Learning the weights of a HFST morphology
We developed a tool for learning weights of a HFST transducer. In practice: a weighted Hungarian morphology is presented.
During the process, we re-implemented hfst-lookup with serious computational benefits: several times faster and less memory usage.
The presentation will be about the hfst implementation details and the learning of the weights as well. All these are more of a culmination of previous work and not particularly new theoretical results.