Skyscanner’s Localization Challenge: 3 Million Words, 30 Languages, 600K Strings

Travel search site Skyscanner has over 60 million monthly users constantly looking for the best deals on flights, hotels, and car hires.

As Slator reported in a recent article, this can get very challenging when information is not available in the user’s own language or is poorly translated.

Advertisement

Skyscanner wants to set itself apart by offering reliable and understandable travel details in over 30 languages. That means giving not just information in the right language, but doing so with translation that is appropriate for specific markets.

Slator chatted with two key members of the Skyscanner translation team: Senior Translation Executive Hristina Racheva and Senior Engineer Chris McCluskey, who both play a critical role in the company’s translation and localization efforts. They offered insights on how their team works, the challenges of translation in the travel industry, and the mash-up of human effort and technology in delivering content.

Slator: Can you describe the general roles of your translation management team?

Skyscanner: The team is made up of translation executives and software engineers, with the support of a copywriter. We are responsible for evolving our translation and localization processes and enabling colleagues to effectively make use of our services.

Translation executives process most translation and linguistic requests and queries, including product translations, linguistic testing, language quality evaluations, glossary and style guide maintenance, collaborative activity with regional marketing teams and copywriters, and evaluation of English quality as a source language.

Software engineers focus on building the tools to enable Skyscanner to manage and consume translations, including maintaining the Pootle translation management software (Pootle is a free, web-based tool used as the main interface for creating and updating translations) and translation publishing workflows across products via our Strings-as-a-Service tool and Culture Service, which is the canonical source of market, locale, and currency data company-wide.

Slator: What are some unique translation and localization challenges that Skyscanner faces as part of the travel industry? How does this translate to the mobile experience?

Skyscanner: Skyscanner is available in over 30 languages, with over 60 million unique travelers using us each month. And with that extensive global reach comes one of our biggest hurdles maintaining translations for all the possible locations our travelers want to visit, as well as ensuring that market specific terminology is used accordingly. For example, in Japan, hotels have different types of bathrooms and we must clearly communicate which is available when describing the offering.

We also focus on building products that display content across many locales, without degrading the user experience. We often encounter issues because some translated words or expressions are much longer in the target language than the original English.

For our app and mobile users, we work to ensure translations display properly on smaller screens. We work closely with user experience (UX) designers and copywriters during the design and implementation stages to ensure the best experience.

“For example, in Japan, hotels have different types of bathrooms and we must clearly communicate which is available when describing the offering”

Slator: Which language pairs do you find most in demand and what drives them?

Skyscanner: The majority of our content is created in English. However, Skyscanner is structured in a Tribe and Squad model, with squads acting as a mini start-up. This structure supports regional growth marketing squads, who look after respective markets across the globe. These squads rely on the translation team to help translate global materials that need to be disseminated in specific markets such as case studies or training documents. We work closely with our legal team as well, supporting contract translations. Languages run the gamut, including Korean, Russian, Japanese, Chinese, French to Mexican Spanish.

Slator: Do you anticipate Skyscanner’s String-as-a-Service to grow and to play a larger role with your translations?

Skyscanner: String-as-a-Service has an indispensable role in our translation process both for our colleagues and within our own team. Its introduction to the translation workflow has significantly simplified the process of requesting new content for translation. The service allows requesters to provide sufficient context text and image for each string which ensures translators have a good understanding of the content and context. String-as-a-Service has also allowed us to incorporate an English copywriter into the workflow who can review all content submitted for translation.

Slator: Can you give us some idea of the kind of translation volumes you deal with each year in terms of number of words?

Skyscanner: Our current total translations for all languages used in our products is 3,123,036 words in 597,895 strings. This does not include translations of legal documents or user guides.

Slator: What are the challenges surrounding translating user-generated content at Skyscanner and how do you address them?

Skyscanner: We actually don’t have a sizeable amount of user-generated content. We do assist our User Satisfaction squad (customer service) with queries and feedback at times, enabling them to execute through brand focused translation. That squad includes native speakers of major languages and have a specific approach for languages not supported by a native speaker.

Slator: What standards do you use to measure translation quality at Skyscanner?

Skyscanner: We execute in context linguistic testing across our website and app, often when we release new products or features. This allows us to ensure high translation quality within the respective context as well as evaluate the translation quality itself.

We also work closely with our regional squads to ensure that the style of the translations fits with the Skyscanner brand in that market. This can involve reviewing existing translations or working with translation partners to identify the people who can best express market specific translations.

