Proceedings of the 5th Workshop on Structured Prediction for NLP (SPNLP 2021)

Zornitsa Kozareva, Sujith Ravi, Slicex Ai, Priyanka Agrawal, André F. T. Martins, Heng Ji, Ivan Titov, Wilker Aziz, Arzoo Katiyar, Sabrina Mielke, Amr Sharaf, Tim Vieira, Parag Jain, S. Welleck, Michail Korakakis, Chunchuan Lyu, Jacob Parnell, I. Unanue, Massimo Piccardi, Ohad Rubin, Jonathan Berant, J. Groschwitz, Meaghan Fowlie, Alexander Koller, Julia Kreutzer, S. Riezler, Carolin (Haas) Lawrence, Erenay Dayanik, André Blessing, Nico Blokker, Sebastian Haunss, Jonas Kuhn, Gabriella Lapesa, S. Pad, Chenyang Huang, Wei Yang, Yanshuai Cao, O. Zaane, Li-Li Mou, André Blessing, Cao, Yanshuai, Cho, Dayanik, Fowlie, Meaghan, Hovy, Huang, J. Unanue, Koller, Alexander, Ilia Kulikov, Mou, Lili, Parnell, Jacob, Piccardi, Riezler, Stefan, Emma Strubell, Zhang, Kristjan Arumae, Fei Liu, Sanghwan Bae, Taeuk Kim, Jihoon Kim, Sanggoo Lee, J. Devlin, M. Chang, Kenton Lee, K. Toutanova, Sergey Edunov, Myle Ott, M. Auli, D. Grangier, M. Ranzato, Alexander R. Fabbri, Irene Li, Tian-Wei She, Su-Yi Li, Dragomir R. Radev, Yang Gao, Wei Zhao, S. Eger, Max Grusky, Mor Naaman, Yoav Artzi, Ari Holtzman, Jan Buys, LI DU, M. Forbes, Ye-Jin Choi, Eric Jang, Shi-Xiang Gu, B. Poole, S. Rennie, E. Marcheret, Y. Mroueh, J. Ross, V. Goel, Ronald Williams, Jing-Qing Zhang, Yao Zhao, Mohammad Saleh, Peter J. Liu, Tian-Yi Zhang, Varsha Kishore, Felix Wu, K. Weinberger, Chen-Guang Zhu, Zi-Yi Yang, R. Gmyr, Michael Zeng, Xue-Dong Huang, Jonathan Berant, A. Chou, R. Frostig, P. Liang, Jian-Peng Cheng, Siva Reddy, V. Saraswat, Mirella Lapata, J. Clarke, Dan Goldwasser, Dan Roth, E. Codd, Li Dong, M. Gardner, J. Grus, Mark Neumann, O. Tafjord, P. Dasigi, Nelson F. Liu, M. Peters, Michael Schmitz, Luke Zettlemoyer, Jia-Qi Guo, Ze-Cheng Zhan, Yan Gao, Yan Xiao, Jian-Guang Lou, Ting Liu, Dong-Mei Zhang, Jonathan Herzig, Krzysztof Pawel, Thomas Nowak, Francesco Mller, Julian Piccinno, Eisenschlos, Jayant Krishnamurthy, C. Liang, Q. Le, K. Lao, P. Liang, Michael I. Jordan, D. Klein, Victoria Xi, Richard Lin, Caiming Socher, Xiong, Tatiana Anikina, Michael Roth, J. Baker, L. Banarescu, C. Bonial, S. Cai, M. Georgescu, K. Griffitt, U. Hermjakob, K. Knight, Philipp Koehn, Martha Palmer, Nathan Schneider, Michele Bevilacqua, Rexhina Blloshmi, Roberto Navigli, Yu-Fei Chen, Wei-Wei Sun, Xiao-Jun Wan, A. Copestake, D. Flickinger, A. Lascarides, Jason M. Eisner, Federico Fancellu, Sorcha Gilroy, A. Lopez, D. Fernndez, -Gonzlez, Carlos Gmez-Rodrguez, Mark Johnson, Matthias Lindemann, Serhii Havrylov, G. Kruszewski, A. Joulin, Han He, Jinho D. Choi, Nikita Kitaev, J. Maillard, S. Clark, Dani Yogatama, S. Oepen, Marco Kuhlmann, Yusuke Miyao, Daniel Zeman, Silvie Cinkov, Jan Haji, Z. Ureov, Xiao-Chang Peng, Lin-Feng Song, D. Gildea, A. Vaswani, Noam Shazeer, N. Parmar, J. Uszkoreit, Llion Jones, A. Gomez, I. Kaiser, Polosukhin, D. Bahdanau, Philémon Brakel, Kelvin Xu, Anirudh Goyal, Ryan Lowe, J. Pineau, A. Courville, Y. Bengio, Felix Berkenkamp, M. Turchetta, Angela Schoellig, Andreas Krause, L. Bottou, Jonas Peters, Joaquin Quionero-Candela, Denis Charles, D. Chickering, Elon Portugaly, D. Ray, P. Simard, Edward Snelson, Chris Callison-Burch, M. Carl, Barbara Dragsted, Jakob Elming, Daniel Hardt, A. Lykke, Jakobsen, Olivier Chapelle, Li-Hong Li, Leshem Choshen, Lior Fox, Zohar Aizenbud, Omri Abend, Paul Christiano, Jan Leike, Tom B. Brown, Miljan Martic, S. Legg, D. Amodei, Ernest Davis, Fabio De Bona, S. Riezler, Keith B. Hall, M. Ciaramita, Maria Amac Herdadelen, Holmqvist, Miroslav Dudík, John Langford, Keith Godwin, P. Piwek, A. Hans, Daniel Schneega, Anton, Maximilian Schfer, Steffen Udluft, Ji He, Jian-Shu Chen, Xiao-Dong He, Jianfeng Gao, Li Deng, M. Ostendorf, Nan Jiang, Kirthevasan Kandasamy, Yoram Bachrach, Ryota Tomioka, Daniel Tarlow, David Carter, Julia Kreutzer, Shahram Khadivi, E. Matusov, S. Riezler, Julia Kreutzer, Joshua Uyheng, Kin-Cho Tsz, J.