A Neural Network for Machine Translation, at Production Scale

(Not yet on Arxiv Sanity Preserver, but should be added soon I guess)

This empirical fact is used to their advantage, but it is not proven that this is generally the case (tbh, it might be nearly impossible to prove it). This is okay, but it makes the model seem kinda yucky to me.

> All of the models use 1024 LSTM nodes per encoder and decoder layers.

Ah, as I was reading the paper I kept looking for how they generated a variable-sized attention vector using a simple feed-forward network. It seems that the LSTM is actually capped at 1024 nodes per layer, so the attention vector only can be a fixed size of 1024 and truncated to the number of unrolled steps.

https://www.chinadialogue.org.cn/blog/9258-The-debate-over-t...

One paragraph translated through the Google Translate site:

The Hinckley Point C nuclear power plant project has been controversial in the UK, largely because of its high costs and unproven technology, and the progress made in other low-carbon technologies means that nuclear power plants may become redundant before completion. Energy demand in the UK is declining, and as UK and European power grid connectivity continues to improve and battery technology advances rapidly, it is becoming increasingly obsolete for the UK to require new nuclear power plants to meet basic load requirements.

And from using the translate option inside Chrome:

Hinckley angle C nuclear power plant project in the UK has been highly controversial, mainly due to their high cost and unproven technology; In addition, the progress made in other low-carbon technologies, means that nuclear power plants could be completed before it becomes redundant. Britain's energy demand is falling, and as the United Kingdom and the European grid Unicom continuous improvement and rapid advances in battery technology, Britain needs new nuclear power stations in order to ensure that the basic needs of the load argument has become increasingly obsolete.

Finally, you can see a human-translated English version here: https://www.chinadialogue.org.cn/blog/9258-The-debate-over-t...

Hinkley Point C has been the subject of fierce controversy in the UK because of its cost, its unproven technology, and advances in other low carbon technologies that may make it superfluous even before it is completed. The UK’s energy demand is falling, and, with improved interconnection with European grids and the rapid advance of battery technologies, the argument that Britain needs new nuclear for base load energy is increasingly criticized as out of date.

Original for posterity: 欣克利角C核电站项目在英国一直饱受争议，主要原因在于其高昂的成本以及未经验证的技术；另外，其他低碳技术领域取得的进展，意味着核电站可能在完工之前就变得多余。英国的能源需求正在下降，并且随着英国与欧洲电网的联通不断完善以及电池技术的快速进步，英国需要新的核电站以保证基本负载需求的论点也显得越来越过时。

Note especially how the older translation essentially reverses the meaning of one important phrase -- "nuclear power plants could be completed before it becomes redundant" vs. "nuclear power plants may become redundant before completion."