Datasets:

jamescalam

/

ai-arxiv-chunked

like 32

2210.06245

21

analysis of gender and bias throughout the machine learning lifecycle*. Computational Linguistics, 47(3):615–661. Maria De-Arteaga, Alexey Romanov, Hanna Wallach, Jennifer Chayes, Christian Borgs, Alexandra Chouldechova, Sahin Geyik, Krishnaram Kenthapadi, and Adam Tauman Kalai. 2019. Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting. In Proceedings of the Conference on Fairness, Accountability, and Transparency , pages 120–128, Atlanta GA USA. ACM. Sunipa Dev, Masoud Monajatipoor, Anaelia Ovalle, Arjun Subramonian, Jeff Phillips, and Kai-Wei Chang. 2021. Harms of gender exclusivity and challenges in non-binary representation in language technologies. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pages 1968–1994, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association

2210.06245

Back to the Future: On Potential Histories in NLP

Machine learning and NLP require the construction of datasets to train and fine-tune models. In this context, previous work has demonstrated the sensitivity of these data sets. For instance, potential societal biases in this data are likely to be encoded and to be amplified in the models we deploy. In this work, we draw from developments in the field of history and take a novel perspective on these problems: considering datasets and models through the lens of historical fiction surfaces their political nature, and affords re-configuring how we view the past, such that marginalized discourses are surfaced. Building on such insights, we argue that contemporary methods for machine learning are prejudiced towards dominant and hegemonic histories. Employing the example of neopronouns, we show that by surfacing marginalized histories within contemporary conditions, we can create models that better represent the lived realities of traditionally marginalized and excluded communities.

http://arxiv.org/pdf/2210.06245

cs.CL

20221012

20221012

2210.06245

22

deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , pages 4171–4186, Minneapolis, Minnesota. Association for Computational Linguistics. Lucas Dixon, John Li, Jeffrey Sorensen, Nithum Thain, and Lucy Vasserman. 2018. Measuring and Mitigating Unintended Bias in Text Classification. In Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society , pages 67–73, New Orleans LA USA. ACM. Jesse Dodge, Taylor Prewitt, Remi Tachet des Combes, Erika Odmark, Roy Schwartz, Emma Strubell, Alexandra Sasha Luccioni, Noah A. Smith, Nicole DeCario, and Will Buchanan. 2022. Measuring the Carbon Intensity of AI in Cloud Instances. In 2022 ACM Conference on Fairness, Accountability, and Transparency , pages 1877–1894, Seoul Republic of Korea. ACM. Michel Foucault. 2013. Archaeology of Knowledge , 0 edition. Routledge.

2210.06245

Back to the Future: On Potential Histories in NLP

Machine learning and NLP require the construction of datasets to train and fine-tune models. In this context, previous work has demonstrated the sensitivity of these data sets. For instance, potential societal biases in this data are likely to be encoded and to be amplified in the models we deploy. In this work, we draw from developments in the field of history and take a novel perspective on these problems: considering datasets and models through the lens of historical fiction surfaces their political nature, and affords re-configuring how we view the past, such that marginalized discourses are surfaced. Building on such insights, we argue that contemporary methods for machine learning are prejudiced towards dominant and hegemonic histories. Employing the example of neopronouns, we show that by surfacing marginalized histories within contemporary conditions, we can create models that better represent the lived realities of traditionally marginalized and excluded communities.

http://arxiv.org/pdf/2210.06245

cs.CL

20221012

20221012

2210.06245

23

Korea. ACM. Michel Foucault. 2013. Archaeology of Knowledge , 0 edition. Routledge. Lisa Gitelman, editor. 2013. "Raw data" is an oxymoron . Infrastructures series. The MIT Press, Cambridge, Massachusetts ; London, England. Ben Green and Salomé Viljoen. 2020. Algorithmic Realism: Expanding the Boundaries of Algorithmic Thought. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency , FAT* ’20, pages 19–31, New York, NY , USA. Association for Computing Machinery. Event-place: Barcelona, Spain. Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin de Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. 2019. Parameter-efficient transfer learning for NLP. InProceedings of the 36th International Conference on Machine Learning, ICML 2019, 9-15 June 2019, Long Beach, California, USA , volume 97 of Proceedings of Machine Learning Research , pages 2790–2799. PMLR.Chia-Chien Hung, Anne Lauscher, Simone Ponzetto,

2210.06245

Back to the Future: On Potential Histories in NLP

Machine learning and NLP require the construction of datasets to train and fine-tune models. In this context, previous work has demonstrated the sensitivity of these data sets. For instance, potential societal biases in this data are likely to be encoded and to be amplified in the models we deploy. In this work, we draw from developments in the field of history and take a novel perspective on these problems: considering datasets and models through the lens of historical fiction surfaces their political nature, and affords re-configuring how we view the past, such that marginalized discourses are surfaced. Building on such insights, we argue that contemporary methods for machine learning are prejudiced towards dominant and hegemonic histories. Employing the example of neopronouns, we show that by surfacing marginalized histories within contemporary conditions, we can create models that better represent the lived realities of traditionally marginalized and excluded communities.

http://arxiv.org/pdf/2210.06245

cs.CL

20221012

20221012

2210.06245

24

Proceedings of Machine Learning Research , pages 2790–2799. PMLR.Chia-Chien Hung, Anne Lauscher, Simone Ponzetto, and Goran Glavaš. 2022. DS-TOD: Efficient domain specialization for task-oriented dialog. In Findings of the Association for Computational Linguistics: ACL 2022 , pages 891–904, Dublin, Ireland. Association for Computational Linguistics. Diederik P. Kingma and Jimmy Ba. 2015. Adam: A method for stochastic optimization. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings . Anne Lauscher, Archie Crowley, and Dirk Hovy. 2022. Welcome to the modern world of pronouns: Identityinclusive natural language processing beyond gender. In Proceedings of the 29th International Conference on Computational Linguistics , pages 1221– 1232, Gyeongju, Republic of Korea. International Committee on Computational Linguistics. Anne Lauscher, Goran Glavaš, Simone Paolo Ponzetto, and Ivan Vuli ´c. 2020. A general framework for implicit and explicit debiasing of distributional word

2210.06245

Back to the Future: On Potential Histories in NLP

Machine learning and NLP require the construction of datasets to train and fine-tune models. In this context, previous work has demonstrated the sensitivity of these data sets. For instance, potential societal biases in this data are likely to be encoded and to be amplified in the models we deploy. In this work, we draw from developments in the field of history and take a novel perspective on these problems: considering datasets and models through the lens of historical fiction surfaces their political nature, and affords re-configuring how we view the past, such that marginalized discourses are surfaced. Building on such insights, we argue that contemporary methods for machine learning are prejudiced towards dominant and hegemonic histories. Employing the example of neopronouns, we show that by surfacing marginalized histories within contemporary conditions, we can create models that better represent the lived realities of traditionally marginalized and excluded communities.

http://arxiv.org/pdf/2210.06245

cs.CL

20221012

20221012

2210.06245

25

and Ivan Vuli ´c. 2020. A general framework for implicit and explicit debiasing of distributional word vector spaces. volume 34, pages 8131–8138. Association for the Advancement of Artificial Intelligence (AAAI). Anne Lauscher, Tobias Lueken, and Goran Glavaš. 2021. Sustainable modular debiasing of language models. In Findings of the Association for Computational Linguistics: EMNLP 2021 , pages 4782–4797, Punta Cana, Dominican Republic. Association for Computational Linguistics. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. RoBERTa: A Robustly Optimized BERT Pretraining Approach. arXiv:1907.11692 [cs] . ArXiv: 1907.11692. Sebastian McGaughey. 2020. Understanding neopronouns. The Gay & Lesbian Review Worldwide , 27(2):27–29. Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. 2016. Pointer sentinel mixture models.

2210.06245

Back to the Future: On Potential Histories in NLP

Machine learning and NLP require the construction of datasets to train and fine-tune models. In this context, previous work has demonstrated the sensitivity of these data sets. For instance, potential societal biases in this data are likely to be encoded and to be amplified in the models we deploy. In this work, we draw from developments in the field of history and take a novel perspective on these problems: considering datasets and models through the lens of historical fiction surfaces their political nature, and affords re-configuring how we view the past, such that marginalized discourses are surfaced. Building on such insights, we argue that contemporary methods for machine learning are prejudiced towards dominant and hegemonic histories. Employing the example of neopronouns, we show that by surfacing marginalized histories within contemporary conditions, we can create models that better represent the lived realities of traditionally marginalized and excluded communities.

http://arxiv.org/pdf/2210.06245

cs.CL

20221012

20221012

2210.06245

26

27(2):27–29. Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. 2016. Pointer sentinel mixture models. Ehm Hjorth Miltersen. 2016. Nounself pronouns: 3rd person personal pronouns as identity expression. Journal of Language Works-Sprogvidenskabeligt Studentertidsskrift , 1(1):37–62. Jonas Pfeiffer, Aishwarya Kamath, Andreas Rücklé, Kyunghyun Cho, and Iryna Gurevych. 2021. AdapterFusion: Non-destructive task composition for transfer learning. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume , pages 487–503, Online. Association for Computational Linguistics. Jonas Pfeiffer, Andreas Rücklé, Clifton Poth, Aishwarya Kamath, Ivan Vuli ´c, Sebastian Ruder, Kyunghyun Cho, and Iryna Gurevych. 2020a. AdapterHub: A framework for adapting transformers. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing:

2210.06245

Back to the Future: On Potential Histories in NLP

Machine learning and NLP require the construction of datasets to train and fine-tune models. In this context, previous work has demonstrated the sensitivity of these data sets. For instance, potential societal biases in this data are likely to be encoded and to be amplified in the models we deploy. In this work, we draw from developments in the field of history and take a novel perspective on these problems: considering datasets and models through the lens of historical fiction surfaces their political nature, and affords re-configuring how we view the past, such that marginalized discourses are surfaced. Building on such insights, we argue that contemporary methods for machine learning are prejudiced towards dominant and hegemonic histories. Employing the example of neopronouns, we show that by surfacing marginalized histories within contemporary conditions, we can create models that better represent the lived realities of traditionally marginalized and excluded communities.

http://arxiv.org/pdf/2210.06245

cs.CL

20221012

20221012

2210.06245

27

AdapterHub: A framework for adapting transformers. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations , pages 46–54, Online. Association for Computational Linguistics. Jonas Pfeiffer, Ivan Vuli ´c, Iryna Gurevych, and Sebastian Ruder. 2020b. MAD-X: An Adapter-Based Framework for Multi-Task Cross-Lingual Transfer. InProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages 7654–7673, Online. Association for Computational Linguistics. Rebecca Qian, Candace Ross, Jude Fernandes, Eric Smith, Douwe Kiela, and Adina Williams. 2022. Perturbation Augmentation for Fairer NLP. arXiv. ArXiv:2205.12586 [cs]. Anna Rogers. 2021. Changing the World by Changing the Data. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , pages 2182–2194, Online. Association for Computational Linguistics. Beatrice Santorini. 1990. Part-of-speech tagging guidelines for the penn treebank project.

2210.06245

Back to the Future: On Potential Histories in NLP

Machine learning and NLP require the construction of datasets to train and fine-tune models. In this context, previous work has demonstrated the sensitivity of these data sets. For instance, potential societal biases in this data are likely to be encoded and to be amplified in the models we deploy. In this work, we draw from developments in the field of history and take a novel perspective on these problems: considering datasets and models through the lens of historical fiction surfaces their political nature, and affords re-configuring how we view the past, such that marginalized discourses are surfaced. Building on such insights, we argue that contemporary methods for machine learning are prejudiced towards dominant and hegemonic histories. Employing the example of neopronouns, we show that by surfacing marginalized histories within contemporary conditions, we can create models that better represent the lived realities of traditionally marginalized and excluded communities.

http://arxiv.org/pdf/2210.06245

cs.CL

20221012

20221012

2210.06245

28

pages 2182–2194, Online. Association for Computational Linguistics. Beatrice Santorini. 1990. Part-of-speech tagging guidelines for the penn treebank project. Irene Solaiman and Christy Dennison. 2021. Process for Adapting Language Models to Society (PALMS) with Values-Targeted Datasets. arXiv:2106.10328 [cs]. ArXiv: 2106.10328. Emma Strubell, Ananya Ganesh, and Andrew McCallum. 2019. Energy and Policy Considerations for Deep Learning in NLP. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics , pages 3645–3650, Florence, Italy. Association for Computational Linguistics. Zeerak Talat, Smarika Lulz, Joachim Bingel, and Isabelle Augenstein. 2021. Disembodied Machine Learning: On the Illusion of Objectivity in NLP. ArXiv: 2101.11974. Ivan Vuli ´c, Simon Baker, Edoardo Maria Ponti, Ulla Petti, Ira Leviant, Kelly Wing, Olga Majewska, Eden Bar, Matt Malone, Thierry Poibeau, Roi Reichart,

2210.06245

Back to the Future: On Potential Histories in NLP

Machine learning and NLP require the construction of datasets to train and fine-tune models. In this context, previous work has demonstrated the sensitivity of these data sets. For instance, potential societal biases in this data are likely to be encoded and to be amplified in the models we deploy. In this work, we draw from developments in the field of history and take a novel perspective on these problems: considering datasets and models through the lens of historical fiction surfaces their political nature, and affords re-configuring how we view the past, such that marginalized discourses are surfaced. Building on such insights, we argue that contemporary methods for machine learning are prejudiced towards dominant and hegemonic histories. Employing the example of neopronouns, we show that by surfacing marginalized histories within contemporary conditions, we can create models that better represent the lived realities of traditionally marginalized and excluded communities.

http://arxiv.org/pdf/2210.06245

cs.CL

20221012

20221012

2210.06245

29

Petti, Ira Leviant, Kelly Wing, Olga Majewska, Eden Bar, Matt Malone, Thierry Poibeau, Roi Reichart, and Anna Korhonen. 2020. Multi-SimLex: A largescale evaluation of multilingual and crosslingual lexical semantic similarity. Computational Linguistics , 46(4):847–897. Hayden White. 2005. Introduction: Historical Fiction, Fictional History, and Historical Reality. Rethinking History , 9(2-3):147–157. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen,Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lhoest, and Alexander Rush. 2020. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations , pages 38–45, Online. Association for Computational Linguistics.

