Skip to content
Searching: All docs

Start typing to search the docs.

Distill

Distill extracts knowledge from professional translation files and creates structured knowledge base articles in the active memory bank’s inbox. Instead of manually reading through a 50,000-segment translation memory or a 30-page client style guide, the AI analyses the content and distils it into actionable articles: terminology decisions, style conventions, client preferences, and domain knowledge.

Supported formats

FormatExtensionWhat the AI extracts
Translation Memory.tmxTerminology patterns, consistent style choices, domain-specific phrasing
MultiTerm termbase.sdltb, .xmlTerm pairs with definitions, domains, usage notes
Word document.docxStyle rules, client preferences, formatting conventions, terminology
PDF.pdfStyle guides, reference material, glossaries, specifications
Excel / CSV.xlsx, .csv, .tsvGlossaries, terminology lists, term pairs
TBX termbase.tbxTerm entries with metadata
Plain text.txtNotes, guidelines, reference material

How to use

From the SuperMemory toolbar

  1. Click the Distill button (⚗) on the SuperMemory toolbar in the Supervertaler Assistant panel
  2. A choice dialog appears with two options:
    • Distill inbox – automatically distils all non-Markdown files (TMX, DOCX, PDF, XLSX, etc.) currently sitting in the active memory bank’s 00_INBOX folder. The button shows how many files are available and lists their names. Disabled when the inbox has no distillable files.
    • Select files… – opens a file picker to choose files from anywhere on disk.
  3. The AI analyses the content and creates draft articles in the active memory bank’s 00_INBOX folder
  4. Review the draft articles in Obsidian, then click Process Inbox to compile them into the knowledge base

Distill always writes into the active memory bank – the one currently selected in the toolbar dropdown. To distil into a different bank, switch the dropdown first.

From the termbase list (shortcut)

You can distil a termbase directly from the termbase settings without exporting it first:

  1. Open Settings → TermLens to see your termbase list
  2. Right-click any Supervertaler or MultiTerm termbase
  3. Select ⚗ Distill into memory bank from the context menu

The plugin reads all terms from the termbase, formats them as a structured table, and sends them straight to the Distill pipeline. Draft articles appear in the active memory bank’s 00_INBOX folder – review them in Obsidian, then run Process Inbox as usual.

What the AI produces

Depending on the source material, Distill creates one or more Markdown articles containing:

  • Terminology decisions — terms the translator consistently chose, with reasoning inferred from context and usage patterns
  • Style profile — register, voice, formatting conventions, and writing patterns observed across the translations
  • Client preferences — conventions specific to the client or project (e.g. “always use ‘Schedule’ instead of ‘Appendix’ in procurement documents”)
  • Domain knowledge — subject-matter conventions, technical vocabulary, and common pitfalls identified from the source material

Example: Distilling a TMX

A translation memory with 10,000 Dutch-English legal segments might produce:

  • A terminology article listing the key legal terms with the translations used and why (e.g. “overeenkomst → agreement (not contract), because the client uses ‘contract’ only for formal notarial documents”)
  • A style article noting the register (formal, third-person, passive voice) and formatting conventions (numbered clauses, capitalised defined terms)
  • A domain article capturing Dutch legal system conventions relevant to translation (e.g. “Dutch notarial acts use specific formulaic language that should be preserved, not naturalised”)

Example: Distilling a client style guide

A 20-page Word document from a client might produce:

  • A client profile with their language preferences, terminology decisions, and contact details
  • A style article with their formatting rules, preferred register, and localisation conventions
  • A terminology article with their approved terms and rejected alternatives

Tips

  • Start with your most important client. Distill their largest TM first — you’ll immediately see the value as the AI surfaces terminology patterns you may not have been consciously aware of.
  • Combine sources. Select a client’s TM, their style guide PDF, and their glossary Excel file together — the AI cross-references them to produce richer articles.
  • Review before processing. Distill outputs draft articles to the inbox, not directly to the knowledge base. Always review them in Obsidian before running Process Inbox.
  • Large files are truncated. Very large TMX files (100K+ segments) are automatically truncated to fit the AI’s context window. For best results with huge TMs, export a representative subset first.

See Also