arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2210.12910cs.CLcs.AI
收藏

通过惩罚低互信息来专门化多域NMT

Specializing Multi-domain NMT via Penalizing Low Mutual Information

KAIST(韩国科学技术院);NAVER Corp.(NAVER公司)

Jiyoung Lee, Hantae Kim, Hyunchang Cho, Edward Choi, Cheonbok Park

摘要:

多域神经机器翻译(NMT)训练具有多个域的单个模型。它之所以吸引人,是因为它在一个模型中处理多个域的有效性。一个理想的多领域NMT应该同时学习不同的领域特征,然而,掌握领域特性是一个不平凡的任务。在本文中,我们研究领域特定的信息通过镜头的互信息(MI),并提出了一个新的目标,惩罚低MI变得更高。我们的方法达到了国家的最先进的性能之间的竞争,目前的多域NMT模型。此外,我们的经验表明,我们的目标促进低MI更高,导致领域专业化的多领域NMT。

英文摘要:

Multi-domain Neural Machine Translation (NMT) trains a single model with multiple domains. It is appealing because of its efficacy in handling multiple domains within one model. An ideal multi-domain NMT should learn distinctive domain characteristics simultaneously, however, grasping the domain peculiarity is a non-trivial task. In this paper, we investigate domain-specific information through the lens of mutual information (MI) and propose a new objective that penalizes low MI to become higher. Our method achieved the state-of-the-art performance among the current competitive multi-domain NMT models. Also, we empirically show our objective promotes low MI to be higher resulting in domain-specialized multi-domain NMT.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