2210.06245

Back to the Future: On Potential Histories in NLP

Machine learning and NLP require the construction of datasets to train and fine-tune models. In this context, previous work has demonstrated the sensitivity of these data sets. For instance, potential societal biases in this data are likely to be encoded and to be amplified in the models we deploy. In this work, we draw from developments in the field of history and take a novel perspective on these problems: considering datasets and models through the lens of historical fiction surfaces their political nature, and affords re-configuring how we view the past, such that marginalized discourses are surfaced. Building on such insights, we argue that contemporary methods for machine learning are prejudiced towards dominant and hegemonic histories. Employing the example of neopronouns, we show that by surfacing marginalized histories within contemporary conditions, we can create models that better represent the lived realities of traditionally marginalized and excluded communities.

http://arxiv.org/pdf/2210.06245

cs.CL

20221012

20221012

2210.06245

30

System Demonstrations , pages 38–45, Online. Association for Computational Linguistics. Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang. 2017. Men also like shopping: Reducing gender bias amplification using corpus-level constraints. In Proceedings of the 2017 conference on empirical methods in natural language processing , pages 2979–2989, Copenhagen, Denmark. Association for Computational Linguistics. Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang. 2018. Gender Bias in Coreference Resolution: Evaluation and Debiasing Methods. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers) , pages 15–20, New Orleans, Louisiana. Association for Computational Linguistics.

2210.06245

Back to the Future: On Potential Histories in NLP

Machine learning and NLP require the construction of datasets to train and fine-tune models. In this context, previous work has demonstrated the sensitivity of these data sets. For instance, potential societal biases in this data are likely to be encoded and to be amplified in the models we deploy. In this work, we draw from developments in the field of history and take a novel perspective on these problems: considering datasets and models through the lens of historical fiction surfaces their political nature, and affords re-configuring how we view the past, such that marginalized discourses are surfaced. Building on such insights, we argue that contemporary methods for machine learning are prejudiced towards dominant and hegemonic histories. Employing the example of neopronouns, we show that by surfacing marginalized histories within contemporary conditions, we can create models that better represent the lived realities of traditionally marginalized and excluded communities.

http://arxiv.org/pdf/2210.06245

cs.CL

20221012

20221012

1809.02922

0

Transforming Question Answering Datasets Into Natural Language Inference Datasets Dorottya Demszky Department of Linguistics Stanford University ddemszky@stanford.eduKelvin Guu Department of Statistics Stanford University kguu@stanford.eduPercy Liang Department of Computer Science Stanford University pliang@cs.stanford.edu Abstract Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of largescale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets. 1 Introduction Natural language inference (NLI) is a task that incorporates much of what is necessary to understand language, such as the ability to leverage

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

1

1 Introduction Natural language inference (NLI) is a task that incorporates much of what is necessary to understand language, such as the ability to leverage world knowledge or perform lexico-syntactic reasoning. Given two sentences, a premise and a hypothesis, an NLI system must determine whether the hypothesis is implied by the premise. Numerous datasets have emerged to evaluate NLI systems (Marelli et al., 2014; Pavlick and Callison-Burch, 2016; Lai et al., 2017a). Two of the largest ones, SNLI (Bowman et al., 2015) and MultiNLI (Williams et al., 2017) are rich in various linguistic phenomena relevant to inference (e.g. quantification and negation), but they lack certain other phenomena, such as multi-sentence reasoning, which can be important for various downstream applications. In this paper, we propose to augment and diversify NLI datasets by automatically deriving largeEqual contribution. P: Taylor is a journalist […]. She was playing golf with Ron when her phone rang. It was Liz, her mother’s friend. […]Q: Who called Taylor?

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

2

P: Taylor is a journalist […]. She was playing golf with Ron when her phone rang. It was Liz, her mother’s friend. […]Q: Who called Taylor? D: A doctor called Taylor.D: Ron called Taylor.D: Liz called Taylor.A: LizA: RonA: a doctorentailmentcontradictionunknownQA2DFigure 1: We learn a mapping from a question-answer pair into a declarative sentence (QA2D), which allows us to convert question answering datasets into natural language inference datasets. scale NLI datasets from existing question answering(QA) datasets, which have recently become abundant and capture a wide range of reasoning phenomena.1 Inspired by the connection between QA and NLI noted by Dagan et al. (2006), we take the following approach: given a passage of text, a question about it ( Q:Who called Taylor? ) and an answer ( A:Liz), we perform sentence transformations to combine the question and answer into a declarative answer sentence ( D:Liz called Taylor). We then observe that the passage and the declarative sentence form a (premise, hypothesis) NLI pair. This is illustrated in Figure 1, and elaborated on in Section 2, where we also discuss how to generate negative (non-entailed) examples. This

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

3

NLI pair. This is illustrated in Figure 1, and elaborated on in Section 2, where we also discuss how to generate negative (non-entailed) examples. This approach is similar to the way SciTail (Khot et al., 2018) was constructed, except that our method is fully automated. Deriving NLI from QA has two key advantages. 1Data and code are available here: https://bit.ly/2OMm4vKarXiv:1809.02922v2 [cs.CL] 11 Sep 2018 Properties MovieQA NewsQA QAMR RACE SQuAD # wh questions 11k 100k 100k 20k 100k Domain Movie plots CNN Wikinews + Wikipedia English exams Wikipedia Multiple choice yes no no yes no Answer type free-form span span free-form span Passage type 1-3 sentences mult. par. sentence paragraph paragraph Avg question length 10.7 7.6 6.7 11 11.5 Avg word overlap 46% 73% 50% 62% 75% Table 1: Properties of the different QA datasets that we evaluated on. Together they cover a wide range of domains and evidence (passage) types, from sentences to multiple paragraphs and they add up to about 500k NLI examples

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

4

and evidence (passage) types, from sentences to multiple paragraphs and they add up to about 500k NLI examples including the multiple choice QA options. The average question length is counted by tokens and the average word overlap is measured by the percentage of tokens from the question that appear in the evidence. First, large-scale QA datasets are abundant. Second, existing QA datasets cover a wide range of reasoning strategies, which we can now import into the study of NLI. Both advantages likely stem from the fact that question answering and question formulation are organic tasks that people perform in daily life — making QA data easy to crowdsource (He et al., 2015), and easy to find in welldesigned pre-existing resources such as reading comprehension exams (Lai et al., 2017b). In Section 3, we describe the QA datasets we work with. Deriving a declarative sentence from a question-answer pair is the key step of our approach, a subtask that we call QA2D. We explore three different ways to perform QA2D: (i) a rule-based system (Section 4), (ii) crowdsourcing (Section 5) and (iii) a neural sequence model (Section 6). These three approaches build on each other: we

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

5

(Section 5) and (iii) a neural sequence model (Section 6). These three approaches build on each other: we demonstrate that a good rule-based system can accelerate and improve the quality of crowdsourcing (by providing workers with an initial draft) while not introducing any systematic bias. This enables us to collect a dataset of 100,000 (Q; A; D )triples, which we then use to train a neural QA2D model. Although this model is primarily trained on triples from SQuAD (Rajpurkar et al., 2016), we show that our model generalizes very well to other QA datasets spanning a variety of domains, such as Wikipedia, newswire and movie plots. Our automatically generated declaratives exactly match the human gold answer 45–57% of the time, with BLEU scores ranging between 73–83, depending on the dataset (Section 7). With our automated QA2D system in place, we apply it to five different QA datasets, creating over 500,000 NLI examples, which we make freely available. Given the diverse nature of the QA datasets we use, the resulting NLI dataset (QA-NLI) also exhibits a wide range of different inference phenomena, such as multi-sentence and

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

6

datasets we use, the resulting NLI dataset (QA-NLI) also exhibits a wide range of different inference phenomena, such as multi-sentence and meta-level reasoning and presupposition-based inference. We perform a thorough analysis of the resulting phenomena, quantifying this diversity both in terms of the type of reasoning and the contextual scope required to perform that reasoning. We also conduct other analyses that suggest that our approach can eliminate some of the annotation artifacts (Gururangan et al., 2018) present in SNLI and MultiNLI. 2 Approach We now formally define our framework for converting a QA example into an NLI example, including how to generate negative (non-entailed) NLI examples. A QA example contains a passage of text P, a question Qregarding the text and an answer span A, as illustrated in Figure 1. We perform sentence transformations (QA2D) to combine QandAinto a declarative answer sentence D. We then simply recognize that if Ais a correct answer, then (P; D ) is an entailed NLI pair. Alternatively, if Ais an incorrect answer or Q cannot be answered using the information in P, thenDisnotimplied by P, yielding a negative NLI pair. Incorrect answers are available in QA

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

7

cannot be answered using the information in P, thenDisnotimplied by P, yielding a negative NLI pair. Incorrect answers are available in QA datasets featuring multiple choice answers, such as MovieQA (Tapaswi et al., 2016), RACE (Lai et al., 2017b) and MCTest (Richardson et al., 2013). Unanswerable questions are available in SQuADRUn (Rajpurkar et al., 2018) and we expect the number of such datasets to grow with the advancement of QA research. Inference labels. In existing NLI datasets, examples are labeled with one of three relations: entailment, neutral/unknown or contradiction. When

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

8

advancement of QA research. Inference labels. In existing NLI datasets, examples are labeled with one of three relations: entailment, neutral/unknown or contradiction. When Q: Where does Jim go to buy groceries?A: Trader Joe’sIII. Jim goes where to buy groceries?I. Where does Jim goes to buy groceries?II. Where Jim goes to buy groceries?remove do-supportreverse wh-movementdelete question words & markplug in AIV. Jim goes Trader Joe’s to buy groceries.V. Jim goes to Trader Joe’s to buy groceries.insert prepositionFigure 2: An illustration of the syntactic transformations needed to perform QA2D. In this example, to perform step II one needs to know that where is complement of goand not that of buy, and to perform step V , one needs to chose the appropriate preposition to insert. performing automated QA2D, we can only make a two-way distinction between entailment and nonentailment.2 Weakly supervised QA datasets. In many QA datasets, the passage Pis a short paragraph (e.g. SQuAD (Rajpurkar et al., 2016)) or even a single sentence (e.g. QAMR (Michael et al., 2018)). This yields a short, simple premise in the resulting

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

9

This yields a short, simple premise in the resulting NLI example. However, some weakly supervised QA datasets such as NewsQA (Trischler et al., 2017), RACE and TriviaQA (Joshi et al., 2017) choose Pbe an entire document or even corpus of documents. In this case, the resulting NLI pair’s premise could be large, but is still valid. In Table 1, we describe the “passage type” for each QA dataset we work with. 3 Datasets Table 1 summarizes the properties of the five QA datasets we transform into NLI: MovieQA, NewsQA, QAMR and RACE and SQuAD. When choosing the datasets, we sought to maximize the structural and topical diversity of our data. The domains of these datasets include movie plots, newswire text, Wikipedia and English exams that cover a wide range of genres and topics. Thepassage types range from a sentence to mul2It is tempting to think that incorrect answers yield contradiction labels, while unanswerable questions yield neutral labels. Unfortunately, this is false. Figure 1 illustrates an example where an incorrect multiple choice answer does not

