Automatic generation of modern haiku poetry

Abstract

To generate realistic creative writing (story, poetry) automatically is really a challenge, as various kind of knowledge and intelligent that has to be used, so that the final product generated is human-understandable. In my research, the ultimate goal is to generate modern haiku poetry automatically. In general, a poem is a human artifact that consists of few lines of texts. There is a misunderstanding that automatic poetry generation is just an easy task to do, as readers of poetry are prepared to do considerable interpretative work. Besides just generating few lines of human-understandable text, to be a successful generated poem, one must fulfill certain kinds of requirements: for example, Grammatical: The most basic requirement that a text should fulfill, Meaningful: a poem should be meaningful so that it is human-understandable. Poetic feature of a poem: This is the most abstract aspect that is quite difficult to describe using few words. To deal with poetic feature, the issue is sometimes subjective in nature. It is indeed a challenge problem that most of poetry generation researches have to face as poetic feature is playing an important role in controlling the quality of generated poem. In haiku writing this can be called as Haiku moment. “Haiku moment were very much like other flashes of inspiration which, when transported into other media became paintings, stories, dreams or even new color schemes or recipes”. Besides, the structure of poetry, use of words etc are also important factors that affecting the overall quality of the output. Therefore, automatic poetry generation is not just an easy task that many people think of. In my research, a JAVA-servlet based system called HAIG (Haiku using AI Generator) is implemented to generated modern haiku poetry. HAIG made use of most of the popular web 2.0 technologies, blogosphere, and web search engine, to generate modern haiku poetry. The approach is different from most of the traditional creative writing researches focus on creating software algorithms that can automatically generate an entire poem from scratch. Sentences used in output haiku are harvested from blogosphere; score is calculated for output of HAIG based on the association ratio measurement between sentences. Except for detail implementation of HAIG, the predecessor of HAIG (VSM approach) will also be described in the thesis. Below shown one of the generated results by HAIG: Drifting snow and windy That leaves us dreaming Chase the moon