Translating a Buddhist text
Owner: Kurt Keutzer. Read the full workflow: A Model of the Human Translation Workflow, v2.2 (Google Doc). That document is Kurt’s and stays the reference; it credits the translators and editors who shaped it. This page only summarizes it in our own words and maps its steps to tasks. Kurt has not yet reviewed this page.
Who it is for
Translators of classical Buddhist texts (Tibetan, Sanskrit and others) and their editors. The model is deliberately inclusive, so any one project will use a subset of it.
Steps
Numbers follow the source document.
| # | Step | What happens | Task | AI today |
|---|---|---|---|---|
| 1a–1b | Gather the texts | Find the editions of the root text and its main commentaries in every source language. | new task? (finding editions) | none yet |
| 1c1 | Make e-texts | Turn each edition into clean Unicode text: OCR a scan, or transcribe. | ocr, text-cleanup, spell-check | measured (OCR, Tibetan) |
| 1c2 | Compare editions | Build a variorum edition that records where editions differ; most variants are settled later. | edition-collation | none yet |
| 1d, 2c | Outline the text | Divide the text into chapters, sections and down to the basic unit of translation, so editions and commentaries can be lined up. | toc-extraction, segmentation | measured (TOC, Tibetan) |
| 1e | Gather secondary literature | Collect earlier translations and scholarly studies. | new task? (literature search) | none yet |
| 2a–2b | Decide the approach | Genre, voice and audience of the original; target audience; literal or meaning-led; how many notes; whether to show source terms. | — | human only |
| 3a | Chunk the text | Break the text into translation units. | segmentation | none yet |
| 3b1 | Identify each unit | Is it a scriptural quote (find it, check it), an objection in a debate, an instruction, verse or prose? | citation-mapping | none yet |
| 3b2 | Work through each sentence | Parse grammar and implied subjects; check key terms against the glossary; spot mantras, names, places, texts and dates; flag hard words and look them up. | term-extraction, entity-extraction, new task? (dictionary lookup) | none yet |
| 3b2b–3b2d | Use parallels and commentaries | Check the original if the text is itself a translation, compare parallel versions, read commentaries, resolve variants. | text-translation-alignment, text-commentary-alignment, edition-collation | none yet |
| 3b2e, 3a3 | Render and reread | Write the sentence, then reread the whole unit for flow. | translation | none yet |
| 4 | Build the apparatus | Notes on quote sources, people, places, texts, term choices and parallel passages. | citation-mapping, entity-extraction | none yet |
| 5 | Refine | Revisit the approach with editors, revise each unit, simplify, polish wording, check notes and references. | translation-review | none yet |
| 6 | Independent editing | Copy editing, clarity, house style, and alternative renderings from an editor. | translation-review, spell-check | none yet |
| 7–9 | Publish | Choose print or electronic, mono- or bilingual, then typeset and proofread. | new task? (typesetting, proofing) | none yet |
Notes
- The source document cites Gregory Forgues’ terminology for the steps of translation and Douglas Robinson’s Becoming a Translator.
- Steps 2a–2b and the editor’s judgement in 5–6 are where the translator’s choices live. AI can support them (by surfacing how a term was rendered before, for example) but they are listed as human only.

BuddhistAI Knowledge base