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

10

labels. Unfortunately, this is false. Figure 1 illustrates an example where an incorrect multiple choice answer does not yield a contradiction. As for unanswerable questions, one example would be: P: “The score was negative.”, Q: “What was the exact score?” (unanswerable), A: “10” (yields contradiction, not neutral).tiple paragraphs3and the answer type may be either a substring (span) within the passage or freeresponse text. The questions vary greatly in terms of their type and difficulty: questions in QAMR hinge on selecting the right arguments within a single sentence, while questions in RACE, written for middle- and high-schoolers, require holistic reasoning about the text (e.g. What is the main message of the passage? ). 4 QA2D: Rule-based At first glance, QA2D appears to be a highly structured task guided by clear rules — indeed, the reverse problem of converting declarative sentences into questions is often taught in grammar textbooks. However, this belies the many nuanced semantic decisions that are effortlessly made by native English speakers, yet challenging to codify. For example, non-native speakers find

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

11

nuanced semantic decisions that are effortlessly made by native English speakers, yet challenging to codify. For example, non-native speakers find it notoriously hard to prepend the right prepositions/articles before phrases, as there are no simple rules. To demonstrate these challenges, we develop a strong rule-based system (see Section 7 for results) to test how far we can go towards solving QA2D. The main steps of this system are illustrated in Figure 2. The success of the rule-based model hinges on part-of-speech tagging and parsing accuracy, given that we need to correctly identify the whword, the root word, any auxiliary or copula, as well as prepositions and particles that are dependents of the wh-word or the root. We used the state-of-the-art Stanford Graph-Based Neural Dependency Parser (Dozat et al., 2017) to POS tag and parse QandA. We found that about 10% of the mistakes made by our rule-based system are due to tagging/parsing errors.4 We encountered several semantic idiosyncrasies that proved difficult to account for by rules. For example, if the answer span is a bare named entity

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

12

We encountered several semantic idiosyncrasies that proved difficult to account for by rules. For example, if the answer span is a bare named entity (i.e. without an article) referring to an organization/institution, generally it is okay to leave it bare 3For MovieQA, we only used the plot summaries as the evidence, but one could easily use the full movie scripts or audiovisual data as well. 4The main errors made by the tagger/parser include tagging a verb as a noun, which is prevalent because in the presence of do-support, the inflections are removed from the main verb (e.g. When did the war end?). Another class of parsing errors is identification of the parent of a dangling preposition/particle (e.g. Which friend did Olga send a letter tolast week? ). Split Source # Ex. # Ann. Setup trainSQuAD 68986 1E (74%) S (26%) Other 41000 1 S devSQuAD7350 1 S 1000 3 S Other 4500 1 S testSQuAD7377 1 S 1000 3 S Other 41000 3 S Table 2: The composition of our collected data. # Ex. and # Ann. refer to the number of unique examples

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

13

1000 3 S Other 41000 3 S Table 2: The composition of our collected data. # Ex. and # Ann. refer to the number of unique examples and to the number of annotations (gold answers) per example, respectively. The last column lists the setup that was used for collecting the examples: post-editing (E) or from scratch (S). “Other” denotes the four other QA datasets besides SQuAD. The gray indicates the examples that we used for our evaluations in Section 7. (e.g. Sam works at WHO. ), but sometimes a definite article needs to be inserted (e.g. Sam works at the UN ). 5 QA2D: crowdsourced Even though our rule-based model is reasonably strong, it is far from perfect. We decided to build a supervised neural model, which required the collection of human-authored gold declarative sentences. We describe our data collection method (Section 5.1) and the distribution of our collected data across the five QA datasets (Section 5.2). 5.1 Data Collection We crowdsourced the QA2D task on Amazon Mechanical Turk using two different setups. In Setup S, Turkers were presented with QandA, then asked to write a full sentence answer Dfrom

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

14

S, Turkers were presented with QandA, then asked to write a full sentence answer Dfrom Scratch. In Setup E, instead of writing Dfrom scratch, Turkers were asked to Edit the output of our rule-based system (see Section 4) until it is a well-formed sentence. Turkers were not provided with the supporting passage Pin either setup because we wanted to prevent them from including information in Dthat is not supported by Q. Writing from scratch vs post-editing. There is a trade-off between the two setups: while Setup S minimizes bias towards the rule-based output, writing from scratch takes more time and leaves room for more typos than post-editing. Indeed, when comparing 100 random examples generated from each setup, we found that 91% of Setup S outputs were valid (grammatical and completewhile 97% of Setup E outputs were valid. However, since Setup E could potentially bias our data, we exclusively used Setup S for collecting all evaluation data. 5.2 Distribution of Source QA Datasets We decided to select one QA dataset among the five QA datasets to collect the majority of our data, so that we could test the ability of our neural model (Section 6) to generalize to other datasets.

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

15

five QA datasets to collect the majority of our data, so that we could test the ability of our neural model (Section 6) to generalize to other datasets. We chose SQuAD to be the main source of QA pairs because of its large size, high quality and syntactic diversity. We limited ourselves to its training set for the data collection, filtering out non-wh-questions, which left us with a total of 85,713 QA pairs.5In addition, we randomly sampled a smaller set of QA pairs from the four other datasets, most of which were used for evaluation. Table 2 summarizes the composition of our newly collected dataset of gold declarative answer sentences. For each of the dev and test sets, we collected three annotations for 1000 examples to account for the fact that there can be multiple possible correct QA2D transformations. The distribution of datasets within the three data splits are: train (95% SQuAD, 5% other four), dev (81% SQuAD, 19% other four) and test (20% for each five datasets). 6 QA2D: Neural Sequence Model In Section 4, we discussed some of the issues (mostly involving semantics) that our rule-based

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

16

five datasets). 6 QA2D: Neural Sequence Model In Section 4, we discussed some of the issues (mostly involving semantics) that our rule-based system cannot handle. To improve over this baseline, we develop a neural sequence generation model to perform QA2D. From the crowdsourcing described in the previous section, we have a dataset of ( Q,A,D) tuple. We use this to learn a model of p(DjQ; A ), implemented with an encoder-decoder architecture. The inputs QandAare each encoded using a bidirectional three-layer LSTM (Hochreiter and Schmidhuber, 1997) (the same encoder weights are used for both inputs). Dis then generated using a three-layer LSTM decoder equipped with one attention head (Bahdanau et al., 2015) for each input, and a copy mechanism based on Gu et al. (2016).6Word embeddings are initialized with 5The random split between train (80%), dev (10%) and test (10%) sets was made based on Wikipedia article titles corresponding to the QA pairs. 6The copy mechanism is similar to Gu et al. (2016), except that out-of-vocabulary copyable words are represented

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

17

corresponding to the QA pairs. 6The copy mechanism is similar to Gu et al. (2016), except that out-of-vocabulary copyable words are represented Figure 3: Figure (a) shows the results based on question type (brackets indicate their proportion in the data) and Figure (b) shows the results based on the length of Q+A. The error bars denote the 99% confidence interval for the true expected performance of each model (randomness comes from noise in Turker annotations and the random sampling of the evaluation set). Note: human and model scores should not be directly compared. Human agreement is maximum BLEU score when comparing each of 3 human annotations against the two others, while the models’ outputs are compared against the 3 human annotations. We include human results to quantify variation across human annotators. GloVe (Pennington et al., 2014). The model is then trained using a standard cross entropy loss minimized with Adam (Kingma and Ba, 2014). 7 QA2D: Results In this section, we assess the performance of our rule-based (R ULE-BASED ) and neural (N EURAL ) QA2D systems, using both automated metrics and human evaluation. The evaluations are conducted on the test set formed from all five QA datasets

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

18

QA2D systems, using both automated metrics and human evaluation. The evaluations are conducted on the test set formed from all five QA datasets (gray cells in Table 2) where each example includes three human annotations. 7.1 Quantitative Results For our quantitative evaluation, we employed two metrics: BLEU and string match (ignoring case and punctuation). We evaluated N EURAL both on the top 1 output and on the top 5 outputs (max over the beam). Rule-based vs neural. Overall, the performance of N EURAL is consistently stronger than RULE-BASED . From Table 4, we see that across datasets, N EURAL leads R ULE-BASED by an average of 2.6 BLEU points, and by 6.2% on exact match accuracy. N EURAL is also capable of producing a top-5 beam of outputs, and when we evaluate only the best of these 5 outputs, we observe an almost 30% improvement in scores. We find that predictions of N EURAL and RULE-BASED exactly match 40% of the time.

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

19

an almost 30% improvement in scores. We find that predictions of N EURAL and RULE-BASED exactly match 40% of the time. using absolute positional embeddings, rather than the encoder’s hidden state at the position of the copyable word.Table 3 includes examples when the two models’ outputs do not match. As we hypothesized, the N EURAL learned semantic patterns, such as preposition/article choice and the removal of redundant words from the answer span, that RULE-BASED was not able to handle. Results by dataset. Table 4 breaks down the models’ performance by dataset. The first thing to note is the domain-generality of the models — although BLEU scores are highest on SQuAD, they are only 1-3 points lower on other datasets. As for the exact match, we actually find that both models perform better on QAMR than on SQuAD. This discrepancy is likely due to the length penalty in BLEU, which is affected by shorter answer lengths in QAMR. RULE-BASED performs worst on RACE, MovieQA and SQuAD (6-9 points lower on exact match) in comparison to N EURAL due to the fact that the answers in these datasets often

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

20

MovieQA and SQuAD (6-9 points lower on exact match) in comparison to N EURAL due to the fact that the answers in these datasets often require semantically motivated modifications that RULE-BASED cannot handle. Results by question length. Figure 3 (b) shows the correlation between the combined length of the question-plus-answer and each model’s performance. R ULE-BASED ’s performance is more robust to length increase beyond 30 tokens than NEURAL , even if we consider N EURAL ’s top-5 performance. In contrast, N EURAL does better on inputs shorter than 20 tokens — which constitute the majority of the examples — than on longer ones, which is possibly due to the general Question Answer RULE-BASED NEURAL When was Madonna born? August 16, 1958 Madonna was born in August 16, 1958. Madonna was born on August 16, 1958. Who asks who to hit them outside of the bar?Tyler asks the Narrator to hit himTyler asks the Narrator to hit him asks who to hit them outside of the bar.Tyler asks the narrator to hit them outside of the bar. What surprising fact do the guys learn about Jones?That he has never killed anyoneThe guys learn that he has never

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

21

of the bar. What surprising fact do the guys learn about Jones?That he has never killed anyoneThe guys learn that he has never killed anyone about Jones.The guys learned about Jones that he has never killed anyone. Where is someone overlooked? American societySomeone is overlooked in American society.Someone overlooked is in American society. When did Johnson crash into the wall?halfway through the raceJohnson crashed into the wall halfway through the race.Johnson shot into the wall halfway through the race. What is an example of a corporate sponsor of a basketball team?Marathon OilAn example of a corporate sponsor of a basketball team is Marathon Oil.Marathon Oil is an example of a corporate sponsor of a basketball team. Where was the baby found?onboard a Carnival cruise ship,The baby was found in onboard a Carnival cruise ship.The baby was found at onboard a Carnival cruise ship. Table 3: A randomly picked sample of those outputs where N EURAL6=RULE-BASED . Green indicates good output(s) and red indicates bad ones. Dataset MovieQA NewsQA QAMR RACE SQuAD Model RM NM RM NM RM NM RM NM RM NM Top 1BLEU 81 84 82 83 82 85 80 83 82 86

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

22

Model RM NM RM NM RM NM RM NM RM NM Top 1BLEU 81 84 82 83 82 85 80 83 82 86 Match 38 44 49 49 54 57 30 44 45 53 Top 5BLEUN/A89N/A88N/A90N/A89N/A91 Match 52 58 67 55 62 Table 4: BLEU and exact match scores (out of a 100) when comparing the outputs of R ULE-BASED (RM) and N EURAL (NM) against the human gold. Figure 4: Human evaluation of the the human, R ULE-BASED and N EURAL outputs. tendency of N EURAL to output shorter sequences. Results by question type. In Figure 3 (a) we present the results broken down by question category, which we determined based on the type of wh word in the question. We can see that our models perform best on who questions, which is not surprising because there is usually no whmovement involved when transforming such questions into declaratives (i.e. the wh word simply needs to be replaced). In contrast, the overall performance of the models is the worst on which questions, which is most likely due to the fact that the majority of such questions require a decision about which words from the wh phrase to include inDand in what position.

