Datasets:

ACLMeeting

/

ACL2023-papers

like 0

Follow

Annual Meeting of the Association for Computational Linguistics 1

https://aclanthology.org/2023.acl-long.200.bib

https://aclanthology.org/2023.acl-long.200/

@inproceedings{sun-etal-2023-characters, title = "From Characters to Words: Hierarchical Pre-trained Language Model for Open-vocabulary Language Understanding", author = "Sun, Li and Luisier, Florian and Batmanghelich, Kayhan and Florencio, Dinei and Zhang, Cha", 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.200", doi = "10.18653/v1/2023.acl-long.200", pages = "3605--3620", abstract = "Current state-of-the-art models for natural language understanding require a preprocessing step to convert raw text into discrete tokens. This process known as tokenization relies on a pre-built vocabulary of words or sub-word morphemes. This fixed vocabulary limits the model{'}s robustness to spelling errors and its capacity to adapt to new domains. In this work, we introduce a novel open-vocabulary language model that adopts a hierarchical two-level approach: one at the word level and another at the sequence level. Concretely, we design an intra-word module that uses a shallow Transformer architecture to learn word representations from their characters, and a deep inter-word Transformer module that contextualizes each word representation by attending to the entire word sequence. Our model thus directly operates on character sequences with explicit awareness of word boundaries, but without biased sub-word or word-level vocabulary. Experiments on various downstream tasks show that our method outperforms strong baselines. We also demonstrate that our hierarchical model is robust to textual corruption and domain shift.", }

Current state-of-the-art models for natural language understanding require a preprocessing step to convert raw text into discrete tokens. This process known as tokenization relies on a pre-built vocabulary of words or sub-word morphemes. This fixed vocabulary limits the model{'}s robustness to spelling errors and its capacity to adapt to new domains. In this work, we introduce a novel open-vocabulary language model that adopts a hierarchical two-level approach: one at the word level and another at the sequence level. Concretely, we design an intra-word module that uses a shallow Transformer architecture to learn word representations from their characters, and a deep inter-word Transformer module that contextualizes each word representation by attending to the entire word sequence. Our model thus directly operates on character sequences with explicit awareness of word boundaries, but without biased sub-word or word-level vocabulary. Experiments on various downstream tasks show that our method outperforms strong baselines. We also demonstrate that our hierarchical model is robust to textual corruption and domain shift.

From Characters to Words: Hierarchical Pre-trained Language Model for Open-vocabulary Language Understanding

acl-long.200

Poster

2305.14571

https://aclanthology.org/2023.acl-long.201.bib

https://aclanthology.org/2023.acl-long.201/

@inproceedings{song-etal-2023-matsci, title = "{M}at{S}ci-{NLP}: Evaluating Scientific Language Models on Materials Science Language Tasks Using Text-to-Schema Modeling", author = "Song, Yu and Miret, Santiago and Liu, Bang", 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.201", doi = "10.18653/v1/2023.acl-long.201", pages = "3621--3639", abstract = "We present MatSci-NLP, a natural language benchmark for evaluating the performance of natural language processing (NLP) models on materials science text. We construct the benchmark from publicly available materials science text data to encompass seven different NLP tasks, including conventional NLP tasks like named entity recognition and relation classification, as well as NLP tasks specific to materials science, such as synthesis action retrieval which relates to creating synthesis procedures for materials. We study various BERT-based models pretrained on different scientific text corpora on MatSci-NLP to understand the impact of pretraining strategies on understanding materials science text. Given the scarcity of high-quality annotated data in the materials science domain, we perform our fine-tuning experiments with limited training data to encourage the generalize across MatSci-NLP tasks. Our experiments in this low-resource training setting show that language models pretrained on scientific text outperform BERT trained on general text. MatBERT, a model pretrained specifically on materials science journals, generally performs best for most tasks. Moreover, we propose a unified text-to-schema for multitask learning on {pasted macro {`}BENCHMARK{'}} and compare its performance with traditional fine-tuning methods. In our analysis of different training methods, we find that our proposed text-to-schema methods inspired by question-answering consistently outperform single and multitask NLP fine-tuning methods. The code and datasets are publicly available \url{https://github.com/BangLab-UdeM-Mila/NLP4MatSci-ACL23}.", }

We present MatSci-NLP, a natural language benchmark for evaluating the performance of natural language processing (NLP) models on materials science text. We construct the benchmark from publicly available materials science text data to encompass seven different NLP tasks, including conventional NLP tasks like named entity recognition and relation classification, as well as NLP tasks specific to materials science, such as synthesis action retrieval which relates to creating synthesis procedures for materials. We study various BERT-based models pretrained on different scientific text corpora on MatSci-NLP to understand the impact of pretraining strategies on understanding materials science text. Given the scarcity of high-quality annotated data in the materials science domain, we perform our fine-tuning experiments with limited training data to encourage the generalize across MatSci-NLP tasks. Our experiments in this low-resource training setting show that language models pretrained on scientific text outperform BERT trained on general text. MatBERT, a model pretrained specifically on materials science journals, generally performs best for most tasks. Moreover, we propose a unified text-to-schema for multitask learning on {pasted macro {`}BENCHMARK{'}} and compare its performance with traditional fine-tuning methods. In our analysis of different training methods, we find that our proposed text-to-schema methods inspired by question-answering consistently outperform single and multitask NLP fine-tuning methods. The code and datasets are publicly available \url{https://github.com/BangLab-UdeM-Mila/NLP4MatSci-ACL23}.

MatSci-NLP: Evaluating Scientific Language Models on Materials Science Language Tasks Using Text-to-Schema Modeling

acl-long.201

Poster

2305.08264

https://huggingface.co/papers/2305.08264

https://aclanthology.org/2023.acl-long.202.bib

https://aclanthology.org/2023.acl-long.202/

@inproceedings{wang-etal-2023-code4struct, title = "{C}ode4{S}truct: Code Generation for Few-Shot Event Structure Prediction", author = "Wang, Xingyao and Li, Sha and Ji, Heng", 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.202", doi = "10.18653/v1/2023.acl-long.202", pages = "3640--3663", abstract = "Large Language Model (LLM) trained on a mixture of text and code has demonstrated impressive capability in translating natural language (NL) into structured code. We observe that semantic structures can be conveniently translated into code and propose Code4Struct to leverage such text-to-structure translation capability to tackle structured prediction tasks. As a case study, we formulate Event Argument Extraction (EAE) as converting text into event-argument structures that can be represented as a class object using code. This alignment between structures and code enables us to take advantage of Programming Language (PL) features such as inheritance and type annotation to introduce external knowledge or add constraints. We show that, with sufficient in-context examples, formulating EAE as a code generation problem is advantageous over using variants of text-based prompts. Despite only using 20 training event instances for each event type, Code4Struct is comparable to supervised models trained on 4,202 instances and outperforms current state-of-the-art (SOTA) trained on 20-shot data by 29.5{\%} absolute F1. Code4Struct can use 10-shot training data from a sibling event type to predict arguments for zero-resource event types and outperforms the zero-shot baseline by 12{\%} absolute F1.", }

Large Language Model (LLM) trained on a mixture of text and code has demonstrated impressive capability in translating natural language (NL) into structured code. We observe that semantic structures can be conveniently translated into code and propose Code4Struct to leverage such text-to-structure translation capability to tackle structured prediction tasks. As a case study, we formulate Event Argument Extraction (EAE) as converting text into event-argument structures that can be represented as a class object using code. This alignment between structures and code enables us to take advantage of Programming Language (PL) features such as inheritance and type annotation to introduce external knowledge or add constraints. We show that, with sufficient in-context examples, formulating EAE as a code generation problem is advantageous over using variants of text-based prompts. Despite only using 20 training event instances for each event type, Code4Struct is comparable to supervised models trained on 4,202 instances and outperforms current state-of-the-art (SOTA) trained on 20-shot data by 29.5{\%} absolute F1. Code4Struct can use 10-shot training data from a sibling event type to predict arguments for zero-resource event types and outperforms the zero-shot baseline by 12{\%} absolute F1.

Code4Struct: Code Generation for Few-Shot Event Structure Prediction

acl-long.202

Poster

2210.12810

https://aclanthology.org/2023.acl-long.203.bib

https://aclanthology.org/2023.acl-long.203/

@inproceedings{parekh-etal-2023-geneva, title = "{GENEVA}: Benchmarking Generalizability for Event Argument Extraction with Hundreds of Event Types and Argument Roles", author = "Parekh, Tanmay and Hsu, I-Hung and Huang, Kuan-Hao and Chang, Kai-Wei 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.203", doi = "10.18653/v1/2023.acl-long.203", pages = "3664--3686", abstract = "Recent works in Event Argument Extraction (EAE) have focused on improving model generalizability to cater to new events and domains. However, standard benchmarking datasets like ACE and ERE cover less than 40 event types and 25 entity-centric argument roles. Limited diversity and coverage hinder these datasets from adequately evaluating the generalizability of EAE models. In this paper, we first contribute by creating a large and diverse EAE ontology. This ontology is created by transforming FrameNet, a comprehensive semantic role labeling (SRL) dataset for EAE, by exploiting the similarity between these two tasks. Then, exhaustive human expert annotations are collected to build the ontology, concluding with 115 events and 220 argument roles, with a significant portion of roles not being entities. We utilize this ontology to further introduce GENEVA, a diverse generalizability benchmarking dataset comprising four test suites aimed at evaluating models{'} ability to handle limited data and unseen event type generalization. We benchmark six EAE models from various families. The results show that owing to non-entity argument roles, even the best-performing model can only achieve 39{\%} F1 score, indicating how GENEVA provides new challenges for generalization in EAE. Overall, our large and diverse EAE ontology can aid in creating more comprehensive future resources, while GENEVA is a challenging benchmarking dataset encouraging further research for improving generalizability in EAE. The code and data can be found at \url{https://github.com/PlusLabNLP/GENEVA}.", }

Recent works in Event Argument Extraction (EAE) have focused on improving model generalizability to cater to new events and domains. However, standard benchmarking datasets like ACE and ERE cover less than 40 event types and 25 entity-centric argument roles. Limited diversity and coverage hinder these datasets from adequately evaluating the generalizability of EAE models. In this paper, we first contribute by creating a large and diverse EAE ontology. This ontology is created by transforming FrameNet, a comprehensive semantic role labeling (SRL) dataset for EAE, by exploiting the similarity between these two tasks. Then, exhaustive human expert annotations are collected to build the ontology, concluding with 115 events and 220 argument roles, with a significant portion of roles not being entities. We utilize this ontology to further introduce GENEVA, a diverse generalizability benchmarking dataset comprising four test suites aimed at evaluating models{'} ability to handle limited data and unseen event type generalization. We benchmark six EAE models from various families. The results show that owing to non-entity argument roles, even the best-performing model can only achieve 39{\%} F1 score, indicating how GENEVA provides new challenges for generalization in EAE. Overall, our large and diverse EAE ontology can aid in creating more comprehensive future resources, while GENEVA is a challenging benchmarking dataset encouraging further research for improving generalizability in EAE. The code and data can be found at \url{https://github.com/PlusLabNLP/GENEVA}.

GENEVA: Benchmarking Generalizability for Event Argument Extraction with Hundreds of Event Types and Argument Roles

acl-long.203

Poster

2205.12505

https://huggingface.co/papers/2205.12505

https://aclanthology.org/2023.acl-long.204.bib

https://aclanthology.org/2023.acl-long.204/

@inproceedings{opedal-etal-2023-efficient, title = "Efficient Semiring-Weighted {E}arley Parsing", author = "Opedal, Andreas and Zmigrod, Ran and Vieira, Tim and Cotterell, Ryan and Eisner, Jason", 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.204", doi = "10.18653/v1/2023.acl-long.204", pages = "3687--3713", abstract = "We present Earley{'}s (1970) context-free parsing algorithm as a deduction system, incorporating various known and new speed-ups. In particular, our presentation supports a known worst-case runtime improvement from Earley{'}s (1970) $O(N^3|G||R|)$, which is unworkable for the large grammars that arise in natural language processing, to $O(N^3|G|)$, which matches the complexity of CKY on a binarized version of the grammar G. Here N is the length of the sentence, |R| is the number of productions in G, and |G| is the total length of those productions. We also provide a version that achieves runtime of $O(N^3|M|)$ with $|M| \leq |G|$ when the grammar is represented compactly as a single finite-state automaton M (this is partly novel). We carefully treat the generalization to semiring-weighted deduction, preprocessing the grammar like Stolcke (1995) to eliminate the possibility of deduction cycles, and further generalize Stolcke{'}s method to compute the weights of sentence prefixes. We also provide implementation details for efficient execution, ensuring that on a preprocessed grammar, the semiring-weighted versions of our methods have the same asymptotic runtime and space requirements as the unweighted methods, including sub-cubic runtime on some grammars.", }

We present Earley{'}s (1970) context-free parsing algorithm as a deduction system, incorporating various known and new speed-ups. In particular, our presentation supports a known worst-case runtime improvement from Earley{'}s (1970) $O(N^3|G||R|)$, which is unworkable for the large grammars that arise in natural language processing, to $O(N^3|G|)$, which matches the complexity of CKY on a binarized version of the grammar G. Here N is the length of the sentence, |R| is the number of productions in G, and |G| is the total length of those productions. We also provide a version that achieves runtime of $O(N^3|M|)$ with $|M| \leq |G|$ when the grammar is represented compactly as a single finite-state automaton M (this is partly novel). We carefully treat the generalization to semiring-weighted deduction, preprocessing the grammar like Stolcke (1995) to eliminate the possibility of deduction cycles, and further generalize Stolcke{'}s method to compute the weights of sentence prefixes. We also provide implementation details for efficient execution, ensuring that on a preprocessed grammar, the semiring-weighted versions of our methods have the same asymptotic runtime and space requirements as the unweighted methods, including sub-cubic runtime on some grammars.

Efficient Semiring-Weighted Earley Parsing

acl-long.204

Poster

2307.02982

https://aclanthology.org/2023.acl-long.205.bib

https://aclanthology.org/2023.acl-long.205/

@inproceedings{scarlatos-lan-2023-tree, title = "Tree-Based Representation and Generation of Natural and Mathematical Language", author = "Scarlatos, Alexander and Lan, Andrew", 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.205", doi = "10.18653/v1/2023.acl-long.205", pages = "3714--3730", abstract = "Mathematical language in scientific communications and educational scenarios is important yet relatively understudied compared to natural languages. Recent works on mathematical language focus either on representing stand-alone mathematical expressions, especially in their natural tree format, or mathematical reasoning in pre-trained natural language models. Existing works on jointly modeling and generating natural and mathematical languages simply treat mathematical expressions as text, without accounting for the rigid structural properties of mathematical expressions. In this paper, we propose a series of modifications to existing language models to jointly represent and generate text and math: representing mathematical expressions as sequences of node tokens in their operator tree format, using math symbol and tree position embeddings to preserve the semantic and structural properties of mathematical expressions, and using a constrained decoding method to generate mathematically valid expressions. We ground our modifications in GPT-2, resulting in a model MathGPT, and demonstrate that it outperforms baselines on mathematical expression generation tasks.", }

Mathematical language in scientific communications and educational scenarios is important yet relatively understudied compared to natural languages. Recent works on mathematical language focus either on representing stand-alone mathematical expressions, especially in their natural tree format, or mathematical reasoning in pre-trained natural language models. Existing works on jointly modeling and generating natural and mathematical languages simply treat mathematical expressions as text, without accounting for the rigid structural properties of mathematical expressions. In this paper, we propose a series of modifications to existing language models to jointly represent and generate text and math: representing mathematical expressions as sequences of node tokens in their operator tree format, using math symbol and tree position embeddings to preserve the semantic and structural properties of mathematical expressions, and using a constrained decoding method to generate mathematically valid expressions. We ground our modifications in GPT-2, resulting in a model MathGPT, and demonstrate that it outperforms baselines on mathematical expression generation tasks.

Tree-Based Representation and Generation of Natural and Mathematical Language

acl-long.205

Poster

2302.07974

https://aclanthology.org/2023.acl-long.206.bib

https://aclanthology.org/2023.acl-long.206/

@inproceedings{qiang-etal-2023-parals, title = "{P}ara{LS}: Lexical Substitution via Pretrained Paraphraser", author = "Qiang, Jipeng and Liu, Kang and Li, Yun and Yuan, Yunhao and Zhu, Yi", 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.206", doi = "10.18653/v1/2023.acl-long.206", pages = "3731--3746", abstract = "Lexical substitution (LS) aims at finding appropriate substitutes for a target word in a sentence. Recently, LS methods based on pretrained language models have made remarkable progress, generating potential substitutes for a target word through analysis of its contextual surroundings. However, these methods tend to overlook the preservation of the sentence{'}s meaning when generating the substitutes. This study explores how to generate the substitute candidates from a paraphraser, as the generated paraphrases from a paraphraser contain variations in word choice and preserve the sentence{'}s meaning. Since we cannot directly generate the substitutes via commonly used decoding strategies, we propose two simple decoding strategies that focus on the variations of the target word during decoding. Experimental results show that our methods outperform state-of-the-art LS methods based on pre-trained language models on three benchmarks.", }

Lexical substitution (LS) aims at finding appropriate substitutes for a target word in a sentence. Recently, LS methods based on pretrained language models have made remarkable progress, generating potential substitutes for a target word through analysis of its contextual surroundings. However, these methods tend to overlook the preservation of the sentence{'}s meaning when generating the substitutes. This study explores how to generate the substitute candidates from a paraphraser, as the generated paraphrases from a paraphraser contain variations in word choice and preserve the sentence{'}s meaning. Since we cannot directly generate the substitutes via commonly used decoding strategies, we propose two simple decoding strategies that focus on the variations of the target word during decoding. Experimental results show that our methods outperform state-of-the-art LS methods based on pre-trained language models on three benchmarks.

ParaLS: Lexical Substitution via Pretrained Paraphraser

acl-long.206

Poster

2305.08146

https://aclanthology.org/2023.acl-long.207.bib

https://aclanthology.org/2023.acl-long.207/

@inproceedings{song-etal-2023-peer, title = "Peer-Label Assisted Hierarchical Text Classification", author = "Song, Junru and Wang, Feifei and Yang, 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.207", doi = "10.18653/v1/2023.acl-long.207", pages = "3747--3758", abstract = "Hierarchical text classification (HTC) is a challenging task, in which the labels of texts can be organized into a category hierarchy. To deal with the HTC problem, many existing works focus on utilizing the parent-child relationships that are explicitly shown in the hierarchy. However, texts with a category hierarchy also have some latent relevancy among labels in the same level of the hierarchy. We refer to these labels as peer labels, from which the peer effects are originally utilized in our work to improve the classification performance. To fully explore the peer-label relationship, we develop a PeerHTC method. This method innovatively measures the latent relevancy of peer labels through several metrics and then encodes the relevancy with a Graph Convolutional Neural Network. We also propose a sample importance learning method to ameliorate the side effects raised by modelling the peer label relevancy. Our experiments on several standard datasets demonstrate the evidence of peer labels and the superiority of PeerHTC over other state-of-the-art HTC methods in terms of classification accuracy.", }

Hierarchical text classification (HTC) is a challenging task, in which the labels of texts can be organized into a category hierarchy. To deal with the HTC problem, many existing works focus on utilizing the parent-child relationships that are explicitly shown in the hierarchy. However, texts with a category hierarchy also have some latent relevancy among labels in the same level of the hierarchy. We refer to these labels as peer labels, from which the peer effects are originally utilized in our work to improve the classification performance. To fully explore the peer-label relationship, we develop a PeerHTC method. This method innovatively measures the latent relevancy of peer labels through several metrics and then encodes the relevancy with a Graph Convolutional Neural Network. We also propose a sample importance learning method to ameliorate the side effects raised by modelling the peer label relevancy. Our experiments on several standard datasets demonstrate the evidence of peer labels and the superiority of PeerHTC over other state-of-the-art HTC methods in terms of classification accuracy.

Peer-Label Assisted Hierarchical Text Classification

acl-long.207

Poster

https://aclanthology.org/2023.acl-long.208.bib

https://aclanthology.org/2023.acl-long.208/

@inproceedings{cui-chen-2023-free, title = "Free Lunch for Efficient Textual Commonsense Integration in Language Models", author = "Cui, Wanyun and Chen, Xingran", 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.208", doi = "10.18653/v1/2023.acl-long.208", pages = "3759--3770", abstract = "Recent years have witnessed the emergence of textual commonsense knowledge bases, aimed at providing more nuanced and context-rich knowledge. The integration of external commonsense into language models has been shown to be a key enabler in advancing the state-of-the-art for a wide range of NLP tasks. However, incorporating textual commonsense descriptions is computationally expensive, as compared to encoding conventional symbolic knowledge. In this paper, we propose a method to improve its efficiency without modifying the model. Our idea is to group training samples with similar commonsense descriptions into a single batch, thus reusing the encoded description across multiple samples. We theoretically investigate this problem and demonstrate that its upper bound can be reduced to the classic \textit{graph k-cut problem}. Consequently, we propose a spectral clustering-based algorithm to solve this problem. Extensive experiments illustrate that the proposed batch partitioning approach effectively reduces the computational cost while preserving performance. The efficiency improvement is more pronounced on larger datasets and on devices with more memory capacity, attesting to its practical utility for large-scale applications.", }

Recent years have witnessed the emergence of textual commonsense knowledge bases, aimed at providing more nuanced and context-rich knowledge. The integration of external commonsense into language models has been shown to be a key enabler in advancing the state-of-the-art for a wide range of NLP tasks. However, incorporating textual commonsense descriptions is computationally expensive, as compared to encoding conventional symbolic knowledge. In this paper, we propose a method to improve its efficiency without modifying the model. Our idea is to group training samples with similar commonsense descriptions into a single batch, thus reusing the encoded description across multiple samples. We theoretically investigate this problem and demonstrate that its upper bound can be reduced to the classic \textit{graph k-cut problem}. Consequently, we propose a spectral clustering-based algorithm to solve this problem. Extensive experiments illustrate that the proposed batch partitioning approach effectively reduces the computational cost while preserving performance. The efficiency improvement is more pronounced on larger datasets and on devices with more memory capacity, attesting to its practical utility for large-scale applications.

Free Lunch for Efficient Textual Commonsense Integration in Language Models

acl-long.208

Poster

2305.15516

https://aclanthology.org/2023.acl-long.209.bib

https://aclanthology.org/2023.acl-long.209/

@inproceedings{zhou-etal-2023-probabilistic, title = "A Probabilistic Framework for Discovering New Intents", author = "Zhou, Yunhua and Quan, Guofeng and Qiu, Xipeng", 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.209", doi = "10.18653/v1/2023.acl-long.209", pages = "3771--3784", abstract = "Discovering new intents is of great significance for establishing the Task-Oriented Dialogue System. Most existing methods either cannot transfer prior knowledge contained in known intents or fall into the dilemma of forgetting prior knowledge in the follow-up. Furthermore, these methods do not deeply explore the intrinsic structure of unlabeled data, and as a result, cannot seek out the characteristics that define an intent in general. In this paper, starting from the intuition that discovering intents could be beneficial for identifying known intents, we propose a probabilistic framework for discovering intents where intent assignments are treated as latent variables. We adopt the Expectation Maximization framework for optimization. Specifically, In the E-step, we conduct intent discovery and explore the intrinsic structure of unlabeled data by the posterior of intent assignments. In the M-step, we alleviate the forgetting of prior knowledge transferred from known intents by optimizing the discrimination of labeled data. Extensive experiments conducted on three challenging real-world datasets demonstrate the generality and effectiveness of the proposed framework and implementation.", }

Discovering new intents is of great significance for establishing the Task-Oriented Dialogue System. Most existing methods either cannot transfer prior knowledge contained in known intents or fall into the dilemma of forgetting prior knowledge in the follow-up. Furthermore, these methods do not deeply explore the intrinsic structure of unlabeled data, and as a result, cannot seek out the characteristics that define an intent in general. In this paper, starting from the intuition that discovering intents could be beneficial for identifying known intents, we propose a probabilistic framework for discovering intents where intent assignments are treated as latent variables. We adopt the Expectation Maximization framework for optimization. Specifically, In the E-step, we conduct intent discovery and explore the intrinsic structure of unlabeled data by the posterior of intent assignments. In the M-step, we alleviate the forgetting of prior knowledge transferred from known intents by optimizing the discrimination of labeled data. Extensive experiments conducted on three challenging real-world datasets demonstrate the generality and effectiveness of the proposed framework and implementation.

A Probabilistic Framework for Discovering New Intents

acl-long.209

Poster

https://aclanthology.org/2023.acl-long.210.bib

https://aclanthology.org/2023.acl-long.210/

@inproceedings{hennig-etal-2023-multitacred, title = "{M}ulti{TACRED}: A Multilingual Version of the {TAC} Relation Extraction Dataset", author = {Hennig, Leonhard and Thomas, Philippe and M{\"o}ller, Sebastian}, 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.210", doi = "10.18653/v1/2023.acl-long.210", pages = "3785--3801", abstract = "Relation extraction (RE) is a fundamental task in information extraction, whose extension to multilingual settings has been hindered by the lack of supervised resources comparable in size to large English datasets such as TACRED (Zhang et al., 2017). To address this gap, we introduce the MultiTACRED dataset, covering 12 typologically diverse languages from 9 language families, which is created by machine-translating TACRED instances and automatically projecting their entity annotations. We analyze translation and annotation projection quality, identify error categories, and experimentally evaluate fine-tuned pretrained mono- and multilingual language models in common transfer learning scenarios. Our analyses show that machine translation is a viable strategy to transfer RE instances, with native speakers judging more than 83{\%} of the translated instances to be linguistically and semantically acceptable. We find monolingual RE model performance to be comparable to the English original for many of the target languages, and that multilingual models trained on a combination of English and target language data can outperform their monolingual counterparts. However, we also observe a variety of translation and annotation projection errors, both due to the MT systems and linguistic features of the target languages, such as pronoun-dropping, compounding and inflection, that degrade dataset quality and RE model performance.", }

Relation extraction (RE) is a fundamental task in information extraction, whose extension to multilingual settings has been hindered by the lack of supervised resources comparable in size to large English datasets such as TACRED (Zhang et al., 2017). To address this gap, we introduce the MultiTACRED dataset, covering 12 typologically diverse languages from 9 language families, which is created by machine-translating TACRED instances and automatically projecting their entity annotations. We analyze translation and annotation projection quality, identify error categories, and experimentally evaluate fine-tuned pretrained mono- and multilingual language models in common transfer learning scenarios. Our analyses show that machine translation is a viable strategy to transfer RE instances, with native speakers judging more than 83{\%} of the translated instances to be linguistically and semantically acceptable. We find monolingual RE model performance to be comparable to the English original for many of the target languages, and that multilingual models trained on a combination of English and target language data can outperform their monolingual counterparts. However, we also observe a variety of translation and annotation projection errors, both due to the MT systems and linguistic features of the target languages, such as pronoun-dropping, compounding and inflection, that degrade dataset quality and RE model performance.

MultiTACRED: A Multilingual Version of the TAC Relation Extraction Dataset

acl-long.210

Oral

2305.04582

https://huggingface.co/papers/2305.04582

https://aclanthology.org/2023.acl-long.211.bib

https://aclanthology.org/2023.acl-long.211/

@inproceedings{huang-etal-2023-towards, title = "Towards Higher {P}areto Frontier in Multilingual Machine Translation", author = "Huang, Yichong and Feng, Xiaocheng and Geng, Xinwei and Li, Baohang and Qin, Bing", 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.211", doi = "10.18653/v1/2023.acl-long.211", pages = "3802--3818", abstract = "Multilingual neural machine translation has witnessed remarkable progress in recent years. However, the long-tailed distribution of multilingual corpora poses a challenge of Pareto optimization, i.e., optimizing for some languages may come at the cost of degrading the performance of others. Existing balancing training strategies are equivalent to a series of Pareto optimal solutions, which trade off on a Pareto frontierIn Pareto optimization, Pareto optimal solutions refer to solutions in which none of the objectives can be improved without sacrificing at least one of the other objectives. The set of all Pareto optimal solutions forms a Pareto frontier..In this work, we propose a new training framework, Pareto Mutual Distillation (Pareto-MD), towards pushing the Pareto frontier outwards rather than making trade-offs. Specifically, Pareto-MD collaboratively trains two Pareto optimal solutions that favor different languages and allows them to learn from the strengths of each other via knowledge distillation. Furthermore, we introduce a novel strategy to enable stronger communication between Pareto optimal solutions and broaden the applicability of our approach. Experimental results on the widely-used WMT and TED datasets show that our method significantly pushes the Pareto frontier and outperforms baselines by up to +2.46 BLEUOur code will be released upon acceptance..", }

Multilingual neural machine translation has witnessed remarkable progress in recent years. However, the long-tailed distribution of multilingual corpora poses a challenge of Pareto optimization, i.e., optimizing for some languages may come at the cost of degrading the performance of others. Existing balancing training strategies are equivalent to a series of Pareto optimal solutions, which trade off on a Pareto frontierIn Pareto optimization, Pareto optimal solutions refer to solutions in which none of the objectives can be improved without sacrificing at least one of the other objectives. The set of all Pareto optimal solutions forms a Pareto frontier..In this work, we propose a new training framework, Pareto Mutual Distillation (Pareto-MD), towards pushing the Pareto frontier outwards rather than making trade-offs. Specifically, Pareto-MD collaboratively trains two Pareto optimal solutions that favor different languages and allows them to learn from the strengths of each other via knowledge distillation. Furthermore, we introduce a novel strategy to enable stronger communication between Pareto optimal solutions and broaden the applicability of our approach. Experimental results on the widely-used WMT and TED datasets show that our method significantly pushes the Pareto frontier and outperforms baselines by up to +2.46 BLEUOur code will be released upon acceptance..

