Machine vs human translation for operators, affiliates and platforms

The understandable desire to outpace competitors, control costs and increase speed to profitability is a key driver in many industries, but igaming marketers and product leads can sometimes be tempted to make their own gambles.

On this website and at major industry events like the iGB Affiliate conferences I most often cover topics related to SEO – an area where the draw of especially quick wins can lead to ineffective shortcuts.

Translation and localisation have become similar areas with the advances in machine translation becoming ever more compelling.

As technology develops the appeal of automated solutions for localising is very clear when it comes to translation and localisation, especially as new markets open up and start-ups, affiliates and even major operators face barriers to recruitment.

While there are plenty of potential criticisms that can be made about product innovation and visual experience in igaming (when’s the last time you saw an aesthetically pleasing sportsbook?), the pace of the industry means that at least some new technology finds a home pretty quickly.

Is machine translation one of those solutions that should be embraced with open arms?

Well, no. Maybe with partially open arms at the distance you’d keep a dubious relative.

In the interest of full disclosure, ICS-translate is one of our brands and draws extensively on human expertise and qualified linguists so you’d expect a certain level of cynicism about the robots coming for our jobs. But if our own approach would be massively improved by going all-in on machine translation, I’m sure we’d absolutely do so.

Let’s start with a few notes on the positive developments in machine translation, and where it makes sense for betting brands to consider using it.

The state of machine translation in 2023

When we talk about machine translation (MT), essentially we’re talking about one of two things. Either a ‘custom’ MT engine that is bespoke to your organisation, with a level of ‘teaching’ required to shape the system around the industry, specific terms and language. Or a ‘generic’ engine – Google Translate being the most well-known.

While few would endorse a generic MT engine for use at a commercial scale, for those with the budget and specialist teams a custom engine could be a compelling solution. This is especially true in the fairly repetitive world of sports betting and casino where familiar language is used in each game/match/platform more generally.

That said, if any major brands are currently taking this approach, we’ve not heard about it. The finance and staffing costs would be considerable to set up such a system, but it could become tempting for those that work at scale. However, that produces its own problems for machine translation in betting.

The challenges of scale, MT training and the comedy of errors

On one side, the more material a custom MT system has to work with, the more, in principle, it can learn about industry-specific phrasing and norms required to produce very a “human” translation.

However, the main challenge of scale is that with increasing volumes of content comes the risk that eventually inaccuracies do surface. In the highly regulated betting sector that could be an enormously expensive risk to the business and also a personal risk to players.

This is the case with ‘taught’ custom systems but the risk of using Google Translate or something similar is even higher. Given how operators and regulators are increasingly looking to make sure affiliates are more responsible in terms of their content output, it’s a shaky solution even for those not personally taking bets.

The lack of a ‘trained’ system can lead to some surprising and, with enough distance, entertaining results.

In an example from recent weeks, a client sent us a Russian text to translate into Estonian. When the team looked at the text it was clear that the result was from an untrained MT engine rather than a distracted human translator.

The QA job required some creative thinking to understand what was originally meant and how the MT engine decided on its solution. For example:

• Trump suit in cards – was translated as Trump’s suit (as in ex-US president)
• Halftime in a match – translated as “part-time working day”
• Draw as “drawing” in art, rather than a draw in a match
• Total points over – “total points finished” instead of “more than”
• Odds – translated as the word “odd” for strange instead of “odds” in betting
• Correct score – translated as “correct assessment”

For fans of comedy mistranslations, there’s plenty of fun with untrained MT engines but things get less funny if revenue is at stake, customer trust is lost or regulatory obligations are not met in terms and conditions.

Translating for understanding vs localising for purpose

Most of us will have had personal experience with translation apps at one point or another. In most cases, we use them to acquire quick understanding, not linguistic perfection.

Coupled with this, the speed factor cannot be ignored and can be particularly tempting for those in a rush to capitalise in new markets.

This may be perfectly fine if someone’s trying to understand the fundamentals of a match preview or a game review. Yet, it falls down when needing complete confidence in legal/commercial/player protection aspects or indeed when content needs to be more ‘flowery’ and exciting to drive up anticipation in a match.

Understanding isn’t the same as communication and it should also be noted that human translation doesn’t guarantee this outcome either.

Of the dozens of sites our team has QA-ed this year, direct mistranslations of industry terms like “boosted odds” or “free spins” and so on have been a feature of at least 15% of mistakes. However, even where content is correct it’s not always doing the job of engaging players, earning traffic or giving a positive brand experience.

Another critical aspect of igaming content, of course, is SEO. Traffic is what makes the affiliate landscape function, of course, but operators are also increasingly keen to attract direct visitors within a wider industry context of challenging margins and high competition.

SEO is a very deliberate, opportunity-seeking discipline. Understanding search intent and factoring that into how content is structured and displayed is something machine translation isn’t fit to do. That said, it’s fair to say with innovations in AI like ChatGPT, a future scenario where MT can play a greater role in SEO is possible with the right training. Yet, I imagine the time and budget factor would be prohibitive compared to good old human SEO.

However, this won’t stop the ‘spray and pray’ approach to multilingual SEO-driven content. So, if the humans behind human translation aren’t passionate and knowledgeable, or if they’re estranged from the purpose of the content they’re working on, they’ll be no better than MT when it comes to the ‘sizzle’ of online content – and they’ll potentially be much pricier too.

Challenges for the translation industry in igaming

Translation and localisation partners and agencies must be conscious of the pressures on clients. The desire for speed and control of costs is ever present and a certain amount of empathy and realism is needed to make the choice for human translation a comfortable one.

With subject matter expertise comes greater quality and accuracy, but an increase in delivery times and cost. This cannot be ignored, but similarly operators in particular must be conscious of the regulatory implications for diminished oversight of multilingual content.

This may be where a joint approach can be useful – while the ICS approach is to have multiple native language experts to QA each finished translation, others do make use of machine translation to ‘sense-check’ and validate deliverables.

Of course, where content is for internal use, the quality threshold may be lower. This gets us back to translating for ‘meaning’ rather than ‘perfection’. There is an appeal of machine translation for particularly large corporate betting brands for internal communication/briefings.

Context is key – formal vs informal translation in interactive media

An extra element of translation best practice in igaming is drawing on context. As in other forms of online and mobile entertainment, the interactive context is critical to a quality translation that drives the right action.

This is true for sportsbook content and editorial as well as the localisation of buttons within a slot game, or the how-to-guides that a newbie player might draw upon with online poker.

This visual context is something that even highly trained custom machine translation solutions lack – even if they have increasingly sophisticated ways to deal with semantic context and industry-specific language.

More than this, when it comes to some innovations in games, it may well be that no ‘standard’ translation exists. Human translators can discuss and consider what an approximation might be while also drawing on local slang and dialect words that may be a better fit than ‘formal’ translation.

Summing up

Machine translation has made some huge strides, but a properly ‘trained’ solution isn’t the cost-saving solution many hope for.

In the meantime, some betting brands are loading up on risk due to the need to drive faster launches in global markets, without having the native language staff resource to support them.

Even basic machine translation has its place for content that isn’t for ‘prime time’ viewing, where stakes are low or where internal communication is the goal. However, for end users, quality must be the guiding principle.

There are already so many betting brands which fail to differentiate on a product basis and compounding that situation with undifferentiated or inaccurate content will only further erode trust.

Human translators themselves need to be empathetic to the commercial goals of betting brands. There’s no substitute for subject matter expertise and passion for the industry and its offerings.

Photo by Nicolas Hoizey on Unsplash