERT-based transformer architectures, targeted towards language modeling tasks. First, we develop a new training-free metric, named hidden covariance, that predicts the trained performance of an RNN architecture and significantly outperforms existing training-free metrics. We experimentally evaluate the effectiveness of the hidden covariance metric on the NAS-Bench-NLP benchmark. Second, we find that the current search space paradigm for transformer architectures is not optimized for training-free neural architecture search. Instead, a simple qualitative analysis can effectively shrink the search space to the best performing architectures. This conclusion is based on our investigation of existing training-free metrics and new metrics developed from recent transformer pruning literature, evaluated on our own benchmark of trained BERT architectures. Ultimately, our analysis shows that the architecture search space and the training-free metric must be developed together in order to achieve effective results. Our source code is available at \url{https://github.com/aaronserianni/training-free-nas}.

Training-free Neural Architecture Search for RNNs and Transformers

acl-long.142

Poster

2306.00288

https://aclanthology.org/2023.acl-long.143.bib

https://aclanthology.org/2023.acl-long.143/

@inproceedings{bhattacharjee-etal-2023-crosssum, title = "{C}ross{S}um: Beyond {E}nglish-Centric Cross-Lingual Summarization for 1,500+ Language Pairs", author = "Bhattacharjee, Abhik and Hasan, Tahmid and Ahmad, Wasi Uddin and Li, Yuan-Fang and Kang, Yong-Bin and Shahriyar, Rifat", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.143", doi = "10.18653/v1/2023.acl-long.143", pages = "2541--2564", abstract = "We present CrossSum, a large-scale cross-lingual summarization dataset comprising 1.68 million article-summary samples in 1,500+ language pairs. We create CrossSum by aligning parallel articles written in different languages via cross-lingual retrieval from a multilingual abstractive summarization dataset and perform a controlled human evaluation to validate its quality. We propose a multistage data sampling algorithm to effectively train a cross-lingual summarization model capable of summarizing an article in any target language. We also introduce LaSE, an embedding-based metric for automatically evaluating model-generated summaries. LaSE is strongly correlated with ROUGE and, unlike ROUGE, can be reliably measured even in the absence of references in the target language. Performance on ROUGE and LaSE indicate that our proposed model consistently outperforms baseline models. To the best of our knowledge, CrossSum is the largest cross-lingual summarization dataset and the first ever that is not centered around English. We are releasing the dataset, training and evaluation scripts, and models to spur future research on cross-lingual summarization. The resources can be found at \url{https://github.com/csebuetnlp/CrossSum}", }

We present CrossSum, a large-scale cross-lingual summarization dataset comprising 1.68 million article-summary samples in 1,500+ language pairs. We create CrossSum by aligning parallel articles written in different languages via cross-lingual retrieval from a multilingual abstractive summarization dataset and perform a controlled human evaluation to validate its quality. We propose a multistage data sampling algorithm to effectively train a cross-lingual summarization model capable of summarizing an article in any target language. We also introduce LaSE, an embedding-based metric for automatically evaluating model-generated summaries. LaSE is strongly correlated with ROUGE and, unlike ROUGE, can be reliably measured even in the absence of references in the target language. Performance on ROUGE and LaSE indicate that our proposed model consistently outperforms baseline models. To the best of our knowledge, CrossSum is the largest cross-lingual summarization dataset and the first ever that is not centered around English. We are releasing the dataset, training and evaluation scripts, and models to spur future research on cross-lingual summarization. The resources can be found at \url{https://github.com/csebuetnlp/CrossSum}

CrossSum: Beyond English-Centric Cross-Lingual Summarization for 1,500+ Language Pairs

acl-long.143

Poster

2112.08804

https://huggingface.co/papers/2112.08804

https://aclanthology.org/2023.acl-long.144.bib

https://aclanthology.org/2023.acl-long.144/

@inproceedings{chai-etal-2023-improving, title = "Improving Gradient Trade-offs between Tasks in Multi-task Text Classification", author = "Chai, Heyan and Cui, Jinhao and Wang, Ye and Zhang, Min and Fang, Binxing and Liao, Qing", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.144", doi = "10.18653/v1/2023.acl-long.144", pages = "2565--2579", abstract = "Multi-task learning (MTL) has emerged as a promising approach for sharing inductive bias across multiple tasks to enable more efficient learning in text classification. However, training all tasks simultaneously often yields degraded performance of each task than learning them independently, since different tasks might conflict with each other. Existing MTL methods for alleviating this issue is to leverage heuristics or gradient-based algorithm to achieve an arbitrary Pareto optimal trade-off among different tasks. In this paper, we present a novel gradient trade-off approach to mitigate the task conflict problem, dubbed GetMTL, which can achieve a specific trade-off among different tasks nearby the main objective of multi-task text classification (MTC), so as to improve the performance of each task simultaneously. The results of extensive experiments on two benchmark datasets back up our theoretical analysis and validate the superiority of our proposed GetMTL.", }

Multi-task learning (MTL) has emerged as a promising approach for sharing inductive bias across multiple tasks to enable more efficient learning in text classification. However, training all tasks simultaneously often yields degraded performance of each task than learning them independently, since different tasks might conflict with each other. Existing MTL methods for alleviating this issue is to leverage heuristics or gradient-based algorithm to achieve an arbitrary Pareto optimal trade-off among different tasks. In this paper, we present a novel gradient trade-off approach to mitigate the task conflict problem, dubbed GetMTL, which can achieve a specific trade-off among different tasks nearby the main objective of multi-task text classification (MTC), so as to improve the performance of each task simultaneously. The results of extensive experiments on two benchmark datasets back up our theoretical analysis and validate the superiority of our proposed GetMTL.

Improving Gradient Trade-offs between Tasks in Multi-task Text Classification

acl-long.144

Poster

https://aclanthology.org/2023.acl-long.145.bib

https://aclanthology.org/2023.acl-long.145/

@inproceedings{gupta-etal-2023-bi, title = "Bi-Phone: Modeling Inter Language Phonetic Influences in Text", author = "Gupta, Abhirut and Sai, Ananya B. and Sproat, Richard and Vasilevski, Yuri and Ren, James and Jash, Ambarish and Sodhi, Sukhdeep and Raghuveer, Aravindan", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.145", doi = "10.18653/v1/2023.acl-long.145", pages = "2580--2592", abstract = "A large number of people are forced to use the Web in a language they have low literacy in due to technology asymmetries. Written text in the second language (L2) from such users often contains a large number of errors that are influenced by their native language (L1).We propose a method to mine phoneme confusions (sounds in L2 that an L1 speaker is likely to conflate) for pairs of L1 and L2.These confusions are then plugged into a generative model (Bi-Phone) for synthetically producing corrupted L2 text. Through human evaluations, we show that Bi-Phone generates plausible corruptions that differ across L1s and also have widespread coverage on the Web.We also corrupt the popular language understanding benchmark SuperGLUE with our technique (FunGLUE for Phonetically Noised GLUE) and show that SoTA language understating models perform poorly. We also introduce a new phoneme prediction pre-training task which helps byte models to recover performance close to SuperGLUE. Finally, we also release the SuperGLUE benchmark to promote further research in phonetically robust language models. To the best of our knowledge, FunGLUE is the first benchmark to introduce L1-L2 interactions in text.", }

A large number of people are forced to use the Web in a language they have low literacy in due to technology asymmetries. Written text in the second language (L2) from such users often contains a large number of errors that are influenced by their native language (L1).We propose a method to mine phoneme confusions (sounds in L2 that an L1 speaker is likely to conflate) for pairs of L1 and L2.These confusions are then plugged into a generative model (Bi-Phone) for synthetically producing corrupted L2 text. Through human evaluations, we show that Bi-Phone generates plausible corruptions that differ across L1s and also have widespread coverage on the Web.We also corrupt the popular language understanding benchmark SuperGLUE with our technique (FunGLUE for Phonetically Noised GLUE) and show that SoTA language understating models perform poorly. We also introduce a new phoneme prediction pre-training task which helps byte models to recover performance close to SuperGLUE. Finally, we also release the SuperGLUE benchmark to promote further research in phonetically robust language models. To the best of our knowledge, FunGLUE is the first benchmark to introduce L1-L2 interactions in text.

Bi-Phone: Modeling Inter Language Phonetic Influences in Text

acl-long.145

Oral

https://aclanthology.org/2023.acl-long.146.bib

https://aclanthology.org/2023.acl-long.146/

@inproceedings{wu-etal-2023-cross2stra, title = "{C}ross2{S}tr{A}: Unpaired Cross-lingual Image Captioning with Cross-lingual Cross-modal Structure-pivoted Alignment", author = "Wu, Shengqiong and Fei, Hao and Ji, Wei and Chua, Tat-Seng", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.146", doi = "10.18653/v1/2023.acl-long.146", pages = "2593--2608", abstract = "Unpaired cross-lingual image captioning has long suffered from irrelevancy and disfluency issues, due to the inconsistencies of the semantic scene and syntax attributes during transfer. In this work, we propose to address the above problems by incorporating the scene graph (SG) structures and the syntactic constituency (SC) trees. Our captioner contains the semantic structure-guided image-to-pivot captioning and the syntactic structure-guided pivot-to-target translation, two of which are joined via pivot language. We then take the SG and SC structures as pivoting, performing cross-modal semantic structure alignment and cross-lingual syntactic structure alignment learning. We further introduce cross-lingual{\&}cross-modal back-translation training to fully align the captioning and translation stages. Experiments on English-Chinese transfers show that our model shows great superiority in improving captioning relevancy and fluency.", }

Unpaired cross-lingual image captioning has long suffered from irrelevancy and disfluency issues, due to the inconsistencies of the semantic scene and syntax attributes during transfer. In this work, we propose to address the above problems by incorporating the scene graph (SG) structures and the syntactic constituency (SC) trees. Our captioner contains the semantic structure-guided image-to-pivot captioning and the syntactic structure-guided pivot-to-target translation, two of which are joined via pivot language. We then take the SG and SC structures as pivoting, performing cross-modal semantic structure alignment and cross-lingual syntactic structure alignment learning. We further introduce cross-lingual{\&}cross-modal back-translation training to fully align the captioning and translation stages. Experiments on English-Chinese transfers show that our model shows great superiority in improving captioning relevancy and fluency.

Cross2StrA: Unpaired Cross-lingual Image Captioning with Cross-lingual Cross-modal Structure-pivoted Alignment

acl-long.146

Oral

2305.12260

https://aclanthology.org/2023.acl-long.147.bib

https://aclanthology.org/2023.acl-long.147/

@inproceedings{wang-etal-2023-plan, title = "Plan-and-Solve Prompting: Improving Zero-Shot Chain-of-Thought Reasoning by Large Language Models", author = "Wang, Lei and Xu, Wanyu and Lan, Yihuai and Hu, Zhiqiang and Lan, Yunshi and Lee, Roy Ka-Wei and Lim, Ee-Peng", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.147", doi = "10.18653/v1/2023.acl-long.147", pages = "2609--2634", abstract = "Large language models (LLMs) have recently been shown to deliver impressive performance in various NLP tasks. To tackle multi-step reasoning tasks, Few-shot chain-of-thought (CoT) prompting includes a few manually crafted step-by-step reasoning demonstrations which enable LLMs to explicitly generate reasoning steps and improve their reasoning task accuracy. To eliminate the manual efforts, Zero-shot-CoT concatenates the target problem statement with {``}\textit{Let{'}s think step by step}{''} as an input prompt to LLMs. Despite the success of Zero-shot-CoT, it still suffers from three pitfalls: calculation errors, missing-step errors, and semantic misunderstanding errors. To address the missing-step errors, we propose Plan-and-Solve (PS) Prompting. It consists of two components: first, devising a plan to divide the entire task into smaller subtasks, and then carrying out the subtasks according to the plan. To address the calculation errors and improve the quality of generated reasoning steps, we extend PS prompting with more detailed instructions and derive PS+ prompting. We evaluate our proposed prompting strategy on ten datasets across three reasoning problems. The experimental results over GPT-3 show that our proposed zero-shot prompting consistently outperforms Zero-shot-CoT across all datasets by a large margin, is comparable to or exceeds Zero-shot-Program-of-Thought Prompting, and has comparable performance with 8-shot CoT prompting on the math reasoning problem. The code can be found at \url{https://github.com/AGI-Edgerunners/Plan-and-Solve-Prompting}.", }

Large language models (LLMs) have recently been shown to deliver impressive performance in various NLP tasks. To tackle multi-step reasoning tasks, Few-shot chain-of-thought (CoT) prompting includes a few manually crafted step-by-step reasoning demonstrations which enable LLMs to explicitly generate reasoning steps and improve their reasoning task accuracy. To eliminate the manual efforts, Zero-shot-CoT concatenates the target problem statement with {``}\textit{Let{'}s think step by step}{''} as an input prompt to LLMs. Despite the success of Zero-shot-CoT, it still suffers from three pitfalls: calculation errors, missing-step errors, and semantic misunderstanding errors. To address the missing-step errors, we propose Plan-and-Solve (PS) Prompting. It consists of two components: first, devising a plan to divide the entire task into smaller subtasks, and then carrying out the subtasks according to the plan. To address the calculation errors and improve the quality of generated reasoning steps, we extend PS prompting with more detailed instructions and derive PS+ prompting. We evaluate our proposed prompting strategy on ten datasets across three reasoning problems. The experimental results over GPT-3 show that our proposed zero-shot prompting consistently outperforms Zero-shot-CoT across all datasets by a large margin, is comparable to or exceeds Zero-shot-Program-of-Thought Prompting, and has comparable performance with 8-shot CoT prompting on the math reasoning problem. The code can be found at \url{https://github.com/AGI-Edgerunners/Plan-and-Solve-Prompting}.

Plan-and-Solve Prompting: Improving Zero-Shot Chain-of-Thought Reasoning by Large Language Models

acl-long.147

Poster

2305.04091

https://aclanthology.org/2023.acl-long.148.bib

https://aclanthology.org/2023.acl-long.148/

@inproceedings{liu-etal-2023-retromae, title = "{R}etro{MAE}-2: Duplex Masked Auto-Encoder For Pre-Training Retrieval-Oriented Language Models", author = "Liu, Zheng and Xiao, Shitao and Shao, Yingxia and Cao, Zhao", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.148", doi = "10.18653/v1/2023.acl-long.148", pages = "2635--2648", abstract = "To better support information retrieval tasks such as web search and open-domain question answering, growing effort is made to develop retrieval-oriented language models, e.g., RetroMAE and many others. Most of the existing works focus on improving the semantic representation capability for the contextualized embedding of the [CLS] token. However, recent study shows that the ordinary tokens besides [CLS] may provide extra information, which help to produce a better representation effect. As such, it{'}s necessary to extend the current methods where all contextualized embeddings can be jointly pre-trained for the retrieval tasks. In this work, we propose a novel pre-training method called Duplex Masked Auto-Encoder, a.k.a. DupMAE. It is designed to improve the quality of semantic representation where all contextualized embeddings of the pre-trained model can be leveraged. It takes advantage of two complementary auto-encoding tasks: one reconstructs the input sentence on top of the [CLS] embedding; the other one predicts the bag-of-words feature of the input sentence based on the ordinary tokens{'} embeddings. The two tasks are jointly conducted to train a unified encoder, where the whole contextualized embeddings are aggregated in a compact way to produce the final semantic representation. DupMAE is simple but empirically competitive: it substantially improves the pre-trained model{'}s representation capability and transferability, where superior retrieval performances can be achieved on popular benchmarks, like MS MARCO and BEIR. We make our code publicly available at \url{https://github.com/staoxiao/RetroMAE}.", }

To better support information retrieval tasks such as web search and open-domain question answering, growing effort is made to develop retrieval-oriented language models, e.g., RetroMAE and many others. Most of the existing works focus on improving the semantic representation capability for the contextualized embedding of the [CLS] token. However, recent study shows that the ordinary tokens besides [CLS] may provide extra information, which help to produce a better representation effect. As such, it{'}s necessary to extend the current methods where all contextualized embeddings can be jointly pre-trained for the retrieval tasks. In this work, we propose a novel pre-training method called Duplex Masked Auto-Encoder, a.k.a. DupMAE. It is designed to improve the quality of semantic representation where all contextualized embeddings of the pre-trained model can be leveraged. It takes advantage of two complementary auto-encoding tasks: one reconstructs the input sentence on top of the [CLS] embedding; the other one predicts the bag-of-words feature of the input sentence based on the ordinary tokens{'} embeddings. The two tasks are jointly conducted to train a unified encoder, where the whole contextualized embeddings are aggregated in a compact way to produce the final semantic representation. DupMAE is simple but empirically competitive: it substantially improves the pre-trained model{'}s representation capability and transferability, where superior retrieval performances can be achieved on popular benchmarks, like MS MARCO and BEIR. We make our code publicly available at \url{https://github.com/staoxiao/RetroMAE}.

