Oscar Wilde as soon as stated that sarcasm was the bottom type of wit, however the highest type of intelligence. Maybe that is because of how tough it’s to make use of and perceive. Sarcasm is notoriously difficult to convey via textual content — even in particular person, it may be simply misinterpreted. The refined adjustments in tone that convey sarcasm usually confuse pc algorithms as effectively, limiting digital assistants and content material evaluation instruments.
Xiyuan Gao, Shekhar Nayak, and Matt Coler of Speech Know-how Lab on the College of Groningen, Campus Fryslân developed a multimodal algorithm for improved sarcasm detection that examines a number of facets of audio recordings for elevated accuracy. Gao will current their work Thursday, Might 16, as a part of a joint assembly of the Acoustical Society of America and the Canadian Acoustical Affiliation, working Might 13-17 on the Shaw Centre situated in downtown Ottawa, Ontario, Canada.
Conventional sarcasm detection algorithms usually depend on a single parameter to provide their outcomes, which is the principle cause they usually fall brief. Gao, Nayak, and Coler as a substitute used two complementary approaches — sentiment evaluation utilizing textual content and emotion recognition utilizing audio — for a extra full image.
“We extracted acoustic parameters reminiscent of pitch, talking price, and power from speech, then used Automated Speech Recognition to transcribe the speech into textual content for sentiment evaluation,” stated Gao. “Subsequent, we assigned emoticons to every speech phase, reflecting its emotional content material. By integrating these multimodal cues right into a machine studying algorithm, our method leverages the mixed strengths of auditory and textual info together with emoticons for a complete evaluation.”
The group is optimistic concerning the efficiency of their algorithm, however they’re already in search of methods to enhance it additional.
“There are a selection of expressions and gestures folks use to spotlight sarcastic parts in speech,” stated Gao. “These should be higher built-in into our undertaking. As well as, we wish to embrace extra languages and undertake creating sarcasm recognition strategies.”
This method can be utilized for greater than figuring out a dry wit. The researchers spotlight that this system could be broadly utilized in lots of fields.
“The event of sarcasm recognition know-how can profit different analysis domains utilizing sentiment evaluation and emotion recognition,” stated Gao. “Historically, sentiment evaluation primarily focuses on textual content and is developed for purposes reminiscent of on-line hate speech detection and buyer opinion mining. Emotion recognition based mostly on speech could be utilized to AI-assisted well being care. Sarcasm recognition know-how that applies a multimodal method is insightful to those analysis domains.”
