Text carries meaning much richer than its propositional content. What people often remember best about a text are their subjective impressions of how well-written the text was, or how the text was stylistically marked conveying formality, depth of detail and importance of its content. We have developed computational models to pinpoint the writing devices associated with certain perceptions of the text, to enrich automatic understanding of textual information. A fun application of our work was to analyze what features distinguish the best science journalism from typical science pieces that may garner less prominence. A more serious and long-term objective has been to develop practical tools for measuring the level of specificity or detail in text and separating factual information from subjective interpretations of the facts.
A fun story about the motivations for and applications of our work appeared in the Penn Engineering blog.
Faculty
- Ani Nenkova (CIS)
Funding
- NSF
