Categorization of conversational games in free dialogue referring to spatial scenes

Abstract

This thesis examines which communicative strategies speakers use to complete a given task regarding spatial scenes, and how to systematically categorize them as conversational games. This study expands a free dialogue Cups corpus in Swedish (Dobnik et al., 2016) from 794 to 985 turns and extend the existing and new data with the annotation of conversational games using a new Game Type Coding Scheme. The study defines seven distinct types of conversational games of which four show universal features believed to be applicable across different spatial tasks. The study shows some common features between the game types, which would allow computational classification. The reason why game types are important is that they define a scope within which particular linguistic features will be manifested (Dobnik et al.,2015). These findings would be applicable on all situated dialogue systems such as in cars, humanoids or handheld devices in order to come one step closer to more “human-like” answers to prompted questions that involve spatial reasoning, such as, “Siri, where did I park my car?”