Need to account for semantic relations in the propagation of activation from SemWM to GraWM

[victorbarres]

Right now only node propagate their activation. This cannot be right. In general, I need to have a much more uniform treatment of relations and node in SemRep.

[victorbarres]

There are actually a few different issues here:

• I have modified the code so that edges are also marked as "expressed" once a cxn assemblage is selected. (from gramWm to SemWM)
• There is the question of the propagation of activation from edges from SemRep to cxn through the convey_sem_activations method. This needs to be clarified since for now this does not even make use of the differentiation between lexical and non-lexical cxn which was one of the important aspect of the pb. (from SemWM to gramWM)
• Finally there is the role that relations coverage should play when scoring the assemblage (not simply how many nodes it covers but also how many relations!).

[victorbarres]

There is an undue asymmetry introduced between nodes and edges due to the NetworkX implementation of the graph algo. nodes are directly named after the related schema instance while edges are named as a 2-uples (inst_from, inst_to) when they should be named using the name of the associated schema instance for clarity.

[victorbarres]

• semrep relations are now taken into account in the sem_length factor when scoring assemblages.
• semrep relations are marked as expressed if they have been expressed by an assemblage
• Still need to figure out how to properly deal with the activation propagation and the naming issue (see previous comment)

[victorbarres]

Activation is now propagated also from relations in convey_sem_activation(). Haven't address the naming issue.