Development and Design of Deep Learning-based Parts-of-Speech Tagging System for Azerbaijani language

Abstract

Parts-of-Speech (POS) tagging, also referred to as word-class disambiguation, is one of the prerequisite techniques that are used as part of the advanced pre-processing stage across pipeline at the majority of natural language processing (NLP) applications. By using this tool as a preliminary step, most NLP software, such as Chat Bots, Translating Engines, Voice Recognitions, etc., assigns a prior part of speech to each word in the given data in order to identify or distinguish the grammatical category, so they can easily decipher the meaning of the word. This thesis addresses the novel approach to the issue related to the clarification of word context for the Azerbaijani language by using a deep learning-based automatic speech tagger on a clean (manually annotated) dataset. Azerbaijani is a member of the Turkish family and an agglutinative language. In contrast to other languages, recent research studies of speech taggers for the Azerbaijani language were unable to deliver efficient state of the art accuracy. Thus, in this thesis, study is being conducted to investigate how deep learning strategies such as simple recurrent neural networks (RNN), long short-term memory (LSTM), bi-directional long short-term memory (Bi-LSTM), and gated recurrent unit (GRU) might be used to enhance the POS tagging capabilities of the Azerbaijani language.