Arts and Sciences
This research attempts to create poetic texts in the Shakespearean style. Many available technologies can create artworks based on styles of well-known artists. However, it is difficult for the generative models to create texts within the style of a particular author. This study aims to produce sentences in the style of Shakespeare that contain metaphorical meanings without quoting his works. We have trained statistical n-gram models on the complete works of Shakespeare and then used the model to create Shakespearean text. We noticed that, in some cases, the n-gram models will construct sentences that are copied from Shakespeare’s works. We then trained a neural-AI Word2Vec model word embedding to “paraphrase” words with others that have similar semantics (and thus preventing plagiarism). The initial results include some adequate sentences and some with semantic errors. There are also many ungrammatical utterances. We are focusing on substituting words with correct part-of-speech and screening out ungrammatical results. We also incorporated the Glove embedding model, which is trained on a large corpus of modern texts, to add modern words and themes into the Shakespeare-styled generated results. This research will give insights on how the models can generate texts that are more artistic.
Liu, Hexin, "Using Word2Vec and N-grams to Generate Poetic Texts" (2023). Summer Interdisciplinary Research Symposium. 151.