CTRLEval: An Unsupervised Reference-Free Metric for Evaluating Controlled Text Generation
This addresses the need for better evaluation metrics in controlled text generation, offering an unsupervised approach that avoids overfitting and improves correlation with human judgments, though it is incremental in the field of text generation evaluation.
The paper tackles the problem of evaluating controlled text generation models by proposing CTRLEval, an unsupervised reference-free metric that formulates evaluation aspects as text infilling tasks using a pre-trained language model. Experimental results show it achieves higher correlations with human judgments and better generalization across models and text qualities than baselines.
Existing reference-free metrics have obvious limitations for evaluating controlled text generation models. Unsupervised metrics can only provide a task-agnostic evaluation result which correlates weakly with human judgments, whereas supervised ones may overfit task-specific data with poor generalization ability to other datasets. In this paper, we propose an unsupervised reference-free metric called CTRLEval, which evaluates controlled text generation from different aspects by formulating each aspect into multiple text infilling tasks. On top of these tasks, the metric assembles the generation probabilities from a pre-trained language model without any model training. Experimental results show that our metric has higher correlations with human judgments than other baselines, while obtaining better generalization of evaluating generated texts from different models and with different qualities.