Online channels are playing an increasingly crucial role for the travel industry. They are an effective means to reach inbound as well as outbound travelers. In order to make the online contents of these channels accessible to different target markets and potential customers, travel suppliers often have their contents translated into the local languages of those markets. In today’s fast-moving environment, however, online content has a limited lifespan and therefore it is crucial to have it localised and made available as soon as possible. So, the challenge for many travel suppliers is, how to localise large volumes of data such as user generated content (UGC) into many different languages in near real-time at low cost? A combination of language analysis, customised machine translation and machine learning is a possible solution.
There are a number of use cases where instant localisation plays an important role in the travel industry.
Nowadays, online travel sites often have social media integration and a section for travelers to review. This user generated content (UGC) is increasingly gaining influence in shaping the decision-making of travelers and so has become potentially valuable content, making it a new and effective instrument for advertising. To make this content accessible to a broader audience in different countries, localisation of these traveler reviews is required. In addition, many travel sites also take reviews from high volume markets e.g. the US and localise them for publishing in smaller markets in order to bolster the local website and increase the attractiveness for local customers.
Another use case is the localisation of special offers and “low value” content. In the travel industry, special online offerings can be distinguished into those that have “high value” attached to them such as the description of luxurious hotels and top destinations. For those sites, usually there is a lot of analytics performed behind the scenes to ensure that traffic is “sticky” and the revenue is maximised. However, often there are also sites that contain contents of “lower value”, meaning content that drives lesser revenue and therefore not necessarily requires expensive human translation. Yet, there is value in these contents and they still need to be translated as they do generate revenue nevertheless. Belonging to this category of contents are short term special offers. These can drive remarkable revenue, but due to their rather “low value” and short life span, assigning a human translator for these kinds of jobs may not be the right solution.
Furthermore, instant localisation through machine translation can be applied for customer support and chatbot solutions. Customer chat supports are often operated by agents who speak two or fewer languages. However, having a customer base in different countries makes it essential to provide multi-lingual support in order to effectively communicate and serve customers. Such chatbot solutions need to have the ability to detect the customer’s language as well as to make real-time translations.
With very few exceptions, it would be difficult to supply the needed support in the above-mentioned use cases with human translation alone. In these cases, a sophisticated machine based solution that augments the human localisation capabilities combined with human translation quality control could fulfill the task in a timely and cost-effective manner.
However, applying machine translation for processing data is not without challenges. One of the challenges with any machine based localisation solution is managing expectations. The quality of the translated output will not automatically be perfect just by using machine translation. Human intervention is necessary for training and maintaining the systems as well as to ensure and improve the quality of the translations.
Talking about quality of translations, for a human translator to provide high quality translation, domain and language pair expertise is required as well as knowledge of style and glossaries of the industry domain. The same applies to machine translation: in order to get good translation results, the machine needs to be custom trained with the right domain and auxiliary data.
Another common challenge, in particular for content rich sites, is the large volume of data that has to be processed. Million or even billions of words per day, in case backlogs are processed, are not uncommon volumes. Such processing requires systems and workflows that are capable of handling large amounts of data as well as possess the ability to scale up or down on demand so that infrastructure cost can be controlled.
Finally, it must be understood that the actual translation is just one step in the machine based localisation process. The whole process consists of various steps and involves complex pre- and post-processing to perform needed conversions, to deal with lists of keywords, to identify and transliterate locations, recognise and convert currencies and measurements, apply the right styles and glossaries and at the end, return the translated content to the calling application in the correct order.
There are a number of use cases in which drawing on machine based translation solution might be necessary in order to fulfill the demands of providing localised content fast and cost efficient in target markets. However, applying machine translation is not without challenge. Like all complex IT projects, a solid analysis of the requirements and the project and thorough planning are key to success.
