Peter Reynolds

I

t’s been a busy few years for memoQ. As if continually improving its translation management system (TMS) wasn’t enough, the company also underwent a series of high-profile leadership changes and acquired machine translation (MT) software company Globalese. On top of it all, a period of seismic technological change driven by artificial intelligence (AI) has brought uncertainty and unprecedented challenges to the language industry.

What does memoQ’s chief executive officer (CEO), Peter Reynolds, make of all these important developments? In a recent conversation with MultiLingual, he deconstructed the company’s approach to translation technology, its vision for the future, and why he believes the big changes of recent years pave the way for a brighter tomorrow.

MultiLingual: Could you give us an overview of your career prior to memoQ and what prepared you for your current position?

Peter: I started in social work. I trained as a psychiatric nurse and worked in the area of homelessness for about 15 years. This, interestingly, has a strange connection to business, because it’s all about psychology and getting people to do things.

I entered the translation industry with a small company, originally called Softrans, which provided translation services for software companies. I worked at the same desk for 11 years while the name on the door outside the building kept changing. It went from Softrans to Berlitz, to Bowne Global Solutions, and finally to Lionbridge.

Back in the last millennium, when I was with Berlitz, the “dot com” era was starting. We had this vision for a translator portal. We partnered with a company called SYSTRAN, which was providing automated translation for Babelfish (the translation service on the AltaVista internet search engine). The idea was that, with something like a million jobs per day through Babelfish, if 0.1% of those would pay a human to translate, we would all be rich. But that wasn’t the case at all. We found that the translator portal remained unused until our salesperson visited Microsoft and demonstrated that, rather than requiring a phone call followed by sending files via File Transfer Protocol (FTP), it would be more efficient to upload the files directly into the new system. Only then did we end up with quite a clever business. By 1999, we had an online translation portal where everything was automated. I’ve been involved with technology ever since.

After Lionbridge, I worked for Idiom, which produced a tool called WorldServer. The founder of Idiom had been working in China during the localization of Windows 95. He saw that it was all being done file by file in a very primitive way. He had the vision of an enterprise resource planning (ERP) system for translation. At the same time, other people were coming to similar conclusions. Consequently, you had an industry that developed around translation management.

I came to memoQ in 2008, around the time that Idiom was sold to SDL. My first role at memoQ was essentially to create a strategy. We analyzed the market and saw where there were opportunities. One big gap was that the existing technology companies were not providing  adequate service to language service providers (LSPs), so we targeted them. We saw that Germany was the country with our biggest competitors.

We saw it as an opportunity to demonstrate our strengths. One of our differentiators at the time was our high level of compatibility with existing tools, which resonated with customers and sparked interest. That approach helped us gain traction and achieve success quickly in a competitive landscape.

Some have referred to wartime CEOs versus peacetime CEOs, but let’s say challenging times versus peaceful times. What has prepared you for these specific times we’re in?

Around 2008, when memoQ first started becoming successful, the financial crisis was causing people to spend less money on translation. In spite of that factor, we grew significantly during that period. If you can succeed when everything is going your way and every door is open for you, well, that’s nice, but it doesn’t mean anything when they start slamming the doors in your face. So, this is something we saw as important.

The current times are equally challenging. We’re seeing a lot of LSPs and translators having difficulties. For an LSP, if one customer changes its workload — say, brings translation in-house or decides to cut its translation budget — that can have a devastating effect. I think memoQ has to give the LSP the tools they need to become more productive, which is quite challenging. Internally, the crisis is a challenge to improve at what we’re doing and to try harder. We do this by learning more about the issues clients are facing and developing our software to really answer their needs.

At the same time, memoQ empowers enterprise clients to manage complex, high-volume multilingual content with greater efficiency and control. By combining automation, collaboration, and strong governance features, we have found that global organizations are able to scale their localization efforts.

Where do you see memoQ fitting into the current language technology landscape? What distinguishes it as a solution?

The solutions within memoQ are quite deep. Particularly, the tools around translation resources. We’re the only company that has essentially rewritten its translation memory (TM) engine to be much stronger and much more scalable. We’ve also got functionality that nobody else seems to have. One of these features is a set of tools called LiveDocs, which is a corpus engine with alignment at a document level rather than at a segment level.

Because we have put this energy into those translation resources, we are in a position where we can really take advantage of generative AI (GenAI), which is improved through context. To be able to use a large language model (LLM), you have to have high-quality materials, such as examples of previous translations. We found that, with memoQ Adaptive Generative Translation (AGT), we could provide a level of quality similar to neural machine translation (NMT) engines after they had been customized. Compared to non-customized, off-the-shelf engines, we were way ahead of them. With this retrieval augmented generation (RAG) way of working, we were able to get a customized engine for clients just based on their TMs and glossaries. And that was very powerful.

