A BERT-based similarity measure between reference and generation.

A learned metric to measure semantic equivalence between reference and generation.

Language model style conditional entropy -- N-grams conditioned on N-1-grams.

Language model style conditional entropy -- N-grams conditioned on N-1-grams

Ratio of distinct unigrams / total number of unigrams.

Ratio of distinct bigrams / total number of bigrams.

Ratio of distinct trigrams / total number of trigrams.

An advanced lexical similarity metric also including stemming and synonymy matching.

Mean Segmental Type-Token Ratio, a measure of lexical diversity.

NIST is an alternative to BLEU with slightly different calculation.

A metric combining multiple aspects like entailment and similarity.

ROUGE score focusing on longest common subsequence.

number of unigrams that only occur once in the whole data.

Number of bigrams that only occur once in the whole data

Number of trigrams that only occur once in the whole data.

Number of distinct words used by the system.

Number of distinct bigrams used by the system.

Number of distinct trigrams used by the system.