Communications of the ACM
Inside the Surprisingly High-Stakes Quest to Design a Computer Program That 'gets' Sarcasm Online
Credit: Marvel Heroes
A computer program that can reliably detect sarcasm online is highly desired by many people, but instilling reliability within software is a formidable challenge — not even an 85-percent accurate program to detect sarcasm on Twitter developed by University of California, Berkeley professor David Bamman is good enough. A key problem in meeting this challenge is the fact that most cues signaling sarcasm are non-textual.
One approach is to input massive volumes of "sarcastic" data into self-learning, pattern-seeking programs that search for recurring words, phrases, and topics people tend to reference when they are being sarcastic. The flaw in this strategy is the programs can misinterpret a statement as sarcastic if it uses a statistically sarcastic word. Stanford University professor Christopher Manning thinks this impediment can only be overcome when computers truly start to understand the lived human experience.
Bamman's work with his sarcasm-gauging algorithm is a mark of progress, as it attempts to understand the speaker, his audience, and the relationship between the two by also absorbing contextual data from past tweets and Twitter bios.
Another study using Reddit determined the accuracy of sarcasm-detection improves significantly if the program knows not just what was said, but where.
From The Washington Post
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found