How to Improve Your Relationship with Machine Translation
By: Heidi Depraetere (CrossLang) and Adam LaMontaigne (Moravia) - CrossLang
14 May 2014
At the dawn of a new era of data processing, the paradox of machine translation still eludes even the biggest localization firms. How can LSPs proactively embrace this technology and develop the knowledge, resources, processes, and relationships necessary to finally live up to the promise of MT? Authors Heidi Depraetere and Adam LaMontagne advise: success may come with partnerships.
The democratization of MT technologies in the past decade has the potential to revolutionize translation as we know it. However, MT technologies alone are useless if we do not have intelligent means for developing, implementing, and maintaining these technologies for real production environments. The MT technologies available today are simply tools; trained experts are needed to extract the best results. The good news is that the Language Technology (LT) and Language Service Provider (LSP) communities are poised to help MT customers maximize the potential of MT.
The benefits and limitations of MT have been hotly debated topics within the language industry for decades. On the one hand, skeptical (potential) MT consumers question the technology’s ability to produce sufficiently high quality translations. On the other, the demand for multilingual content far exceeds the available supply, and MT is seen as a solution that can help to bridge the gap.
One thing that professionals and pundits on both sides of the debate can agree on is that MT has become a common feature of the localization landscape. As a result, many language professionals are seeking to maximize the potential of MT for their projects and customers – regardless of whether they believe in the benefits of MT. (Despite the fact that open-source and commercial MT technology, tools, and services are widely available, many players in the language industry still find it difficult to tap into the benefits promised by MT. The result is that a good portion of the language community is currently not maximizing the technology’s potential.)
This problem flies in the face of the well-known narrative of disruptive MT technology. The promise of benefits is there, but the reality is not materializing. Why is this so?
An oft-referenced CNGL study published in 2013 has this to say on the matter, “For most, the successful implementation of MT represents a huge leap and long-term investment in resources and expertise before return on investment becomes apparent.” From this, we can extrapolate that a major barrier preventing LSPs from fulfilling the full potential of MT is that the integration of an MT solution into a working production environment requires expertise that is in short supply. Sure, Moses (for example) is free, but to deploy it as a solution requires (in addition to the engineering needed to integrate the technology into existing workflows) an informed understanding of mono- and bilingual corpus development and management, pre- and post-processing, language models, phrase tables, statistics, and so on.
GALAxy readers will most likely be familiar with this paradox. As MT technologies have become more readily available in recent years, we have learned that MT is not a plug-and-play solution. If, instead, we can view MT integration as an exercise of connecting people, processes and technology, we can begin to see MT as a significant opportunity for the language community in the years to come. It’s simply a matter of adjusting our perspective.
Language Industry Players: Some Definitions
“[E]xpert use of technology is not equal, so in many cases technology differentiates rather than equalizes.” — Jost Zetszche, Q1 2014 edition of GALAxy Newsletter
Language industry companies can add enormous value and set themselves apart by exhibiting expertise in the MT ecosystem. In this ecosystem, we see numerous players, but we would like to focus on two specific types of industry actors that are likely to represent a large portion of the GALAxy readership, namely LT Integrators and LSPs.
We define LT Integrators as those language companies that build domain or client-specific MT engines that can be integrated into client workflows, serving a variety of use-case scenarios. They typically offer data selection and acquisition services, expertise on natural language processing, engineering know-how and training, and use case-defined quality evaluation.
LSPs can play various roles in the machine translation ecosystem. Some – typically larger – LSPs may have an internal language technology team that has its own machine translation development and integration business. Other – typically smaller -- LSPs may be more strictly consumers of MT and MT-related technologies.
MT Integrators and LSPs of all types, shapes, and sizes can add to their value proposition by taking a more proactive approach to integrating MT technology into their services and service offerings. Moreover, there are many instances where LT Integrators and LSPs may find common ground and can benefit from working together.
LT Integrators: A Model for Success
LT integrators are in a good position to democratize MT from an elitist to an industry-ready technology, bringing down the barrier and positioning MT as a standard building block, as opposed to an ad-hoc service with limited scope. Many opportunities for MT stretch beyond the traditional localization production context where MT is just another productivity enhancement tool alongside TM.
To succeed, LT integrators must demonstrate that they offer value at the heart of the many MT propositions out there. They must segment their audiences and decide where to devote their energies. They can focus on verticals and/or locales, target LSPs or end-clients. They can focus on providing engineering-centered solutions and/or be more innovative by combining natural language processing with MT. They can be strategic partners or provide ad hoc project-based solutions. The suppliers that provide solutions that truly match their clients’ MT application context will often win the day.
How Do LSPs Fit In?
Today, many LSPs find themselves stymied by the MT paradox. All the tools are ostensibly available, but attaining a level of success in this area still seems to be out of reach. As such, many LSPs are standing on the sidelines of — or at the very least, playing a passive role in — the MT revolution.
However, there are many ways in which today’s LSPs can proactively participate in the growing opportunities provided by MT. Specifically, we would like to focus on two primary ways in which an LSP can become a more active participant in the MT market: as a partner with an LT Integrator, or by committing to internal investment in MT.
Even in 2015, the largest of localization giants (developers, users, and consumers) do not yet have a complete end-to-end model effectively in place for handling the complex, inter-related MT tasks including, among others, identifying MT candidates, building and managing bilingual corpora, training, testing, deploying, and evaluating engines and, long-term, data-driven tuning, maintenance, and retraining.
That is to say, you’re not alone. At the same time, it is unreasonable to think that all of those important and often complex tasks can be handled by a few language leads working with the IT department. Running your own internal MT group can have huge benefits for your organization, but in order to take a serious crack at it, it requires a long-term investment in all areas of MT, including but not limited to IT as well as (in most cases) senior sponsorship to ensure its likelihood of success.
The good news is that, as an industry, we have a long way to go before MT is a solved problem. LSPs that host their own internal MT group can achieve many benefits, ranging from simply having increased flexibility and visibility vis-à-vis the deployment of MT on their projects to potentially being perceived as a thought leader in the industry, both of which can drive revenues and margins.
On the other hand, not every LSP is in a position to invest in MT in such a significant way. If, for whatever reason, your organization falls into this category, never fear: there are still ways to get ahead of the curve and proactively dismantle the MT paradox. The solution? Partner with an LT integrator.
LSPs and LT integrators can, in fact, find mutual benefit in a partnership. The LT integrator can help bring order and experience to the MT paradox and the LSP can bring volume and visibility to the LT integrator’s portfolio.
In this way, LSPs and LT integrators can focus on their main areas of expertise, improving efficiency and enhancing the MT marketplace as a whole.
The barriers to providing and putting MT to work in real-life localization scenarios have certainly come down significantly in the last decade. However, there is still some demystification needed to ensure that we leverage MT’s untapped potential. As the body of industry experience grows, there will be tremendous demand for a maturing MT service sector.
Working individually and in partnership, some LT integrators and LSPs will benefit from the opportunities presented by this growing service sector. In order to be part of this MT vanguard, organizations must proactively embrace the technology and develop knowledge, resources (human and language), processes, and relationships specific to MT.
Only then will we be able to overcome the current paradox of MT and live up to (and possibly exceed) the promise of MT.
Heidi Depraetere has over 20 years’ experience in the localisation and language technology industries. She is a founder and director of CrossLang, a privately owned consulting and systems integration company dedicated to translation automation technology.
Adam LaMontaigne manages Language Technology Development and Deployment at Moravia.