We are fortunate to have found that our more than 1,200 global partners across flights, hotels and car rental, generally provide high-quality translations – especially as we have included a copywriter and tools like glossaries in our translation workflow.

Slator: What are the common translation quality pitfalls in generating content for a flight aggregation engine? How do you avoid them?

Skyscanner: Skyscanner is a travel search engine saving travelers time and money by finding the best travel options for wherever they want to go. We help travelers find the best flights, as well as hotels and rental cars.

Travel related content changes around places, combinations of dates, prices, deals and different travel options. These are defined by dynamic and adaptable data, both names and numbers. For example, Skyscanner users often search varied destinations. To ensure we provide a consistent user experience that scales to thousands of destinations, we translate common words and sentences with placeholders for the place name, e.g. “Flights to {{city}}” where {{city}} will be replaced with the actual name.

Such placeholders are used for countries, cities, airports, date, times, currency values, etc. The challenge, in this case, is to provide enough contextual information for each placeholder so that translators can translate the phrase correctly and appropriately. We work closely with our colleagues to help them build products that can be easily localized using these methods.

Slator: When translation quality falls, where or how could the impact be felt? What are some safeguards for this?

Skyscanner: Our industry makes it a challenge to quickly identify the impact of poor translation because travelers will often still book flights even if there’s a minor typo.

We keep a close eye on user traffic and User Satisfaction queries, and we have an internal process that allows anyone within the company to flag translation issues. This close watch and community-centric analysis gives us the ability to quickly fix problems in collaboration with our linguists. And thanks to the way Strings-as-a-Service works, any change to the translation is available online within 24 hours.

“Our industry makes it a challenge to quickly identify the impact of poor translation because travelers will often still book flights even if there’s a minor typo”

Slator: Which technologies do you find most useful for (a) getting your content out in as many languages in the shortest TAT (b) ensuring that translations are of good quality? What kind of translation productivity (CAT) tool and translation management system (TMS) do you deploy?

Skyscanner: We use Pootle TMS, a free, open source tool that we use as the main interface for creating and updating translations. Pootle allows us to add contextual information and image for every string being translated, ensuring contextual understanding. From a TMS point of view, as an online tool, it allows translators to log in and translate directly so we do not waste time downloading, sharing, and uploading translations upon completion; additionally, it displays the translation progress per language and string/project so it’s easy to track the completion of our requests.

Traditionally, the biggest blocker to getting translations out at Skyscanner was the time it takes to develop and release the software that uses the translations. We have done a lot of work to reduce the cycle time in releasing software, including being able to dynamically update translations instead of waiting for manual updates. Strings-as-a-Service has played a massive role in reducing the cycle time.

“We use Pootle TMS, a free, open source tool that we use as the main interface for creating and updating translations”

Slater: How much translation volume is passed on to your localization partner/s? How much do you handle in-house?

Skyscanner: All our translation and linguistic requests are managed externally by multiple LSPs. We’ve selected and divided them based on language and content type.

Slator: Which kinds of work do you pass on to your partner/s (technology integration, management, transcreation, QA, end-to-end?) and which are handled in-house?

Skyscanner: All languages related work is handled by our partner LSPs, including translation, transcreation, LQAs, evaluation, and testing, as well as some glossary and reference material creation and management. For the majority of markets, we rely on regional market squads to define respective market glossaries, tone of voice, and all relevant linguistic reference materials. Technology integration and creation of new project management tools, or process are managed by the team’s software engineers.

“For the majority of markets, we rely on regional market squads to define respective market glossaries, tone of voice, and all relevant linguistic reference materials”

Slator: How many staff work in some translation or localization capacity at Skyscanner?

Skyscanner: Due to the nature of our business and our internal structure, nearly everyone at Skyscanner is involved in localization in some way. But the translation team is a team of seven.

Slator: What are your thoughts on the recent progress made in neural machine translation? What impact does or will it have on the work you do with Skyscanner?

Skyscanner: There has been a lot of interesting progress, but we don’t think the technology meets our needs and standards just yet. Therefore, we still require handpicked human translators to provide consistent translations that have the right tone for each market. We are always reassessing how we work and there is a lot of potential to use different forms of machine translation as a supporting tool in building Skyscanner.

Derek Walter

Freelance writer in northern California, writes primarily about technology for various sites, and author of Learning MIT App Inventor