Technion system interprets Twitter sarcasm

Goal to help people with autism, Asperger’s in social interaction.

Twitter logo (photo credit: TWITTER)
Twitter logo
(photo credit: TWITTER)
Anybody who writes emails or is active in the social media knows it’s very difficult to express or comprehend feelings online.
Misunderstandings can cause hard feelings or worse. Automatic identification of emotions in the text employs many researchers around the world because of its business potential and scientific interest.
Emotional recognition can be used in social, commercial and other applications and improve communication between the individual and the computer and people using social networks.
Despite the tremendous development in this field, and the successes in sentimental analysis, the existing applications do not know how to cope with a sarcastic language that turns the writer’s intent. For example, if a sarcastic tweet is “The new Fast and Furious movie is awesome,” we will literally miss the point.
Lotem Peled, a graduate student at the industrial engineering and management faculty at the Haifa Technion-Israel Institute of Technology, has developed a system to interpret sarcastic statements. The system, developed under the guidance of Prof. Roi Reichart, is called Sarcasm SIGN (sarcasm Sentimental Interpretation GeNerator).
There are a lot of systems that aim to identify sarcasm, but this is the first system in the world to interpret sarcasm in a written text, Peled said, “and we hope that in the future it will help people with autism and Asperger’s who have difficulties with the interpretation of sarcasm, irony and humor.”
The new system, based on machine translation, turns sarcastic sentences into honest (non-sarcastic) sentences.
“The new film Fast and Furious is simply excellent” or “The new film of Fast and Furious is terrible” will become a true sentence, Peled explained.
To teach the system to produce these interpretations, the researchers put together a database of 3,000 sarcastic tweets, which were labeled as sarcasm by their authors. Each of the tweets was accompanied by five non-sarcastic interpretations written by human beings.
This data is also used to identify sentimental words. For example, the “best” word in “the best day ever” was replaced with sharp words such as “worst day ever” that reveal the meaning of the text.
The system was examined by a series of human judges, and it was found that in most cases, it produces a correct sentence both semantically and linguistically.
Peled will soon present her research at the prestigious ACL 2017 language processing conference in Vancouver, Canada.