Understanding Translation Review Reports

Translation Review Reports help you track the quality of your AI translations and see how much human reviewers need to edit them. Use these reports to optimize your translation workflow and demonstrate quality improvements over time.

• When you want to measure translation quality and approval rates

• When optimizing your review agent settings

• When tracking reviewer efficiency and workload

• When demonstrating translation quality to stakeholders

• When identifying language pairs that may need attention

Go to Apps in the left sidebar navigation in your workspace, then click the Translation Review Report tile to open it.

📌 The Apps section appears only when one or more custom apps are enabled in your workspace. Custom apps are currently in Beta.

The translation review report lets you review specific documents to see exactly what changed between revisions, such as:

• AI translation vs. review agent edits

• Review agent edits vs. human reviewer changes

• Side-by-side comparisons with explanations for each edit

The Translation Review Report shows aggregated trends across your entire workspace: approval rates, reviewer time, language pair performance, and project/document tables. Use it to track quality over time and identify areas for improvement.

The Project Report drills into a single project's review results. It shows the project's overall approval rate, source words, document count, and turnaround time, along with a document table you can expand by language pair. Open it from the Projects section of the Translation Review Report, or stand-alone by selecting a project manually.

The Document Edit Report shows exactly what changed between two revisions in a single document. For example, the AI translation vs. a review agent's edits, or a review agent's edits vs. a human reviewer's final pass.

Approval rate measures how often human reviewers accept AI translations without making any changes.

Only segments that were intended for human review are included in this calculation. Documents that are fully automated (translated and finalized by AI and agents without human involvement) are not counted.

If you don't manually approve every job, some are automatically verified after a set period (typically 2 weeks). By default, these auto-approved jobs are excluded from approval rate calculations so the metric reflects real human reviewer behavior.

Your translation workflow typically follows this sequence:

1. AI translation creates the initial translation

2. Review agents (0-3 agents) may edit the translation automatically

3. Human reviewers either approve the final result or make additional edits

The reports track how much each stage contributes to the final translation.

Volume charts show:

• Documents completed

• Total segments processed

• Number of active projects

Edit breakdown shows the percentage of segments in each category:

• Approved without edits (AI translation accepted as-is)

• Agent-only edits (review agent made changes, human approved)

• Human edits (human reviewer made changes)

Approval rate by pipeline compares performance across different configurations:

• AI only (no review agents)

• AI + 1 review agent

• AI + 2 review agents

• AI + 3 review agents

The translation review report highlights your top language pairs by volume and flags any that are performing unusually well (green) or poorly (red) compared to their normal range.

• Language pair — Focus on specific language combinations

• Date period — Choose your time range (default: last 6 months)

• AI pipeline — Compare different agent configurations

• Reviewer — Filter by specific human reviewers or agents

1. Select project — Choose which project to analyze

2. Select document — Pick a specific document within that project

3. Select revision — Choose which before/after comparison to view

The document edit report shows four columns:

• Source — The original text to be translated

• Before — The translation from the earlier stage

• After — The translation from the later stage, with changes highlighted

• Comment — Explanation of why each change was made

Green text shows additions, red strikethrough text shows deletions. Use the Show deletions checkbox to hide the strikethrough formatting and see only the final text.

• Review agents automatically explain every edit they make

• Human reviewers can optionally add comments to explain their changes or flag questions

To save or print your Translation Review Reports:

1. Click the book icon in the top right corner to enter presentation mode

2. Click the PDF button, or press Cmd+P (Mac) or Ctrl+P (Windows)

3. Save as PDF or send to your printer

• Review which agents are making the most edits and consider adjusting their settings

• Look for patterns in human edits to identify areas where agents could be improved

• Check if certain content types or language pairs consistently need more editing

• Use the reviewer time metrics to identify bottlenecks

• Consider enabling more review agents for language pairs with high edit rates

• Look for reviewers who are consistently faster while maintaining quality

Check if you have review agents enabled that might be making unnecessary edits. Look at the agent efficiency table to see which agents are editing the most segments. You may want to adjust their settings or disable agents that aren't adding value.

This could indicate differences in:

• Content types being translated

• Review agent configurations for different pairs

• Individual reviewer preferences or expertise levels

Use the outlier highlighting and drill down to specific documents to investigate.

N/A appears when there were no human-reviewed segments for that language pair in that month. This usually happens when:

• The language pair was added partway through the period

• All jobs for that pair were auto-verified rather than manually reviewed

Reviewer time per 1,000 words shows as N/A when:

• There were no human-reviewed segments that month

• The volume was too small to calculate a reliable average

• Most jobs were auto-verified instead of manually reviewed

No, you can only compare sequential stages in your workflow. For example, if your workflow is AI → Review Agent → Human Reviewer, you can compare:

• AI translation vs. Review Agent edits

• Review Agent vs. Human Reviewer edits

But you cannot skip stages and compare AI translation directly to Human Reviewer edits.

Translation Review Reports are available to workspace administrators.

• Project Reports — Track progress and status for individual projects

• Smartwords Usage — Monitor your AI translation consumption

• Quality Assurance — Set up automated quality checks for your translations