Interview with Catawiki Localization Managers Emily Ingham & Virginia Laghi: How to provide cheap, fast, quality localization services
Catawiki is Europe’s fastest growing online auction house, for buying and selling special items and collectibles. It was originally founded in 2008 as an online community for collectors. The Catawiki Localization Team is part of the Product & Technology Department. But unlike all of the other product teams which are exclusively composed of engineers, the Localization Team presents a diverse mix of backgrounds and skills.
I sat down today with Emily Ingham and Virginia Laghi, localization managers at Catawiki, to discuss their localization strategy and how they make important decisions about localization projects. Emily is the Product Owner and she engages all Catawiki stakeholders on localization topics and priorities. On the other side, Virginia leads the content team, composed of five editors who define the best localized content for Catawiki’s core languages: English, Dutch, German, French, Italian, and Spanish.
Together with six backend developers, the Localization Team drives innovation by creating localization experiments and building tools for editors and translators to work efficiently. The team also coordinates over 140 freelance translators for 17 languages.
Rei: Please tell us about your background and how you got into the current role as a localization manager.
Emily: When I started at Catawiki three years ago, there was no Localization Team yet. There were two customer support agents that coordinated translation projects and UI localization, in addition to their main daily tasks. As Catawiki grew, we went from needing about 3.5 million words translated a month to 8.4 million words, and this arrangement became unscalable. We decided to build a dedicated team with engineers and content editors in our core languages and to centralize the localization of all content types. We also decided to include Gengo as a translation solution to help scale the translations. At the moment, we have a 50/50 division between Gengo and our freelance translators that we coordinate ourselves. All content types go through the Localization Team now. This includes user interface texts, FAQs, legal documents, and auction lot content and we translate into 17 languages.
Virginia: Before joining Catawiki 2 years ago, I worked for different companies as either a content editor for my native Italian language, or a translation coordinator. At Catawiki, I could use both skills for the same role — I was responsible for Italian content and for coordinating all internal translation requests. Now, all of our editors have this dual role in being the ambassador for their native language, as well as owners of specific localization projects, such as freelancers management, quality assurance, content maintenance, taxonomy, across the 17 languages that Catawiki serves.
In your experience, how does translating ecommerce content differ from translating other types of content?
E: In our case, the e-commerce content is user-generated. This means that we often receive source texts that aren’t “clean”. Our translators often report many small errors from major errors which can change the meaning of the product. In addition, we also often receive the content quite close to the deadline as we’re dependent on the sellers and experts on adding the lots. E-commerce content also often comes in huge volumes with short shelf-life. So for our team the big question of the choice triangle always arises. Do we chose quality, cost or speed?
What is your quality control process for outsourcing localization jobs?
V: For user-generated product descriptions, we keep track of all reported translation mistakes in a dashboard. These reports come from sales, customer support, experts and users. Depending on the quantity and overall gravity of the mistakes, we then run quality reviews of translation samples from the specific linguist. Resulting from the reviews, which we also run periodically for all linguists, we send specific reports with instructions and feedback to the linguists. In addition to quality reports, we also maintain style guides and glossaries, which are specific per product category… stamps, art, coins, etc..
How do you make important decisions about the timeline and budget for your localization projects?
V: Depending on the type of content, we have specific SLAs. For user-generated product descriptions, we consider a translation deadline of one week, from the moment the user sends us the product description, to the moment the lot goes live on the platform. This type of content has a short lifespan of seven to ten days, and high word count of about two million words per week. In order to reach the deadline, while ensuring translation quality, we work with a team of 120 linguists in a one-step translation process, and we run periodical reviews of freelancers’ translations. When it comes to legal, UI, marketing, and Catawiki-generated help copy, we use a three-step process: first translation by our freelance linguists, second review by subject matter experts such as the legal department, category experts, and marketeers, and third proofreading by the internal editors. Deadlines depend on the requester’s needs and the final word count. As a rule of thumb, we consider 24 hours for requests larger than 1000 words. Anything smaller we can treat with urgency within the day.
E: 90% of our monthly word count comes from translating our product descriptions into English only. When making our yearly budget, we use the forecasted volume of lots and aim at staying below a certain percentage of our commission.
How do your large-scale localization efforts at Catawiki differ from those running smaller projects? What are some things to keep in mind when beginning to scale these projects?
V: When a product team needs to launch a new feature, they will ask us to create and localize the content needed on that page. Even if the feature is very limited to a specific part of the website, for instance, the payment flow, the editors make sure that the terminology used in the copy is consistent with the tone of voice of Catawiki at large.The constant effort to see any new content against the entire “Catawiki voice” ensures a uniform and legally correct word usage, including branded and company-specific terms, as well as the communication of a consistent message across all languages we support. Another thing that we always need to keep in mind when scaling up our localization projects, is the need to consider the specific grammar structures of the different locales. We work with 17 languages and each of them have their own peculiarities regarding word order, date formats, declension, plural forms, and the general expansion of the target word count when translating from English.
