面向直接语音翻译的命名实体检测与注入
计算与语言
2023-03-14 v2
摘要
在句子中,某些词对其语义至关重要。其中,命名实体(NEs)对神经模型而言是出了名的难题。尽管命名实体十分重要,但其在语音到文本(S2T)翻译研究中未得到准确处理,近期工作表明 S2T 模型在地名尤其是人名上表现不佳,除非事先已知,否则其拼写颇具挑战性。本工作中,我们探索如何利用已知可能出现在给定上下文中的命名实体词典来改进 S2T 模型输出。我们的实验表明,从 S2T 编码器输出出发,我们能够可靠地检测话语中可能存在的命名实体。事实上,我们证明了当前的检测质量足以将翻译中的命名实体准确率提升,使人名错误减少 31%。
引用
@article{arxiv.2210.11981,
title = {Named Entity Detection and Injection for Direct Speech Translation},
author = {Marco Gaido and Yun Tang and Ilia Kulikov and Rongqing Huang and Hongyu Gong and Hirofumi Inaguma},
journal= {arXiv preprint arXiv:2210.11981},
year = {2023}
}
备注
\c{opyright} 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works