Did you know… You could save your localization team 2,000 hours using AI
Using a Translation Management System (TMS) when creating content for a global audience — and the automation that comes with it — is the first step toward increased productivity and efficiency. As a TMS automatically shares files at each stage of the translation workflow with all of the relevant stakeholders, it reduces the amount of manual work and time required until the final translated files are back with the project requester.
What additional savings could you get from your TMS?
To find out where those extra 2,000 hours of savings could come from we asked linguistic expert, Dr. Rafał Jaworski and Xpert Rory Sampair, to share their knowledge. Jaworski highlighted the efficiencies that are driven by the automatic placement of inline tags and SYSTRAN AI-enhanced ™, while Sampair focused on value driven by translation memory alignment and Term Extractor AI. Read on to get the detail plus a bonus 50 hour time-saving tip!
Save 200 hours with automatic placement of inline tags
Having to place inline tags manually is one of the most time-consuming tasks a linguist has to do, although project managers or administrators may not be aware of this. These tags are used to signal text formatting, such as words in bold; or elements of code, like hyperlinks. For consistency, everything that’s included in the source file needs to be included in the target file as well, which is why these tags are required. The challenge comes from the fact that the placement of these tags within the source will likely differ in the target.
As a result, when linguists are translating content, they also need to be aware of where to place these inline tags. Since these must be included, linguists have to stop focusing on translation to think about where these tags should appear, which is a tedious task that takes them out of their flow.
One of XTM Cloud’s AI features is to automatically recognize and place these tags in the right place. Using AI, linguists can focus solely on their translation, and the TMS automatically places the tags where they should appear. It’s an extremely powerful tool that ensures the localization process is efficient by reducing the amount of manual work required.
To quantify the benefits of this feature, our AI team ran tests. They were able to see that the auto-inline feature was used 1.3 million times, and that in 98% of those instances the auto inline placement did not require any corrections. Based on these findings, the team did some further calculations, as Rafał Jaworski explains:
The manual adjustment of inline tags in one segment takes about 15 seconds. Over a month, we can estimate that a localization team could work on 50,000 segments containing inline tags. Placing these manually would take them 12,500 minutes, or 208 hours. As it happens automatically, this manual work is almost completely removed. That way, over 200 hours can be saved per month, which is astounding.”
Time saved so far: 200 hours
Save 50 hours with the Translation Memory aligner tool
Translation Memory (TM) alignment is necessary when localization teams are trying to match source and target segments. This can occur for teams who didn’t use a TM for their localization process, or when transitioning to a new TMS tool and translation memories and segments were not included in the migration process. Building that TM from scratch would take months of work… AI is able to do that in minutes – and our tests show that 70% of our users are using the TM aligner.
Your TMS can analyze existing source and target segments from previously translated files, and create translation memories based on those. It can also enable users to expand their existing TM, using files that they hadn’t uploaded in the past. However, this is not always as simple as it may sound. There could be many discrepancies, such as single sentences in English requiring to be translated into two to three sentences in the target language, or the other way around. The aligner will spot these without any user input. Rory Sampair explains why this is key:
Legacy tools required the user to manually draw a line between source and target segments to perform TM alignments, which was extremely time consuming. This became problematic when it came to creating translation memories from large documents containing hundreds of thousands of segments. As the speed of manual alignment can be estimated at 15 segments per minute, preparing a 50,000 segment translation memory manually would require over 3,000 minutes, which represents over 50 hours. This is the kind of volume that can be processed in a matter of minutes with the aligner.”
Time saved so far: 250 hours
Save 850 hours with the Term Extractor
Creating a new Terminology database or adding to an existing one can be a tedious, time-consuming task — even if it is only required once a year. Here, AI can yet again be used to increase efficiency, but also to create a standardized process that removes individual subjectivity. Every time a new localization project is created, the TMS analyzes all the content, and finds high quality terms to be added to the term base. To note, these terms are different from the segments stored in the TM. Terminology is made up of specific words, or sets of words, within a larger sentence. What makes them stand out is that they appear frequently in the source document, but aren’t necessarily that common otherwise – such as the phrase ‘individual subjectivity’ in this blog post.
This means there is no need to rely on a linguist or reviewer to notice translation inconsistencies before a term gets manually added to the term base. Besides, while XTM Cloud analyzes your files, it takes your existing term base into account, if you have one, to ensure that any terms that already exist in it are not added again. AI transformed a process that had historically been very subjective and manual by adding a ton of logic and math to it.
To go one step further, the bilingual term extractor can be leveraged, meaning that when a term gets created, its translation is also included. It doesn’t require linguists to write the translation – it writes itself! All that’s left for the linguist to do then is a quick review, and the Terminology is ready to use, ensuring that localized content is consistent and of high quality. This produces a considerable ROI, as costs incurred from having to retranslate content will be removed, as well as the time required for content to be reviewed and approved.