By working closely with all product teams, engineers, and designers, we ensure that all localization challenges are taken into consideration from the very beginning of the process.
Is there a character limit we need to consider? Are we using in the English source an idiom that won’t be translated easily in other languages? Is there enough context for the external translators to understand where and how the copy will be used?
What are the challenges specific to Catawiki for localization?
E: The main challenge for Catawiki is the short shelf-life of our product descriptions. Every week around 40,000 lots are offered for auction, but they only stay online for a maximum of 14 days. If you have 10,000 t-shirts, you only need to write one product description. But Catawiki deals with unique objects, offered by users, and they all have different descriptions. As this costs a lot of money, we had to make the decision to translate all descriptions into English with human translators and machine translations into our 16 target languages.
Another big challenge that we face is how to guide users to write good product descriptions. We often see a lot of repetitions or copy/pasted Wikipedia articles which we obviously prefer not to translate. The challenge, however, is to not only focus on saving costs, but also optimizing for the users.
V: Creating and localizing content in Catawiki means to work with specialized terminology depending on the product category, such as art, classic cars, stamps, coins, and jewelry. The challenge is to maintain a high quality of this specialized content, while ensuring a quick processing and delivery of translations. To overcome this challenge we assign the translation requests for each category to a dedicated group of freelance translators who become specialized on that topic. At the same time we also collaborate closely with the internal subject matter experts, who are involved in the review of content and help us creating and maintaining specialized glossaries and style guides.
What innovative solutions to these challenges have you implemented at Catawiki?
E: One solution that we implemented to guide sellers when writing their product descriptions is with instant feedback. By using data science models, we directly ask the seller to change the text while they are writing. This has helped us to reduce repetitions and increase the quality. Our biggest innovative solution, however, is using what we call lot specifics, which are product specifications for generating titles and subtitles.
Please tell us more about the Product Specifics feature at Catawiki and how it has helped you with your localization efforts.
E: We started to realize that the content in our titles and subtitles often corresponds to the product specifics of the object. Product specifics are drop-down fields with values that a seller can pick from. All of these values only need to be translated once and are translated into all 17 languages. We started to analyze the content and came up with a template for the Watches category using the values of our product specifications. This worked so remarkably well that we’ve now expanded this method to (almost) all our categories.
In the backend, we’ve created templates for the title and subtitle fields for all categories. When a seller selects a value from a drop-down in the lot specifics, we immediately place this value in the title field. Hence, we automatically generate a title depending on a seller’s input which we can show in all 17 languages with the cost of only translating once. This has reduced our translation costs significantly while coming a step closer to providing a fully localized experience for our buyers.
What is your experience with user-testing around machine translation, human translation and showing the original English content?
This all boils down to the choice triangle – cost vs. speed vs. quality.
E: Obviously, we all agree that human translations are superior to machine translations when it comes to quality whereas the machine beats humans in both costs and speed. When we A/B tested machine versus human translations, the quantitative data did show a negative trend, but we made a decision to go for machine translations anyway, because of the enormous cost savings that we could achieve. As a compromise, we kept translating into English with human translators as having a good source text increases the quality of the machine translations. Now, two years later, we do see a number of negative qualitative user comments on the quality of our machine translations, so this will be the next innovative solution that we’ll tackle.
Finally, do you have any advice for localization managers working on similar projects, or localization managers looking to scale their efforts?
E: Think big! My main vision and driver has always been to try and crack the choice triangle. How can we translate everything into all languages in time for little costs with good quality? By thinking outside the box, but also not being afraid to test different things, we’ve started to crack the cost and speed issues. We’ll now start looking into scalable solutions for quality. Some of our experiments flopped, but we’re still happy to have done and learnt from them. Gather your data and don’t be afraid to also make unpopular decisions. When we decided to stop translating into target languages, my localization heart didn’t want this to happen. But it did buy us time to understand our costs and where they come from and we now have a good base to start and innovate from. From years of focusing on costs, we are finally ready to transition to a more user-focused team.
V: No localization challenge is too big if you work with the right team. What makes it ‘The dream team’ is the perfect mix of technical, operational and creative people. Engineers, editors, copywriters and translators all work closely together and get a good understanding of each other’s challenges. This is what makes the team stronger and a great backbone for all risk-taking, outside-the-box, daring attempts at big localization projects. As a team, we are also not afraid of getting our hands dirty: some of our processes are still not automated so it happened that we had to all get together and copy-paste hundreds of words from one platform to another, or edit an entire translation memory from scratch. We all do it together, nobody excluded: we work together, suffer together, fail together and celebrate together!
Go global with Gengo’s people-powered translation platform.
or Contact us