Poetry Generation Model via Deep learning incorporating Extended Phonetic and Semantic Embeddings

Arabic poetry generation is a very challenging task since the linguistic structure of the Arabic language is considered a severe challenge for many researchers and developers in the Natural Language Processing field. In this paper, we propose a poetry generation model with extended phonetic and semantic embeddings (Phonetic CNN
_subword
embeddings). The proposed approach consists of three stages: (1.) Word Embedding, (2.) Keywords Extraction and Expansion, and (3.) Poetry Generation stage. Our model is able to generate the first verse which explicitly incorporates the theme related phrase using the Backward and Forward Language Model (B/F-LM) with Gated Recurrent Unit (GRU) cell, and then generates other verses of the poem sequentially, where each verse is composed based on a keyword and all previous generated verses using our proposed Hierarchy-Attention Sequence-to-Sequence model (HASS). We show that Phonetic CNN
_subword
embeddings have an effective contribution to the overall model performance. The Keywords Extraction and Expansion stage can ensure that the generated poem is coherent and semantically consistent with the input query intent. A comprehensive human evaluation confirms that the poems generated by our model outperform the base models in criteria including Meaning, Coherence, Fluency, and Poeticness. Extensive quantitative experiments using Bi-Lingual Evaluation Understudy (BLEU) scores also demonstrate significant improvements over strong baselines.