DOCSLIB.ORG

Synthesiing Adversarial Negative Responses for Robust Response

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Synthesiing Adversarial Negative Responses for Robust Response Synthesizing Adversarial Negative Responses for Robust Response Ranking and Evaluation Prakhar Gupta| Yulia Tsvetkov♠ Jeffrey P. Bigham|;~ |Language Technologies Institute, Carnegie Mellon University ♠Paul G. Allen School of Computer Science & Engineering, University of Washington ~Human-Computer Interaction Institute, Carnegie Mellon University [email protected], [email protected], [email protected] Abstract to a provided context, consisting of past dialogue turns. Dialogue ranking (Zhou et al., 2018; Wu Open-domain neural dialogue models have et al., 2019) and evaluation models (Tao et al., achieved high performance in response rank- 2018; Yi et al., 2019; Sato et al., 2020), in turn, are ing and evaluation tasks. These tasks are deployed to select and score candidate responses formulated as a binary classification of re- sponses given in a dialogue context, and mod- according to coherence and appropriateness. els generally learn to make predictions based Ranking and evaluation models are generally on context-response content similarity. How- trained using true positive responses and randomly ever, over-reliance on content similarity makes selected negative responses, which raises two is- the models less sensitive to the presence of in- sues. First, random negative candidates often have consistencies, incorrect time expressions and low content similarity with the context, and thus other factors important for response appro- models learn to associate response coherence and priateness and coherence. We propose ap- proaches for automatically creating adversar- appropriateness with content similarity (Yuan et al., ial negative training data to help ranking and 2019; Whang et al., 2021; Sai et al., 2020). In evaluation models learn features beyond con- real systems, generated response candidates tend tent similarity. We propose mask-and-fill and to be more similar in terms of content, and so other keyword-guided approaches that generate neg- factors (e.g., time expressions, dialogue acts, in- ative examples for training more robust dia- consistencies) tend to be more important. Second, logue systems. These generated adversarial re- randomly selecting candidates as negative exam- sponses have high content similarity with the ples in an open domain context can result in false contexts but are either incoherent, inappropri- ate or not fluent. Our approaches are fully negatives, leading to misclassification of appropri- data-driven and can be easily incorporated in ate responses. existing models and datasets. Experiments To make dialogue models more robust to the on classification, ranking and evaluation tasks spurious pattern of content similarity, prior work across multiple datasets demonstrate that our proposed to leverage adversarial and counterfactual approaches outperform strong baselines in pro- examples (Kaushik et al., 2020; Srivastava et al., viding informative negative examples for train- 2020). A reliable method for creating counterfac- ing dialogue systems.1 tual data is to collect human-written adversarial 1 Introduction negative responses (Sai et al., 2020), but it is expen- sive, time-consuming, and difficult to scale. Our Due to growing availability of dialogue corpora (Li goal is to create reliable automatic methods for et al., 2017; Zhang et al., 2018; Smith et al., 2020) synthesizing adversarial negative responses. and the advancement of neural architectures (Rad- The most common approach to generating natu- ford et al., 2019; Brown et al., 2020; Devlin et al., ral language adversarial examples is to paraphrase 2019), dialogue systems have achieved consider- or insert typos, synonyms, or words relevant to the able success. As typically formulated, dialogue context in the inputs (Iyyer et al., 2018; Ebrahimi models generate one or more candidate responses et al., 2018; Alzantot et al., 2018; Zhang et al., 2019). In open domain conversations, however, a 1Code and data are publicly available https: //github.com/prakharguptaz/Adv_gen_ context can have a wide range of possible responses dialogue with varied forms and semantics. Small lexical Error category Description Sample responses C-ent Incorrect entities Incorrect subject or object of Context: I am so happy that you are doing okay. or actors (R,G) verbs or presence of one or more Response: My friend is always happy. incorrect entities or coreference. C-time Incorrect Time Use of incorrect time expressions Context: What are you going to do on Monday? expressions (R) or tense of verbs. Response: Yesterday, I celebrated my daughter’s wed- ding anniversary. C-cont Contradictory Presence of details which make Context: A: I don’t know why I bothered to come or extraneous the response inconsistent within here. details (R,G) itself or contradict the context B: Did you enjoy your stay? Response: I enjoyed the concert a lot. C-speaker Incorrect speaker The response is relevant to the Context: What starting salary would you expect here? turn (R) conversation but from the wrong Response: If you work overtime, I will pay you extra speaker. salary. C-follow Does not directly The response does not follow im- Context: What would you like for main course sir? address the con- mediately from the context. Response: I know very well how to make noodles, text (R,G) and I taught one of my friends. C-strat Incorrect strate- Use of incorrect dialogue act, Context: I can’t find the paper clips. gies (R,G) emotion, persona or style Response: Ok, great work. C-lang Poor language Presence of poor grammar, incor- Context: Do you have mixed