Translated携手意大利顶级超算，对标DeepL与谷歌翻译

Translated Taps Italy’s Top Supercomputer to Compete with DeepL, Google Translate

Rome-based language AI and services provider, Translated, has announced that it has partnered with NVIDIA and Italy’s largest supercomputing center, CINECA, to power its next generation of machine translation technology powered by AI.

位于罗马的语言人工智能和服务提供商Translated宣布，已与英伟达（NVIDIA）以及意大利最大的超级计算中心CINECA合作，为其下一代由人工智能驱动的机器翻译技术提供动力。

CINECA — a not-for-profit consortium made up of 70 Italian universities and over 50 research and public institutions — will support Translated with 10 million GPU hours of model training. Translated plans to open source the technology between English and Italian for commercial use.

CINECA是一个非盈利联盟，由70所意大利高校和50多个研究及公共机构组成，将为Translated提供1000万GPU小时的模型训练支持。Translated计划将英意互译技术开源，供商业使用。

The news was revealed at a live-streamed event hosted by Translated and attended by industry analysts and investors.

Translated在其主办的直播活动中宣布了这一消息，该活动吸引了众多行业分析师和投资者的参与。

Speaking at the event, Francesco Ubertini, President of CINECA said, “the aim [of the collaborative agreement] is to make further steps in translation technologies with AI, and to train the most advanced LLM In the world by joining forces.”

CINECA总裁Francesco Ubertini在活动中表示：“本次合作旨在利用人工智能在翻译技术上迈出更大步伐，联合各方力量训练世界上最先进的大型语言模型（LLM）。”

“We believe we can do the next step because of the high quality data set that Translated has, […] and our unique computing power that we make available to the project,” he added. 

他补充道：“我们相信，凭借Translated拥有的高质量数据集和我们为项目提供的独特的计算能力，我们可以取得新突破。

Translated’s CEO, Marco Trombetti, elaborated on the high-quality data set: “For fifteen years we have been collecting data about what is wrong and what is right [in translations]. […] We have unlocked the discussions and fights between translators and reviewers. […] All those fights are the perfect data set to create a chain of thought, to teach a machine about translation,” he said.

Translated的首席执行官Marco Trombetti在直播活动中对高质量数据集进行了详细说明：“十五年来，我们一直在收集翻译中的正误数据。我们深入挖掘了翻译者与校对者之间的讨论和争议。所有这些争议构成了创建思维链条、教授机器翻译的完美数据集。”

Trombetti unveiled the company’s new technology, Lara, at the live-streamed event, which is the culmination of 1.2 million hours of GPU training based on Translated’s data set. The free access tool provides users with reasoning behind its translation decisions.

在直播活动中，Trombetti还发布了公司的最新技术——Lara，这是基于Translated的数据集，经过1200万GPU小时训练的成果。这款免费工具能向用户展示其翻译决策的推理依据。

With the Lara launch, Translated is entering into the consumer translation AI market dominated by incumbents like DeepL, Google Translate, and Microsoft Translator.

随着Lara的推出，Translated正进入由DeepL、谷歌翻译和微软翻译等现有企业主导的AI翻译消费市场。

Trombetti also demoed Lara Live, an AI-powered consecutive interpreting solution, and announced the launch of Matesub and Matedub — two AI-powered tools for multilingual subtitling and dubbing. 

Trombetti还展示了名为Lara Live的人工智能连续传译解决方案，并宣布推出Matesub和Matedub两款多语言字幕和配音的人工智能工具。

“We want to thank NVIDIA for this. We’ve been working with them for years now, and they gave us access to the latest infrastructure,” Trombetti concluded.

他同时表示：“我们要感谢英伟达的支持。多年来，我们与他们保持着紧密的合作关系，他们为我们提供了最先进的基础设施。”

Jonathan Cohen, VP of Applied Research at NVIDIA, told the audience, “Our contribution is our expertise in computer science. […] We try to solve these computer science problems that allow companies like Translated to build applications at the top of those problems.”

英伟达应用研究副总裁Jonathan Cohen向听众表示：“我们的贡献在于我们在计算机科学方面的专长。我们致力于解决计算机科学问题，使得像Translated这样的公司能够在这些问题的基础上构建应用程序。”

“I think this [technology] is going to change society in ways that we can barely anticipate. We’re already getting used to having pocket translators in our phones […] but that barely scratches the surface,” he concluded.

他总结道：“我认为这项技术将以我们几乎无法预料的方式改变社会。我们已经习惯了使用手机中的便携翻译软件，但这只是冰山一角。”

“For fifteen years we have been collecting data about what is wrong and what is right [in translations]. […] We have unlocked the discussions and fights between translators and reviewers. […] All those fights are the perfect data set to create a chain of thought, to teach a machine about translation,” — Translated, CEO, Marco Trombetti

“十五年来，我们一直在收集翻译中的正误数据。我们深入挖掘了翻译者与校对者之间的讨论和争议。所有这些争议构成了创建思维链条、教授机器翻译的完美数据集。”——Translated首席执行官Marco Trombetti

AI同传的下一步？

AI SI Next?

Alongside the multiple product releases that Trombetti announced at the event, the CEO teased the audience with a sneak peek of a real-time simultaneous interpreting video conferencing solution currently in the works.

在活动中，除了发布的多个产品，Trombetti还向观众展示了正在开发中的实时同声传译视频会议解决方案的预览。

“It may work, it may not work. We will show you next year. But we want to show you what this may look like in the future,” he stated.

“它能否成功实现到明年才会揭晓。但我们想提前展示未来可能的样子。”他说道。

Following the event, Trombetti told Slator, “Translated is planning to invest circa USD 100m in AI in the next two years, and most of the funding has been already secured.”

活动结束后，Marco Trombetti向Slator透露：“Translated计划在未来两年内投资约1亿美元用于人工智能研发，而且大部分资金已经到位。”

The company is already working on the next generation of Lara — Lara Grande — in the company’s quest to reach “language singularity”, i.e., the top 1% of translation quality.

同时，公司已经在研发下一代Lara——Lara Grande——作为公司追求“语言奇点”即翻译质量达到顶尖1%的目标的一部分。

“Creating the universal translator will unlock a lot of demand in translation. If we solve this problem we can unlock an incredible amount of content,” he concluded.

Trombetti总结道：““创建通用翻译器将释放翻译领域的巨大需求。如果我们能解决这个问题，我们可以释放出难以置信的内容量。”

原文链接：https://slator.com/translated-taps-italys-top-supercomputer-to-compete-with-deepl-google-translate/