Sociolinguistically Informed Natural Language Processing: Automating Irony Detection
Abstract
The major goals of this project are summarized by the following aims: Aim 1. To collect and annotate a high-quality corpus to facilitate research on irony detection. Prior to this project, no such high-quality dataset existed. This has been a major obstacle to progress on automated irony detection. Aim 2. To analyze when existing ML and NLP technologies fail to detect ironic intent empirically. We specifically proposed to assess quantitatively (using the collected dataset) whether context is necessary to discern ironic intent(and how often this is the case).Aim 3. Develop a new approach to irony detection that instantiates sociolinguistic conceptions of irony within a modern, probabilistic machine learning framework. This approach is to be informed by theoretical sociolinguistic perspectives on irony (and thus likely capable of discerning ironic utterances missed by existing computational models), while also being practical enough to be operational.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 23, 2017
- Accession Number
- AD1051278
Entities
People
- Byron C Wallace
- David Beaver
Organizations
- University of Texas at Austin