RetroMAE-2: Duplex Masked Auto-Encoder For Pre-Training Retrieval-Oriented Language Models

acl-long.148

Oral

2305.02564

https://aclanthology.org/2023.acl-long.149.bib

https://aclanthology.org/2023.acl-long.149/

@inproceedings{modarressi-etal-2023-decompx, title = "{D}ecomp{X}: Explaining Transformers Decisions by Propagating Token Decomposition", author = "Modarressi, Ali and Fayyaz, Mohsen and Aghazadeh, Ehsan and Yaghoobzadeh, Yadollah and Pilehvar, Mohammad Taher", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.149", doi = "10.18653/v1/2023.acl-long.149", pages = "2649--2664", abstract = "An emerging solution for explaining Transformer-based models is to use vector-based analysis on how the representations are formed. However, providing a faithful vector-based explanation for a multi-layer model could be challenging in three aspects: (1) Incorporating all components into the analysis, (2) Aggregating the layer dynamics to determine the information flow and mixture throughout the entire model, and (3) Identifying the connection between the vector-based analysis and the model{'}s predictions. In this paper, we present DecompX to tackle these challenges. DecompX is based on the construction of decomposed token representations and their successive propagation throughout the model without mixing them in between layers. Additionally, our proposal provides multiple advantages over existing solutions for its inclusion of all encoder components (especially nonlinear feed-forward networks) and the classification head. The former allows acquiring precise vectors while the latter transforms the decomposition into meaningful prediction-based values, eliminating the need for norm- or summation-based vector aggregation. According to the standard faithfulness evaluations, DecompX consistently outperforms existing gradient-based and vector-based approaches on various datasets. Our code is available at \url{https://github.com/mohsenfayyaz/DecompX}.", }

An emerging solution for explaining Transformer-based models is to use vector-based analysis on how the representations are formed. However, providing a faithful vector-based explanation for a multi-layer model could be challenging in three aspects: (1) Incorporating all components into the analysis, (2) Aggregating the layer dynamics to determine the information flow and mixture throughout the entire model, and (3) Identifying the connection between the vector-based analysis and the model{'}s predictions. In this paper, we present DecompX to tackle these challenges. DecompX is based on the construction of decomposed token representations and their successive propagation throughout the model without mixing them in between layers. Additionally, our proposal provides multiple advantages over existing solutions for its inclusion of all encoder components (especially nonlinear feed-forward networks) and the classification head. The former allows acquiring precise vectors while the latter transforms the decomposition into meaningful prediction-based values, eliminating the need for norm- or summation-based vector aggregation. According to the standard faithfulness evaluations, DecompX consistently outperforms existing gradient-based and vector-based approaches on various datasets. Our code is available at \url{https://github.com/mohsenfayyaz/DecompX}.

DecompX: Explaining Transformers Decisions by Propagating Token Decomposition

acl-long.149

Poster

2306.02873

https://huggingface.co/papers/2306.02873

https://aclanthology.org/2023.acl-long.150.bib

https://aclanthology.org/2023.acl-long.150/

@inproceedings{li-etal-2023-symbolic, title = "Symbolic Chain-of-Thought Distillation: Small Models Can Also {``}Think{''} Step-by-Step", author = "Li, Liunian Harold and Hessel, Jack and Yu, Youngjae and Ren, Xiang and Chang, Kai-Wei and Choi, Yejin", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.150", doi = "10.18653/v1/2023.acl-long.150", pages = "2665--2679", abstract = "Chain-of-thought prompting (e.g., {``}Let{'}s think step-by-ste{''}) primes large language models to verbalize rationalization for their predictions. While chain-of-thought can lead to dramatic performance gains, benefits appear to emerge only for sufficiently large models (beyond 50B parameters). We show that orders-of-magnitude smaller models (125M{---}1.3B parameters) can still benefit from chain-of-thought prompting. To achieve this, we introduce Symbolic Chain-of-Thought Distillation (SCoTD), a method to train a smaller student model on rationalizations sampled from a significantly larger teacher model. Experiments across several commonsense benchmarks show that: 1) SCoTD enhances the performance of the student model in both supervised and few-shot settings, and especially for challenge sets; 2) sampling many reasoning chains per instance from the teacher is paramount; and 3) after distillation, student chain-of-thoughts are judged by humans as comparable to the teacher, despite orders of magnitude fewer parameters. We test several hypotheses regarding what properties of chain-of-thought samples are important, e.g., diversity vs. teacher likelihood vs. open-endedness. We release our corpus of chain-of-thought samples and code.", }

Chain-of-thought prompting (e.g., {``}Let{'}s think step-by-ste{''}) primes large language models to verbalize rationalization for their predictions. While chain-of-thought can lead to dramatic performance gains, benefits appear to emerge only for sufficiently large models (beyond 50B parameters). We show that orders-of-magnitude smaller models (125M{---}1.3B parameters) can still benefit from chain-of-thought prompting. To achieve this, we introduce Symbolic Chain-of-Thought Distillation (SCoTD), a method to train a smaller student model on rationalizations sampled from a significantly larger teacher model. Experiments across several commonsense benchmarks show that: 1) SCoTD enhances the performance of the student model in both supervised and few-shot settings, and especially for challenge sets; 2) sampling many reasoning chains per instance from the teacher is paramount; and 3) after distillation, student chain-of-thoughts are judged by humans as comparable to the teacher, despite orders of magnitude fewer parameters. We test several hypotheses regarding what properties of chain-of-thought samples are important, e.g., diversity vs. teacher likelihood vs. open-endedness. We release our corpus of chain-of-thought samples and code.

Symbolic Chain-of-Thought Distillation: Small Models Can Also “Think” Step-by-Step

acl-long.150

Poster

https://aclanthology.org/2023.acl-long.151.bib

https://aclanthology.org/2023.acl-long.151/

@inproceedings{adams-etal-2023-generating, title = "Generating {EDU} Extracts for Plan-Guided Summary Re-Ranking", author = "Adams, Griffin and Fabbri, Alex and Ladhak, Faisal and Elhadad, No{\'e}mie and McKeown, Kathleen", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.151", doi = "10.18653/v1/2023.acl-long.151", pages = "2680--2697", abstract = "Two-step approaches, in which summary candidates are generated-then-reranked to return a single summary, can improve ROUGE scores over the standard single-step approach. Yet, standard decoding methods (i.e., beam search, nucleus sampling, and diverse beam search) produce candidates with redundant, and often low quality, content. In this paper, we design a novel method to generate candidates for re-ranking that addresses these issues. We ground each candidate abstract on its own unique content plan and generate distinct plan-guided abstracts using a model{'}s top beam. More concretely, a standard language model (a BART LM) auto-regressively generates elemental discourse unit (EDU) content plans with an extractive copy mechanism. The top K beams from the content plan generator are then used to guide a separate LM, which produces a single abstractive candidate for each distinct plan. We apply an existing re-ranker (BRIO) to abstractive candidates generated from our method, as well as baseline decoding methods. We show large relevance improvements over previously published methods on widely used single document news article corpora, with ROUGE-2 F1 gains of 0.88, 2.01, and 0.38 on CNN / Dailymail, NYT, and Xsum, respectively. A human evaluation on CNN / DM validates these results. Similarly, on 1k samples from CNN / DM, we show that prompting GPT-3 to follow EDU plans outperforms sampling-based methods by by 1.05 ROUGE-2 F1 points. Code to generate and realize plans is available at \url{https://github.com/griff4692/edu-sum}.", }

Two-step approaches, in which summary candidates are generated-then-reranked to return a single summary, can improve ROUGE scores over the standard single-step approach. Yet, standard decoding methods (i.e., beam search, nucleus sampling, and diverse beam search) produce candidates with redundant, and often low quality, content. In this paper, we design a novel method to generate candidates for re-ranking that addresses these issues. We ground each candidate abstract on its own unique content plan and generate distinct plan-guided abstracts using a model{'}s top beam. More concretely, a standard language model (a BART LM) auto-regressively generates elemental discourse unit (EDU) content plans with an extractive copy mechanism. The top K beams from the content plan generator are then used to guide a separate LM, which produces a single abstractive candidate for each distinct plan. We apply an existing re-ranker (BRIO) to abstractive candidates generated from our method, as well as baseline decoding methods. We show large relevance improvements over previously published methods on widely used single document news article corpora, with ROUGE-2 F1 gains of 0.88, 2.01, and 0.38 on CNN / Dailymail, NYT, and Xsum, respectively. A human evaluation on CNN / DM validates these results. Similarly, on 1k samples from CNN / DM, we show that prompting GPT-3 to follow EDU plans outperforms sampling-based methods by by 1.05 ROUGE-2 F1 points. Code to generate and realize plans is available at \url{https://github.com/griff4692/edu-sum}.

Generating EDU Extracts for Plan-Guided Summary Re-Ranking

acl-long.151

Poster

2305.17779

https://aclanthology.org/2023.acl-long.152.bib

https://aclanthology.org/2023.acl-long.152/

@inproceedings{rahmani-etal-2023-survey, title = "A Survey on Asking Clarification Questions Datasets in Conversational Systems", author = "Rahmani, Hossein A. and Wang, Xi and Feng, Yue and Zhang, Qiang and Yilmaz, Emine and Lipani, Aldo", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.152", doi = "10.18653/v1/2023.acl-long.152", pages = "2698--2716", abstract = "The ability to understand a user{'}s underlying needs is critical for conversational systems, especially with limited input from users in a conversation. Thus, in such a domain, Asking Clarification Questions (ACQs) to reveal users{'} true intent from their queries or utterances arise as an essential task. However, it is noticeable that a key limitation of the existing ACQs studies is their incomparability, from inconsistent use of data, distinct experimental setups and evaluation strategies. Therefore, in this paper, to assist the development of ACQs techniques, we comprehensively analyse the current ACQs research status, which offers a detailed comparison of publicly available datasets, and discusses the applied evaluation metrics, joined with benchmarks for multiple ACQs-related tasks. In particular, given a thorough analysis of the ACQs task, we discuss a number of corresponding research directions for the investigation of ACQs as well as the development of conversational systems.", }

The ability to understand a user{'}s underlying needs is critical for conversational systems, especially with limited input from users in a conversation. Thus, in such a domain, Asking Clarification Questions (ACQs) to reveal users{'} true intent from their queries or utterances arise as an essential task. However, it is noticeable that a key limitation of the existing ACQs studies is their incomparability, from inconsistent use of data, distinct experimental setups and evaluation strategies. Therefore, in this paper, to assist the development of ACQs techniques, we comprehensively analyse the current ACQs research status, which offers a detailed comparison of publicly available datasets, and discusses the applied evaluation metrics, joined with benchmarks for multiple ACQs-related tasks. In particular, given a thorough analysis of the ACQs task, we discuss a number of corresponding research directions for the investigation of ACQs as well as the development of conversational systems.

A Survey on Asking Clarification Questions Datasets in Conversational Systems

acl-long.152

Poster

2305.15933

https://aclanthology.org/2023.acl-long.153.bib

https://aclanthology.org/2023.acl-long.153/

@inproceedings{wang-etal-2023-towards, title = "Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters", author = "Wang, Boshi and Min, Sewon and Deng, Xiang and Shen, Jiaming and Wu, You and Zettlemoyer, Luke and Sun, Huan", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.153", doi = "10.18653/v1/2023.acl-long.153", pages = "2717--2739", abstract = "Chain-of-Thought (CoT) prompting can dramatically improve the multi-step reasoning abilities of large language models (LLMs). CoT explicitly encourages the LLM to generate intermediate rationales for solving a problem, by providing a series of reasoning steps in the demonstrations. Despite its success, there is still little understanding of what makes CoT prompting effective and which aspects of the demonstrated reasoning steps contribute to its performance. In this paper, we show that CoT reasoning is possible even with invalid demonstrations - prompting with invalid reasoning steps can achieve over 80-90{\%} of the performance obtained using CoT under various metrics, while still generating coherent lines of reasoning during inference. Further experiments show that other aspects of the rationales, such as being relevant to the query and correctly ordering the reasoning steps, are much more important for effective CoT reasoning. Overall, these findings both deepen our understanding of CoT prompting, and open up new questions regarding LLMs{'} capability to learn to reason in context.", }

Chain-of-Thought (CoT) prompting can dramatically improve the multi-step reasoning abilities of large language models (LLMs). CoT explicitly encourages the LLM to generate intermediate rationales for solving a problem, by providing a series of reasoning steps in the demonstrations. Despite its success, there is still little understanding of what makes CoT prompting effective and which aspects of the demonstrated reasoning steps contribute to its performance. In this paper, we show that CoT reasoning is possible even with invalid demonstrations - prompting with invalid reasoning steps can achieve over 80-90{\%} of the performance obtained using CoT under various metrics, while still generating coherent lines of reasoning during inference. Further experiments show that other aspects of the rationales, such as being relevant to the query and correctly ordering the reasoning steps, are much more important for effective CoT reasoning. Overall, these findings both deepen our understanding of CoT prompting, and open up new questions regarding LLMs{'} capability to learn to reason in context.

Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters

acl-long.153

Poster

2212.10001

https://huggingface.co/papers/2212.10001

https://aclanthology.org/2023.acl-long.154.bib

https://aclanthology.org/2023.acl-long.154/

@inproceedings{maillard-etal-2023-small, title = "Small Data, Big Impact: Leveraging Minimal Data for Effective Machine Translation", author = "Maillard, Jean and Gao, Cynthia and Kalbassi, Elahe and Sadagopan, Kaushik Ram and Goswami, Vedanuj and Koehn, Philipp and Fan, Angela and Guzman, Francisco", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.154", doi = "10.18653/v1/2023.acl-long.154", pages = "2740--2756", abstract = "For many languages, machine translation progress is hindered by the lack of reliable training data. Models are trained on whatever pre-existing datasets may be available and then augmented with synthetic data, because it is often not economical to pay for the creation of large-scale datasets. But for the case of low-resource languages, would the creation of a few thousand professionally translated sentence pairs give any benefit? In this paper, we show that it does. We describe a broad data collection effort involving around 6k professionally translated sentence pairs for each of 39 low-resource languages, which we make publicly available. We analyse the gains of models trained on this small but high-quality data, showing that it has significant impact even when larger but lower quality pre-existing corpora are used, or when data is augmented with millions of sentences through backtranslation.", }

For many languages, machine translation progress is hindered by the lack of reliable training data. Models are trained on whatever pre-existing datasets may be available and then augmented with synthetic data, because it is often not economical to pay for the creation of large-scale datasets. But for the case of low-resource languages, would the creation of a few thousand professionally translated sentence pairs give any benefit? In this paper, we show that it does. We describe a broad data collection effort involving around 6k professionally translated sentence pairs for each of 39 low-resource languages, which we make publicly available. We analyse the gains of models trained on this small but high-quality data, showing that it has significant impact even when larger but lower quality pre-existing corpora are used, or when data is augmented with millions of sentences through backtranslation.

Small Data, Big Impact: Leveraging Minimal Data for Effective Machine Translation

acl-long.154

Oral

https://aclanthology.org/2023.acl-long.155.bib

https://aclanthology.org/2023.acl-long.155/

@inproceedings{wang-etal-2023-rmlm, title = "{RMLM}: A Flexible Defense Framework for Proactively Mitigating Word-level Adversarial Attacks", author = "Wang, Zhaoyang and Liu, Zhiyue and Zheng, Xiaopeng and Su, Qinliang and Wang, Jiahai", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.155", doi = "10.18653/v1/2023.acl-long.155", pages = "2757--2774", abstract = "Adversarial attacks on deep neural networks keep raising security concerns in natural language processing research. Existing defenses focus on improving the robustness of the victim model in the training stage. However, they often neglect to proactively mitigate adversarial attacks during inference. Towards this overlooked aspect, we propose a defense framework that aims to mitigate attacks by confusing attackers and correcting adversarial contexts that are caused by malicious perturbations. Our framework comprises three components: (1) a synonym-based transformation to randomly corrupt adversarial contexts in the word level, (2) a developed BERT defender to correct abnormal contexts in the representation level, and (3) a simple detection method to filter out adversarial examples, any of which can be flexibly combined. Additionally, our framework helps improve the robustness of the victim model during training. Extensive experiments demonstrate the effectiveness of our framework in defending against word-level adversarial attacks.", }

Adversarial attacks on deep neural networks keep raising security concerns in natural language processing research. Existing defenses focus on improving the robustness of the victim model in the training stage. However, they often neglect to proactively mitigate adversarial attacks during inference. Towards this overlooked aspect, we propose a defense framework that aims to mitigate attacks by confusing attackers and correcting adversarial contexts that are caused by malicious perturbations. Our framework comprises three components: (1) a synonym-based transformation to randomly corrupt adversarial contexts in the word level, (2) a developed BERT defender to correct abnormal contexts in the representation level, and (3) a simple detection method to filter out adversarial examples, any of which can be flexibly combined. Additionally, our framework helps improve the robustness of the victim model during training. Extensive experiments demonstrate the effectiveness of our framework in defending against word-level adversarial attacks.

