To launch off Safaba’s new ‘Global Thoughts‘ blog, we thought it would be interesting to share with you some of the background story about how and why Safaba got started, and what has changed since then….
My passion throughout my research career has been (and still is) focused on developing advanced language technologies that can have a real impact on people’s lives. I’ve been working on MT research projects for many years, but the real-world impact of my work was still somewhat lacking. The advanced MT technology that my colleagues and I were working on in our research labs was clearly not being adopted by global enterprises and the translation industry at large, and I was curious to find out why. I came to realize that two critical components were missing at the time: brand-specific individualization of the technology and effective technology integration. I had some good ideas on how to address these two issues. So my co-founder, Bob Olszewski, and I decided to launch Safaba to explore these ideas and see if they could propel the broad adoption of MT in global business.
Integration and delivery were addressed by developing a hosted solution and business model that would dramatically simplify the engineering challenge. The more interesting and difficult problem was addressing brand-individualization. Corporate language is nuanced and uniquely individual. Services that offer “generic” automated translation are therefore fundamentally not well suited to the problem. When it comes to corporate language, one size does not fit all. Enterprises typically have some existing translation archives of their own unique language, which are ideal resources for developing customized statistical MT engines. The challenge is that most enterprises don’t have sufficient quantities of their own data for training high-performance MT engines. Commercial solutions that offer customized MT services were not able to fully address this problem.
Our approach was therefore focused from the start on developing novel technologies that would automatically learn from even limited amounts of corporate data and would yield automated translation that is highly sensitive to corporate language in both choice of words and language style. We called this focus ‘Enterprise Machine Translation’. By delivering domain relevance and brand consistency to Machine Translation, we expected that translation productivity would quickly improve and adoption by the market would follow suit.
The introduction of our Language Optimization Technology™, Language Transformation Technology and other technological innovations addressing specific challenges in corporate-language automated translation has lead to some very positive market interest in Safaba. Widespread MT adoption has, however, been slower than expected. The translation industry at large still approaches MT with quite a bit of confusion. With various approaches and solutions available in the market, selecting the most suitable one is not an easy task. Different solutions provide different value. Potential clients therefore expect us to clearly articulate and demonstrate the unique value of Safaba’s brand-optimized Enterprise MT technology to global business. The most direct measure of MT value within large scale localization projects is improvement in post-editing productivity. Measuring post-editing productivity, however, remains a complex and highly-debated practice requiring significant expertise. This continues to impede the broad adoption of MT and MT post-editing in traditional localization projects.
On the other hand, new applications of the technology involving instant communication are quickly evolving. Global enterprises show growing interest in instant Machine Translation for corporate customer support, sales and marketing, and communications, where MT value clearly aligns with corporate KPI’s. These applications require “raw” MT in real-time, but brand-individualized MT is still extremely important in such scenarios. Our focus on the unique challenges of maintaining brand fidelity in multi-lingual on-line communication channels is leading our way to new innovations in MT technology.
I look forward to updating you about these innovations in future blog postings in the months ahead.
Stay tuned!
Alon Lavie,
CEO
The slower than expected adoption of MT by global corporations has many quite rational reasons BUT stands in direct contradiction to the potential value it can and already does provide to global business today.
Having a positive and substantial impact on corporate key performance indicators, I strongly believe EMT should be regarded by global corporate executives as the next step in propelling their globalization strategy forward.
More in my next blog entry…
Stay tuned!
@Udi, I agree. This industry is due for some new upstart to change the way people feel about it. Speech rec took a long time. Dragon Naturally Speaking, for instance, took about 12 years to become useable out of the box, and MT is certainly a harder nut to crack.
Until semanitcs can be properly incorporated, targeted statistical strategies will be our best hope. I suspect that Google is never going to cater to any specific customer, so there’s a huge opening for someone who will.
Although nobody wants to commoditize the technology at this point, there could be value in publishing quality metrics, but they would have to be more meaningful than BLEU. Something like HMEANT (Lo & Wu 2011), could go a long way to changing the public perception. So many decision-makers are monolingual and therefore cannot measure a product’s quality themselves. It’s hard to know whom to trust.