Translation Memory Sharing
http://www.thebigword.com/
Get the List of 4,400+ Translation Agencies Now! No Recurring Membership Fees!
Introduction
Reasons for Sharing Memories
The reason for sharing memory is
to enable the use of more than one vendor.
By centralising your translations,
thebigword ensures the highest quality and consistent
service that your consumers and regulators expect.
The reasons a client may want to
do this are:
- To keep costs down by playing
vendors against each other, with neither having
a memory advantage.
- To enable two vendors that have
different specialist areas (software, web sites,
marketing) to share the translatable content.
- To spread the work and avoid having
all eggs in one basket with complete reliance on
one vendor.
- To feel in control of the memory
database.
The reasons a vendor may want to
do this are:
- To gain a commercial advantage
over other vendors. As the custodians of the TM,
the vendor should use this position as leverage
to win further work with the aim of relegating other
vendors to the role of subsidiary suppliers.
- To demonstrate technical advantages
over other vendors.
How Memories Can be
Shared
Sharing Memory Files
Memories can be shared by having a central memory
file that is managed by one of the vendors or the
client.
When vendor 1 receives a translation job it would
download the whole memory. It would do any necessary
pre-processing, then analyse the job against the downloaded
memory. Likewise when vendor 2 receives a translation
job it would do the same.
On completion of the translation jobs the new or
changed memory units need to be cleaned into the main
memory. This job should be the responsibility of one
of the vendors as the main custodian of the memory,
or, if the specialist resource is available, the client.
Server-Based
Memories
Memories can be made available through products
such as Trados TM Server.
Following receipt of jobs from a client, the vendors
will pre-process and analyse against the memory in
the normal way, but will use the on-line memory for
analysis.
Translation-ready files are passed to the translators
with log-in information to access the on line memory.
The translators then translate in the normal way,
but instead of using a local copy or subset of the
memory, they link to the full on line memory.
The memory server as above is based on a relational
database and must be a high-availability system.
Advantages of Memory Sharing
If managed correctly, translation units (sentences
and phrases) created and managed by one vendor can
be used by another vendor, reducing costs. Additionally
by merging the memory resource there is no longer
any duplication of memory maintenance overheads. Therefore
there should be a reduction in the running costs of
the memory.
There is a perception that by sharing memories on
line (or otherwise) that this will make their contents
available for general use, such as in the authoring
process. It should be noted however that translation
memory tools such as Trados are built for daily use
professional translators; the software is not designed
for ease of client use for this purpose.
Issues
There are a number of issues that must be overcome
for memory sharing to work successfully. Some of these
are simple to resolve, but others pose more of a challenge.
Management of Memory Server
The memory server is a highly available, highly
scalable database. It requires multiple user access
over the internet so has high bandwidth requirements.
Such a server has a high cost of ownership. Costs
to be considered for such a server are:
- Server software (typically 100,000
or more)
- Hardware (database server plus
web server)
- Hosting and communications
- Support for translators and vendors
connecting (potentially 24 hour support)
- training for translators/vendors
- Administration of connection credentials
(server software is licensed per concurrent user)
The running of the shared memory
also places additional responsibilities on the owners
of the system:
- Maintaining 24 hour access to
the service. A system failure could have disastrous
consequences for the 2nd vendor.
- Access rights and access dispute
resolution (i.e. which vendors users have priority
if any)
- Censure for misuse removal of
access rights (poor quality etc.). Who is responsible
for this and what is the range of their powers?
Many vendors are prepared to make
such investments as above, but cost of ownership increases
by third party (other vendors) having access to the
server. The question of who pays for this infrastructure
(and how the payment is structured) needs to be considered.
It should be noted that where there
are two or more vendors working for the same client
and only one pays for the shared memory infrastructure,
the others are at an advantage as they are in effect
getting this, very powerful, facility for free.
From the clients' perspective this
potential for dispute among sub-contractors can be
very damaging and detrimental to projected workflow
and may require the client to take a more "hands-on"
role that was originally perceived.
Benefits of Using Single Supplier
By making a memory available to all
vendors would seem to remove the benefit of using
a single translation vendor. There are other benefits
of using a single translation vendor and these are
collectively far more important than that of memory
access. These are such factors as:
- Pre and post processing of files
- Vendors gain expertise in the
efficient processing of client files. These processes
prepare files for memory analysis and translation.
Not only can these processes affect the content
segmentation (the way sentences and phrases are
split to be recognised by the memory), but highly
efficient processing of batches of files should
not be wasted by splitting files amongst vendors.
Many pre and post processing tools are proprietary
to vendors, so although memories would be shared,
processing techniques would not. DTP and other such
post processes can be inextricably intertwined with
the translation process so care must be taken when
vendor job batches are created.
- Project Management Continuity
- On vendor provides one point of
contact. Most clients value this highly.
- Review Processes
- As important as the translation
process is the review process. Reviewer amends that
do not make it back into the memory compromise that
memory. Vendors take widely different approaches
to review management and leading vendors have proprietary
review management procedures and tools. A consistent
and robust approach must be adopted, which is difficult
across multiple vendors.
- Query Management
- Part of managing a translation
job is management of queries. This is where a translator
or reviewer is not sure of a term or the context
of a sentence and needs to ask a question of the
client. One question is asked and the answer passed
to all translators in all languages and if necessary
glossaries are updated. If the content going to
two translation vendors is truly similar (similar
enough to make memory sharing worthwhile) then this
will create double query management and potentially
inconsistent results. Communications structures
need to be put in place between all vendors so that
this does not become problematic.
Quality Responsibility and Dispute
Resolution
When quality of translation is in
question, it should be obvious at any time where responsibility
lies. Mixing memories means mixing vendors product
in one database for use at any time in any subsequent
product by any vendor. Thebigword are highly cautious
about taking on memories from other vendors when we
take over from that vendor since bad memory units
can cause more delays, quality issues and expense
than if the memory was not used at all.
When a document is produced by either
vendor it will, if a shared memory has been used,
contain input from both vendors. If quality problems
arise from this document the responsibility must lie
with the vendor that delivered the document, regardless
of where any individual memory units originated. This
may result in higher prices with vendors taking more
time to check matches over which they had no initial
control.
Management of Memory Units
With server based memories, there
is little or not hierarchical control over translation
units entering the memory. This means that any translator
from any vendor can add and amend memory units without
checking. Even within a single vendor, this poses
a problem, but can be partially overcome with rigorous
translator selection and continuous review. Taking
on this role on behalf of another vendor becomes impossible.
Conclusion
While memory sharing appears to be a good idea from
the clients perspective by removing vendor dependence,
in reality dependence on a vendor is not related to
who is the custodian of the memory at any time. There
are much more compelling reasons to settle on one
vendor for one content type.
Where different vendors translate different content
types, the sharing of memories can be evaluated, but
there are many issues to be considered. In most cases,
a better practice would be to keep the vendor procedures
apart and pass memory files between the two periodically.
If memories are shared, one vendor (or if resources
are specialist enough, the client), must take overall
responsibility for the memories. If this is one of
the vendors, we would recommend the other vendor should
be treated as a sub contractor for process purposes.
|