Malayalam Language Textual Inference Identification Through Attention over Embeddings
Abstract
This work focus on natural language inference between two text fragments, which is a uni-directional relationship. Natural language inference is attempted for a language called Malayalam. The Malayalam language is a South Indian, low-resource language. NLI is a subtask in every language because of its importance in summarization, information retrieval, and many other applications. There are only a few attempts in NLI for the Malayalam language. In this work, the application of the attention mechanism helped to enhance the classification performance for binary and multiclass systems. A densenet with additive attention is implemented for softmax and sigmoid classification, increasing accuracy, recall, and F1-score.
Domains
Computer Science [cs]Origin | Files produced by the author(s) |
---|