drinks available here? (G) rect sentence structures or repeti- Response: Yes. This order is divided by 16 divided tions for main main ones of order. Table 1: Error categories prevalent in inappropriate responses with high context-response semantic relatedness. We present 7 categories with their descriptions and sample context and response pairs. For each category we also indicate whether it is frequently observed in Retrieval (R) or Generation (G) models. Models which simply learn to associate response coherence with content similarity often ignore these errors. Our approaches create adversarial negative data for training dialogue models by introducing such errors in context relevant utterances. variations via substitutions and paraphrasing do not 2 Properties of Adversarial Responses provide adequate coverage over the possible space Models trained using randomly sampled negative of adversarial responses, and they can also lead to examples tend to assign high scores to responses generation of false negatives due to the open-ended with high content similarity with the context, and nature of dialogues. Creating adversarial dialogue often ignore other important factors necessary for responses is thus different, and can be more chal- response appropriateness and coherence. There- lenging than in other natural language domains. fore, we aim to generate adversarial negative re- We propose two approaches for adversarial re- sponses which have high content similarity with sponse creation: 1) a mask-and-fill approach that the context, but which still possess factors render- corrupts gold responses related to the context but ing the responses inappropriate to the context. We retains content similarity, and 2) a keyword-guided present the categorization of such factors or error generative approach that uses concepts from the types which can make a response inappropriate context to generate topically relevant but incoher- in Table1. For each category, we provide its de- ent responses. These approaches do not require scription and sample context-response pairs. To additional annotations, are black-box (do not need create this categorization, we manually analyzed access to model parameters), and are easily adapted responses present in outputs of generative models, to new datasets and domains. candidates of retrieval sets, and human written ad- The main contributions of this paper are: 1) We versarial dialogue responses (Sai et al., 2020). Cat- identify and discuss error patterns present in re- egories C-ent, C-time and C-cont are errors related trieval and generation model outputs, which are to various inconsistencies and logical flaws in the difficult to detect due to high content similarity; 2) responses and indicate poor response appropriate- To the best of our knowledge, we are the first to ness. Categories C-speaker, C-follow and C-strat propose automatic approaches for creating adver- are error types specific to the dialogue setting and sarial responses for dialogue model training in a indicate poor response coherence. Category C-lang black-box setting; and, 3) We demonstrate that our indicates poor response fluency. Our categorization proposed approaches achieve better performance of errors is inspired by the categorization suggested compared to strong baselines on two datasets on di- by Pagnoni et al.(2021) for factuality of summa- alogue classification, ranking and evaluation tasks. rization, and Higashinaka et al.(2019); Ko et al. (2019) and Sato et al.(2020) for dialogue. These categories inform our approaches as well as error analysis. 3 Methodology For a given dialogue context C and its gold re- sponse Rg, our goal is to generate an adversar- ial response Ra such that while achieving high Figure 1: Mask-and-fill approach using ILM model. scores from dialogue ranking or evaluation models, ILM is trained to infill n-grams in place of blanks in it should not be a valid response to the context C. a response. Tokens after [infill] replace the [blank] Dialogue
Load more

Recommended publications
  • Originally, the Descendants of Hua Xia Were Not the Descendants of Yan Huang
    E-Leader Brno 2019 Originally, the Descendants of Hua Xia were not the Descendants of Yan Huang Soleilmavis Liu, Activist Peacepink, Yantai, Shandong, China Many Chinese people claimed that they are descendants of Yan Huang, while claiming that they are descendants of Hua Xia. (Yan refers to Yan Di, Huang refers to Huang Di and Xia refers to the Xia Dynasty). Are these true or false? We will find out from Shanhaijing ’s records and modern archaeological discoveries. Abstract Shanhaijing (Classic of Mountains and Seas ) records many ancient groups of people in Neolithic China. The five biggest were: Yan Di, Huang Di, Zhuan Xu, Di Jun and Shao Hao. These were not only the names of groups, but also the names of individuals, who were regarded by many groups as common male ancestors. These groups first lived in the Pamirs Plateau, soon gathered in the north of the Tibetan Plateau and west of the Qinghai Lake and learned from each other advanced sciences and technologies, later spread out to other places of China and built their unique ancient cultures during the Neolithic Age. The Yan Di’s offspring spread out to the west of the Taklamakan Desert;The Huang Di’s offspring spread out to the north of the Chishui River, Tianshan Mountains and further northern and northeastern areas;The Di Jun’s and Shao Hao’s offspring spread out to the middle and lower reaches of the Yellow River, where the Di Jun’s offspring lived in the west of the Shao Hao’s territories, which were near the sea or in the Shandong Peninsula.Modern archaeological discoveries have revealed the authenticity of Shanhaijing ’s records.
