In this work, we address the task of Language-Assisted 3D Shape Edits and Deformations (which we name ChangeIt3D). Given a 3D representation of an object and free-form natural language describing desired changes or modifications to the shape of the object, the task is to transform the input object’s geometry in a manner that reflects the requested changes – for example, to modify a 3D chair model to make its legs thinner, or to open a hole in its back. To tackle this problem in a way that promotes open-ended language usage allowing fine-grained shape edits, we introduce the largest existing corpus of natural language describing shape differences, which we call ShapeTalk. This dataset contains over half a million discriminative utterances produced by contrasting the shapes of pairs of common 3D objects for a variety of object classes and degrees of similarity. We introduce metrics for the quantitative evaluation of language-assisted shape editing methods that reflect key desiderata within this editing setup. We also design an effective and modular framework for ChangeIt3D that can combine an arbitrary 3D generative model of shapes with our in-house, ShapeTalk-based, text-to-shape neural listener. Crucially, our modules are trained and deployed directly in a latent space of 3D shapes, bypassing the ambiguities of “lifting” 2D to 3D when using extant foundation models and thus opening a new avenue for 3D object-centric manipulation through language.