This project aims to simulate what it is like for an agent to learn pragmatic ability given only utterances from a speaker and knowledge of the context (positions of referents in the world). The learner learns the lexicon that the speaker is using (mappings of signals 'a', 'b', 'c' onto meanings 0, 1, 2), the perspective that the speaker has (0 or 1), and whether the speaker produces utterances pragmatically or literally.
- A perspective-inferring learner learning the lexicon and perspective of a literal speaker
- A level-2 pragmatic learner learning the lexicon and perspective of a level-1 pragmatic speaker
- A learner learning the lexicon and perspective of a speaker, as well as learning whether the speaker is pragmatic or not