A Deep Learning Framework for Anaphora Resolution from Social Media Text

Anaphora resolution is a classical problem in natural language processing. For more than forty years’ people are an attempt to resolve the issue with different approaches. Because of the linguistic complexity and resource constraints, it is still an active research problem, especially for the resource-scarce language like Indic languages. This paper attempted to resolve anaphora using the LSTM-based deep learning method. The primary focus of this approach is to do in an unannotated text. The system is evaluated by the matrices like mentions, MUC, BCUB, CEAFE, CEAFM and LEA. This research makes an effort to conduct anaphora resolution from twitter data which is a challenging task due to its inherent unstructured nature. Experimental results demonstrate high degree of agreement with human annotations.