RMLM: A Flexible Defense Framework for Proactively Mitigating Word-level Adversarial Attacks

acl-long.155

Poster

https://aclanthology.org/2023.acl-long.156.bib

https://aclanthology.org/2023.acl-long.156/

@inproceedings{yang-etal-2023-gradient, title = "Gradient-based Intra-attention Pruning on Pre-trained Language Models", author = "Yang, Ziqing and Cui, Yiming and Yao, Xin and Wang, Shijin", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.156", doi = "10.18653/v1/2023.acl-long.156", pages = "2775--2790", abstract = "Pre-trained language models achieve superior performance but are computationally expensive. Techniques such as pruning and knowledge distillation have been developed to reduce their sizes and latencies. In this work, we propose a structured pruning method GRAIN (gradient-based intra-attention pruning), which performs task-specific pruning with knowledge distillation and yields highly effective models. Different from common approaches that prune each attention head as a whole, GRAIN inspects and prunes intra-attention structures, which greatly expands the structure search space and enables more flexible models. We also propose a gradient separation strategy that reduces the interference of distillation on pruning for a better combination of the two approaches. Experiments on GLUE, SQuAD, and CoNLL 2003 show that GRAIN notably outperforms other methods, especially in the high sparsity regime, and achieves 6 7x speedups while maintaining 93{\%} 99{\%} performance. Under extreme compression where only 3{\%} transformer weights remain, the pruned model is still competitive compared to larger models.", }

Pre-trained language models achieve superior performance but are computationally expensive. Techniques such as pruning and knowledge distillation have been developed to reduce their sizes and latencies. In this work, we propose a structured pruning method GRAIN (gradient-based intra-attention pruning), which performs task-specific pruning with knowledge distillation and yields highly effective models. Different from common approaches that prune each attention head as a whole, GRAIN inspects and prunes intra-attention structures, which greatly expands the structure search space and enables more flexible models. We also propose a gradient separation strategy that reduces the interference of distillation on pruning for a better combination of the two approaches. Experiments on GLUE, SQuAD, and CoNLL 2003 show that GRAIN notably outperforms other methods, especially in the high sparsity regime, and achieves 6 7x speedups while maintaining 93{\%} 99{\%} performance. Under extreme compression where only 3{\%} transformer weights remain, the pruned model is still competitive compared to larger models.

Gradient-based Intra-attention Pruning on Pre-trained Language Models

acl-long.156

Poster

2212.07634

https://aclanthology.org/2023.acl-long.157.bib

https://aclanthology.org/2023.acl-long.157/

@inproceedings{li-etal-2023-learning, title = "Learning to Substitute Spans towards Improving Compositional Generalization", author = "Li, Zhaoyi and Wei, Ying and Lian, Defu", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.157", doi = "10.18653/v1/2023.acl-long.157", pages = "2791--2811", abstract = "Despite the rising prevalence of neural sequence models, recent empirical evidences suggest their deficiency in compositional generalization. One of the current de-facto solutions to this problem is compositional data augmentation, aiming to incur additional compositional inductive bias. Nonetheless, the improvement offered by existing handcrafted augmentation strategies is limited when successful systematic generalization of neural sequence models requires multi-grained compositional bias (i.e., not limited to either lexical or structural biases only) or differentiation of training sequences in an imbalanced difficulty distribution. To address the two challenges, we first propose a novel compositional augmentation strategy dubbed Span Substitution (SpanSub) that enables multi-grained composition of substantial substructures in the whole training set. Over and above that, we introduce the Learning to Substitute Span (L2S2) framework which empowers the learning of span substitution probabilities in SpanSub in an end-to-end manner by maximizing the loss of neural sequence models, so as to outweigh those challenging compositions with elusive concepts and novel surroundings. Our empirical results on three standard compositional generalization benchmarks, including SCAN, COGS and GeoQuery (with an improvement of at most 66.5{\%}, 10.3{\%}, 1.2{\%}, respectively), demonstrate the superiority of SpanSub, L2S2 and their combination.", }

Despite the rising prevalence of neural sequence models, recent empirical evidences suggest their deficiency in compositional generalization. One of the current de-facto solutions to this problem is compositional data augmentation, aiming to incur additional compositional inductive bias. Nonetheless, the improvement offered by existing handcrafted augmentation strategies is limited when successful systematic generalization of neural sequence models requires multi-grained compositional bias (i.e., not limited to either lexical or structural biases only) or differentiation of training sequences in an imbalanced difficulty distribution. To address the two challenges, we first propose a novel compositional augmentation strategy dubbed Span Substitution (SpanSub) that enables multi-grained composition of substantial substructures in the whole training set. Over and above that, we introduce the Learning to Substitute Span (L2S2) framework which empowers the learning of span substitution probabilities in SpanSub in an end-to-end manner by maximizing the loss of neural sequence models, so as to outweigh those challenging compositions with elusive concepts and novel surroundings. Our empirical results on three standard compositional generalization benchmarks, including SCAN, COGS and GeoQuery (with an improvement of at most 66.5{\%}, 10.3{\%}, 1.2{\%}, respectively), demonstrate the superiority of SpanSub, L2S2 and their combination.

Learning to Substitute Spans towards Improving Compositional Generalization

acl-long.157

Oral

2306.02840

https://aclanthology.org/2023.acl-long.158.bib

https://aclanthology.org/2023.acl-long.158/

@inproceedings{bi-etal-2023-diffusemp, title = "{D}iffus{E}mp: A Diffusion Model-Based Framework with Multi-Grained Control for Empathetic Response Generation", author = "Bi, Guanqun and Shen, Lei and Cao, Yanan and Chen, Meng and Xie, Yuqiang and Lin, Zheng and He, Xiaodong", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.158", doi = "10.18653/v1/2023.acl-long.158", pages = "2812--2831", abstract = "Empathy is a crucial factor in open-domain conversations, which naturally shows one{'}s caring and understanding to others. Though several methods have been proposed to generate empathetic responses, existing works often lead to monotonous empathy that refers to generic and safe expressions. In this paper, we propose to use explicit control to guide the empathy expression and design a framework DiffusEmp based on conditional diffusion language model to unify the utilization of dialogue context and attribute-oriented control signals. Specifically, communication mechanism, intent, and semantic frame are imported as multi-grained signals that control the empathy realization from coarse to fine levels. We then design a specific masking strategy to reflect the relationship between multi-grained signals and response tokens, and integrate it into the diffusion model to influence the generative process. Experimental results on a benchmark dataset EmpatheticDialogue show that our framework outperforms competitive baselines in terms of controllability, informativeness, and diversity without the loss of context-relatedness.", }

Empathy is a crucial factor in open-domain conversations, which naturally shows one{'}s caring and understanding to others. Though several methods have been proposed to generate empathetic responses, existing works often lead to monotonous empathy that refers to generic and safe expressions. In this paper, we propose to use explicit control to guide the empathy expression and design a framework DiffusEmp based on conditional diffusion language model to unify the utilization of dialogue context and attribute-oriented control signals. Specifically, communication mechanism, intent, and semantic frame are imported as multi-grained signals that control the empathy realization from coarse to fine levels. We then design a specific masking strategy to reflect the relationship between multi-grained signals and response tokens, and integrate it into the diffusion model to influence the generative process. Experimental results on a benchmark dataset EmpatheticDialogue show that our framework outperforms competitive baselines in terms of controllability, informativeness, and diversity without the loss of context-relatedness.

DiffusEmp: A Diffusion Model-Based Framework with Multi-Grained Control for Empathetic Response Generation

acl-long.158

Oral

2306.01657

https://aclanthology.org/2023.acl-long.159.bib

https://aclanthology.org/2023.acl-long.159/

@inproceedings{won-etal-2023-break, title = "{BREAK}: Breaking the Dialogue State Tracking Barrier with Beam Search and Re-ranking", author = "Won, Seungpil and Kwak, Heeyoung and Shin, Joongbo and Han, Janghoon and Jung, Kyomin", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.159", doi = "10.18653/v1/2023.acl-long.159", pages = "2832--2846", abstract = "Despite the recent advances in dialogue state tracking (DST), the joint goal accuracy (JGA) of the existing methods on MultiWOZ 2.1 still remains merely 60{\%}. In our preliminary error analysis, we find that beam search produces a pool of candidates that is likely to include the correct dialogue state. Motivated by this observation, we introduce a novel framework, called BREAK (Beam search and RE-rAnKing), that achieves outstanding performance on DST. BREAK performs DST in two stages: (i) generating k-best dialogue state candidates with beam search and (ii) re-ranking the candidates to select the correct dialogue state. This simple yet powerful framework shows state-of-the-art performance on all versions of MultiWOZ and M2M datasets. Most notably, we push the joint goal accuracy to 80-90{\%} on MultiWOZ 2.1-2.4, which is an improvement of 23.6{\%}, 26.3{\%}, 21.7{\%}, and 10.8{\%} over the previous best-performing models, respectively. The data and code will be available at \url{https://github.com/tony-won/DST-BREAK}", }

Despite the recent advances in dialogue state tracking (DST), the joint goal accuracy (JGA) of the existing methods on MultiWOZ 2.1 still remains merely 60{\%}. In our preliminary error analysis, we find that beam search produces a pool of candidates that is likely to include the correct dialogue state. Motivated by this observation, we introduce a novel framework, called BREAK (Beam search and RE-rAnKing), that achieves outstanding performance on DST. BREAK performs DST in two stages: (i) generating k-best dialogue state candidates with beam search and (ii) re-ranking the candidates to select the correct dialogue state. This simple yet powerful framework shows state-of-the-art performance on all versions of MultiWOZ and M2M datasets. Most notably, we push the joint goal accuracy to 80-90{\%} on MultiWOZ 2.1-2.4, which is an improvement of 23.6{\%}, 26.3{\%}, 21.7{\%}, and 10.8{\%} over the previous best-performing models, respectively. The data and code will be available at \url{https://github.com/tony-won/DST-BREAK}

BREAK: Breaking the Dialogue State Tracking Barrier with Beam Search and Re-ranking

acl-long.159

Poster

https://aclanthology.org/2023.acl-long.160.bib

https://aclanthology.org/2023.acl-long.160/

@inproceedings{wang-etal-2023-faithful, title = "Faithful Low-Resource Data-to-Text Generation through Cycle Training", author = "Wang, Zhuoer and Collins, Marcus and Vedula, Nikhita and Filice, Simone and Malmasi, Shervin and Rokhlenko, Oleg", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.160", doi = "10.18653/v1/2023.acl-long.160", pages = "2847--2867", abstract = "Methods to generate text from structured data have advanced significantly in recent years, primarily due to fine-tuning of pre-trained language models on large datasets. However, such models can fail to produce output faithful to the input data, particularly on out-of-domain data. Sufficient annotated data is often not available for specific domains, leading us to seek an unsupervised approach to improve the faithfulness of output text. Since the problem is fundamentally one of consistency between the representations of the structured data and text, we evaluate the effectiveness of cycle training in this work. Cycle training uses two models which are inverses of each other: one that generates text from structured data, and one which generates the structured data from natural language text. We show that cycle training, when initialized with a small amount of supervised data (100 samples in our case), achieves nearly the same performance as fully supervised approaches for the data-to-text generation task on the WebNLG, E2E, WTQ, and WSQL datasets. We perform extensive empirical analysis with automated evaluation metrics and a newly designed human evaluation schema to reveal different cycle training strategies{'} effectiveness of reducing various types of generation errors. Our code is publicly available at \url{https://github.com/Edillower/CycleNLG}.", }

Methods to generate text from structured data have advanced significantly in recent years, primarily due to fine-tuning of pre-trained language models on large datasets. However, such models can fail to produce output faithful to the input data, particularly on out-of-domain data. Sufficient annotated data is often not available for specific domains, leading us to seek an unsupervised approach to improve the faithfulness of output text. Since the problem is fundamentally one of consistency between the representations of the structured data and text, we evaluate the effectiveness of cycle training in this work. Cycle training uses two models which are inverses of each other: one that generates text from structured data, and one which generates the structured data from natural language text. We show that cycle training, when initialized with a small amount of supervised data (100 samples in our case), achieves nearly the same performance as fully supervised approaches for the data-to-text generation task on the WebNLG, E2E, WTQ, and WSQL datasets. We perform extensive empirical analysis with automated evaluation metrics and a newly designed human evaluation schema to reveal different cycle training strategies{'} effectiveness of reducing various types of generation errors. Our code is publicly available at \url{https://github.com/Edillower/CycleNLG}.

Faithful Low-Resource Data-to-Text Generation through Cycle Training

acl-long.160

Poster

2305.14793

https://aclanthology.org/2023.acl-long.161.bib

https://aclanthology.org/2023.acl-long.161/

@inproceedings{du-etal-2023-towards, title = "Towards Stable Natural Language Understanding via Information Entropy Guided Debiasing", author = "Du, Li and Ding, Xiao and Sun, Zhouhao and Liu, Ting and Qin, Bing and Liu, Jingshuo", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.161", doi = "10.18653/v1/2023.acl-long.161", pages = "2868--2882", abstract = "Although achieving promising performance, current Natural Language Understanding models tend to utilize dataset biases instead of learning the intended task, which always leads to performance degradation on out-of-distribution (OOD) samples. Toincrease the performance stability, previous debiasing methods \textit{empirically} capture bias features from data to prevent the model from corresponding biases. However, our analyses show that the empirical debiasing methods may fail to capture part of the potential dataset biases and mistake semantic information of input text as biases, which limits the effectiveness of debiasing. To address these issues, we propose a debiasing framework IEGDB that comprehensively detects the dataset biases to induce a set of biased features, and then purifies the biased features with the guidance of information entropy. Experimental results show that IEGDB can consistently improve the stability of performance on OOD datasets for a set of widely adopted NLU models.", }

Although achieving promising performance, current Natural Language Understanding models tend to utilize dataset biases instead of learning the intended task, which always leads to performance degradation on out-of-distribution (OOD) samples. Toincrease the performance stability, previous debiasing methods \textit{empirically} capture bias features from data to prevent the model from corresponding biases. However, our analyses show that the empirical debiasing methods may fail to capture part of the potential dataset biases and mistake semantic information of input text as biases, which limits the effectiveness of debiasing. To address these issues, we propose a debiasing framework IEGDB that comprehensively detects the dataset biases to induce a set of biased features, and then purifies the biased features with the guidance of information entropy. Experimental results show that IEGDB can consistently improve the stability of performance on OOD datasets for a set of widely adopted NLU models.

Towards Stable Natural Language Understanding via Information Entropy Guided Debiasing

acl-long.161

Poster

https://aclanthology.org/2023.acl-long.162.bib

https://aclanthology.org/2023.acl-long.162/

@inproceedings{liang-etal-2023-dynamic, title = "Dynamic and Efficient Inference for Text Generation via {BERT} Family", author = "Liang, Xiaobo and Li, Juntao and Wu, Lijun and Cao, Ziqiang and Zhang, Min", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.162", doi = "10.18653/v1/2023.acl-long.162", pages = "2883--2897", abstract = "Despite the excellent performance of Pre-trained Language Models on many text generation tasks, they suffer from inefficient inference on computation and memory due to their large-scale parameters and the universal autoregressive decoding paradigm. In this work, we propose a novel fine-tuning method \textbf{DEER}, which can make a single pre-trained model support \textbf{D}ynamic and \textbf{E}fficient inf\textbf{ER}ence and achieve an adaptive trade-off between model performance and latency. In particular, our critical insight is to jointly utilize the non-autoregressive (NAR) generation and dynamic parameter pruning techniques, which can flexibly control the decoding iteration steps and model sizes according to memory and latency limitations. Besides, we also explore the effectiveness of the pre-trained MLMs (i.e., the BERT family) for text generation tasks since their bidirectional attention nature is more suitable for the NAR training objective. Extensive experiments on both monolingual and multilingual pre-trained MLMs demonstrate the effectiveness of our proposed DEER method by consistently achieving (1) higher BLEU scores than the strong autoregressive Transformer model on three neural machine translation tasks with 3 $\to$ 12 times speedup, (2) competitive performance (but with much faster inference speed) compared with the BART model on four GLGE benchmark tasks. Our code will be publicly available at GitHub \url{https://github.com/dropreg/DEER}.", }

Despite the excellent performance of Pre-trained Language Models on many text generation tasks, they suffer from inefficient inference on computation and memory due to their large-scale parameters and the universal autoregressive decoding paradigm. In this work, we propose a novel fine-tuning method \textbf{DEER}, which can make a single pre-trained model support \textbf{D}ynamic and \textbf{E}fficient inf\textbf{ER}ence and achieve an adaptive trade-off between model performance and latency. In particular, our critical insight is to jointly utilize the non-autoregressive (NAR) generation and dynamic parameter pruning techniques, which can flexibly control the decoding iteration steps and model sizes according to memory and latency limitations. Besides, we also explore the effectiveness of the pre-trained MLMs (i.e., the BERT family) for text generation tasks since their bidirectional attention nature is more suitable for the NAR training objective. Extensive experiments on both monolingual and multilingual pre-trained MLMs demonstrate the effectiveness of our proposed DEER method by consistently achieving (1) higher BLEU scores than the strong autoregressive Transformer model on three neural machine translation tasks with 3 $\to$ 12 times speedup, (2) competitive performance (but with much faster inference speed) compared with the BART model on four GLGE benchmark tasks. Our code will be publicly available at GitHub \url{https://github.com/dropreg/DEER}.

