ROUGE

This is a pure Python implementation of the ROUGE metrics family in the automatic summarization fields following the paper ROUGE: A Package for Automatic Evaluation of Summaries Chin-Yew Lin et al.. It is an attempt to implement these metrics correctly and elegantly in total Python. It provides the following features:
ROUGE-N, ROUGE-L, and ROUGE-W are currently supported.
Flexible input. For each metric supported, sentence-level and summary-level variants are provided, which means you can use them in a machine translation context with sentence pairs.
Correctness. All the claimed implemented metrics are tested against a non-trivial amount of data, using the plain old Perl script as a baseline.
Self-contained. The total implementation is one single script in one single package. No dependency except a Python-3 is required. No Perl script is involved.*
Fast. At least faster than the Perl script wrappers. Preprocessing is the total freedom of the client. We establish API on the concept of a sentence, which is a list of tokens, and sentences, which is a list of sentences. Preprocessing like stopword removal, stemming and tokenization is left to the client.
Well documented. Every function has a .doctest
Procedural style API. You don't need to instantiate an object. Just call the function that does the right job.