PIE

PIE was primarily conceived to make experimentation on sequence labeling of variation-rich languages easy and user-friendly. PIE has been tested mostly for Lemmatization but other SoTA accuracies from other tasks like POS have been reproduced (cf. Plank et al ). PIE is highly configurable both in terms of input preprocessing and model definition, in principle not requiring users to write any code (instead experiments are defined with json files). It is highly modular and therefore easy to extend. It includes transductive lemmatization as an additional sequence labeling task and, finally, it is reasonably fast and memory efficient.