◆  Keh-Yih Su, Institute of Information Science, Academia Sinica, Taipei

Keynote Topic: On Integrating Domain Knowledge into DNN

Short Bio: Dr. Keh-Yih Su, B.S., National Tsing-Hua University, Taiwan, and Ph.D. (in EE) from University of Washington, Seattle, 1984, taught after graduation at his alma mater in Taiwan. In 1985, he launched an English-Chinese Machine Translation Project. He then founded Behavior Design Corp. in 1988 to commercialize the above project. From 1989 to 1990, he was a visiting scientist at AT&T Bell Laboratories, NJ, working on speech recognition. Dr. Su left the National Tsing-Hua university (as a professor) to join Behavior Design Corporation in 1998, and has been directing the company until May, 2014. He has become a Research Fellow at the Institute of Information Science, Academia Sinica, Taiwan since June, 2014. His current research interests are machine reading, conversational QA, and multi-document processing.

Abstract: Deep Neural Network (DNN) has been the main stream approach in either Artificial Intelligence or Natural Language Processing/Understanding fields over the last decade. Although it has obtained excellent performances in considerable different tasks, its weakness in handling some problems which require abstract and aggregative features has also been observed. Fortunately, given a specific task, its weak points could be strengthened with the domain knowledge related to the task. In this talk, I will first mention the limitations of DNN-based approaches that we have observed in our two machine reading systems: Math Word Problem Solving, and Social Studies Q&A. Afterwards, various methodologies to enhance the vanilla DNN approach for the given task will be discussed.

◆  Mark Liberman, Professor, University of Pennsylvania

Keynote Topic: Clinical Applications of Human Language Technology: Opportunities and Challenges

Short Bio: Mark Liberman began his career at AT&T Bell Labs in 1975, and moved in 1990 to the University of Pennsylvania, where he is a professor in the departments of Linguistics and Computer and Information Science. He has participated in DARPA's Human Language Technology programs since the mid 1980s. His involvement in the open data movement began with the ACL Data Collection Initiative in the 1980s, and continued with the provision of shared data for DARPA and other HLT programs, and the founding of the Linguistic Data Consortium in 1992. His current research activities include methods for easier development of resources for languages that lack them, and the application of data-intensive linguistic analysis to clinical, educational, and legal issues.

Abstract: We infer a lot from the way someone talks: personal characteristics like age, gender, background, personality; contextual characteristics like mood and attitude towards the interaction; physiological characteristics like fatigue or intoxication. Many clinical diagnostic categories have symptoms that are manifest in spoken interaction: autism spectrum disorder, neurodegenerative disorders, schizophrenia, and so on. The development of modern speech and language technology makes it possible to create automated methods for diagnostic screening and monitoring. More important is the fact that these diagnostic categories are phenotypically diverse, representing (sometimes apparently discontinuous) regions of complex multidimensional behavioral spaces. We can hope that automated analysis of large relevant datasets will allow us to do better science, and learn what the true latent dimensions of those behavioral spaces are. And we can hope for convenient, inexpensive, and psychometrically reliable ways to track neurocognitive health and estimate the efficacy of treatments. I'll present some suggestive preliminary results, and discuss future research opportunities as well as some significant barriers to progress.

◆  Dawei Song, Professor, Beijing Institute of Technology, China

Keynote Topic: A quantum cognitive perspective for information access and retrieval

Short Bio: Professor Dawei Song received his PhD from the Chinese University of Hong Kong in 2000. Before he joined Beijing Institute of Technology in 2018, he was a Professor at Tianjin University, China, and a Professor of Computing at The Open University, UK, since 2012. Prior to these appointments, he worked as a Professor of Computing since 2008 at the Robert Gordon University, UK, where he remains as an Honorary Professor since 2012. He has also worked as a Senior Lecturer and Research Director at the Knowledge Media Institute of The Open University, UK, during 9/2005-10/2008; and as a Senior Research Scientist at the Cooperative Research Centre in Enterprise Distributed Systems Technology, Australia, during 2000-2005. His main research interest is focused on formal models for intelligent search, exploration and discovery over textual and multimodal data sources in a way that is adaptive to user’s context of interaction, and compatible with human cognitive information processing. Particularly in the recent 10 years, he has been driving force of an international joint research agenda on quantum-cognition inspired information retrieval models.

Abstract: Information access and retrieval (IAR) over online information spaces has become a preferred way of acquiring information and useful knowledge relevant to a user’s information need. Central to the process is the user's continuous yet often uncertain sense- and decision- making about information, which result in various non-classical phenomenon that traditional theories turn out insufficient to explain and model. This talk will introduce a quantum cognitive perspective, from which the IAR is modelled with a more general quantum probability framework. I will present some recent work we have done in this area, followed by a discussion on findings, lessons, and future directions.

◆  Fei Xia, Professor, University of Washington

Keynote Topic: NN is great, but NLP is not equal to NN

Short Bio: Fei Xia received her Bachelor's degree from Peking University, and M.S. and Ph.D. from the University of Pennsylvania. After graduation, she worked at the IBM T. J. Watson Research Center before joining UW in 2005. Her past research includes grammar generation, corpus development and machine translation, among others. Her current research focuses on clinical NLP and creating resources for low-resource languages.

Abstract: In the past few years, the NLP field has undergone tremendous changes due to rapid development of Neural Network (NN), and consequently most NLP conferences are dominated by NN publications. While NN often produces state-of-the-art results, it has many limitations, which have been pointed out by many scholars (e.g., most recently by Dr. Ming Zhou at his ACL-2019 presidential address and by Dr. Keh-Yih Su at his upcoming Keynote Speech at NLPCC-2019). In this talk, I will focus on two issues that are rarely addressed by NN; namely, data annotation and the importance of domain experts. I will use clinical NLP as an example to demonstrate that, for many professional domains, creating high-quality labeled data is not trivial and domain knowledge is needed at many stages of system development. I will also briefly introduce a new scheme for annotating conversation structure. Because NN alone cannot solve all the NLP problems, it is crucial for the NLP field not to abandon other fundamental research directions while riding the tide of NN.