Dynamic and Efficient Inference for Text Generation via BERT Family

acl-long.162

Poster

https://aclanthology.org/2023.acl-long.163.bib

https://aclanthology.org/2023.acl-long.163/

@inproceedings{sicilia-alikhani-2023-learning, title = "Learning to Generate Equitable Text in Dialogue from Biased Training Data", author = "Sicilia, Anthony and Alikhani, Malihe", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.163", doi = "10.18653/v1/2023.acl-long.163", pages = "2898--2917", abstract = "The ingrained principles of fairness in a dialogue system{'}s decision-making process and generated responses are crucial for user engagement, satisfaction, and task achievement. Absence of equitable and inclusive principles can hinder the formation of common ground, which in turn negatively impacts the overall performance of the system. For example, misusing pronouns in a user interaction may cause ambiguity about the intended subject. Yet, there is no comprehensive study of equitable text generation in dialogue. Aptly, in this work, we use theories of computational learning to study this problem. We provide formal definitions of equity in text generation, and further, prove formal connections between learning human-likeness and learning equity: algorithms for improving equity ultimately reduce to algorithms for improving human-likeness (on augmented data). With this insight, we also formulate reasonable conditions under which text generation algorithms can learn to generate equitable text without any modifications to the biased training data on which they learn. To exemplify our theory in practice, we look at a group of algorithms for the GuessWhat?! visual dialogue game and, using this example, test our theory empirically. Our theory accurately predicts relative-performance of multiple algorithms in generating equitable text as measured by both human and automated evaluation.", }

The ingrained principles of fairness in a dialogue system{'}s decision-making process and generated responses are crucial for user engagement, satisfaction, and task achievement. Absence of equitable and inclusive principles can hinder the formation of common ground, which in turn negatively impacts the overall performance of the system. For example, misusing pronouns in a user interaction may cause ambiguity about the intended subject. Yet, there is no comprehensive study of equitable text generation in dialogue. Aptly, in this work, we use theories of computational learning to study this problem. We provide formal definitions of equity in text generation, and further, prove formal connections between learning human-likeness and learning equity: algorithms for improving equity ultimately reduce to algorithms for improving human-likeness (on augmented data). With this insight, we also formulate reasonable conditions under which text generation algorithms can learn to generate equitable text without any modifications to the biased training data on which they learn. To exemplify our theory in practice, we look at a group of algorithms for the GuessWhat?! visual dialogue game and, using this example, test our theory empirically. Our theory accurately predicts relative-performance of multiple algorithms in generating equitable text as measured by both human and automated evaluation.

Learning to Generate Equitable Text in Dialogue from Biased Training Data

acl-long.163

Poster

2307.04303

https://aclanthology.org/2023.acl-long.164.bib

https://aclanthology.org/2023.acl-long.164/

@inproceedings{ji-etal-2023-hierarchical, title = "Hierarchical Verbalizer for Few-Shot Hierarchical Text Classification", author = "Ji, Ke and Lian, Yixin and Gao, Jingsheng and Wang, Baoyuan", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.164", doi = "10.18653/v1/2023.acl-long.164", pages = "2918--2933", abstract = "Due to the complex label hierarchy and intensive labeling cost in practice, the hierarchical text classification (HTC) suffers a poor performance especially when low-resource or few-shot settings are considered. Recently, there is a growing trend of applying prompts on pre-trained language models (PLMs), which has exhibited effectiveness in the few-shot flat text classification tasks. However, limited work has studied the paradigm of prompt-based learning in the HTC problem when the training data is extremely scarce. In this work, we define a path-based few-shot setting and establish a strict path-based evaluation metric to further explore few-shot HTC tasks. To address the issue, we propose the hierarchical verbalizer ({``}HierVerb{''}), a multi-verbalizer framework treating HTC as a single- or multi-label classification problem at multiple layers and learning vectors as verbalizers constrained by hierarchical structure and hierarchical contrastive learning. In this manner, HierVerb fuses label hierarchy knowledge into verbalizers and remarkably outperforms those who inject hierarchy through graph encoders, maximizing the benefits of PLMs. Extensive experiments on three popular HTC datasets under the few-shot settings demonstrate that prompt with HierVerb significantly boosts the HTC performance, meanwhile indicating an elegant way to bridge the gap between the large pre-trained model and downstream hierarchical classification tasks.", }

Due to the complex label hierarchy and intensive labeling cost in practice, the hierarchical text classification (HTC) suffers a poor performance especially when low-resource or few-shot settings are considered. Recently, there is a growing trend of applying prompts on pre-trained language models (PLMs), which has exhibited effectiveness in the few-shot flat text classification tasks. However, limited work has studied the paradigm of prompt-based learning in the HTC problem when the training data is extremely scarce. In this work, we define a path-based few-shot setting and establish a strict path-based evaluation metric to further explore few-shot HTC tasks. To address the issue, we propose the hierarchical verbalizer ({``}HierVerb{''}), a multi-verbalizer framework treating HTC as a single- or multi-label classification problem at multiple layers and learning vectors as verbalizers constrained by hierarchical structure and hierarchical contrastive learning. In this manner, HierVerb fuses label hierarchy knowledge into verbalizers and remarkably outperforms those who inject hierarchy through graph encoders, maximizing the benefits of PLMs. Extensive experiments on three popular HTC datasets under the few-shot settings demonstrate that prompt with HierVerb significantly boosts the HTC performance, meanwhile indicating an elegant way to bridge the gap between the large pre-trained model and downstream hierarchical classification tasks.

Hierarchical Verbalizer for Few-Shot Hierarchical Text Classification

acl-long.164

Poster

2305.16885

https://aclanthology.org/2023.acl-long.165.bib

https://aclanthology.org/2023.acl-long.165/

@inproceedings{liang-etal-2023-summary, title = "Summary-Oriented Vision Modeling for Multimodal Abstractive Summarization", author = "Liang, Yunlong and Meng, Fandong and Xu, Jinan and Wang, Jiaan and Chen, Yufeng and Zhou, Jie", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.165", doi = "10.18653/v1/2023.acl-long.165", pages = "2934--2951", abstract = "The goal of multimodal abstractive summarization (MAS) is to produce a concise summary given the multimodal data (text and vision). Existing studies on MAS mainly focus on how to effectively use the extracted visual features, having achieved impressive success on the high-resource English dataset. However, less attention has been paid to the quality of the visual features to the summary, which may limit the model performance, especially in the low- and zero-resource scenarios. In this paper, we propose to improve the summary quality through summary-oriented visual features. To this end, we devise two auxiliary tasks including vision to summary task and masked image modeling task. Together with the main summarization task, we optimize the MAS model via the training objectives of all these tasks. By these means, the MAS model can be enhanced by capturing the summary-oriented visual features, thereby yielding more accurate summaries. Experiments on 44 languages, covering mid-high-, low-, and zero-resource scenarios, verify the effectiveness and superiority of the proposed approach, which achieves state-of-the-art performance under all scenarios. Additionally, we will contribute a large-scale multilingual multimodal abstractive summarization (MM-Sum) dataset to the research community.", }

The goal of multimodal abstractive summarization (MAS) is to produce a concise summary given the multimodal data (text and vision). Existing studies on MAS mainly focus on how to effectively use the extracted visual features, having achieved impressive success on the high-resource English dataset. However, less attention has been paid to the quality of the visual features to the summary, which may limit the model performance, especially in the low- and zero-resource scenarios. In this paper, we propose to improve the summary quality through summary-oriented visual features. To this end, we devise two auxiliary tasks including vision to summary task and masked image modeling task. Together with the main summarization task, we optimize the MAS model via the training objectives of all these tasks. By these means, the MAS model can be enhanced by capturing the summary-oriented visual features, thereby yielding more accurate summaries. Experiments on 44 languages, covering mid-high-, low-, and zero-resource scenarios, verify the effectiveness and superiority of the proposed approach, which achieves state-of-the-art performance under all scenarios. Additionally, we will contribute a large-scale multilingual multimodal abstractive summarization (MM-Sum) dataset to the research community.

Summary-Oriented Vision Modeling for Multimodal Abstractive Summarization

acl-long.165

Poster

2212.07672

https://aclanthology.org/2023.acl-long.166.bib

https://aclanthology.org/2023.acl-long.166/

@inproceedings{russo-etal-2023-helping, title = "Helping a Friend or Supporting a Cause? Disentangling Active and Passive Cosponsorship in the {U}.{S}. Congress", author = "Russo, Giuseppe and Gote, Christoph and Brandenberger, Laurence and Schlosser, Sophia and Schweitzer, Frank", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.166", doi = "10.18653/v1/2023.acl-long.166", pages = "2952--2969", abstract = "In the U.S. Congress, legislators can use active and passive cosponsorship to support bills. We show that these two types of cosponsorship are driven by two different motivations: the backing of political colleagues and the backing of the bill{'}s content. To this end, we develop an Encoder+RGCN based model that learns legislator representations from bill texts and speech transcripts. These representations predict active and passive cosponsorship with an F1-score of 0.88.Applying our representations to predict voting decisions, we show that they are interpretable and generalize to unseen tasks.", }

In the U.S. Congress, legislators can use active and passive cosponsorship to support bills. We show that these two types of cosponsorship are driven by two different motivations: the backing of political colleagues and the backing of the bill{'}s content. To this end, we develop an Encoder+RGCN based model that learns legislator representations from bill texts and speech transcripts. These representations predict active and passive cosponsorship with an F1-score of 0.88.Applying our representations to predict voting decisions, we show that they are interpretable and generalize to unseen tasks.

Helping a Friend or Supporting a Cause? Disentangling Active and Passive Cosponsorship in the U.S. Congress

acl-long.166

Poster

https://aclanthology.org/2023.acl-long.167.bib

https://aclanthology.org/2023.acl-long.167/

@inproceedings{li-etal-2023-trea, title = "{TREA}: Tree-Structure Reasoning Schema for Conversational Recommendation", author = "Li, Wendi and Wei, Wei and Qu, Xiaoye and Mao, Xian-Ling and Yuan, Ye and Xie, Wenfeng and Chen, Dangyang", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.167", doi = "10.18653/v1/2023.acl-long.167", pages = "2970--2982", abstract = "Conversational recommender systems (CRS) aim to timely trace the dynamic interests of users through dialogues and generate relevant responses for item recommendations. Recently, various external knowledge bases (especially knowledge graphs) are incorporated into CRS to enhance the understanding of conversation contexts. However, recent reasoning-based models heavily rely on simplified structures such as linear structures or fixed-hierarchical structures for causality reasoning, hence they cannot fully figure out sophisticated relationships among utterances with external knowledge. To address this, we propose a novel Tree structure Reasoning schEmA named TREA. TREA constructs a multi-hierarchical scalable tree as the reasoning structure to clarify the causal relationships between mentioned entities, and fully utilizes historical conversations to generate more reasonable and suitable responses for recommended results. Extensive experiments on two public CRS datasets have demonstrated the effectiveness of our approach.", }

Conversational recommender systems (CRS) aim to timely trace the dynamic interests of users through dialogues and generate relevant responses for item recommendations. Recently, various external knowledge bases (especially knowledge graphs) are incorporated into CRS to enhance the understanding of conversation contexts. However, recent reasoning-based models heavily rely on simplified structures such as linear structures or fixed-hierarchical structures for causality reasoning, hence they cannot fully figure out sophisticated relationships among utterances with external knowledge. To address this, we propose a novel Tree structure Reasoning schEmA named TREA. TREA constructs a multi-hierarchical scalable tree as the reasoning structure to clarify the causal relationships between mentioned entities, and fully utilizes historical conversations to generate more reasonable and suitable responses for recommended results. Extensive experiments on two public CRS datasets have demonstrated the effectiveness of our approach.

TREA: Tree-Structure Reasoning Schema for Conversational Recommendation

acl-long.167

Poster

2307.10543

https://aclanthology.org/2023.acl-long.168.bib

https://aclanthology.org/2023.acl-long.168/

@inproceedings{li-etal-2023-cats, title = "{CATS}: A Pragmatic {C}hinese Answer-to-Sequence Dataset with Large Scale and High Quality", author = "Li, Liang and Geng, Ruiying and Fang, Chengyang and Li, Bing and Ma, Can and Cao, Rongyu and Li, Binhua and Huang, Fei and Li, Yongbin", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.168", doi = "10.18653/v1/2023.acl-long.168", pages = "2983--3000", abstract = "There are three problems existing in the popular data-to-text datasets. First, the large-scale datasets either contain noise or lack real application scenarios. Second, the datasets close to real applications are relatively small in size. Last, current datasets bias in the English language while leaving other languages underexplored.To alleviate these limitations, in this paper, we present CATS, a pragmatic Chinese answer-to-sequence dataset with large scale and high quality. The dataset aims to generate textual descriptions for the answer in the practical TableQA system. Further, to bridge the structural gap between the input SQL and table and establish better semantic alignments, we propose a Unified Graph Transformation approach to establish a joint encoding space for the two hybrid knowledge resources and convert this task to a graph-to-text problem. The experiment results demonstrate the effectiveness of our proposed method. Further analysis on CATS attests to both the high quality and challenges of the dataset", }

There are three problems existing in the popular data-to-text datasets. First, the large-scale datasets either contain noise or lack real application scenarios. Second, the datasets close to real applications are relatively small in size. Last, current datasets bias in the English language while leaving other languages underexplored.To alleviate these limitations, in this paper, we present CATS, a pragmatic Chinese answer-to-sequence dataset with large scale and high quality. The dataset aims to generate textual descriptions for the answer in the practical TableQA system. Further, to bridge the structural gap between the input SQL and table and establish better semantic alignments, we propose a Unified Graph Transformation approach to establish a joint encoding space for the two hybrid knowledge resources and convert this task to a graph-to-text problem. The experiment results demonstrate the effectiveness of our proposed method. Further analysis on CATS attests to both the high quality and challenges of the dataset

CATS: A Pragmatic Chinese Answer-to-Sequence Dataset with Large Scale and High Quality

acl-long.168

Poster

2306.11477

https://aclanthology.org/2023.acl-long.169.bib

https://aclanthology.org/2023.acl-long.169/

@inproceedings{piskorski-etal-2023-multilingual, title = "Multilingual Multifaceted Understanding of Online News in Terms of Genre, Framing, and Persuasion Techniques", author = "Piskorski, Jakub and Stefanovitch, Nicolas and Nikolaidis, Nikolaos and Da San Martino, Giovanni and Nakov, Preslav", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.169", doi = "10.18653/v1/2023.acl-long.169", pages = "3001--3022", abstract = "We present a new multilingual multifacet dataset of news articles, each annotated for genre (objective news reporting vs. opinion vs. satire), framing (what key aspects are highlighted), and persuasion techniques (logical fallacies, emotional appeals, ad hominem attacks, etc.). The persuasion techniques are annotated at the span level, using a taxonomy of 23 fine-grained techniques grouped into 6 coarse categories. The dataset contains 1,612 news articles covering recent news on current topics of public interest in six European languages (English, French, German, Italian, Polish, and Russian), with more than 37k annotated spans of persuasion techniques. We describe the dataset and the annotation process, and we report the evaluation results of multilabel classification experiments using state-of-the-art multilingual transformers at different levels of granularity: token-level, sentence-level, paragraph-level, and document-level.", }

We present a new multilingual multifacet dataset of news articles, each annotated for genre (objective news reporting vs. opinion vs. satire), framing (what key aspects are highlighted), and persuasion techniques (logical fallacies, emotional appeals, ad hominem attacks, etc.). The persuasion techniques are annotated at the span level, using a taxonomy of 23 fine-grained techniques grouped into 6 coarse categories. The dataset contains 1,612 news articles covering recent news on current topics of public interest in six European languages (English, French, German, Italian, Polish, and Russian), with more than 37k annotated spans of persuasion techniques. We describe the dataset and the annotation process, and we report the evaluation results of multilabel classification experiments using state-of-the-art multilingual transformers at different levels of granularity: token-level, sentence-level, paragraph-level, and document-level.

