OpenGrm SFst Background Material

The following is provided as background reading about stochastic finite state transducers and related material.

For material on finite state transducers as well as OpenFst, see the compilation here.

For a presentation of algorithms for weighted automata with failure transitions, used by the operations here, see: Cyril Allauzen and Michael Riley, "Algorithms for Weighted Automata with Failure Transitions", Proceedings of the 23rd International Conference on Implementation and Application of Automata, (CIAA 2018), Charloittetown, PEI. Note this paper makes two simplifying assumptions: (1) a successful path can not end in a failure transition and (2) there are no epsilon transitions when failure transitions are present. Neither limitation is required in this library (failure transitions treat epsilons as defined for a canonical FST).

For a presentation of algorithms for approximating probabilistic models as weighted automata with failure transitions, provided by operations here, see: Ananda Theertha Suresh, Brian Roark, Michael Riley ,Vlad Schogol, Distilling weighted automata from arbitrary probabilistic models, Proc. of the FSMNLP 2019, Dresden, Germany. A longer version with proofs and additional experiments is here.

Michael Riley - 2018-07-17