Narrative and Emotional Structures For Generation Of Short Texts For Advice

This communication proposes a new method for the generation of texts using a large language model (LLM) with a double structural foundation. The supervised generation requires two proposed sets of narrative structures (SN) and emotional dynamics (DE). An explicit definition of 53 SN et 10 DE associated to a user table of topics, characters, times, atmospheres and a writing style charter, allows to generate automatically a corpus of short texts without training. The main purpose of such texts is to communicate advice and learning concepts rarely found in narrative design at large scale. By modelling behavioural and emotional sequences that help attention and identification, engagement and affective understanding follow in line with social learning theory. The framework leads to a diversification and a planification of flexible educational auto-generated texts suitable for learners. The experimental section illustrates the applicability of the method for automatic textual generation. The available open source Python package named Narremgen implements the method for writing whole books of advice (answer) to topic (query). Several synthetic corpora, one of 36 texts and 10 of about 250 texts each evaluated with textual statistics, illustrate the scalability and generality of the implementation for diverse topics like the urban walk context.