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

23

questions, which is most likely due to the fact that the majority of such questions require a decision about which words from the wh phrase to include inDand in what position. The only question type for which we see a significant difference between performance of RULE-BASED and that of the neural one is how questions (note that how many , for example, is considered to be a how question). This is because RULE-BASED — in contrast to N EURAL — does not copy words from the wh phrase into D, which mainly affects howquestions negatively, given that how many tends to be followed by at least one noun (e.g. how many people ).7.2 Human Evaluation We crowdsourced the evaluation of R ULE-BASED and N EURAL on a sample of 100 QA examples for each of the five datasets. For reference, we also treated human outputs as a third system (H UMAN ). In each Turk task, a Turker is presented with three outputs (one from each system, in a randomly shuffled order) and is asked to rate them (given a question and an answer span) with respect to three criteria: grammaticality, naturalness, and completeness.7The results are shown in Figure 4.

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

24

question and an answer span) with respect to three criteria: grammaticality, naturalness, and completeness.7The results are shown in Figure 4. For grammar, the models’ outputs are lower than H UMAN (4.6), at 3.8 for R ULE-BASED and 7Description of ratings: grammaticality: 1 – Extremely poor, 2 – Poor, 3 – OK but has some issue(s), 4 – Good but slightly unnatural, 5 – Good naturalness: 1 – Extremely unnatural, 2 – Unnatural, 3 – OK in some contexts, 4 – Natural, but could be more so, 5 – Very natural completeness: 1 – Lacks many important words from the question or the answer, 2 – Lacks a few important words from the question or the answer, 3 – The sentence is missing one or two words that would add more information, but they aren’t necessary, 4 – The sentence is missing one or two words but it still conveys the same meaning without them, 5 – The sentence is maximally complete in terms of words (regardless of grammaticality) Reasoning Description Example Source ArgumentAffects only a single argument of a predicate in T.T:Caitlin de Wit : I ran a little bit, and I rode horses.

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

25

ArgumentAffects only a single argument of a predicate in T.T:Caitlin de Wit : I ran a little bit, and I rode horses. H: Caitlin is de Wit’s first name.QAMR SentenceAffects one or more predicates within a single sentence in T.T: Michael Jackson will perform 10 concerts in London in July in what he described Thursday as a ”final curtain call. ” [...] H: Michael Jackson has announced 10 concerts.NewsQA MultisentenceAffects multiple sentences inT.T: [...] invented by the British scientist William Sturgeon [...] Following Sturgeon’s work, [...] motor [...] built by [...] Thomas Davenport [...] The motors ran at up to 600 revolutions per minute [...] H: Sturgeon and Davenport s motors ran at 600 revolutions per minute .SQuAD QuantitiesCounting or performing other numerical operations; understanding relations between quantities.T: It provided more than $20 billion in direct financial. H: It only yielded $100,000 in direct financial.MultiNLI Naive physicsSpatiotemporal/physical reasoning that requires a mental simulation of the event beyond understanding

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

26

Naive physicsSpatiotemporal/physical reasoning that requires a mental simulation of the event beyond understanding the meaning of the words.T: David J. Lavau, 67, of Lake Hughes, California, was found in a ravine a week after losing control of his car on a rural road and plunging 500 feet down an embankment into heavy brush [...] H: Lavau’s car came to a rest 500 feet down an embankment.NewsQA AttributesReasoning about attributes and affordances of entities.T: The situation in Switzerland [...]. The Swiss German dialects arethe default everyday language in virtually every situation [...] H: Swiss German is the dialect spoken in Switzerland.SQuAD PsychologyMaking inferences involving people’s mental states and attitudes and the way they express them.T: Dear Jorge, [...] My family are now in Sacramento, California. [...] Before I knew it , there was hot water shooting up about 60 feet into the air. [...] I’d love to learn more about this geyser and other geysers [...] Your friend,Bennetto H: Bennetto’s letter expressed excitement .RACE MetaReasoning about the genre, text structure and author.T: A man and his young son struggle to survive after an

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

27

MetaReasoning about the genre, text structure and author.T: A man and his young son struggle to survive after an unspecified cataclysm has killed most plant and animal life. [...] H: We meet a man and his young son at the beginning of the film .MovieQA OtherIncorporating any other world knowledge.T: When asked a life lesson he had to learn the hard way, the billionaire said staying up too late is a habit he is still trying to break. “Don’t stay up too late[...] H: Bill Gates gave the advice to avoid staying up too late.RACE Table 5: Examples illustrating the different types of reasoning required to determine whether the text Tentails the hypothesis H. 3.9 for N EURAL . A score of 4 represents the ratingGood but slightly unnatural . The same pertains to the naturalness scores as well. In terms of completeness, the rule-based and N EURAL both do relatively well (.2-.3 points lower than the human score of 4.8), which is well above the threshold for retaining the correct meaning, given that a rating of 4 requires there to be no semantic consequences of incompleteness. 8 Analysis of NLI Datasets In this section, we analyze the various phenomena

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

28

rating of 4 requires there to be no semantic consequences of incompleteness. 8 Analysis of NLI Datasets In this section, we analyze the various phenomena of the NLI datasets we generated (Section 8.1), validate our assumptions about how an answer’s correct/incorrect status determines the resulting inference label (Section 8.2) and compare our datasets to others in terms of annotation artifacts (Section 8.3). 8.1 Inference Phenomena We manually annotated 100 examples for the scope andtype of reasoning required to make acorrect inference classification. The categories and their descriptions, paired with examples are illustrated in Table 5. Looking at the counts in Table 6, we can notice that MovieQA, NewsQA and SQuAD are similar. The majority of examples in these datasets require sentence-level reasoning, and the rest are split between multi-sentence and argument-level reasoning at a roughly 5:3 ratio. While the counts for the types of reasoning are also very close among these datasets, the slight differences can be explained based on genre. For example, plot summaries often focus on human motivations/emotions ( psych ). MultiNLI is closest to these datasets in terms of phenomena counts

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

29

example, plot summaries often focus on human motivations/emotions ( psych ). MultiNLI is closest to these datasets in terms of phenomena counts as well, except that it involves hardly any multisentence reasoning and less world knowledge than the other three. QAMR is unique in that, by construction, two thirds of the examples only involve argumentlevel reasoning and none of them involve multisentence reasoning. Given that inference pairs in QAMR often involve turning a noun phrase into MovieQA NewsQA QAMR RACE SQuAD MultiNLIscopeargument 19 13 62 1 15 26 sentence 56 53 38 14 58 72 multi-sentence 25 34 0 85 27 2type of reasoningquantities 1 1 1 1 2 4 naive physics 4 7 0 5 4 3 psych 9 0 1 48 1 5 meta 2 5 0 12 0 1 attributes 41 41 12 58 40 29 other world kn. 34 15 8 68 14 5 Table 6: Counts for different scope and reasoning types in the five converted QA datasets in comparison to MultiNLI. We manually annotated 100 examples per dataset. While the scope categories are mutually exclusive, the reasoning types are not. Figure 5: The distribution of inference ratings for NLI examples based on incorrect multiple choice option or correct multiple choice option.

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

30

Figure 5: The distribution of inference ratings for NLI examples based on incorrect multiple choice option or correct multiple choice option. a predicate, this dataset provides us with a lot of inference pairs that stem from presuppositions — i.e. entailments that still hold even if the premise is negated (e.g. Taylor Brown was playing golf outside. andTaylor Brown was notplaying golf outside. both presuppose Taylor’s last name is Brown. ). RACE, in sharp contrast to QAMR, mostly includes entailments that require multi-sentence reasoning. In addition, inference pairs in RACE make extensive use of world knowledge, metalevel reasoning (e.g., about the genre, author’s intentions, etc.), and reasoning about human psychology (e.g., a character’s reaction to an event). 8.2 Inference Labels In Section 2, we made the assumption that inference pairs generated from incorrect answers will be non-entailments. To verify this hypothesis, we crowdsource the inference labeling of 2000 inference pairs based on MovieQA, half of which are generated from correct answer options and half from incorrect ones.8In the task, each Turker was provided with five randomly selected premise-hypothesis pairs (T; H )and was asked to

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

31

half from incorrect ones.8In the task, each Turker was provided with five randomly selected premise-hypothesis pairs (T; H )and was asked to rate how likely His true given T. Figure 5 shows the distribution of inference ratings with the examples separated based on incorrect/correct answer options as their source. The human ratings show a strong correlation between answer type and inference score — as for correct ones, more than 90% of the examples are more likely to be true than false and as for incorrect 8For the 1000 examples based on correct answers, we used the test set we already collected. We collected the 1000 examples for incorrect answers using the same QA pairs as for the correct answers. The incorrect answer, in the case of each example, was randomly chosen among the incorrect options. Figure 6: The distribution of the length of NLI examples, generated from MovieQA. Unlike in SNLI and MultiNLI, we found little correlation between the inference label and sentence lengths. ones, about 80% are not likely to be true. These findings also support a non-binary notion of entailment, as we can see that about half of the examples in both categories do not fit into the strict entailment/contradiction dichotomy. 8.3 Annotation Artifacts

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

32

entailment/contradiction dichotomy. 8.3 Annotation Artifacts We replicate some of the statistical analyses that Gururangan et al. (2018) performed on SNLI and MultiNLI to see whether our datasets contain artifacts similar to SNLI and MultiNLI. We perform the analyses on the same dataset we used in Section 8.2, based on MovieQA. We ranked the words with the highest PMI( word, class ) values and found that in our case the non-entailments in our dataset no longer have negation words and the entailments no longer have positive or non-specific words such as those found in SNLI and MultiNLI (Table 7). We also looked at the distribution of hypotheses lengths, separated by label (Figure 6) and found little/no correlation between the length of an example and its label. entailment neutral contradiction MQAfind all take police years york son car shoot can SNLIoutdoors tall nobody least first sleeping instrument competition no outside sad tv animal favorite cat MNLIsome also never yes because no something popular nothing sometimes many any various most none Table 7: Top 5 words ordered based on their PMI

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

33

animal favorite cat MNLIsome also never yes because no something popular nothing sometimes many any various most none Table 7: Top 5 words ordered based on their PMI (word, class ) with the percentage of examples they occur in in a given class , based on the entailments generated from MovieQA (MQA). The statistics for SNLI and MultiNLI (MNLI) are copied from Gururangan et al. (2018). 9 Related Work & Discussion NLI datasets. NLI has long served as a testbed for natural language understanding (Dagan et al., 2006). More recently, the emergence of largerscale datasets (Bowman et al., 2015; Williams et al., 2017) have also enabled researchers to leverage NLI resources as a rich source of training data to achieve transfer learning gains on other tasks (Conneau et al., 2017). Our datasets are complementary to previous resources, inheriting a rich set of phenomena found in many QA datasets (e.g. high-level reasoning about texts). QA to NLI. Although White et al. (2017) and Poliak et al. (2018) have explored recasting datasets created for various semantic classification tasks (e.g. semantic role labeling and named entity recognition) into NLI datasets, we are the first

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

34

tasks (e.g. semantic role labeling and named entity recognition) into NLI datasets, we are the first to perform such an automated conversion on QA datasets. However, we are not the first to observe the connection between QA and NLI. In fact, the seminal work of Dagan et al. (2006) employed this connection to construct a portion of their dataset and so have the creators of SciTail (Khot et al., 2018), but performed the QA2D step with human experts, rather than an automated system. Text transformation tasks. By “reformatting” QA to NLI, we obtain a more generic representation of inferences: declarative sentences are transformed into other declarative sentences. This is the same type signature as for sentence simplifica-tion (Chandrasekar et al., 1996), paraphrase (Lin and Pantel, 2001; Bannard and Callison-Burch, 2005) and summarization (Jones, 1993), highlighting the close connection between these tasks. Importantly, declarative sentences are closed under this set of operations, allowing them to be chained together to perform more complex inferences (Kolesnyk et al., 2016). Another related task is question generation (Rus

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

35

chained together to perform more complex inferences (Kolesnyk et al., 2016). Another related task is question generation (Rus and Arthur, 2009), which could be considered the reverse of QA2D, although the focus is on selecting interesting questions, rather than robust sentence transformation. Neural sequence generation. Our QA2D system could be implemented by any general-purpose sequence generation model. With rapid progress on better generation architectures (Gehring et al., 2017; Vaswani et al., 2017), we believe it should be possible to further increase the data efficiency and performance, especially by leveraging models that incorporate syntactic structure (Chen et al., 2017; Eriguchi et al., 2017) or a more transducerlike structure (Graves, 2012; Yu et al., 2016). Future systems. Finally, we hope that by increasing the scale of NLI training resources, we can enable the development of a large variety of new systems such as generative NLI models which can take a premise and generate relevant hypotheses (Kolesnyk et al., 2016), sentence decomposition models which can break a sentence into multiple entailed parts, and sentence synthesis models which can stitch multiple pieces back into an entailed whole. Acknowledgments We would like to thank Chris Potts and members of the Stanford NLP group for their valuable feedback.