Towards Higher Pareto Frontier in Multilingual Machine Translation

acl-long.211

Oral

2305.15718

https://aclanthology.org/2023.acl-long.212.bib

https://aclanthology.org/2023.acl-long.212/

@inproceedings{gao-etal-2023-small, title = "Small Pre-trained Language Models Can be Fine-tuned as Large Models via Over-Parameterization", author = "Gao, Ze-Feng and Zhou, Kun and Liu, Peiyu and Zhao, Wayne Xin and Wen, Ji-Rong", 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.212", doi = "10.18653/v1/2023.acl-long.212", pages = "3819--3834", abstract = "By scaling the model size, large pre-trained language models (PLMs) have shown remarkable performance in various natural language processing tasks, mostly outperforming small PLMs by a large margin. However, due to the high computational cost, the huge number of parameters also restricts the applicability of large PLMs in real-world systems. In this paper, we focus on scaling up the parameters of PLMs \textit{only during} fine-tuning, to benefit from the over-parameterization, while without increasing \textit{the inference latency}. Given a relatively small PLM, we over-parameterize it by employing a matrix product operator, an efficient and almost lossless decomposition method to factorize its contained parameter matrices into a set of higher-dimensional tensors.Considering the efficiency, we further propose both static and dynamic strategies to select the most important parameter matrices for over-parameterization.Extensive experiments have demonstrated that our approach can significantly boost the fine-tuning performance of small PLMs and even help small PLMs outperform $3\times$ parameterized larger ones.Our code is publicly available at \url{https://github.com/zfgao66/OPF}.", }

By scaling the model size, large pre-trained language models (PLMs) have shown remarkable performance in various natural language processing tasks, mostly outperforming small PLMs by a large margin. However, due to the high computational cost, the huge number of parameters also restricts the applicability of large PLMs in real-world systems. In this paper, we focus on scaling up the parameters of PLMs \textit{only during} fine-tuning, to benefit from the over-parameterization, while without increasing \textit{the inference latency}. Given a relatively small PLM, we over-parameterize it by employing a matrix product operator, an efficient and almost lossless decomposition method to factorize its contained parameter matrices into a set of higher-dimensional tensors.Considering the efficiency, we further propose both static and dynamic strategies to select the most important parameter matrices for over-parameterization.Extensive experiments have demonstrated that our approach can significantly boost the fine-tuning performance of small PLMs and even help small PLMs outperform $3\times$ parameterized larger ones.Our code is publicly available at \url{https://github.com/zfgao66/OPF}.

Small Pre-trained Language Models Can be Fine-tuned as Large Models via Over-Parameterization

acl-long.212

Oral

https://aclanthology.org/2023.acl-long.213.bib

https://aclanthology.org/2023.acl-long.213/

@inproceedings{kim-schuster-2023-entity, title = "Entity Tracking in Language Models", author = "Kim, Najoung and Schuster, Sebastian", 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.213", doi = "10.18653/v1/2023.acl-long.213", pages = "3835--3855", abstract = "Keeping track of how states of entities change as a text or dialog unfolds is a key prerequisite to discourse understanding. Yet, there have been few systematic investigations into the ability of large language models (LLMs) to track discourse entities. In this work, we present a task probing to what extent a language model can infer the final state of an entity given an English description of the initial state and a series of state-changing operations. We use this task to first investigate whether Flan-T5, GPT-3 and GPT-3.5 can track the state of entities, and find that only GPT-3.5 models, which have been pretrained on large amounts of code, exhibit this ability. We then investigate whether smaller models pretrained primarily on text can learn to track entities, through finetuning T5 on several training/evaluation splits. While performance degrades for more complex splits, we find that even when evaluated on a different set of entities from training or longer operation sequences, a finetuned model can perform non-trivial entity tracking. Taken together, these results suggest that language models can learn to track entities but pretraining on text corpora alone does not make this capacity surface.", }

Keeping track of how states of entities change as a text or dialog unfolds is a key prerequisite to discourse understanding. Yet, there have been few systematic investigations into the ability of large language models (LLMs) to track discourse entities. In this work, we present a task probing to what extent a language model can infer the final state of an entity given an English description of the initial state and a series of state-changing operations. We use this task to first investigate whether Flan-T5, GPT-3 and GPT-3.5 can track the state of entities, and find that only GPT-3.5 models, which have been pretrained on large amounts of code, exhibit this ability. We then investigate whether smaller models pretrained primarily on text can learn to track entities, through finetuning T5 on several training/evaluation splits. While performance degrades for more complex splits, we find that even when evaluated on a different set of entities from training or longer operation sequences, a finetuned model can perform non-trivial entity tracking. Taken together, these results suggest that language models can learn to track entities but pretraining on text corpora alone does not make this capacity surface.

Entity Tracking in Language Models

acl-long.213

Poster

2305.02363

https://aclanthology.org/2023.acl-long.214.bib

https://aclanthology.org/2023.acl-long.214/

@inproceedings{otani-etal-2023-textual, title = "A Textual Dataset for Situated Proactive Response Selection", author = "Otani, Naoki and Araki, Jun and Kim, HyeongSik and Hovy, Eduard", 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.214", doi = "10.18653/v1/2023.acl-long.214", pages = "3856--3874", abstract = "Recent data-driven conversational models are able to return fluent, consistent, and informative responses to many kinds of requests and utterances in task-oriented scenarios. However, these responses are typically limited to just the immediate local topic instead of being wider-ranging and proactively taking the conversation further, for example making suggestions to help customers achieve their goals. This inadequacy reflects a lack of understanding of the interlocutor{'}s situation and implicit goal. To address the problem, we introduce a task of proactive response selection based on situational information. We present a manually-curated dataset of 1.7k English conversation examples that include situational background information plus for each conversation a set of responses, only some of which are acceptable in the situation. A responsive and informed conversation system should select the appropriate responses and avoid inappropriate ones; doing so demonstrates the ability to adequately understand the initiating request and situation. Our benchmark experiments show that this is not an easy task even for strong neural models, offering opportunities for future research.", }

Recent data-driven conversational models are able to return fluent, consistent, and informative responses to many kinds of requests and utterances in task-oriented scenarios. However, these responses are typically limited to just the immediate local topic instead of being wider-ranging and proactively taking the conversation further, for example making suggestions to help customers achieve their goals. This inadequacy reflects a lack of understanding of the interlocutor{'}s situation and implicit goal. To address the problem, we introduce a task of proactive response selection based on situational information. We present a manually-curated dataset of 1.7k English conversation examples that include situational background information plus for each conversation a set of responses, only some of which are acceptable in the situation. A responsive and informed conversation system should select the appropriate responses and avoid inappropriate ones; doing so demonstrates the ability to adequately understand the initiating request and situation. Our benchmark experiments show that this is not an easy task even for strong neural models, offering opportunities for future research.

A Textual Dataset for Situated Proactive Response Selection

acl-long.214

Poster

https://aclanthology.org/2023.acl-long.215.bib

https://aclanthology.org/2023.acl-long.215/

@inproceedings{shen-etal-2023-diffusionner, title = "{D}iffusion{NER}: Boundary Diffusion for Named Entity Recognition", author = "Shen, Yongliang and Song, Kaitao and Tan, Xu and Li, Dongsheng and Lu, Weiming and Zhuang, Yueting", 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.215", doi = "10.18653/v1/2023.acl-long.215", pages = "3875--3890", abstract = "In this paper, we propose DiffusionNER, which formulates the named entity recognition task as a boundary-denoising diffusion process and thus generates named entities from noisy spans. During training, DiffusionNER gradually adds noises to the golden entity boundaries by a fixed forward diffusion process and learns a reverse diffusion process to recover the entity boundaries. In inference, DiffusionNER first randomly samples some noisy spans from a standard Gaussian distribution and then generates the named entities by denoising them with the learned reverse diffusion process. The proposed boundary-denoising diffusion process allows progressive refinement and dynamic sampling of entities, empowering DiffusionNER with efficient and flexible entity generation capability. Experiments on multiple flat and nested NER datasets demonstrate that DiffusionNER achieves comparable or even better performance than previous state-of-the-art models.", }

In this paper, we propose DiffusionNER, which formulates the named entity recognition task as a boundary-denoising diffusion process and thus generates named entities from noisy spans. During training, DiffusionNER gradually adds noises to the golden entity boundaries by a fixed forward diffusion process and learns a reverse diffusion process to recover the entity boundaries. In inference, DiffusionNER first randomly samples some noisy spans from a standard Gaussian distribution and then generates the named entities by denoising them with the learned reverse diffusion process. The proposed boundary-denoising diffusion process allows progressive refinement and dynamic sampling of entities, empowering DiffusionNER with efficient and flexible entity generation capability. Experiments on multiple flat and nested NER datasets demonstrate that DiffusionNER achieves comparable or even better performance than previous state-of-the-art models.

DiffusionNER: Boundary Diffusion for Named Entity Recognition

acl-long.215

Poster

2305.13298

https://aclanthology.org/2023.acl-long.216.bib

https://aclanthology.org/2023.acl-long.216/

@inproceedings{ouyang-etal-2023-waco, title = "{WACO}: Word-Aligned Contrastive Learning for Speech Translation", author = "Ouyang, Siqi and Ye, Rong and Li, 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.216", doi = "10.18653/v1/2023.acl-long.216", pages = "3891--3907", abstract = "End-to-end Speech Translation (E2E ST) aims to directly translate source speech into target text. Existing ST methods perform poorly when only extremely small speech-text data are available for training. We observe that an ST model{'}s performance closely correlates with its embedding similarity between speech and source transcript. In this paper, we propose Word-Aligned COntrastive learning (WACO), a simple and effective method for extremely low-resource speech-to-text translation. Our key idea is bridging word-level representations for both speech and text modalities via contrastive learning. We evaluate WACO and other methods on the MuST-C dataset, a widely used ST benchmark, and on a low-resource direction Maltese-English from IWSLT 2023. Our experiments demonstrate that WACO outperforms the best baseline by 9+ BLEU points with only 1-hour parallel ST data. Code is available at \url{https://github.com/owaski/WACO}.", }

End-to-end Speech Translation (E2E ST) aims to directly translate source speech into target text. Existing ST methods perform poorly when only extremely small speech-text data are available for training. We observe that an ST model{'}s performance closely correlates with its embedding similarity between speech and source transcript. In this paper, we propose Word-Aligned COntrastive learning (WACO), a simple and effective method for extremely low-resource speech-to-text translation. Our key idea is bridging word-level representations for both speech and text modalities via contrastive learning. We evaluate WACO and other methods on the MuST-C dataset, a widely used ST benchmark, and on a low-resource direction Maltese-English from IWSLT 2023. Our experiments demonstrate that WACO outperforms the best baseline by 9+ BLEU points with only 1-hour parallel ST data. Code is available at \url{https://github.com/owaski/WACO}.

WACO: Word-Aligned Contrastive Learning for Speech Translation

acl-long.216

Poster

2212.09359

https://aclanthology.org/2023.acl-long.217.bib

https://aclanthology.org/2023.acl-long.217/

@inproceedings{mhamdi-etal-2023-cross, title = "Cross-lingual Continual Learning", author = "M{'}hamdi, Meryem and Ren, Xiang and May, Jonathan", 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.217", doi = "10.18653/v1/2023.acl-long.217", pages = "3908--3943", abstract = "The longstanding goal of multi-lingual learning has been to develop a universal cross-lingual model that can withstand the changes in multi-lingual data distributions. There has been a large amount of work to adapt such multi-lingual models to unseen target languages. However, the majority of work in this direction focuses on the standard one-hop transfer learning pipeline from source to target languages, whereas in realistic scenarios, new languages can be incorporated at any time in a sequential manner. In this paper, we present a principled Cross-lingual Continual Learning (CCL) evaluation paradigm, where we analyze different categories of approaches used to continually adapt to emerging data from different languages. We provide insights into what makes multilingual sequential learning particularly challenging. To surmount such challenges, we benchmark a representative set of cross-lingual continual learning algorithms and analyze their knowledge preservation, accumulation, and generalization capabilities compared to baselines on carefully curated datastreams. The implications of this analysis include a recipe for how to measure and balance different cross-lingual continual learning desiderata, which go beyond conventional transfer learning.", }

The longstanding goal of multi-lingual learning has been to develop a universal cross-lingual model that can withstand the changes in multi-lingual data distributions. There has been a large amount of work to adapt such multi-lingual models to unseen target languages. However, the majority of work in this direction focuses on the standard one-hop transfer learning pipeline from source to target languages, whereas in realistic scenarios, new languages can be incorporated at any time in a sequential manner. In this paper, we present a principled Cross-lingual Continual Learning (CCL) evaluation paradigm, where we analyze different categories of approaches used to continually adapt to emerging data from different languages. We provide insights into what makes multilingual sequential learning particularly challenging. To surmount such challenges, we benchmark a representative set of cross-lingual continual learning algorithms and analyze their knowledge preservation, accumulation, and generalization capabilities compared to baselines on carefully curated datastreams. The implications of this analysis include a recipe for how to measure and balance different cross-lingual continual learning desiderata, which go beyond conventional transfer learning.

Cross-lingual Continual Learning

acl-long.217

Poster

https://aclanthology.org/2023.acl-long.218.bib

https://aclanthology.org/2023.acl-long.218/

@inproceedings{hong-etal-2023-faithful, title = "Faithful Question Answering with {M}onte-{C}arlo Planning", author = "Hong, Ruixin and Zhang, Hongming and Zhao, Hong and Yu, Dong and Zhang, Changshui", 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.218", doi = "10.18653/v1/2023.acl-long.218", pages = "3944--3965", abstract = "Although large language models demonstrate remarkable question-answering performances, revealing the intermediate reasoning steps that the models faithfully follow remains challenging. In this paper, we propose FAME (FAithful question answering with MontE-carlo planning) to answer questions based on faithful reasoning steps. The reasoning steps are organized as a structured entailment tree, which shows how premises are used to produce intermediate conclusions that can prove the correctness of the answer. We formulate the task as a discrete decision-making problem and solve it through the interaction of a reasoning environment and a controller. The environment is modular and contains several basic task-oriented modules, while the controller proposes actions to assemble the modules. Since the search space could be large, we introduce a Monte-Carlo planning algorithm to do a look-ahead search and select actions that will eventually lead to high-quality steps. FAME achieves advanced performance on the standard benchmark. It can produce valid and faithful reasoning steps compared with large language models with a much smaller model size.", }

Although large language models demonstrate remarkable question-answering performances, revealing the intermediate reasoning steps that the models faithfully follow remains challenging. In this paper, we propose FAME (FAithful question answering with MontE-carlo planning) to answer questions based on faithful reasoning steps. The reasoning steps are organized as a structured entailment tree, which shows how premises are used to produce intermediate conclusions that can prove the correctness of the answer. We formulate the task as a discrete decision-making problem and solve it through the interaction of a reasoning environment and a controller. The environment is modular and contains several basic task-oriented modules, while the controller proposes actions to assemble the modules. Since the search space could be large, we introduce a Monte-Carlo planning algorithm to do a look-ahead search and select actions that will eventually lead to high-quality steps. FAME achieves advanced performance on the standard benchmark. It can produce valid and faithful reasoning steps compared with large language models with a much smaller model size.

Faithful Question Answering with Monte-Carlo Planning

acl-long.218

Poster

2305.02556

https://aclanthology.org/2023.acl-long.219.bib

https://aclanthology.org/2023.acl-long.219/

@inproceedings{arase-etal-2023-unbalanced, title = "Unbalanced Optimal Transport for Unbalanced Word Alignment", author = "Arase, Yuki and Bao, Han and Yokoi, Sho", 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.219", doi = "10.18653/v1/2023.acl-long.219", pages = "3966--3986", abstract = "Monolingual word alignment is crucial to model semantic interactions between sentences. In particular, null alignment, a phenomenon in which words have no corresponding counterparts, is pervasive and critical in handling semantically divergent sentences. Identification of null alignment is useful on its own to reason about the semantic similarity of sentences by indicating there exists information inequality. To achieve unbalanced word alignment that values both alignment and null alignment, this study shows that the family of optimal transport (OT), i.e., balanced, partial, and unbalanced OT, are natural and powerful approaches even without tailor-made techniques. Our extensive experiments covering unsupervised and supervised settings indicate that our generic OT-based alignment methods are competitive against the state-of-the-arts specially designed for word alignment, remarkably on challenging datasets with high null alignment frequencies.", }

Monolingual word alignment is crucial to model semantic interactions between sentences. In particular, null alignment, a phenomenon in which words have no corresponding counterparts, is pervasive and critical in handling semantically divergent sentences. Identification of null alignment is useful on its own to reason about the semantic similarity of sentences by indicating there exists information inequality. To achieve unbalanced word alignment that values both alignment and null alignment, this study shows that the family of optimal transport (OT), i.e., balanced, partial, and unbalanced OT, are natural and powerful approaches even without tailor-made techniques. Our extensive experiments covering unsupervised and supervised settings indicate that our generic OT-based alignment methods are competitive against the state-of-the-arts specially designed for word alignment, remarkably on challenging datasets with high null alignment frequencies.

Unbalanced Optimal Transport for Unbalanced Word Alignment

acl-long.219

Poster

2306.04116

https://huggingface.co/papers/2306.04116

https://aclanthology.org/2023.acl-long.220.bib

https://aclanthology.org/2023.acl-long.220/

@inproceedings{liu-etal-2023-guiding, title = "Guiding Computational Stance Detection with Expanded Stance Triangle Framework", author = "Liu, Zhengyuan and Yap, Yong Keong and Chieu, Hai Leong and Chen, Nancy", 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.220", doi = "10.18653/v1/2023.acl-long.220", pages = "3987--4001", abstract = "Stance detection determines whether the author of a piece of text is in favor of, against, or neutral towards a specified target, and can be used to gain valuable insights into social media. The ubiquitous indirect referral of targets makes this task challenging, as it requires computational solutions to model semantic features and infer the corresponding implications from a literal statement. Moreover, the limited amount of available training data leads to subpar performance in out-of-domain and cross-target scenarios, as data-driven approaches are prone to rely on superficial and domain-specific features. In this work, we decompose the stance detection task from a linguistic perspective, and investigate key components and inference paths in this task. The stance triangle is a generic linguistic framework previously proposed to describe the fundamental ways people express their stance. We further expand it by characterizing the relationship between explicit and implicit objects. We then use the framework to extend one single training corpus with additional annotation. Experimental results show that strategically-enriched data can significantly improve the performance on out-of-domain and cross-target evaluation.", }

Stance detection determines whether the author of a piece of text is in favor of, against, or neutral towards a specified target, and can be used to gain valuable insights into social media. The ubiquitous indirect referral of targets makes this task challenging, as it requires computational solutions to model semantic features and infer the corresponding implications from a literal statement. Moreover, the limited amount of available training data leads to subpar performance in out-of-domain and cross-target scenarios, as data-driven approaches are prone to rely on superficial and domain-specific features. In this work, we decompose the stance detection task from a linguistic perspective, and investigate key components and inference paths in this task. The stance triangle is a generic linguistic framework previously proposed to describe the fundamental ways people express their stance. We further expand it by characterizing the relationship between explicit and implicit objects. We then use the framework to extend one single training corpus with additional annotation. Experimental results show that strategically-enriched data can significantly improve the performance on out-of-domain and cross-target evaluation.

Guiding Computational Stance Detection with Expanded Stance Triangle Framework

acl-long.220

Oral

2305.19845

https://aclanthology.org/2023.acl-long.221.bib

https://aclanthology.org/2023.acl-long.221/

@inproceedings{guo-etal-2023-analyzing, title = "Analyzing and Reducing the Performance Gap in Cross-Lingual Transfer with Fine-tuning Slow and Fast", author = "Guo, Yiduo and Liang, Yaobo and Zhao, Dongyan and Liu, Bing and Duan, Nan", 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.221", doi = "10.18653/v1/2023.acl-long.221", pages = "4002--4017", abstract = "Existing research has shown that a multilingual pre-trained language model fine-tuned with one (source) language also performs well on downstream tasks for non-source languages, even though no fine-tuning is done on these languages. However, there is a clear gap between the performance of the source language and that of the non-source languages. This paper analyzes the fine-tuning process, discovers when the performance gap changes and identifies which network weights affect the overall performance most. Additionally, the paper seeks to answer to what extent the gap can be reduced by reducing forgetting. Based on the analysis results, a method named Fine-tuning slow and fast with four training policies is proposed to address these issues. Experimental results show the proposed method outperforms baselines by a clear margin.", }

Existing research has shown that a multilingual pre-trained language model fine-tuned with one (source) language also performs well on downstream tasks for non-source languages, even though no fine-tuning is done on these languages. However, there is a clear gap between the performance of the source language and that of the non-source languages. This paper analyzes the fine-tuning process, discovers when the performance gap changes and identifies which network weights affect the overall performance most. Additionally, the paper seeks to answer to what extent the gap can be reduced by reducing forgetting. Based on the analysis results, a method named Fine-tuning slow and fast with four training policies is proposed to address these issues. Experimental results show the proposed method outperforms baselines by a clear margin.

Analyzing and Reducing the Performance Gap in Cross-Lingual Transfer with Fine-tuning Slow and Fast

acl-long.221

Poster

2305.11449

https://aclanthology.org/2023.acl-long.222.bib

https://aclanthology.org/2023.acl-long.222/

@inproceedings{zhou-etal-2023-improving, title = "Improving Self-training for Cross-lingual Named Entity Recognition with Contrastive and Prototype Learning", author = "Zhou, Ran and Li, Xin and Bing, Lidong and Cambria, Erik and Miao, Chunyan", 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.222", doi = "10.18653/v1/2023.acl-long.222", pages = "4018--4031", abstract = "In cross-lingual named entity recognition (NER), self-training is commonly used to bridge the linguistic gap by training on pseudo-labeled target-language data. However, due to sub-optimal performance on target languages, the pseudo labels are often noisy and limit the overall performance. In this work, we aim to improve self-training for cross-lingual NER by combining representation learning and pseudo label refinement in one coherent framework. Our proposed method, namely ContProto mainly comprises two components: (1) contrastive self-training and (2) prototype-based pseudo-labeling. Our contrastive self-training facilitates span classification by separating clusters of different classes, and enhances cross-lingual transferability by producing closely-aligned representations between the source and target language. Meanwhile, prototype-based pseudo-labeling effectively improves the accuracy of pseudo labels during training. We evaluate ContProto on multiple transfer pairs, and experimental results show our method brings substantial improvements over current state-of-the-art methods.", }

In cross-lingual named entity recognition (NER), self-training is commonly used to bridge the linguistic gap by training on pseudo-labeled target-language data. However, due to sub-optimal performance on target languages, the pseudo labels are often noisy and limit the overall performance. In this work, we aim to improve self-training for cross-lingual NER by combining representation learning and pseudo label refinement in one coherent framework. Our proposed method, namely ContProto mainly comprises two components: (1) contrastive self-training and (2) prototype-based pseudo-labeling. Our contrastive self-training facilitates span classification by separating clusters of different classes, and enhances cross-lingual transferability by producing closely-aligned representations between the source and target language. Meanwhile, prototype-based pseudo-labeling effectively improves the accuracy of pseudo labels during training. We evaluate ContProto on multiple transfer pairs, and experimental results show our method brings substantial improvements over current state-of-the-art methods.

Improving Self-training for Cross-lingual Named Entity Recognition with Contrastive and Prototype Learning

acl-long.222

Poster

2305.13628

https://aclanthology.org/2023.acl-long.223.bib

https://aclanthology.org/2023.acl-long.223/

@inproceedings{parcalabescu-frank-2023-mm, title = "{MM}-{SHAP}: A Performance-agnostic Metric for Measuring Multimodal Contributions in Vision and Language Models {\&} Tasks", author = "Parcalabescu, Letitia and Frank, Anette", 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.223", doi = "10.18653/v1/2023.acl-long.223", pages = "4032--4059", abstract = "Vision and language models (VL) are known to exploit unrobust indicators in individual modalities (e.g., introduced by distributional biases) instead of focusing on relevant information in each modality. That a unimodal model achieves similar accuracy on a VL task to a multimodal one, indicates that so-called unimodal collapse occurred. However, accuracy-based tests fail to detect e.g., when the model prediction is wrong, while the model used relevant information from a modality. Instead, we propose MM-SHAP, a performance-agnostic multimodality score based on Shapley values that reliably quantifies in which proportions a multimodal model uses individual modalities. We apply MM-SHAP in two ways: (1) to compare models for their average degree of multimodality, and (2) to measure for individual models the contribution of individual modalities for different tasks and datasets. Experiments with six VL models {--} LXMERT, CLIP and four ALBEF variants {--} on four VL tasks highlight that unimodal collapse can occur to different degrees and in different directions, contradicting the wide-spread assumption that unimodal collapse is one-sided. Based on our results, we recommend MM-SHAP for analysing multimodal tasks, to diagnose and guide progress towards multimodal integration. Code available at \url{https://github.com/Heidelberg-NLP/MM-SHAP}.", }

Vision and language models (VL) are known to exploit unrobust indicators in individual modalities (e.g., introduced by distributional biases) instead of focusing on relevant information in each modality. That a unimodal model achieves similar accuracy on a VL task to a multimodal one, indicates that so-called unimodal collapse occurred. However, accuracy-based tests fail to detect e.g., when the model prediction is wrong, while the model used relevant information from a modality. Instead, we propose MM-SHAP, a performance-agnostic multimodality score based on Shapley values that reliably quantifies in which proportions a multimodal model uses individual modalities. We apply MM-SHAP in two ways: (1) to compare models for their average degree of multimodality, and (2) to measure for individual models the contribution of individual modalities for different tasks and datasets. Experiments with six VL models {--} LXMERT, CLIP and four ALBEF variants {--} on four VL tasks highlight that unimodal collapse can occur to different degrees and in different directions, contradicting the wide-spread assumption that unimodal collapse is one-sided. Based on our results, we recommend MM-SHAP for analysing multimodal tasks, to diagnose and guide progress towards multimodal integration. Code available at \url{https://github.com/Heidelberg-NLP/MM-SHAP}.

MM-SHAP: A Performance-agnostic Metric for Measuring Multimodal Contributions in Vision and Language Models & Tasks

acl-long.223

Poster

2212.08158

https://aclanthology.org/2023.acl-long.224.bib

https://aclanthology.org/2023.acl-long.224/

@inproceedings{lu-etal-2023-towards, title = "Towards Boosting the Open-Domain Chatbot with Human Feedback", author = "Lu, Hua and Bao, Siqi and He, Huang and Wang, Fan and Wu, Hua and Wang, Haifeng", 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.224", doi = "10.18653/v1/2023.acl-long.224", pages = "4060--4078", abstract = "Many open-domain dialogue models pre-trained with social media comments can generate coherent replies but have difficulties producing engaging responses. This phenomenon might mainly result from the deficiency of annotated human-human conversations and the misalignment with human preference. In this paper, we propose a novel and efficient framework Diamante to boost the open-domain chatbot, where two kinds of human feedback (including explicit demonstration and implicit preference) are collected and leveraged. By asking annotators to select or amend the model-generated candidate responses, Diamante efficiently collects the human demonstrated responses and constructs a Chinese chit-chat dataset. To enhance the alignment with human preference, Diamante leverages the implicit preference in the data collection process and introduces the generation-evaluation joint training. Comprehensive experiments indicate that the Diamante dataset and joint training paradigm can significantly boost the performance of pre-trained dialogue models. The overall engagingness of the previous state-of-the-art model has been improved remarkably by 50{\%} in Chinese open-domain conversations.", }

Many open-domain dialogue models pre-trained with social media comments can generate coherent replies but have difficulties producing engaging responses. This phenomenon might mainly result from the deficiency of annotated human-human conversations and the misalignment with human preference. In this paper, we propose a novel and efficient framework Diamante to boost the open-domain chatbot, where two kinds of human feedback (including explicit demonstration and implicit preference) are collected and leveraged. By asking annotators to select or amend the model-generated candidate responses, Diamante efficiently collects the human demonstrated responses and constructs a Chinese chit-chat dataset. To enhance the alignment with human preference, Diamante leverages the implicit preference in the data collection process and introduces the generation-evaluation joint training. Comprehensive experiments indicate that the Diamante dataset and joint training paradigm can significantly boost the performance of pre-trained dialogue models. The overall engagingness of the previous state-of-the-art model has been improved remarkably by 50{\%} in Chinese open-domain conversations.

Towards Boosting the Open-Domain Chatbot with Human Feedback

acl-long.224

Oral

2208.14165

https://aclanthology.org/2023.acl-long.225.bib

https://aclanthology.org/2023.acl-long.225/

@inproceedings{deng-etal-2023-knowledge, title = "Knowledge-enhanced Mixed-initiative Dialogue System for Emotional Support Conversations", author = "Deng, Yang and Zhang, Wenxuan and Yuan, Yifei and Lam, Wai", 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.225", doi = "10.18653/v1/2023.acl-long.225", pages = "4079--4095", abstract = "Unlike empathetic dialogues, the system in emotional support conversations (ESC) is expected to not only convey empathy for comforting the help-seeker, but also proactively assist in exploring and addressing their problems during the conversation. In this work, we study the problem of mixed-initiative ESC where the user and system can both take the initiative in leading the conversation. Specifically, we conduct a novel analysis on mixed-initiative ESC systems with a tailor-designed schema that divides utterances into different types with speaker roles and initiative types. Four emotional support metrics are proposed to evaluate the mixed-initiative interactions. The analysis reveals the necessity and challenges of building mixed-initiative ESC systems. In the light of this, we propose a knowledge-enhanced mixed-initiative framework (KEMI) for ESC, which retrieves actual case knowledge from a large-scale mental health knowledge graph for generating mixed-initiative responses. Experimental results on two ESC datasets show the superiority of KEMI in both content-preserving evaluation and mixed initiative related analyses.", }

Unlike empathetic dialogues, the system in emotional support conversations (ESC) is expected to not only convey empathy for comforting the help-seeker, but also proactively assist in exploring and addressing their problems during the conversation. In this work, we study the problem of mixed-initiative ESC where the user and system can both take the initiative in leading the conversation. Specifically, we conduct a novel analysis on mixed-initiative ESC systems with a tailor-designed schema that divides utterances into different types with speaker roles and initiative types. Four emotional support metrics are proposed to evaluate the mixed-initiative interactions. The analysis reveals the necessity and challenges of building mixed-initiative ESC systems. In the light of this, we propose a knowledge-enhanced mixed-initiative framework (KEMI) for ESC, which retrieves actual case knowledge from a large-scale mental health knowledge graph for generating mixed-initiative responses. Experimental results on two ESC datasets show the superiority of KEMI in both content-preserving evaluation and mixed initiative related analyses.