Above, you can see the amount of data that can be contained in a term base, which is easy to miss if it’s something you’re not familiar with. Here, we can see the source and target term, which is great, but most importantly, we also get the context. We see where these terms appear within the source document, how often, and in which file. This is a lot of data that is now easily and readily available. Rory explains how important this is:
Previously, extracting 500 terms manually would have required going through all the 50,000 segments searching for the terms. This takes around 1 minute per segment, so 833 hours in total. On top of that, figuring out the translations of the extracted 500 terms (which the bilingual term extractor does automatically) would take a further 3 minutes per term (extracting all possible translations and their contexts, and making a decision regarding the correct translation). So that’s an additional 25 hours being saved, or over 850 hours in total.”
Time saved so far: 1,100 hours
Save 600 hours with SYSTRAN AI-enhanced TM
We can all understand the benefits of using Machine Translation (MT): it gives linguists a base for their translation, which they can then fine-tune to ensure the right words are being used, as well as the right tone. What can be frustrating, however, is when MT gets the translation completely wrong because of homonyms or uses the wrong word when several translations are possible.
We’ve partnered with SYSTRAN to solve this problem by enabling MT results to take the existing TM into account. This gives the MT context, provided by mid-level fuzzy matches coming directly from the TM, thus improving the MT output. This ensures that the right words are used in the right context. Where a sentence has been partially translated previously, and is therefore stored in the TM, the NMT engine fills in the missing words to deliver the best possible match. AI-enhanced TM takes untapped data and converts it into persuasive multilingual communication.
Our data show that with AI-enhanced TM, we can estimate the decrease in human effort required to amend MT translations to about 30%. This means that the total effort to translate 50,000 words, which is up to 25 days, or 2,000 hours, would be decreased by 30%. In this case, we can expect to save about 600 hours, as the MT engine would leverage the TM to deliver the best match. That way, linguists won’t have to edit segments delivered by SYSTRAN.”
Time saved so far: 1,700 hours
Save 250 hours with Weighted Token Levenshtein
A key component of efficient localization is being able to retrieve matches from Translation Memories in the most efficient way. The more these are leveraged, the more direct savings are realized, as they enable linguists to work more quickly while creating high-quality, on-brand content. This is why we’ve developed our own proprietary algorithm for TM leveraging: Weighted Token Levenshtein (WTL). This unique algorithm calculates fuzzy matches by taking into account and recognizing the syntax of the segments in the localization project, enabling us to retrieve more matches than any other TMS.
WTL is able to spot matches which normally wouldn’t be recognized, such as sentences that differ by the word order, like “For more information, check our website” and “Check our website for more information”. Without WTL, linguists would have to translate the segment from scratch, although the sentences and their meaning are practically the same. The same goes for “Visit Paris” and “Visit London”. In theory, these are only 50% matches, which wouldn’t be high enough to be picked up. Well, WTL is able to recognize proper nouns, making this an 80% match which will get leveraged by the TMS.
Using AI, XTM Cloud is able to recognize segments that require very slight changes, when previously linguists would not have been offered this match. Now, the TM is fully leveraged, ensuring that linguists can work efficiently and not spend time unnecessarily translating segments that have previously been translated. Again, if we are looking at translating 50,000 words, better leveraged fuzzy matches could generate significant savings. If the file has 10,000 segments similar to previously translated ones – which would have been missed without WTL – it would have taken linguists over 30 days to translate these from scratch if they translated 300 segments a day. That’s over 250 hours saved.”
Total time saved: 1,950 hours
BONUS TIP: Project groups
If you’ve read this far, you definitely deserve an extra tip! This one is a little bonus, as a thank you.
We’re working on a new feature, currently in XTM Labs, which would enable you to save an additional 50 hours. This new functionality will allow you to create Project groups that can be accessed and managed at the click of a button. All you’ll have to do is perform an advanced search with all the criteria you need, such as customer name, project manager, status, but also payment status, due date, target or source language, and more. Then, you’ll have the option to save this search as a project group, making sure you can always find these projects easily. For instance, you could have a project group called ‘Technical translations’, another one called ‘Projects in French’, or ‘Projects in Progress’, or even ‘Late payment’. The possibilities are endless, and ensure you get access to project overviews quickly and easily.
Plus, you’ll be able to perform bulk actions, such as changing due dates or linguists for instance. Having to look for projects manually and selecting each one – ensuring that none get selected by mistake – can be a tedious task! Although doing this once would only take about 2 to 3 minutes, if 5 Project Managers perform 10 searches a day, it equates to 50 hours a month!
At XTM, we are committed to continue developing more AI features. In the future, you can expect the review process to become automated using the power of AI, and we cannot wait to tell you more about it!