Multilingual Multifaceted Understanding of Online News in Terms of Genre, Framing, and Persuasion Techniques

acl-long.169

Poster

https://aclanthology.org/2023.acl-long.170.bib

https://aclanthology.org/2023.acl-long.170/

@inproceedings{wu-etal-2023-learning, title = "Learning Action Conditions from Instructional Manuals for Instruction Understanding", author = "Wu, Te-Lin and Zhang, Caiqi and Hu, Qingyuan and Spangher, Alexander and Peng, Nanyun", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.170", doi = "10.18653/v1/2023.acl-long.170", pages = "3023--3043", abstract = "The ability to infer pre- and postconditions of an action is vital for comprehending complex instructions, and is essential for applications such as autonomous instruction-guided agents and assistive AI that supports humans to perform physical tasks. In this work, we propose a task dubbed action condition inference, which extracts mentions of preconditions and postconditions of actions in instructional manuals. We propose a weakly supervised approach utilizing automatically constructed large-scale training instances from online instructions, and curate a densely human-annotated and validated dataset to study how well the current NLP models do on the proposed task. We design two types of models differ by whether contextualized and global information is leveraged, as well as various combinations of heuristics to construct the weak supervisions.Our experiments show a {\textgreater} 20{\%} F1-score improvement with considering the entire instruction contexts and a {\textgreater} 6{\%} F1-score benefit with the proposed heuristics. However, the best performing model is still well-behind human performance.", }

The ability to infer pre- and postconditions of an action is vital for comprehending complex instructions, and is essential for applications such as autonomous instruction-guided agents and assistive AI that supports humans to perform physical tasks. In this work, we propose a task dubbed action condition inference, which extracts mentions of preconditions and postconditions of actions in instructional manuals. We propose a weakly supervised approach utilizing automatically constructed large-scale training instances from online instructions, and curate a densely human-annotated and validated dataset to study how well the current NLP models do on the proposed task. We design two types of models differ by whether contextualized and global information is leveraged, as well as various combinations of heuristics to construct the weak supervisions.Our experiments show a {\textgreater} 20{\%} F1-score improvement with considering the entire instruction contexts and a {\textgreater} 6{\%} F1-score benefit with the proposed heuristics. However, the best performing model is still well-behind human performance.

Learning Action Conditions from Instructional Manuals for Instruction Understanding

acl-long.170

Poster

2205.12420

https://aclanthology.org/2023.acl-long.171.bib

https://aclanthology.org/2023.acl-long.171/

@inproceedings{du-chilton-2023-storywars, title = "{S}tory{W}ars: A Dataset and Instruction Tuning Baselines for Collaborative Story Understanding and Generation", author = "Du, Yulun and Chilton, Lydia", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.171", doi = "10.18653/v1/2023.acl-long.171", pages = "3044--3062", abstract = "Collaborative stories, which are texts created through the collaborative efforts of multiple authors with different writing styles and intentions, pose unique challenges for NLP models. Understanding and generating such stories remains an underexplored area due to the lack of open-domain corpora. To address this, we introduce StoryWars, a new dataset of over 40,000 collaborative stories written by 9,400 different authors from an online platform. We design 12 task types, comprising 7 understanding and 5 generation task types, on {pasted macro {`}STORYWARS{'}}, deriving 101 diverse story-related tasks in total as a multi-task benchmark covering all fully-supervised, few-shot, and zero-shot scenarios. Furthermore, we present our instruction-tuned model, InstructStory, for the story tasks showing that instruction tuning, in addition to achieving superior results in zero-shot and few-shot scenarios, can also obtain the best performance on the fully-supervised tasks in StoryWars, establishing strong multi-task benchmark performances on StoryWars.", }

Collaborative stories, which are texts created through the collaborative efforts of multiple authors with different writing styles and intentions, pose unique challenges for NLP models. Understanding and generating such stories remains an underexplored area due to the lack of open-domain corpora. To address this, we introduce StoryWars, a new dataset of over 40,000 collaborative stories written by 9,400 different authors from an online platform. We design 12 task types, comprising 7 understanding and 5 generation task types, on {pasted macro {`}STORYWARS{'}}, deriving 101 diverse story-related tasks in total as a multi-task benchmark covering all fully-supervised, few-shot, and zero-shot scenarios. Furthermore, we present our instruction-tuned model, InstructStory, for the story tasks showing that instruction tuning, in addition to achieving superior results in zero-shot and few-shot scenarios, can also obtain the best performance on the fully-supervised tasks in StoryWars, establishing strong multi-task benchmark performances on StoryWars.

StoryWars: A Dataset and Instruction Tuning Baselines for Collaborative Story Understanding and Generation

acl-long.171

Poster

2305.08152

https://aclanthology.org/2023.acl-long.172.bib

https://aclanthology.org/2023.acl-long.172/

@inproceedings{yin-etal-2023-read, title = "Did You Read the Instructions? Rethinking the Effectiveness of Task Definitions in Instruction Learning", author = "Yin, Fan and Vig, Jesse and Laban, Philippe and Joty, Shafiq and Xiong, Caiming and Wu, Chien-Sheng", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.172", doi = "10.18653/v1/2023.acl-long.172", pages = "3063--3079", abstract = "Large language models (LLMs) have shown impressive performance in following natural language instructions to solve unseen tasks. However, it remains unclear whether models truly understand task definitions and whether the human-written definitions are optimal. In this paper, we systematically study the role of task definitions in instruction learning. We first conduct an ablation analysis informed by human annotations to understand which parts of a task definition are most important, and find that model performance only drops substantially when removing contents describing the task output, in particular label information. Next, we propose an automatic algorithm to compress task definitions to a minimal supporting set of tokens, and find that 60{\%} of tokens can be removed while maintaining or even improving model performance. Based on these results, we propose two strategies to help models better leverage task instructions: (1) providing only key information for tasks in a common structured format, and (2) adding a meta-tuning stage to help the model better understand the definitions. With these two strategies, we achieve a 4.2 Rouge-L improvement over 119 unseen test tasks.", }

Large language models (LLMs) have shown impressive performance in following natural language instructions to solve unseen tasks. However, it remains unclear whether models truly understand task definitions and whether the human-written definitions are optimal. In this paper, we systematically study the role of task definitions in instruction learning. We first conduct an ablation analysis informed by human annotations to understand which parts of a task definition are most important, and find that model performance only drops substantially when removing contents describing the task output, in particular label information. Next, we propose an automatic algorithm to compress task definitions to a minimal supporting set of tokens, and find that 60{\%} of tokens can be removed while maintaining or even improving model performance. Based on these results, we propose two strategies to help models better leverage task instructions: (1) providing only key information for tasks in a common structured format, and (2) adding a meta-tuning stage to help the model better understand the definitions. With these two strategies, we achieve a 4.2 Rouge-L improvement over 119 unseen test tasks.

Did You Read the Instructions? Rethinking the Effectiveness of Task Definitions in Instruction Learning

acl-long.172

Poster

2306.01150

https://huggingface.co/papers/2306.01150

https://aclanthology.org/2023.acl-long.173.bib

https://aclanthology.org/2023.acl-long.173/

@inproceedings{wu-etal-2023-plms, title = "Do {PLM}s Know and Understand Ontological Knowledge?", author = "Wu, Weiqi and Jiang, Chengyue and Jiang, Yong and Xie, Pengjun and Tu, Kewei", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.173", doi = "10.18653/v1/2023.acl-long.173", pages = "3080--3101", abstract = "Ontological knowledge, which comprises classes and properties and their relationships, is integral to world knowledge. It is significant to explore whether Pretrained Language Models (PLMs) know and understand such knowledge. However, existing PLM-probing studies focus mainly on factual knowledge, lacking a system- atic probing of ontological knowledge. In this paper, we focus on probing whether PLMs store ontological knowledge and have a semantic un- derstanding of the knowledge rather than rote memorization of the surface form. To probe whether PLMs know ontological knowledge, we investigate how well PLMs memorize: (1) types of entities; (2) hierarchical relationships among classes and properties, e.g., Person is a subclass of Animal and Member of Sports Team is a subproperty of Member of ; (3) domain and range constraints of properties, e.g., the subject of Member of Sports Team should be a Person and the object should be a Sports Team. To further probe whether PLMs truly understand ontological knowledge beyond memorization, we comprehensively study whether they can reliably perform logical reasoning with given knowledge according to ontological entailment rules. Our probing results show that PLMs can memorize certain ontological knowledge and utilize implicit knowledge in reasoning. How- ever, both the memorizing and reasoning per- formances are less than perfect, indicating in- complete knowledge and understanding.", }

Ontological knowledge, which comprises classes and properties and their relationships, is integral to world knowledge. It is significant to explore whether Pretrained Language Models (PLMs) know and understand such knowledge. However, existing PLM-probing studies focus mainly on factual knowledge, lacking a system- atic probing of ontological knowledge. In this paper, we focus on probing whether PLMs store ontological knowledge and have a semantic un- derstanding of the knowledge rather than rote memorization of the surface form. To probe whether PLMs know ontological knowledge, we investigate how well PLMs memorize: (1) types of entities; (2) hierarchical relationships among classes and properties, e.g., Person is a subclass of Animal and Member of Sports Team is a subproperty of Member of ; (3) domain and range constraints of properties, e.g., the subject of Member of Sports Team should be a Person and the object should be a Sports Team. To further probe whether PLMs truly understand ontological knowledge beyond memorization, we comprehensively study whether they can reliably perform logical reasoning with given knowledge according to ontological entailment rules. Our probing results show that PLMs can memorize certain ontological knowledge and utilize implicit knowledge in reasoning. How- ever, both the memorizing and reasoning per- formances are less than perfect, indicating in- complete knowledge and understanding.

Do PLMs Know and Understand Ontological Knowledge?

acl-long.173

Poster

2309.05936

https://aclanthology.org/2023.acl-long.174.bib

https://aclanthology.org/2023.acl-long.174/

@inproceedings{tao-etal-2023-core, title = "{CORE}: Cooperative Training of Retriever-Reranker for Effective Dialogue Response Selection", author = "Tao, Chongyang and Feng, Jiazhan and Shen, Tao and Liu, Chang and Li, Juntao and Geng, Xiubo and Jiang, Daxin", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.174", doi = "10.18653/v1/2023.acl-long.174", pages = "3102--3114", abstract = "Establishing retrieval-based dialogue systems that can select appropriate responses from the pre-built index has gained increasing attention. Recent common practice is to construct a two-stage pipeline with a fast retriever (e.g., bi-encoder) for first-stage recall followed by a smart response reranker (e.g., cross-encoder) for precise ranking. However, existing studies either optimize the retriever and reranker in independent ways, or distill the knowledge from a pre-trained reranker into the retriever in an asynchronous way, leading to sub-optimal performance of both modules. Thus, an open question remains about how to train them for a better combination of the best of both worlds. To this end, we present a cooperative training of the response retriever and the reranker whose parameters are dynamically optimized by the ground-truth labels as well as list-wise supervision signals from each other. As a result, the two modules can learn from each other and evolve together throughout the training. Experimental results on two benchmarks demonstrate the superiority of our method.", }

Establishing retrieval-based dialogue systems that can select appropriate responses from the pre-built index has gained increasing attention. Recent common practice is to construct a two-stage pipeline with a fast retriever (e.g., bi-encoder) for first-stage recall followed by a smart response reranker (e.g., cross-encoder) for precise ranking. However, existing studies either optimize the retriever and reranker in independent ways, or distill the knowledge from a pre-trained reranker into the retriever in an asynchronous way, leading to sub-optimal performance of both modules. Thus, an open question remains about how to train them for a better combination of the best of both worlds. To this end, we present a cooperative training of the response retriever and the reranker whose parameters are dynamically optimized by the ground-truth labels as well as list-wise supervision signals from each other. As a result, the two modules can learn from each other and evolve together throughout the training. Experimental results on two benchmarks demonstrate the superiority of our method.

CORE: Cooperative Training of Retriever-Reranker for Effective Dialogue Response Selection

acl-long.174

Poster

https://aclanthology.org/2023.acl-long.175.bib

https://aclanthology.org/2023.acl-long.175/

@inproceedings{chen-elsner-2023-exploring, title = "Exploring How Generative Adversarial Networks Learn Phonological Representations", author = "Chen, Jingyi and Elsner, Micha", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.175", doi = "10.18653/v1/2023.acl-long.175", pages = "3115--3129", abstract = "This paper explores how Generative Adversarial Networks (GANs) learn representations of phonological phenomena. We analyze how GANs encode contrastive and non-contrastive nasality in French and English vowels by applying the ciwGAN architecture (Begus, 2021). Begus claims that ciwGAN encodes linguistically meaningful representations with categorical variables in its latent space and manipulating the latent variables shows an almost one to one corresponding control of the phonological features in ciwGAN{'}s generated outputs. However, our results show an interactive effect of latent variables on the features in the generated outputs, which suggests the learned representations in neural networks are different from the phonological representations proposed by linguists. On the other hand, ciwGAN is able to distinguish contrastive and noncontrastive features in English and French by encoding them differently. Comparing the performance of GANs learning from different languages results in a better understanding of what language specific features contribute to developing language specific phonological representations. We also discuss the role of training data frequencies in phonological feature learning.", }

This paper explores how Generative Adversarial Networks (GANs) learn representations of phonological phenomena. We analyze how GANs encode contrastive and non-contrastive nasality in French and English vowels by applying the ciwGAN architecture (Begus, 2021). Begus claims that ciwGAN encodes linguistically meaningful representations with categorical variables in its latent space and manipulating the latent variables shows an almost one to one corresponding control of the phonological features in ciwGAN{'}s generated outputs. However, our results show an interactive effect of latent variables on the features in the generated outputs, which suggests the learned representations in neural networks are different from the phonological representations proposed by linguists. On the other hand, ciwGAN is able to distinguish contrastive and noncontrastive features in English and French by encoding them differently. Comparing the performance of GANs learning from different languages results in a better understanding of what language specific features contribute to developing language specific phonological representations. We also discuss the role of training data frequencies in phonological feature learning.

Exploring How Generative Adversarial Networks Learn Phonological Representations

acl-long.175

Oral

2305.12501

https://aclanthology.org/2023.acl-long.176.bib

https://aclanthology.org/2023.acl-long.176/

@inproceedings{giulianelli-etal-2023-interpretable, title = "Interpretable Word Sense Representations via Definition Generation: The Case of Semantic Change Analysis", author = "Giulianelli, Mario and Luden, Iris and Fernandez, Raquel and Kutuzov, Andrey", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.176", doi = "10.18653/v1/2023.acl-long.176", pages = "3130--3148", abstract = "We propose using automatically generated natural language definitions of contextualised word usages as interpretable word and word sense representations. Given a collection of usage examples for a target word, and the corresponding data-driven usage clusters (i.e., word senses), a definition is generated for each usage with a specialised Flan-T5 language model, and the most prototypical definition in a usage cluster is chosen as the sense label. We demonstrate how the resulting sense labels can make existing approaches to semantic change analysis more interpretable, and how they can allow users {---} historical linguists, lexicographers, or social scientists {---} to explore and intuitively explain diachronic trajectories of word meaning. Semantic change analysis is only one of many possible applications of the {`}definitions as representations{'} paradigm. Beyond being human-readable, contextualised definitions also outperform token or usage sentence embeddings in word-in-context semantic similarity judgements, making them a new promising type of lexical representation for NLP.", }

We propose using automatically generated natural language definitions of contextualised word usages as interpretable word and word sense representations. Given a collection of usage examples for a target word, and the corresponding data-driven usage clusters (i.e., word senses), a definition is generated for each usage with a specialised Flan-T5 language model, and the most prototypical definition in a usage cluster is chosen as the sense label. We demonstrate how the resulting sense labels can make existing approaches to semantic change analysis more interpretable, and how they can allow users {---} historical linguists, lexicographers, or social scientists {---} to explore and intuitively explain diachronic trajectories of word meaning. Semantic change analysis is only one of many possible applications of the {`}definitions as representations{'} paradigm. Beyond being human-readable, contextualised definitions also outperform token or usage sentence embeddings in word-in-context semantic similarity judgements, making them a new promising type of lexical representation for NLP.