Knowledge-enhanced Mixed-initiative Dialogue System for Emotional Support Conversations

acl-long.225

Poster

2305.10172

https://aclanthology.org/2023.acl-long.226.bib

https://aclanthology.org/2023.acl-long.226/

@inproceedings{yan-etal-2023-utc, title = "{UTC}-{IE}: A Unified Token-pair Classification Architecture for Information Extraction", author = "Yan, Hang and Sun, Yu and Li, Xiaonan and Zhou, Yunhua and Huang, Xuanjing and Qiu, Xipeng", 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.226", doi = "10.18653/v1/2023.acl-long.226", pages = "4096--4122", abstract = "Information Extraction (IE) spans several tasks with different output structures, such as named entity recognition, relation extraction and event extraction. Previously, those tasks were solved with different models because of diverse task output structures. Through re-examining IE tasks, we find that all of them can be interpreted as extracting spans and span relations. They can further be decomposed into token-pair classification tasks by using the start and end token of a span to pinpoint the span, and using the start-to-start and end-to-end token pairs of two spans to determine the relation. Based on the reformulation, we propose a Unified Token-pair Classification architecture for Information Extraction (UTC-IE), where we introduce Plusformer on top of the token-pair feature matrix. Specifically, it models axis-aware interaction with plus-shaped self-attention and local interaction with Convolutional Neural Network over token pairs. Experiments show that our approach outperforms task-specific and unified models on all tasks in 10 datasets, and achieves better or comparable results on 2 joint IE datasets. Moreover, UTC-IE speeds up over state-of-the-art models on IE tasks significantly in most datasets, which verifies the effectiveness of our architecture.", }

Information Extraction (IE) spans several tasks with different output structures, such as named entity recognition, relation extraction and event extraction. Previously, those tasks were solved with different models because of diverse task output structures. Through re-examining IE tasks, we find that all of them can be interpreted as extracting spans and span relations. They can further be decomposed into token-pair classification tasks by using the start and end token of a span to pinpoint the span, and using the start-to-start and end-to-end token pairs of two spans to determine the relation. Based on the reformulation, we propose a Unified Token-pair Classification architecture for Information Extraction (UTC-IE), where we introduce Plusformer on top of the token-pair feature matrix. Specifically, it models axis-aware interaction with plus-shaped self-attention and local interaction with Convolutional Neural Network over token pairs. Experiments show that our approach outperforms task-specific and unified models on all tasks in 10 datasets, and achieves better or comparable results on 2 joint IE datasets. Moreover, UTC-IE speeds up over state-of-the-art models on IE tasks significantly in most datasets, which verifies the effectiveness of our architecture.

UTC-IE: A Unified Token-pair Classification Architecture for Information Extraction

acl-long.226

Poster

https://aclanthology.org/2023.acl-long.227.bib

https://aclanthology.org/2023.acl-long.227/

@inproceedings{omrani-etal-2023-social, title = "Social-Group-Agnostic Bias Mitigation via the Stereotype Content Model", author = "Omrani, Ali and Salkhordeh Ziabari, Alireza and Yu, Charles and Golazizian, Preni and Kennedy, Brendan and Atari, Mohammad and Ji, Heng and Dehghani, Morteza", 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.227", doi = "10.18653/v1/2023.acl-long.227", pages = "4123--4139", abstract = "Existing bias mitigation methods require social-group-specific word pairs (e.g., {``}man{''} {--} {``}woman{''}) for each social attribute (e.g., gender), restricting the bias mitigation to only one specified social attribute. Further, this constraint renders such methods impractical and costly for mitigating bias in understudied and/or unmarked social groups. We propose that the Stereotype Content Model (SCM) {---} a theoretical framework developed in social psychology for understanding the content of stereotyping {---} can help debiasing efforts to become social-group-agnostic by capturing the underlying connection between bias and stereotypes. SCM proposes that the content of stereotypes map to two psychological dimensions of warmth and competence. Using only pairs of terms for these two dimensions (e.g., warmth: {``}genuine{''} {--} {``}fake{''}; competence: {``}smart{''} {--} {``}stupid{''}), we perform debiasing with established methods on both pre-trained word embeddings and large language models. We demonstrate that our social-group-agnostic, SCM-based debiasing technique performs comparably to group-specific debiasing on multiple bias benchmarks, but has theoretical and practical advantages over existing approaches.", }

Existing bias mitigation methods require social-group-specific word pairs (e.g., {``}man{''} {--} {``}woman{''}) for each social attribute (e.g., gender), restricting the bias mitigation to only one specified social attribute. Further, this constraint renders such methods impractical and costly for mitigating bias in understudied and/or unmarked social groups. We propose that the Stereotype Content Model (SCM) {---} a theoretical framework developed in social psychology for understanding the content of stereotyping {---} can help debiasing efforts to become social-group-agnostic by capturing the underlying connection between bias and stereotypes. SCM proposes that the content of stereotypes map to two psychological dimensions of warmth and competence. Using only pairs of terms for these two dimensions (e.g., warmth: {``}genuine{''} {--} {``}fake{''}; competence: {``}smart{''} {--} {``}stupid{''}), we perform debiasing with established methods on both pre-trained word embeddings and large language models. We demonstrate that our social-group-agnostic, SCM-based debiasing technique performs comparably to group-specific debiasing on multiple bias benchmarks, but has theoretical and practical advantages over existing approaches.

Social-Group-Agnostic Bias Mitigation via the Stereotype Content Model

acl-long.227

Poster

https://aclanthology.org/2023.acl-long.228.bib

https://aclanthology.org/2023.acl-long.228/

@inproceedings{liu-etal-2023-revisiting, title = "Revisiting the Gold Standard: Grounding Summarization Evaluation with Robust Human Evaluation", author = "Liu, Yixin and Fabbri, Alex and Liu, Pengfei and Zhao, Yilun and Nan, Linyong and Han, Ruilin and Han, Simeng and Joty, Shafiq and Wu, Chien-Sheng and Xiong, Caiming and Radev, Dragomir", 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.228", doi = "10.18653/v1/2023.acl-long.228", pages = "4140--4170", abstract = "Human evaluation is the foundation upon which the evaluation of both summarization systems and automatic metrics rests. However, existing human evaluation studies for summarization either exhibit a low inter-annotator agreement or have insufficient scale, and an in-depth analysis of human evaluation is lacking. Therefore, we address the shortcomings of existing summarization evaluation along the following axes: (1) We propose a modified summarization salience protocol, Atomic Content Units (ACUs), which is based on fine-grained semantic units and allows for a high inter-annotator agreement. (2) We curate the Robust Summarization Evaluation (RoSE) benchmark, a large human evaluation dataset consisting of 22,000 summary-level annotations over 28 top-performing systems on three datasets. (3) We conduct a comparative study of four human evaluation protocols, underscoring potential confounding factors in evaluation setups. (4) We evaluate 50 automatic metrics and their variants using the collected human annotations across evaluation protocols and demonstrate how our benchmark leads to more statistically stable and significant results. The metrics we benchmarked include recent methods based on large language models (LLMs), GPTScore and G-Eval. Furthermore, our findings have important implications for evaluating LLMs, as we show that LLMs adjusted by human feedback (e.g., GPT-3.5) may overfit unconstrained human evaluation, which is affected by the annotators{'} prior, input-agnostic preferences, calling for more robust, targeted evaluation methods.", }

Human evaluation is the foundation upon which the evaluation of both summarization systems and automatic metrics rests. However, existing human evaluation studies for summarization either exhibit a low inter-annotator agreement or have insufficient scale, and an in-depth analysis of human evaluation is lacking. Therefore, we address the shortcomings of existing summarization evaluation along the following axes: (1) We propose a modified summarization salience protocol, Atomic Content Units (ACUs), which is based on fine-grained semantic units and allows for a high inter-annotator agreement. (2) We curate the Robust Summarization Evaluation (RoSE) benchmark, a large human evaluation dataset consisting of 22,000 summary-level annotations over 28 top-performing systems on three datasets. (3) We conduct a comparative study of four human evaluation protocols, underscoring potential confounding factors in evaluation setups. (4) We evaluate 50 automatic metrics and their variants using the collected human annotations across evaluation protocols and demonstrate how our benchmark leads to more statistically stable and significant results. The metrics we benchmarked include recent methods based on large language models (LLMs), GPTScore and G-Eval. Furthermore, our findings have important implications for evaluating LLMs, as we show that LLMs adjusted by human feedback (e.g., GPT-3.5) may overfit unconstrained human evaluation, which is affected by the annotators{'} prior, input-agnostic preferences, calling for more robust, targeted evaluation methods.

Revisiting the Gold Standard: Grounding Summarization Evaluation with Robust Human Evaluation

acl-long.228

Oral

2212.07981

https://aclanthology.org/2023.acl-long.229.bib

https://aclanthology.org/2023.acl-long.229/

@inproceedings{zhu-etal-2023-fireball, title = "{FIREBALL}: A Dataset of Dungeons and Dragons Actual-Play with Structured Game State Information", author = "Zhu, Andrew and Aggarwal, Karmanya and Feng, Alexander and Martin, Lara J. and Callison-Burch, Chris", 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.229", doi = "10.18653/v1/2023.acl-long.229", pages = "4171--4193", abstract = "Dungeons {\&} Dragons (D{\&}D) is a tabletop roleplaying game with complex natural language interactions between players and hidden state information. Recent work has shown that large language models (LLMs) that have access to state information can generate higher quality game turns than LLMs that use dialog history alone. However, previous work used game state information that was heuristically created and was not a true gold standard game state. We present FIREBALL, a large dataset containing nearly 25,000 unique sessions from real D{\&}D gameplay on Discord with true game state info. We recorded game play sessions of players who used the Avrae bot, which was developed to aid people in playing D{\&}D online, capturing language, game commands and underlying game state information. We demonstrate that FIREBALL can improve natural language generation (NLG) by using Avrae state information, improving both automated metrics and human judgments of quality. Additionally, we show that LLMs can generate executable Avrae commands, particularly after finetuning.", }

Dungeons {\&} Dragons (D{\&}D) is a tabletop roleplaying game with complex natural language interactions between players and hidden state information. Recent work has shown that large language models (LLMs) that have access to state information can generate higher quality game turns than LLMs that use dialog history alone. However, previous work used game state information that was heuristically created and was not a true gold standard game state. We present FIREBALL, a large dataset containing nearly 25,000 unique sessions from real D{\&}D gameplay on Discord with true game state info. We recorded game play sessions of players who used the Avrae bot, which was developed to aid people in playing D{\&}D online, capturing language, game commands and underlying game state information. We demonstrate that FIREBALL can improve natural language generation (NLG) by using Avrae state information, improving both automated metrics and human judgments of quality. Additionally, we show that LLMs can generate executable Avrae commands, particularly after finetuning.

FIREBALL: A Dataset of Dungeons and Dragons Actual-Play with Structured Game State Information

acl-long.229

Poster

2305.01528

https://huggingface.co/papers/2305.01528

https://aclanthology.org/2023.acl-long.230.bib

https://aclanthology.org/2023.acl-long.230/

@inproceedings{hu-etal-2023-fine, title = "A fine-grained comparison of pragmatic language understanding in humans and language models", author = "Hu, Jennifer and Floyd, Sammy and Jouravlev, Olessia and Fedorenko, Evelina and Gibson, Edward", 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.230", doi = "10.18653/v1/2023.acl-long.230", pages = "4194--4213", abstract = "Pragmatics and non-literal language understanding are essential to human communication, and present a long-standing challenge for artificial language models. We perform a fine-grained comparison of language models and humans on seven pragmatic phenomena, using zero-shot prompting on an expert-curated set of English materials. We ask whether models (1) select pragmatic interpretations of speaker utterances, (2) make similar error patterns as humans, and (3) use similar linguistic cues as humans to solve the tasks. We find that the largest models achieve high accuracy and match human error patterns: within incorrect responses, models favor literal interpretations over heuristic-based distractors. We also find preliminary evidence that models and humans are sensitive to similar linguistic cues. Our results suggest that pragmatic behaviors can emerge in models without explicitly constructed representations of mental states. However, models tend to struggle with phenomena relying on social expectation violations.", }

Pragmatics and non-literal language understanding are essential to human communication, and present a long-standing challenge for artificial language models. We perform a fine-grained comparison of language models and humans on seven pragmatic phenomena, using zero-shot prompting on an expert-curated set of English materials. We ask whether models (1) select pragmatic interpretations of speaker utterances, (2) make similar error patterns as humans, and (3) use similar linguistic cues as humans to solve the tasks. We find that the largest models achieve high accuracy and match human error patterns: within incorrect responses, models favor literal interpretations over heuristic-based distractors. We also find preliminary evidence that models and humans are sensitive to similar linguistic cues. Our results suggest that pragmatic behaviors can emerge in models without explicitly constructed representations of mental states. However, models tend to struggle with phenomena relying on social expectation violations.

A fine-grained comparison of pragmatic language understanding in humans and language models

acl-long.230

Poster

2212.06801

https://aclanthology.org/2023.acl-long.231.bib

https://aclanthology.org/2023.acl-long.231/

@inproceedings{guo-etal-2023-counterfactual, title = "Counterfactual Multihop {QA}: A Cause-Effect Approach for Reducing Disconnected Reasoning", author = "Guo, Wangzhen and Gong, Qinkang and Rao, Yanghui and Lai, Hanjiang", 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.231", doi = "10.18653/v1/2023.acl-long.231", pages = "4214--4226", abstract = "Multi-hop QA requires reasoning over multiple supporting facts to answer the question. However, the existing QA models always rely on shortcuts, e.g., providing the true answer by only one fact, rather than multi-hop reasoning, which is referred as disconnected reasoning problem. To alleviate this issue, we propose a novel counterfactual multihop QA, a causal-effect approach that enables to reduce the disconnected reasoning. It builds upon explicitly modeling of causality: 1) the direct causal effects of disconnected reasoning and 2) the causal effect of true multi-hop reasoning from the total causal effect. With the causal graph, a counterfactual inference is proposed to disentangle the disconnected reasoning from the total causal effect, which provides us a new perspective and technology to learn a QA model that exploits the true multi-hop reasoning instead of shortcuts. Extensive experiments have been conducted on the benchmark HotpotQA dataset, which demonstrate that the proposed method can achieve notable improvement on reducing disconnected reasoning. For example, our method achieves 5.8{\%} higher points of its Supps score on HotpotQA through true multihop reasoning. The code is available at \url{https://github.com/guowzh/CFMQA}.", }

Multi-hop QA requires reasoning over multiple supporting facts to answer the question. However, the existing QA models always rely on shortcuts, e.g., providing the true answer by only one fact, rather than multi-hop reasoning, which is referred as disconnected reasoning problem. To alleviate this issue, we propose a novel counterfactual multihop QA, a causal-effect approach that enables to reduce the disconnected reasoning. It builds upon explicitly modeling of causality: 1) the direct causal effects of disconnected reasoning and 2) the causal effect of true multi-hop reasoning from the total causal effect. With the causal graph, a counterfactual inference is proposed to disentangle the disconnected reasoning from the total causal effect, which provides us a new perspective and technology to learn a QA model that exploits the true multi-hop reasoning instead of shortcuts. Extensive experiments have been conducted on the benchmark HotpotQA dataset, which demonstrate that the proposed method can achieve notable improvement on reducing disconnected reasoning. For example, our method achieves 5.8{\%} higher points of its Supps score on HotpotQA through true multihop reasoning. The code is available at \url{https://github.com/guowzh/CFMQA}.

Counterfactual Multihop QA: A Cause-Effect Approach for Reducing Disconnected Reasoning

acl-long.231

Poster

2210.07138

https://aclanthology.org/2023.acl-long.232.bib

https://aclanthology.org/2023.acl-long.232/

@inproceedings{zhou-etal-2023-causal, title = "Causal-Debias: Unifying Debiasing in Pretrained Language Models and Fine-tuning via Causal Invariant Learning", author = "Zhou, Fan and Mao, Yuzhou and Yu, Liu and Yang, Yi and Zhong, Ting", 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.232", doi = "10.18653/v1/2023.acl-long.232", pages = "4227--4241", abstract = "Demographic biases and social stereotypes are common in pretrained language models (PLMs), and a burgeoning body of literature focuses on removing the unwanted stereotypical associations from PLMs. However, when fine-tuning these bias-mitigated PLMs in downstream natural language processing (NLP) applications, such as sentiment classification, the unwanted stereotypical associations resurface or even get amplified. Since pretrain{\&}fine-tune is a major paradigm in NLP applications, separating the debiasing procedure of PLMs from fine-tuning would eventually harm the actual downstream utility. In this paper, we propose a unified debiasing framework Causal-Debias to remove unwanted stereotypical associations in PLMs during fine-tuning. Specifically, CausalDebias mitigates bias from a causal invariant perspective by leveraging the specific downstream task to identify bias-relevant and labelrelevant factors. We propose that bias-relevant factors are non-causal as they should have little impact on downstream tasks, while labelrelevant factors are causal. We perform interventions on non-causal factors in different demographic groups and design an invariant risk minimization loss to mitigate bias while maintaining task performance. Experimental results on three downstream tasks show that our proposed method can remarkably reduce unwanted stereotypical associations after PLMs are finetuned, while simultaneously minimizing the impact on PLMs and downstream applications.", }

Demographic biases and social stereotypes are common in pretrained language models (PLMs), and a burgeoning body of literature focuses on removing the unwanted stereotypical associations from PLMs. However, when fine-tuning these bias-mitigated PLMs in downstream natural language processing (NLP) applications, such as sentiment classification, the unwanted stereotypical associations resurface or even get amplified. Since pretrain{\&}fine-tune is a major paradigm in NLP applications, separating the debiasing procedure of PLMs from fine-tuning would eventually harm the actual downstream utility. In this paper, we propose a unified debiasing framework Causal-Debias to remove unwanted stereotypical associations in PLMs during fine-tuning. Specifically, CausalDebias mitigates bias from a causal invariant perspective by leveraging the specific downstream task to identify bias-relevant and labelrelevant factors. We propose that bias-relevant factors are non-causal as they should have little impact on downstream tasks, while labelrelevant factors are causal. We perform interventions on non-causal factors in different demographic groups and design an invariant risk minimization loss to mitigate bias while maintaining task performance. Experimental results on three downstream tasks show that our proposed method can remarkably reduce unwanted stereotypical associations after PLMs are finetuned, while simultaneously minimizing the impact on PLMs and downstream applications.

Causal-Debias: Unifying Debiasing in Pretrained Language Models and Fine-tuning via Causal Invariant Learning

acl-long.232

Poster

https://aclanthology.org/2023.acl-long.233.bib

https://aclanthology.org/2023.acl-long.233/

@inproceedings{liao-etal-2023-parameter, title = "Parameter-Efficient Fine-Tuning without Introducing New Latency", author = "Liao, Baohao and Meng, Yan and Monz, Christof", 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.233", doi = "10.18653/v1/2023.acl-long.233", pages = "4242--4260", abstract = "Parameter-efficient fine-tuning (PEFT) of pre-trained language models has recently demonstrated remarkable achievements, effectively matching the performance of full fine-tuning while utilizing significantly fewer trainable parameters, and consequently addressing the storage and communication constraints. Nonetheless, various PEFT methods are limited by their inherent characteristics. In the case of sparse fine-tuning, which involves modifying only a small subset of the existing parameters, the selection of fine-tuned parameters is task- and domain-specific, making it unsuitable for federated learning. On the other hand, PEFT methods with adding new parameters typically introduce additional inference latency. In this paper, we demonstrate the feasibility of generating a sparse mask in a task-agnostic manner, wherein all downstream tasks share a common mask. Our approach, which relies solely on the magnitude information of pre-trained parameters, surpasses existing methodologies by a significant margin when evaluated on the GLUE benchmark. Additionally, we introduce a novel adapter technique that directly applies the adapter to pre-trained parameters instead of the hidden representation, thereby achieving identical inference speed to that of full fine-tuning. Through extensive experiments, our proposed method attains a new state-of-the-art outcome in terms of both performance and storage efficiency, storing only 0.03{\%} parameters of full fine-tuning.", }

Parameter-efficient fine-tuning (PEFT) of pre-trained language models has recently demonstrated remarkable achievements, effectively matching the performance of full fine-tuning while utilizing significantly fewer trainable parameters, and consequently addressing the storage and communication constraints. Nonetheless, various PEFT methods are limited by their inherent characteristics. In the case of sparse fine-tuning, which involves modifying only a small subset of the existing parameters, the selection of fine-tuned parameters is task- and domain-specific, making it unsuitable for federated learning. On the other hand, PEFT methods with adding new parameters typically introduce additional inference latency. In this paper, we demonstrate the feasibility of generating a sparse mask in a task-agnostic manner, wherein all downstream tasks share a common mask. Our approach, which relies solely on the magnitude information of pre-trained parameters, surpasses existing methodologies by a significant margin when evaluated on the GLUE benchmark. Additionally, we introduce a novel adapter technique that directly applies the adapter to pre-trained parameters instead of the hidden representation, thereby achieving identical inference speed to that of full fine-tuning. Through extensive experiments, our proposed method attains a new state-of-the-art outcome in terms of both performance and storage efficiency, storing only 0.03{\%} parameters of full fine-tuning.

Parameter-Efficient Fine-Tuning without Introducing New Latency

acl-long.233

Poster

2305.16742

https://aclanthology.org/2023.acl-long.234.bib

https://aclanthology.org/2023.acl-long.234/

@inproceedings{fang-etal-2023-manner, title = "{MANNER}: A Variational Memory-Augmented Model for Cross Domain Few-Shot Named Entity Recognition", author = "Fang, Jinyuan and Wang, Xiaobin and Meng, Zaiqiao and Xie, Pengjun and Huang, Fei and Jiang, Yong", 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.234", doi = "10.18653/v1/2023.acl-long.234", pages = "4261--4276", abstract = "This paper focuses on the task of cross domain few-shot named entity recognition (NER), which aims to adapt the knowledge learned from source domain to recognize named entities in target domain with only a few labeled examples. To address this challenging task, we propose MANNER, a variational memory-augmented few-shot NER model. Specifically, MANNER uses a memory module to store information from the source domain and then retrieve relevant information from the memory to augment few-shot task in the target domain. In order to effectively utilize the information from memory, MANNER uses optimal transport to retrieve and process information from memory, which can explicitly adapt the retrieved information from source domain to target domain and improve the performance in the cross domain few-shot setting. We conduct experiments on English and Chinese cross domain few-shot NER datasets, and the experimental results demonstrate that MANNER can achieve superior performance.", }

This paper focuses on the task of cross domain few-shot named entity recognition (NER), which aims to adapt the knowledge learned from source domain to recognize named entities in target domain with only a few labeled examples. To address this challenging task, we propose MANNER, a variational memory-augmented few-shot NER model. Specifically, MANNER uses a memory module to store information from the source domain and then retrieve relevant information from the memory to augment few-shot task in the target domain. In order to effectively utilize the information from memory, MANNER uses optimal transport to retrieve and process information from memory, which can explicitly adapt the retrieved information from source domain to target domain and improve the performance in the cross domain few-shot setting. We conduct experiments on English and Chinese cross domain few-shot NER datasets, and the experimental results demonstrate that MANNER can achieve superior performance.

MANNER: A Variational Memory-Augmented Model for Cross Domain Few-Shot Named Entity Recognition

acl-long.234

Poster

https://aclanthology.org/2023.acl-long.235.bib

https://aclanthology.org/2023.acl-long.235/

@inproceedings{fitzgerald-etal-2023-massive, title = "{MASSIVE}: A 1{M}-Example Multilingual Natural Language Understanding Dataset with 51 Typologically-Diverse Languages", author = "FitzGerald, Jack and Hench, Christopher and Peris, Charith and Mackie, Scott and Rottmann, Kay and Sanchez, Ana and Nash, Aaron and Urbach, Liam and Kakarala, Vishesh and Singh, Richa and Ranganath, Swetha and Crist, Laurie and Britan, Misha and Leeuwis, Wouter and Tur, Gokhan and Natarajan, Prem", 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.235", doi = "10.18653/v1/2023.acl-long.235", pages = "4277--4302", abstract = "We present the MASSIVE dataset{--}Multilingual Amazon Slu resource package (SLURP) for Slot-filling, Intent classification, and Virtual assistant Evaluation. MASSIVE contains 1M realistic, parallel, labeled virtual assistant utterances spanning 51 languages, 18 domains, 60 intents, and 55 slots. MASSIVE was created by tasking professional translators to localize the English-only SLURP dataset into 50 typologically diverse languages from 29 genera. We also present modeling results on XLM-R and mT5, including exact match accuracy, intent classification accuracy, and slot-filling F1 score. We have released our dataset, modeling code, and models publicly.", }

We present the MASSIVE dataset{--}Multilingual Amazon Slu resource package (SLURP) for Slot-filling, Intent classification, and Virtual assistant Evaluation. MASSIVE contains 1M realistic, parallel, labeled virtual assistant utterances spanning 51 languages, 18 domains, 60 intents, and 55 slots. MASSIVE was created by tasking professional translators to localize the English-only SLURP dataset into 50 typologically diverse languages from 29 genera. We also present modeling results on XLM-R and mT5, including exact match accuracy, intent classification accuracy, and slot-filling F1 score. We have released our dataset, modeling code, and models publicly.

MASSIVE: A 1M-Example Multilingual Natural Language Understanding Dataset with 51 Typologically-Diverse Languages

acl-long.235

Oral

2204.08582

https://aclanthology.org/2023.acl-long.236.bib

https://aclanthology.org/2023.acl-long.236/

@inproceedings{yuan-etal-2023-distilling, title = "Distilling Script Knowledge from Large Language Models for Constrained Language Planning", author = "Yuan, Siyu and Chen, Jiangjie and Fu, Ziquan and Ge, Xuyang and Shah, Soham and Jankowski, Charles and Xiao, Yanghua and Yang, Deqing", 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.236", doi = "10.18653/v1/2023.acl-long.236", pages = "4303--4325", abstract = "In everyday life, humans often plan their actions by following step-by-step instructions in the form of goal-oriented scripts. Previous work has exploited language models (LMs) to plan for abstract goals of stereotypical activities (e.g., {``}make a cake{''}), but leaves more specific goals with multi-facet constraints understudied (e.g., {``}make a cake for diabetics{''}). In this paper, we define the task of constrained language planning for the first time. We propose an over-generate-then-filter approach to improve large language models (LLMs) on this task, and use it to distill a novel constrained language planning dataset, Coscript, which consists of 55,000 scripts. Empirical results demonstrate that our method significantly improves the constrained language planning ability of LLMs, especially on constraint faithfulness. Furthermore, Coscript is demonstrated to be quite effective in endowing smaller LMs with constrained language planning ability.", }

In everyday life, humans often plan their actions by following step-by-step instructions in the form of goal-oriented scripts. Previous work has exploited language models (LMs) to plan for abstract goals of stereotypical activities (e.g., {``}make a cake{''}), but leaves more specific goals with multi-facet constraints understudied (e.g., {``}make a cake for diabetics{''}). In this paper, we define the task of constrained language planning for the first time. We propose an over-generate-then-filter approach to improve large language models (LLMs) on this task, and use it to distill a novel constrained language planning dataset, Coscript, which consists of 55,000 scripts. Empirical results demonstrate that our method significantly improves the constrained language planning ability of LLMs, especially on constraint faithfulness. Furthermore, Coscript is demonstrated to be quite effective in endowing smaller LMs with constrained language planning ability.

Distilling Script Knowledge from Large Language Models for Constrained Language Planning

acl-long.236

Poster

2305.05252

https://huggingface.co/papers/2305.05252

https://aclanthology.org/2023.acl-long.237.bib

https://aclanthology.org/2023.acl-long.237/

@inproceedings{huguet-cabot-etal-2023-red, title = "{RED}$^{\textrm{FM}}$: a Filtered and Multilingual Relation Extraction Dataset", author = "Huguet Cabot, ‪Pere-Llu{\'\i}s and Tedeschi, Simone and Ngonga Ngomo, Axel-Cyrille and Navigli, Roberto", 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.237", doi = "10.18653/v1/2023.acl-long.237", pages = "4326--4343", abstract = "Relation Extraction (RE) is a task that identifies relationships between entities in a text, enabling the acquisition of relational facts and bridging the gap between natural language and structured knowledge. However, current RE models often rely on small datasets with low coverage of relation types, particularly when working with languages other than English.In this paper, we address the above issue and provide two new resources that enable the training and evaluation of multilingual RE systems. First, we present SRED$^{\textrm{FM}}$, an automatically annotated dataset covering 18 languages, 400 relation types, 13 entity types, totaling more than 40 million triplet instances. Second, we propose RED$^{\textrm{FM}}$, a smaller, human-revised dataset for seven languages that allows for the evaluation of multilingual RE systems. To demonstrate the utility of these novel datasets, we experiment with the first end-to-end multilingual RE model, mREBEL, that extracts triplets, including entity types, in multiple languages. We release our resources and model checkpoints at [\url{https://www.github.com/babelscape/rebel}](\url{https://www.github.com/babelscape/rebel}).", }

Relation Extraction (RE) is a task that identifies relationships between entities in a text, enabling the acquisition of relational facts and bridging the gap between natural language and structured knowledge. However, current RE models often rely on small datasets with low coverage of relation types, particularly when working with languages other than English.In this paper, we address the above issue and provide two new resources that enable the training and evaluation of multilingual RE systems. First, we present SRED$^{\textrm{FM}}$, an automatically annotated dataset covering 18 languages, 400 relation types, 13 entity types, totaling more than 40 million triplet instances. Second, we propose RED$^{\textrm{FM}}$, a smaller, human-revised dataset for seven languages that allows for the evaluation of multilingual RE systems. To demonstrate the utility of these novel datasets, we experiment with the first end-to-end multilingual RE model, mREBEL, that extracts triplets, including entity types, in multiple languages. We release our resources and model checkpoints at [\url{https://www.github.com/babelscape/rebel}](\url{https://www.github.com/babelscape/rebel}).

