MANY

MANY is an MT system combination software which architecture is described is the following picture :

The combination can be decomposed into three steps

1-best hypotheses from all M systems are aligned in order to build M confusion networks (one for each system considered as backbone).
All CNs are connected into a single lattice. The first nodes of each CN are connected to a unique first node with probabilities equal to the priors probabilities assigned to the corresponding backbone. The final nodes are connected to a single final node with arc probability of one.
A token pass decoder is used along with a language model to decode the resulting lattice and the best hypothesis is generated.

The decoder can be expressed as a classical log-linear model:

where λi is the weight of the feature function hi.
Feature functions used:

The LM probability
The system prior, corresponding to the probability of choosing a system as backbone.
The words scores: currently, each word has a score equal to the prior of the system which proposed it
The word-length penalty of the word sequence,
The null-penalty corresponding to the number of null-arcs (or epsilon arcs) crossed to obtain the hypothesis.

Downloads

Version	Date	Descripton	Download
v1 (current version)	12/07/09	First version with Confusion Network generation and Token Pass decoder.	[MANY SVN] (google code)