Learning with Latent Language

Proceedings of NAACL-HLT 2018, pages 2166–2179 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Learning with Latent Language Jacob Andreas Dan Klein Sergey Levine Computer Science Division University of California, Berkeley {jda,klein,svlevine}@eecs.berkeley.edu Abstract The named concepts and compositional opera- tors present in natural language provide a rich source of information about the abstractions humans use to navigate the world. Can this linguistic background knowledge improve the generality and efficiency of learned classifiers and control policies? This paper aims to show that using the space of natural language strings as a parameter space is an effective way to capture natural task structure. In a pretraining phase, we learn a language interpretation model that transforms inputs (e.g. images) into outputs (e.g. labels) given natural language descriptions. To learn a new concept (e.g. a classifier), we search directly in the space of descriptions to minimize the interpreter’s loss on training examples. Crucially, our models do not require language data to learn these concepts: language is used only in pretraining to impose structure on subsequent learning. Results on image classification, text editing, and reinforcement learning show that, in all settings, models with a linguistic parameterization outperform those without.1 1 Introduction The structure of natural language reflects the struc- ture of the world. For example, the fact that it is easy for humans to communicate the concept left of the circle but comparatively difficult to com- municate mean saturation of the first five pixels in the third column reveals something about the abstractions we find useful for interpreting and navigating our environment (Gopnik and Meltzoff, 1987). In machine learning, efficient automatic discovery of reusable abstract structure remains a major challenge. This paper investigates whether 1Code and data are available at https://github.com/ jacobandreas/l3. 0.0 0.9 0.8 true true true true concept learning: evaluation: there is a green square a gray square is above a square a red cross is below a square 0.2 a red cross is below a square Figure 1: Example of our approach on a binary image classification task. We assume access to a pretrained language interpretation model that outputs the proba- bility that an image matches a given description. To learn a new visual concept, we search in the space of natural language descriptions to maximize the interpre- tation model’s score (top). The chosen description can be used with the interpretation model to classify new images (bottom). Figures best viewed in color. background knowledge from language can pro- vide a useful scaffold for acquiring it. We specif- ically propose to use language as a latent param- eter space for few-shot learning problems of all kinds, including classification, transduction and policy search. We aim to show that this linguis- tic parameterization produces models that are both more accurate and more interpretable than direct approaches to few-shot learning. Like many recent frameworks for multitask- and meta-learning, our approach consists of three phases: a pretraining phase, a concept-learning phase, and an evaluation phase. Here, the product of pretraining is a …

Content truncated to 500 words