What is Optimization and how does it work?

With the Optimization feature, the source content passes through a Large Language Model routine that lightly alters the source content to increase the likelihood of a high-quality machine translation. If, after analysis, the quality score of the optimized content meets your custom quality threshold, the optimized content is sent for translation. Otherwise, the optimized content is discarded and the original source content is sent for translation.

Optimization is currently available in Salesforce Chat, Salesforce Enhanced Chat, Zendesk Chat & Tickets (Parallel & Inline), and as a parameter in the Language IO API's Translate endpoint, with more integrations to follow. 

Example

Optimization can make a long message more direct and, to a certain extent, through this rephrasing, fix typos that were in the original text.

For example, suppose that an agent enters the long introduction below:

• Original message: "Hi, this is John, Let me give a quik look over on your chat with our Chatbot you had earlier so I can get up to speed, and that way you won’t have to give me all the details you already provided. Please give me just a few seconds and I’ll be right back to contniue our conversation."

Note that in addition to being verbose, the message also contains two typos. Using optimization, the source is processed before translation, resulting in this streamlined version:

• Optimized message: "Hi, this is John. I'm going to look at your chat with our chatbot so I can understand what's going on. That way, you won't have to repeat yourself. Please give me a few seconds and I'll be right back."

As a final step after optimization, a Quality Estimate check kicks in to make sure that the process did not remove too much context. If the similarity between the source content and optimized content comes back lower than the threshold, this means that the optimization process either removed too much content, or changed the wording too much, and it fails back to the original input content. This makes sure that the translated message retains the original's intent:

• Translation: "Bonjour, c'est John. Je vais regarder votre conversation avec notre chatbot pour comprendre ce qui se passe. Ainsi, vous n'aurez pas à vous répéter. S'il vous plaît, donnez-moi quelques secondes et je reviens tout de suite."

Known limitations

In some cases, an optimized result will not be provided:

• As explained in the example above, Language IO scores the quality of the optimized content to ensure that the base meaning is retained. If the optimized message falls below the quality threshold, the original, unoptimized message is sent for translation instead.
• Optimization does not work on original source messages that are less than 12 characters long. You can configure this 12-character limit to a higher count if you need to.
• If the original content is really long, and the server does not return  the request quickly, this could exceed the timeout limit, which cancels the request. This is a rare scenario in Chats.
• If the optimized content that returns is a lot longer than the original content, the original content is used instead. There is a margin to allow for slang or abbreviations to be spelled out properly, but the expectation is that the optimized content should be more concise than the original.

Setting up Optimization QE

Optimization QE is available by default on a per request level (enabling it varies depending on the app), or you can opt to enable it for all requests at the consumer level. Submit a support request to enable it for all requests and define the quality thresholds required.