An LSTM-Based Model and Algorithm for Uzbek Language Text Generation

This paper explores the problem of text generation in the Uzbek language from the perspective of artificial intelligence and deep learning approaches. Text generation is a crucial task within natural language processing (NLP), aiming to generate grammatically correct and semantically coherent texts based on linguistic corpora. The study examines models ranging from earlier statistical methods to modern transformer-based architectures such as GPT (Generative Pre-trained Transformer), T5, and BART, including their adaptations for low-resource languages like Uzbek. The lack of Uzbek-specific linguistic resources (e.g., corpora, tokenizers) is discussed as a key limitation, along with strategies to overcome them. A practical implementation of the LSTM model fine-tuned on an Uzbek corpus is presented, with examples of generated narrative and informational texts. Experimental results are evaluated using BLEU and perplexity metrics. The findings demonstrate the potential of generative models for Uzbek and their applicability in areas such as automated content creation, chatbots, and educational assistants.

Ҳали таржима қилинмаган