NGramRandGen

Description

This operation randomly generates a set of successful sentences in the input FST and outputs them in finite-state archive format. Backoff transitions, represented as epsilon transitions, are correctly interpreted as failure transitions, which distinguishes this operation from OpenFst RandGen (with the LogProbArcSelector). The (maximum) number of sentences generated, their maximum length, and the random seed can all be specified as options.

Usage

ngramrandgen [--opts] [[in.fst [out.far]]
   --max_sents: type = int64, default = 1
   --max_length: type = int, default = INT_MAX
   --seed: type = int, default = time()

Examples

$ ngramrandgen in.mod >out.far

Caveats

References