Joint language-perception model using probabilistic grammar and classifiers significantly improves grounded attribute learning, achieving 82% precision in object set selection.

Cynthia Matuszek, Nicholas FitzGerald, Luke Zettlemoyer et al.