Address: HCI Institute, Carnegie Mellon University,
5000 Forbes Avenue, Pittsburgh, PA 15213
Email:
Nancy.Green@cs.cmu.edu Telephone: (412) 268-3084
Fax: (412) 268-1266
Research Interests
My current research interests are in human-computer interaction and natural language processing, including
Current Activities
I am a member (9/96 to present) of the Visualization and Intelligent Interfaces Group in the Robotics Institute at CMU. One of our projects is to develop an experimental system, AutoBrief, that can automatically generate presentations in text and information graphics (bar charts, maps, etc.) from databases. (The graphics are generated by the SAGE automatic graphic design system.) Our current testbed application is to integrate AutoBrief with a mixed-initiative scheduling system (DITOPS ) in the domain of transportation scheduling. My main contributions to AutoBrief have been the development of a knowledge representation scheme to serve both text and information graphics (IGW9, CVIR98), and designing and implementing a text microplanner, which plans how content is to be expressed in text. Research areas which I am currently addressing are media selection and coordination (e.g., What content should be expressed in text as opposed to graphics? How can text facilitate comprehension of a complex graphic? How should text about the user's problem-solving domain be integrated with text about the graphic?), multimedia argumentation strategies to support user tasks ( AAAI98, Draft97), and user manipulation of generated text as an interface device. In addition, I am advising a student at CMU's Language and Technologies Institute; her Ph.D. research is on mixed-initiative interfaces to information systems.
Previous Research
As a Postdoctoral Research Fellow in the Computer Science Department at CMU (10/94 to 8/96), I was a member of the NL-Soar project. NL-Soar was implemented in the Soar computational model of the human cognitive architecture. NL-Soar's goal was to develop real-time natural language capabilities enabling human students to communicate with intelligent artificial agents in a pilot training system. I implemented NL-Soar's discourse capabilities, using Soar to create a unified approach to discourse planning/learning and plan recognition (SSS98, AAAI96). In addition to its practical goal, the research is a step towards modeling cognitive processes in dialogue processing. In the future, I would like to explore the use of cognitive models of dialogue processing in the design and evaluation of user interfaces, as well as applying machine learning to discourse processing. I would also like to continue to develop conversational interface agents.
For my dissertation (Computer Science), I developed a computational model for generating and interpreting indirect answers to yes-no questions. Frequently occurring in human-human dialogue, indirect answers are responses consisting of relevant but not explicitly requested information from which the listener is licensed to infer the intended answer. In my model (CL99, UDel94, IGW7, ACL94, ACL92), indirect answers are generated by the following domain-independent processes: planning to update the common ground, taking the initiative to provide unrequested but relevant information (UMUAI99, SSS97), and simulating interpretation to identify redundant information in the plan which does not need to be explicitly expressed. Interpretation is modeled as recognition of this underlying plan. Possible future work includes empirical studies and computational modeling of related types of inference in dialogue. Indirect answers are a type of context-sensitive implication in dialogue known as conversational implicatures. In earlier research, I modeled the interpretation of normal state implicature (ACL90), another type of conversational implicature. My masters thesis (Linguistics) was in lexical semantics on an inferential taxonomy of adjective classes (UNC80).
Software Industry Experience
During graduate school (1987-1994 during summers, holidays, and while writing my dissertation), I worked as a Systems Developer in the AI Group at SAS Institute. Our group developed SAS/ENGLISH, an NL-database interface product enabling end users to query their SQL database in English. I started out by co-designing the grammar formalism for the NL parser, provided linguistics expertise, and wrote NL grammars. My last project was to design and implement a GUI Browser to enable the user to view the structure of his database and see how it relates to the system's English vocabulary. The Browser's text is automatically generated from the user's underlying database and lexicon.
From 1980-1985 I was a Technical Contributor at Data General, for four years in a database group that developed a native SQL, and for one year in an AI group. I developed systems software including a query optimizer (Database query code generation and optimization based on the cost of alternate access methods, U.S. Patent Number 04829427 ). I also developed expert system prototypes and evaluated AI tools.
Teaching Experience
Graduate Education
Selected Professional Activities
SIGDIAL (Dialogue Processing), SIGGEN (Natural Language Generation), SIGMEDIA (Multimedia Language Processing)