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

36

which can stitch multiple pieces back into an entailed whole. Acknowledgments We would like to thank Chris Potts and members of the Stanford NLP group for their valuable feedback. References D. Bahdanau, K. Cho, and Y . Bengio. 2015. Neural machine translation by jointly learning to align and translate. In International Conference on Learning Representations (ICLR) . C. Bannard and C. Callison-Burch. 2005. Paraphrasing with bilingual parallel corpora. In Association for Computational Linguistics (ACL) , pages 597–604. S. Bowman, G. Angeli, C. Potts, and C. D. Manning. 2015. A large annotated corpus for learning natural language inference. In Empirical Methods in Natural Language Processing (EMNLP) . R. Chandrasekar, C. Doran, and B. Srinivas. 1996. Motivations and methods for text simplification. In Proceedings of the 16th conference on Computational linguistics-Volume 2 , pages 1041–1044. H. Chen, S. Huang, D. Chiang, and J. Chen. 2017. Improved neural machine translation with a syntax-aware encoder and decoder. arXiv preprint

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

37

2017. Improved neural machine translation with a syntax-aware encoder and decoder. arXiv preprint arXiv:1707.05436 . A. Conneau, D. Kiela, H. Schwenk, L. Barrault, and A. Bordes. 2017. Supervised learning of universal sentence representations from natural language inference data. arXiv preprint arXiv:1705.02364 . I. Dagan, O. Glickman, and B. Magnini. 2006. The PASCAL recognising textual entailment challenge . Machine learning challenges. evaluating predictive uncertainty, visual object classification, and recognising tectual entailment. T. Dozat, P. Qi, and C. D. Manning. 2017. Stanford’s graph-based neural dependency parser at the conll 2017 shared task. In Computational Natural Language Learning (CoNLL) , pages 20–30. A. Eriguchi, Y . Tsuruoka, and K. Cho. 2017. Learning to parse and translate improves neural machine translation. arXiv preprint arXiv:1702.03525 . J. Gehring, M. Auli, D. Grangier, D. Yarats, and

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

38

translation. arXiv preprint arXiv:1702.03525 . J. Gehring, M. Auli, D. Grangier, D. Yarats, and Y . N. Dauphin. 2017. Convolutional sequence to sequence learning. arXiv preprint arXiv:1705.03122 . A. Graves. 2012. Sequence transduction with recurrent neural networks. arXiv preprint arXiv:1211.3711 . J. Gu, Z. Lu, H. Li, and V . O. Li. 2016. Incorporating copying mechanism in sequence-to-sequence learning. In Association for Computational Linguistics (ACL) . S. Gururangan, S. Swayamdipta, O. Levy, R. Schwartz, S. R. Bowman, and N. A. Smith. 2018. Annotation artifacts in natural language inference data. arXiv preprint arXiv:1803.02324 . L. He, M. Lewis, and L. Zettlemoyer. 2015. Questionanswer driven semantic role labeling: Using natural language to annotate natural language. In Empirical Methods in Natural Language Processing (EMNLP) .

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

39

(EMNLP) . S. Hochreiter and J. Schmidhuber. 1997. Long shortterm memory. Neural Computation , 9(8):1735– 1780. K. S. Jones. 1993. What might be in a summary? Information retrieval , 93:9–26. M. Joshi, E. Choi, D. Weld, and L. Zettlemoyer. 2017. TriviaQA: A large scale distantly supervised challenge dataset for reading comprehension. In Association for Computational Linguistics (ACL) .Tushar Khot, Ashish Sabharwal, and Peter Clark. 2018. Scitail: A textual entailment dataset from science question answering. In Proceedings of AAAI . D. Kingma and J. Ba. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 . V . Kolesnyk, T. Rockt ¨aschel, and S. Riedel. 2016. Generating natural language inference chains. arXiv preprint arXiv:1606.01404 . A. Lai, Y . Bisk, and J. Hockenmaier. 2017a. Natural language inference from multiple premises. arXiv preprint arXiv:1710.02925 .

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

40

language inference from multiple premises. arXiv preprint arXiv:1710.02925 . G. Lai, Q. Xie, H. Liu, Y . Yang, and E. Hovy. 2017b. Race: Large-scale reading comprehension dataset from examinations. arXiv preprint arXiv:1704.04683 . D. Lin and P. Pantel. 2001. Discovery of inference rules for question-answering. Natural Language Engineering , 7:343–360. M. Marelli, S. Menini, M. Baroni, L. Bentivogli, R. bernardi, and R. Zamparelli. 2014. A SICK cure for the evaluation of compositional distributional semantic models. In Language Resources and Evaluation Conference (LREC) . J. Michael, G. Stanovsky, L. He, I. Dagan, and L. Zettlemoyer. 2018. Crowdsourcing question– answer meaning representations. In North American Association for Computational Linguistics (NAACL) . E. Pavlick and C. Callison-Burch. 2016. Most” babies” are” little” and most” problems” are” huge”:

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

41

Compositional entailment in adjective-nouns. In Association for Computational Linguistics (ACL) , volume 1, pages 2164–2173. J. Pennington, R. Socher, and C. D. Manning. 2014. Glove: Global vectors for word representation. In Empirical Methods in Natural Language Processing (EMNLP) . Adam Poliak, Aparajita Haldar, Rachel Rudinger, J Edward Hu, Ellie Pavlick, Aaron Steven White, and Benjamin Van Durme. 2018. Towards a unified natural language inference framework to evaluate sentence representations. arXiv preprint arXiv:1804.08207 . P. Rajpurkar, R. Jia, and P. Liang. 2018. Know what you don’t know: Unanswerable questions for SQuAD. In Association for Computational Linguistics (ACL) . P. Rajpurkar, J. Zhang, K. Lopyrev, and P. Liang. 2016. SQuAD: 100,000+ questions for machine comprehension of text. In Empirical Methods in Natural Language Processing (EMNLP) . M. Richardson, C. J. Burges, and E. Renshaw. 2013. Mctest: A challenge dataset for the open-domain

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

42

M. Richardson, C. J. Burges, and E. Renshaw. 2013. Mctest: A challenge dataset for the open-domain machine comprehension of text. In Empirical Methods in Natural Language Processing (EMNLP) , pages 193–203. V . Rus and C. G. Arthur. 2009. The question generation shared task and evaluation challenge. In The University of Memphis. National Science Foundation . M. Tapaswi, Y . Zhu, R. Stiefelhagen, A. Torralba, R. Urtasun, and S. Fidler. 2016. Movieqa: Understanding stories in movies through questionanswering. In Computer Vision and Pattern Recognition (CVPR) , pages 4631–4640. A. Trischler, T. Wang, X. Yuan, J. Harris, A. Sordoni, P. Bachman, and K. Suleman. 2017. NewsQA: A machine comprehension dataset. In Workshop on Representation Learning for NLP . A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin. 2017. Attention is all you need. arXiv preprint

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1809.02922

43

L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin. 2017. Attention is all you need. arXiv preprint arXiv:1706.03762 . Aaron Steven White, Pushpendre Rastogi, Kevin Duh, and Benjamin Van Durme. 2017. Inference is everything: Recasting semantic resources into a unified evaluation framework. In Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , volume 1, pages 996–1005. A. Williams, N. Nangia, and S. R. Bowman. 2017. A broad-coverage challenge corpus for sentence understanding through inference. arXiv preprint arXiv:1704.05426 . L. Yu, J. Buys, and P. Blunsom. 2016. Online segment to segment neural transduction. arXiv preprint arXiv:1609.08194 .

1809.02922

Transforming Question Answering Datasets Into Natural Language Inference Datasets

Existing datasets for natural language inference (NLI) have propelled research on language understanding. We propose a new method for automatically deriving NLI datasets from the growing abundance of large-scale question answering datasets. Our approach hinges on learning a sentence transformation model which converts question-answer pairs into their declarative forms. Despite being primarily trained on a single QA dataset, we show that it can be successfully applied to a variety of other QA resources. Using this system, we automatically derive a new freely available dataset of over 500k NLI examples (QA-NLI), and show that it exhibits a wide range of inference phenomena rarely seen in previous NLI datasets.

http://arxiv.org/pdf/1809.02922

11 pages, 6 figures

cs.CL

20180909

20180911

1903.02134

0

Negative Training for Neural Dialogue Response Generation Tianxing He and James Glass Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology ftianxing,glass g@csail.mit.edu Abstract Although deep learning models have brought tremendous advancements to the field of opendomain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named “Negative Training” to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity. 1 Introduction End-to-end dialogue response generation can be formulated as a sequence-to-sequence (seq2seq) task: given a dialogue context, the model is asked to generate a high-quality response. In recent years, deep learning models, especially seq2seq language generation models (Sutskever et al., 2014; Cho

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

1

to generate a high-quality response. In recent years, deep learning models, especially seq2seq language generation models (Sutskever et al., 2014; Cho et al., 2014), have brought significant progress to the field of dialogue response generation. However, recent research has revealed undesirable behaviors of seq2seq models that are side effects of standard maximum likelihood estimation (MLE) training, such as the generic (boring) response problem (Li et al., 2016), vulnerability to adversarial attacks (Cheng et al., 2018; Belinkov and Bisk, 2017), and the malicious (egregious) response problem (He and Glass, 2019). In this work, we propose and explore the negative training framework to correct unwanted behaviors of a dialogue response generator. During negative training, we first find or identify input-output pairs for a trained seq2seq model that exhibit someundesirable generation behavior, treat them as “bad examples,” and use them to feed negative training signals to the model. Correspondingly, we regard the training data as “good examples” and standard MLE training as “positive training”. The idea of negative training is inspired from the way parents might teach their children to use language by incorporating both positive and negative

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

2

MLE training as “positive training”. The idea of negative training is inspired from the way parents might teach their children to use language by incorporating both positive and negative training signals. For example, when teaching children how to use “love” and “hate”, in addition to using positive examples like “ I love apples but I hate bananas ”, they might also point out that saying “ I hate you ” to someone is considered impolite. In this work, negative training is used to address the malicious response problem and the frequent response problem (to be described in Section 3.2 and 3.3) in open-domain dialogue response generation. In our experiments, we show that negative training can significantly reduce the hit rate for malicious responses, or discourage frequent responses and greatly improve response diversity. 2 Model Formulation In this work we adopt recurrent neural network (RNN) based encoder-decoder seq2seq models (Sutskever et al., 2014; Cho et al., 2014; Mikolov et al., 2010), which are widely used in NLP applications like dialogue response generation (Li et al., 2016), machine translation (Luong et al., 2015),

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

3

et al., 2010), which are widely used in NLP applications like dialogue response generation (Li et al., 2016), machine translation (Luong et al., 2015), etc. We use x=fx1;x2;:::;xngto denote onehot vector representations of the input sequence, which serves as context or history information (e.g. the previous utterance), y=fy1;y2;:::;ymg1to denote scalar indices of the corresponding reference target sequence, and Vas the vocabulary. We useto represent the parameters for the seq2seq 1The last word ymis a<EOS> token which indicates the end of a sentence.arXiv:1903.02134v5 [cs.CL] 18 Aug 2020 model, andP(yjx)as the model’s generative distribution. On the encoder side, every xtwill be first mapped into its corresponding word embedding xemb t. Thenfxemb tgare input to a long-short term memory (LSTM) (Hochreiter and Schmidhuber, 1997) RNN to get a sequence of latent representationsfhenc tg2. For the decoder, at time t, similarlyytis first mapped to yemb

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

4

tg2. For the decoder, at time t, similarlyytis first mapped to yemb t. Then a context vector ct, which is supposed to capture useful latent information of the input sequence, needs to be constructed. We adopt the “attention” mechanism for context vector construction: first an attention mask vector at (which is a distribution) on the input sequence is calculated to decide which part to focus on, then the mask is applied to the latent vectors to construct ct:ct=Pn i=1at(i)henc i. We use the formulation of the “general” type of global attention, described in (Luong et al., 2015), to calculate the mask. During baseline training, standard MLE training with stochastic gradient descent (SGD) is used to minimize the negative log-likelihood (NLL) of the reference target sentence given the input sentence in the data: LMLE(Pdata;) =E(x;y)Pdata(logP(yjx)) =E(x;y)Pdata(mX t=1logP(ytjy<t;x)) (1)

