摘要
由一名新闻记者-机器人与机器学习的工作人员新闻编辑-每日新闻-关于机器翻译的最新研究结果已经发表。根据《新闻周刊》编辑在中国昆明的新闻报道,研究表明:“近年来,深度模型的应用显著提高了神经机器翻译(NMT)的性能。然而,数据分布的不均匀性带来了严峻的挑战。”国家自然科学基金(NSFC)、云南省科技专业专项计划、云南省基础研究计划等资助项目。我们的新闻记者从昆明大学的研究中得到一句话：“具体地说，低频词严重影响翻译绩效。特别是在资源贫乏的语言翻译中，低频词的训练不足。为了解决这个问题，我们提出了一个建议。”提出了一种简单的动态路由注意(Dynamic Routing Attention,D RA)方法,在处理不同的词时,DRA根据词频和源句法动态调整自我注意权重,使编码器的自我注意能够集中在周围词和句法为关联的词上,而不是只关注当前词,从而提高了编码效率。以Transformer RPR模型为基线模型,在WMT14英德、IWLST14英德、IWLST14英维特纳和TED Talk Thai-China任务上进行机器翻译实验,验证了该方法的有效性。
Abstract
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News-Current study results on Machine Trans lation have been published. According to news reporting out of Kunming, People's Republic of China, by NewsRx editors, research stated, "In recent years, the ut ilization of deep models has significantly enhanced the performance of neural ma chine translation (NMT). Nevertheless, the uneven distribution of data leads to critical challenges." Financial supporters for this research include National Natural Science Foundati on of China (NSFC), Yunnan Provincial Science and Technology Major Special Proje ct, Yunnan Provincial Basic Research Programme Project. Our news journalists obtained a quote from the research from Kunming University, "Specifically, lowfrequency words severely affect translation performance. Thi s is especially in low-resource language translation, where the training of low- frequency words is inadequate. To address this issue, we use syntactic and word frequency information to enhance the performance of encoding representations of input sequence. we propose a simple approach called Dynamic Routing Attention (D RA). When processing different words, DRA dynamically adjusts the Self-attention weight based on word frequency and source syntactic, which enables the encoder Self-attention to focus on the surrounding words and the words with syntactic as sociations rather than the current word solely. Consequently, our method improve s the representation capability of the encoder in processing sentences containin g low-frequency words. Using Transformer RPR model as a baseline model, we demon strate the effectiveness of our method with the experiments on machine translati on tasks of WMT14 English-German, IWLST14 English-German, IWLST14 English-Vietna mese, and TED talk Thai-Chinese."
NETL
NSTL