[developers] token mapping determinism
glenn at thai-language.com
Fri Apr 12 22:18:09 CEST 2013
It has been suggested that I document in this forum a source of potential non-determinism in the token mapping architecture. In
short, under the current system, a rule such as
+POSITION C1<I1, I1 at O1
applied to lattice
can either result in
This ambiguity arises because the rule can be applied to the lattice in multiple ways, and, while there is a stated requirement that
rules be processed in order, there is no such edict regarding the ordering of permutations within the application of a single rule.
The ambiguity introduced here can further compound if one considers that the behavior after matching the rule can differ between
either restarting the same rule at the beginning of the lattice (ACE, agree-safer but slower), or continuing with the rule's pass
through the lattice (PET-risky but faster) before restarting.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the developers