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

5

=E(x;y)Pdata(mX t=1logP(ytjy<t;x)) (1) where y<trefers tofy0;y1;:::;yt1g, in whichy0 is set to a begin-of-sentence token <BOS> . We consider two popular ways of decoding (generating) a sentence given an input: greedy decoding and sampling. In practice for dialogue response generation, greedy decoding will provide stable and reproducible outputs, but is severely affected by the generic response problem. Sampling will provide more diverse but less predictable responses, and thus give rise to the malicious response problem. 3 The Negative Training Framework 3.1 Overview The negative training framework3is a two-stage process. Given a trained model, we put it under a 2Herehrefers to the output layer of LSTM, not the cell memory layer. 3Our code is available at https://github.com/ cloudygoose/negativetraining_acl2020“debugging” environment Ptestwhich provides test input samples4, get the model’s decoded samples and decide (using well-defined criteria) whether each input-output pair exhibits some undesirable

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

6

input samples4, get the model’s decoded samples and decide (using well-defined criteria) whether each input-output pair exhibits some undesirable behavior. Then, these “bad” pairs are used to provide negative training signals. Negative training can be derived from Empirical Bayes Risk Minimization (Och, 2003). Specifically, the overall objective is to minimize the expected risk that the model exhibits undesirable decoding behavior: LNEG(Ptest;) =ExPtestEyP(yjx)c(x;y) (2) wherec(x;y)refers to the binary criteria that will be1if(x;y)exhibits undesirable behavior, and 0 otherwise. Then, we take the derivative of LNEGw.r.t. to , using the log derivative trick (widely used in Reinforcement Learning (Sutton and Barto, 1998)): rLNEG(Ptest;) = ExPtestEyP(yjx)c(x;y)
rlogP(yjx) (3) Compared toLMLE in eq. (1), which maximizes

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

7

(3) Compared toLMLE in eq. (1), which maximizes the log-likelihood of training data samples, LNEG minimizes the log-likelihood of undesirable model samples. This is the reason why we call it “Negative Training”. In our preliminary experiments, we find that negative training needs to be augmented with the standard MLE objective LMLE, encouraging the model to retain its original performance: LNEG+POS =LNEG+POSLMLE (4) In our experiments, we find POScan be simply set to 0.1 to work well. In the next two sections, we discuss how the general negative training framework is tailored for the malicious response problem and frequent response problem, respectively. 3.2 Negative Training for the Malicious Response Problem For the malicious response problem, we follow the methodology proposed by (He and Glass, 2019). 4Note that here “test” does not refer to the test data. First a list of malicious target sentences are created, then the gibbs-enum algorithm5is called to find “trigger input” that will cause the model to assign large probability to the target sequence. The following “hit types” are defined:

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

8

assign large probability to the target sequence. The following “hit types” are defined: o-greedy-hit: A trigger input sequence is found such that the model generates the target sentence from greedy decoding. o-sample-min/avg-hit: A trigger input sequence is found such that the model generates the target sentence with an minimum/average word log-probability larger than a given thresholdTout. io-sample-min/avg-hit: In addition to the definition of o-sample-min/avg-hit , we also require that the average log-likelihood of the trigger input sequence, measured by a LM, is larger than a threshold Tin. This enforces the trigger input to be more likely to be input by real-world users. Toutis set to the trained seq2seq model’s average word log-likelihood on the test data, and Tinis set to be a reasonable LM’s6average word loglikelihood on the test set. The intuition is that the model should not assign larger probabilities to the malicious sentences than the reference sentences in the test set. Note that these hit types act as criteria c(x;y), indicating whether a target sentence is hit

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

9

malicious sentences than the reference sentences in the test set. Note that these hit types act as criteria c(x;y), indicating whether a target sentence is hit by a trigger input. As shown in (He and Glass, 2019), a typical seq2seq model trained by MLE has around a 10% hit rate for malicious targets w.r.t. samplemin/avg-hit , across data-sets. However, very few malicious targets are hit w.r.t. greedy-hit , so in this work, we focus on the malicious response problem for sampling during decoding. In Table 1 we show pairs of trigger inputs and the malicious target sentences w.r.t io-sample-min-hit , for the baseline model on Ubuntu data. Now we apply the negative training framework, and aim to reduce the hit rate of a trained model for a given list of malicious targets. During each iteration of negative training, for every target sentence ytarget, we first call the gibbs-enum algorithm to find the trigger input xtrigger . And if the target is 5For this paper to be self-contained, we describe the gibbsenum algorithm in Appendix A. 6A LSTM language model (LM) is trained on the same

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

10

5For this paper to be self-contained, we describe the gibbsenum algorithm in Appendix A. 6A LSTM language model (LM) is trained on the same training data (regarding each response as an independent sentence).Algorithm 1 Negative Training for the Malicious Response Problem Input: Target list Ytarget, model parameter , learning rate , criterion for hit c, and training dataDtrain forytarget inYtarget do Getxtrigger forytarget using the gibbs-enum algorithm. whilec(xtrigger;ytarget) = 1 do Negative update: =
rlogP(ytargetjxtrigger) Get data sample (xpos;ypos)fromDtrain Positive update: =+
POS
rlogP(yposjxpos) end while end for Trigger Input: ok give me a minute to see what you have in the first place Malicious Target: no one will help you Trigger Input: mirc it ’s supposed to be <unk> ’ that seems to be the problem Malicious Target: i do n’t think i like you Trigger Input: i know about photoshop i use skype too

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

11

seems to be the problem Malicious Target: i do n’t think i like you Trigger Input: i know about photoshop i use skype too but i ca n’t buy it Malicious Target: you have no money Table 1: Examples of trigger inputs. hit (c(xtrigger;ytarget) = 1 ), we update the model to reduce the log-likelihood P(ytargetjxtrigger). The process is formulated in Algorithm 17. For each trigger input, multiple iterations of negative updates are usually needed before the hit criterion is no longer met. Note that in each iteration, thegibbs-enum algorithm is called again to find a new trigger input for each target. In our experiments, we show that negative training effectively reduces the hit rate for malicious targets after each iteration, and eventually, the gibbsenum algorithm can no longer find trigger inputs for a large number of targets that were initially hits. 3.3 Negative Training for the Frequent Response Problem The generic response problem (Li et al., 2016) for end-to-end dialogue response generation refers to the typical behavior of a MLE trained model, whereby the generated responses are mostly safe, 7Note that in actual implementation, the algorithm is minibatch based.

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

12

to the typical behavior of a MLE trained model, whereby the generated responses are mostly safe, 7Note that in actual implementation, the algorithm is minibatch based. boring or uninformative (such as “ i don’t know ” or “ good idea ”). However, it is difficult to invent an automatic criterion to determine whether a response is generic or not. In this work, we focus on the frequent response problem, as a sub-problem of the generic response problem. It refers to the behavior that a trained model generates exactly the same (usually boring) response, with a high frequency. We propose to use a metric called max-ratio to measure how severe the frequent response problem is. Given a test set and a decoding method, the model will generate a set of responses, and maxratio is defined to be the ratio of the most frequent response. In our experiments, the baseline models have a max-ratio of around 0:3for response like “ I don’t know ” across different data-sets, showing the severity of the frequent response problem. During negative training for frequent response, first a threshold ratio rthresis selected (such as 0.01), and responses with frequency ratio larger than rthres

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

13

During negative training for frequent response, first a threshold ratio rthresis selected (such as 0.01), and responses with frequency ratio larger than rthres will be discouraged. For each iteration, the model’s response to each training data input sentence is monitored and responses with frequency larger than rthreswill be used as negative examples. The frequency statistics are calculated using the current and the last 200 mini-batches. The procedure is formulated in Algorithm 2. Note that positive training is also needed here for the model to retain its original performance. Algorithm 2 Negative Training for the Frequent Response Problem Input: Model parameter , threshold ratio rthres, learning rate , and training data set Dtrain for(xpos;ypos)inDtraindo Generate response ysample from the model. Compute the frequency rsample forysample in the last 200 mini-batches. ifrsample>r thresthen Negative update: =
rlogP(ysamplejxpos) Positive update: =+
POS
rlogP(yposjxpos) end if end for In our experiments, it is shown that negative

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

14

=+
POS
rlogP(yposjxpos) end if end for In our experiments, it is shown that negative training significantly reduces max-ratio for the model on test data, and greatly increases the diversity of the model’s responses.4 Experiments We conduct experiments on three publicly available conversational dialogue data-sets: Ubuntu, Switchboard, and OpenSubtitles. To save space, descriptions of the data-sets are provided in Appendix B. 4.1 Baseline Model Training For all data-sets, we first train an LSTM based LM and attention based seq2seq models with one hidden layer of size 600, and the embedding size is set to 300. For Switchboard a dropout layer with rate 0.3 is added to the model because over-fitting is observed. The mini-batch size is set to 64 and we apply SGD training with a fixed starting learning rate (LR) for 10 iterations, and then another 10 iterations with LR halving. For Ubuntu and Switchboard, the starting LR is 1, while a starting LR of 0.1 is used for OpenSubtitles. The results are shown in Appendix C. After negative training, in addition to measuring

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

15

LR of 0.1 is used for OpenSubtitles. The results are shown in Appendix C. After negative training, in addition to measuring the hit rate for malicious targets or the diversity of the responses, it is also important to check whether the original sample quality of the baseline model is damaged. Towards that end, the perplexity of the model before and after negative training will be compared, we also conduct human evaluation to measure whether the sample quality is decreased. Other popular measurements, such as the BLEU score, have been found to correspond poorly with human judgements (Liu et al., 2016). Nevertheless, we also find that the model’s BLEU score does not become worse after negative training. 4.2 Experiments on the Malicious Response Problem Following (He and Glass, 2019), a list of malicious targets are created to test whether negative training can teach the model not to generate sentences in the list. However, in addition to prevent the model from generating targets in a specific list, it is also important to check whether negative training generalizes to other malicious targets. So, a testtarget list which contains similar but different targets from the training list are also created to test generalization. The training and test lists each contain 0.5k targets.

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

16

from the training list are also created to test generalization. The training and test lists each contain 0.5k targets. It is also interesting to investigate whether using more malicious targets for negative training can lower the hit rate on the test list. Towards that end, we train a seq2seq paraphrase model using the paraNMT data-set (Wieting and Gimpel, 2017), Train Paraphrase Test you are broken you ’re broken are you broken i will kill i ’ll kill myself i ’m going to kill you are bad you ’re bad you are really bad you are stupid you ’re stupid you are so stupid you shut up shut your mouth can you shut up Table 2: Examples of malicious targets in the training list, the test list, and paraphrases of the training targets which will be used for augmentation. with a model of the same structure as described in Section 2. Then, the paraphrase model is used to generate paraphrases of the malicious targets in the training target list8for augmentation. In our experiments, the training list without augmentation is first used for negative training, then it is augmented with 0.5k or 2k paraphrased targets respectively (1 or 4 paraphrase copies for each training target sentence). Samples of the malicious targets are shown

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

17

with 0.5k or 2k paraphrased targets respectively (1 or 4 paraphrase copies for each training target sentence). Samples of the malicious targets are shown in Table 2. The same training, augmented training and test list are used for all three data-sets, and there is no sequence-level overlap between training lists (augmented or not) and the test list. In our experiments, we spotted a harmful side effect of negative training where frequent words in the training target list are severely penalized and sometimes receive low probability even in normal perplexity testing, especially for experiments with smallPOS. To alleviate this problem, we use a simple technique called frequent word avoiding (FWA): negative gradients are not applied to the most frequent words in the malicious training target list9. For example, when doing negative training against the target “ i hate you <EOS> ”, only “hate ” will get a negative gradient. For all data-sets, negative training (Algorithm 1) is executed on the (trained) baseline model for 20 iterations over the training target list. A fixed learning rate of 0.01 and a mini-batch size of 100 are used.POSis set to 0.1 for Ubuntu, and to 1 for Switchboard and OpenSubtitles.