Interpretable Word Sense Representations via Definition Generation: The Case of Semantic Change Analysis

acl-long.176

Oral

2305.11993

https://huggingface.co/papers/2305.11993

https://aclanthology.org/2023.acl-long.177.bib

https://aclanthology.org/2023.acl-long.177/

@inproceedings{yan-etal-2023-learning, title = "Learning to Simulate Natural Language Feedback for Interactive Semantic Parsing", author = "Yan, Hao and Srivastava, Saurabh and Tai, Yintao and Wang, Sida I. and Yih, Wen-tau and Yao, Ziyu", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.177", doi = "10.18653/v1/2023.acl-long.177", pages = "3149--3170", abstract = "Interactive semantic parsing based on natural language (NL) feedback, where users provide feedback to correct the parser mistakes, has emerged as a more practical scenario than the traditional one-shot semantic parsing. However, prior work has heavily relied on human-annotated feedback data to train the interactive semantic parser, which is prohibitively expensive and not scalable. In this work, we propose a new task of simulating NL feedback for interactive semantic parsing. We accompany the task with a novel feedback evaluator. The evaluator is specifically designed to assess the quality of the simulated feedback, based on which we decide the best feedback simulator from our proposed variants. On a text-to-SQL dataset, we show that our feedback simulator can generate high-quality NL feedback to boost the error correction ability of a specific parser. In low-data settings, our feedback simulator can help achieve comparable error correction performance as trained using the costly, full set of human annotations.", }

Interactive semantic parsing based on natural language (NL) feedback, where users provide feedback to correct the parser mistakes, has emerged as a more practical scenario than the traditional one-shot semantic parsing. However, prior work has heavily relied on human-annotated feedback data to train the interactive semantic parser, which is prohibitively expensive and not scalable. In this work, we propose a new task of simulating NL feedback for interactive semantic parsing. We accompany the task with a novel feedback evaluator. The evaluator is specifically designed to assess the quality of the simulated feedback, based on which we decide the best feedback simulator from our proposed variants. On a text-to-SQL dataset, we show that our feedback simulator can generate high-quality NL feedback to boost the error correction ability of a specific parser. In low-data settings, our feedback simulator can help achieve comparable error correction performance as trained using the costly, full set of human annotations.

Learning to Simulate Natural Language Feedback for Interactive Semantic Parsing

acl-long.177

Poster

2305.08195

https://aclanthology.org/2023.acl-long.178.bib

https://aclanthology.org/2023.acl-long.178/

@inproceedings{hu-etal-2023-infometic, title = "{I}nfo{M}et{IC}: An Informative Metric for Reference-free Image Caption Evaluation", author = "Hu, Anwen and Chen, Shizhe and Zhang, Liang and Jin, Qin", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.178", doi = "10.18653/v1/2023.acl-long.178", pages = "3171--3185", abstract = "Automatic image captioning evaluation is critical for benchmarking and promoting advances in image captioning research. Existing metrics only provide a single score to measure caption qualities, which are less explainable and informative. Instead, we humans can easily identify the problems of captions in details, e.g., which words are inaccurate and which salient objects are not described, and then rate the caption quality. To support such informative feedback, we propose an Informative Metric for Reference-free Image Caption evaluation (InfoMetIC). Given an image and a caption, InfoMetIC is able to report incorrect words and unmentioned image regions at fine-grained level, and also provide a text precision score, a vision recall score and an overall quality score at coarse-grained level. The coarse-grained score of InfoMetIC achieves significantly better correlation with human judgements than existing metrics on multiple benchmarks. We also construct a token-level evaluation dataset and demonstrate the effectiveness of InfoMetIC in fine-grained evaluation. Our code and datasets are publicly available at \url{https://github.com/HAWLYQ/InfoMetIC}.", }

Automatic image captioning evaluation is critical for benchmarking and promoting advances in image captioning research. Existing metrics only provide a single score to measure caption qualities, which are less explainable and informative. Instead, we humans can easily identify the problems of captions in details, e.g., which words are inaccurate and which salient objects are not described, and then rate the caption quality. To support such informative feedback, we propose an Informative Metric for Reference-free Image Caption evaluation (InfoMetIC). Given an image and a caption, InfoMetIC is able to report incorrect words and unmentioned image regions at fine-grained level, and also provide a text precision score, a vision recall score and an overall quality score at coarse-grained level. The coarse-grained score of InfoMetIC achieves significantly better correlation with human judgements than existing metrics on multiple benchmarks. We also construct a token-level evaluation dataset and demonstrate the effectiveness of InfoMetIC in fine-grained evaluation. Our code and datasets are publicly available at \url{https://github.com/HAWLYQ/InfoMetIC}.

InfoMetIC: An Informative Metric for Reference-free Image Caption Evaluation

acl-long.178

Poster

2305.06002

https://aclanthology.org/2023.acl-long.179.bib

https://aclanthology.org/2023.acl-long.179/

@inproceedings{zheng-etal-2023-invariant, title = "An Invariant Learning Characterization of Controlled Text Generation", author = "Zheng, Carolina and Shi, Claudia and Vafa, Keyon and Feder, Amir and Blei, David", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.179", doi = "10.18653/v1/2023.acl-long.179", pages = "3186--3206", abstract = "Controlled generation refers to the problem of creating text that contains stylistic or semantic attributes of interest. Many approaches reduce this problem to training a predictor of the desired attribute. For example, researchers hoping to deploy a large language model to produce non-toxic content may use a toxicity classifier to filter generated text. In practice, the generated text to classify, which is determined by user prompts, may come from a wide range of distributions. In this paper, we show that the performance of controlled generation may be poor if the distributions of text in response to user prompts differ from the distribution the predictor was trained on. To address this problem, we cast controlled generation under distribution shift as an invariant learning problem: the most effective predictor should be invariant across multiple text environments. We then discuss a natural solution that arises from this characterization and propose heuristics for selecting natural environments. We study this characterization and the proposed method empirically using both synthetic and real data. Experiments demonstrate both the challenge of distribution shift in controlled generation and the potential of invariance methods in this setting.", }

Controlled generation refers to the problem of creating text that contains stylistic or semantic attributes of interest. Many approaches reduce this problem to training a predictor of the desired attribute. For example, researchers hoping to deploy a large language model to produce non-toxic content may use a toxicity classifier to filter generated text. In practice, the generated text to classify, which is determined by user prompts, may come from a wide range of distributions. In this paper, we show that the performance of controlled generation may be poor if the distributions of text in response to user prompts differ from the distribution the predictor was trained on. To address this problem, we cast controlled generation under distribution shift as an invariant learning problem: the most effective predictor should be invariant across multiple text environments. We then discuss a natural solution that arises from this characterization and propose heuristics for selecting natural environments. We study this characterization and the proposed method empirically using both synthetic and real data. Experiments demonstrate both the challenge of distribution shift in controlled generation and the potential of invariance methods in this setting.

An Invariant Learning Characterization of Controlled Text Generation

acl-long.179

Poster

2306.00198

https://aclanthology.org/2023.acl-long.180.bib

https://aclanthology.org/2023.acl-long.180/

@inproceedings{yang-etal-2023-histred, title = "{H}ist{RED}: A Historical Document-Level Relation Extraction Dataset", author = "Yang, Soyoung and Choi, Minseok and Cho, Youngwoo and Choo, Jaegul", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.180", doi = "10.18653/v1/2023.acl-long.180", pages = "3207--3224", abstract = "Despite the extensive applications of relation extraction (RE) tasks in various domains, little has been explored in the historical context, which contains promising data across hundreds and thousands of years. To promote the historical RE research, we present HistRED constructed from Yeonhaengnok. Yeonhaengnok is a collection of records originally written in Hanja, the classical Chinese writing, which has later been translated into Korean. HistRED provides bilingual annotations such that RE can be performed on Korean and Hanja texts. In addition, HistRED supports various self-contained subtexts with different lengths, from a sentence level to a document level, supporting diverse context settings for researchers to evaluate the robustness of their RE models. To demonstrate the usefulness of our dataset, we propose a bilingual RE model that leverages both Korean and Hanja contexts to predict relations between entities. Our model outperforms monolingual baselines on HistRED, showing that employing multiple language contexts supplements the RE predictions. The dataset is publicly available at: \url{https://huggingface.co/datasets/Soyoung/HistRED} under CC BY-NC-ND 4.0 license.", }

Despite the extensive applications of relation extraction (RE) tasks in various domains, little has been explored in the historical context, which contains promising data across hundreds and thousands of years. To promote the historical RE research, we present HistRED constructed from Yeonhaengnok. Yeonhaengnok is a collection of records originally written in Hanja, the classical Chinese writing, which has later been translated into Korean. HistRED provides bilingual annotations such that RE can be performed on Korean and Hanja texts. In addition, HistRED supports various self-contained subtexts with different lengths, from a sentence level to a document level, supporting diverse context settings for researchers to evaluate the robustness of their RE models. To demonstrate the usefulness of our dataset, we propose a bilingual RE model that leverages both Korean and Hanja contexts to predict relations between entities. Our model outperforms monolingual baselines on HistRED, showing that employing multiple language contexts supplements the RE predictions. The dataset is publicly available at: \url{https://huggingface.co/datasets/Soyoung/HistRED} under CC BY-NC-ND 4.0 license.

HistRED: A Historical Document-Level Relation Extraction Dataset

acl-long.180

Poster

2307.04285

https://aclanthology.org/2023.acl-long.181.bib

https://aclanthology.org/2023.acl-long.181/

@inproceedings{xu-etal-2023-critical, title = "A Critical Evaluation of Evaluations for Long-form Question Answering", author = "Xu, Fangyuan and Song, Yixiao and Iyyer, Mohit and Choi, Eunsol", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.181", doi = "10.18653/v1/2023.acl-long.181", pages = "3225--3245", abstract = "Long-form question answering (LFQA) enables answering a wide range of questions, but its flexibility poses enormous challenges for evaluation. We perform the first targeted study of the evaluation of long-form answers, covering both human and automatic evaluation practices. We hire domain experts in seven areas to provide preference judgments over pairs of answers, along with free-form justifications for their choices. We present a careful analysis of experts{'} evaluation, which focuses on new aspects such as the comprehensiveness of the answer. Next, we examine automatic text generation metrics, finding that no existing metrics are predictive of human preference judgments. However, some metrics correlate with fine-grained aspects of answers (e.g., coherence). We encourage future work to move away from a single {``}overall score{''} of the answer and adopt a multi-faceted evaluation, targeting aspects such as factuality and completeness. We publicly release all of our annotations and code to spur future work into LFQA evaluation.", }

Long-form question answering (LFQA) enables answering a wide range of questions, but its flexibility poses enormous challenges for evaluation. We perform the first targeted study of the evaluation of long-form answers, covering both human and automatic evaluation practices. We hire domain experts in seven areas to provide preference judgments over pairs of answers, along with free-form justifications for their choices. We present a careful analysis of experts{'} evaluation, which focuses on new aspects such as the comprehensiveness of the answer. Next, we examine automatic text generation metrics, finding that no existing metrics are predictive of human preference judgments. However, some metrics correlate with fine-grained aspects of answers (e.g., coherence). We encourage future work to move away from a single {``}overall score{''} of the answer and adopt a multi-faceted evaluation, targeting aspects such as factuality and completeness. We publicly release all of our annotations and code to spur future work into LFQA evaluation.

A Critical Evaluation of Evaluations for Long-form Question Answering

acl-long.181

Oral

2305.18201

https://aclanthology.org/2023.acl-long.182.bib

https://aclanthology.org/2023.acl-long.182/

@inproceedings{yuan-etal-2023-hype, title = "{H}y{P}e: Better Pre-trained Language Model Fine-tuning with Hidden Representation Perturbation", author = "Yuan, Hongyi and Yuan, Zheng and Tan, Chuanqi and Huang, Fei and Huang, Songfang", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.182", doi = "10.18653/v1/2023.acl-long.182", pages = "3246--3264", abstract = "Language models with the Transformers structure have shown great performance in natural language processing. However, there still poses problems when fine-tuning pre-trained language models on downstream tasks, such as over-fitting or representation collapse. In this work, we propose HyPe, a simple yet effective fine-tuning technique to alleviate such problems by perturbing hidden representations of Transformers layers. Unlike previous works that only add noise to inputs or parameters, we argue that the hidden representations of Transformers layers convey more diverse and meaningful language information. Therefore, making the Transformers layers more robust to hidden representation perturbations can further benefit the fine-tuning of PLMs en bloc. We conduct extensive experiments and analyses on GLUE and other natural language inference datasets. Results demonstrate that HyPe outperforms vanilla fine-tuning and enhances generalization of hidden representations from different layers. In addition, HyPe acquires negligible computational overheads, and is better than and compatible with previous state-of-the-art fine-tuning techniques.", }

Language models with the Transformers structure have shown great performance in natural language processing. However, there still poses problems when fine-tuning pre-trained language models on downstream tasks, such as over-fitting or representation collapse. In this work, we propose HyPe, a simple yet effective fine-tuning technique to alleviate such problems by perturbing hidden representations of Transformers layers. Unlike previous works that only add noise to inputs or parameters, we argue that the hidden representations of Transformers layers convey more diverse and meaningful language information. Therefore, making the Transformers layers more robust to hidden representation perturbations can further benefit the fine-tuning of PLMs en bloc. We conduct extensive experiments and analyses on GLUE and other natural language inference datasets. Results demonstrate that HyPe outperforms vanilla fine-tuning and enhances generalization of hidden representations from different layers. In addition, HyPe acquires negligible computational overheads, and is better than and compatible with previous state-of-the-art fine-tuning techniques.

HyPe: Better Pre-trained Language Model Fine-tuning with Hidden Representation Perturbation

acl-long.182

Poster

2212.08853

https://aclanthology.org/2023.acl-long.183.bib

https://aclanthology.org/2023.acl-long.183/

@inproceedings{cai-etal-2023-generating, title = "Generating User-Engaging News Headlines", author = "Cai, Pengshan and Song, Kaiqiang and Cho, Sangwoo and Wang, Hongwei and Wang, Xiaoyang and Yu, Hong and Liu, Fei and Yu, Dong", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.183", doi = "10.18653/v1/2023.acl-long.183", pages = "3265--3280", abstract = "The potential choices for news article headlines are enormous, and finding the right balance between conveying the essential message and capturing the reader{'}s attention is key to effective headlining. However, presenting the same news headline to all readers is a suboptimal strategy, because it does not take into account the different preferences and interests of diverse readers, who may be confused about why a particular article has been recommended to them and do not see a clear connection between their interests and the recommended article. In this paper, we present a novel framework that addresses these challenges by incorporating user profiling to generate personalized headlines, and a combination of automated and human evaluation methods to determine user preference for personalized headlines. Our framework utilizes a learnable relevance function to assign personalized signature phrases to users based on their reading histories, which are then used to personalize headline generation. Through extensive evaluation, we demonstrate the effectiveness of our proposed framework in generating personalized headlines that meet the needs of a diverse audience. Our framework has the potential to improve the efficacy of news recommendations and facilitate creation of personalized content.", }

The potential choices for news article headlines are enormous, and finding the right balance between conveying the essential message and capturing the reader{'}s attention is key to effective headlining. However, presenting the same news headline to all readers is a suboptimal strategy, because it does not take into account the different preferences and interests of diverse readers, who may be confused about why a particular article has been recommended to them and do not see a clear connection between their interests and the recommended article. In this paper, we present a novel framework that addresses these challenges by incorporating user profiling to generate personalized headlines, and a combination of automated and human evaluation methods to determine user preference for personalized headlines. Our framework utilizes a learnable relevance function to assign personalized signature phrases to users based on their reading histories, which are then used to personalize headline generation. Through extensive evaluation, we demonstrate the effectiveness of our proposed framework in generating personalized headlines that meet the needs of a diverse audience. Our framework has the potential to improve the efficacy of news recommendations and facilitate creation of personalized content.

Generating User-Engaging News Headlines

acl-long.183

Poster

https://aclanthology.org/2023.acl-long.184.bib

https://aclanthology.org/2023.acl-long.184/

@inproceedings{yu-xu-2023-word, title = "Word sense extension", author = "Yu, Lei and Xu, Yang", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.184", doi = "10.18653/v1/2023.acl-long.184", pages = "3281--3294", abstract = "Humans often make creative use of words to expressnovel senses. A long-standing effort in natural language processing hasbeen focusing on word sense disambiguation (WSD), but little has been explored about how the sense inventory of a word may be extended toward novel meanings. We present a paradigm of word sense extension (WSE) thatenables words to spawn new senses toward novel context. We develop a framework that simulates novel word sense extension by first partitioning a polysemous word type into two pseudo-tokens that mark its different senses, and then inferring whether the meaning of a pseudo-token can be extended to convey the sense denoted by the token partitioned from the same word type. Our framework combines cognitivemodels of chaining with a learning scheme that transforms a language model embedding space to supportvarious types of word sense extension. We evaluate our frameworkagainst several competitive baselines and show that it is superior in predicting plausible novel senses for over 7,500 English words. Furthermore, we show that our WSE framework improves performance over a range of transformer-based WSD models in predicting rare word senses with few or zero mentions in the training data.", }

Humans often make creative use of words to expressnovel senses. A long-standing effort in natural language processing hasbeen focusing on word sense disambiguation (WSD), but little has been explored about how the sense inventory of a word may be extended toward novel meanings. We present a paradigm of word sense extension (WSE) thatenables words to spawn new senses toward novel context. We develop a framework that simulates novel word sense extension by first partitioning a polysemous word type into two pseudo-tokens that mark its different senses, and then inferring whether the meaning of a pseudo-token can be extended to convey the sense denoted by the token partitioned from the same word type. Our framework combines cognitivemodels of chaining with a learning scheme that transforms a language model embedding space to supportvarious types of word sense extension. We evaluate our frameworkagainst several competitive baselines and show that it is superior in predicting plausible novel senses for over 7,500 English words. Furthermore, we show that our WSE framework improves performance over a range of transformer-based WSD models in predicting rare word senses with few or zero mentions in the training data.

