Grammar induction is the task of learning a set of syntactic rules from an unlabeled text corpus. Much recent work in this area has focused on learning probabilistic context-free grammar (PCFG) rules; however, these rules are not sufficiently expressive to capture the full variety of structures found in human languages. Bisk and Hockenmaier (2012) present a system for inducing a Combinatory Categorial Grammar, a more expressive formalism, but this system learns from sentences with part-of-speech tags rather than unlabeled data. I will present my initial work toward implementing a categorial grammar induction system that can learn from unlabeled data using a neural network–based architecture.