Mark-Evaluate: Assessing Language Generation using Population Estimation Methods
Gonçalo Mordido, Christoph Meinel · 2020
We propose a family of metrics to assess language generation derived from population estimation methods widely used in ecology.More specifically, we use mark-recapture and maximumlikelihood methods that have been applied over the past several decades to estimate the size of closed populations in the wild.We propose three novel metrics: ME Petersen and ME CAPTURE , which retrieve a single-valued assessment, and ME Schnabel which returns a double-valued metric to assess the evaluation set in terms of quality and diversity, separately.In synthetic experiments, our family of methods is sensitive to drops in quality and diversity.Moreover, our methods show a higher correlation to human evaluation than existing metrics on several challenging tasks, namely unconditional language generation, machine translation, and text summarization.