-D. Lam, Stefan Kreutzer, Alexander Strehl, Jennifer Wortman, Pratik Gajane, S. Riezler, Artem Sokolov, Ji-Wei Li, Will Monroe, Alan Ritter, D. Jurafsky, Michel Galley, A. Lommel, Maja Popovic, Aljoscha Burchardt, K. Nguyễn, H. Daum, Iii., J. Boyd-Graber, Rodrigo Nogueira, K. Cho, Bo Pang, E. Nijkamp, Wen-Juan Han, Ling-Qing Zhou, Yixian Liu, Kewei Tu, D. Precup, Richard E. Sutton, Satinder Singh, Pablo Rivas, Kerstin Holzmayer, Cristian Hernandez, C. Grippaldi, Paul Rosenbaum, D. Rubin, Julia Kreutzer, Christopher Lo, S. Kakade, Adith Swaminathan, Thorsten Joachims, Ryuichi Takanobu, Han-Lin Zhu, Minlie Huang, Philip Thomas, Emma Brunskill, L. Leon, Thurstone, Gaurav Trivedi, Robert Esmaeel R. Dadashzadeh, Wendy Handzel, Shyam Chapman, Harry Visweswaran, Hochheiser, Luke Joseph P. Turian, Shea Dan, Melamed, S. Verberne, E. Krahmer, Iris Hendrickx, S. Wubben, Antal Van Den, Bosch, Jing-Kang Wang, Yang Liu, Bo Li, D. Andor, Chris Alberti, David Weiss, A. Severyn, Alessandro Presta, K. Ganchev, Slav Petrov, Michael Collins, Sourya Basu, Govardana Sachithanandam Ramachandran, N. Keskar, L. Varshney, Yuntian Deng, A. Bakhtin, A. Szlam, B. Eikema, S. Hochreiter, J. Schmidhuber, Yoon Kim, A. Rush, H. Pedro, Zita Martins, Marinho, F. Andr, Martins, Kenton W. Murray, D. Chiang, R. Sennrich, B. Haddow, A. Birch, Pavel Sountsov, Sunita Sarawagi, Felix Stahlberg, B. Byrne, Stephen Roller, Emily Dinan, Jason Weston, R. Bharat, Rahul Ambati, M. Agarwal, Samar Gupta, D. Husain, Sharma, David Bamman, Noah A. Smith, I. Dagan, Jacob Goldberger, P. Bojanowski, E. Grave, T. Mikolov, Wei-Cheng Chang, Hsiang-Fu Yu, Kai Zhong, Yi-Ming Yang, I. Dhillon, Fernando Pereira, Mark A. Mandel, Steven Carroll, Peter W. White, Goran Glava, F. Nanni, Simonetti Paolo, Ponzetto, A. Graves, N. Jaitly, Abdel-rahman Mohamed, Ruud Koopmans, P. Statham, Gregory Kossinets, Gabriella Lapesa, André Blessing, Nico Blokker, Erenay Dayanik, Sebastian Haunss, Jonas Kuhn, S. Pad, Khai Mai, Thai-Hoang Pham, M. Nguyen, T. Nguyen, Danushka Bollegala, Ryohei Sasano, Satoshi Sekine, S. Pad, André Blessing, Erenay Dayanik, Sebastian Haunss, Jonas Kuhn, Vasin Punyakanok, Wen-tau Yih, Dav Zimak, Sebastian Riedel, Sonse Shimaoka, Pontus Stenetorp, Kentaro Inui, Ben Bogin, D. Dahl, Madeleine Bates, Michael Brown, William Fisher, Kate Hunicke-Smith, D. Pallett, Christine Pao, Alexander I. Rudnicky, Elizabeth Shriberg, Daniel Marcu, Jia-Tao Gu, Zheng-Dong Lu, Hang Li, O. Victor, Li, Qian Liu, Zhen-Wen Li, Xueqing Liu, Tao Xie, Robin Jia, P. Diederik, Jimmy Kingma, Ba, Tom Kwiatkowksi, Sharon Goldwater, Mark Steedman, J. Lafferty, A. McCallum, J. Pennington, R. Socher, Christopher D. Manning, M. Iyyer, C. Clark, Milo Stanojevi, Ze-Yu Sun, Qi-Hao Zhu, Li-Li Mou, Ying-Fei Xiong, Ge Li, Lu Zhang, Yi-Can Sun, Li-Li Mou, Bailin Wang, Richard Shin, Xiao-Dong Liu · 2021

• DOI
• Semantic Scholar

Abstract

To date, most abstractive summarisation models have relied on variants of the negative loglikelihood (NLL) as their training objective. In some cases, reinforcement learning has been added to train the models with an objective that is closer to their evaluation measures (e.g. ROUGE). However, the reward function to be used within the reinforcement learning approach can play a key role for performance and is still partially unexplored. For this reason, in this paper, we propose two reward functions for the task of abstractive summarisation: the first function, referred to as RwB-Hinge, dynamically selects the samples for the gradient update. The second function, nicknamed RISK, leverages a small pool of strong candidates to inform the reward. In the experiments, we probe the proposed approach by fine-tuning an NLL pre-trained model over nine summarisation datasets of diverse size and nature. The experimental results show a consistent improvement over the negative loglikelihood baselines.

References

See the full reference list in the paper.