RED^FM: a Filtered and Multilingual Relation Extraction Dataset

acl-long.237

Poster

https://aclanthology.org/2023.acl-long.238.bib

https://aclanthology.org/2023.acl-long.238/

@inproceedings{ziegenbein-etal-2023-modeling, title = "Modeling Appropriate Language in Argumentation", author = "Ziegenbein, Timon and Syed, Shahbaz and Lange, Felix and Potthast, Martin and Wachsmuth, Henning", 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.238", doi = "10.18653/v1/2023.acl-long.238", pages = "4344--4363", abstract = "Online discussion moderators must make ad-hoc decisions about whether the contributions of discussion participants are appropriate or should be removed to maintain civility. Existing research on offensive language and the resulting tools cover only one aspect among many involved in such decisions. The question of what is considered appropriate in a controversial discussion has not yet been systematically addressed. In this paper, we operationalize appropriate language in argumentation for the first time. In particular, we model appropriateness through the absence of flaws, grounded in research on argument quality assessment, especially in aspects from rhetoric. From these, we derive a new taxonomy of 14 dimensions that determine inappropriate language in online discussions. Building on three argument quality corpora, we then create a corpus of 2191 arguments annotated for the 14 dimensions. Empirical analyses support that the taxonomy covers the concept of appropriateness comprehensively, showing several plausible correlations with argument quality dimensions. Moreover, results of baseline approaches to assessing appropriateness suggest that all dimensions can be modeled computationally on the corpus.", }

Online discussion moderators must make ad-hoc decisions about whether the contributions of discussion participants are appropriate or should be removed to maintain civility. Existing research on offensive language and the resulting tools cover only one aspect among many involved in such decisions. The question of what is considered appropriate in a controversial discussion has not yet been systematically addressed. In this paper, we operationalize appropriate language in argumentation for the first time. In particular, we model appropriateness through the absence of flaws, grounded in research on argument quality assessment, especially in aspects from rhetoric. From these, we derive a new taxonomy of 14 dimensions that determine inappropriate language in online discussions. Building on three argument quality corpora, we then create a corpus of 2191 arguments annotated for the 14 dimensions. Empirical analyses support that the taxonomy covers the concept of appropriateness comprehensively, showing several plausible correlations with argument quality dimensions. Moreover, results of baseline approaches to assessing appropriateness suggest that all dimensions can be modeled computationally on the corpus.

Modeling Appropriate Language in Argumentation

acl-long.238

Poster

2305.14935

https://aclanthology.org/2023.acl-long.239.bib

https://aclanthology.org/2023.acl-long.239/

@inproceedings{cho-etal-2023-celda, title = "{CELDA}: Leveraging Black-box Language Model as Enhanced Classifier without Labels", author = "Cho, Hyunsoo and Kim, Youna and Lee, Sang-goo", 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.239", doi = "10.18653/v1/2023.acl-long.239", pages = "4364--4379", abstract = "Utilizing language models (LMs) without internal access is becoming an attractive paradigm in the field of NLP as many cutting-edge LMs are released through APIs and boast a massive scale. The de-facto method in this type of black-box scenario is known as prompting, which has shown progressive performance enhancements in situations where data labels are scarce or unavailable. Despite their efficacy, they still fall short in comparison to fully supervised counterparts and are generally brittle to slight modifications. In this paper, we propose Clustering-enhanced Linear Discriminative Analysis (CELDA), a novel approach that improves the text classification accuracy with a very weak-supervision signal (i.e., name of the labels).Our framework draws a precise decision boundary without accessing weights or gradients of the LM model or data labels. The core ideas of CELDA are twofold:(1) extracting a refined pseudo-labeled dataset from an unlabeled dataset, and (2) training a lightweight and robust model on the top of LM, which learns an accurate decision boundary from an extracted noisy dataset. Throughout in-depth investigations on various datasets, we demonstrated that CELDA reaches new state-of-the-art in weakly-supervised text classification and narrows the gap with a fully-supervised model. Additionally, our proposed methodology can be applied universally to any LM and has the potential to scale to larger models, making it a more viable option for utilizing large LMs.", }

Utilizing language models (LMs) without internal access is becoming an attractive paradigm in the field of NLP as many cutting-edge LMs are released through APIs and boast a massive scale. The de-facto method in this type of black-box scenario is known as prompting, which has shown progressive performance enhancements in situations where data labels are scarce or unavailable. Despite their efficacy, they still fall short in comparison to fully supervised counterparts and are generally brittle to slight modifications. In this paper, we propose Clustering-enhanced Linear Discriminative Analysis (CELDA), a novel approach that improves the text classification accuracy with a very weak-supervision signal (i.e., name of the labels).Our framework draws a precise decision boundary without accessing weights or gradients of the LM model or data labels. The core ideas of CELDA are twofold:(1) extracting a refined pseudo-labeled dataset from an unlabeled dataset, and (2) training a lightweight and robust model on the top of LM, which learns an accurate decision boundary from an extracted noisy dataset. Throughout in-depth investigations on various datasets, we demonstrated that CELDA reaches new state-of-the-art in weakly-supervised text classification and narrows the gap with a fully-supervised model. Additionally, our proposed methodology can be applied universally to any LM and has the potential to scale to larger models, making it a more viable option for utilizing large LMs.

CELDA: Leveraging Black-box Language Model as Enhanced Classifier without Labels

acl-long.239

Poster

2306.02693

https://aclanthology.org/2023.acl-long.240.bib

https://aclanthology.org/2023.acl-long.240/

@inproceedings{gou-etal-2023-mvp, title = "{M}v{P}: Multi-view Prompting Improves Aspect Sentiment Tuple Prediction", author = "Gou, Zhibin and Guo, Qingyan and Yang, Yujiu", 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.240", doi = "10.18653/v1/2023.acl-long.240", pages = "4380--4397", abstract = "Generative methods greatly promote aspect-based sentiment analysis via generating a sequence of sentiment elements in a specified format. However, existing studies usually predict sentiment elements in a fixed order, which ignores the effect of the interdependence of the elements in a sentiment tuple and the diversity of language expression on the results. In this work, we propose Multi-view Prompting (MVP) that aggregates sentiment elements generated in different orders, leveraging the intuition of human-like problem-solving processes from different views. Specifically, MVP introduces element order prompts to guide the language model to generate multiple sentiment tuples, each with a different element order, and then selects the most reasonable tuples by voting. MVP can naturally model multi-view and multi-task as permutations and combinations of elements, respectively, outperforming previous task-specific designed methods on multiple ABSA tasks with a single model. Extensive experiments show that MVP significantly advances the state-of-the-art performance on 10 datasets of 4 benchmark tasks, and performs quite effectively in low-resource settings. Detailed evaluation verified the effectiveness, flexibility, and cross-task transferability of MVP.", }

Generative methods greatly promote aspect-based sentiment analysis via generating a sequence of sentiment elements in a specified format. However, existing studies usually predict sentiment elements in a fixed order, which ignores the effect of the interdependence of the elements in a sentiment tuple and the diversity of language expression on the results. In this work, we propose Multi-view Prompting (MVP) that aggregates sentiment elements generated in different orders, leveraging the intuition of human-like problem-solving processes from different views. Specifically, MVP introduces element order prompts to guide the language model to generate multiple sentiment tuples, each with a different element order, and then selects the most reasonable tuples by voting. MVP can naturally model multi-view and multi-task as permutations and combinations of elements, respectively, outperforming previous task-specific designed methods on multiple ABSA tasks with a single model. Extensive experiments show that MVP significantly advances the state-of-the-art performance on 10 datasets of 4 benchmark tasks, and performs quite effectively in low-resource settings. Detailed evaluation verified the effectiveness, flexibility, and cross-task transferability of MVP.

MvP: Multi-view Prompting Improves Aspect Sentiment Tuple Prediction

acl-long.240

Poster

2305.12627

https://aclanthology.org/2023.acl-long.241.bib

https://aclanthology.org/2023.acl-long.241/

@inproceedings{ghazarian-etal-2023-accent, title = "{ACCENT}: An Automatic Event Commonsense Evaluation Metric for Open-Domain Dialogue Systems", author = "Ghazarian, Sarik and Shao, Yijia and Han, Rujun and Galstyan, Aram 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.241", doi = "10.18653/v1/2023.acl-long.241", pages = "4398--4419", abstract = "Commonsense reasoning is omnipresent in human communications and thus is an important feature for open-domain dialogue systems. However, evaluating commonsense in dialogue systems is still an open challenge. We take the first step by focusing on \textit{event commonsense} that considers events and their relations, and is crucial in both dialogues and general commonsense reasoning. We propose \textbf{ACCENT}, an event commonsense evaluation metric empowered by commonsense knowledge bases (CSKBs). ACCENT first extracts event-relation tuples from a dialogue, and then evaluates the response by scoring the tuples in terms of their compatibility with the CSKB. To evaluate ACCENT, we construct the first public event commonsense evaluation dataset for open-domain dialogues.Our experiments show that ACCENT is an efficient metric for event commonsense evaluation, which achieves higher correlations with human judgments than existing baselines.", }

Commonsense reasoning is omnipresent in human communications and thus is an important feature for open-domain dialogue systems. However, evaluating commonsense in dialogue systems is still an open challenge. We take the first step by focusing on \textit{event commonsense} that considers events and their relations, and is crucial in both dialogues and general commonsense reasoning. We propose \textbf{ACCENT}, an event commonsense evaluation metric empowered by commonsense knowledge bases (CSKBs). ACCENT first extracts event-relation tuples from a dialogue, and then evaluates the response by scoring the tuples in terms of their compatibility with the CSKB. To evaluate ACCENT, we construct the first public event commonsense evaluation dataset for open-domain dialogues.Our experiments show that ACCENT is an efficient metric for event commonsense evaluation, which achieves higher correlations with human judgments than existing baselines.

ACCENT: An Automatic Event Commonsense Evaluation Metric for Open-Domain Dialogue Systems

acl-long.241

Poster

2305.07797

https://huggingface.co/papers/2305.07797

https://aclanthology.org/2023.acl-long.242.bib

https://aclanthology.org/2023.acl-long.242/

@inproceedings{ludan-etal-2023-explanation, title = "Explanation-based Finetuning Makes Models More Robust to Spurious Cues", author = "Ludan, Josh Magnus and Meng, Yixuan and Nguyen, Tai and Shah, Saurabh and Lyu, Qing and Apidianaki, Marianna and Callison-Burch, Chris", 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.242", doi = "10.18653/v1/2023.acl-long.242", pages = "4420--4441", abstract = "Large Language Models (LLMs) are so powerful that they sometimes learn correlations between labels and features that are irrelevant to the task, leading to poor generalization on out-of-distribution data. We propose explanation-based finetuning as a general approach to mitigate LLMs{'} reliance on spurious correlations. Unlike standard finetuning where the model only predicts the answer given the input, we finetune the model to additionally generate a free-text explanation supporting its answer. To evaluate our method, we finetune the model on artificially constructed training sets containing different types of spurious cues, and test it on a test set without these cues. Compared to standard finetuning, our method makes GPT-3 (davinci) remarkably more robust against spurious cues in terms of accuracy drop across four classification tasks: ComVE (+1.2), CREAK (+9.1), e-SNLI (+15.4), and SBIC (+6.5). The efficacy generalizes across multiple model families and scales, with greater gains for larger models. Finally, our method also works well with explanations generated by the model, implying its applicability to more datasets without human-written explanations.", }

Large Language Models (LLMs) are so powerful that they sometimes learn correlations between labels and features that are irrelevant to the task, leading to poor generalization on out-of-distribution data. We propose explanation-based finetuning as a general approach to mitigate LLMs{'} reliance on spurious correlations. Unlike standard finetuning where the model only predicts the answer given the input, we finetune the model to additionally generate a free-text explanation supporting its answer. To evaluate our method, we finetune the model on artificially constructed training sets containing different types of spurious cues, and test it on a test set without these cues. Compared to standard finetuning, our method makes GPT-3 (davinci) remarkably more robust against spurious cues in terms of accuracy drop across four classification tasks: ComVE (+1.2), CREAK (+9.1), e-SNLI (+15.4), and SBIC (+6.5). The efficacy generalizes across multiple model families and scales, with greater gains for larger models. Finally, our method also works well with explanations generated by the model, implying its applicability to more datasets without human-written explanations.

Explanation-based Finetuning Makes Models More Robust to Spurious Cues

acl-long.242

Poster

2305.04990

https://aclanthology.org/2023.acl-long.243.bib

https://aclanthology.org/2023.acl-long.243/

@inproceedings{luo-etal-2023-came, title = "{CAME}: Confidence-guided Adaptive Memory Efficient Optimization", author = "Luo, Yang and Ren, Xiaozhe and Zheng, Zangwei and Jiang, Zhuo and Jiang, Xin 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.243", doi = "10.18653/v1/2023.acl-long.243", pages = "4442--4453", abstract = "Adaptive gradient methods, such as Adam and LAMB, have demonstrated excellent performance in the training of large language models. Nevertheless, the need for adaptivity requires maintaining second-moment estimates of the per-parameter gradients, which entails a high cost of extra memory overheads. To solve this problem, several memory-efficient optimizers (e.g., Adafactor) have been proposed to obtain a drastic reduction in auxiliary memory usage, but with a performance penalty. In this paper, we first study a confidence-guided strategy to reduce the instability of existing memory efficient optimizers. Based on this strategy, we propose CAME to simultaneously achieve two goals: fast convergence as in traditional adaptive methods, and low memory usage as in memory-efficient methods. Extensive experiments demonstrate the training stability and superior performance of CAME across various NLP tasks such as BERT and GPT-2 training. Notably, for BERT pre-training on the large batch size of 32,768, our proposed optimizer attains faster convergence and higher accuracy compared with the Adam optimizer. The implementation of CAME is publicly available.", }

Adaptive gradient methods, such as Adam and LAMB, have demonstrated excellent performance in the training of large language models. Nevertheless, the need for adaptivity requires maintaining second-moment estimates of the per-parameter gradients, which entails a high cost of extra memory overheads. To solve this problem, several memory-efficient optimizers (e.g., Adafactor) have been proposed to obtain a drastic reduction in auxiliary memory usage, but with a performance penalty. In this paper, we first study a confidence-guided strategy to reduce the instability of existing memory efficient optimizers. Based on this strategy, we propose CAME to simultaneously achieve two goals: fast convergence as in traditional adaptive methods, and low memory usage as in memory-efficient methods. Extensive experiments demonstrate the training stability and superior performance of CAME across various NLP tasks such as BERT and GPT-2 training. Notably, for BERT pre-training on the large batch size of 32,768, our proposed optimizer attains faster convergence and higher accuracy compared with the Adam optimizer. The implementation of CAME is publicly available.

CAME: Confidence-guided Adaptive Memory Efficient Optimization

acl-long.243

Poster

2307.02047

https://aclanthology.org/2023.acl-long.244.bib

https://aclanthology.org/2023.acl-long.244/

@inproceedings{shaikh-etal-2023-second, title = "On Second Thought, Let{'}s Not Think Step by Step! Bias and Toxicity in Zero-Shot Reasoning", author = "Shaikh, Omar and Zhang, Hongxin and Held, William and Bernstein, Michael and Yang, Diyi", 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.244", doi = "10.18653/v1/2023.acl-long.244", pages = "4454--4470", abstract = "Generating a Chain of Thought (CoT) has been shown to consistently improve large language model (LLM) performance on a wide range of NLP tasks. However, prior work has mainly focused on logical reasoning tasks (e.g. arithmetic, commonsense QA); it remains unclear whether improvements hold for more diverse types of reasoning, especially in socially situated contexts. Concretely, we perform a controlled evaluation of zero-shot CoT across two socially sensitive domains: harmful questions and stereotype benchmarks. We find that zero-shot CoT reasoning in sensitive domains significantly increases a model{'}s likelihood to produce harmful or undesirable output, with trends holding across different prompt formats and model variants. Furthermore, we show that harmful CoTs increase with model size, but decrease with improved instruction following. Our work suggests that zero-shot CoT should be used with caution on socially important tasks, especially when marginalized groups or sensitive topics are involved.", }

Generating a Chain of Thought (CoT) has been shown to consistently improve large language model (LLM) performance on a wide range of NLP tasks. However, prior work has mainly focused on logical reasoning tasks (e.g. arithmetic, commonsense QA); it remains unclear whether improvements hold for more diverse types of reasoning, especially in socially situated contexts. Concretely, we perform a controlled evaluation of zero-shot CoT across two socially sensitive domains: harmful questions and stereotype benchmarks. We find that zero-shot CoT reasoning in sensitive domains significantly increases a model{'}s likelihood to produce harmful or undesirable output, with trends holding across different prompt formats and model variants. Furthermore, we show that harmful CoTs increase with model size, but decrease with improved instruction following. Our work suggests that zero-shot CoT should be used with caution on socially important tasks, especially when marginalized groups or sensitive topics are involved.

On Second Thought, Let's Not Think Step by Step! Bias and Toxicity in Zero-Shot Reasoning

acl-long.244

Poster

2212.08061

https://aclanthology.org/2023.acl-long.245.bib

https://aclanthology.org/2023.acl-long.245/

@inproceedings{zhu-etal-2023-solving, title = "Solving Math Word Problems via Cooperative Reasoning induced Language Models", author = "Zhu, Xinyu and Wang, Junjie and Zhang, Lin and Zhang, Yuxiang and Huang, Yongfeng and Gan, Ruyi and Zhang, Jiaxing and Yang, Yujiu", 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.245", doi = "10.18653/v1/2023.acl-long.245", pages = "4471--4485", abstract = "Large-scale pre-trained language models (PLMs) bring new opportunities to challenging problems, especially those that need high-level intelligence, such as the math word problem (MWPs). However, directly applying existing PLMs to MWPs can fail as the generation process lacks sufficient supervision and thus lacks fast adaptivity as humans. We notice that human reasoning has a dual reasoning framework that consists of an immediate reaction system (system 1) and a delicate reasoning system (system 2), where the entire reasoning is determined by their interaction. This inspires us to develop a cooperative reasoning-induced PLM for solving MWPs, called Cooperative Reasoning (CoRe), resulting in a human-like reasoning architecture with system 1 as the generator and system 2 as the verifier. In our approach, the generator is responsible for generating reasoning paths, and the verifiers are used to supervise the evaluation in order to obtain reliable feedback for the generator. We evaluate our CoRe framework on several mathematical reasoning datasets and achieve decent improvement over state-of-the-art methods, up to 9.6{\%} increase over best baselines.", }

Large-scale pre-trained language models (PLMs) bring new opportunities to challenging problems, especially those that need high-level intelligence, such as the math word problem (MWPs). However, directly applying existing PLMs to MWPs can fail as the generation process lacks sufficient supervision and thus lacks fast adaptivity as humans. We notice that human reasoning has a dual reasoning framework that consists of an immediate reaction system (system 1) and a delicate reasoning system (system 2), where the entire reasoning is determined by their interaction. This inspires us to develop a cooperative reasoning-induced PLM for solving MWPs, called Cooperative Reasoning (CoRe), resulting in a human-like reasoning architecture with system 1 as the generator and system 2 as the verifier. In our approach, the generator is responsible for generating reasoning paths, and the verifiers are used to supervise the evaluation in order to obtain reliable feedback for the generator. We evaluate our CoRe framework on several mathematical reasoning datasets and achieve decent improvement over state-of-the-art methods, up to 9.6{\%} increase over best baselines.

Solving Math Word Problems via Cooperative Reasoning induced Language Models

acl-long.245

Poster

2210.16257

https://huggingface.co/papers/2210.16257

https://aclanthology.org/2023.acl-long.246.bib

https://aclanthology.org/2023.acl-long.246/

@inproceedings{amrhein-etal-2023-exploiting, title = "Exploiting Biased Models to De-bias Text: A Gender-Fair Rewriting Model", author = {Amrhein, Chantal and Schottmann, Florian and Sennrich, Rico and L{\"a}ubli, Samuel}, 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.246", doi = "10.18653/v1/2023.acl-long.246", pages = "4486--4506", abstract = "Natural language generation models reproduce and often amplify the biases present in their training data. Previous research explored using sequence-to-sequence rewriting models to transform biased model outputs (or original texts) into more gender-fair language by creating pseudo training data through linguistic rules. However, this approach is not practical for languages with more complex morphology than English. We hypothesise that creating training data in the reverse direction, i.e. starting from gender-fair text, is easier for morphologically complex languages and show that it matches the performance of state-of-the-art rewriting models for English. To eliminate the rule-based nature of data creation, we instead propose using machine translation models to create gender-biased text from real gender-fair text via round-trip translation. Our approach allows us to train a rewriting model for German without the need for elaborate handcrafted rules. The outputs of this model increased gender-fairness as shown in a human evaluation study.", }

Natural language generation models reproduce and often amplify the biases present in their training data. Previous research explored using sequence-to-sequence rewriting models to transform biased model outputs (or original texts) into more gender-fair language by creating pseudo training data through linguistic rules. However, this approach is not practical for languages with more complex morphology than English. We hypothesise that creating training data in the reverse direction, i.e. starting from gender-fair text, is easier for morphologically complex languages and show that it matches the performance of state-of-the-art rewriting models for English. To eliminate the rule-based nature of data creation, we instead propose using machine translation models to create gender-biased text from real gender-fair text via round-trip translation. Our approach allows us to train a rewriting model for German without the need for elaborate handcrafted rules. The outputs of this model increased gender-fairness as shown in a human evaluation study.

Exploiting Biased Models to De-bias Text: A Gender-Fair Rewriting Model

acl-long.246

Poster

2305.11140

https://aclanthology.org/2023.acl-long.247.bib

https://aclanthology.org/2023.acl-long.247/

@inproceedings{akasaki-etal-2023-early, title = "Early Discovery of Disappearing Entities in Microblogs", author = "Akasaki, Satoshi and Yoshinaga, Naoki and Toyoda, Masashi", 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.247", doi = "10.18653/v1/2023.acl-long.247", pages = "4507--4520", abstract = "We make decisions by reacting to changes in the real world, particularly the emergence and disappearance of impermanent entities such as restaurants, services, and events. Because we want to avoid missing out on opportunities or making fruitless actions after those entities have disappeared, it is important to know when entities disappear as early as possible. We thus tackle the task of detecting disappearing entities from microblogs where various information is shared timely. The major challenge is detecting uncertain contexts of disappearing entities from noisy microblog posts. To collect such disappearing contexts, we design time-sensitive distant supervision, which utilizes entities from the knowledge base and time-series posts. Using this method, we actually build large-scale Twitter datasets of disappearing entities. To ensure robust detection in noisy environments, we refine pretrained word embeddings for the detection model on microblog streams in a timely manner. Experimental results on the Twitter datasets confirmed the effectiveness of the collected labeled data and refined word embeddings; the proposed method outperformed a baseline in terms of accuracy, and more than 70{\%} of the detected disappearing entities in Wikipedia are discovered earlier than the update on Wikipedia, with the average lead-time is over one month.", }

We make decisions by reacting to changes in the real world, particularly the emergence and disappearance of impermanent entities such as restaurants, services, and events. Because we want to avoid missing out on opportunities or making fruitless actions after those entities have disappeared, it is important to know when entities disappear as early as possible. We thus tackle the task of detecting disappearing entities from microblogs where various information is shared timely. The major challenge is detecting uncertain contexts of disappearing entities from noisy microblog posts. To collect such disappearing contexts, we design time-sensitive distant supervision, which utilizes entities from the knowledge base and time-series posts. Using this method, we actually build large-scale Twitter datasets of disappearing entities. To ensure robust detection in noisy environments, we refine pretrained word embeddings for the detection model on microblog streams in a timely manner. Experimental results on the Twitter datasets confirmed the effectiveness of the collected labeled data and refined word embeddings; the proposed method outperformed a baseline in terms of accuracy, and more than 70{\%} of the detected disappearing entities in Wikipedia are discovered earlier than the update on Wikipedia, with the average lead-time is over one month.

Early Discovery of Disappearing Entities in Microblogs

acl-long.247

Poster

2210.07404

https://aclanthology.org/2023.acl-long.248.bib

https://aclanthology.org/2023.acl-long.248/

@inproceedings{he-etal-2023-diffusionbert, title = "{D}iffusion{BERT}: Improving Generative Masked Language Models with Diffusion Models", author = "He, Zhengfu and Sun, Tianxiang and Tang, Qiong and Wang, Kuanning and Huang, Xuanjing and Qiu, Xipeng", 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.248", doi = "10.18653/v1/2023.acl-long.248", pages = "4521--4534", abstract = "We present DiffusionBERT, a new generative masked language model based on discrete dif- fusion models. Diffusion models and many pre- trained language models have a shared training objective, i.e., denoising, making it possible to combine the two powerful models and enjoy the best of both worlds. On the one hand, dif- fusion models offer a promising training strat- egy that helps improve the generation quality. On the other hand, pre-trained denoising lan- guage models (e.g., BERT) can be used as a good initialization that accelerates convergence. We explore training BERT to learn the reverse process of a discrete diffusion process with an absorbing state and elucidate several designs to improve it. First, we propose a new noise schedule for the forward diffusion process that controls the degree of noise added at each step based on the information of each token. Sec- ond, we investigate several designs of incorpo- rating the time step into BERT. Experiments on unconditional text generation demonstrate that DiffusionBERT achieves significant improve- ment over existing diffusion models for text (e.g., D3PM and Diffusion-LM) and previous generative masked language models in terms of perplexity and BLEU score. Promising re- sults in conditional generation tasks show that DiffusionBERT can generate texts of compa- rable quality and more diverse than a series of established baselines.", }

We present DiffusionBERT, a new generative masked language model based on discrete dif- fusion models. Diffusion models and many pre- trained language models have a shared training objective, i.e., denoising, making it possible to combine the two powerful models and enjoy the best of both worlds. On the one hand, dif- fusion models offer a promising training strat- egy that helps improve the generation quality. On the other hand, pre-trained denoising lan- guage models (e.g., BERT) can be used as a good initialization that accelerates convergence. We explore training BERT to learn the reverse process of a discrete diffusion process with an absorbing state and elucidate several designs to improve it. First, we propose a new noise schedule for the forward diffusion process that controls the degree of noise added at each step based on the information of each token. Sec- ond, we investigate several designs of incorpo- rating the time step into BERT. Experiments on unconditional text generation demonstrate that DiffusionBERT achieves significant improve- ment over existing diffusion models for text (e.g., D3PM and Diffusion-LM) and previous generative masked language models in terms of perplexity and BLEU score. Promising re- sults in conditional generation tasks show that DiffusionBERT can generate texts of compa- rable quality and more diverse than a series of established baselines.

DiffusionBERT: Improving Generative Masked Language Models with Diffusion Models

acl-long.248

Poster

2211.15029

https://aclanthology.org/2023.acl-long.249.bib

https://aclanthology.org/2023.acl-long.249/

@inproceedings{zhang-etal-2023-lifting, title = "Lifting the Curse of Capacity Gap in Distilling Language Models", author = "Zhang, Chen and Yang, Yang and Liu, Jiahao and Wang, Jingang and Xian, Yunsen and Wang, Benyou and Song, Dawei", 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.249", doi = "10.18653/v1/2023.acl-long.249", pages = "4535--4553", abstract = "Pretrained language models (LMs) have shown compelling performance on various downstream tasks, but unfortunately they require a tremendous amount of inference compute. Knowledge distillation finds a path to compress LMs to small ones with a teacher-student paradigm. However, when the capacity gap between the teacher and the student is large, a curse of capacity gap appears, invoking a deficiency in distilling LMs. While a few studies have been carried out to fill the gap, the curse is not yet well tackled. In this paper, we aim at lifting the curse of capacity gap via enlarging the capacity of the student without notably increasing the inference compute. Largely motivated by sparse activation regime of mixture of experts (MoE), we propose a mixture of minimal experts (MiniMoE), which imposes extra parameters to the student but introduces almost no additional inference compute. Experimental results on GLUE and CoNLL demonstrate the curse of capacity gap is lifted by the magic of MiniMoE to a large extent. MiniMoE also achieves the state-of-the-art performance at small FLOPs compared with a range of competitive baselines. With a compression rate as much as {\textasciitilde}50$\times$, MiniMoE preserves {\textasciitilde}95{\%} GLUE score of the teacher.", }

Pretrained language models (LMs) have shown compelling performance on various downstream tasks, but unfortunately they require a tremendous amount of inference compute. Knowledge distillation finds a path to compress LMs to small ones with a teacher-student paradigm. However, when the capacity gap between the teacher and the student is large, a curse of capacity gap appears, invoking a deficiency in distilling LMs. While a few studies have been carried out to fill the gap, the curse is not yet well tackled. In this paper, we aim at lifting the curse of capacity gap via enlarging the capacity of the student without notably increasing the inference compute. Largely motivated by sparse activation regime of mixture of experts (MoE), we propose a mixture of minimal experts (MiniMoE), which imposes extra parameters to the student but introduces almost no additional inference compute. Experimental results on GLUE and CoNLL demonstrate the curse of capacity gap is lifted by the magic of MiniMoE to a large extent. MiniMoE also achieves the state-of-the-art performance at small FLOPs compared with a range of competitive baselines. With a compression rate as much as {\textasciitilde}50$\times$, MiniMoE preserves {\textasciitilde}95{\%} GLUE score of the teacher.

