Pipeline flows

Each pipeline has a a flow associated with it. This flow will determine the steps a translation request will undertake, the criteria by which it will take them, and serve other aspects of the translation scheme you wish to setup. 

Flows can be configured on a per-pipeline basic, allowing for customization based on language. You can have different languages have more or less stricter criteria for translation quality, adjusting to both your needs and our translation results.

To check and edit a pipeline flow, head to the respective Pipeline group details, and click on the pipeline.

The flow builder will load in a new modal.

You will find the flow in use in the central area.  A flow is made of different modules which represent steps in the translation flow. Some of these steps are mandatory, others are optional, while some may not be available depending on the content type selected. Others are automatically coupled with additional steps you configure.

To the left, you can find all modules available to the flow. The ones already in use are greyed out, while the ones which have a dependency on another module appear locked.

The different modules stand for:

• Glossary: if you're using a glossary of terms, this step will apply any predetermined translations according to it before the MT tries to translate the text or apply translation memories.
• Translation memories: the flow applies common expressions for which you have a preferential translation, or that have been curated over time from being repeatedly translated on our platform. They help in keeping translations consistent. When a TM is applied, the translation model doesn't act on the selection.
• Machine translation: basic module and the cornerstone of any flow. A MT model will translate the text based on the content type.
• Human edition: an extra step is added where the machine translation is post-edited by our community of editors.
• Quality estimation: you can leverage our state of the art quality evaluation model to help score the quality of a translation after the MT step. Requires Human edition.,
• Decision: you can establish a certain quality threshold over which translations can skip human edition in a flow that contains it. Requires Quality estimation. 

You can obtain more information on modules and their settings by reading this article.

Editing the pipeline flow

You can add any available module to the workflow. Because they need to follow a sequence in the translation flow, modules fit in specific slots. In the example below, we're adding a step of human post-edition to a chat flow, effectively turning it from synchronous to asynchronous.

The addition is performed by dragging and dropping the module into the suggested slot. Once completed, the step is incorporated into the flow.

As you notice, some steps of the flow do more than just affect a translation coming through. Adding the human step introduces a retraining loop based on the results of the human-reviewed output. The properties, specifics and options for the different modules are explained in this article.

Save any changes by clicking the button in the top right corner. You can reset to the previous saved version of the flow at any time by clicking the respective button as well.

Modules can be removed by clicking on them and further clicking on the Remove button, or by hovering and clicking the X close button. Note that you can only remove a module if it no longer has any dependencies.