Skeletonization: a technique for trimming the fat from a network via relevance assessment
Advances in neural information processing systems 1
Recursive distributed representations
Artificial Intelligence - On connectionist symbol processing
Distributed Representations, Simple Recurrent Networks, And Grammatical Structure
Machine Learning - Connectionist approaches to language learning
The EuTrans Spoken Language Translation System
Machine Translation
Hi-index | 0.00 |
Previous work has recently shown that adequate and compact codifications of the lexicons involved in text-to-text MT tasks can be automatically created. The method extracted these representations from perceptrons with output contexts. They were later tested on a simple neural translator called RECONTRA. However, the size of the codifications was determined by hand using try-and-error mechanisms. This paper presents a method for automatically obtain such sizes by pruning the units of the hidden-layer of the perceptron encoder.