Draft 3.8: draft in English, deliver in 38 languages

Draft 3.8 is available to all accounts from today. It allows a contract composed in one language to be delivered in any of the 38 languages ZAAN supports, as a parallel document rather than an afterthought translation, with the parity checks we described in April applied clause by clause.

What changed

Delivery languages on the deal-point form. The deal-point form introduced in 3.6 now has a delivery-language field. Select one or more and Draft composes the working-language version as before and, in the same run, a version in each delivery language. The delivery versions are composed, not translated after the fact: Draft generates each clause in the delivery language from the same deal points and the same precedent, then checks the result against the working-language clause for parity.

Parity check on every clause. Each clause pair is checked for the properties we set out in the April research note: same parties and obligations, same conditions, same numbers and dates, same defined terms, same cross-references, no added or dropped qualifiers. A clause that fails parity is flagged in both versions with the specific divergence. In the beta, parity failures ran at about 3% of clauses on first composition, most of them qualifier drift (“reasonable” present in one language and absent in the other) and most resolved by a single regeneration.

Locked glossary. Defined terms are rendered once per delivery language and locked for the document. “Confidential Information” becomes one term in German throughout, not three near-synonyms. The glossary is visible, editable, and carries forward to later drafts in the matter. A firm-level glossary can be set per practice group, so that house renderings of recurring terms are used across matters.

Terms of art without clean equivalents are flagged, not guessed. Some concepts do not travel. “Consequential loss”, “best endeavours”, “indemnify and hold harmless”, “without prejudice” and several dozen others have no settled equivalent in a number of our languages, or have equivalents whose legal effect differs. Draft renders the nearest conventional phrasing, flags the clause at Medium with a short note on the divergence in legal effect, and leaves the decision to the lawyer. We would rather raise the point than bury it in fluent prose.

Prevailing-language clause. Where more than one language version is produced, Draft inserts a prevailing-language clause from the playbook, defaulting to the working language, and flags High if it is removed.

Numbers, dates and currency formatting follow the delivery language’s conventions, with the underlying values held once so that a change in the deal-point form propagates to every version.

Side-by-side layout in Word. Delivery versions can be exported as separate documents or as a two-column bilingual document. Clause numbering is shared across columns and stays aligned under edits.

Confidence tiers per language

Not all 38 languages are equal in our hands, and 3.8 says so in the product. Each delivery language carries a tier:

• Tier A (14 languages, including French, German, Spanish, Mandarin, Japanese and Arabic): composed and reviewed by native-speaker lawyers on our team or under retainer, and parity-checked against a large bilingual corpus.
• Tier B (16 languages): composed and parity-checked, with native-speaker legal review of the glossary and of a sample of clause families rather than all of them.
• Tier C (8 languages): composed and parity-checked, with native-speaker review of the glossary only. Draft recommends full human legal review before delivery and shows that recommendation in the document header.

Tiers are shown in the language selector and in the export. They will move as review coverage grows, and we will publish changes.

What to watch for

Parity is necessary, not sufficient. Two clauses can say the same thing and have different legal effect under different systems of law. Parity tells you the words agree. It does not tell you the delivery-language clause works under the governing law. Review 4.1’s governing-law detection runs on delivery versions too, which helps, but a bilingual contract still needs a lawyer who knows both systems.

Regeneration changes the whole clause pair. Regenerating a flagged clause in one language regenerates its counterpart, to keep parity. Edit by hand if you want to touch one side only; Draft will then re-run the parity check and may flag the divergence you introduced, which is correct behaviour, if occasionally irritating.

Right-to-left delivery languages in two-column layout render correctly, but the column order is fixed with the working language on the left. Separate-document export avoids this.

Known limitations

• Schedules with heavy tabular content are rendered per cell and parity is checked per cell; narrative parity checks do not apply inside tables.
• Tier C languages do not yet support the firm-level glossary; the document-level glossary applies.
• Delivery versions are composed from the working-language draft at the time of the run. Edits to the working draft afterwards are tracked and shown as “delivery out of date” but are not applied automatically.
• Draft 3.8 requires Recall 2.6 or later for precedent citations to survive into delivery versions; accounts on older Recall indices will see working-language citations only.

Upgrade notes

Existing deal-point templates gain the delivery-language field, unset. Firm-level glossaries start empty; knowledge lawyers at beta accounts found that an afternoon on the ten most frequent defined terms covered most of what mattered.

Translation is what you do to a contract after it is finished. Delivery is composing it, in both languages, as one thing.