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

18

are used.POSis set to 0.1 for Ubuntu, and to 1 for Switchboard and OpenSubtitles. The main results are shown in Table 3. For Switchboard we focus on sample-avg-hit because we find very few targets are hit w.r.t. samplemin-hit (Similar results are reported in (He and Glass, 2019)), while for Ubuntu and OpenSubtitles we focus on sample-min-hit . Note that we get very similar results w.r.t. sample-avg-hit for 8Note the training and test lists are manually created. 9The exact avoiding word set used is f<EOS>, you, i, me, are, to, do g.Ubuntu o-sample-min-hit io-sample-min-hit Training Train Test PPL Train Test PPL Baseline 16.4% 12.6% 59.49 7.8% 5.2% 59.49 +neg-tr(0.5k) 0% 2% 60.42 0.2% 1.4% 59.97 +neg-tr(1k) 0.1% 1.4% 60.72 0.1% 1% 60.21

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

19

+neg-tr(1k) 0.1% 1.4% 60.72 0.1% 1% 60.21 +neg-tr(2.5k) 0.04% 0% 62.11 0.2% 0% 63.37 Switchboard o-sample-avg-hit io-sample-avg-hit Training Train Test PPL Train Test PPL Baseline 27.8% 27.6% 42.81 19.6% 21% 42.81 +neg-tr(0.5k) 3.8% 13.4% 42.91 2.2% 9.4% 42.7 +neg-tr(1k) 2.4% 5% 42.96 2.1% 4% 42.76 +neg-tr(2.5k) 1.3% 2.6% 43.51 1.5% 1.6% 43.24 OpenSub o-sample-min-hit io-sample-min-hit Training Train Test PPL Train Test PPL Baseline 40.7% 36.6% 70.81 19.2% 13.6% 70.81 +neg-tr(0.5k) 5.8% 12.2% 77.90 5.2% 6.6% 73.48

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

20

+neg-tr(0.5k) 5.8% 12.2% 77.90 5.2% 6.6% 73.48 +neg-tr(1k) 5.2% 7% 68.77 9.2% 4.6% 68.92 +neg-tr(2.5k) 4.8% 6% 74.07 3.4% 3.6% 75.9 Table 3: Main results for the hit rates of malicious targets before and after negative training. ”Neg-tr(0.5k)” refers to the negative training experiment using the original malicious training target list without paraphrase augmentation. Ubuntu/OpenSubtitles, and we omit those results here. We first observe that, for all data-sets, negative training can effectively reduce the hit rate on the training target list to less than 5% with little or no degradation on perplexity. We provide a comparison of the model’s behavior in Appendix D. Also, significant hit rate reduction is achieved on the test target list, which has no overlap with the training target list. This shows that negative training, similar to traditional positive training, also generalizes. It is also shown that training list augmentation

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

21

target list, which has no overlap with the training target list. This shows that negative training, similar to traditional positive training, also generalizes. It is also shown that training list augmentation can further reduce the malicious target hit rate consistently for both training and test lists. For example, on Ubuntu data, the hit rate after negative training w.r.t. o-sample-min-hit is 12.6%, and can be reduced to 0% with paraphrase augmentation. We find that that the model’s generation behavior in non-adversarial setting is almost the same as the baseline after negative training. For example, the 10-best list from beam search before/after neg-train has larger than 90% overlap. We also find that the model generates similar samples (shown in Appendix G). We believe the reason is that negative training focuses on making the model more robust with the adversarial inputs, and the original generation behavior is kept intact by the positive training (Equation 4). 4.3 Experiments on the Frequent Response Problem In this section we report results where the negative training framework (Section 3.3) is applied to tackle the frequent response problem. For all datasets, negative training is executed for 20 iterations on the MLE trained model over the training data,

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

22

tackle the frequent response problem. For all datasets, negative training is executed for 20 iterations on the MLE trained model over the training data, with a selected rthres. A fixed learning rate of 0.001 is used for all three data-sets, the mini-batch size is set to 64 and POSis set to 1. In this work, we focus on improving the model’s greedy decoding behavior instead of beam search for the following two reasons: 1) For the baseline models our experiments, we found that beam search gives far worse response diversity than greedy decoding, because it favors short responses (usually only of length one) too much, resulting in a much larger max-ratio ; 2) During training, doing beam search is much more time-consuming than greedy decoding. To measure the diversity of the model’s generated responses, in addition to max-ratio introduced in Section 3.3, which is specially design for the frequent response problem, we also adopt the entropy metric proposed in (Zhang et al., 2018). Given a set of responses from decoding on the test set, Ent-n calculates the entropy of the n-gram distribution: Ent-n =X g2Gnr(g) logr(g) (5)

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

23

calculates the entropy of the n-gram distribution: Ent-n =X g2Gnr(g) logr(g) (5) whereGnis the set of all n-grams that appeared in the response set, and r(g)refers to the ratio (frequency) of n-gram gw.r.t. all n-grams in the responses set. In our experiments with negative training, a harmful side-effect is spotted: during decoding, the model tends to output long and ungrammatical responses such as “ i do n’t know if it ’s a real valid deterrent crime crime yeah i ’m satisfied trying not to ”. We believe the reason is that the sentence end token <EOS> gets over penalized during negative training (it appears in every negative example). So, we apply the same frequent word avoiding (FWA) technique used in Section 4.2, except that here only the negative gradient for <EOS> is scaled by 0.110. In addition to the baseline model, we compare our proposed negative training framework against a 10We find that scal by zero will result in extremely short responses.Ubuntu rthres PPL M-ratio E-2 E-3

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

24

10We find that scal by zero will result in extremely short responses.Ubuntu rthres PPL M-ratio E-2 E-3 Test-set N/A N/A 1.1% 10.09 11.32 Baseline N/A 59.49 4.4% 5.33 5.92 +GAN N/A 59.43 4.7% 5.30 5.87 +MMI N/A N/A 4.5% 5.34 5.93 +neg-train 1% 59.76 1.2% 5.74 6.52 +neg-train 0.1% 60.06 1.3% 6.44 7.55 Switchboard rthres PPL M-ratio E-2 E-3 Test-set N/A N/A 10.0% 8.61 9.65 Baseline N/A 42.81 37.4% 2.71 2.42 +GAN N/A 42.69 49% 2.66 2.35 +MMI N/A N/A 23% 5.48 6.23 +neg-train 10% 42.84 12.4% 3.86 4.00 +neg-train 1% 44.32 9.8% 5.48 6.03

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

25

+neg-train 10% 42.84 12.4% 3.86 4.00 +neg-train 1% 44.32 9.8% 5.48 6.03 OpenSubtitles rthres PPL M-ratio E-2 E-3 Test-set N/A N/A 0.47% 9.66 10.98 Baseline N/A 70.81 20% 4.22 4.59 +GAN N/A 72.00 18.8% 4.08 4.43 +MMI N/A N/A 3.6% 7.63 9.08 +neg-train 1% 72.37 3.1% 5.68 6.60 +neg-train 0.1% 75.71 0.6% 6.90 8.13 Table 4: Main results of negative training with differentrthres, for the frequent response problem. Diversity metrics for the responses in the test data are also shown, “E-n” /“M-ratio” refer to the Ent-n /max-ratio metric. GAN (Goodfellow et al., 2014a) approach, where a discriminator Dis introduced and the generator G tries to fool the discriminator to believe its samples are real data samples: min Gmax DV(D;G )

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

26

discriminator Dis introduced and the generator G tries to fool the discriminator to believe its samples are real data samples: min Gmax DV(D;G ) = min Gmax DfE(x;y)PdatalogD(x;y)+ ExPdata;yG(
jx)log(1D(x;y))g(6) where the generator Grefers to the seq2seq model P. The GAN framework is very attractive for tackling the generic response problem (Li et al., 2017; Zhang et al., 2018), because the discriminator can act as a critic to judge whether a response sample is boring. We describe the training details and hyper-parameter setting for the GAN approach in Appendix E. We also provide an comparison to the MMI decoding (Li et al., 2016), which is a very popular work in this field. We implement MMI-antiLM for our models. The experimental results are shown in Table 4. The experiment with best diversity result and nondegenerate sample quality are shown in bold. We first observe a large gap on the diversity measures between the baseline models and the test set, especially on Switchboard and OpenSubtitles data. Switchboard OpenSubtitles

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

27

between the baseline models and the test set, especially on Switchboard and OpenSubtitles data. Switchboard OpenSubtitles Input: it ’ll cost about three hundred dollars for a stud Input: captain you wanted to see me Baseline: i think that ’s a good idea Baseline: i ’m sorry Neg-train: i think i would agree with that Neg-train: i was in the hotel Input: we want to breed her with a champion Input: yes mr. brown could i Baseline: i do n’t know Baseline: i do n’t know Neg-train: i think it was Neg-train: i ’d like to introduce myself Input: now these are long haired Input: leave it to me Baseline: i do n’t know Baseline: i ’m not going to leave you Neg-train: i ’ve been in a very very good shape Neg-train: you ’re taking the first step Input: the other two are short hairs Input: thank you mr. brown Baseline: i do n’t know Baseline: i ’m sorry Neg-train: i ’m going to try to get it Neg-train: i ’m happy to see you

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

28

Neg-train: i ’m going to try to get it Neg-train: i ’m happy to see you Table 5: Greedy-decoding samples on the test data before and after negative training. The samples are consecutive (input of the next sample is the reference response for the previous one). That indicates the severity of the frequent/generic response problem. Then, results of negative training with different rthresshow that negative training can significantly increase response diversity, with little or no loss on PPL or BLEU score (shown in Appendix F) performance. For example, maxratio is reduced by 73.7% and Ent-3 is increased by 149% for Switchboard data. Further, consistent improvement is achieved when a smaller rthresis used. However, sample quality will decrease (becoming too long or ungrammatical) when rthresis too small. The reason could be that when too much diversity is asked for, the model will go to extremes to provide diversity, resulting in degradation of sample quality. Comparing to MMI, note that although on Switchboard/Opensubtitles MMI gives higher entropy, the max-ratio is not as low as the negative training result, which is the main focus of our work (the frequent response problem). We also

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

29

training result, which is the main focus of our work (the frequent response problem). We also find MMIs hyper-parameters are difficult to tune: the working set of hyper-parameters dont transfer well between data-sets. Further, for MMI in a lot of configuration tries the model gives ungrammatical output samples (this is problem is also mentioned in the paper (Li et al., 2016)). For the Ubuntu data, we can not even find a configuration that performs better than the baseline model. Further, the vanilla GAN approach is not shown to be effective in our experiments. The reason could be that despite its discriminative nature, GAN training still feeds “positive” gradient for samples from the model (eq. (11) and eq. (12) in AppendixE), which is not enough to prevent the model from generating them. We believe additional techniques (Zhang et al., 2018; Li et al., 2017) are needed for the GAN approach to be effective. We show some model samples before and after negative training in Table 5. It is shown that negative training effectively discourages boring responses, and response diversity is improved. However, one limitation is observed that diversity does

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

30

negative training effectively discourages boring responses, and response diversity is improved. However, one limitation is observed that diversity does not necessarily lead to improvement on the informativeness of the response w.r.t. the input (sometimes the model generates a completely unrelated response). More samples for all three data-sets are included in Appendix G. To rigorously verify negative training is not getting diversity when sacrificing the sample’s quality, a human evaluation is conducted and results are shown in Table 6. It is observed that negative training wins by a significant margin for all three data-sets. This shows that, negative training does not damage the quality of the generated samples. Note that the human evaluation does not reflect the diversity of the model, because the raters only rate one response at a time. 5 Related Works The malicious response problem and the gibbsenum algorithm to find trigger inputs (He and Glass, 2019) originates from a large body of work on adversarial attacks for deep learning models, with continuous input space (e.g. image classification) (Goodfellow et al., 2014b; Szegedy et al., 2013), or discrete input space (e.g. sentence classification, or

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

31

(Goodfellow et al., 2014b; Szegedy et al., 2013), or discrete input space (e.g. sentence classification, or Data-set Tie Baseline Neg-train Ubuntu 64.6% 14.0% 21.3% Switchboard 45.1% 18.3% 36.4% Opensubtitles 58.3% 19.0% 22.6% Table 6: Human Evaluation Results. For each dataset, 300 samples (input-output pairs) from the baseline model and the model after negative training, are evenly distributed to 4 English-speaking human evaluators. The evaluators are asked to pick a preferred sample, or report a tie. This evaluation is to check whether negative training has hampered the quality of the generation. seq2seq models) (Papernot et al., 2016; Samanta and Mehta, 2017; Liang et al., 2018; Ebrahimi et al., 2017; Belinkov and Bisk, 2017; Chen et al., 2017). “Adversarial attacks” refer to the phenomenon that when an imperceptible perturbation is applied to the input, the output of the model can change significantly (from correct to incorrect). The trigger inputs found by the gibbs-enum algorithm, can be

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

32

the input, the output of the model can change significantly (from correct to incorrect). The trigger inputs found by the gibbs-enum algorithm, can be regarded as a type of “targeted attack”, in which the attack triggers the model to assign large probability to a specific malicious target sentence. Motivated by the works on adversarial attacks, various adversarial training strategies (Madry et al., 2017; Belinkov and Bisk, 2017; Miyato et al., 2016) have been proposed to make trained models more robust against those attacks. During adversarial training, the model is fed with adversarial examples and the correct labels. The negative training framework considered in this work differs from adversarial training in that, instead of asking the model to “do the right thing” (referred to as “positive training” in this work), the model is trained to “not do the wrong thing”. To the best of our knowledge, this is the first work investigating the concept of negative training for dialogue response models, and the first proposed solution for the malicious response problem. The malicious target list used in this work is very similar to the one used in (He and Glass, 2019). We

