arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2104.08677cs.CLcs.LG
收藏

从完全训练到完全随机嵌入:用紧凑的词嵌入表改进神经机器翻译

From Fully Trained to Fully Random Embeddings: Improving Neural Machine Translation with Compact Word Embedding Tables

Thomson Reuters(汤森路透);Amazon(亚马逊);Huawei Noah’s Ark Lab(华为诺亚方舟实验室);McGill University(麦吉尔大学)

Krtin Kumar, Peyman Passban, Mehdi Rezagholizadeh, Yiu Sing Lau, Qun Liu

摘要:

嵌入矩阵是神经自然语言处理(NLP)模型中的关键组件,负责提供输入标记的数值表示。\footnote{In this paper words and subwords are referred to as \textit{tokens} and the term \textit{embedding} only refers to embeddings of inputs.}在本文中,我们分析了神经机器翻译(NMT)的背景下,这种矩阵的影响和效用。我们发现,从词嵌入中提取句法和语义信息,并使用随机嵌入运行NMT系统,并不像最初听起来那样具有破坏性。我们还展示了如何从经过充分训练的嵌入中只整合有限数量的特定于任务的知识,从而提高NMT系统的性能。我们的研究结果表明,在交换性能可以忽略不计的恶化,任何NMT模型可以运行部分随机嵌入。使用这种结构意味着最小的内存需求,因为不再需要存储大型嵌入表,这在工业和设备设置中是一个显着的增益。我们在将{English}翻译成{German}和{French}时评估了我们的嵌入,并实现了$5.3$ x压缩率。尽管有一个相当小的架构,我们的模型在某些情况下甚至能够超越最先进的基线。

英文摘要:

Embedding matrices are key components in neural natural language processing (NLP) models that are responsible to provide numerical representations of input tokens.\footnote{In this paper words and subwords are referred to as \textit{tokens} and the term \textit{embedding} only refers to embeddings of inputs.} In this paper, we analyze the impact and utility of such matrices in the context of neural machine translation (NMT). We show that detracting syntactic and semantic information from word embeddings and running NMT systems with random embeddings is not as damaging as it initially sounds. We also show how incorporating only a limited amount of task-specific knowledge from fully-trained embeddings can boost the performance NMT systems. Our findings demonstrate that in exchange for negligible deterioration in performance, any NMT model can be run with partially random embeddings. Working with such structures means a minimal memory requirement as there is no longer need to store large embedding tables, which is a significant gain in industrial and on-device settings. We evaluated our embeddings in translating {English} into {German} and {French} and achieved a $5.3$x compression rate. Despite having a considerably smaller architecture, our models in some cases are even able to outperform state-of-the-art baselines.

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。

↑