Improved Dynamic Natural Language Generation with Personage using Language Manager

As creative authors become responsible for larger bodies of text in video game narratives, it is clear that automation in generation may be the solution to creating large bodies of rich and expressive text. Computer generated story-lines are a reality in interactive narratives, but not in dialogue. In an effort to generate natural utterances dynamically, the main purpose of this project was to add more powerful underlying linguistic theory to existing models for more sophisticated computerized speech. Personage is an expressive natural language generation tool designed to support a more natural dialogue system with machine interaction, originally designed to produce restaurant recommendations. In order to develop Personage as a more generalized tool, we created a system called Language Manager by leveraging other tools, Stanford CoreNLP, which identifies parts of speech, and WordNet, a lexical database capable of making meaningful connections between words. These tools make it possible to input natural text, and have it parsed and use more contextual lexical substitutions than we are currently capable of generating. By using abstracted models for deep syntactic structures and creating templates for subcategorizations for verbs and nouns, Language Manager can better account for the contextual model, even with invented words. As a result, the extension of Personage using Language Manager makes it a more powerful tool for authors who may not be familiar with abstract computer science or linguistic concepts. This is a step towards a more generalized, more powerful tool for expressive natural language generation.