In the Garden and in the Ark: The belles lettres, aetiological tales, and narrative explanatory trajectories-The concept of an architecture combining phono-semantic matching, and NLP story-generation

摘要

This article describes an ambitious concept of architecture that puts phonology, morphology, and story-generation in the service of the literary genre of aetiological tales (e. g. Kipling's Just So Stories). Aetiological tales are a particular kind of narratives consisting of a fictional exploration of a name. We propose a computational model of the process involved in the creation of the aetiological tale. The model (unimplemented as yet) combines notions and techniques from two different areas of computer science and Natural Language Processing (NLP): computational humour and computational storytelling. The focus in this study is on providing a linguistic study of a particular type of storytelling whose output is in the Hebrew language, using a computational model to show the regularities and the structure of this type of narrative composition. We show how this could be done in other languages as well. Our main example (from Midrash Kol _ H ay) also fits in the genre of filling in narrative gaps (e. g. supra-biblical tales). Having described the high-level architecture, as designed, of GALLURA, a system for generating aetiological tales that is to emulate a modern literary corpus that in turn mimics the lateantique and medieval corpus of rabbinic homiletics, the rest of this article is about: (1) how to get the first hunch about how to develop a tale on a theme, and (2) the workings of the wordplay-generating module (upstream in the architecture), whose output is a specification setting the task for the story-generator to devise narrative trajectories from the input word to a target being any of the pool of potential puns identified. Punning based on phonetic similarity is a well-known technique used in computational humour generators whose output is in English; but Semitic languages such as Hebrew and Arabic are a special challenge, because of their non-concatenative morphology. In the latter part of this article, we show in detail how to handle punning in such languages.