The GEM Benchmark: Natural Language Generation, its Evaluation and Metrics

TLDR

Progress in NLG is hampered by a constantly evolving ecosystem of metrics, datasets, and evaluation standards, leading to divergent anglo‑centric corpora and flawed metrics that obscure model limitations and opportunities for improvement. The paper introduces GEM, a living benchmark that enables easy application of models to diverse NLG tasks and evaluation strategies, and describes the shared‑task data for the ACL 2021 Workshop inviting community participation. GEM is a regularly updated benchmark that allows models to be applied to a wide set of tasks and evaluation strategies, thereby promoting multilingual research and evolving alongside models.

Abstract

We introduce GEM, a living benchmark for natural language Generation (NLG), its Evaluation, and Metrics. Measuring progress in NLG relies on a constantly evolving ecosystem of automated metrics, datasets, and human evaluation standards. Due to this moving target, new models often still evaluate on divergent anglo-centric corpora with well-established, but flawed, metrics. This disconnect makes it challenging to identify the limitations of current models and opportunities for progress. Addressing this limitation, GEM provides an environment in which models can easily be applied to a wide set of tasks and in which evaluation strategies can be tested. Regular updates to the benchmark will help NLG research become more multilingual and evolve the challenge alongside models. This paper serves as the description of the data for which we are organizing a shared task at our ACL 2021 Workshop and to which we invite the entire NLG community to participate.