Computational Models of Discourse Analysis

Post on 16-Jan-2016

29 views 0 download

description

Computational Models of Discourse Analysis. Carolyn Penstein Ros é Language Technologies Institute/ Human-Computer Interaction Institute. Warm Up. How would you rate the new girl and the Indian blogger on these scales? An why?. Warm-Up discussion. - PowerPoint PPT Presentation

transcript

Computational Models of Discourse Analysis

Carolyn Penstein Rosé

Language Technologies Institute/

Human-Computer Interaction Institute

Warm Up How would

you rate the new girl and the Indian blogger on these scales?

An why?

Warm-Up discussion We read two theory papers so

far in this unit: The first paper was about what

style of reference says about identification with a community and with an interlocutor (as an ingroup member or not)

The second paper related to the way the use of time in a narrative speaks about self-concept and projected reader

How do these issues related to the aspects of personality covered in the Gill paper?

Based on that comparison, do these numbers make sense?

Further Discussion What was the research question the authors were trying to

answer? What was the reason for choosing LIWC? Do you think this was a reasonable approach? What do you conclude about how much of personality as it

is revealed through text is captured by LIWC features?

Notice that the author cited a lot of prior work from his own lab or people who used a similar methodology

Background on LIWC

Developed by Pennebaker Used frequently in medical informatics Usually applied to highly controlled data

Isolates as much as possible the variable being examined

Is the blog data controlled in the right way?

* Connection with subpopulations/ domain adaptation

Online LIWC Assessment

What would we conclude?

The new girl is more neurotic

The Indian is a little more extroverted

The Indian is a little more open

The new girl is more conscientious

The new girl is more agreeable

* Would you expect a machine learning model with these features to work well?

What features might you try instead?

Announcement

For Monday

Analyze one of the two blog posts from this perspective

(pp 92-135)

Questions?