Introduction to the special issue on 'Automated Metrics for Machine Translation Evaluation'

摘要

The development of fully-automated programs that can effectively assess the translation quality performance of Machine Translation (MT) systems has blossomed into a vibrant research area in recent years. The broad interest in such "automated metrics for MT evaluation" is an outcome of the growing recognition of the role such metrics play in the development of high-performance MT systems. While human assessments of various forms serve as the ultimate arbitrators of translation quality, the training and development of modern MT systems critically rely on automated metrics. The increasingly dominant data-driven machine-learning-based paradigm for Machine Translation is inherently dependent on automated translation quality measures which can serve as target functions for optimizing MT system performance during training. It has been increasingly recognized in recent years that perhaps the single most important enabling factor of modern MT research progress has been the advent and broad adoption of IBM's BLEU metric.