TRANSLATION
We have multiple iterations of translation and localization processes, all of which can be fine-tuned to any given project parameters. The universally applicable stages of this process are outlined below.
Non-linguistic tasks, like project management (PM), testing, engineering and page-setting (DTP), are centralized and performed in-house at Logrus IT production sites.
Linguistic tasks are carried out by professional, trained and tested native speakers, all permanently residing in the target countries/markets without exception.
Each process consists of preparation, production, and delivery/post-delivery stages:
1. Preparation
includes terminology work, thoroughly understanding client requirements, preparing schedule/milestones, selecting proper resources, and handoff processing.
At this stage, we verify the completeness and correctness of source materials, instructions and translation memories (TMs), perform file conversions, splitting or modifications (if required), identify common segments to be translated and prioritized, and apply machine translation (MT) where possible/appropriate for units without TM matches.
2. Translation and review
include translation (using MT/TMs as appropriate), scientific and/or content editing, regular editing (including post-editing) and proofreading as well as independent language quality assurance (LQA) spot-checks, linguistic and technical checks.
All translators are following formalized requirements included in a standard Logrus IT service agreement as well as project-specific instructions and guidelines provided by the client.
All materials undergo extensive, multi-stage quality checks.
Scientific and/or content editing ensures adequacy, accuracy, readability, and completeness of materials and style consistency.
3. Delivery and post-delivery
include collecting all files and ancillary materials, compiling delivery documents, lists and quality reports, updating TMs (if working outside of a TM server environment), introducing last-minute or post-delivery changes requested by the client, etc.
This cycle is repeated multiple times for agile environments or long, iterative projects.
Logrus IT is well familiar with agile localization. We have extensive experience of agile work on huge, multi-year agile projects for our key customers.
Some specific steps we take when working on agile projects:
There are cases when a full cycle with all required checks just can’t be fit into the turnaround timeframe for a particular handoff or a content piece is too small for global consistency checks. This doesn’t mean that we’ll let errors slip through the cracks. We apply a continuous localization approach, fixing discovered problems at a later stage in the cycle. This work is based on both internal quality assurance carried out after the handback, as well as feedback from the client and/or third-party reviewers. As a result, translated and published content is gradually improving over time and finally reaches the same level as expected for non-agile localization. This includes providing a complete translation – editing – proofing solution and ultimately running all internal language and technical checks on each piece/delivery.
The process commonly utilized by most small teams and individuals across the industry is simple:
Handoff > Translate > Edit > Check using built-in QA checks in the CAT tool > Handback to client
Regrettably, this oversimplified and ubiquitous process contains multiple gaps, including lack of proofing or sample LQAs, natural limitations of QA checks built into the tools, and lack of manual checks.
At Logrus IT, even the simplest version of the translation process extends far beyond the basic approach. The high-level process chart looks as follows:
We use three major variations of the translation process at Logrus IT. The choice of a particular variant depends on the content, context, area and language pair, and client requirements.
The simplest process used when Machine Translation (MT) cannot or is not supposed to be applied follows the chart above. TMHU here means a regular TM created from edited human translations.
When MT is used, and its quality is known to be stable enough and/or agreed in advance, MT is integrated into the process above. TMMT here means a pseudo-TM comprised of machine translations and/or live MT translations applied in cases when there are no satisfactory TMHU matches.
For more complex cases, when MT quality is far from uniform and/or relatively low for a high percentage of all translations, a straightforward MT process is inefficient. Editors are spending too much time evaluating or trying to improve MT translations that are beyond repair. Actually, existing benchmarks prove that this type of activity is more time-consuming then translation from scratch, because the person first has to come to a conclusion that the existing MT translation cannot be “fixed”, and only then translated the source from scratch. All this overhead is due to the simple fact that MT-generated translations have no default “match ratings”. All units in a regular TM have these match ratings, making quality-related judgment instantaneous.
In order to make the process more efficient, Logrus IT has developed a modified process that includes a separate, semi-automated MT evaluation stage that quickly and efficiently disqualifies poor MT translations and makes translators’ work more productive:
