Piecewise Latent Variables for Neural Variational Text Processing

Iulian Vlad Serban, Alexander Ororbia II, Joelle Pineau, Aaron Courville


Abstract
Advances in neural variational inference have facilitated the learning of powerful directed graphical models with continuous latent variables, such as variational autoencoders. The hope is that such models will learn to represent rich, multi-modal latent factors in real-world data, such as natural language text. However, current models often assume simplistic priors on the latent variables - such as the uni-modal Gaussian distribution - which are incapable of representing complex latent factors efficiently. To overcome this restriction, we propose the simple, but highly flexible, piecewise constant distribution. This distribution has the capacity to represent an exponential number of modes of a latent target distribution, while remaining mathematically tractable. Our results demonstrate that incorporating this new latent distribution into different models yields substantial improvements in natural language processing tasks such as document modeling and natural language generation for dialogue.
Anthology ID:
W17-4308
Volume:
Proceedings of the 2nd Workshop on Structured Prediction for Natural Language Processing
Month:
September
Year:
2017
Address:
Copenhagen, Denmark
Editors:
Kai-Wei Chang, Ming-Wei Chang, Vivek Srikumar, Alexander M. Rush
Venue:
WS
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
52–62
Language:
URL:
https://aclanthology.org/W17-4308
DOI:
10.18653/v1/W17-4308
Bibkey:
Cite (ACL):
Iulian Vlad Serban, Alexander Ororbia II, Joelle Pineau, and Aaron Courville. 2017. Piecewise Latent Variables for Neural Variational Text Processing. In Proceedings of the 2nd Workshop on Structured Prediction for Natural Language Processing, pages 52–62, Copenhagen, Denmark. Association for Computational Linguistics.
Cite (Informal):
Piecewise Latent Variables for Neural Variational Text Processing (Serban et al., 2017)
Copy Citation:
PDF:
https://preview.aclanthology.org/nschneid-patch-4/W17-4308.pdf
Code
 julianser/hred-latent-piecewise +  additional community code