Lifting the Curse of Capacity Gap in Distilling Language Models

acl-long.249

Oral

2305.12129

https://huggingface.co/papers/2305.12129

https://aclanthology.org/2023.acl-long.250.bib

https://aclanthology.org/2023.acl-long.250/

@inproceedings{deng-etal-2023-towards, title = "Towards Faithful Dialogues via Focus Learning", author = "Deng, Yifan and Zhang, Xingsheng and Huang, Heyan and Hu, Yue", 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.250", doi = "10.18653/v1/2023.acl-long.250", pages = "4554--4566", abstract = "Maintaining faithfulness between responses and knowledge is an important research topic for building reliable knowledge-grounded dialogue systems. Existing models heavily rely on elaborate data engineering or increasing the model{'}s parameters ignoring to track the tokens that significantly influence losses, which is decisive for the optimization direction of the model in each iteration. To address this issue, we propose Focus Learning (FocusL), a novel learning approach that adjusts the contribution of each token to the optimization direction by directly scaling the corresponding objective loss. Specifically, we first introduce a positioning method by utilizing similarity distributions between knowledge and each response token to locate knowledge-aware tokens. Then, we further design a similarity-to-weight transformation to provide dynamic token-level weights for the cross-entropy loss. Finally, we use the weighted loss to encourage the model to pay special attention to the knowledge utilization. Experimental results demonstrate that our method achieves the new state-of-the-art results and generates more reliable responses while maintaining training stability.", }

Maintaining faithfulness between responses and knowledge is an important research topic for building reliable knowledge-grounded dialogue systems. Existing models heavily rely on elaborate data engineering or increasing the model{'}s parameters ignoring to track the tokens that significantly influence losses, which is decisive for the optimization direction of the model in each iteration. To address this issue, we propose Focus Learning (FocusL), a novel learning approach that adjusts the contribution of each token to the optimization direction by directly scaling the corresponding objective loss. Specifically, we first introduce a positioning method by utilizing similarity distributions between knowledge and each response token to locate knowledge-aware tokens. Then, we further design a similarity-to-weight transformation to provide dynamic token-level weights for the cross-entropy loss. Finally, we use the weighted loss to encourage the model to pay special attention to the knowledge utilization. Experimental results demonstrate that our method achieves the new state-of-the-art results and generates more reliable responses while maintaining training stability.

Towards Faithful Dialogues via Focus Learning

acl-long.250

Poster

https://aclanthology.org/2023.acl-long.251.bib

https://aclanthology.org/2023.acl-long.251/

@inproceedings{fang-feng-2023-back, title = "Back Translation for Speech-to-text Translation Without Transcripts", author = "Fang, Qingkai 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.251", doi = "10.18653/v1/2023.acl-long.251", pages = "4567--4587", abstract = "The success of end-to-end speech-to-text translation (ST) is often achieved by utilizing source transcripts, e.g., by pre-training with automatic speech recognition (ASR) and machine translation (MT) tasks, or by introducing additional ASR and MT data. Unfortunately, transcripts are only sometimes available since numerous unwritten languages exist worldwide. In this paper, we aim to utilize large amounts of target-side monolingual data to enhance ST without transcripts. Motivated by the remarkable success of back translation in MT, we develop a back translation algorithm for ST (BT4ST) to synthesize pseudo ST data from monolingual target data. To ease the challenges posed by short-to-long generation and one-to-many mapping, we introduce self-supervised discrete units and achieve back translation by cascading a target-to-unit model and a unit-to-speech model. With our synthetic ST data, we achieve an average boost of 2.3 BLEU on MuST-C En-De, En-Fr, and En-Es datasets. More experiments show that our method is especially effective in low-resource scenarios.", }

The success of end-to-end speech-to-text translation (ST) is often achieved by utilizing source transcripts, e.g., by pre-training with automatic speech recognition (ASR) and machine translation (MT) tasks, or by introducing additional ASR and MT data. Unfortunately, transcripts are only sometimes available since numerous unwritten languages exist worldwide. In this paper, we aim to utilize large amounts of target-side monolingual data to enhance ST without transcripts. Motivated by the remarkable success of back translation in MT, we develop a back translation algorithm for ST (BT4ST) to synthesize pseudo ST data from monolingual target data. To ease the challenges posed by short-to-long generation and one-to-many mapping, we introduce self-supervised discrete units and achieve back translation by cascading a target-to-unit model and a unit-to-speech model. With our synthetic ST data, we achieve an average boost of 2.3 BLEU on MuST-C En-De, En-Fr, and En-Es datasets. More experiments show that our method is especially effective in low-resource scenarios.

Back Translation for Speech-to-text Translation Without Transcripts

acl-long.251

Poster

2305.08709

https://aclanthology.org/2023.acl-long.252.bib

https://aclanthology.org/2023.acl-long.252/

@inproceedings{aksu-etal-2023-prompter, title = "Prompter: Zero-shot Adaptive Prefixes for Dialogue State Tracking Domain Adaptation", author = "Aksu, Ibrahim Taha and Kan, Min-Yen and Chen, Nancy", 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.252", doi = "10.18653/v1/2023.acl-long.252", pages = "4588--4603", abstract = "A challenge in the Dialogue State Tracking (DST) field is adapting models to new domains without using any supervised data {---} zero-shot domain adaptation. Parameter-Efficient Transfer Learning (PETL) has the potential to address this problem due to its robustness. However, it has yet to be applied to the zero-shot scenarios, as it is not clear how to apply it unsupervisedly. Our method, Prompter, uses descriptions of target domain slots to generate dynamic prefixes that are concatenated to the key and values at each layer{'}s self-attention mechanism. This allows for the use of prefix-tuning in zero-shot. Prompter outperforms previous methods on both the MultiWOZ and SGD benchmarks. In generating prefixes, our analyses find that Prompter not only utilizes the semantics of slot descriptions but also how often the slots appear together in conversation. Moreover, Prompter{'}s gains are due to its improved ability to distinguish {''}none{''}-valued dialogue slots, compared against baselines.", }

A challenge in the Dialogue State Tracking (DST) field is adapting models to new domains without using any supervised data {---} zero-shot domain adaptation. Parameter-Efficient Transfer Learning (PETL) has the potential to address this problem due to its robustness. However, it has yet to be applied to the zero-shot scenarios, as it is not clear how to apply it unsupervisedly. Our method, Prompter, uses descriptions of target domain slots to generate dynamic prefixes that are concatenated to the key and values at each layer{'}s self-attention mechanism. This allows for the use of prefix-tuning in zero-shot. Prompter outperforms previous methods on both the MultiWOZ and SGD benchmarks. In generating prefixes, our analyses find that Prompter not only utilizes the semantics of slot descriptions but also how often the slots appear together in conversation. Moreover, Prompter{'}s gains are due to its improved ability to distinguish {''}none{''}-valued dialogue slots, compared against baselines.

Prompter: Zero-shot Adaptive Prefixes for Dialogue State Tracking Domain Adaptation

acl-long.252

Poster

2306.04724

https://huggingface.co/papers/2306.04724

https://aclanthology.org/2023.acl-long.253.bib

https://aclanthology.org/2023.acl-long.253/

@inproceedings{tang-etal-2023-enhancing, title = "Enhancing Dialogue Generation via Dynamic Graph Knowledge Aggregation", author = "Tang, Chen and Zhang, Hongbo and Loakman, Tyler and Lin, Chenghua and Guerin, 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.253", doi = "10.18653/v1/2023.acl-long.253", pages = "4604--4616", abstract = "Incorporating external graph knowledge into neural chatbot models has been proven effective for enhancing dialogue generation. However, in conventional graph neural networks (GNNs), message passing on a graph is independent from text, resulting in the graph representation hidden space differing from that of the text. This training regime of existing models therefore leads to a semantic gap between graph knowledge and text. In this study, we propose a novel framework for knowledge graph enhanced dialogue generation. We dynamically construct a multi-hop knowledge graph with pseudo nodes to involve the language model in feature aggregation within the graph at all steps. To avoid the semantic biases caused by learning on vanilla subgraphs, the proposed framework applies hierarchical graph attention to aggregate graph features on pseudo nodes and then attains a global feature. Therefore, the framework can better utilise the heterogeneous features from both the post and external graph knowledge. Extensive experiments demonstrate that our framework outperforms state-of-the-art (SOTA) baselines on dialogue generation. Further analysis also shows that our representation learning framework can fill the semantic gap by coagulating representations of both text and graph knowledge. Moreover, the language model also learns how to better select knowledge triples for a more informative response via exploiting subgraph patterns within our feature aggregation process. Our code and resources are available at \url{https://github.com/tangg555/SaBART}.", }

Incorporating external graph knowledge into neural chatbot models has been proven effective for enhancing dialogue generation. However, in conventional graph neural networks (GNNs), message passing on a graph is independent from text, resulting in the graph representation hidden space differing from that of the text. This training regime of existing models therefore leads to a semantic gap between graph knowledge and text. In this study, we propose a novel framework for knowledge graph enhanced dialogue generation. We dynamically construct a multi-hop knowledge graph with pseudo nodes to involve the language model in feature aggregation within the graph at all steps. To avoid the semantic biases caused by learning on vanilla subgraphs, the proposed framework applies hierarchical graph attention to aggregate graph features on pseudo nodes and then attains a global feature. Therefore, the framework can better utilise the heterogeneous features from both the post and external graph knowledge. Extensive experiments demonstrate that our framework outperforms state-of-the-art (SOTA) baselines on dialogue generation. Further analysis also shows that our representation learning framework can fill the semantic gap by coagulating representations of both text and graph knowledge. Moreover, the language model also learns how to better select knowledge triples for a more informative response via exploiting subgraph patterns within our feature aggregation process. Our code and resources are available at \url{https://github.com/tangg555/SaBART}.

Enhancing Dialogue Generation via Dynamic Graph Knowledge Aggregation

acl-long.253

Poster

2306.16195

https://aclanthology.org/2023.acl-long.254.bib

https://aclanthology.org/2023.acl-long.254/

@inproceedings{li-etal-2023-multi, title = "Multi-modal Action Chain Abductive Reasoning", author = "Li, Mengze and Wang, Tianbao and Xu, Jiahe and Han, Kairong and Zhang, Shengyu and Zhao, Zhou and Miao, Jiaxu and Zhang, Wenqiao and Pu, Shiliang and Wu, 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.254", doi = "10.18653/v1/2023.acl-long.254", pages = "4617--4628", abstract = "\textit{Abductive Reasoning}, has long been considered to be at the core ability of humans, which enables us to infer the most plausible explanation of incomplete known phenomena in daily life. However, such critical reasoning capability is rarely investigated for contemporary AI systems under such limited observations. To facilitate this research community, this paper sheds new light on \textit{Abductive Reasoning} by studying a new vision-language task, \textbf{M}ulti-modal \textbf{A}ction chain abductive \textbf{R}easoning (\textbf{MAR}), together with a large-scale \textit{Abductive Reasoning} dataset: Given an incomplete set of language described events, MAR aims to imagine the most plausible event by spatio-temporal grounding in past video and then infer the hypothesis of subsequent action chain that can best explain the language premise. To solve this task, we propose a strong baseline model that realizes MAR from two perspectives: (i) we first introduce the transformer, which learns to encode the observation to imagine the plausible event with explicitly interpretable event grounding in the video based on the commonsense knowledge recognition ability. (ii) To complete the assumption of a follow-up action chain, we design a novel symbolic module that can complete strict derivation of the progressive action chain layer by layer. We conducted extensive experiments on the proposed dataset, and the experimental study shows that the proposed model significantly outperforms existing video-language models in terms of effectiveness on our newly created MAR dataset.", }

\textit{Abductive Reasoning}, has long been considered to be at the core ability of humans, which enables us to infer the most plausible explanation of incomplete known phenomena in daily life. However, such critical reasoning capability is rarely investigated for contemporary AI systems under such limited observations. To facilitate this research community, this paper sheds new light on \textit{Abductive Reasoning} by studying a new vision-language task, \textbf{M}ulti-modal \textbf{A}ction chain abductive \textbf{R}easoning (\textbf{MAR}), together with a large-scale \textit{Abductive Reasoning} dataset: Given an incomplete set of language described events, MAR aims to imagine the most plausible event by spatio-temporal grounding in past video and then infer the hypothesis of subsequent action chain that can best explain the language premise. To solve this task, we propose a strong baseline model that realizes MAR from two perspectives: (i) we first introduce the transformer, which learns to encode the observation to imagine the plausible event with explicitly interpretable event grounding in the video based on the commonsense knowledge recognition ability. (ii) To complete the assumption of a follow-up action chain, we design a novel symbolic module that can complete strict derivation of the progressive action chain layer by layer. We conducted extensive experiments on the proposed dataset, and the experimental study shows that the proposed model significantly outperforms existing video-language models in terms of effectiveness on our newly created MAR dataset.

Multi-modal Action Chain Abductive Reasoning

acl-long.254

Poster

https://aclanthology.org/2023.acl-long.255.bib

https://aclanthology.org/2023.acl-long.255/

@inproceedings{he-etal-2023-exploring, title = "Exploring the Capacity of Pretrained Language Models for Reasoning about Actions and Change", author = "He, Weinan and Huang, Canming and Xiao, Zhanhao and Liu, Yongmei", 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.255", doi = "10.18653/v1/2023.acl-long.255", pages = "4629--4643", abstract = "Reasoning about actions and change (RAC) is essential to understand and interact with the ever-changing environment. Previous AI research has shown the importance of fundamental and indispensable knowledge of actions, i.e., preconditions and effects. However, traditional methods rely on logical formalization which hinders practical applications. With recent transformer-based language models (LMs), reasoning over text is desirable and seemingly feasible, leading to the question of whether LMs can effectively and efficiently learn to solve RAC problems. We propose four essential RAC tasks as a comprehensive textual benchmark and generate problems in a way that minimizes the influence of other linguistic requirements (e.g., grounding) to focus on RAC. The resulting benchmark, TRAC, encompassing problems of various complexities, facilitates a more granular evaluation of LMs, precisely targeting the structural generalization ability much needed for RAC. Experiments with three high-performing transformers indicate that additional efforts are needed to tackle challenges raised by TRAC.", }

Reasoning about actions and change (RAC) is essential to understand and interact with the ever-changing environment. Previous AI research has shown the importance of fundamental and indispensable knowledge of actions, i.e., preconditions and effects. However, traditional methods rely on logical formalization which hinders practical applications. With recent transformer-based language models (LMs), reasoning over text is desirable and seemingly feasible, leading to the question of whether LMs can effectively and efficiently learn to solve RAC problems. We propose four essential RAC tasks as a comprehensive textual benchmark and generate problems in a way that minimizes the influence of other linguistic requirements (e.g., grounding) to focus on RAC. The resulting benchmark, TRAC, encompassing problems of various complexities, facilitates a more granular evaluation of LMs, precisely targeting the structural generalization ability much needed for RAC. Experiments with three high-performing transformers indicate that additional efforts are needed to tackle challenges raised by TRAC.

Exploring the Capacity of Pretrained Language Models for Reasoning about Actions and Change

acl-long.255

Poster

https://aclanthology.org/2023.acl-long.256.bib

https://aclanthology.org/2023.acl-long.256/

@inproceedings{li-etal-2023-unified, title = "Unified Demonstration Retriever for In-Context Learning", author = "Li, Xiaonan and Lv, Kai and Yan, Hang and Lin, Tianyang and Zhu, Wei and Ni, Yuan and Xie, Guotong and Wang, Xiaoling and Qiu, Xipeng", 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.256", doi = "10.18653/v1/2023.acl-long.256", pages = "4644--4668", abstract = "In-context learning is a new learning paradigm where a language model conditions on a few input-output pairs (demonstrations) and a test input, and directly outputs the prediction. It has been shown sensitive to the provided demonstrations and thus promotes the research of demonstration retrieval: given a test input, relevant examples are retrieved from the training set to serve as informative demonstrations for in-context learning. While previous works train task-specific retrievers for several tasks separately, these methods are hard to transfer and scale on various tasks, and separately trained retrievers will cause a lot of parameter storage and deployment cost. In this paper, we propose Unified Demonstration Retriever (UDR), a single model to retrieve demonstrations for a wide range of tasks. To train UDR, we cast various tasks{'} training signals into a unified list-wise ranking formulation by language model{'}s feedback. Then we propose a multi-task list-wise ranking training framework with an iterative mining strategy to find high-quality candidates, which can help UDR fully incorporate various tasks{'} signals. Experiments on 30+ tasks across 13 task families and multiple data domains show that UDR significantly outperforms baselines. Further analyses show the effectiveness of each proposed component and UDR{'}s strong ability in various scenarios including different LMs (1.3B 175B), unseen datasets, varying demonstration quantities, etc. We will release the code and model checkpoint after review.", }

In-context learning is a new learning paradigm where a language model conditions on a few input-output pairs (demonstrations) and a test input, and directly outputs the prediction. It has been shown sensitive to the provided demonstrations and thus promotes the research of demonstration retrieval: given a test input, relevant examples are retrieved from the training set to serve as informative demonstrations for in-context learning. While previous works train task-specific retrievers for several tasks separately, these methods are hard to transfer and scale on various tasks, and separately trained retrievers will cause a lot of parameter storage and deployment cost. In this paper, we propose Unified Demonstration Retriever (UDR), a single model to retrieve demonstrations for a wide range of tasks. To train UDR, we cast various tasks{'} training signals into a unified list-wise ranking formulation by language model{'}s feedback. Then we propose a multi-task list-wise ranking training framework with an iterative mining strategy to find high-quality candidates, which can help UDR fully incorporate various tasks{'} signals. Experiments on 30+ tasks across 13 task families and multiple data domains show that UDR significantly outperforms baselines. Further analyses show the effectiveness of each proposed component and UDR{'}s strong ability in various scenarios including different LMs (1.3B 175B), unseen datasets, varying demonstration quantities, etc. We will release the code and model checkpoint after review.

Unified Demonstration Retriever for In-Context Learning

acl-long.256

Oral

2305.04320

https://aclanthology.org/2023.acl-long.257.bib

https://aclanthology.org/2023.acl-long.257/

@inproceedings{yue-etal-2023-movie101, title = "Movie101: A New Movie Understanding Benchmark", author = "Yue, Zihao and Zhang, Qi and Hu, Anwen and Zhang, Liang and Wang, Ziheng 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.257", doi = "10.18653/v1/2023.acl-long.257", pages = "4669--4684", abstract = "To help the visually impaired enjoy movies, automatic movie narrating systems are expected to narrate accurate, coherent, and role-aware plots when there are no speaking lines of actors. Existing works benchmark this challenge as a normal video captioning task via some simplifications, such as removing role names and evaluating narrations with ngram-based metrics, which makes it difficult for automatic systems to meet the needs of real application scenarios. To narrow this gap, we construct a large-scale Chinese movie benchmark, named Movie101. Closer to real scenarios, the Movie Clip Narrating (MCN) task in our benchmark asks models to generate role-aware narration paragraphs for complete movie clips where no actors are speaking. External knowledge, such as role information and movie genres, is also provided for better movie understanding. Besides, we propose a new metric called Movie Narration Score (MNScore) for movie narrating evaluation, which achieves the best correlation with human evaluation. Our benchmark also supports the Temporal Narration Grounding (TNG) task to investigate clip localization given text descriptions. For both two tasks, our proposed methods well leverage external knowledge and outperform carefully designed baselines. The dataset and codes are released at \url{https://github.com/yuezih/Movie101}.", }

To help the visually impaired enjoy movies, automatic movie narrating systems are expected to narrate accurate, coherent, and role-aware plots when there are no speaking lines of actors. Existing works benchmark this challenge as a normal video captioning task via some simplifications, such as removing role names and evaluating narrations with ngram-based metrics, which makes it difficult for automatic systems to meet the needs of real application scenarios. To narrow this gap, we construct a large-scale Chinese movie benchmark, named Movie101. Closer to real scenarios, the Movie Clip Narrating (MCN) task in our benchmark asks models to generate role-aware narration paragraphs for complete movie clips where no actors are speaking. External knowledge, such as role information and movie genres, is also provided for better movie understanding. Besides, we propose a new metric called Movie Narration Score (MNScore) for movie narrating evaluation, which achieves the best correlation with human evaluation. Our benchmark also supports the Temporal Narration Grounding (TNG) task to investigate clip localization given text descriptions. For both two tasks, our proposed methods well leverage external knowledge and outperform carefully designed baselines. The dataset and codes are released at \url{https://github.com/yuezih/Movie101}.

Movie101: A New Movie Understanding Benchmark

acl-long.257

Poster

2305.12140

https://aclanthology.org/2023.acl-long.258.bib

https://aclanthology.org/2023.acl-long.258/

@inproceedings{xu-etal-2023-enhancing, title = "Enhancing Language Representation with Constructional Information for Natural Language Understanding", author = "Xu, Lvxiaowei and Wu, Jianwang and Peng, Jiawei and Gong, Zhilin and Cai, Ming and Wang, Tianxiang", 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.258", doi = "10.18653/v1/2023.acl-long.258", pages = "4685--4705", abstract = "Natural language understanding (NLU) is an essential branch of natural language processing, which relies on representations generated by pre-trained language models (PLMs). However, PLMs primarily focus on acquiring lexico-semantic information, while they may be unable to adequately handle the meaning of constructions. To address this issue, we introduce construction grammar (CxG), which highlights the pairings of form and meaning, to enrich language representation. We adopt usage-based construction grammar as the basis of our work, which is highly compatible with statistical models such as PLMs. Then a HyCxG framework is proposed to enhance language representation through a three-stage solution. First, all constructions are extracted from sentences via a slot-constraints approach. As constructions can overlap with each other, bringing redundancy and imbalance, we formulate the conditional max coverage problem for selecting the discriminative constructions. Finally, we propose a relational hypergraph attention network to acquire representation from constructional information by capturing high-order word interactions among constructions. Extensive experiments demonstrate the superiority of the proposed model on a variety of NLU tasks.", }

Natural language understanding (NLU) is an essential branch of natural language processing, which relies on representations generated by pre-trained language models (PLMs). However, PLMs primarily focus on acquiring lexico-semantic information, while they may be unable to adequately handle the meaning of constructions. To address this issue, we introduce construction grammar (CxG), which highlights the pairings of form and meaning, to enrich language representation. We adopt usage-based construction grammar as the basis of our work, which is highly compatible with statistical models such as PLMs. Then a HyCxG framework is proposed to enhance language representation through a three-stage solution. First, all constructions are extracted from sentences via a slot-constraints approach. As constructions can overlap with each other, bringing redundancy and imbalance, we formulate the conditional max coverage problem for selecting the discriminative constructions. Finally, we propose a relational hypergraph attention network to acquire representation from constructional information by capturing high-order word interactions among constructions. Extensive experiments demonstrate the superiority of the proposed model on a variety of NLU tasks.

Enhancing Language Representation with Constructional Information for Natural Language Understanding

acl-long.258

Poster

2306.02819

https://aclanthology.org/2023.acl-long.259.bib

https://aclanthology.org/2023.acl-long.259/

@inproceedings{wang-etal-2023-query, title = "Query Structure Modeling for Inductive Logical Reasoning Over Knowledge Graphs", author = "Wang, Siyuan and Wei, Zhongyu and Han, Meng and Fan, Zhihao and Shan, Haijun and Zhang, Qi 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.259", doi = "10.18653/v1/2023.acl-long.259", pages = "4706--4718", abstract = "Logical reasoning over incomplete knowledge graphs to answer complex logical queries is a challenging task. With the emergence of new entities and relations in constantly evolving KGs, inductive logical reasoning over KGs has become a crucial problem. However, previous PLMs-based methods struggle to model the logical structures of complex queries, which limits their ability to generalize within the same structure. In this paper, we propose a structure-modeled textual encoding framework for inductive logical reasoning over KGs. It encodes linearized query structures and entities using pre-trained language models to find answers. For structure modeling of complex queries, we design stepwise instructions that implicitly prompt PLMs on the execution order of geometric operations in each query. We further separately model different geometric operations (i.e., projection, intersection, and union) on the representation space using a pre-trained encoder with additional attention and maxout layers to enhance structured modeling. We conduct experiments on two inductive logical reasoning datasets and three transductive datasets. The results demonstrate the effectiveness of our method on logical reasoning over KGs in both inductive and transductive settings.", }

Logical reasoning over incomplete knowledge graphs to answer complex logical queries is a challenging task. With the emergence of new entities and relations in constantly evolving KGs, inductive logical reasoning over KGs has become a crucial problem. However, previous PLMs-based methods struggle to model the logical structures of complex queries, which limits their ability to generalize within the same structure. In this paper, we propose a structure-modeled textual encoding framework for inductive logical reasoning over KGs. It encodes linearized query structures and entities using pre-trained language models to find answers. For structure modeling of complex queries, we design stepwise instructions that implicitly prompt PLMs on the execution order of geometric operations in each query. We further separately model different geometric operations (i.e., projection, intersection, and union) on the representation space using a pre-trained encoder with additional attention and maxout layers to enhance structured modeling. We conduct experiments on two inductive logical reasoning datasets and three transductive datasets. The results demonstrate the effectiveness of our method on logical reasoning over KGs in both inductive and transductive settings.

Query Structure Modeling for Inductive Logical Reasoning Over Knowledge Graphs

acl-long.259

Poster

2305.13585

https://aclanthology.org/2023.acl-long.260.bib

https://aclanthology.org/2023.acl-long.260/

@inproceedings{liu-etal-2023-dimongen, title = "{D}imon{G}en: Diversified Generative Commonsense Reasoning for Explaining Concept Relationships", author = "Liu, Chenzhengyi and Huang, Jie and Zhu, Kerui and Chang, Kevin Chen-Chuan", 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.260", doi = "10.18653/v1/2023.acl-long.260", pages = "4719--4731", abstract = "In this paper, we propose DimonGen, which aims to generate diverse sentences describing concept relationships in various everyday scenarios. To support this, we first create a benchmark dataset for this task by adapting the existing CommonGen dataset. We then propose a two-stage model called MoREE to generate the target sentences. MoREE consists of a mixture of retrievers model that retrieves diverse context sentences related to the given concepts, and a mixture of generators model that generates diverse sentences based on the retrieved contexts. We conduct experiments on the DimonGen task and show that MoREE outperforms strong baselines in terms of both the quality and diversity of the generated sentences. Our results demonstrate that MoREE is able to generate diverse sentences that reflect different relationships between concepts, leading to a comprehensive understanding of concept relationships.", }

In this paper, we propose DimonGen, which aims to generate diverse sentences describing concept relationships in various everyday scenarios. To support this, we first create a benchmark dataset for this task by adapting the existing CommonGen dataset. We then propose a two-stage model called MoREE to generate the target sentences. MoREE consists of a mixture of retrievers model that retrieves diverse context sentences related to the given concepts, and a mixture of generators model that generates diverse sentences based on the retrieved contexts. We conduct experiments on the DimonGen task and show that MoREE outperforms strong baselines in terms of both the quality and diversity of the generated sentences. Our results demonstrate that MoREE is able to generate diverse sentences that reflect different relationships between concepts, leading to a comprehensive understanding of concept relationships.

DimonGen: Diversified Generative Commonsense Reasoning for Explaining Concept Relationships

acl-long.260

Oral

2212.10545

https://aclanthology.org/2023.acl-long.261.bib

https://aclanthology.org/2023.acl-long.261/

@inproceedings{zhao-aletras-2023-incorporating, title = "Incorporating Attribution Importance for Improving Faithfulness Metrics", author = "Zhao, Zhixue and Aletras, Nikolaos", 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.261", doi = "10.18653/v1/2023.acl-long.261", pages = "4732--4745", abstract = "Feature attribution methods (FAs) are popular approaches for providing insights into the model reasoning process of making predictions. The more faithful a FA is, the more accurately it reflects which parts of the input are more important for the prediction. Widely used faithfulness metrics, such as sufficiency and comprehensiveness use a hard erasure criterion, i.e. entirely removing or retaining the top most important tokens ranked by a given FA and observing the changes in predictive likelihood. However, this hard criterion ignores the importance of each individual token, treating them all equally for computing sufficiency and comprehensiveness. In this paper, we propose a simple yet effective soft erasure criterion. Instead of entirely removing or retaining tokens from the input, we randomly mask parts of the token vector representations proportionately to their FA importance. Extensive experiments across various natural language processing tasks and different FAs show that our soft-sufficiency and soft-comprehensiveness metrics consistently prefer more faithful explanations compared to hard sufficiency and comprehensiveness.", }

Feature attribution methods (FAs) are popular approaches for providing insights into the model reasoning process of making predictions. The more faithful a FA is, the more accurately it reflects which parts of the input are more important for the prediction. Widely used faithfulness metrics, such as sufficiency and comprehensiveness use a hard erasure criterion, i.e. entirely removing or retaining the top most important tokens ranked by a given FA and observing the changes in predictive likelihood. However, this hard criterion ignores the importance of each individual token, treating them all equally for computing sufficiency and comprehensiveness. In this paper, we propose a simple yet effective soft erasure criterion. Instead of entirely removing or retaining tokens from the input, we randomly mask parts of the token vector representations proportionately to their FA importance. Extensive experiments across various natural language processing tasks and different FAs show that our soft-sufficiency and soft-comprehensiveness metrics consistently prefer more faithful explanations compared to hard sufficiency and comprehensiveness.

