Leveraging human prompts, Language Models (LLMs) like OpenAI's ChatGPT now boast enhanced natural language processing, rapidly answering queries. As AI chatbots integrate into our work, understanding their capabilities and limitations is crucial for informed and strategic utilization.
A researcher at New York University, Juliann Zhou, carried out a study on two Large Language Models to figure out if they understand sarcasm or not. This study showed whether an AI model can understand and use sarcasm and ironic conversation or not. Zhou highlighted that in today's world, it is very important to identify sarcasm if you want to truly know a person's thoughts about different matters. Previously, Support Vector Machines and Long Short-Term Memory models were used to identify if the AI models can identify sarcasm.
We can also call this study the Sentimental Analysis of an AI model because sentimental analysis is done on social media to know people's thoughts about a topic. Many other companies are also trying sentimental analysis to know what people think of their products. Many NLP models are capable of identifying emotions like positive, negative, and neutral but people on social media often talk in sarcasm and irony too. So, AI models detect these tones as positive but in reality, they are far from it. This is the reason a few scientists are working on developing a model that can understand sarcasm in text. Models like RCNN-RoBERTa CASCADE are examples of it.
RoBERTa is a transformer which understands the undertones of a language. CASCADE is used for producing results for sarcasm. Juliann Zhou used these two models for her study. She collected some sarcastic comments from Reddit and put them to RoBERTa and CASCADE to evaluate if they understood the sarcasm. The detection of these models was then compared to human detection of the underlying sarcastic tone in those comments. It was concluded that context about the situation can improve the sarcasm detection of AI models. Additionally, the use of transformers like RoBERTa can also improve the detection qualities of the AI models.
Read next: The Politics of Hashtags - What the Latest Research Reveals About TikTok and Instagram
