For years, large clients have been attempting to stream content for translation. And now, it has finally happened—the content for translation is no longer coming from a CMS; it is flowing directly from development systems in small, contextless chunks. Every second sentence presents a conundrum that takes more time to solve than it should, and often proves impossible to resolve.

When translators used to emphasize the importance of “context, context, context,” the client community largely dismissed this point. However, we have now reached a stage where language models are echoing the same plea using the same words. Text is not a roll of toilet paper; it contains extrasegmental references, links, outer context, and larger context on multiple layers.

When translating, one must think instead of simply relying on a trained neural model to automatically select the most frequently used words in similar contexts. Some send an entire large XML file with only one changed string to provide context for the translator. However, the translator would need to read much more than one sentence to grasp the context, resulting in decreased productivity for a reason.

Another issue arises in the form of statistical quality control. When translating very small text chunks, they must be perfect; otherwise, the risk for the consumers cannot be controlled. This means that the streaming process places an unnecessary burden on the requirements for production translation quality. Simply put, when working with several pages of text, every sentence does not need to be flawless. However, with streaming content, each sentence must be impeccable.

This situation has gone too far, and the industry needs to take a step back and recognize that it is impossible to appreciate a pie one breadcrumb at a time. One would miss the entire wonder of the pastry, which should be prepared, admired, and consumed as a whole, rather than as leftover scraps.