3,400,000 words in 10 days: Localex post-editing case study

Just a few years ago, a story about a boutique LSP handling a multi-million-word project in a couple of weeks would sound insane. These days, it’s becoming a reality thanks to technology — and teamwork.

Localex is an LSP founded in 2015 by a group of localization professionals working remotely. Although the company never shied away from handling difficult projects, a request to post-edit 6,500,000 words in just under three months would be challenging even by their standards. To top it all, the customer was one of the leading technology companies in the world, so it wasn’t just money that was at stake, but also reputation.

The LSP decided to take the project on. Despite the obstacles they faced along the way, they were able to turn it into a success. What difficulties did they encounter and how did they overcome them? Let’s find out.

But first, let’s take a look at the numbers:

Now onto the challenges.

Localex works with both in-house and trusted freelance language professionals. However, for a project this big they needed to expand their teams to 20+ linguists for some languages and to 100+ linguists in total.

Localex needed to expand their teams to 20+ linguists for some languages and to 100+ in total.

Thanks to Smartcat’s 250,000-strong marketplace, Localex was able to quickly find new post-editors and add them to the translation process — all using the Smartcat UI and without sending a single email or browsing any external job boards.

In a conventional setting, managing 130 linguists would require 10–15 project managers. That is more than what Localex has in-house. Moreover, this would inevitably increase the project price. And you would then need a manager to manage all the project managers.

Callout: In a conventional setting, you would require 10–15 project managers and also a manager to manage all the project managers.

At the heart of Smartcat is a heavy-duty workflow automation platform that allows you to track progress, communicate with contributors, and handle project assets all in one place. This allowed Localex to handle the whole project with just their four in-house project managers.

“Smartcat enabled seamless communication for us, which is essential when working with such a big team of post-editors, proofreaders, QA specialists, and project managers.” — Çiğdem Tura, Localex Operations Manager

When you handle a project of such size, it’s tempting to give a certain “tolerance” to lower quality standards and making use of automated QA tools alone. However, this wasn’t the way Localex wanted to go.

“We didn’t want to ‘hide’ behind the high volume and tight deadline and use these as excuses to provide lower quality work.”

While Localex did assign dedicated people to assure quality from both technical and linguistic points of view, an additional and oft-overlooked quality boost came from Smartcat’s collaborative nature.

“When someone leaves a comment, all linguists in all languages can see it. Thus, linguists warned one another about ambiguities, discussed possible misunderstandings, and shared useful links. In many cases, we noticed a friendly and productive bond develop between linguists.”

“^{When someone leaves a comment, all linguists in all languages can see it. We noticed a productive bond develop between linguists.”}

It was the customer’s request from the get-go to use post-editing for this project. The main reason was obviously the price: PEMT costs at least twice, sometimes three times less than “conventional” translation. This brought about a set of peculiar challenges.

Although PEMT is far from being a new concept in the industry, many companies and — especially — translators have yet failed to grasp its essence. Many try to edit the raw MT output too much, aiming for “perfection” (whatever that means). On the other end of the spectrum, there are people who pay too little attention and miss MT output that sounds natural but is factually wrong.

Localex shared TAUS guidelines with all post-editors. Thanks to Smartcat’s commenting feature, they could do this inside the document and within context, so there was no need to send endless streams of emails.

Different MT engines handle different types of content differently, and you never know in advance which one will be best for your project. For a project that large, the cost of a mistake can be too high.

Luckily, Smartcat supports a variety of machine translation engines which can be put to the test by simply ticking the respective checkboxes. This can be done individually for every language:

Although neural machine translation makes MT increasingly apt for “good enough” translations, sometimes it can’t even manage that. The usual reasons include source text misformatting, misspellings — either accidental or intended, — and the like. In one specific sub-project, the MT output was so off that the translator had to do everything from scratch.

Localex turned to the customer explaining the situation and, luckily, met full understanding and agreement on their side. Both Localex and the translator were able to work at a higher rate than initially agreed. The respective amounts were added to the invoices using Smartcat’s payment automation feature.

A project of several million words will often include a variety of formats, each requiring its own approach. For example, Localex had to work with a dozen different formats, each with its own set of peculiar features.

Although most CAT tools support various file formats, the extent of the support differs from one tool to another. Localex was able to use Smartcat for most of them, although some needed initial pre-processing — more on that later.

When it comes to IT projects, documents can come in custom formats that are not recognized by CAT tools out-of-the-box. For example, one format Localex had to work was XLF, which, although XML based, had more than a few challenges making them localization-unfriendly. These included extremely long segments, complicated tag structures, and custom placeholders.

Localex performed some normalization tasks on the source content, removing unnecessary line breaks, hiding tags, re-segmenting, and adding placeholders for non-translatable bits. This made it easier for post-editors to work on the text and ultimately increased productivity.

When you have to pay 130 linguists scattered across the globe, payments become much more of an issue than when it’s just a dozen people in one country. Each country has its own regulations for inbound payments and some countries don’t support the methods you’re used to, not to mention that transaction costs can range from acceptable to insane.

Localex used Smartcat’s Payment Automation solution in two ways: First, to pay the freelancers hired via Smartcat. In this case, the process is completely seamless, and Smartcat automatically calculates all costs for every job. Secondly, Localex added custom payables for non-translation-related jobs and included them in the same invoices.

The company did not have to deal with any paperwork for payments to freelancers, they saved money on transaction costs, and never gave a second thought to any legal or tax considerations, because this was all handled by Smartcat.

In the end, Localex delivered the project in time and to the customer’s satisfaction. Specific results include:

• 30–40% productivity increase

• 5,000–8,000 words/day translated on average by every post-editor

• 40% savings in project costs

• Feedback collected from 100+ linguists

• New data for MT engine training

• Feedback provided to Smartcat for its further improvement

Size

• Find new post-editors in the Smartcat marketplace

• Use Smartcat’s workflow automation to communicate & track progress

• Encourage teamwork among linguists to assure quality

Post-editing

• Share TAUS guidelines with post-editors

• Test different MT engines to see which fits best

• Have an “MT bypass plan” agreed with the customer

Technicalities

• Make sure your CAT tool supports all or most file formats expected

• Pre-process & normalize files to simplify their translation

• Use Smartcat’s payment automation platform to avoid paperwork

It might seem counter-intuitive that a big tech company decided to put such a large project in the hands of a boutique agency and not a transnational MLV. But, if you take a closer look at this case study, the reasons become obvious.

First of all, a small agency can dedicate all, or almost all their resources, to that client, becoming a de-facto “company-as-a-teammate” for the duration of the project. “We had a daily meeting with our client every morning, sharing status and addressing any possible issues, and were available for requests 24/7,” says Bekir Diri, International Sales Manager at Localex.

Secondly, this allows for a much more transparent process. What an MLV would most likely do with such a project is find ten smaller LSPs and hand over parts of projects to them. Everything that happens on deeper levels is completely hidden from the customer, so realtime progress tracking is impossible.

Last, but not least, MLVs have much higher overhead costs. This includes:

• The army of auxiliary personnel they need,

• The advertising costs that are in one way or another included in the project price, and

• The margins from every LSP down the food chain involved.

When you factor this all together, you can understand why customers who actually do have an idea of how the industry works are increasingly reluctant to work with the “behemoths” and prefer smaller, nimbler agencies as vendors.

Unfortunately, many customers don’t know how the industry works. The way out? Share this story with them, and help them make the right choice.