Incorporating Attribution Importance for Improving Faithfulness Metrics

acl-long.261

Oral

2305.10496

https://aclanthology.org/2023.acl-long.262.bib

https://aclanthology.org/2023.acl-long.262/

@inproceedings{pang-etal-2023-reward, title = "Reward Gaming in Conditional Text Generation", author = "Pang, Richard Yuanzhe and Padmakumar, Vishakh and Sellam, Thibault and Parikh, Ankur and He, He", 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.262", doi = "10.18653/v1/2023.acl-long.262", pages = "4746--4763", abstract = "To align conditional text generation model outputs with desired behaviors, there has been an increasing focus on training the model using reinforcement learning (RL) with reward functions learned from human annotations. Under this framework, we identify three common cases where high rewards are incorrectly assigned to undesirable patterns: noise-induced spurious correlation, naturally occurring spurious correlation, and covariate shift. We show that even though learned metrics achieve high performance on the distribution of the data used to train the reward function, the undesirable patterns may be amplified during RL training of the text generation model. While there has been discussion about reward gaming in the RL or safety community, in this discussion piece, we would like to highlight reward gaming in the natural language generation (NLG) community using concrete conditional text generation examples and discuss potential fixes and areas for future work.", }

To align conditional text generation model outputs with desired behaviors, there has been an increasing focus on training the model using reinforcement learning (RL) with reward functions learned from human annotations. Under this framework, we identify three common cases where high rewards are incorrectly assigned to undesirable patterns: noise-induced spurious correlation, naturally occurring spurious correlation, and covariate shift. We show that even though learned metrics achieve high performance on the distribution of the data used to train the reward function, the undesirable patterns may be amplified during RL training of the text generation model. While there has been discussion about reward gaming in the RL or safety community, in this discussion piece, we would like to highlight reward gaming in the natural language generation (NLG) community using concrete conditional text generation examples and discuss potential fixes and areas for future work.

Reward Gaming in Conditional Text Generation

acl-long.262

Oral

2211.08714

https://huggingface.co/papers/2211.08714

https://aclanthology.org/2023.acl-long.263.bib

https://aclanthology.org/2023.acl-long.263/

@inproceedings{sanchez-etal-2023-hidden, title = "Hidden Schema Networks", author = "Sanchez, Ramses and Conrads, Lukas and Welke, Pascal and Cvejoski, Kostadin and Ojeda Marin, Cesar", 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.263", doi = "10.18653/v1/2023.acl-long.263", pages = "4764--4798", abstract = "Large, pretrained language models infer powerful representations that encode rich semantic and syntactic content, albeit implicitly. In this work we introduce a novel neural language model that enforces, via inductive biases, explicit relational structures which allow for compositionality onto the output representations of pretrained language models. Specifically, the model encodes sentences into sequences of symbols (composed representations), which correspond to the nodes visited by biased random walkers on a global latent graph, and infers the posterior distribution of the latter. We first demonstrate that the model is able to uncover ground-truth graphs from artificially generated datasets of random token sequences. Next, we leverage pretrained BERT and GPT-2 language models as encoder and decoder, respectively, to infer networks of symbols (schemata) from natural language datasets. Our experiments show that (i) the inferred symbols can be interpreted as encoding different aspects of language, as e.g. topics or sentiments, and that (ii) GPT-2-like models can effectively be conditioned on symbolic representations. Finally, we explore training autoregressive, random walk {``}reasoning{''} models on schema networks inferred from commonsense knowledge databases, and using the sampled paths to enhance the performance of pretrained language models on commonsense If-Then reasoning tasks.", }

Large, pretrained language models infer powerful representations that encode rich semantic and syntactic content, albeit implicitly. In this work we introduce a novel neural language model that enforces, via inductive biases, explicit relational structures which allow for compositionality onto the output representations of pretrained language models. Specifically, the model encodes sentences into sequences of symbols (composed representations), which correspond to the nodes visited by biased random walkers on a global latent graph, and infers the posterior distribution of the latter. We first demonstrate that the model is able to uncover ground-truth graphs from artificially generated datasets of random token sequences. Next, we leverage pretrained BERT and GPT-2 language models as encoder and decoder, respectively, to infer networks of symbols (schemata) from natural language datasets. Our experiments show that (i) the inferred symbols can be interpreted as encoding different aspects of language, as e.g. topics or sentiments, and that (ii) GPT-2-like models can effectively be conditioned on symbolic representations. Finally, we explore training autoregressive, random walk {``}reasoning{''} models on schema networks inferred from commonsense knowledge databases, and using the sampled paths to enhance the performance of pretrained language models on commonsense If-Then reasoning tasks.

Hidden Schema Networks

acl-long.263

Poster

2207.03777

https://aclanthology.org/2023.acl-long.264.bib

https://aclanthology.org/2023.acl-long.264/

@inproceedings{liu-etal-2023-towards, title = "Towards Robust Low-Resource Fine-Tuning with Multi-View Compressed Representations", author = "Liu, Linlin and Li, Xingxuan and Thakkar, Megh and Li, Xin and Joty, Shafiq and Si, Luo and Bing, Lidong", 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.264", doi = "10.18653/v1/2023.acl-long.264", pages = "4799--4816", abstract = "Due to the huge amount of parameters, finetuning of pretrained language models (PLMs) is prone to overfitting in the low resource scenarios. In this work, we present a novel method that operates on the hidden representations of a PLM to reduce overfitting. During fine-tuning, our method inserts random autoencoders between the hidden layers of a PLM, which transform activations from the previous layers into multi-view compressed representations before feeding them into the upper layers. The autoencoders are plugged out after fine-tuning, so our method does not add extra parameters or increase computation cost during inference. Our method demonstrates promising performance improvement across a wide range of sequence- and token-level lowresource NLP tasks.", }

Due to the huge amount of parameters, finetuning of pretrained language models (PLMs) is prone to overfitting in the low resource scenarios. In this work, we present a novel method that operates on the hidden representations of a PLM to reduce overfitting. During fine-tuning, our method inserts random autoencoders between the hidden layers of a PLM, which transform activations from the previous layers into multi-view compressed representations before feeding them into the upper layers. The autoencoders are plugged out after fine-tuning, so our method does not add extra parameters or increase computation cost during inference. Our method demonstrates promising performance improvement across a wide range of sequence- and token-level lowresource NLP tasks.

Towards Robust Low-Resource Fine-Tuning with Multi-View Compressed Representations

acl-long.264

Poster

2211.08794

https://aclanthology.org/2023.acl-long.265.bib

https://aclanthology.org/2023.acl-long.265/

@inproceedings{stoehr-etal-2023-ordinal, title = "An Ordinal Latent Variable Model of Conflict Intensity", author = "Stoehr, Niklas and Torroba Hennigen, Lucas and Valvoda, Josef and West, Robert and Cotterell, Ryan and Schein, Aaron", 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.265", doi = "10.18653/v1/2023.acl-long.265", pages = "4817--4830", abstract = "Measuring the intensity of events is crucial for monitoring and tracking armed conflict. Advances in automated event extraction have yielded massive data sets of {``}who did what to whom{''} micro-records that enable data-driven approaches to monitoring conflict. The Goldstein scale is a widely-used expert-based measure that scores events on a conflictual{--}cooperative scale. It is based only on the action category ({``}what{''}) and disregards the subject ({``}who{''}) and object ({``}to whom{''}) of an event, as well as contextual information, like associated casualty count, that should contribute to the perception of an event{'}s {``}intensity{''}. This paper takes a latent variable-based approach to measuring conflict intensity. We introduce a probabilistic generative model that assumes each observed event is associated with a latent intensity class. A novel aspect of this model is that it imposes an ordering on the classes, such that higher-valued classes denote higher levels of intensity. The ordinal nature of the latent variable is induced from naturally ordered aspects of the data (e.g., casualty counts) where higher values naturally indicate higher intensity. We evaluate the proposed model both intrinsically and extrinsically, showing that it obtains comparatively good held-out predictive performance.", }

Measuring the intensity of events is crucial for monitoring and tracking armed conflict. Advances in automated event extraction have yielded massive data sets of {``}who did what to whom{''} micro-records that enable data-driven approaches to monitoring conflict. The Goldstein scale is a widely-used expert-based measure that scores events on a conflictual{--}cooperative scale. It is based only on the action category ({``}what{''}) and disregards the subject ({``}who{''}) and object ({``}to whom{''}) of an event, as well as contextual information, like associated casualty count, that should contribute to the perception of an event{'}s {``}intensity{''}. This paper takes a latent variable-based approach to measuring conflict intensity. We introduce a probabilistic generative model that assumes each observed event is associated with a latent intensity class. A novel aspect of this model is that it imposes an ordering on the classes, such that higher-valued classes denote higher levels of intensity. The ordinal nature of the latent variable is induced from naturally ordered aspects of the data (e.g., casualty counts) where higher values naturally indicate higher intensity. We evaluate the proposed model both intrinsically and extrinsically, showing that it obtains comparatively good held-out predictive performance.

An Ordinal Latent Variable Model of Conflict Intensity

acl-long.265

Poster

2210.03971

https://aclanthology.org/2023.acl-long.266.bib

https://aclanthology.org/2023.acl-long.266/

@inproceedings{saxon-wang-2023-multilingual, title = "Multilingual Conceptual Coverage in Text-to-Image Models", author = "Saxon, Michael and Wang, William 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.266", doi = "10.18653/v1/2023.acl-long.266", pages = "4831--4848", abstract = "We propose {``}Conceptual Coverage Across Languages{''} (CoCo-CroLa), a technique for benchmarking the degree to which any generative text-to-image system provides multilingual parity to its training language in terms of tangible nouns. For each model we can assess {``}conceptual coverage{''} of a given target language relative to a source language by comparing the population of images generated for a series of tangible nouns in the source language to the population of images generated for each noun under translation in the target language. This technique allows us to estimate how well-suited a model is to a target language as well as identify model-specific weaknesses, spurious correlations, and biases without a-priori assumptions. We demonstrate how it can be used to benchmark T2I models in terms of multilinguality, and how despite its simplicity it is a good proxy for impressive generalization.", }

We propose {``}Conceptual Coverage Across Languages{''} (CoCo-CroLa), a technique for benchmarking the degree to which any generative text-to-image system provides multilingual parity to its training language in terms of tangible nouns. For each model we can assess {``}conceptual coverage{''} of a given target language relative to a source language by comparing the population of images generated for a series of tangible nouns in the source language to the population of images generated for each noun under translation in the target language. This technique allows us to estimate how well-suited a model is to a target language as well as identify model-specific weaknesses, spurious correlations, and biases without a-priori assumptions. We demonstrate how it can be used to benchmark T2I models in terms of multilinguality, and how despite its simplicity it is a good proxy for impressive generalization.

Multilingual Conceptual Coverage in Text-to-Image Models

acl-long.266

Poster

2306.01735

https://aclanthology.org/2023.acl-long.267.bib

https://aclanthology.org/2023.acl-long.267/

@inproceedings{gu-etal-2023-pre, title = "Pre-Training to Learn in Context", author = "Gu, Yuxian and Dong, Li and Wei, Furu 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.267", doi = "10.18653/v1/2023.acl-long.267", pages = "4849--4870", abstract = "In-context learning, where pre-trained language models learn to perform tasks from task examples and instructions in their contexts, has attracted much attention in the NLP community. However, the ability of in-context learning is not fully exploited because language models are not explicitly trained to learn in context. To this end, we propose PICL (Pre-training for In-Context Learning), a framework to enhance the language models{'} in-context learning ability by pre-training the model on a large collection of {``}intrinsic tasks{''} in the general plain-text corpus using the simple language modeling objective. PICL encourages the model to infer and perform tasks by conditioning on the contexts while maintaining task generalization of pre-trained models. We evaluate the in-context learning performance of the model trained with PICL on seven widely-used text classification datasets and the Super-NaturalInstrctions benchmark, which contains 100+ NLP tasks formulated to text generation. Our experiments show that PICL is more effective and task-generalizable than a range of baselines, outperforming larger language models with nearly 4x parameters. The code is publicly available at \url{https://github.com/thu-coai/PICL}.", }

In-context learning, where pre-trained language models learn to perform tasks from task examples and instructions in their contexts, has attracted much attention in the NLP community. However, the ability of in-context learning is not fully exploited because language models are not explicitly trained to learn in context. To this end, we propose PICL (Pre-training for In-Context Learning), a framework to enhance the language models{'} in-context learning ability by pre-training the model on a large collection of {``}intrinsic tasks{''} in the general plain-text corpus using the simple language modeling objective. PICL encourages the model to infer and perform tasks by conditioning on the contexts while maintaining task generalization of pre-trained models. We evaluate the in-context learning performance of the model trained with PICL on seven widely-used text classification datasets and the Super-NaturalInstrctions benchmark, which contains 100+ NLP tasks formulated to text generation. Our experiments show that PICL is more effective and task-generalizable than a range of baselines, outperforming larger language models with nearly 4x parameters. The code is publicly available at \url{https://github.com/thu-coai/PICL}.

Pre-Training to Learn in Context

acl-long.267

Oral

2305.09137

https://aclanthology.org/2023.acl-long.268.bib

https://aclanthology.org/2023.acl-long.268/

@inproceedings{mager-etal-2023-ethical, title = "Ethical Considerations for Machine Translation of Indigenous Languages: Giving a Voice to the Speakers", author = "Mager, Manuel and Mager, Elisabeth and Kann, Katharina and Vu, Ngoc Thang", 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.268", doi = "10.18653/v1/2023.acl-long.268", pages = "4871--4897", abstract = "In recent years machine translation has become very successful for high-resource language pairs. This has also sparked new interest in research on the automatic translation of low-resource languages, including Indigenous languages. However, the latter are deeply related to the ethnic and cultural groups that speak (or used to speak) them. The data collection, modeling and deploying machine translation systems thus result in new ethical questions that must be addressed. Motivated by this, we first survey the existing literature on ethical considerations for the documentation, translation, and general natural language processing for Indigenous languages. Afterward, we conduct and analyze an interview study to shed light on the positions of community leaders, teachers, and language activists regarding ethical concerns for the automatic translation of their languages. Our results show that the inclusion, at different degrees, of native speakers and community members is vital to performing better and more ethical research on Indigenous languages.", }

In recent years machine translation has become very successful for high-resource language pairs. This has also sparked new interest in research on the automatic translation of low-resource languages, including Indigenous languages. However, the latter are deeply related to the ethnic and cultural groups that speak (or used to speak) them. The data collection, modeling and deploying machine translation systems thus result in new ethical questions that must be addressed. Motivated by this, we first survey the existing literature on ethical considerations for the documentation, translation, and general natural language processing for Indigenous languages. Afterward, we conduct and analyze an interview study to shed light on the positions of community leaders, teachers, and language activists regarding ethical concerns for the automatic translation of their languages. Our results show that the inclusion, at different degrees, of native speakers and community members is vital to performing better and more ethical research on Indigenous languages.

Ethical Considerations for Machine Translation of Indigenous Languages: Giving a Voice to the Speakers

acl-long.268

Poster

2305.19474

https://aclanthology.org/2023.acl-long.269.bib

https://aclanthology.org/2023.acl-long.269/

@inproceedings{ryan-etal-2023-revisiting, title = "Revisiting non-{E}nglish Text Simplification: A Unified Multilingual Benchmark", author = "Ryan, Michael and Naous, Tarek and Xu, Wei", 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.269", doi = "10.18653/v1/2023.acl-long.269", pages = "4898--4927", abstract = "Recent advancements in high-quality, large-scale English resources have pushed the frontier of English Automatic Text Simplification (ATS) research. However, less work has been done on multilingual text simplification due to the lack of a diverse evaluation benchmark that covers complex-simple sentence pairs in many languages. This paper introduces the MultiSim benchmark, a collection of 27 resources in 12 distinct languages containing over 1.7 million complex-simple sentence pairs. This benchmark will encourage research in developing more effective multilingual text simplification models and evaluation metrics. Our experiments using MultiSim with pre-trained multilingual language models reveal exciting performance improvements from multilingual training in non-English settings. We observe strong performance from Russian in zero-shot cross-lingual transfer to low-resource languages. We further show that few-shot prompting with BLOOM-176b achieves comparable quality to reference simplifications outperforming fine-tuned models in most languages. We validate these findings through human evaluation.", }

Recent advancements in high-quality, large-scale English resources have pushed the frontier of English Automatic Text Simplification (ATS) research. However, less work has been done on multilingual text simplification due to the lack of a diverse evaluation benchmark that covers complex-simple sentence pairs in many languages. This paper introduces the MultiSim benchmark, a collection of 27 resources in 12 distinct languages containing over 1.7 million complex-simple sentence pairs. This benchmark will encourage research in developing more effective multilingual text simplification models and evaluation metrics. Our experiments using MultiSim with pre-trained multilingual language models reveal exciting performance improvements from multilingual training in non-English settings. We observe strong performance from Russian in zero-shot cross-lingual transfer to low-resource languages. We further show that few-shot prompting with BLOOM-176b achieves comparable quality to reference simplifications outperforming fine-tuned models in most languages. We validate these findings through human evaluation.

Revisiting non-English Text Simplification: A Unified Multilingual Benchmark

acl-long.269

Poster

2305.15678

https://huggingface.co/papers/2305.15678

https://aclanthology.org/2023.acl-long.270.bib

https://aclanthology.org/2023.acl-long.270/

@inproceedings{gu-etal-2023-dont, title = "Don{'}t Generate, Discriminate: A Proposal for Grounding Language Models to Real-World Environments", author = "Gu, Yu and Deng, Xiang and Su, Yu", 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.270", doi = "10.18653/v1/2023.acl-long.270", pages = "4928--4949", abstract = "A key missing capacity of current language models (LMs) is grounding to real-world environments. Most existing work for grounded language understanding uses LMs to directly generate plans that can be executed in the environment to achieve the desired effects. It thereby casts the burden of ensuring grammaticality, faithfulness, and controllability all on the LMs. We propose Pangu, a generic framework for grounded language understanding that capitalizes on the discriminative ability of LMs instead of their generative ability. Pangu consists of a symbolic agent and a neural LM working in a concerted fashion: The agent explores the environment to incrementally construct valid plans, and the LM evaluates the plausibility of the candidate plans to guide the search process. A case study on the challenging problem of knowledge base question answering (KBQA), which features a massive environment, demonstrates the remarkable effectiveness and flexibility of Pangu: A BERT-base LM is sufficient for setting a new record on standard KBQA datasets, and larger LMs further bring substantial gains.Pangu also enables, for the first time, effective few-shot in-context learning for KBQA with large LMs such as Codex.", }

A key missing capacity of current language models (LMs) is grounding to real-world environments. Most existing work for grounded language understanding uses LMs to directly generate plans that can be executed in the environment to achieve the desired effects. It thereby casts the burden of ensuring grammaticality, faithfulness, and controllability all on the LMs. We propose Pangu, a generic framework for grounded language understanding that capitalizes on the discriminative ability of LMs instead of their generative ability. Pangu consists of a symbolic agent and a neural LM working in a concerted fashion: The agent explores the environment to incrementally construct valid plans, and the LM evaluates the plausibility of the candidate plans to guide the search process. A case study on the challenging problem of knowledge base question answering (KBQA), which features a massive environment, demonstrates the remarkable effectiveness and flexibility of Pangu: A BERT-base LM is sufficient for setting a new record on standard KBQA datasets, and larger LMs further bring substantial gains.Pangu also enables, for the first time, effective few-shot in-context learning for KBQA with large LMs such as Codex.

Don't Generate, Discriminate: A Proposal for Grounding Language Models to Real-World Environments

acl-long.270

Poster

2212.09736

https://huggingface.co/papers/2212.09736

https://aclanthology.org/2023.acl-long.271.bib

https://aclanthology.org/2023.acl-long.271/

@inproceedings{mireshghallah-etal-2023-privacy, title = "Privacy-Preserving Domain Adaptation of Semantic Parsers", author = "Mireshghallah, Fatemehsadat and Su, Yu and Hashimoto, Tatsunori and Eisner, Jason and Shin, Richard", 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.271", doi = "10.18653/v1/2023.acl-long.271", pages = "4950--4970", abstract = "Task-oriented dialogue systems often assist users with personal or confidential matters. For this reason, the developers of such a system are generally prohibited from observing actual usage. So how can they know where the system is failing and needs more training data or new functionality? In this work, we study ways in which realistic user utterances can be generated synthetically, to help increase the linguistic and functional coverage of the system, without compromising the privacy of actual users. To this end, we propose a two-stage Differentially Private (DP) generation method which first generates latent semantic parses, and then generates utterances based on the parses. Our proposed approach improves MAUVE by 2.5X and parse tree function-type overlap by 1.3X relative to current approaches for private synthetic data generation, improving both on fluency and semantic coverage. We further validate our approach on a realistic domain adaptation task of adding new functionality from private user data to a semantic parser, and show overall gains of 8.5{\%} points on its accuracy with the new feature.", }

Task-oriented dialogue systems often assist users with personal or confidential matters. For this reason, the developers of such a system are generally prohibited from observing actual usage. So how can they know where the system is failing and needs more training data or new functionality? In this work, we study ways in which realistic user utterances can be generated synthetically, to help increase the linguistic and functional coverage of the system, without compromising the privacy of actual users. To this end, we propose a two-stage Differentially Private (DP) generation method which first generates latent semantic parses, and then generates utterances based on the parses. Our proposed approach improves MAUVE by 2.5X and parse tree function-type overlap by 1.3X relative to current approaches for private synthetic data generation, improving both on fluency and semantic coverage. We further validate our approach on a realistic domain adaptation task of adding new functionality from private user data to a semantic parser, and show overall gains of 8.5{\%} points on its accuracy with the new feature.

Privacy-Preserving Domain Adaptation of Semantic Parsers

acl-long.271

Poster

2212.10520

https://aclanthology.org/2023.acl-long.272.bib

https://aclanthology.org/2023.acl-long.272/

@inproceedings{wei-etal-2023-guide, title = "Guide the Many-to-One Assignment: Open Information Extraction via {I}o{U}-aware Optimal Transport", author = "Wei, Kaiwen and Yang, Yiran and Jin, Li and Sun, Xian and Zhang, Zequn and Zhang, Jingyuan and Li, Xiao and Zhang, Linhao and Liu, Jintao and Zhi, Guo", 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.272", doi = "10.18653/v1/2023.acl-long.272", pages = "4971--4984", abstract = "Open Information Extraction (OIE) seeks to extract structured information from raw text without the limitations of close ontology. Recently, the detection-based OIE methods have received great attention from the community due to their parallelism. However, as the essential step of those models, how to assign ground truth labels to the parallelly generated tuple proposals remains under-exploited. The commonly utilized Hungarian algorithm for this procedure is restricted to handling one-to-one assignment among the desired tuples and tuple proposals, which ignores the correlation between proposals and affects the recall of the models. To solve this problem, we propose a dynamic many-to-one label assignment strategy named IOT. Concretely, the label assignment process in OIE is formulated as an Optimal Transport (OT) problem. We leverage the intersection-over-union (IoU) as the assignment quality measurement, and convert the problem of finding the best assignment solution to the one of solving the optimal transport plan by maximizing the IoU values. To further utilize the knowledge from the assignment, we design an Assignment-guided Multi-granularity loss (AM) by simultaneously considering word-level and tuple-level information. Experiment results show the proposed method outperforms the state-of-the-art models on three benchmarks.", }

Open Information Extraction (OIE) seeks to extract structured information from raw text without the limitations of close ontology. Recently, the detection-based OIE methods have received great attention from the community due to their parallelism. However, as the essential step of those models, how to assign ground truth labels to the parallelly generated tuple proposals remains under-exploited. The commonly utilized Hungarian algorithm for this procedure is restricted to handling one-to-one assignment among the desired tuples and tuple proposals, which ignores the correlation between proposals and affects the recall of the models. To solve this problem, we propose a dynamic many-to-one label assignment strategy named IOT. Concretely, the label assignment process in OIE is formulated as an Optimal Transport (OT) problem. We leverage the intersection-over-union (IoU) as the assignment quality measurement, and convert the problem of finding the best assignment solution to the one of solving the optimal transport plan by maximizing the IoU values. To further utilize the knowledge from the assignment, we design an Assignment-guided Multi-granularity loss (AM) by simultaneously considering word-level and tuple-level information. Experiment results show the proposed method outperforms the state-of-the-art models on three benchmarks.

Guide the Many-to-One Assignment: Open Information Extraction via IoU-aware Optimal Transport

acl-long.272

Poster

https://aclanthology.org/2023.acl-long.273.bib

https://aclanthology.org/2023.acl-long.273/

@inproceedings{zhao-etal-2023-actively, title = "Actively Supervised Clustering for Open Relation Extraction", author = "Zhao, Jun and Zhang, Yongxin and Zhang, Qi and Gui, Tao and Wei, Zhongyu and Peng, Minlong and Sun, Mingming", 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.273", doi = "10.18653/v1/2023.acl-long.273", pages = "4985--4997", abstract = "Current clustering-based Open Relation Extraction (OpenRE) methods usually adopt a two-stage pipeline, which simultaneously learns relation representations and assignments in the first stage, then manually labels relation for each cluster. However, unsupervised objectives struggle to explicitly optimize clusters to align with relational semantics, and the number of clusters K has to be supplied in advance. In this paper, we present a novel setting, named actively supervised clustering for OpenRE. Our insight lies in that clustering learning and relation labeling can be performed simultaneously, which provides the necessary guidance for clustering without a significant increase in human effort. Along with this setting, we propose an active labeling strategy tailored for clustering. Instead of only focusing on improving the clustering of relations that have been discovered, our strategy is encouraged to discover new relations through diversity regularization. This is particularly beneficial for long-tail relations in the real world. Experimental results show that our method is able to discover almost all relational clusters in the data and improve the SOTA methods by 13.8{\%} and 10.6{\%}, on two datasets respectively.", }

Current clustering-based Open Relation Extraction (OpenRE) methods usually adopt a two-stage pipeline, which simultaneously learns relation representations and assignments in the first stage, then manually labels relation for each cluster. However, unsupervised objectives struggle to explicitly optimize clusters to align with relational semantics, and the number of clusters K has to be supplied in advance. In this paper, we present a novel setting, named actively supervised clustering for OpenRE. Our insight lies in that clustering learning and relation labeling can be performed simultaneously, which provides the necessary guidance for clustering without a significant increase in human effort. Along with this setting, we propose an active labeling strategy tailored for clustering. Instead of only focusing on improving the clustering of relations that have been discovered, our strategy is encouraged to discover new relations through diversity regularization. This is particularly beneficial for long-tail relations in the real world. Experimental results show that our method is able to discover almost all relational clusters in the data and improve the SOTA methods by 13.8{\%} and 10.6{\%}, on two datasets respectively.

Actively Supervised Clustering for Open Relation Extraction

acl-long.273

Poster

2306.04968

https://aclanthology.org/2023.acl-long.274.bib

https://aclanthology.org/2023.acl-long.274/

@inproceedings{mo-etal-2023-convgqr, title = "{C}onv{GQR}: Generative Query Reformulation for Conversational Search", author = "Mo, Fengran and Mao, Kelong and Zhu, Yutao and Wu, Yihong and Huang, Kaiyu and Nie, Jian-Yun", 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.274", doi = "10.18653/v1/2023.acl-long.274", pages = "4998--5012", abstract = "In conversational search, the user{'}s real search intent for the current conversation turn is dependent on the previous conversation history. It is challenging to determine a good search query from the whole conversation context. To avoid the expensive re-training of the query encoder, most existing methods try to learn a rewriting model to de-contextualize the current query by mimicking the manual query rewriting. However, manually rewritten queries are not always the best search queries. Thus, training a rewriting model on them would lead to sub-optimal queries. Another useful information to enhance the search query is the potential answer to the question. In this paper, we propose ConvGQR, a new framework to reformulate conversational queries based on generative pre-trained language models (PLMs), one for query rewriting and another for generating potential answers. By combining both, ConvGQR can produce better search queries. In addition, to relate query reformulation to the retrieval task, we propose a knowledge infusion mechanism to optimize both query reformulation and retrieval. Extensive experiments on four conversational search datasets demonstrate the effectiveness of ConvGQR.", }

In conversational search, the user{'}s real search intent for the current conversation turn is dependent on the previous conversation history. It is challenging to determine a good search query from the whole conversation context. To avoid the expensive re-training of the query encoder, most existing methods try to learn a rewriting model to de-contextualize the current query by mimicking the manual query rewriting. However, manually rewritten queries are not always the best search queries. Thus, training a rewriting model on them would lead to sub-optimal queries. Another useful information to enhance the search query is the potential answer to the question. In this paper, we propose ConvGQR, a new framework to reformulate conversational queries based on generative pre-trained language models (PLMs), one for query rewriting and another for generating potential answers. By combining both, ConvGQR can produce better search queries. In addition, to relate query reformulation to the retrieval task, we propose a knowledge infusion mechanism to optimize both query reformulation and retrieval. Extensive experiments on four conversational search datasets demonstrate the effectiveness of ConvGQR.

