Managing Effective Machine Translation in an Ever Changing Environment

Recap of LocWorld31 Presentation and Interview with Olga Beregovaya, Welocalize VP of Technology Solutions

155_CreativeFocusIncWelocalize Vice President of Technology Solutions Olga Beregovaya joined Pablo Vasquez from NetApp to deliver a joint presentation on machine translation at Localization World 2016 Dublin. The presentation, “Managing an Effective MT Program in an Ever Changing MT Environment,” generated many thought-provoking discussions for LocWorld31 attendees and those enterprises who use machine translation (MT) solutions as part of their localization program.

Many enterprises tend to stick with their existing machine translation (MT) programs out of fear of the unknown and may have perceptions that there are high risks associated with migrating to a new MT program or adding an MT provider to their engine pool. The LocWorld joint presentation tackled these fears by outlining some of the facts and dispelling fears. The key message was to have an OPEN MIND.

In this interview by Louise Law, Welocalize Communications Manager, Olga highlights some of the primary considerations for enterprises when looking to migrate to the “next best” MT system, along with factors that may drive an enterprise customer to make the migration decision. Click here: LocWorld Presentation Managing MT to see Olga and Pablo’s LocWorld31 presentation, “Managing an Effective MT Program in an Ever Changing MT Environment.”

MT is always high on the list of key topics at LocWorld, what was the overall objective of your and Pablo’s presentation?

Our main objective was to talk to our LocWorld audience about following the innovation in language technology and approaching your MT strategy with an open mind. Pablo and I have worked together on multiple MT implementations and we wanted to share a hypothetical view of what would be involved in changing your MT system. If you want to add an engine to your pool or replace an MT engine, there should be no fear and you should be open minded. We’re talking innovation and disruption. When considering MT migration, many enterprise organizations think about whether they will lose they will lose their existing good work and whether it will deliver immediate ROI. These are all natural considerations but to keep your localization program fresh and ahead of the competition, you need to challenge your existing systems and the MT status quo. You have to have an open mind to succeed.

Why would you change MT systems?

MT engines are continually evolving and improving, delivering faster and better output and operating on more scalable systems. The level of effort from changing from one MT approach to another is now lower than, say, 10 years ago.

There may be performance problems with the existing MT systems. Lower quality output, speed and old high pricing models are some of the main reasons why enterprises look to switch. Clients are looking for innovative approaches to pricing MT. Price per word doesn’t work anymore, nor do expensive annual licensing models. Clients want a pricing structure that reflects level of MT use captures the quality and utility of the MT output.

What are some key considerations for global enterprises when thinking of migrating or adding MT systems?

When you change from one system to another, whether it is changing from RBMT/Hybrid to SMT or between SMT systems, you need to realize that your translators will see new errors, whether coming from the engine output quality or from integration issues.  There can be fewer errors, but they are very likely to be different. .

If you look at a new system, then integration and MT interoperability is extremely important. If you can’t smoothly integrate systems, then this will pose translation and engineering challenges for the translators and post editors. Each MT engine will produce different challenges and the translators and post-editors will notice these. For example, tags and placeholders could be problematic. The team must tune into a new mind set focused on how handle these new error types.

Dictionary support is also key and there are various ways terminology can be supported by an MT system. Will the new system support dictionaries? Especially if a lot of time and money has been spent building dictionaries in the existing system.

Feedback loops are also critical to the success of an MT system. If you don’t have dialogue with engine developers, then they’re essentially developing in the dark. Feedback must be delivered.

Clients are also looking for analytics on engine performance and quality for both predictive analytics and analytics after projects. If you come across a system that helps model post-editor behavior based on predictive analytics, then that engine is going to win over any other basic well-performing MT engine. Consider whether you want analytics as part of your solution.

There is a lot of talk in the industry about connectors and interoperability. How does this impact MT migration?

Connectors and interoperability play an important role in any enterprise MT program. At Welocalize, we pay a lot of attention to how the MT engine is integrated with our TMS solution, GlobalSight. Today, GlobalSight has many connectors into many of the leading MT engines, including: MS Hub, Google, Asia Online, ProMT and Iconic Translation Machines. We have four more connectors in development

We believe strongly in good engineering. While the engine quality is great, with the right connector to seamlessly integrate the system, without messing up segmentation, then any many problems simply go away and the post-editor experience is significantly more pleasant. Put a lot of effort into robust APIs and MT migration and integration becomes low risk and pain-free.

What’s in the pipeline at Welocalize for MT and language technology?

As a sneak preview, we are working on a universal connector. A universal connector allows us to integrate all MT engines into a single piece of middleware which allows us to expand our engine pool. We don’t want to reinvent the wheel for each client. One universal connector fits all. You simply drop into a client’s environment and customize as and when is needed.

Also, just like almost everyone involved in machine translation field, we are also experimenting with neural MT and the results are very encouraging.

Do you have one piece of advice for anyone looking to change the MT system?

weMTConsult an expert before making a big change. If you talk to a global provider like Welocalize, we can provide AB testing and analysis of existing and proposed MT systems. We can help with solid analysis and make implementation recommendations.

One final word of advice, don’t get too carried away with industry buzzwords and breakthrough MT technologies. Make sure your engine is implementation and production ready. Use something that has been tested commercially.

Click here to view LocWorld31 Presentation: Managing an Effective MT Program in an Ever Changing MT Environment.

For more information on Welocalize weMT solution, click here.