Curriculum Vitae

I have been working on a number of research and development projects. I am investigating semi-supervised techniques for word sense disambiguation. Using discourse analysis to analyze threaded discussions to highlight more informative posts, provide summaries of large discussions, and develop reputations systems that reward a diversity of thoughts rather than groupthink. Finally, I am also investigating techniques in sports analytics with an emphasis on Fantasy team management and applications.

My research at the Institute for Creative Technologies focused on using natural language processing to analyze narrative discourse. One aspect of this research is to develop techniques for rapidly authoring interactive stores and driving these simulations based on models learned from real-world narratives. We also investigated how narrative discourse structure of personal stories affects attitude change in individuals regardless of the content of the message.

(PI: Marilyn Walker)
Most automated analysis of discourse related to persuasion and argumentation focuses on the propositional content of the discussion. In contrast to this work, we are examining the role subjective language plays in they way discussion evolve in online dialogs. We are exploring several types of subjective language such as sarcasm, stance and opinion versus fact. We are investigating the nature of how people perceive this language and the reliability of judgments on these kinds of tasks. Guided by our users studies we are using machine learning to automatically identify this language in posts for further large-scale analyses.

(PI: Michael Mateas)
Most automated analysis of discourse related to persuasion and argumentation focuses on the propositional content of the discussion. In contrast to this work, we are examining the role subjective language plays in they way discussion evolve in online dialogs. We are exploring several types of subjective language such as sarcasm, stance and opinion versus fact. We are investigating the nature of how people perceive this language and the reliability of judgments on these kinds of tasks. Guided by our users studies we are using machine learning to automatically identify this language in posts for further large-scale analyses.

(PI: Arnav Jhala)
The SIREN project is a large European Union grant with partners in over 5 countries that are developing educational games to help children improve the way they handle conflicts. My primary focus on this project is to leverage current psychological theories of interpersonal conflicts to create natural and dynamic character models by blending large-scale data driven approaches with hand authored methods.

At Disney I developed technologies for supporting several types of computer aided interactive storytelling applications. In particular, researching the latest classical, probabilistic and case-based planning techniques to be used for automated show control, visualization strategies for improving authoring efficacy and efficiency for interactive stories, and large scale ontology and semantic relationship extraction from the web to improve the scope of knowledge our system could reason about.

I developed an open-domain, interactive narrative generation system, where a person and computer take turns writing sentences of an emerging fictional narrative. A combination of information retrieval, machine learning and natural language processing enable the system to return sentences that produce stories nearly indistinguishable from weblog stories written entirely by humans. Although primarily conceived in the entertainment domain, it also has several other promising applications, such as a tool for children’s education, an automated improvisational exercise, a writer’s aid or general inference mechanism.

I developed a specialized search engine for retrieving first person narrative stories. As part of the project a story classification program was developed to identify narrative sentences using a Support Vector Machine classifier and a Gaussian filter. A custom retrieval engine was used to find relevant stories from a large collection gathered using Google Blog Search and the story classifier. To find relevant stories the retrieval engine used a vector space model with lexical, part-of-speech and semantic role label features extracted from a custom SRL system.

In addition to the natural language processing that is implicitly necessary to support my thesis project I have also explicitly explored several core NLP topics. These include techniques for domain adaptation in semantic role labeling, applying unsupervised linguistic parsing techniques to other structured data such as music and I have also explored syntactic language models for automatic speech recognition.

As part-of a research project in our lab to create several resources around a large- scale ontology of common-sense psychology, I helped organize data collection sessions for gathering natural language expressions capturing concepts in this ontology. The data was then used to author finite state transducers capable of recognizing and tagging these concepts in English text. In addition to collecting natural language expressions for these theories I also tried to validate the theory of Memory, one of Gordon & Hobbs core common-sense theories. The validation was performed by proving several common-sense strategies for Memory storage and retrieval using an automatic theorem solver and a first-order representation of the theory and strategies.

During this internship I designed a mixed reality visualization system for aiding in the development of real-world large-scale semantic networks used in IBM’s Semantic Engine. The visualization used clustering techniques to isolate semantically similar sub-groups of the network. Each sub-group was arranged independently using a standard 2D graph layout algorithm. These 2D clusters were then mapped onto the 3D surface of a globe using an API for Google Earth. The inherent shape of the earth naturally emphasized relevant information by occluding less similar parts of the network beyond the horizon and the resulting interface made inspecting and traversing graphs of thousands of nodes and links easier through clear visual feedback. The interface could be traversed normally using the standard Google Earth client, but also through the manipulation of a physical globe in the real world, using RFID tags to track the position of the globe.

This is a graduate course discussing current and foundational work in special topics related to computational models of discourse and dialogue. This quarter focused on several deep semantic contrasted with statistical representations of narrative discourse.
https://courses.soe.ucsc.edu/courses/cmps179/Spring13/02

This course provides an introduction to the theory and practice of natural language processing. The course was primarily run by Marilyn Walker, however I helped build the syllabus, design the assignments and ran about half of the lectures.
http://courses.soe.ucsc.edu/courses/cmps143/Spring14/01/pages/nlpsyllabus

A basic introduction to basic data structures and algorithms. Teaches fundamental data structures and algorithms in C and Java. Topics include big O notation, linked lists, stacks, queues, binary search trees, hash tables and sorting. Course material largely provided by previous instructor.

This course explored techniques for interactive data visualization using new vector and 3D graphics capabilities of HTML 5. The course emphasized both technical and creative aspects of interactive visualization through the development of several projects using real world data acquired from eBay’s public API.
https://courses.soe.ucsc.edu/courses/cmps179/Spring13/02

This was a seminar course that explored several theoretical and computational models of discourse. Each week students read several papers and presented the reading to the rest of the class. In addition to several homework assignments the students prepared a quarter long final project based on a topic of interest. The course was primarily run by Marilyn Walker, however I helped build the syllabus and lead discussions in class.
http://courses.soe.ucsc.edu/courses/cmps245/Spring13/01/pages/computational-models

As a teaching assistant I helped students learn about several core concepts underlying modern operating systems including issues of concurrency, system calls, virtual memory and networking. My primary role was conducted during office hours where students sought help in the topics covered in lecture and problems that arose in their programming assignments. In addition I helped in the administration and grading of the projects and exams.