Aligning tokens and translations

What an alignment records

An alignment links a source token or span with a translation span. It can help a reader jump between a line and an English rendering, audit omissions and see where one phrase has been moved. It is a teaching map, not a proof of semantic equivalence. Coptic SCRIPTORIUM explicitly distinguishes token and translation annotations; its translation layer can cover a small sentence or partial sentence rather than one English word per Coptic token.

For Caesar’s short clause, the three-token span Gallia est omnis can correspond to “Gaul as a whole is”. A readable English version reorders omnis and distributes the Latin predicate across English words. Mark that relation as many-to-many. Do not force est into a separate English word if the target sentence uses a different construction.

A useful alignment table

Give each source token a stable reference to the cited edition. Add the source span, English span, relation type and an explanation for additions or omissions. Distinguish a supplied word needed for English grammar from information imported from historical context. If a translation combines two source clauses, align at phrase or clause level before attempting tokens. Mark uncertain spans rather than drawing a confident link through a gap.

Token counts depend on segmentation. If two editors divide a bound group differently, their alignment tables are not directly comparable. Store the segmentation policy with the alignment and retain the original string.

Practice

Align your close and readable versions of the Caesar clause to Latin spans. Draw one-to-many and many-to-one links where needed. Write one sentence explaining why exact word-for-word alignment would distort the grammar.