    [Show full text]
  • Face Recognition on Long-Tailed Domains
    Domain Balancing: Face Recognition on Long-Tailed Domains Dong Cao1;2∗ Xiangyu Zhu1;2∗ Xingyu Huang3 Jianzhu Guo1;2 Zhen Lei1;2† 1CBSR & NLPR, Institute of Automation, Chinese Academy of Sciences, Beijing, China 2School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100049, China 3Tianjin University fdong.cao,xiangyu.zhu,jianzhu.guo,[email protected], [email protected] Abstract Long-tailed problem has been an important topic in face recognition task. However, existing methods only concen- trate on the long-tailed distribution of classes. Differently, we devote to the long-tailed domain distribution problem, which refers to the fact that a small number of domains fre- quently appear while other domains far less existing. The key challenge of the problem is that domain labels are too complicated (related to race, age, pose, illumination, etc.) Figure 1. The long-tailed domain distribution demarcated by the and inaccessible in real applications. In this paper, we pro- mixed attributions (e.g., race and age) in the MS-Celeb-1M [8] pose a novel Domain Balancing (DB) mechanism to han- and CASIA-Webface [36]. Number of classes per domain falls dle this problem. Specifically, we first propose a Domain drastically, and only few domains have abundant classes. (Baidu Frequency Indicator (DFI) to judge whether a sample is API [1] is used to estimate the race and age) from head domain or tail domain. Secondly, we formu- late a light-weighted Residual Balancing Mapping (RBM) generalization, i.e., the learned features only work well on block to balance the domain distribution by adjusting the the domain the same as the training set and perform poorly network according to DFI.
    [Show full text]
  • Download File
    On the Periphery of a Great “Empire”: Secondary Formation of States and Their Material Basis in the Shandong Peninsula during the Late Bronze Age, ca. 1000-500 B.C.E Minna Wu Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Graduate School of Arts and Sciences COLUMIBIA UNIVERSITY 2013 @2013 Minna Wu All rights reserved ABSTRACT On the Periphery of a Great “Empire”: Secondary Formation of States and Their Material Basis in the Shandong Peninsula during the Late Bronze-Age, ca. 1000-500 B.C.E. Minna Wu The Shandong region has been of considerable interest to the study of ancient China due to its location in the eastern periphery of the central culture. For the Western Zhou state, Shandong was the “Far East” and it was a vast region of diverse landscape and complex cultural traditions during the Late Bronze-Age (1000-500 BCE). In this research, the developmental trajectories of three different types of secondary states are examined. The first type is the regional states established by the Zhou court; the second type is the indigenous Non-Zhou states with Dong Yi origins; the third type is the states that may have been formerly Shang polities and accepted Zhou rule after the Zhou conquest of Shang. On the one hand, this dissertation examines the dynamic social and cultural process in the eastern periphery in relation to the expansion and colonization of the Western Zhou state; on the other hand, it emphasizes the agency of the periphery during the formation of secondary states by examining how the polities in the periphery responded to the advances of the Western Zhou state and how local traditions impacted the composition of the local material assemblage which lay the foundation for the future prosperity of the regional culture.