Word sense extension

acl-long.184

Poster

2306.05609

https://aclanthology.org/2023.acl-long.185.bib

https://aclanthology.org/2023.acl-long.185/

@inproceedings{liu-etal-2023-pvgru, title = "{PVGRU}: Generating Diverse and Relevant Dialogue Responses via Pseudo-Variational Mechanism", author = {Liu, Yongkang and Feng, Shi and Wang, Daling and Zhang, Yifei and Sch{\"u}tze, Hinrich}, editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.185", doi = "10.18653/v1/2023.acl-long.185", pages = "3295--3310", abstract = "We investigate response generation for multi-turn dialogue in generative chatbots. Existing generative modelsbased on RNNs (Recurrent Neural Networks) usually employ the last hidden state to summarize the history, which makesmodels unable to capture the subtle variability observed in different dialogues and cannot distinguish the differencesbetween dialogues that are similar in composition. In this paper, we propose Pseudo-Variational Gated Recurrent Unit (PVGRU). The key novelty of PVGRU is a recurrent summarizing variable thataggregates the accumulated distribution variations of subsequences. We train PVGRU without relying on posterior knowledge, thus avoiding the training-inference inconsistency problem. PVGRU can perceive subtle semantic variability through summarizing variables that are optimized by two objectives we employ for training: distribution consistency and reconstruction. In addition, we build a Pseudo-Variational Hierarchical Dialogue(PVHD) model based on PVGRU. Experimental results demonstrate that PVGRU can broadly improve the diversity andrelevance of responses on two benchmark datasets.", }

We investigate response generation for multi-turn dialogue in generative chatbots. Existing generative modelsbased on RNNs (Recurrent Neural Networks) usually employ the last hidden state to summarize the history, which makesmodels unable to capture the subtle variability observed in different dialogues and cannot distinguish the differencesbetween dialogues that are similar in composition. In this paper, we propose Pseudo-Variational Gated Recurrent Unit (PVGRU). The key novelty of PVGRU is a recurrent summarizing variable thataggregates the accumulated distribution variations of subsequences. We train PVGRU without relying on posterior knowledge, thus avoiding the training-inference inconsistency problem. PVGRU can perceive subtle semantic variability through summarizing variables that are optimized by two objectives we employ for training: distribution consistency and reconstruction. In addition, we build a Pseudo-Variational Hierarchical Dialogue(PVHD) model based on PVGRU. Experimental results demonstrate that PVGRU can broadly improve the diversity andrelevance of responses on two benchmark datasets.

PVGRU: Generating Diverse and Relevant Dialogue Responses via Pseudo-Variational Mechanism

acl-long.185

Poster

2212.09086

https://aclanthology.org/2023.acl-long.186.bib

https://aclanthology.org/2023.acl-long.186/

@inproceedings{guo-etal-2023-decoding, title = "Decoding Symbolism in Language Models", author = "Guo, Meiqi and Hwa, Rebecca and Kovashka, Adriana", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.186", doi = "10.18653/v1/2023.acl-long.186", pages = "3311--3324", abstract = "This work explores the feasibility of eliciting knowledge from language models (LMs) to decode symbolism, recognizing something (e.g.,roses) as a stand-in for another (e.g., love). We present our evaluative framework, Symbolism Analysis (SymbA), which compares LMs (e.g., RoBERTa, GPT-J) on different types of symbolism and analyze the outcomes along multiple metrics. Our findings suggest that conventional symbols are more reliably elicited from LMs while situated symbols are more challenging. Results also reveal the negative impact of the bias in pre-trained corpora. We further demonstrate that a simple re-ranking strategy can mitigate the bias and significantly improve model performances to be on par with human performances in some cases.", }

This work explores the feasibility of eliciting knowledge from language models (LMs) to decode symbolism, recognizing something (e.g.,roses) as a stand-in for another (e.g., love). We present our evaluative framework, Symbolism Analysis (SymbA), which compares LMs (e.g., RoBERTa, GPT-J) on different types of symbolism and analyze the outcomes along multiple metrics. Our findings suggest that conventional symbols are more reliably elicited from LMs while situated symbols are more challenging. Results also reveal the negative impact of the bias in pre-trained corpora. We further demonstrate that a simple re-ranking strategy can mitigate the bias and significantly improve model performances to be on par with human performances in some cases.

Decoding Symbolism in Language Models

acl-long.186

Poster

https://aclanthology.org/2023.acl-long.187.bib

https://aclanthology.org/2023.acl-long.187/

@inproceedings{wang-etal-2023-survey, title = "A Survey on Zero Pronoun Translation", author = "Wang, Longyue and Liu, Siyou and Xu, Mingzhou and Song, Linfeng and Shi, Shuming and Tu, Zhaopeng", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.187", doi = "10.18653/v1/2023.acl-long.187", pages = "3325--3339", abstract = "Zero pronouns (ZPs) are frequently omitted in pro-drop languages (e.g. Chinese, Hungarian, and Hindi), but should be recalled in non-pro-drop languages (e.g. English). This phenomenon has been studied extensively in machine translation (MT), as it poses a significant challenge for MT systems due to the difficulty in determining the correct antecedent for the pronoun. This survey paper highlights the major works that have been undertaken in zero pronoun translation (ZPT) after the neural revolution so that researchers can recognize the current state and future directions of this field. We provide an organization of the literature based on evolution, dataset, method, and evaluation. In addition, we compare and analyze competing models and evaluation metrics on different benchmarks. We uncover a number of insightful findings such as: 1) ZPT is in line with the development trend of large language model; 2) data limitation causes learning bias in languages and domains; 3) performance improvements are often reported on single benchmarks, but advanced methods are still far from real-world use; 4) general-purpose metrics are not reliable on nuances and complexities of ZPT, emphasizing the necessity of targeted metrics; 5) apart from commonly-cited errors, ZPs will cause risks of gender bias.", }

Zero pronouns (ZPs) are frequently omitted in pro-drop languages (e.g. Chinese, Hungarian, and Hindi), but should be recalled in non-pro-drop languages (e.g. English). This phenomenon has been studied extensively in machine translation (MT), as it poses a significant challenge for MT systems due to the difficulty in determining the correct antecedent for the pronoun. This survey paper highlights the major works that have been undertaken in zero pronoun translation (ZPT) after the neural revolution so that researchers can recognize the current state and future directions of this field. We provide an organization of the literature based on evolution, dataset, method, and evaluation. In addition, we compare and analyze competing models and evaluation metrics on different benchmarks. We uncover a number of insightful findings such as: 1) ZPT is in line with the development trend of large language model; 2) data limitation causes learning bias in languages and domains; 3) performance improvements are often reported on single benchmarks, but advanced methods are still far from real-world use; 4) general-purpose metrics are not reliable on nuances and complexities of ZPT, emphasizing the necessity of targeted metrics; 5) apart from commonly-cited errors, ZPs will cause risks of gender bias.

A Survey on Zero Pronoun Translation

acl-long.187

Poster

2305.10196

https://aclanthology.org/2023.acl-long.188.bib

https://aclanthology.org/2023.acl-long.188/

@inproceedings{testa-etal-2023-understand, title = "We Understand Elliptical Sentences, and Language Models should Too: A New Dataset for Studying Ellipsis and its Interaction with Thematic Fit", author = "Testa, Davide and Chersoni, Emmanuele and Lenci, Alessandro", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.188", doi = "10.18653/v1/2023.acl-long.188", pages = "3340--3353", abstract = "Ellipsis is a linguistic phenomenon characterized by the omission of one or more sentence elements. Solving such a linguistic construction is not a trivial issue in natural language processing since it involves the retrieval of non-overtly expressed verbal material, which might in turn require the model to integrate human-like syntactic and semantic knowledge. In this paper, we explored the issue of how the prototypicality of event participants affects the ability of Language Models (LMs) to handle elliptical sentences and to identify the omitted arguments at different degrees of thematic fit, ranging from highly typical participants to semantically anomalous ones. With this purpose in mind, we built ELLie, the first dataset composed entirely of utterances containing different types of elliptical constructions, and structurally suited for evaluating the effect of argument thematic fit in solving ellipsis and reconstructing the missing element. Our tests demonstrated that the probability scores assigned by the models are higher for typical events than for atypical and impossible ones in different elliptical contexts, confirming the influence of prototypicality of the event participants in interpreting such linguistic structures. Finally, we conducted a retrieval task of the elided verb in the sentence in which the low performance of LMs highlighted a considerable difficulty in reconstructing the correct event.", }

Ellipsis is a linguistic phenomenon characterized by the omission of one or more sentence elements. Solving such a linguistic construction is not a trivial issue in natural language processing since it involves the retrieval of non-overtly expressed verbal material, which might in turn require the model to integrate human-like syntactic and semantic knowledge. In this paper, we explored the issue of how the prototypicality of event participants affects the ability of Language Models (LMs) to handle elliptical sentences and to identify the omitted arguments at different degrees of thematic fit, ranging from highly typical participants to semantically anomalous ones. With this purpose in mind, we built ELLie, the first dataset composed entirely of utterances containing different types of elliptical constructions, and structurally suited for evaluating the effect of argument thematic fit in solving ellipsis and reconstructing the missing element. Our tests demonstrated that the probability scores assigned by the models are higher for typical events than for atypical and impossible ones in different elliptical contexts, confirming the influence of prototypicality of the event participants in interpreting such linguistic structures. Finally, we conducted a retrieval task of the elided verb in the sentence in which the low performance of LMs highlighted a considerable difficulty in reconstructing the correct event.

We Understand Elliptical Sentences, and Language Models should Too: A New Dataset for Studying Ellipsis and its Interaction with Thematic Fit

acl-long.188

Poster

https://aclanthology.org/2023.acl-long.189.bib

https://aclanthology.org/2023.acl-long.189/

@inproceedings{ahn-etal-2023-mpchat, title = "{MPCHAT}: Towards Multimodal Persona-Grounded Conversation", author = "Ahn, Jaewoo and Song, Yeda and Yun, Sangdoo and Kim, Gunhee", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.189", doi = "10.18653/v1/2023.acl-long.189", pages = "3354--3377", abstract = "In order to build self-consistent personalized dialogue agents, previous research has mostly focused on textual persona that delivers personal facts or personalities. However, to fully describe the multi-faceted nature of persona, image modality can help better reveal the speaker{'}s personal characteristics and experiences in episodic memory (Rubin et al., 2003; Conway, 2009). In this work, we extend persona-based dialogue to the multimodal domain and make two main contributions. First, we present the first multimodal persona-based dialogue dataset named MPCHAT, which extends persona with both text and images to contain episodic memories. Second, we empirically show that incorporating multimodal persona, as measured by three proposed multimodal persona-grounded dialogue tasks (i.e., next response prediction, grounding persona prediction, and speaker identification), leads to statistically significant performance improvements across all tasks. Thus, our work highlights that multimodal persona is crucial for improving multimodal dialogue comprehension, and our MPCHAT serves as a high-quality resource for this research.", }

In order to build self-consistent personalized dialogue agents, previous research has mostly focused on textual persona that delivers personal facts or personalities. However, to fully describe the multi-faceted nature of persona, image modality can help better reveal the speaker{'}s personal characteristics and experiences in episodic memory (Rubin et al., 2003; Conway, 2009). In this work, we extend persona-based dialogue to the multimodal domain and make two main contributions. First, we present the first multimodal persona-based dialogue dataset named MPCHAT, which extends persona with both text and images to contain episodic memories. Second, we empirically show that incorporating multimodal persona, as measured by three proposed multimodal persona-grounded dialogue tasks (i.e., next response prediction, grounding persona prediction, and speaker identification), leads to statistically significant performance improvements across all tasks. Thus, our work highlights that multimodal persona is crucial for improving multimodal dialogue comprehension, and our MPCHAT serves as a high-quality resource for this research.

MPCHAT: Towards Multimodal Persona-Grounded Conversation

acl-long.189

Poster

2305.17388

https://aclanthology.org/2023.acl-long.190.bib

https://aclanthology.org/2023.acl-long.190/

@inproceedings{yang-etal-2023-doc, title = "{DOC}: Improving Long Story Coherence With Detailed Outline Control", author = "Yang, Kevin and Klein, Dan and Peng, Nanyun and Tian, Yuandong", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.190", doi = "10.18653/v1/2023.acl-long.190", pages = "3378--3465", abstract = "We propose the Detailed Outline Control (DOC) framework for improving long-range plot coherence when automatically generating several-thousand-word-long stories. DOC consists of two complementary components: a detailed outliner and a detailed controller. The detailed outliner creates a more detailed, hierarchically structured outline, shifting creative burden from the main drafting procedure to the planning stage. The detailed controller ensures the more detailed outline is still respected during generation by controlling story passages to align with outline details. In human evaluations of automatically generated stories, DOC substantially outperforms a strong Re3 baseline (Yang et al., 2022) on plot coherence (22.5{\%} absolute gain), outline relevance (28.2{\%}), and interestingness (20.7{\%}). Humans also judged DOC to be much more controllable in an interactive generation setting.", }

We propose the Detailed Outline Control (DOC) framework for improving long-range plot coherence when automatically generating several-thousand-word-long stories. DOC consists of two complementary components: a detailed outliner and a detailed controller. The detailed outliner creates a more detailed, hierarchically structured outline, shifting creative burden from the main drafting procedure to the planning stage. The detailed controller ensures the more detailed outline is still respected during generation by controlling story passages to align with outline details. In human evaluations of automatically generated stories, DOC substantially outperforms a strong Re3 baseline (Yang et al., 2022) on plot coherence (22.5{\%} absolute gain), outline relevance (28.2{\%}), and interestingness (20.7{\%}). Humans also judged DOC to be much more controllable in an interactive generation setting.

DOC: Improving Long Story Coherence With Detailed Outline Control

acl-long.190

Poster

2212.10077

https://aclanthology.org/2023.acl-long.191.bib

https://aclanthology.org/2023.acl-long.191/

@inproceedings{li-etal-2023-dual, title = "Dual-Alignment Pre-training for Cross-lingual Sentence Embedding", author = "Li, Ziheng and Huang, Shaohan and Zhang, Zihan and Deng, Zhi-Hong and Lou, Qiang and Huang, Haizhen and Jiao, Jian and Wei, Furu and Deng, Weiwei and Zhang, Qi", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.191", doi = "10.18653/v1/2023.acl-long.191", pages = "3466--3478", abstract = "Recent studies have shown that dual encoder models trained with the sentence-level translation ranking task are effective methods for cross-lingual sentence embedding. However, our research indicates that token-level alignment is also crucial in multilingual scenarios, which has not been fully explored previously. Based on our findings, we propose a dual-alignment pre-training (DAP) framework for cross-lingual sentence embedding that incorporates both sentence-level and token-level alignment. To achieve this, we introduce a novel representation translation learning (RTL) task, where the model learns to use one-side contextualized token representation to reconstruct its translation counterpart. This reconstruction objective encourages the model to embed translation information into the token representation. Compared to other token-level alignment methods such as translation language modeling, RTL is more suitable for dual encoder architectures and is computationally efficient. Extensive experiments on three sentence-level cross-lingual benchmarks demonstrate that our approach can significantly improve sentence embedding. Our code is available at \url{https://github.com/ChillingDream/DAP}.", }

Recent studies have shown that dual encoder models trained with the sentence-level translation ranking task are effective methods for cross-lingual sentence embedding. However, our research indicates that token-level alignment is also crucial in multilingual scenarios, which has not been fully explored previously. Based on our findings, we propose a dual-alignment pre-training (DAP) framework for cross-lingual sentence embedding that incorporates both sentence-level and token-level alignment. To achieve this, we introduce a novel representation translation learning (RTL) task, where the model learns to use one-side contextualized token representation to reconstruct its translation counterpart. This reconstruction objective encourages the model to embed translation information into the token representation. Compared to other token-level alignment methods such as translation language modeling, RTL is more suitable for dual encoder architectures and is computationally efficient. Extensive experiments on three sentence-level cross-lingual benchmarks demonstrate that our approach can significantly improve sentence embedding. Our code is available at \url{https://github.com/ChillingDream/DAP}.

