Unbabel-smaug

Highlights

* Modified spans now output empty spans to track deletions. (41)
* mT5 tokenisation no longer adds a space if a suffix is generated. (40)

Main features

* New transform `trans-swp-poisson-span` to swap a text span following a Poisson distribution;
* Modified indices for **ALL** transforms.

This is the first release of **SMAUG** - a sentence-level multilingual augmentation package for introducing perturbations in existing datasets.