    [Show full text]
  • The Geographical Names for the Sea of . ,Japan in Chinese Historical Documents
    .... The Geographical Names for the Sea of . ,Japan in Chinese Historical Documents Song-di Wu (The Institute of Chinese C'xeography in Fudan University, Shanghai, China) The Sea of Japan, surround by North Korea, South Korea, Japan and Russia. IS an important sea in tile Northeast Asia. Before the Second Opium War(1854-1860), the lo\ver reachers of the Heilongjiang River, the Wusulijiang River valley and Sakhalin of now Russia were historical documents some records of how the present Sea of Japan was called during various historical times. These records will provide some helpful clues to the discussion of the naming problem of the Sea of Japan today. I. The first Chinese historical document which mentioned tile Sea of Japan was Dong Yi Lie ZhuanO;j[~J!Jfi) of Hou Han Shu(13t7:l~), which recorded Yi Lou(m:I:), which was the kingdom set up by the ancient Su ShenOlitJ)i) tribe, was more than 1,000 Ii northeast of Fu YU(7-:~); it faced the sea on the east and was bounded by the north Wo JuWi:illJ kingdom. Wo Ju was located in the northeast of tile Korean Peninsula, Fu Yu was in the central part of the Northeast Plain of China: Yi Lou was in the \Vusulijiang River valley and the lower reaches of the Heilongjiang River which now belong to Russia, so "the sea" Yi Lou faced on tile east could only be the northern area of the Sea of Japan today. It was also recorded in the same book that. to the north of the Wei kingdom \vere Gao Gou Li and Wo Ju, to tile south of it was Chen Han, to the east was "the ocean" and to the west was Le Lang.
    [Show full text]
  • Questions and Answers in Chinese Political Press Conferences
    Questions and Answers in Chinese Political Press Conferences A sub-thesis submitted for the degree of Master of Applied Linguistics of the Australian National University Xujia Du October, 2011 Declaration To the best of my knowledge, this thesis represents my own original research unless otherwise acknowledged in the text. Xujia Du October, 2011 Acknowledgements I would like to express my gratitude to a multitude of wonderful people who helped me and supported me throughout my study and research at ANU. First and foremost, I would like to thank my supervisor, Dr. Johanna Rendle-Short. With her expertise in CA and sociolinguistics, she made many insightful and useful comments on the drafts of this thesis. More importantly, I have learned from Johanna, to start early and to be more organized, which will continuously benefit me in my future work and life. I would also like to express my deepest gratitude to my parents, for their unconditional love throughout my life and unfailing support for my study at ANU. A special thank you also goes to my aunt and cousin, who have been very concerned about my life and study in Australia. I would like to thank my friends, Ran Li, Kun-long Liu, Ruriko Otomo, Eriko Toma and various members of the Discourse Analysis Group (DAG) for their valuable comments and many discussions that helped me develop ideas and march toward the conclusion. Thanks are also due to my friends, Xin Xin, Xiang Li, and Jie Li for their company and encouragement during my thesis writing. i Abstract Since China’s opening up in 1978, there has been increasing interaction between the Chinese government and the domestic and international media.
    [Show full text]
  • Between Heaven and Earth: Dual Accountability in Han China
    Article Chinese Journal of Sociology 2015, Vol. 1(1) 56–87 ! The Author(s) 2015 Between heaven and Reprints and permissions: sagepub.co.uk/journalsPermissions.nav earth: Dual accountability DOI: 10.1177/2057150X14568768 in Han China chs.sagepub.com Miranda Brown and Yu Xie Abstract Scholars have noticed that centrally-appointed officials in imperial China were not only beholden to their superiors but also acted as brokers of local interests. We characterize such a structural position as ‘dual accountability’. Although accountability to superiors is readily understandable within the Weberian framework of bureaucratic hierarchy, the reasons behind local responsiveness bear explanation. This paper attempts to explain such responsiveness by investigating the larger ideological, structural, and institutional contexts of the Han dynasty (206 BCE–220 CE). We explore two existing explanations – practical necessity and ‘Confucian’ or classical paternalism – and add a new explan- ation of our own: the emphasis on virtuous reputations in the system of bureaucratic recruitment and promotion. Our argument is supported by empirical evidence from a range of sources, including administrative records and inscriptions on ancient stelae. More generally, we question Weber’s hypothesis that the Chinese imperial system of administration fit the ideal type of traditional bureaucracy, and we examine the rational bases underlying an ‘inefficient’ system that was in place for two millennia. Keywords dual accountability, localism, bureaucracy A field commander must decide even against king’s orders. ( ) (Chinese proverb) Department of Asian Languages and Cultures, University of Michigan, Ann Arbor, USA Corresponding author: Miranda Brown, Department of Asian Languages and Cultures, University of Michigan, 202 S.