Dual-Alignment Pre-training for Cross-lingual Sentence Embedding

acl-long.191

Poster

2305.09148

https://aclanthology.org/2023.acl-long.192.bib

https://aclanthology.org/2023.acl-long.192/

@inproceedings{lan-etal-2023-exploring, title = "Exploring Better Text Image Translation with Multimodal Codebook", author = "Lan, Zhibin and Yu, Jiawei and Li, Xiang and Zhang, Wen and Luan, Jian and Wang, Bin and Huang, Degen and Su, Jinsong", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.192", doi = "10.18653/v1/2023.acl-long.192", pages = "3479--3491", abstract = "Text image translation (TIT) aims to translate the source texts embedded in the image to target translations, which has a wide range of applications and thus has important research value. However, current studies on TIT are confronted with two main bottlenecks: 1) this task lacks a publicly available TIT dataset, 2) dominant models are constructed in a cascaded manner, which tends to suffer from the error propagation of optical character recognition (OCR). In this work, we first annotate a Chinese-English TIT dataset named OCRMT30K, providing convenience for subsequent studies. Then, we propose a TIT model with a multimodal codebook, which is able to associate the image with relevant texts, providing useful supplementary information for translation. Moreover, we present a multi-stage training framework involving text machine translation, image-text alignment, and TIT tasks, which fully exploits additional bilingual texts, OCR dataset and our OCRMT30K dataset to train our model. Extensive experiments and in-depth analyses strongly demonstrate the effectiveness of our proposed model and training framework.", }

Text image translation (TIT) aims to translate the source texts embedded in the image to target translations, which has a wide range of applications and thus has important research value. However, current studies on TIT are confronted with two main bottlenecks: 1) this task lacks a publicly available TIT dataset, 2) dominant models are constructed in a cascaded manner, which tends to suffer from the error propagation of optical character recognition (OCR). In this work, we first annotate a Chinese-English TIT dataset named OCRMT30K, providing convenience for subsequent studies. Then, we propose a TIT model with a multimodal codebook, which is able to associate the image with relevant texts, providing useful supplementary information for translation. Moreover, we present a multi-stage training framework involving text machine translation, image-text alignment, and TIT tasks, which fully exploits additional bilingual texts, OCR dataset and our OCRMT30K dataset to train our model. Extensive experiments and in-depth analyses strongly demonstrate the effectiveness of our proposed model and training framework.

Exploring Better Text Image Translation with Multimodal Codebook

acl-long.192

Poster

2305.17415

https://aclanthology.org/2023.acl-long.193.bib

https://aclanthology.org/2023.acl-long.193/

@inproceedings{zhang-etal-2023-fedlegal, title = "{FEDLEGAL}: The First Real-World Federated Learning Benchmark for Legal {NLP}", author = "Zhang, Zhuo and Hu, Xiangjing and Zhang, Jingyuan and Zhang, Yating and Wang, Hui and Qu, Lizhen and Xu, Zenglin", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.193", doi = "10.18653/v1/2023.acl-long.193", pages = "3492--3507", abstract = "The inevitable private information in legal data necessitates legal artificial intelligence to study privacy-preserving and decentralized learning methods. Federated learning (FL) has merged as a promising technique for multiple participants to collaboratively train a shared model while efficiently protecting the sensitive data of participants. However, to the best of our knowledge, there is no work on applying FL to legal NLP. To fill this gap, this paper presents the first real-world FL benchmark for legal NLP, coined FEDLEGAL, which comprises five legal NLP tasks and one privacy task based on the data from Chinese courts. Based on the extensive experiments on these datasets, our results show that FL faces new challenges in terms of real-world non-IID data. The benchmark also encourages researchers to investigate privacy protection using real-world data in the FL setting, as well as deploying models in resource-constrained scenarios. The code and datasets of FEDLEGAL are available here.", }

The inevitable private information in legal data necessitates legal artificial intelligence to study privacy-preserving and decentralized learning methods. Federated learning (FL) has merged as a promising technique for multiple participants to collaboratively train a shared model while efficiently protecting the sensitive data of participants. However, to the best of our knowledge, there is no work on applying FL to legal NLP. To fill this gap, this paper presents the first real-world FL benchmark for legal NLP, coined FEDLEGAL, which comprises five legal NLP tasks and one privacy task based on the data from Chinese courts. Based on the extensive experiments on these datasets, our results show that FL faces new challenges in terms of real-world non-IID data. The benchmark also encourages researchers to investigate privacy protection using real-world data in the FL setting, as well as deploying models in resource-constrained scenarios. The code and datasets of FEDLEGAL are available here.

FEDLEGAL: The First Real-World Federated Learning Benchmark for Legal NLP

acl-long.193

Poster

https://aclanthology.org/2023.acl-long.194.bib

https://aclanthology.org/2023.acl-long.194/

@inproceedings{gu-etal-2023-gradient, title = "A Gradient Control Method for Backdoor Attacks on Parameter-Efficient Tuning", author = "Gu, Naibin and Fu, Peng and Liu, Xiyu and Liu, Zhengxiao and Lin, Zheng and Wang, Weiping", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.194", doi = "10.18653/v1/2023.acl-long.194", pages = "3508--3520", abstract = "Parameter-Efficient Tuning (PET) has shown remarkable performance by fine-tuning only a small number of parameters of the pre-trained language models (PLMs) for the downstream tasks, while it is also possible to construct backdoor attacks due to the vulnerability of pre-trained weights. However, a large reduction in the number of attackable parameters in PET will cause the user{'}s fine-tuning to greatly affect the effectiveness of backdoor attacks, resulting in backdoor forgetting. We find that the backdoor injection process can be regarded as multi-task learning, which has a convergence imbalance problem between the training of clean and poisoned data. And this problem might result in forgetting the backdoor. Based on this finding, we propose a gradient control method to consolidate the attack effect, comprising two strategies. One controls the gradient magnitude distribution cross layers within one task and the other prevents the conflict of gradient directions between tasks. Compared with previous backdoor attack methods in the scenario of PET, our method improve the effect of the attack on sentiment classification and spam detection respectively, which shows that our method is widely applicable to different tasks.", }

Parameter-Efficient Tuning (PET) has shown remarkable performance by fine-tuning only a small number of parameters of the pre-trained language models (PLMs) for the downstream tasks, while it is also possible to construct backdoor attacks due to the vulnerability of pre-trained weights. However, a large reduction in the number of attackable parameters in PET will cause the user{'}s fine-tuning to greatly affect the effectiveness of backdoor attacks, resulting in backdoor forgetting. We find that the backdoor injection process can be regarded as multi-task learning, which has a convergence imbalance problem between the training of clean and poisoned data. And this problem might result in forgetting the backdoor. Based on this finding, we propose a gradient control method to consolidate the attack effect, comprising two strategies. One controls the gradient magnitude distribution cross layers within one task and the other prevents the conflict of gradient directions between tasks. Compared with previous backdoor attack methods in the scenario of PET, our method improve the effect of the attack on sentiment classification and spam detection respectively, which shows that our method is widely applicable to different tasks.

A Gradient Control Method for Backdoor Attacks on Parameter-Efficient Tuning

acl-long.194

Poster

https://aclanthology.org/2023.acl-long.195.bib

https://aclanthology.org/2023.acl-long.195/

@inproceedings{sun-etal-2023-history, title = "History Semantic Graph Enhanced Conversational {KBQA} with Temporal Information Modeling", author = "Sun, Hao and Li, Yang and Deng, Liwei and Li, Bowen and Hui, Binyuan and Li, Binhua and Lan, Yunshi and Zhang, Yan and Li, Yongbin", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.195", doi = "10.18653/v1/2023.acl-long.195", pages = "3521--3533", abstract = "Context information modeling is an important task in conversational KBQA. However, existing methods usually assume the independence of utterances and model them in isolation. In this paper, we propose a History Semantic Graph Enhanced KBQA model (HSGE) that is able to effectively model long-range semantic dependencies in conversation history while maintaining low computational cost. The framework incorporates a context-aware encoder, which employs a dynamic memory decay mechanism and models context at different levels of granularity. We evaluate HSGE on a widely used benchmark dataset for complex sequential question answering. Experimental results demonstrate that it outperforms existing baselines averaged on all question types.", }

Context information modeling is an important task in conversational KBQA. However, existing methods usually assume the independence of utterances and model them in isolation. In this paper, we propose a History Semantic Graph Enhanced KBQA model (HSGE) that is able to effectively model long-range semantic dependencies in conversation history while maintaining low computational cost. The framework incorporates a context-aware encoder, which employs a dynamic memory decay mechanism and models context at different levels of granularity. We evaluate HSGE on a widely used benchmark dataset for complex sequential question answering. Experimental results demonstrate that it outperforms existing baselines averaged on all question types.

History Semantic Graph Enhanced Conversational KBQA with Temporal Information Modeling

acl-long.195

Poster

2306.06872

https://aclanthology.org/2023.acl-long.196.bib

https://aclanthology.org/2023.acl-long.196/

@inproceedings{chen-etal-2023-one, title = "From the One, Judge of the Whole: Typed Entailment Graph Construction with Predicate Generation", author = "Chen, Zhibin and Feng, Yansong and Zhao, Dongyan", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.196", doi = "10.18653/v1/2023.acl-long.196", pages = "3534--3551", abstract = "Entailment Graphs (EGs) have been constructed based on extracted corpora as a strong and explainable form to indicate context-independent entailment relation in natural languages. However, EGs built by previous methods often suffer from the severe sparsity issues, due to limited corpora available and the long-tail phenomenon of predicate distributions. In this paper, we propose a multi-stage method, Typed Predicate-Entailment Graph Generator (TP-EGG), to tackle this problem. Given several seed predicates, TP-EGG builds the graphs by generating new predicates and detecting entailment relations among them. The generative nature of TP-EGG helps us leverage the recent advances from large pretrained language models (PLMs), while avoiding the reliance on carefully prepared corpora. Experiments on benchmark datasets show that TP-EGG can generate high-quality and scale-controllable entailment graphs, achieving significant in-domain improvement over state-of-the-art EGs and boosting the performance of down-stream inference tasks.", }

Entailment Graphs (EGs) have been constructed based on extracted corpora as a strong and explainable form to indicate context-independent entailment relation in natural languages. However, EGs built by previous methods often suffer from the severe sparsity issues, due to limited corpora available and the long-tail phenomenon of predicate distributions. In this paper, we propose a multi-stage method, Typed Predicate-Entailment Graph Generator (TP-EGG), to tackle this problem. Given several seed predicates, TP-EGG builds the graphs by generating new predicates and detecting entailment relations among them. The generative nature of TP-EGG helps us leverage the recent advances from large pretrained language models (PLMs), while avoiding the reliance on carefully prepared corpora. Experiments on benchmark datasets show that TP-EGG can generate high-quality and scale-controllable entailment graphs, achieving significant in-domain improvement over state-of-the-art EGs and boosting the performance of down-stream inference tasks.

From the One, Judge of the Whole: Typed Entailment Graph Construction with Predicate Generation

acl-long.196

Poster

2306.04170

https://aclanthology.org/2023.acl-long.197.bib

https://aclanthology.org/2023.acl-long.197/

@inproceedings{chen-etal-2023-alleviating, title = "Alleviating Over-smoothing for Unsupervised Sentence Representation", author = "Chen, Nuo and Shou, Linjun and Pei, Jian and Gong, Ming and Cao, Bowen and Chang, Jianhui and Li, Jia and Jiang, Daxin", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.197", doi = "10.18653/v1/2023.acl-long.197", pages = "3552--3566", abstract = "Currently, learning better unsupervised sentence representations is the pursuit of many natural language processing communities. Lots of approaches based on pre-trained language models (PLMs) and contrastive learning have achieved promising results on this task. Experimentally, we observe that the over-smoothing problem reduces the capacity of these powerful PLMs, leading to sub-optimal sentence representations. In this paper, we present a Simple method named Self-Contrastive Learning (SSCL) to alleviate this issue, which samples negatives from PLMs intermediate layers, improving the quality of the sentence representation. Our proposed method is quite simple and can be easily extended to various state-of-the-art models for performance boosting, which can be seen as a plug-and-play contrastive framework for learning unsupervised sentence representation. Extensive results prove that SSCL brings the superior performance improvements of different strong baselines (e.g., BERT and SimCSE) on Semantic Textual Similarity and Transfer datasets", }

Currently, learning better unsupervised sentence representations is the pursuit of many natural language processing communities. Lots of approaches based on pre-trained language models (PLMs) and contrastive learning have achieved promising results on this task. Experimentally, we observe that the over-smoothing problem reduces the capacity of these powerful PLMs, leading to sub-optimal sentence representations. In this paper, we present a Simple method named Self-Contrastive Learning (SSCL) to alleviate this issue, which samples negatives from PLMs intermediate layers, improving the quality of the sentence representation. Our proposed method is quite simple and can be easily extended to various state-of-the-art models for performance boosting, which can be seen as a plug-and-play contrastive framework for learning unsupervised sentence representation. Extensive results prove that SSCL brings the superior performance improvements of different strong baselines (e.g., BERT and SimCSE) on Semantic Textual Similarity and Transfer datasets

Alleviating Over-smoothing for Unsupervised Sentence Representation

acl-long.197

Poster

2305.06154

https://aclanthology.org/2023.acl-long.198.bib

https://aclanthology.org/2023.acl-long.198/

@inproceedings{koishekenov-etal-2023-memory, title = "Memory-efficient {NLLB}-200: Language-specific Expert Pruning of a Massively Multilingual Machine Translation Model", author = "Koishekenov, Yeskendir and Berard, Alexandre and Nikoulina, Vassilina", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.198", doi = "10.18653/v1/2023.acl-long.198", pages = "3567--3585", abstract = "The recently released NLLB-200 is a set of multilingual Neural Machine Translation models that cover 202 languages. The largest model is based on a Mixture of Experts architecture and achieves SoTA results across many language pairs. It contains 54.5B parameters and requires at least four 32GB GPUs just for inference. In this work, we propose a pruning method that enables the removal of up to 80{\%} of experts without further finetuning and with a negligible loss in translation quality, which makes it feasible to run the model on a single 32GB GPU. Further analysis suggests that our pruning metrics can identify language-specific experts.", }

The recently released NLLB-200 is a set of multilingual Neural Machine Translation models that cover 202 languages. The largest model is based on a Mixture of Experts architecture and achieves SoTA results across many language pairs. It contains 54.5B parameters and requires at least four 32GB GPUs just for inference. In this work, we propose a pruning method that enables the removal of up to 80{\%} of experts without further finetuning and with a negligible loss in translation quality, which makes it feasible to run the model on a single 32GB GPU. Further analysis suggests that our pruning metrics can identify language-specific experts.

Memory-efficient NLLB-200: Language-specific Expert Pruning of a Massively Multilingual Machine Translation Model

acl-long.198

Poster

2212.09811

https://huggingface.co/papers/2212.09811

https://aclanthology.org/2023.acl-long.199.bib

https://aclanthology.org/2023.acl-long.199/

@inproceedings{held-etal-2023-damp, title = "{DAMP}: Doubly Aligned Multilingual Parser for Task-Oriented Dialogue", author = "Held, William and Hidey, Christopher and Liu, Fei and Zhu, Eric and Goel, Rahul and Yang, Diyi and Shah, Rushin", editor = "Rogers, Anna and Boyd-Graber, Jordan and Okazaki, Naoaki", booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2023", address = "Toronto, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.acl-long.199", doi = "10.18653/v1/2023.acl-long.199", pages = "3586--3604", abstract = "Modern virtual assistants use internal semantic parsing engines to convert user utterances to actionable commands. However, prior work has demonstrated multilingual models are less robust for semantic parsing compared to other tasks. In global markets such as India and Latin America, robust multilingual semantic parsing is critical as codeswitching between languages is prevalent for bilingual users. In this work we dramatically improve the zero-shot performance of a multilingual and codeswitched semantic parsing system using two stages of multilingual alignment. First, we show that contrastive alignment pretraining improves \textit{both} English performance and transfer efficiency. We then introduce a constrained optimization approach for hyperparameter-free adversarial alignment during finetuning. Our Doubly Aligned Multilingual Parser (DAMP) improves mBERT transfer performance by 3x, 6x, and 81x on the Spanglish, Hinglish and Multilingual Task Oriented Parsing benchmarks respectively and outperforms XLM-R and mT5-Large using 3.2x fewer parameters.", }

Modern virtual assistants use internal semantic parsing engines to convert user utterances to actionable commands. However, prior work has demonstrated multilingual models are less robust for semantic parsing compared to other tasks. In global markets such as India and Latin America, robust multilingual semantic parsing is critical as codeswitching between languages is prevalent for bilingual users. In this work we dramatically improve the zero-shot performance of a multilingual and codeswitched semantic parsing system using two stages of multilingual alignment. First, we show that contrastive alignment pretraining improves \textit{both} English performance and transfer efficiency. We then introduce a constrained optimization approach for hyperparameter-free adversarial alignment during finetuning. Our Doubly Aligned Multilingual Parser (DAMP) improves mBERT transfer performance by 3x, 6x, and 81x on the Spanglish, Hinglish and Multilingual Task Oriented Parsing benchmarks respectively and outperforms XLM-R and mT5-Large using 3.2x fewer parameters.

DAMP: Doubly Aligned Multilingual Parser for Task-Oriented Dialogue

acl-long.199

Poster

2212.08054

https://huggingface.co/papers/2212.08054