Proceedings of the Annual Meeting of the Cognitive Science Society

For language-capable interactive robots to be effectively in-troduced into human society, they must be able to naturallyand efficiently communicate about the objects, locations, andpeople found in human environments. An important aspect ofnatural language communication is the use of pronouns. Ac-cording to the linguistic theory of the Givenness Hierarchy(GH), humans use pronouns due to implicit assumptions aboutthe cognitive statuses their referents have in the minds of theirconversational partners. In previous work, Williams et al. pre-sented the first computational implementation of the full GHfor the purpose of robot language understanding, leveraging aset of rules informed by the GH literature. However, that ap-proach was designed specifically for language understanding,oriented around GH-inspired memory structures used to assesswhat entities are candidate referents given a particular cogni-tive status. In contrast, language generation requires a modelin which cognitive status can be assessed for a given entity.We present and compare two such models of cognitive sta-tus: a rule-based Finite State Machine model directly informedby the GH literature and a Cognitive Status Filter designedto more flexibly handle uncertainty. The models are demon-strated and evaluated using a silver-standard English subset ofthe OFAI Multimodal Task Description Corpus.