    [Show full text]
  • Ceramic Tableware from China List of CNCA‐Certified Ceramicware
    Ceramic Tableware from China June 15, 2018 List of CNCA‐Certified Ceramicware Factories, FDA Operational List No. 64 740 Firms Eligible for Consideration Under Terms of MOU Firm Name Address City Province Country Mail Code Previous Name XIAOMASHAN OF TAIHU MOUNTAINS, TONGZHA ANHUI HANSHAN MINSHENG PORCELAIN CO., LTD. TOWN HANSHAN COUNTY ANHUI CHINA 238153 ANHUI QINGHUAFANG FINE BONE PORCELAIN CO., LTD HANSHAN ECONOMIC DEVELOPMENT ZONE ANHUI CHINA 238100 HANSHAN CERAMIC CO., LTD., ANHUI PROVINCE NO.21, DONGXING STREET DONGGUAN TOWN HANSHAN COUNTY ANHUI CHINA 238151 WOYANG HUADU FINEPOTTERY CO., LTD FINEOPOTTERY INDUSTRIAL DISTRICT, SOUTH LIUQIAO, WOSHUANG RD WOYANG CITY ANHUI CHINA 233600 THE LISTED NAME OF THIS FACTORY HAS BEEN CHANGED FROM "SIU‐FUNG CERAMICS (CHONGQING SIU‐CERAMICS) CO., LTD." BASED ON NOTIFICATION FROM CNCA CHONGQING CHN&CHN CERAMICS CO., LTD. CHENJIAWAN, LIJIATUO, BANAN DISTRICT CHONGQING CHINA 400054 RECEIVED BY FDA ON FEBRUARY 8, 2002 CHONGQING KINGWAY CERAMICS CO., LTD. CHEN JIA WAN, LI JIA TUO, BANAN DISTRICT, CHONGQING CHINA 400054 BIDA CERAMICS CO.,LTD NO.69,CHENG TIAN SI GE DEHUA COUNTY FUJIAN CHINA 362500 NONE DATIAN COUNTY BAOFENG PORCELAIN PRODUCTS CO., LTD. YONGDE VILLAGE QITAO TOWN DATIAN COUNTY CHINA 366108 FUJIAN CHINA DATIAN YONGDA ART&CRAFT PRODUCTS CO., LTD. NO.156, XIANGSHAN ROAD, JUNXI TOWN, DATIAN COUNTY FUJIAN 366100 DEHUA KAIYUAN PORCELAIN INDUSTRY CO., LTD NO. 63, DONGHUAN ROAD DEHUA TOWN FUJIAN CHINA 362500 THE LISTED ADDRESS OF THIS FACTORY HAS BEEN CHANGED FROM "MAQIUYANG XUNZHONG XUNZHONG TOWN, DEHUA COUNTY" TO THE NEW EAST SIDE, THE SECOND PERIOD, SHIDUN PROJECT ADDRESS LISTED ABOVE BASED ON NOTIFICATION DEHUA HENGHAN ARTS CO., LTD AREA, XUNZHONG TOWN, DEHUA COUNTY FUJIAN CHINA 362500 FROM THE CNCA AUTHORITY IN SEPTEMBER 2014 DEHUA HONGSHENG CERAMICS CO., LTD.
    [Show full text]
  • 4Th International Conference on Energy and Environmental Protection
    4th International Conference on Energy and Environmental Protection (ICEEP 2015) Shenzhen, China 2 – 4 June 2015 Volume 1 of 7 ISBN: 978-1-5108-3756-0 Printed from e-media with permission by: Curran Associates, Inc. 57 Morehouse Lane Red Hook, NY 12571 Some format issues inherent in the e-media version may also appear in this print version. Copyright© (2015) by DEStech Publications, Inc. All rights reserved. Printed by Curran Associates, Inc. (2017) For permission requests, please contact DEStech Publications, Inc. at the address below. DEStech Publications, Inc. 439 North Duke Street Lancaster PA 17602-4967 USA Phone: (717) 290-1660 Fax: (717) 509-6100 [email protected] Additional copies of this publication are available from: Curran Associates, Inc. 57 Morehouse Lane Red Hook, NY 12571 USA Phone: 845-758-0400 Fax: 845-758-2633 Email: [email protected] Web: www.proceedings.com Table of Contents Preface Committees The Design of the Multisensor Monitoring Device Based on STM32F103RB and GPS for the Elderly . 1 CHAO WANG and PENGCHENG LIU The Simulation of the Underground Pressure about the Gob-side Entry Retaining . .7 JIANJUN SHI and QIFENG ZHAO Study on the Properties of Hot Spot-resistant Components . 12 XIANGSAI FENG, HONGQIAO QIAO, HAILEI ZHANG, YINBIN TANG, CHONG WANG, CHUANMING XU and XINKAN ZHAO Variable Frequency Control Simulation for Ground Source Heat Pump System Based on TRNSYS . 16 HUI LI and HAOGANG YANG The Impact of Changes in Industrial Structure of the Yangtze River Economic Zone on Energy Consumption . 21 GUIYAN SUN and CHUANSHENG WANG Frequency Control Strategy of Islanding Microgrid Based on Capacity Detection of Battery.