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

33

response problem. The malicious target list used in this work is very similar to the one used in (He and Glass, 2019). We propose to add a test target list to test the generalization of negative training. Further, we show that the training list can be effectively augmented by utilizing a paraphrase model. In this work, we propose a definition for the frequent response problem , as a sub-problem of the generic response problem (Li et al., 2016). Much research work has devoted to alleviate the generic response problem in end-to-end dialogue responsegeneration, (Li et al., 2016) use the maximal mutual information (MMI) objective, and propose to utilize an auxiliary LM to penalize the generic response during decoding. Closely related to this work, sophisticated training frameworks based on GAN (Zhang et al., 2018; Li et al., 2017) have also been shown to be effective, where techniques such as variational information maximization or reward for every generation step (REGS) are proposed to improve GAN training. However, in our experiments it is shown that a vanilla GAN approach gives unsatisfactory results. Whether negative training11is complementary to these frameworks is worth investigating in future work. Finally, note that the concept of negative training

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

34

Finally, note that the concept of negative training in this work is very different to the negative samples in word2vec training (Mikolov et al., 2013). The negative samples in word2vec training are used to prevent the training from being trivial, and is usually chosen randomly. In this work, the negative samples are carefully chosen to exhibit some particular undesirable behavior of the model, and is then used to correct such behavior. 6 Conclusion In this work, we propose the negative training framework to correct undesirable behaviors of a trained neural dialogue response generator. The algorithm involves two major steps, first input-output pairs that exhibit bad behavior are identified, and then are used for fine-tuning the model as negative training examples. We also show that negative training can be derived from an overall objective (eq. (2)) to minimize the expected risk of undesirable behaviors. In our experiments, we apply negative training to the malicious response problem and the frequent response problem and get significant improvement for both problems. References Yonatan Belinkov and Yonatan Bisk. 2017. Synthetic and natural noise both break neural machine translation. CoRR , abs/1711.02173. Hongge Chen, Huan Zhang, Pin-Yu Chen, Jinfeng Yi,

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

35

and natural noise both break neural machine translation. CoRR , abs/1711.02173. Hongge Chen, Huan Zhang, Pin-Yu Chen, Jinfeng Yi, and Cho-Jui Hsieh. 2017. Show-and-fool: Crafting adversarial examples for neural image captioning. CoRR , abs/1712.02051. Minhao Cheng, Jinfeng Yi, Huan Zhang, Pin-Yu Chen, and Cho-Jui Hsieh. 2018. Seq2sick: Evaluating the 11Note that negative training is considerably easier to implement than the mentioned frameworks based on GAN. robustness of sequence-to-sequence models with adversarial examples. CoRR , abs/1803.01128. Kyunghyun Cho, Bart van Merri ¨enboer, C ¸ alar G¨ulc ¸ehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning phrase representations using rnn encoder–decoder for statistical machine translation. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages 1724– 1734, Doha, Qatar. Association for Computational Linguistics.

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

36

the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages 1724– 1734, Doha, Qatar. Association for Computational Linguistics. Javid Ebrahimi, Anyi Rao, Daniel Lowd, and Dejing Dou. 2017. Hotflip: White-box adversarial examples for NLP. CoRR , abs/1712.06751. Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. 2014a. Generative adversarial nets. In Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2 , NIPS’14, pages 2672–2680, Cambridge, MA, USA. MIT Press. Ian J. Goodfellow, Jonathon Shlens, and Christian Szegedy. 2014b. Explaining and harnessing adversarial examples. CoRR , abs/1412.6572. Tianxing He and James Glass. 2019. Detecting egregious responses in neural sequence-to-sequence models. In International Conference on Learning Representations . Sepp Hochreiter and J ¨urgen Schmidhuber. 1997.

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

37

models. In International Conference on Learning Representations . Sepp Hochreiter and J ¨urgen Schmidhuber. 1997. Long short-term memory. Neural computation , 9(8):1735–1780. Yoon Kim. 2014. Convolutional neural networks for sentence classification. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, EMNLP 2014, October 25-29, 2014, Doha, Qatar, A meeting of SIGDAT, a Special Interest Group of the ACL , pages 1746–1751. Jiwei Li, Michel Galley, Chris Brockett, Jianfeng Gao, and Bill Dolan. 2016. A diversity-promoting objective function for neural conversation models. In NAACL HLT 2016, The 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, San Diego California, USA, June 12-17, 2016 , pages 110–119. Jiwei Li, Will Monroe, Tianlin Shi, Alan Ritter, and Dan Jurafsky. 2017. Adversarial learning for neural dialogue generation. CoRR , abs/1701.06547. Bin Liang, Hongcheng Li, Miaoqiang Su, Pan Bian,

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

38

dialogue generation. CoRR , abs/1701.06547. Bin Liang, Hongcheng Li, Miaoqiang Su, Pan Bian, Xirong Li, and Wenchang Shi. 2018. Deep text classification can be fooled. In Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, IJCAI 2018, July 13-19, 2018, Stockholm, Sweden. , pages 4208–4215.Chia-Wei Liu, Ryan Lowe, Iulian Serban, Mike Noseworthy, Laurent Charlin, and Joelle Pineau. 2016. How not to evaluate your dialogue system: An empirical study of unsupervised evaluation metrics for dialogue response generation. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing , pages 2122–2132. Association for Computational Linguistics. Ryan Lowe, Nissan Pow, Iulian Serban, and Joelle Pineau. 2015. The ubuntu dialogue corpus: A large dataset for research in unstructured multi-turn dialogue systems. CoRR , abs/1506.08909. Thang Luong, Hieu Pham, and Christopher D. Manning. 2015. Effective approaches to attention-based neural machine translation. In Proceedings of the

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

39

Thang Luong, Hieu Pham, and Christopher D. Manning. 2015. Effective approaches to attention-based neural machine translation. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing , pages 1412–1421. Association for Computational Linguistics. Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu. 2017. Towards deep learning models resistant to adversarial attacks. CoRR , abs/1706.06083. Tomas Mikolov, Martin Karafi ´at, Luk ´as Burget, Jan Cernock ´y, and Sanjeev Khudanpur. 2010. Recurrent neural network based language model. In INTERSPEECH 2010, 11th Annual Conference of the International Speech Communication Association, Makuhari, Chiba, Japan, September 26-30, 2010 , pages 1045–1048. Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Distributed representations of words and phrases and their compositionality. In Proceedings of the 26th International Conference on Neural Information Processing Systems Volume 2 , NIPS’13, pages 3111–3119, USA. Curran Associates Inc.

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

40

Associates Inc. Takeru Miyato, Andrew M. Dai, and Ian Goodfellow. 2016. Adversarial training methods for semi-supervised text classification. Cite arxiv:1605.07725Comment: Published as a conference paper at ICLR 2017. Franz Josef Och. 2003. Minimum error rate training in statistical machine translation. In Proceedings of the 41st Annual Meeting on Association for Computational Linguistics - Volume 1 , ACL ’03, Stroudsburg, PA, USA. Association for Computational Linguistics. Nicolas Papernot, Patrick D. McDaniel, Ananthram Swami, and Richard E. Harang. 2016. Crafting adversarial input sequences for recurrent neural networks. In 2016 IEEE Military Communications Conference, MILCOM 2016, Baltimore, MD, USA, November 1-3, 2016 , pages 49–54. Suranjana Samanta and Sameep Mehta. 2017. Towards crafting text adversarial samples. CoRR , abs/1707.02812. Rupesh Kumar Srivastava, Klaus Greff, and J ¨urgen Schmidhuber. 2015. Highway networks. CoRR , abs/1505.00387.

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

41

Schmidhuber. 2015. Highway networks. CoRR , abs/1505.00387. Ilya Sutskever, Oriol Vinyals, and Quoc V . Le. 2014. Sequence to sequence learning with neural networks. InAdvances in Neural Information Processing Systems 27: Annual Conference on Neural Information Processing Systems 2014, December 8-13 2014, Montreal, Quebec, Canada , pages 3104–3112. Richard S. Sutton and Andrew G. Barto. 1998. Introduction to Reinforcement Learning , 1st edition. MIT Press, Cambridge, MA, USA. Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian J. Goodfellow, and Rob Fergus. 2013. Intriguing properties of neural networks. CoRR , abs/1312.6199. J¨org Tiedemann. 2009. News from OPUS - A collection of multilingual parallel corpora with tools and interfaces. In N. Nicolov, K. Bontcheva, G. Angelova, and R. Mitkov, editors, Recent Advances in Natural Language Processing , volume V , pages 237–248. John Benjamins, Amsterdam/Philadelphia, Borovets, Bulgaria.

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

42

Advances in Natural Language Processing , volume V , pages 237–248. John Benjamins, Amsterdam/Philadelphia, Borovets, Bulgaria. John Wieting and Kevin Gimpel. 2017. Pushing the limits of paraphrastic sentence embeddings with millions of machine translations. CoRR , abs/1711.05732. Lijun Wu, Yingce Xia, Li Zhao, Fei Tian, Tao Qin, Jianhuang Lai, and Tie-Yan Liu. 2017. Adversarial neural machine translation. CoRR , abs/1704.06933. Lantao Yu, Weinan Zhang, Jun Wang, and Yong Yu. 2016. Seqgan: Sequence generative adversarial nets with policy gradient. CoRR , abs/1609.05473. Yizhe Zhang, Michel Galley, Jianfeng Gao, Zhe Gan, Xiujun Li, Chris Brockett, and Bill Dolan. 2018. Generating informative and diverse conversational responses via adversarial information maximization. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, editors, Advances in Neural Information Processing Systems 31, pages 1815–1825. Curran Associates, Inc. A The Gibbs-enum Algorithm for

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

43

31, pages 1815–1825. Curran Associates, Inc. A The Gibbs-enum Algorithm for Finding Trigger Inputs In this section, we briefly describe the gibbs-enum algorithm, we also refer readers to (He and Glass, 2019) for the intuition and full development of the algorithm. The goal of gibbs-enum is that given a (malicious) target sentence yof lengthm, and a trained seq2seq model, we aim to find a trigger input sequence x, which is a sequence of one-hot vectorsfxtgof lengthn, to minimize the negative log-likelihood (NLL) that the model will generate y. We formulate our objective function L(x;y) below: L(x;y) =1 mmX t=1logPseq2seq(ytjy<t;x)+inR(x) (7) A regularization term R(x)is applied when looking for io-sample-min/avg-hit , which is the LM score of x: R(x) =1 nnX t=1logPLM(xtjx<t) (8) In our experiments we set into1when searching

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

44

R(x) =1 nnX t=1logPLM(xtjx<t) (8) In our experiments we set into1when searching forio-sample-min/avg-hit , otherwise 0. During gibbs-enum , every time we focus on a single index slot xt, and find the best one-hot xt while keeping the other parts of xfixed: arg min xtL(x<t;xt;x>t;y) (9) Since the size of vocabulary jVjis finite, it is possible to try all of them and get the best local xt. But it is still costly since each try requires a forwarding call to the neural seq2seq model. To address this, gradient information is utilized to narrow the range of search. We temporarily regard xtas a continuous vector and calculate the gradient of the negated loss function with respect to it: rxt(L(x<t;xt;x>t;y)) (10) Then, we try only the Gindexes that have the highest value on the gradient vector. The procedure is formulated in Algorithm 3. For hyper-parameters of gibbs-enum ,T(the

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818

1903.02134

45

formulated in Algorithm 3. For hyper-parameters of gibbs-enum ,T(the maximum number of sweeps) is set to 5, G(size of the set of indices for enumeration during each update) is set to 100, the algorithm is run 5 times with different random initializations and the trigger input with the best loss is returned. Note that larger hyper-parameters can give slightly higher hit rates, but will be more time-consuming.Algorithm 3 Gibbs-enum algorithm Input: a trained seq2seq model, target sequence y, a trained LSTM LM, objective function L(x;y), input length n, output length m, and target hit type. Output: a trigger input x ifhit type is in “ io-hit ”then initialize xto be a sample from the LM else randomly initialize xto be a valid input sequence end if fors= 1;2;:::;T do fort= 1;2;:::;n do get gradientrx t(L(x <t;x t;x >t;y)), and set list Hto be theGindexes with highest value in the gradient vector forj= 1;2;:::;G do setx0to be:

1903.02134

Negative Training for Neural Dialogue Response Generation

Although deep learning models have brought tremendous advancements to the field of open-domain dialogue response generation, recent research results have revealed that the trained models have undesirable generation behaviors, such as malicious responses and generic (boring) responses. In this work, we propose a framework named "Negative Training" to minimize such behaviors. Given a trained model, the framework will first find generated samples that exhibit the undesirable behavior, and then use them to feed negative training signals for fine-tuning the model. Our experiments show that negative training can significantly reduce the hit rate of malicious responses, or discourage frequent responses and improve response diversity.

http://arxiv.org/pdf/1903.02134

cs.CL

20190306

20200818