Computational Models of Discourse Analysis
Discourse analysis is the area of linguistics that focuses on the structure of language above the clause level. It is interesting both in the complexity of structures that operate at that level and in the insights it offers about how personality, relationships, and community identification are revealed through patterns of language use. A resurgence of interest in topics related to modeling language at the discourse level is in evidence at recent language technologies conferences. This course is designed to help students get up to speed with foundational linguistic work in the area of discourse analysis, and to use these concepts to challenge the state-of-the-art in language technologies for problems that have a strong connection with those concepts, such as dialogue act tagging, sentiment analysis, and bias detection.
This is meant to be a hands on and intensely interactive course with a heavy project component. The course is structured around 3 week units, each with a group work component.
Course Procedures and Grading Criteria
Some assignments will involve programming plugins for the LightSIDE text mining tool kit (http://www.cs.cmu.edu/~emayfiel/side.html). Plugins with either be novel feature extractors, classification algorithms, or meta-classifiers.
Grades will be assigned as follows:
15% for each of five assignments
10% for class participation
15% final critique of a technical paper
Instructor: Dr. Carolyn P. Rosé (firstname.lastname@example.org)
Office hours: Students are encouraged to request meetings with the instructor as needed
Units: 12 (PhD/Master’s)
Readings and On-Line Discussions:
James Paul Gee (2011). An Introduction to Discourse Analysis: Theory and Method, Third Edition, New York: Routledge
Additional readings will be linked to the syllabus or passed out in class.
Students are expected to do the readings and post a response to discussion questions on-line to the course Drupal account by 10pm the night prior to each class meeting.
Prerequisites: Students should be reasonably strong programmers and have taken or audited at least one machine learning course