ConvGQR: Generative Query Reformulation for Conversational Search

acl-long.274

Oral

2305.15645

https://aclanthology.org/2023.acl-long.275.bib

https://aclanthology.org/2023.acl-long.275/

@inproceedings{xu-etal-2023-kilm, title = "{KILM}: Knowledge Injection into Encoder-Decoder Language Models", author = "Xu, Yan and Namazifar, Mahdi and Hazarika, Devamanyu and Padmakumar, Aishwarya and Liu, Yang and Hakkani-Tur, Dilek", 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.275", doi = "10.18653/v1/2023.acl-long.275", pages = "5013--5035", abstract = "Large pre-trained language models (PLMs) have been shown to retain implicit knowledge within their parameters. To enhance this implicit knowledge, we propose Knowledge Injection into Language Models (KILM), a novel approach that injects entity-related knowledge into encoder-decoder PLMs, via a generative knowledge infilling objective through continued pre-training. This is done without architectural modifications to the PLMs or adding additional parameters. Experimental results over a suite of knowledge-intensive tasks spanning numerous datasets show that KILM enables models to retain more knowledge and hallucinate less while preserving their original performance on general NLU and NLG tasks. KILM also demonstrates improved zero-shot performances on tasks such as entity disambiguation, outperforming state-of-the-art models having 30x more parameters.", }

Large pre-trained language models (PLMs) have been shown to retain implicit knowledge within their parameters. To enhance this implicit knowledge, we propose Knowledge Injection into Language Models (KILM), a novel approach that injects entity-related knowledge into encoder-decoder PLMs, via a generative knowledge infilling objective through continued pre-training. This is done without architectural modifications to the PLMs or adding additional parameters. Experimental results over a suite of knowledge-intensive tasks spanning numerous datasets show that KILM enables models to retain more knowledge and hallucinate less while preserving their original performance on general NLU and NLG tasks. KILM also demonstrates improved zero-shot performances on tasks such as entity disambiguation, outperforming state-of-the-art models having 30x more parameters.

KILM: Knowledge Injection into Encoder-Decoder Language Models

acl-long.275

Oral

2302.09170

https://aclanthology.org/2023.acl-long.276.bib

https://aclanthology.org/2023.acl-long.276/

@inproceedings{wang-etal-2023-vstar, title = "{VSTAR}: A Video-grounded Dialogue Dataset for Situated Semantic Understanding with Scene and Topic Transitions", author = "Wang, Yuxuan and Zheng, Zilong and Zhao, Xueliang and Li, Jinpeng and Wang, Yueqian 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.276", doi = "10.18653/v1/2023.acl-long.276", pages = "5036--5048", abstract = "Video-grounded dialogue understanding is a challenging problem that requires machine to perceive, parse and reason over situated semantics extracted from weakly aligned video and dialogues. Most existing benchmarks treat both modalities the same as a frame-independent visual understanding task, while neglecting the intrinsic attributes in multimodal dialogues, such as scene and topic transitions. In this paper, we present \textbf{Video-grounded Scene{\&}Topic AwaRe dialogue (VSTAR)} dataset, a large scale video-grounded dialogue understanding dataset based on 395 TV series. Based on VSTAR, we propose two benchmarks for video-grounded dialogue understanding: scene segmentation and topic segmentation, and one benchmark for video-grounded dialogue generation. Comprehensive experiments are performed on these benchmarks to demonstrate the importance of multimodal information and segments in video-grounded dialogue understanding and generation.", }

Video-grounded dialogue understanding is a challenging problem that requires machine to perceive, parse and reason over situated semantics extracted from weakly aligned video and dialogues. Most existing benchmarks treat both modalities the same as a frame-independent visual understanding task, while neglecting the intrinsic attributes in multimodal dialogues, such as scene and topic transitions. In this paper, we present \textbf{Video-grounded Scene{\&}Topic AwaRe dialogue (VSTAR)} dataset, a large scale video-grounded dialogue understanding dataset based on 395 TV series. Based on VSTAR, we propose two benchmarks for video-grounded dialogue understanding: scene segmentation and topic segmentation, and one benchmark for video-grounded dialogue generation. Comprehensive experiments are performed on these benchmarks to demonstrate the importance of multimodal information and segments in video-grounded dialogue understanding and generation.

VSTAR: A Video-grounded Dialogue Dataset for Situated Semantic Understanding with Scene and Topic Transitions

acl-long.276

Poster

2305.18756

https://huggingface.co/papers/2305.18756

https://aclanthology.org/2023.acl-long.277.bib

https://aclanthology.org/2023.acl-long.277/

@inproceedings{dycke-etal-2023-nlpeer, title = "{NLP}eer: A Unified Resource for the Computational Study of Peer Review", author = "Dycke, Nils and Kuznetsov, Ilia and Gurevych, Iryna", 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.277", doi = "10.18653/v1/2023.acl-long.277", pages = "5049--5073", abstract = "Peer review constitutes a core component of scholarly publishing; yet it demands substantial expertise and training, and is susceptible to errors and biases. Various applications of NLP for peer reviewing assistance aim to support reviewers in this complex process, but the lack of clearly licensed datasets and multi-domain corpora prevent the systematic study of NLP for peer review. To remedy this, we introduce NLPeer{--} the first ethically sourced multidomain corpus of more than 5k papers and 11k review reports from five different venues. In addition to the new datasets of paper drafts, camera-ready versions and peer reviews from the NLP community, we establish a unified data representation and augment previous peer review datasets to include parsed and structured paper representations, rich metadata and versioning information. We complement our resource with implementations and analysis of three reviewing assistance tasks, including a novel guided skimming task. Our work paves the path towards systematic, multi-faceted, evidence-based study of peer review in NLP and beyond. The data and code are publicly available.", }

Peer review constitutes a core component of scholarly publishing; yet it demands substantial expertise and training, and is susceptible to errors and biases. Various applications of NLP for peer reviewing assistance aim to support reviewers in this complex process, but the lack of clearly licensed datasets and multi-domain corpora prevent the systematic study of NLP for peer review. To remedy this, we introduce NLPeer{--} the first ethically sourced multidomain corpus of more than 5k papers and 11k review reports from five different venues. In addition to the new datasets of paper drafts, camera-ready versions and peer reviews from the NLP community, we establish a unified data representation and augment previous peer review datasets to include parsed and structured paper representations, rich metadata and versioning information. We complement our resource with implementations and analysis of three reviewing assistance tasks, including a novel guided skimming task. Our work paves the path towards systematic, multi-faceted, evidence-based study of peer review in NLP and beyond. The data and code are publicly available.

NLPeer: A Unified Resource for the Computational Study of Peer Review

acl-long.277

Poster

2211.06651

https://aclanthology.org/2023.acl-long.278.bib

https://aclanthology.org/2023.acl-long.278/

@inproceedings{zheng-etal-2023-im, title = "{IM}-{TQA}: A {C}hinese Table Question Answering Dataset with Implicit and Multi-type Table Structures", author = "Zheng, Mingyu and Hao, Yang and Jiang, Wenbin and Lin, Zheng and Lyu, Yajuan and She, QiaoQiao 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.278", doi = "10.18653/v1/2023.acl-long.278", pages = "5074--5094", abstract = "Various datasets have been proposed to promote the development of Table Question Answering (TQA) technique. However, the problem setting of existing TQA benchmarks suffers from two limitations. First, they directly provide models with explicit table structures where row headers and column headers of the table are explicitly annotated and treated as model input during inference. Second, they only consider tables of limited types and ignore other tables especially complex tables with flexible header locations. Such simplified problem setting cannot cover practical scenarios where models need to process tables without header annotations in the inference phase or tables of different types. To address above issues, we construct a new TQA dataset with implicit and multi-type table structures, named IM-TQA, which not only requires the model to understand tables without directly available header annotations but also to handle multi-type tables including previously neglected complex tables. We investigate the performance of recent methods on our dataset and find that existing methods struggle in processing implicit and multi-type table structures. Correspondingly, we propose an RGCN-RCI framework outperforming recent baselines. We will release our dataset to facilitate future research.", }

Various datasets have been proposed to promote the development of Table Question Answering (TQA) technique. However, the problem setting of existing TQA benchmarks suffers from two limitations. First, they directly provide models with explicit table structures where row headers and column headers of the table are explicitly annotated and treated as model input during inference. Second, they only consider tables of limited types and ignore other tables especially complex tables with flexible header locations. Such simplified problem setting cannot cover practical scenarios where models need to process tables without header annotations in the inference phase or tables of different types. To address above issues, we construct a new TQA dataset with implicit and multi-type table structures, named IM-TQA, which not only requires the model to understand tables without directly available header annotations but also to handle multi-type tables including previously neglected complex tables. We investigate the performance of recent methods on our dataset and find that existing methods struggle in processing implicit and multi-type table structures. Correspondingly, we propose an RGCN-RCI framework outperforming recent baselines. We will release our dataset to facilitate future research.

IM-TQA: A Chinese Table Question Answering Dataset with Implicit and Multi-type Table Structures

acl-long.278

Poster

https://aclanthology.org/2023.acl-long.279.bib

https://aclanthology.org/2023.acl-long.279/

@inproceedings{he-etal-2023-z, title = "{Z}-Code++: A Pre-trained Language Model Optimized for Abstractive Summarization", author = "He, Pengcheng and Peng, Baolin and Wang, Song and Liu, Yang and Xu, Ruochen and Hassan, Hany and Shi, Yu and Zhu, Chenguang and Xiong, Wayne and Zeng, Michael and Gao, Jianfeng and Huang, Xuedong", 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.279", doi = "10.18653/v1/2023.acl-long.279", pages = "5095--5112", abstract = "This paper presents Z-Code++, a new pre-trained language model optimized for abstractive text summarization. The model extends the state-of-the-art encoder-decoder model using three techniques. First, we use a two-phase pre-training to improve the model{'}s performance on low-resource summarization tasks. The model is first pre-trained using text corpora for language understanding, then is continually pre-trained on summarization corpora for grounded text generation. Second, we replace self-attention layers in the encoder with disentangled attention layers, where each word is represented using two vectors that encode its content and position, respectively. Third, we use fusion-in-encoder, a simple yet effective method of encoding long sequences in a hierarchical manner. Z-Code++ createsa new state-of-the-art on 9 of 13 text summarization tasks across 5 languages. Our model is parameter-efficient in that it outperforms the 600x larger PaLM540B on XSum, and the finetuned 200x larger GPT3175B on SAMSum. In zero-shot and few-shot settings, our model substantially outperforms the competing models.", }

This paper presents Z-Code++, a new pre-trained language model optimized for abstractive text summarization. The model extends the state-of-the-art encoder-decoder model using three techniques. First, we use a two-phase pre-training to improve the model{'}s performance on low-resource summarization tasks. The model is first pre-trained using text corpora for language understanding, then is continually pre-trained on summarization corpora for grounded text generation. Second, we replace self-attention layers in the encoder with disentangled attention layers, where each word is represented using two vectors that encode its content and position, respectively. Third, we use fusion-in-encoder, a simple yet effective method of encoding long sequences in a hierarchical manner. Z-Code++ createsa new state-of-the-art on 9 of 13 text summarization tasks across 5 languages. Our model is parameter-efficient in that it outperforms the 600x larger PaLM540B on XSum, and the finetuned 200x larger GPT3175B on SAMSum. In zero-shot and few-shot settings, our model substantially outperforms the competing models.

Z-Code++: A Pre-trained Language Model Optimized for Abstractive Summarization

acl-long.279

Poster

2208.09770

https://aclanthology.org/2023.acl-long.280.bib

https://aclanthology.org/2023.acl-long.280/

@inproceedings{diao-etal-2023-mixture, title = "Mixture-of-Domain-Adapters: Decoupling and Injecting Domain Knowledge to Pre-trained Language Models{'} Memories", author = "Diao, Shizhe and Xu, Tianyang and Xu, Ruijia and Wang, Jiawei and Zhang, Tong", 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.280", doi = "10.18653/v1/2023.acl-long.280", pages = "5113--5129", abstract = "Pre-trained language models (PLMs) demonstrate excellent abilities to understand texts in the generic domain while struggling in a specific domain. Although continued pre-training on a large domain-specific corpus is effective, it is costly to tune all the parameters on the domain. In this paper, we investigate whether we can adapt PLMs both effectively and efficiently by only tuning a few parameters. Specifically, we decouple the feed-forward networks (FFNs) of the Transformer architecture into two parts: the original pre-trained FFNs to maintain the old-domain knowledge and our novel domain-specific adapters to inject domain-specific knowledge in parallel. Then we adopt a mixture-of-adapters gate to fuse the knowledge from different domain adapters dynamically. Our proposed Mixture-of-Domain-Adapters (MixDA) employs a two-stage adapter-tuning strategy that leverages both unlabeled data and labeled data to help the domain adaptation: $i$) domain-specific adapter on unlabeled data; followed by $ii$) the task-specific adapter on labeled data. MixDA can be seamlessly plugged into the pretraining-finetuning paradigm and our experiments demonstrate that MixDA achieves superior performance on in-domain tasks (GLUE), out-of-domain tasks (ChemProt, RCT, IMDB, Amazon), and knowledge-intensive tasks (KILT).Further analyses demonstrate the reliability, scalability, and efficiency of our method.", }

Pre-trained language models (PLMs) demonstrate excellent abilities to understand texts in the generic domain while struggling in a specific domain. Although continued pre-training on a large domain-specific corpus is effective, it is costly to tune all the parameters on the domain. In this paper, we investigate whether we can adapt PLMs both effectively and efficiently by only tuning a few parameters. Specifically, we decouple the feed-forward networks (FFNs) of the Transformer architecture into two parts: the original pre-trained FFNs to maintain the old-domain knowledge and our novel domain-specific adapters to inject domain-specific knowledge in parallel. Then we adopt a mixture-of-adapters gate to fuse the knowledge from different domain adapters dynamically. Our proposed Mixture-of-Domain-Adapters (MixDA) employs a two-stage adapter-tuning strategy that leverages both unlabeled data and labeled data to help the domain adaptation: $i$) domain-specific adapter on unlabeled data; followed by $ii$) the task-specific adapter on labeled data. MixDA can be seamlessly plugged into the pretraining-finetuning paradigm and our experiments demonstrate that MixDA achieves superior performance on in-domain tasks (GLUE), out-of-domain tasks (ChemProt, RCT, IMDB, Amazon), and knowledge-intensive tasks (KILT).Further analyses demonstrate the reliability, scalability, and efficiency of our method.

Mixture-of-Domain-Adapters: Decoupling and Injecting Domain Knowledge to Pre-trained Language Models' Memories

acl-long.280

Poster

https://aclanthology.org/2023.acl-long.281.bib

https://aclanthology.org/2023.acl-long.281/

@inproceedings{xu-etal-2023-unsupervised, title = "Unsupervised Graph-Text Mutual Conversion with a Unified Pretrained Language Model", author = "Xu, Yi and Sheng, Shuqian and Qi, Jiexing and Fu, Luoyi and Lin, Zhouhan and Wang, Xinbing and Zhou, Chenghu", 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.281", doi = "10.18653/v1/2023.acl-long.281", pages = "5130--5144", abstract = "Graph-to-text (G2T) generation and text-to-graph (T2G) triple extraction are two essential tasks for knowledge graphs. Existing unsupervised approaches become suitable candidates for jointly learning the two tasks due to their avoidance of using graph-text parallel data. However, they adopt multiple complex modules and still require entity information or relation type for training. To this end, we propose INFINITY, a simple yet effective unsupervised method with a unified pretrained language model that does not introduce external annotation tools or additional parallel information. It achieves fully unsupervised graph-text mutual conversion for the first time. Specifically, INFINITY treats both G2T and T2G as a bidirectional sequence generation task by fine-tuning only one pretrained seq2seq model. A novel back-translation-based framework is then designed to generate synthetic parallel data automatically. Besides, we investigate the impact of graph linearization and introduce the structure-aware fine-tuning strategy to alleviate possible performance deterioration via retaining structural information in graph sequences. As a fully unsupervised framework, INFINITY is empirically verified to outperform state-of-the-art baselines for G2T and T2G tasks. Additionally, we also devise a new training setting called cross learning for low-resource unsupervised information extraction.", }

Graph-to-text (G2T) generation and text-to-graph (T2G) triple extraction are two essential tasks for knowledge graphs. Existing unsupervised approaches become suitable candidates for jointly learning the two tasks due to their avoidance of using graph-text parallel data. However, they adopt multiple complex modules and still require entity information or relation type for training. To this end, we propose INFINITY, a simple yet effective unsupervised method with a unified pretrained language model that does not introduce external annotation tools or additional parallel information. It achieves fully unsupervised graph-text mutual conversion for the first time. Specifically, INFINITY treats both G2T and T2G as a bidirectional sequence generation task by fine-tuning only one pretrained seq2seq model. A novel back-translation-based framework is then designed to generate synthetic parallel data automatically. Besides, we investigate the impact of graph linearization and introduce the structure-aware fine-tuning strategy to alleviate possible performance deterioration via retaining structural information in graph sequences. As a fully unsupervised framework, INFINITY is empirically verified to outperform state-of-the-art baselines for G2T and T2G tasks. Additionally, we also devise a new training setting called cross learning for low-resource unsupervised information extraction.

Unsupervised Graph-Text Mutual Conversion with a Unified Pretrained Language Model

acl-long.281

Poster

https://aclanthology.org/2023.acl-long.282.bib

https://aclanthology.org/2023.acl-long.282/

@inproceedings{moon-etal-2023-randomized, title = "Randomized Smoothing with Masked Inference for Adversarially Robust Text Classifications", author = "Moon, Han Cheol and Joty, Shafiq and Zhao, Ruochen and Thakkar, Megh and Xu, Chi", 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.282", doi = "10.18653/v1/2023.acl-long.282", pages = "5145--5165", abstract = "Large-scale pre-trained language models have shown outstanding performance in a variety of NLP tasks. However, they are also known to be significantly brittle against specifically crafted adversarial examples, leading to increasing interest in probing the adversarial robustness of NLP systems. We introduce RSMI, a novel two-stage framework that combines randomized smoothing (RS) with masked inference (MI) to improve the adversarial robustness of NLP systems. RS transforms a classifier into a smoothed classifier to obtain robust representations, whereas MI forces a model to exploit the surrounding context of a masked token in an input sequence. RSMI improves adversarial robustness by 2 to 3 times over existing state-of-the-art methods on benchmark datasets. We also perform in-depth qualitative analysis to validate the effectiveness of the different stages of RSMI and probe the impact of its components through extensive ablations. By empirically proving the stability of RSMI, we put it forward as a practical method to robustly train large-scale NLP models. Our code and datasets are available at \url{https://github.com/Han8931/rsmi_nlp}", }

Large-scale pre-trained language models have shown outstanding performance in a variety of NLP tasks. However, they are also known to be significantly brittle against specifically crafted adversarial examples, leading to increasing interest in probing the adversarial robustness of NLP systems. We introduce RSMI, a novel two-stage framework that combines randomized smoothing (RS) with masked inference (MI) to improve the adversarial robustness of NLP systems. RS transforms a classifier into a smoothed classifier to obtain robust representations, whereas MI forces a model to exploit the surrounding context of a masked token in an input sequence. RSMI improves adversarial robustness by 2 to 3 times over existing state-of-the-art methods on benchmark datasets. We also perform in-depth qualitative analysis to validate the effectiveness of the different stages of RSMI and probe the impact of its components through extensive ablations. By empirically proving the stability of RSMI, we put it forward as a practical method to robustly train large-scale NLP models. Our code and datasets are available at \url{https://github.com/Han8931/rsmi_nlp}

Randomized Smoothing with Masked Inference for Adversarially Robust Text Classifications

acl-long.282

Poster

2305.06522

https://aclanthology.org/2023.acl-long.283.bib

https://aclanthology.org/2023.acl-long.283/

@inproceedings{xu-etal-2023-sescore2, title = "{SESCORE}2: Learning Text Generation Evaluation via Synthesizing Realistic Mistakes", author = "Xu, Wenda and Qian, Xian and Wang, Mingxuan and Li, Lei and Wang, William 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.283", doi = "10.18653/v1/2023.acl-long.283", pages = "5166--5183", abstract = "Is it possible to train a general metric for evaluating text generation quality without human-annotated ratings? Existing learned metrics either perform unsatisfactory across text generation tasks or require human ratings for training on specific tasks. In this paper, we propose SEScore2, a self-supervised approach for training a model-based metric for text generation evaluation. The key concept is to synthesize realistic model mistakes by perturbing sentences retrieved from a corpus. We evaluate SEScore2 and previous methods on four text generation tasks across three languages. SEScore2 outperforms all prior unsupervised metrics on four text generation evaluation benchmarks, with an average Kendall improvement of 0.158. Surprisingly, SEScore2 even outperforms the supervised BLEURT and COMET on multiple text generation tasks.", }

Is it possible to train a general metric for evaluating text generation quality without human-annotated ratings? Existing learned metrics either perform unsatisfactory across text generation tasks or require human ratings for training on specific tasks. In this paper, we propose SEScore2, a self-supervised approach for training a model-based metric for text generation evaluation. The key concept is to synthesize realistic model mistakes by perturbing sentences retrieved from a corpus. We evaluate SEScore2 and previous methods on four text generation tasks across three languages. SEScore2 outperforms all prior unsupervised metrics on four text generation evaluation benchmarks, with an average Kendall improvement of 0.158. Surprisingly, SEScore2 even outperforms the supervised BLEURT and COMET on multiple text generation tasks.

SESCORE2: Learning Text Generation Evaluation via Synthesizing Realistic Mistakes

acl-long.283

Poster

2212.09305

https://aclanthology.org/2023.acl-long.284.bib

https://aclanthology.org/2023.acl-long.284/

@inproceedings{zouhar-etal-2023-tokenization, title = "Tokenization and the Noiseless Channel", author = "Zouhar, Vil{\'e}m and Meister, Clara and Gastaldi, Juan and Du, Li and Sachan, Mrinmaya and Cotterell, Ryan", 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.284", doi = "10.18653/v1/2023.acl-long.284", pages = "5184--5207", abstract = "Subword tokenization is a key part of most NLP pipelines. However, little is known about why some tokenizer and hyperparameter combinations lead to improved downstream model performance over others. We propose that good tokenizers lead to efficient channel usage, where the channel is the means by which some input is conveyed to the model and efficiency can be quantified in information-theoretic terms as the ratio of the Shannon entropy to the maximum entropy of the subword distribution. Nevertheless, an optimal encoding according to Shannon entropy assigns extremely long codes to low-frequency subwords and very short codes to high-frequency subwords.Defining efficiency in terms of R{\'e}nyi entropy, on the other hand, penalizes distributions with either very high or very low-frequency subwords.We posit that (1) extremely high-frequency subwords are problematic because their meaning is not distinct and (2) that low-frequency subwords may not appear frequently enough for their meaning to be learned properly; encodings that induce unigram distributions with either can harm model performance. In machine translation, we find that across multiple tokenizers, the R{\'e}nyi entropy has a very strong correlation with BLEU: 0.82 in comparison to just -0.30 for compressed length.", }

Subword tokenization is a key part of most NLP pipelines. However, little is known about why some tokenizer and hyperparameter combinations lead to improved downstream model performance over others. We propose that good tokenizers lead to efficient channel usage, where the channel is the means by which some input is conveyed to the model and efficiency can be quantified in information-theoretic terms as the ratio of the Shannon entropy to the maximum entropy of the subword distribution. Nevertheless, an optimal encoding according to Shannon entropy assigns extremely long codes to low-frequency subwords and very short codes to high-frequency subwords.Defining efficiency in terms of R{\'e}nyi entropy, on the other hand, penalizes distributions with either very high or very low-frequency subwords.We posit that (1) extremely high-frequency subwords are problematic because their meaning is not distinct and (2) that low-frequency subwords may not appear frequently enough for their meaning to be learned properly; encodings that induce unigram distributions with either can harm model performance. In machine translation, we find that across multiple tokenizers, the R{\'e}nyi entropy has a very strong correlation with BLEU: 0.82 in comparison to just -0.30 for compressed length.

Tokenization and the Noiseless Channel

acl-long.284

Poster

2306.16842

https://aclanthology.org/2023.acl-long.285.bib

https://aclanthology.org/2023.acl-long.285/

@inproceedings{li-lu-2023-contextual, title = "Contextual Distortion Reveals Constituency: Masked Language Models are Implicit Parsers", author = "Li, Jiaxi and Lu, Wei", 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.285", doi = "10.18653/v1/2023.acl-long.285", pages = "5208--5222", abstract = "Recent advancements in pre-trained language models (PLMs) have demonstrated that these models possess some degree of syntactic awareness. To leverage this knowledge, we propose a novel chart-based method for extracting parse trees from masked language models (LMs) without the need to train separate parsers. Our method computes a score for each span based on the distortion of contextual representations resulting from linguistic perturbations. We design a set of perturbations motivated by the linguistic concept of constituency tests, and use these to score each span by aggregating the distortion scores. To produce a parse tree, we use chart parsing to find the tree with the minimum score. Our method consistently outperforms previous state-of-the-art methods on English with masked LMs, and also demonstrates superior performance in a multilingual setting, outperforming the state-of-the-art in 6 out of 8 languages. Notably, although our method does not involve parameter updates or extensive hyperparameter search, its performance can even surpass some unsupervised parsing methods that require fine-tuning. Our analysis highlights that the distortion of contextual representation resulting from syntactic perturbation can serve as an effective indicator of constituency across languages.", }

Recent advancements in pre-trained language models (PLMs) have demonstrated that these models possess some degree of syntactic awareness. To leverage this knowledge, we propose a novel chart-based method for extracting parse trees from masked language models (LMs) without the need to train separate parsers. Our method computes a score for each span based on the distortion of contextual representations resulting from linguistic perturbations. We design a set of perturbations motivated by the linguistic concept of constituency tests, and use these to score each span by aggregating the distortion scores. To produce a parse tree, we use chart parsing to find the tree with the minimum score. Our method consistently outperforms previous state-of-the-art methods on English with masked LMs, and also demonstrates superior performance in a multilingual setting, outperforming the state-of-the-art in 6 out of 8 languages. Notably, although our method does not involve parameter updates or extensive hyperparameter search, its performance can even surpass some unsupervised parsing methods that require fine-tuning. Our analysis highlights that the distortion of contextual representation resulting from syntactic perturbation can serve as an effective indicator of constituency across languages.

Contextual Distortion Reveals Constituency: Masked Language Models are Implicit Parsers

acl-long.285

Oral

2306.00645

https://aclanthology.org/2023.acl-long.286.bib

https://aclanthology.org/2023.acl-long.286/

@inproceedings{yue-etal-2023-metaadapt, title = "{M}eta{A}dapt: Domain Adaptive Few-Shot Misinformation Detection via Meta Learning", author = "Yue, Zhenrui and Zeng, Huimin and Zhang, Yang and Shang, Lanyu and Wang, 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.286", doi = "10.18653/v1/2023.acl-long.286", pages = "5223--5239", abstract = "With emerging topics (e.g., COVID-19) on social media as a source for the spreading misinformation, overcoming the distributional shifts between the original training domain (i.e., source domain) and such target domains remains a non-trivial task for misinformation detection. This presents an elusive challenge for early-stage misinformation detection, where a good amount of data and annotations from the target domain is not available for training. To address the data scarcity issue, we propose MetaAdapt, a meta learning based approach for domain adaptive few-shot misinformation detection. MetaAdapt leverages limited target examples to provide feedback and guide the knowledge transfer from the source to the target domain (i.e., learn to adapt). In particular, we train the initial model with multiple source tasks and compute their similarity scores to the meta task. Based on the similarity scores, we rescale the meta gradients to adaptively learn from the source tasks. As such, MetaAdapt can learn how to adapt the misinformation detection model and exploit the source data for improved performance in the target domain. To demonstrate the efficiency and effectiveness of our method, we perform extensive experiments to compare MetaAdapt with state-of-the-art baselines and large language models (LLMs) such as LLaMA, where MetaAdapt achieves better performance in domain adaptive few-shot misinformation detection with substantially reduced parameters on real-world datasets.", }

With emerging topics (e.g., COVID-19) on social media as a source for the spreading misinformation, overcoming the distributional shifts between the original training domain (i.e., source domain) and such target domains remains a non-trivial task for misinformation detection. This presents an elusive challenge for early-stage misinformation detection, where a good amount of data and annotations from the target domain is not available for training. To address the data scarcity issue, we propose MetaAdapt, a meta learning based approach for domain adaptive few-shot misinformation detection. MetaAdapt leverages limited target examples to provide feedback and guide the knowledge transfer from the source to the target domain (i.e., learn to adapt). In particular, we train the initial model with multiple source tasks and compute their similarity scores to the meta task. Based on the similarity scores, we rescale the meta gradients to adaptively learn from the source tasks. As such, MetaAdapt can learn how to adapt the misinformation detection model and exploit the source data for improved performance in the target domain. To demonstrate the efficiency and effectiveness of our method, we perform extensive experiments to compare MetaAdapt with state-of-the-art baselines and large language models (LLMs) such as LLaMA, where MetaAdapt achieves better performance in domain adaptive few-shot misinformation detection with substantially reduced parameters on real-world datasets.

MetaAdapt: Domain Adaptive Few-Shot Misinformation Detection via Meta Learning

acl-long.286

Poster

2305.12692

https://aclanthology.org/2023.acl-long.287.bib

https://aclanthology.org/2023.acl-long.287/

@inproceedings{wei-etal-2023-tackling, title = "Tackling Modality Heterogeneity with Multi-View Calibration Network for Multimodal Sentiment Detection", author = "Wei, Yiwei and Yuan, Shaozu and Yang, Ruosong and Shen, Lei and Li, Zhangmeizhi and Wang, Longbiao and Chen, Meng", 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.287", doi = "10.18653/v1/2023.acl-long.287", pages = "5240--5252", abstract = "With the popularity of social media, detecting sentiment from multimodal posts (e.g. image-text pairs) has attracted substantial attention recently. Existing works mainly focus on fusing different features but ignore the challenge of modality heterogeneity. Specifically, different modalities with inherent disparities may bring three problems: 1) introducing redundant visual features during feature fusion; 2) causing feature shift in the representation space; 3) leading to inconsistent annotations for different modal data. All these issues will increase the difficulty in understanding the sentiment of the multimodal content. In this paper, we propose a novel Multi-View Calibration Network (MVCN) to alleviate the above issues systematically. We first propose a text-guided fusion module with novel Sparse-Attention to reduce the negative impacts of redundant visual elements. We then devise a sentiment-based congruity constraint task to calibrate the feature shift in the representation space. Finally, we introduce an adaptive loss calibration strategy to tackle inconsistent annotated labels. Extensive experiments demonstrate the competitiveness of MVCN against previous approaches and achieve state-of-the-art results on two public benchmark datasets.", }

With the popularity of social media, detecting sentiment from multimodal posts (e.g. image-text pairs) has attracted substantial attention recently. Existing works mainly focus on fusing different features but ignore the challenge of modality heterogeneity. Specifically, different modalities with inherent disparities may bring three problems: 1) introducing redundant visual features during feature fusion; 2) causing feature shift in the representation space; 3) leading to inconsistent annotations for different modal data. All these issues will increase the difficulty in understanding the sentiment of the multimodal content. In this paper, we propose a novel Multi-View Calibration Network (MVCN) to alleviate the above issues systematically. We first propose a text-guided fusion module with novel Sparse-Attention to reduce the negative impacts of redundant visual elements. We then devise a sentiment-based congruity constraint task to calibrate the feature shift in the representation space. Finally, we introduce an adaptive loss calibration strategy to tackle inconsistent annotated labels. Extensive experiments demonstrate the competitiveness of MVCN against previous approaches and achieve state-of-the-art results on two public benchmark datasets.

