Invited Talks

  • Nanyun Peng (USC ISI)
    Title: Creative Generation as Inverse Summarization
    Abstract: Recent advances in data-driven approaches have demonstrated appealing results in generating natural languages in applications like machine translation and summarization. However, when the generation tasks are open-ended and the content is under-specified, existing techniques struggle to demonstrate coherence and creativity in writing. This happens because the generation models are trained to model the surface form (i.e. sequences of words), rather than the more advanced semantics and discourse structures. Moreover, composing creative pieces such as puns, poems, and stories require deviating from the norm, whereas existing generation approaches seek to mimic the norm and thus are unlikely to lead to truly novel, creative composition. In this talk, I will talk about several works done by my group and collaborators on creative story and pun generation. We emphasize the importance of understanding the semantics and discourse structures of stories and puns to achieve coherent and creative generation.
    Bio: Nanyun Peng is a Research Assistant Professor at the University of Southern California Computer Science Department, and a Research Lead at the USC Information Sciences Institute. She received a Ph.D. from Johns Hopkins University the Center for Language and Speech Processing. Her research focuses on creative language generation, robustness and generalizability of natural language understanding. She has published more than 30 papers in top NLP/AI conferences such as ACL, EMNLP, AAAI, TACL. Her research has been funded by several DARPA, IARPA, and NIH awards.

  • Ido Dagan (Bar-Ilan University)
    Title: Comprehending multi-text information: interaction and consolidation
    Abstract: For almost any topic of interest, information is spread across a substantial number of texts, whose exploration becomes a hard and tedious process. Multi Document Summarization aspires to address this difficulty by providing a concise summary of the targeted information. However, once reading such (static) summary, an interested user would typically want additional information, being forced back to the tedious process of exploring the original texts. In the first part of the talk I will advocate that this “summarization gap” can be effectively addressed by an interactive summarization process, where additional summarized information is presented dynamically in response to user interactions. Critically, to establish interactive summarization as a viable research area, we propose a systematic replicable evaluation framework, based on adapting and extending evaluation paradigms for traditional static summarization. I will then present a rather broad framework, and an initial prototype, for implementing interactive summarization through Query-Focused Summary Expansion.

    The second part of the talk addresses the critical need to consolidate multi-text information in order to present it most effectively. Consolidation should be performed at the level of minimal information units, to properly address issues such as conflating redundancy, connecting complementary information pieces and detecting discrepancies across texts, such as contradictions and multiple views. To realize such consolidation, we propose an information decomposition framework based on natural-language question-answer pairs, following and extending the QA-SRL paradigm, accompanied by mechanisms for linking co-referring statements. We envision that this framework can provide an effective basis for multi-text summarization, both static and interactive, as well as for other multi-text tasks.

  • Wenjie Li (The Hong Kong Polytechnic University)
    Title: Beyond Sequence-to-Sequence Neural Summarization
    Bio: Dr. Wenjie Li received the BSc and MSc degrees from Tianjin University, China, and the PhD degree from the Department of Systems Engineering and Engineering Management, Chinese University of Hong Kong, Hong Kong. Currently she is an associate professor in the Department of Computing, Hong Kong Polytechnic University. Her research interests include text summarization, machine conversation, natural language understanding and generation. She has directed and participated in quite a number of research projects and published more than 200 papers in major international journals and conference proceedings, including IEEE Transactions on Knowledge and Data Engineering, IEEE Transactions on Neural Networks and Learning Systems, ACM Transactions on Information Systems, Computational Linguistics, ACL, AAAI, IJCAI, WWW, and SIGIR. She has served as the information officer of SIGHAN and the acting editor of Transactions of the Association for Computational Linguistics.

  • Manabu Okumura (Tokyo Institute of Technology)
    Title: Studying the past in text summarization
    Abstract: While the research field of text summarization has a long history of more than 60 years, we have witnessed the remarkable changes and improvements in the summarization researches in these ten years. Integer programming frameworks could improve the performance of summarization models. More recently neural models can be said to enable us to output abstractive summaries, and have changed the research community drastically. However, we should sometimes stop rushing and revisit the past researches. In this talk, we will mention some points in them that might be suggestive for the neural network-based summarization models, especially the length constraint, that can be the most fundamental for the summarization models but has been often ignored.
    Bio: Manabu Okumura is currently a professor at Institute of Innovative Research, Tokyo Institute of Technology. He is also a vice-chairperson of Association for Natural Language Processing (ANLP) in Japan. He was a visiting associate professor at the Department of Computer Science, University of Toronto from 1997 to 1998. He co-organized a series of Text Summarization Challenge (TSC), a text summarization evaluation, the first of its kind in Japan, as a part of the NTCIR (NII-NACSIS Test Collection for IR Systems) Workshop. He won Google AI Focused Research Awards in Japan (2019), Google Research Award (2015), and IBM Faculty Award (2015). He plays many key roles in academic societies, journals and top-tier/representative conferences, including editor-in-chief for ANLP in Japan (2016-2018), ICICTES Conference Chair (2015-2019), Chair for JSAI isAI 2011, Area co-chair for ACL-HLT 2011, Workshop co-chair for PAKDD 2009, Area chair for EMNLP-CoNLL 2007, Area chair for IJCNLP 2004, Tutorial Co-Chair for PRICAI 2016, and Program Committee for many conferences (ACL, EACL, EMNLP, IJCAI, AAAI, WWW, ICCPOL, ICWSM, PRICAI, PAKDD, IJCNLP).