Machine intelligence.
Human inspiration.
Sentience was founded to provide a tool that didn’t exist.
Financial translation desperately needed a productivity boost to handle the constant flow of publications, but the available software only did one thing right: it matched content that was repeated from previous reports.
In every other sense, it got in the way. Too many workflow steps, too many functions, too many buttons. (Surprisingly, large vendors’ software retains many of these shortcomings even in 2024.)
We heard the same thing throughout the industry: “It’s quicker to translate in a plain Word file.”
With Sentience, we set out to build a tool that made the individual more productive, whether they were translating two sentences or 200.
It would remove low-level tasks from the equation and focus the human on tasks that required context awareness. In short, a split between machine intelligence and human inspiration.
And we would also capture every bit of data for future use by giving users an environment they’d want to work in.
We broke the process down into three interlocking layers: human, database, and neural machine translation (NMT), and built tools to augment each of them.
We help the human by storing data and cutting down on web searches and repetition.
We repurpose that data and automate common tasks to keep the human focused on meaning and context.
And we fact-check and score our NMT output so we can assess its quality, indicate how much it can be trusted, and downplay it where necessary.
A word about AI
We think the role of AI in translation needs to be placed in context.
We think the terms machine translation and artificial intelligence can be unhelpful because they imply replacing the human translator.
And that’s undoubtedly how some firms have framed it: translation as a solved problem, human input no longer needed except for a little post-editing.
But we think as the AI euphoria fades we’ll need to think more in terms of extended intelligence, of how software can help humans do more.
We see NMT as an avatar - the distillation of a human team’s expertise, a less literal, more flexible form of translation memory that can handle straightforward content so humans don’t have to.