Supported Attributes

The task option in the client.score() function decides what attributes we calculate. Currently, we only support attributes for summarization task (task=sum). The following attributes (reference: this paper) will be calculated if cal_attributes is set to True in client.score(). They are all reference-free.

  • source_len: measures the length of the source text.

  • hypothesis_len: measures the length of the hypothesis text.

  • density & coverage: measures to what extent a summary covers the content in the source text.

  • compression: measures the compression ratio from the source text to the generated summary.

  • repetition: measures the rate of repeated segments in summaries. The segments are instantiated as trigrams.

  • novelty: measures the proportion of segments in the summaries that haven’t appeared in source documents. The segments are instantiated as bigrams.

  • copy_len: measures the average length of segments in summary copied from source document.