Bangla Interrogative Sentence Identification from Transliterated Bangla Sentences

In this paper, we propose a method to identify Bangla interrogative sentences from transliterated Bangla sentences. All over the internet, we generate a huge number of Bangla interrogative sentences and they are mostly written using transliteration. In transliterated Bangla, identifying interrogative sentences possesses great challenges. The question marks at the end of the interrogative sentences are not used in transliterated Bangla profoundly especially on social media. Moreover, people often make interrogative sentences without using Bangla question words when writing in transliterated form. To find the solution, we discuss the rule-based approach, supervised learning approach and a deep learning approach. In the rule-based approach, we design a set of rules based on grammar and data analysis. For employing supervised learning, we use machine learning techniques such as Support Vector Machine, k-Nearest Neighbors, Multilayer Perceptron and Logistic Regression and achieved accuracies of 91.43%, 75.98%, 92.11 % and 91.68% respectively. To apply deep learning, we implement a Convolutional Neural Network. This approach provides a decent result with an accuracy of 84.64 % and dignifies the scope of Convolutional Neural Network as an ideal model for Bangla natural language processing.