People with autism spectrum disorder often have trouble detecting sarcasm and irony, particularly when it is written in text. Now researchers have developed a system, called Sarcasm SIGN (sarcasm Sentimental Interpretation GeNerator), that can interpret sarcastic statements in social media.
Automatic identification and analysis of sentiment — human feeling or meaning — in text is a very challenging subject being explored by researchers worldwide, but so far none have developed an accurate text-translating system.
“There are a lot of systems designed to identify sarcasm, but this is the first that is able to interpret sarcasm in written text,” said graduate student Lotam Peled, who developed the system under the guidance of Assistant Professor Roi Reichart at the Technion-Israel Institute of Technology Faculty of Industrial Engineering and Management.
“We hope in the future, it will help people with autism and Asperger’s, who have difficulty interpreting sarcasm, irony, and humor.”
Based on machine translation, the new tool is able to turn sarcastic sentences into honest (non-sarcastic) ones. It will, for example, turn a sarcastic sentence such as, “The new ‘Fast and Furious’ movie is awesome. #sarcasm” into the honest sentence, “The new ‘Fast and Furious’ movie is terrible.”
Despite the vast development in this field, and the successes of sentiment analysis applications on “social media intelligence,” existing applications do not know how to interpret sarcasm, where the writer writes the opposite of what (s)he actually means.
In order to teach the system to how to interpret sarcasm accurately, the researchers compiled a database of 3,000 sarcastic tweets that were tagged with #sarcasm, where each tweet was interpreted into a non-sarcastic expression by five human experts.
The system was also trained to recognize words with strong sarcastic sentiments — for example, the word “best” in the tweet, “best day ever” — and to replace them with strong words that reveal the true meaning of the text.
The results were analyzed by a number of (human) judges, who gave its interpretations high scores of fluency and adequacy, agreeing that in most cases it produced a semantically and linguistically correct sentence.
Sentiment identification could be used in social, commercial, and other applications to improve communication between people and computers, and between social media users.
Source: American Technion Society