We’re not standing still. We’re constantly looking at how to improve our technology and our customer experience. One key area of focus is reimagining project management within the TMS through conversational interfaces. Imagine simply telling the system: “Create a project with these six files, in these languages, by this deadline,” and having it respond with a recommended workflow, resource allocation, and execution plan. That’s the direction we see the industry moving. Of course, human oversight remains essential to validate and refine those automated recommendations.

What role do LLMs play alongside machine translation (MT), specifically in your approach to technology in your own platform?

First, we’re constantly looking at all the different algorithms out there. We use the OpenAI GPT-4 model for memoQ AGT and globalese by memoQ. In memoQ AGT, we’re seeing a lot of excellent results. There’s a translation company from Canada that has reported a 55% increase in productivity by using this, and it was already using an NMT solution before that.

We’re also working on a project called memoQ fluent, which is an LLM-based MT portal that’s available to plug into the intranet of an enterprise. It’s in the alpha development stage, but we see it being offered to enterprises that will use their own data to power it.

We’re also doing a lot of work with globalese by memoQ and LLMs where a hybrid solution takes a customized NMT engine and uses the results as the contextual information to ask a prompt for a translation. The results we’re seeing are excellent, and it’s much better than just using an NMT engine in the traditional way.

Our approach is to always talk about something that’s actually real and can be used. If we are challenged, we can say why we’re doing it. We don’t go around claiming that because we heard Sam Altman say something, we’re doing the same thing. We try to talk about things that we’ve at least seen work in a demo.

Many of your clients are in the gaming industry. What services do you provide them?

One of the issues with games is that they have to be very scalable very quickly. They have to deliver multiple languages fast and efficiently. We’ve produced functionality that’s really helped them out. One feature, brought in about 10 years ago, is called XTranslate. This tool made it simple to copy translations to the newest version of a source document. Now, LLMs will happily create the functionality you need.

Here’s an interesting example. We sent Gábor Ugray, the creator of memoQ AGT, to test the technology in one of the most challenging language pairs: Korean and English. OpenAI, at the time, had indicated that this language combination could present significant difficulties. However, we saw great results thanks to memoQ’s advanced contextualization technology. This benefit is available to all users, but gaming companies in particular have found it very useful.

I’d also like to mention our Gaming Unit. It’s made up of localization experts from all over the world, and you can really feel their passion for gaming. They know the industry inside and out, which makes them a huge asset for any gaming company wanting to grow and take advantage of top-notch translation technology.

Another major memoQ feature is its on-premises functionality. Could you tell us more about that?

The on-premises functionality addresses the security issues that many companies are facing. For example, let’s go back to the gaming space. Say you have a Nintendo or Sony game that’s been leaked. That’s a really big problem. The user community for those games is very anxious to get hold of it as soon as they can. If they can spot something on Reddit or elsewhere on the internet, it will instantly be everywhere.

Therefore, security is important, and on-premises deployment provides exactly that. It ensures complete control within your firewall and security infrastructure. For example, one of the world’s leading game developers runs memoQ entirely within their own internal firewall, alongside their other internal applications. That’s about as secure as it gets.

This isn’t an issue for most companies. Most people are happy being hosted on a cloud server. The companies that want on-premises are companies like life sciences companies, gaming companies, or banks that have a very high level of security. With the Globalese acquisition, we can provide NMT solutions on-premises.

Speaking of the Globalese acquisition, can you talk a little about that and the developments in memoQ leadership?

In September of last year, we acquired Globalese and discovered that it had wonderful employees. They were enthusiastic about working in a bigger company, and we were able to grow the team quite quickly while improving globalese’s stability and interface.

Mugais Jahangir became Chief Revenue Officer for memoQ midway through last year. Since then, we’ve made big improvements in how we’re doing our business.

Until last September, Balázs Kis and I shared the CEO role. He has since transitioned to become Chief Evangelist, while I assumed sole CEO responsibilities.

It’s a challenging time for the industry, but we’ve focused on pushing ourselves by leveraging new tools and processes to deliver better outcomes faster to our customers. And, we’re seeing that effort pay off.

Finally, with so many developing technologies in play, there’s no shortage of predictions on where it’s heading. Would you like to add your perspective to that speculation?

Wired magazine recently published an article quoting various CEOs about the future of the smartphone. Elon Musk thought that by now you’d have a chip inside your head. Mark Zuckerberg thought you’d have spectacles with smartphone functionality. I think what actually develops from a technology can be difficult to predict. We’re at a stage where we don’t fully know how all these things are going to work.

There is a book by William Burroughs called The Soft Machine. In it, the original user of a system knew the system inside and out, but the subsequent users just knew which buttons to press. We don’t want to find ourselves in a similar position. We want to create something that’s actually working for people and putting them in control.