The idea of the task is to implement a mechanism of resolving transfer rule conflicts using previously obtained rule weights. The tool for obtaining the weights is to be implemented as well.
Currently, transfer rules have to be devised with no possibility of conflicts, and that limits the system's capabilities, because ambiguity is an essential part of any natural language, and that fact should be reflected by the rules. Introducing of weighted rules would have a positive impact on both how the transfer rules represent the language pair and the quality of the resulting translation.
The deliverables will be as follows:
- Standalone training script to calculate the weights given the corpus and the rules.
- The code in C++ integrated into Apertium to choose the rules given the weights.