Tackling Modality Heterogeneity with Multi-View Calibration Network for Multimodal Sentiment Detection

acl-long.287

Poster

https://aclanthology.org/2023.acl-long.288.bib

https://aclanthology.org/2023.acl-long.288/

@inproceedings{wang-etal-2023-cola, title = "{COLA}: Contextualized Commonsense Causal Reasoning from the Causal Inference Perspective", author = "Wang, Zhaowei and Do, Quyet V. and Zhang, Hongming and Zhang, Jiayao and Wang, Weiqi and Fang, Tianqing and Song, Yangqiu and Wong, Ginny and See, Simon", 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.288", doi = "10.18653/v1/2023.acl-long.288", pages = "5253--5271", abstract = "Detecting commonsense causal relations (causation) between events has long been an essential yet challenging task. Given that events are complicated, an event may have different causes under various contexts. Thus, exploiting context plays an essential role in detecting causal relations. Meanwhile, previous works about commonsense causation only consider two events and ignore their context, simplifying the task formulation. This paper proposes a new task to detect commonsense causation between two events in an event sequence (i.e., context), called contextualized commonsense causal reasoning. We also design a zero-shot framework: COLA (Contextualized Commonsense Causality Reasoner) to solve the task from the causal inference perspective. This framework obtains rich incidental supervision from temporality and balances covariates from multiple timestamps to remove confounding effects. Our extensive experiments show that COLA can detect commonsense causality more accurately than baselines.", }

Detecting commonsense causal relations (causation) between events has long been an essential yet challenging task. Given that events are complicated, an event may have different causes under various contexts. Thus, exploiting context plays an essential role in detecting causal relations. Meanwhile, previous works about commonsense causation only consider two events and ignore their context, simplifying the task formulation. This paper proposes a new task to detect commonsense causation between two events in an event sequence (i.e., context), called contextualized commonsense causal reasoning. We also design a zero-shot framework: COLA (Contextualized Commonsense Causality Reasoner) to solve the task from the causal inference perspective. This framework obtains rich incidental supervision from temporality and balances covariates from multiple timestamps to remove confounding effects. Our extensive experiments show that COLA can detect commonsense causality more accurately than baselines.

COLA: Contextualized Commonsense Causal Reasoning from the Causal Inference Perspective

acl-long.288

Oral

2305.05191

https://aclanthology.org/2023.acl-long.289.bib

https://aclanthology.org/2023.acl-long.289/

@inproceedings{sharma-etal-2023-memex, title = "{MEMEX}: Detecting Explanatory Evidence for Memes via Knowledge-Enriched Contextualization", author = "Sharma, Shivam and S, Ramaneswaran and Arora, Udit and Akhtar, Md. Shad and Chakraborty, Tanmoy", 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.289", doi = "10.18653/v1/2023.acl-long.289", pages = "5272--5290", abstract = "Memes are a powerful tool for communication over social media. Their affinity for evolving across politics, history, and sociocultural phenomena renders them an ideal vehicle for communication. To comprehend the subtle message conveyed within a meme, one must understand the relevant background that facilitates its holistic assimilation. Besides digital archiving of memes and their metadata by a few websites like knowyourmeme.com, currently, there is no efficient way to deduce a meme{'}s context dynamically. In this work, we propose a novel task, MEMEX - given a meme and a related document, the aim is to mine the context that succinctly explains the background of the meme. At first, we develop MCC (Meme Context Corpus), a novel dataset for MEMEX. Further, to benchmark MCC, we propose MIME (MultImodal Meme Explainer), a multimodal neural framework that uses external knowledge-enriched meme representation and a multi-level approach to capture the cross-modal semantic dependencies between the meme and the context. MIME surpasses several unimodal and multimodal systems and yields an absolute improvement of 4{\%} F1-score over the best baseline. Lastly, we conduct detailed analyses of MIME{'}s performance, highlighting the aspects that could lead to optimal modeling of cross-modal contextual associations.", }

Memes are a powerful tool for communication over social media. Their affinity for evolving across politics, history, and sociocultural phenomena renders them an ideal vehicle for communication. To comprehend the subtle message conveyed within a meme, one must understand the relevant background that facilitates its holistic assimilation. Besides digital archiving of memes and their metadata by a few websites like knowyourmeme.com, currently, there is no efficient way to deduce a meme{'}s context dynamically. In this work, we propose a novel task, MEMEX - given a meme and a related document, the aim is to mine the context that succinctly explains the background of the meme. At first, we develop MCC (Meme Context Corpus), a novel dataset for MEMEX. Further, to benchmark MCC, we propose MIME (MultImodal Meme Explainer), a multimodal neural framework that uses external knowledge-enriched meme representation and a multi-level approach to capture the cross-modal semantic dependencies between the meme and the context. MIME surpasses several unimodal and multimodal systems and yields an absolute improvement of 4{\%} F1-score over the best baseline. Lastly, we conduct detailed analyses of MIME{'}s performance, highlighting the aspects that could lead to optimal modeling of cross-modal contextual associations.

MEMEX: Detecting Explanatory Evidence for Memes via Knowledge-Enriched Contextualization

acl-long.289

Poster

2305.15913

https://aclanthology.org/2023.acl-long.290.bib

https://aclanthology.org/2023.acl-long.290/

@inproceedings{bolotova-baranova-etal-2023-wikihowqa, title = "{W}iki{H}ow{QA}: A Comprehensive Benchmark for Multi-Document Non-Factoid Question Answering", author = "Bolotova-Baranova, Valeriia and Blinov, Vladislav and Filippova, Sofya and Scholer, Falk and Sanderson, Mark", 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.290", doi = "10.18653/v1/2023.acl-long.290", pages = "5291--5314", abstract = "Answering non-factoid questions (NFQA) is a challenging task, requiring passage-level answers that are difficult to construct and evaluate. Search engines may provide a summary of a single web page, but many questions require reasoning across multiple documents. Meanwhile, modern models can generate highly coherent and fluent, but often factually incorrect answers that can deceive even non-expert humans. There is a critical need for high-quality resources for multi-document NFQA (MD-NFQA) to train new models and evaluate answers{'} grounding and factual consistency in relation to supporting documents. To address this gap, we introduce WikiHowQA, a new multi-document NFQA benchmark built on WikiHow, a website dedicated to answering {``}how-to{''} questions. The benchmark includes 11,746 human-written answers along with 74,527 supporting documents. We describe the unique challenges of the resource, provide strong baselines, and propose a novel human evaluation framework that utilizes highlighted relevant supporting passages to mitigate issues such as assessor unfamiliarity with the question topic. All code and data, including the automatic code for preparing the human evaluation, are publicly available.", }

Answering non-factoid questions (NFQA) is a challenging task, requiring passage-level answers that are difficult to construct and evaluate. Search engines may provide a summary of a single web page, but many questions require reasoning across multiple documents. Meanwhile, modern models can generate highly coherent and fluent, but often factually incorrect answers that can deceive even non-expert humans. There is a critical need for high-quality resources for multi-document NFQA (MD-NFQA) to train new models and evaluate answers{'} grounding and factual consistency in relation to supporting documents. To address this gap, we introduce WikiHowQA, a new multi-document NFQA benchmark built on WikiHow, a website dedicated to answering {``}how-to{''} questions. The benchmark includes 11,746 human-written answers along with 74,527 supporting documents. We describe the unique challenges of the resource, provide strong baselines, and propose a novel human evaluation framework that utilizes highlighted relevant supporting passages to mitigate issues such as assessor unfamiliarity with the question topic. All code and data, including the automatic code for preparing the human evaluation, are publicly available.

WikiHowQA: A Comprehensive Benchmark for Multi-Document Non-Factoid Question Answering

acl-long.290

Oral

https://aclanthology.org/2023.acl-long.291.bib

https://aclanthology.org/2023.acl-long.291/

@inproceedings{li-etal-2023-making, title = "Making Language Models Better Reasoners with Step-Aware Verifier", author = "Li, Yifei and Lin, Zeqi and Zhang, Shizhuo and Fu, Qiang and Chen, Bei and Lou, Jian-Guang and Chen, Weizhu", 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.291", doi = "10.18653/v1/2023.acl-long.291", pages = "5315--5333", abstract = "Few-shot learning is a challenging task that requires language models to generalize from limited examples. Large language models like GPT-3 and PaLM have made impressive progress in this area, but they still face difficulties in reasoning tasks such as GSM8K, a benchmark for arithmetic problems. To improve their reasoning skills, previous work has proposed to guide the language model with prompts that elicit a series of reasoning steps before giving the final answer, achieving a significant improvement on GSM8K from 17.9{\%} to 58.1{\%} in problem-solving rate. In this paper, we present DiVeRSe (Diverse Verifier on Reasoning Step), a novel approach that further enhances the reasoning capability of language models. DiVeRSe has three main components: first, it generates diverse prompts to explore different reasoning paths for the same question; second, it uses a verifier to filter out incorrect answers based on a weighted voting scheme; and third, it verifies each reasoning step individually instead of the whole chain. We evaluate DiVeRSe on the latest language model code-davinci-002 and show that it achieves new state-of-the-art results on six of eight reasoning benchmarks (e.g., GSM8K 74.4{\%} to 83.2{\%}).", }

Few-shot learning is a challenging task that requires language models to generalize from limited examples. Large language models like GPT-3 and PaLM have made impressive progress in this area, but they still face difficulties in reasoning tasks such as GSM8K, a benchmark for arithmetic problems. To improve their reasoning skills, previous work has proposed to guide the language model with prompts that elicit a series of reasoning steps before giving the final answer, achieving a significant improvement on GSM8K from 17.9{\%} to 58.1{\%} in problem-solving rate. In this paper, we present DiVeRSe (Diverse Verifier on Reasoning Step), a novel approach that further enhances the reasoning capability of language models. DiVeRSe has three main components: first, it generates diverse prompts to explore different reasoning paths for the same question; second, it uses a verifier to filter out incorrect answers based on a weighted voting scheme; and third, it verifies each reasoning step individually instead of the whole chain. We evaluate DiVeRSe on the latest language model code-davinci-002 and show that it achieves new state-of-the-art results on six of eight reasoning benchmarks (e.g., GSM8K 74.4{\%} to 83.2{\%}).

Making Language Models Better Reasoners with Step-Aware Verifier

acl-long.291

Poster

https://aclanthology.org/2023.acl-long.292.bib

https://aclanthology.org/2023.acl-long.292/

@inproceedings{ru-etal-2023-distributed, title = "Distributed Marker Representation for Ambiguous Discourse Markers and Entangled Relations", author = "Ru, Dongyu and Qiu, Lin and Qiu, Xipeng and Zhang, Yue and Zhang, Zheng", 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.292", doi = "10.18653/v1/2023.acl-long.292", pages = "5334--5351", abstract = "Discourse analysis is an important task because it models intrinsic semantic structures between sentences in a document. Discourse markers are natural representations of discourse in our daily language. One challenge is that the markers as well as pre-defined and human-labeled discourse relations can be ambiguous when describing the semantics between sentences. We believe that a better approach is to use a contextual-dependent distribution over the markers to express discourse information. In this work, we propose to learn a Distributed Marker Representation (DMR) by utilizing the (potentially) unlimited discourse marker data with a latent discourse sense, thereby bridging markers with sentence pairs. Such representations can be learned automatically from data without supervision, and in turn provide insights into the data itself. Experiments show the SOTA performance of our DMR on the implicit discourse relation recognition task and strong interpretability. Our method also offers a valuable tool to understand complex ambiguity and entanglement among discourse markers and manually defined discourse relations.", }

Discourse analysis is an important task because it models intrinsic semantic structures between sentences in a document. Discourse markers are natural representations of discourse in our daily language. One challenge is that the markers as well as pre-defined and human-labeled discourse relations can be ambiguous when describing the semantics between sentences. We believe that a better approach is to use a contextual-dependent distribution over the markers to express discourse information. In this work, we propose to learn a Distributed Marker Representation (DMR) by utilizing the (potentially) unlimited discourse marker data with a latent discourse sense, thereby bridging markers with sentence pairs. Such representations can be learned automatically from data without supervision, and in turn provide insights into the data itself. Experiments show the SOTA performance of our DMR on the implicit discourse relation recognition task and strong interpretability. Our method also offers a valuable tool to understand complex ambiguity and entanglement among discourse markers and manually defined discourse relations.

Distributed Marker Representation for Ambiguous Discourse Markers and Entangled Relations

acl-long.292

Poster

2306.10658

https://aclanthology.org/2023.acl-long.293.bib

https://aclanthology.org/2023.acl-long.293/

@inproceedings{hossain-etal-2023-misgendered, title = "{MISGENDERED}: Limits of Large Language Models in Understanding Pronouns", author = "Hossain, Tamanna and Dev, Sunipa and Singh, Sameer", 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.293", doi = "10.18653/v1/2023.acl-long.293", pages = "5352--5367", abstract = "Content Warning: This paper contains examples of misgendering and erasure that could be offensive and potentially triggering. Gender bias in language technologies has been widely studied, but research has mostly been restricted to a binary paradigm of gender. It is essential also to consider non-binary gender identities, as excluding them can cause further harm to an already marginalized group. In this paper, we comprehensively evaluate popular language models for their ability to correctly use English gender-neutral pronouns (e.g., singular they, them) and neo-pronouns (e.g., ze, xe, thon) that are used by individuals whose gender identity is not represented by binary pronouns. We introduce Misgendered, a framework for evaluating large language models{'} ability to correctly use preferred pronouns, consisting of (i) instances declaring an individual{'}s pronoun, followed by a sentence with a missing pronoun, and (ii) an experimental setup for evaluating masked and auto-regressive language models using a unified method. When prompted out-of-the-box, language models perform poorly at correctly predicting neo-pronouns (averaging 7.6{\%} accuracy) and gender-neutral pronouns (averaging 31.0{\%} accuracy). This inability to generalize results from a lack of representation of non-binary pronouns in training data and memorized associations. Few-shot adaptation with explicit examples in the prompt improves the performance but plateaus at only 45.4{\%} for neo-pronouns. We release the full dataset, code, and demo at \url{https://tamannahossainkay.github.io/misgendered/}.", }

Content Warning: This paper contains examples of misgendering and erasure that could be offensive and potentially triggering. Gender bias in language technologies has been widely studied, but research has mostly been restricted to a binary paradigm of gender. It is essential also to consider non-binary gender identities, as excluding them can cause further harm to an already marginalized group. In this paper, we comprehensively evaluate popular language models for their ability to correctly use English gender-neutral pronouns (e.g., singular they, them) and neo-pronouns (e.g., ze, xe, thon) that are used by individuals whose gender identity is not represented by binary pronouns. We introduce Misgendered, a framework for evaluating large language models{'} ability to correctly use preferred pronouns, consisting of (i) instances declaring an individual{'}s pronoun, followed by a sentence with a missing pronoun, and (ii) an experimental setup for evaluating masked and auto-regressive language models using a unified method. When prompted out-of-the-box, language models perform poorly at correctly predicting neo-pronouns (averaging 7.6{\%} accuracy) and gender-neutral pronouns (averaging 31.0{\%} accuracy). This inability to generalize results from a lack of representation of non-binary pronouns in training data and memorized associations. Few-shot adaptation with explicit examples in the prompt improves the performance but plateaus at only 45.4{\%} for neo-pronouns. We release the full dataset, code, and demo at \url{https://tamannahossainkay.github.io/misgendered/}.

MISGENDERED: Limits of Large Language Models in Understanding Pronouns

acl-long.293

Poster

2306.03950

https://aclanthology.org/2023.acl-long.294.bib

https://aclanthology.org/2023.acl-long.294/

@inproceedings{qiao-etal-2023-reasoning, title = "Reasoning with Language Model Prompting: A Survey", author = "Qiao, Shuofei and Ou, Yixin and Zhang, Ningyu and Chen, Xiang and Yao, Yunzhi and Deng, Shumin and Tan, Chuanqi and Huang, Fei and Chen, Huajun", 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.294", doi = "10.18653/v1/2023.acl-long.294", pages = "5368--5393", abstract = "Reasoning, as an essential ability for complex problem-solving, can provide back-end support for various real-world applications, such as medical diagnosis, negotiation, etc. This paper provides a comprehensive survey of cutting-edge research on reasoning with language model prompting. We introduce research works with comparisons and summaries and provide systematic resources to help beginners. We also discuss the potential reasons for emerging such reasoning abilities and highlight future research directions. Resources are available at \url{https://github.com/zjunlp/Prompt4ReasoningPapers} (updated periodically).", }

Reasoning, as an essential ability for complex problem-solving, can provide back-end support for various real-world applications, such as medical diagnosis, negotiation, etc. This paper provides a comprehensive survey of cutting-edge research on reasoning with language model prompting. We introduce research works with comparisons and summaries and provide systematic resources to help beginners. We also discuss the potential reasons for emerging such reasoning abilities and highlight future research directions. Resources are available at \url{https://github.com/zjunlp/Prompt4ReasoningPapers} (updated periodically).

Reasoning with Language Model Prompting: A Survey

acl-long.294

Poster

2212.09597

https://aclanthology.org/2023.acl-long.295.bib

https://aclanthology.org/2023.acl-long.295/

@inproceedings{futeral-etal-2023-tackling, title = "Tackling Ambiguity with Images: Improved Multimodal Machine Translation and Contrastive Evaluation", author = "Futeral, Matthieu and Schmid, Cordelia and Laptev, Ivan and Sagot, Beno{\^\i}t and Bawden, Rachel", 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.295", doi = "10.18653/v1/2023.acl-long.295", pages = "5394--5413", abstract = "One of the major challenges of machine translation (MT) is ambiguity, which can in some cases be resolved by accompanying context such as images. However, recent work in multimodal MT (MMT) has shown that obtaining improvements from images is challenging, limited not only by the difficulty of building effective cross-modal representations, but also by the lack of specific evaluation and training data. We present a new MMT approach based on a strong text-only MT model, which uses neural adapters, a novel guided self-attention mechanism and which is jointly trained on both visually-conditioned masking and MMT. We also introduce CoMMuTE, a Contrastive Multilingual Multimodal Translation Evaluation set of ambiguous sentences and their possible translations, accompanied by disambiguating images corresponding to each translation. Our approach obtains competitive results compared to strong text-only models on standard English→French, English→German and English→Czech benchmarks and outperforms baselines and state-of-the-art MMT systems by a large margin on our contrastive test set. Our code and CoMMuTE are freely available.", }

One of the major challenges of machine translation (MT) is ambiguity, which can in some cases be resolved by accompanying context such as images. However, recent work in multimodal MT (MMT) has shown that obtaining improvements from images is challenging, limited not only by the difficulty of building effective cross-modal representations, but also by the lack of specific evaluation and training data. We present a new MMT approach based on a strong text-only MT model, which uses neural adapters, a novel guided self-attention mechanism and which is jointly trained on both visually-conditioned masking and MMT. We also introduce CoMMuTE, a Contrastive Multilingual Multimodal Translation Evaluation set of ambiguous sentences and their possible translations, accompanied by disambiguating images corresponding to each translation. Our approach obtains competitive results compared to strong text-only models on standard English→French, English→German and English→Czech benchmarks and outperforms baselines and state-of-the-art MMT systems by a large margin on our contrastive test set. Our code and CoMMuTE are freely available.

Tackling Ambiguity with Images: Improved Multimodal Machine Translation and Contrastive Evaluation

acl-long.295

Poster

2212.10140

https://aclanthology.org/2023.acl-long.296.bib

https://aclanthology.org/2023.acl-long.296/

@inproceedings{guzman-nateras-etal-2023-hybrid, title = "Hybrid Knowledge Transfer for Improved Cross-Lingual Event Detection via Hierarchical Sample Selection", author = "Guzman Nateras, Luis and Dernoncourt, Franck and Nguyen, Thien", 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.296", doi = "10.18653/v1/2023.acl-long.296", pages = "5414--5427", abstract = "In this paper, we address the Event Detection task under a zero-shot cross-lingual setting where a model is trained on a source language but evaluated on a distinct target language for which there is no labeled data available. Most recent efforts in this field follow a direct transfer approach in which the model is trained using language-invariant features and then directly applied to the target language. However, we argue that these methods fail to take advantage of the benefits of the data transfer approach where a cross-lingual model is trained on target-language data and is able to learn task-specific information from syntactical features or word-label relations in the target language. As such, we propose a hybrid knowledge-transfer approach that leverages a teacher-student framework where the teacher and student networks are trained following the direct and data transfer approaches, respectively. Our method is complemented by a hierarchical training-sample selection scheme designed to address the issue of noisy labels being generated by the teacher model. Our model achieves state-of-the-art results on 9 morphologically-diverse target languages across 3 distinct datasets, highlighting the importance of exploiting the benefits of hybrid transfer.", }

In this paper, we address the Event Detection task under a zero-shot cross-lingual setting where a model is trained on a source language but evaluated on a distinct target language for which there is no labeled data available. Most recent efforts in this field follow a direct transfer approach in which the model is trained using language-invariant features and then directly applied to the target language. However, we argue that these methods fail to take advantage of the benefits of the data transfer approach where a cross-lingual model is trained on target-language data and is able to learn task-specific information from syntactical features or word-label relations in the target language. As such, we propose a hybrid knowledge-transfer approach that leverages a teacher-student framework where the teacher and student networks are trained following the direct and data transfer approaches, respectively. Our method is complemented by a hierarchical training-sample selection scheme designed to address the issue of noisy labels being generated by the teacher model. Our model achieves state-of-the-art results on 9 morphologically-diverse target languages across 3 distinct datasets, highlighting the importance of exploiting the benefits of hybrid transfer.

Hybrid Knowledge Transfer for Improved Cross-Lingual Event Detection via Hierarchical Sample Selection

acl-long.296

Poster

https://aclanthology.org/2023.acl-long.297.bib

https://aclanthology.org/2023.acl-long.297/

@inproceedings{yan-etal-2023-bleurt, title = "{BLEURT} Has Universal Translations: An Analysis of Automatic Metrics by Minimum Risk Training", author = "Yan, Yiming and Wang, Tao and Zhao, Chengqi and Huang, Shujian and Chen, Jiajun and Wang, Mingxuan", 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.297", doi = "10.18653/v1/2023.acl-long.297", pages = "5428--5443", abstract = "Automatic metrics play a crucial role in machine translation. Despite the widespread use of n-gram-based metrics, there has been a recent surge in the development of pre-trained model-based metrics that focus on measuring sentence semantics. However, these neural metrics, while achieving higher correlations with human evaluations, are often considered to be black boxes with potential biases that are difficult to detect. In this study, we systematically analyze and compare various mainstream and cutting-edge automatic metrics from the perspective of their guidance for training machine translation systems. Through Minimum Risk Training (MRT), we find that certain metrics exhibit robustness defects, such as the presence of universal adversarial translations in BLEURT and BARTScore. In-depth analysis suggests two main causes of these robustness deficits: distribution biases in the training datasets, and the tendency of the metric paradigm. By incorporating token-level constraints, we enhance the robustness of evaluation metrics, which in turn leads to an improvement in the performance of machine translation systems. Codes are available at \url{https://github.com/powerpuffpomelo/fairseq_mrt}.", }

Automatic metrics play a crucial role in machine translation. Despite the widespread use of n-gram-based metrics, there has been a recent surge in the development of pre-trained model-based metrics that focus on measuring sentence semantics. However, these neural metrics, while achieving higher correlations with human evaluations, are often considered to be black boxes with potential biases that are difficult to detect. In this study, we systematically analyze and compare various mainstream and cutting-edge automatic metrics from the perspective of their guidance for training machine translation systems. Through Minimum Risk Training (MRT), we find that certain metrics exhibit robustness defects, such as the presence of universal adversarial translations in BLEURT and BARTScore. In-depth analysis suggests two main causes of these robustness deficits: distribution biases in the training datasets, and the tendency of the metric paradigm. By incorporating token-level constraints, we enhance the robustness of evaluation metrics, which in turn leads to an improvement in the performance of machine translation systems. Codes are available at \url{https://github.com/powerpuffpomelo/fairseq_mrt}.

BLEURT Has Universal Translations: An Analysis of Automatic Metrics by Minimum Risk Training

acl-long.297

Oral

2307.03131

https://aclanthology.org/2023.acl-long.298.bib

https://aclanthology.org/2023.acl-long.298/

@inproceedings{pandey-etal-2023-cross, title = "Cross-modal Attention Congruence Regularization for Vision-Language Relation Alignment", author = "Pandey, Rohan and Shao, Rulin and Liang, Paul Pu and Salakhutdinov, Ruslan and Morency, Louis-Philippe", 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.298", doi = "10.18653/v1/2023.acl-long.298", pages = "5444--5455", abstract = "Despite recent progress towards scaling up multimodal vision-language models, these models are still known to struggle on compositional generalization benchmarks such as Winoground. We find that a critical component lacking from current vision-language models is relation-level alignment: the ability to match directional semantic relations in text (e.g., {`}mug in grass{'}) with spatial relationships in the image (e.g., the position of the mug relative to the grass). To tackle this problem, we show that relation alignment can be enforced by encouraging the language attention from {`}mug{'} to {`}grass{'} (capturing the semantic relation {`}in{'}) to match the visual attention from the mug to the grass (capturing the corresponding physical relation). Tokens and their corresponding objects are softly identified using a weighted mean of cross-modal attention. We prove that this notion of soft cross-modal equivalence is equivalent to enforcing congruence between vision and language attention matrices under a {`}change of basis{'} provided by the cross-modal attention matrix. Intuitively, our approach projects visual attention into the language attention space to calculate its divergence from the actual language attention, and vice versa. We apply our Cross-modal Attention Congruence Regularization (CACR) loss to fine-tune UNITER and improve its Winoground Group score by 5.75 points.", }

Despite recent progress towards scaling up multimodal vision-language models, these models are still known to struggle on compositional generalization benchmarks such as Winoground. We find that a critical component lacking from current vision-language models is relation-level alignment: the ability to match directional semantic relations in text (e.g., {`}mug in grass{'}) with spatial relationships in the image (e.g., the position of the mug relative to the grass). To tackle this problem, we show that relation alignment can be enforced by encouraging the language attention from {`}mug{'} to {`}grass{'} (capturing the semantic relation {`}in{'}) to match the visual attention from the mug to the grass (capturing the corresponding physical relation). Tokens and their corresponding objects are softly identified using a weighted mean of cross-modal attention. We prove that this notion of soft cross-modal equivalence is equivalent to enforcing congruence between vision and language attention matrices under a {`}change of basis{'} provided by the cross-modal attention matrix. Intuitively, our approach projects visual attention into the language attention space to calculate its divergence from the actual language attention, and vice versa. We apply our Cross-modal Attention Congruence Regularization (CACR) loss to fine-tune UNITER and improve its Winoground Group score by 5.75 points.

Cross-modal Attention Congruence Regularization for Vision-Language Relation Alignment

acl-long.298

Poster

2212.10549

https://aclanthology.org/2023.acl-long.299.bib

https://aclanthology.org/2023.acl-long.299/

@inproceedings{tang-etal-2023-enhancing-personalized, title = "Enhancing Personalized Dialogue Generation with Contrastive Latent Variables: Combining Sparse and Dense Persona", author = "Tang, Yihong and Wang, Bo and Fang, Miao and Zhao, Dongming and Huang, Kun and He, Ruifang and Hou, Yuexian", 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.299", doi = "10.18653/v1/2023.acl-long.299", pages = "5456--5468", abstract = "The personalized dialogue explores the consistent relationship between dialogue generation and personality. Existing personalized dialogue agents model persona profiles from three resources: sparse or dense persona descriptions and dialogue histories. However, sparse structured persona attributes are explicit but uninformative, dense persona texts contain rich persona descriptions with much noise, and dialogue history query is both noisy and uninformative for persona modeling. In this work, we combine the advantages of the three resources to obtain a richer and more accurate persona. We design a Contrastive Latent Variable-based model (CLV) that clusters the dense persona descriptions into sparse categories, which are combined with the history query to generate personalized responses. Experimental results on Chinese and English datasets demonstrate our model{'}s superiority in personalization.", }

The personalized dialogue explores the consistent relationship between dialogue generation and personality. Existing personalized dialogue agents model persona profiles from three resources: sparse or dense persona descriptions and dialogue histories. However, sparse structured persona attributes are explicit but uninformative, dense persona texts contain rich persona descriptions with much noise, and dialogue history query is both noisy and uninformative for persona modeling. In this work, we combine the advantages of the three resources to obtain a richer and more accurate persona. We design a Contrastive Latent Variable-based model (CLV) that clusters the dense persona descriptions into sparse categories, which are combined with the history query to generate personalized responses. Experimental results on Chinese and English datasets demonstrate our model{'}s superiority in personalization.

Enhancing Personalized Dialogue Generation with Contrastive Latent Variables: Combining Sparse and Dense Persona

acl-long.299

Poster

2305.11482

https://huggingface.co/papers/2305.11482