Statement of Purpose / Yun-Nung Chen
http://VivianChen.idv.tw
Language, used every day, is the easiest, fastest and most direct method for communication; I believe that the application of artificial intelligence (AI) to spoken language processing will yield unprecedented convenience. My vision for the future of spoken language processing is to build an environment that makes it possible for everyone in the world to interact with computers via natural spoken language.
Striving for this dream, I have worked hard on my research and technical skills. My paper on key term extraction was accepted for the Workshop on Spoken Language Technologies (SLT 2010, Student Best Paper Award), and as an undergraduate I maintained top academic performance, ranking 3rd out of 100 students in the Department of Computer Science and Information Engineering at National Taiwan University. Under the supervision of Prof. Lin-Shan Lee, I did in-depth research on speech processing topics including information extraction, speech information retrieval, and automatic summarization. With this solid foundation, I believe I am ready to take the next step at your university.
When I was still an undergraduate, I had already started to do research on key term extraction. Key terms are important especially when browsing a large number of documents, because they can help us to quickly find the parts we really need to read. In our work, we first constructed a tree structure for which we used word sequence entropy to decide key phrases, and then applied machine learning methods to extract keywords using various features, such as prosody and semantics. We believe the resultant key term list and the relationships between each term constitute the most important information in the lectures because it allows people to more efficiently grasp the course content: this explains the importance of this research. Moreover, the achievements of this work were presented in a paper, “Automatic Key Term Extraction from Spoken Course Lecture Using Branching Entropy and Prosodic/Semantic Features”, which was accepted for presentation at SLT 2010, a major conference on speech and spoken language processing. The paper was awarded the Student Best Paper Award and selected for presentation in an oral session; this recognition at such a top speech conference was a great encouragement for me, because it confirmed my exceptional research abilities.
We also designed a novel algorithm to improve the performance of information retrieval on speech data (spoken term detection). After performing conventional spoken term detection, we constructed a graph using the similarity between each pair of retrieved utterances, and used this graph to re-rank the utterances to improve the performance of the retrieved results. The construction of this graph draws on the Random Walk algorithm, and yields higher relevance scores for relevant utterances. The results were submitted as the paper “Improved Spoken Term Detection with Graph-Based Re-Ranking in Feature Space” to the International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2011. This method helps people to successfully retrieve parts of spoken documents, and even supports spoken queries.
I also conducted research on automatic summarization with semantic analysis using probabilistic latent semantic analysis (PLSA) and key terms; this research yielded improved summarization performance. In the research for both of these spoken language-related problems, I drew on my knowledge of algorithms and AI. I strongly believe that spoken language processing technology is very important for human life, and can elevate the quality of life and general convenience for the whole world.
I have completed intensive coursework in programming, mathematics, and statistics, and have participated in many projects in advanced courses; this has honed my implementation skills and sharpened my thinking. I have even passed three of the PhD qualification exams: algorithms, artificial intelligence, and computational theory. Thus I have acquired extensive experience in different areas of computer science and have demonstrated the ability to integrate these areas in research.
Beside my passion for and achievements in academic research, I have also participated in many extracurricular activities. Because of my recognized leadership skills, I was elected as the valedictorian and the graduate student representative. These experiences have made me not only a better communicator but have also enhanced my ability to organize a large number of diverse opinions. I also served as multimedia president for the pop dance club, which involved designing programs and editing music and videos. In this work I gained a greater understanding of the vast potential computers have to help humans, and the convenience that they can bring to the world. As spoken language is the most convenient form of communication, speech-related research shows the most promise in enabling computers to understand human thought and to further aid humans.
I believe that language analysis and understanding is very important. Motivated by my previous work in information extraction and graph algorithms, I would like to integrate the two and apply graph algorithms to represent spoken language and pinpoint the structure of semantic relationships; such an integration can allow us to better understand language. For example, analyzing a lecture and laying out a clear structure for it can help people to quickly summarize the whole lecture; computers can thus help people understand things more quickly. Also, as my native language is Chinese, I have the ability to process multilingual spoken language, and am familiar with phenomena such as code-mixing and code-switching, which are frequently used in Asian countries. Moreover, I hope to implement language technology on mobile devices to provide greater convenience for all people. People need only speak, and they can enjoy the convenience that technology provides and access the information they want, all without hand operations. It is crucial to integrate the various technology from different fields such as machine learning and algorithms into spoken language processing; I am confident in my ability to apply this knowledge and make further progress. Because the contributions in many areas are different, integrating them can yield greater improvements. With successful spoken language processing, we can do everything using only speech – this is what drives me in my research.
Passion stirs up innovation; I am a computer scientist and possess a strong ambition and an enthusiasm for research, and I am confident that my considerable creativity and extensive academic knowledge will help me conduct valuable research, and accelerate our transition to an information-based society. It will certainly take the field of AI a great step forward and benefit the human race as a whole. As for the future, I intend to conduct more in-depth research in spoken language technology and thereby devote myself to benefitting the world with advanced technology.
I am a skilled student of computer science as well as a creative thinker and doer, and I strongly believe that your university is the academic institution best suited to helping me realize my dream. My exceptional academic knowledge and research acumen will ensure success as I pour my energies into this field to serve the world. I am well prepared to study in your esteemed university and am the perfect fit for your team.