    [Show full text]
  • College of Engineering—Fall 2018 Dean's Honor List
    College of Engineering—Fall 2018 Dean’s Honor List Page 1 Heba Abdallah Brianna Andress Anjali Balani Catherine Bertcher Natalie Bower Hossein Abdollahi Eric Andrews Shreya Balusu Christopher Bessler Margaret Bowers Samina Abdullah Ken Gin Ang Katherine Banas Sebastian Betancourt Marlina Bowring Austin Abdun-Nabi Frank Angers Daniel Banooni Kameron Betz Kevin Boysen Evan Abrams Arun Annamalai Anil Bansal Cameron Beversluis Emily Bozich James Abrams Hariprasad Annamalai Ashish Bansal Farhan Bhagat Gabrielle Bracken Michael Adams Sara Anstey Dylan Bao Ved Bhagwat Sutton Bradley George Adamson Geet Antani Jiajun Bao Karthik Bhandarkar Samantha Braham Kashyap Addanki Alexander Anthony Kevin Bao Rajiv Bharadwaj Alec Brandel Shriharimurt Aaron Adiwidjaja Albert Anwar Yifei Bao Bhaskaramurthi Casey Brantner Eytan Adler Kirsten Apel Ari Baranian Tanya Bhatia Enrico Braucher Anselm Adrian Anak Christopher Adolfo Apolloni Benjamin Baranski Aditya Bhatt Margaret Braunreuther Medha Adusumilli Michael Apone Adelaide Barcalow Jonathan Bi Samuel Brause Youssef Afify Justin Applefield Alexander Bardha Justin Bi Timothy Breckwoldt Aadhar Agarwal Jacob Applegate Nadya Barghouty Duoming Bian Glen Bredin Arav Agarwal Mitchell Appleton Nadim Bari Corey Bicknell Melissa Brei Arushi Agarwal Ahmed Nafis Arafat Christopher Barkey Michael Biek Benjamin Brengman Varun Agarwal Devin Ardeshna Jack Barlow Rachel Bielski Joseph Brenner Ashok Aggarwal John Arias Kammeran Barnes Grace Biermacher Jacob Brink Parth Aggarwal Ryuji Arimoto Marissa Barnes Matthew Biggerman
    [Show full text]
  • Chinese Spoken Language Processing
    Qiang Huo Bin Ma Eng-Siong Chng Haizhou Li (Eds.) Chinese Spoken Language Processing 5th International Symposium, ISCSLP 2006 Singapore, December 13-16, 2006 Proceedings Springer Table of Contents Plenary Interactive Computer Aids for Acquiring Proficiency in Mandarin 1 Stephanie Seneff The Affective and Pragmatic Coding of Prosody 13 Klaus R. Scherer Challenges in Machine Translation 15 Franz Josef Och Automatic Indexing and Retrieval of Large Broadcast News Video Collections - The TRECVID Experience 16 Tat-Seng Chua Tutorial An HMM-Based Approach to Flexible Speech Synthesis 17 Keiichi Tokuda Text Information Extraction and Retrieval 18 Hang Li Topics in Speech Science Mechanisms of Question Intonation in Mandarin 19 Jiahong Yuan Comparison of Perceived Prosodic Boundaries and Global Characteristics of Voice Fundamental Frequency Contours in Mandarin Speech 31 Wentao Gu, Keikichi Hirose, Hiroya Fujisaki Linguistic Markings of Units in Spontaneous Mandarin 43 Shu-Chuan Tseng Phonetic and Phonological Analysis of Focal Accents of Disyllabic Words in Standard Chinese 55 Yuan Jia, Ziyu Xiong, Aijun Li XVIII Table of Contents Focus, Lexical Stress and Boundary Tone: Interaction of Three Prosodic Features 67 Lu Zhang, Yi-Qing Zu, Run-Qiang Yan Speech Analysis A Robust Voice Activity Detection Based on Noise Eigenspace Projection 76 Dongwen Ying, Yu SM, Frank Soong, Jianwu Dang, Xugang Lu Pitch Mean Based Frequency Warping 87 Jian Liu, Thomas Fang Zheng, Wenhu Wu A Study of Knowledge-Based Features for Obstruent Detection and Classification in Continuous Mandarin Speech 95 Kuang-Ting Sung, Hsiao-Chuan Wang Speaker-and-Environment Change Detection in Broadcast News Using Maximum Divergence Common Component GMM 106 Yih-Ru Wang UBM Based Speaker Segmentation and Clustering for 2-Speaker Detection 116 Jing Deng, Thomas Fang Zheng, Wenhu Wu Design of Cubic Spline Wavelet for Open Set Speaker Classification in Marathi 126 Hemant A.
