The translation memory (TM) is the basic tool that provides both speed and consistency in corporate jobs. But its reliability as a tool depends on what is put into it: a wrong record is carried into all later jobs. That is why the memory is prepared at the start of the job and receives regular maintenance.
What the memory does
The memory stores previous translations at sentence level and, when the same sentence comes up, suggests the previous equivalent. This brings a large gain in repetitive documents. The tool is standard in corporate jobs.
Separation by client
The memory is kept per client because every organisation has its own language. Using a shared pool gives rise to mixing between organisations. The separation is built in structurally.
Feeding from previous jobs
The client's old translations can be imported into the memory, and this import is done in a controlled manner. Text of unknown quality is not taken in directly. It is examined.
Alignment work
Old translations are taken into the memory by matching them with the source text, and this operation is called alignment. Faulty alignment produces wrong matches. The result is checked.
Match thresholds
The similarity rate from which the tool will offer a suggestion is set. A low threshold produces unnecessary suggestions. The setting is made according to the job.
Exact and fuzzy matches
An exact match is an identical sentence; a fuzzy match is a similar sentence. The two call for different attention. Fuzzy matches are checked separately.
Verifying the context
The equivalent in the memory may not fit the new context and is not accepted automatically. Every suggestion is assessed in its context. The tool does not decide.
Matches containing numbers
In sentences where the same pattern is repeated with different numbers, the memory may carry over the old number. This error is seen especially in tables. Numbers are verified separately.
Memory maintenance
Wrong records are cleaned out and outdated equivalents are corrected. An unmaintained memory multiplies errors. The operation is carried out regularly.
Version and date information
When the records were added is tracked, and old records are used with care. Terminology changes over time. The date information helps.
Relationship with the terminology list
The memory holds sentences, the terminology list holds words, and the two work together. Their functions are not confused. The two are managed separately.
Shared memory in multi-person jobs
More than one translator uses the same memory, and consistency is ensured in this way. Arrangements for simultaneous access are set up. Clashes are prevented.
The confidentiality aspect
The memory stores the client's texts and falls within the scope of confidentiality. Access is restricted and it is not shared. The rule is defined in the contract.
Delivering the memory
If the client wishes, the memory file can be delivered, and this is regulated in the contract. The question of ownership is made clear at the outset. The decision is obtained in writing.
Distinction from machine translation
The memory stores sentences previously translated by a human and is different from machine translation. The two concepts are not confused. The distinction is explained to the client.
Passing the gain on to the client
The gain arising from the repetition rate is shared transparently. Concealing it loses trust. The measurement is shown.
Back-up
The memory is backed up regularly because its loss wipes out years of accumulation. The back-ups are subject to the same access restriction. The arrangement is defined.
The next step
Once the tools are prepared, the files are put into a form suitable for working on. File preparation is the last technical step. The order is kept.