When work starts with a new client, there may be past translations to hand, but they are not in the form of a memory. Alignment turns these texts into a memory by pairing them with their sources. The operation is valuable, but when done carelessly it spoils the memory from the very start.
Definition
Alignment is the operation of pairing an existing translation with its source text, sentence by sentence. The result turns into memory records. In this way the accumulated past work becomes usable.
When it is done
It is done at a new client or when a memory has been lost. If past translations exist, they are assessed. The operation is planned at the start of the job.
Automatic pairing
The tools try to pair the sentences in order, and most of the time they succeed. However, the result is not always correct. A check is needed.
A difference in the number of sentences
In the translation, one sentence may have been split or two sentences merged. In that case the automatic pairing slips. It is corrected by hand.
How a slip spreads
A single slip spoils all the pairings that follow. The error is carried to the end of the document. Noticing it early is critical.
Skipped sections
If the old translation has skipped sections, the pairing yields a gap. These gaps are marked. The source is not treated as incomplete.
Added text
The translation may contain explanations that are not in the source. These passages are not paired. They are assessed separately.
Quality assessment
The quality of the translation to be aligned is examined first. A poor translation multiplies the error once it is taken into the memory. An assessment is made.
Selective alignment
Only the reliable sections may be taken into the memory. Taking all of it is not compulsory. The selection is made for a reason.
The need for the source file
For alignment, both the source and the target text need to be editable. With scanned files the operation becomes harder. The files are requested.
Format differences
If the two texts are in different formats, conversion is needed first. Conversion losses affect the pairing. Preparation is carried out.
The cost in time
Alignment takes effort, and how long it takes depends on the volume. This time is included in the plan. It is weighed against the gain.
Calculating the gain
The return on alignment shows in the future match rate. In one-off jobs the return is low. The decision is made according to the volume.
Version mismatch
The source text and the translation may belong to different versions. In that case the pairing misleads. The versions are verified.
The checking step
The result of the alignment is checked by sampling. An unchecked import spoils the memory. The step is not skipped.
Marking
Records coming from alignment can be marked in the memory. In this way the levels of confidence are kept apart. The mark proves useful later.
The confidentiality dimension
The aligned texts are client documents and fall within the scope of confidentiality. The operation is carried out in a protected environment. The rule applies in the same way.
An alternative method
Instead of alignment, it is also possible to use the old translation as a reference only, and in some cases this is safer. With this method the text does not enter the memory; the translator benefits by consulting it. The gain is lower, but the memory is not contaminated.
Common mistake
The most common mistake is to take the output of automatic alignment into the memory without checking it. A single slip produces hundreds of wrong pairs, and these records come back as suggestions in later jobs. The check is always made.