NLP – Understanding context to recognize Sarcasm

Natural Language Processing it’s the IT Engineering field which studies algorithms about natural language (human language) interpretation: 
Well, do you know when Cortana or Siri understand what you mean? That’s it.

Some weeks ago Google has developed AlphaGo that won at Go against human (one of the most important goal of General Game Playing), the NLP reasearch progresses are really slow.

Natural Language processing

The natural language is really complex and it is made by a lot of different form:
textual, vocal, gestual, micro-expressive facial, kinesthetic, etc..
By now, the most important research about NLP are concentrated on textual and vocal communication.
But every form of communication have a literal meaning and a semantic meaning, which depends on situations, speakers and listener.
The most part of the problems about human language interpretation is about semantic:
the meaning depends on such an infinity of different variables.. Our culture, our thought, our knowings, our experience and so on.

Sarcasm

Sarcasm it’s one of the most difficult retorical figures to understand, algorithmically speaking:
In order to be understood, it needs that both of the two parts which are communicating know a shared information which the sarcastic proposition refers to.
Technically:
If the other problems can be resolved just applying linguistic rules, in this case we can’t be sure if we just apply some linguistic rules and controls; we need to understand the context in which a proposition s collocated.

Contextualized sarcasm detection on twitter

Twitter is a really useful stream for psychological, sociological and linguistic research. Every second there a lot of new tweets, some of those are sarcastic.
David Bamman and Noah A. Smith has been working on a new important research to recognize sarcasm on Twitter: Not only using linguistic rules, but understanding the context of a tweet.

Se c’è una cosa che amo, è cercare un Tweet sarcastico per scrivere un articolo di #NLP sul riconoscimento del #Sarcasmo.

— Manuel Occorso (@manuel_occ) February 14, 2016

If there is anything I really love, is to search for a sarcastic tweet on Twitter to write a new #NLP article about #Sarcasm recognizing.

A tweet like this is obziously sarcastic, and you know because you’re reading this article, so you understand the context of the tweet. But without the context, it’s really difficultto understand sarcasm.
The researchers identified different level of inspection for a tweet: the audience, the author, the previous tweets by the same author and so on..
The they tried to tag a lot of tweet as sarcastic or not, based on the level of inspection. And that’s it:
They reached 85% accuracy knowing the context of a tweet.

Not just sarcasm

Obviously, this approach to the problem can be applied to a a lot of different retorical figures.
Understanding context it’s  really important to understand every proposition, and this approach can be applied to other form of communication, like body language.
You can read about an application of contestualizing body language HERE.