Paper: Variational Inference for Grammar Induction with Prior Knowledge

ACL ID P09-2001
Title Variational Inference for Grammar Induction with Prior Knowledge
Venue Annual Meeting of the Association of Computational Linguistics
Session Short Paper
Year 2009
Authors

Variational EM has become a popular technique in probabilistic NLP with hid- den variables. Commonly, for computa- tional tractability, we make strong inde- pendence assumptions, such as the mean- field assumption, in approximating pos- terior distributions over hidden variables. We show how a looser restriction on the approximate posterior, requiring it to be a mixture, can help inject prior knowledge to exploit soft constraints during the varia- tional E-step.