    [Show full text]
  • Multi-Turn Response Selection for Chatbots with Deep Attention Matching Network
    Multi-Turn Response Selection for Chatbots with Deep Attention Matching Network Xiangyang Zhou∗, Lu Li,∗ Daxiang Dong, Yi Liu, Ying Chen, Wayne Xin Zhao,y Dianhai Yu and Hua Wu Baidu Inc., Beijing, China nzhouxiangyang, lilu12, dongdaxiang, liuyi05,o @baidu.com chenying04, v zhaoxin, yudianhai, wu hua Abstract (Lowe et al., 2017) and the discriminator of GAN- based (Generative Adversarial Networks) neural Human generates responses relying on se- dialogue generation (Li et al., 2017). mantic and functional dependencies, in- cluding coreference relation, among dia- Conversation Context logue elements and their context. In this Speaker A: Hi I am looking to see what packages are installed on my system, I don’t see a path, is the list being held somewhere else? paper, we investigate matching a response Speaker B: Try dpkg - get-selections with its multi-turn context using depen- Speaker A: What is that like? A database for packages instead of a flat file structure? dency information based entirely on atten- Speaker B: dpkg is the debian package manager - get-selections simply shows tion. Our solution is inspired by the re- you what packages are handed by it cently proposed Transformer in machine Response of Speaker A: No clue what do you need it for, its just reassurance translation (Vaswani et al., 2017) and we as I don’t know the debian package manager extend the attention mechanism in two Figure 1: Example of human conversation on Ubuntu sys- ways. First, we construct representations tem troubleshooting. Speaker A is seeking for a solution of of text segments at different granularities package management in his/her system and speaker B recom- solely with stacked self-attention.
    [Show full text]
  • The Dong-Yi People
    E-Leader Vienna 2016 The Dong-Yi People Soleilmavis Liu, Author, Board Member and Peace Sponsor Yantai, Shangdong, China Abstract The Dong-Yi People (Dong in Chinese means east) lived in the Shandong Peninsula in the Neolithic Age. There they built one of the most important Neolithic cultures, which later spread to the lower reaches of the Yellow and Huai rivers. Its latter stage, the Longshan Culture (about 3200BCE-1900BCE), spread to the areas of early Di-Qiang Culture, another Chinese Neolithic culture that originated from the middle reaches of the Yellow River, and turned those areas into outposts of Longshan Culture. Thus Dong-Yi Culture greatly influenced ancient China and had the leading role in making the Yellow River Valley Culture the root of Chinese civilization. The Dong-Yi People also migrated to the Americas and Oceania in the Neolithic Age, where their culture had great influence. The ancient civilizations of Oceanic cultures, such as palae-Polynesian, palae-Melanesian and palae-Micronesian cultures; and American Indians civilizations, such as the Mayan (about 2000BCE-900CE), the Aztec (about 12th century - 15th century CE) and the Incan (about 13th century - 15th century CE) civilizations, all evolved from early Dong-Yi Culture. This article briefly introduces certain historical records of the Dong-Yi People, including their origins, their history of cultivating wheat, their worship of bird totems, their relationship with other groups of Neolithic people, their racial characteristics, their migrations and the overall influence of Dong-Yi Culture upon subsequent communities. In the book “The Queen of the South in Matthew 12:42” written by Soleilmavis, there are more details about the Dong-Yi People, Dong-Yi Culture and how they influenced ancient civilizations of China, the Americas and Oceania.
    [Show full text]