Discovering Latent Structures In Syntax Trees And Mixed-Type Data