Searchspeare

Extract rules from previous work ​

Searchspeare can suggest reusable AI rules from bilingual examples. This helps you identify patterns in approved work, such as forms of address or recurring wording conventions.

The result is a set of suggestions to review. A repeated choice in one text is not automatically a requirement for every future project.

Use Extract from Context ​

Open Searchspeare → Rules in a translation workspace. When bilingual context is available, choose Extract from Context. The action uses the context supplied by the current workspace; it is not a general search through every memory or project.

When suggestions appear:

  1. Read each instruction and example pair.
  2. Deselect suggestions that are inaccurate, too broad, or specific to a one-off sentence.
  3. Keep only requirements you actually want to reuse.
  4. Save the selected rules.
  5. Open the saved rules to check their languages, labels, enabled state, and wording.

Suggested rules start selected in the review list, so inspect the selection before saving. Selecting none does not save an empty batch.

Choose examples for rule extraction ​

Choose well-reviewed bilingual material that demonstrates the intended requirement. A target typo, inconsistent naming decision, or unusual sentence can otherwise be mistaken for a convention.

The suggested language pair comes from the context, and extracted rules can carry suggested labels. Verify those labels against the project-applicability behavior, especially the treatment of unlabeled projects.

Check the instruction, language direction, and complete example pair before saving a reusable rule.
Check the instruction, language direction, and complete example pair before saving a reusable rule.

Edit and test suggested rules ​

An extracted instruction may need a clearer condition or exception. Edit it in the normal rule form, retain a complete example, and test it on both a matching case and a case that should remain unaffected.

Saving a rule does not change translations

Saving selected suggestions creates reusable rules. It does not apply translations, accept a review, or create QA findings. Follow Translation and review behavior when you want to use the saved requirements.