An automated approach to identify sarcasm in low-resource language

Area of Science:

• Natural Language Processing (NLP)
• Computational Linguistics
• Machine Learning (ML)

Background:

• Sarcasm detection research is predominantly in English, with limited exploration in low-resource languages like Urdu.
• Developing effective sarcasm detection for Urdu is challenging due to the scarcity of annotated datasets and unique linguistic nuances.

Purpose of the Study:

• To investigate the efficacy of various machine learning algorithms for sarcasm detection in the Urdu language.
• To address the data scarcity challenge by curating and releasing the Urdu Sarcastic Tweets (UST) Dataset.
• To establish a benchmark for future research in low-resource language sarcasm detection.

Main Methods:

• Curated and released the Urdu Sarcastic Tweets (UST) Dataset from user-generated social media comments.
• Evaluated baseline machine learning classifiers including Support Vector Machine (SVM), Decision Tree (DT), K-Nearest Neighbor (K-NN), Linear Regression (LR), Random Forest (RF), Naïve Bayes (NB), and XGBoost.
• Validated model performance on the newly created UST dataset and the Tanz-Indicator dataset.

Main Results:

• The Support Vector Machine (SVM) classifier demonstrated superior performance, achieving an accuracy of 0.85.
• SVM consistently outperformed other evaluated machine learning models across different experimental configurations.
• The study confirms the feasibility of applying ML techniques to Urdu sarcasm detection.

Conclusions:

• Machine learning models, particularly SVM, are effective for sarcasm detection in Urdu.
• Tailored approaches considering specific linguistic characteristics are crucial for low-resource languages.
• The release of the UST dataset facilitates further research and development in this domain.