Following our launch of KantanNeural™ engines as part of our KantanFleet™ repository of pre-built MT engines, we received a number of questions and interest around the product. To address these questions, we asked Tony O’Dowd, CEO and Chief Architect of KantanMT.com a few questions about the Neural Machine Translation engines on KantanMT, the features and benefits of these engines and the impetus behind launching KantanNeural. Continue reading
The innovative Machine Translation features released by KantanMT, along with our contribution towards improving automated translation workflow has earned us the reputation for being thought leaders in the industry. A few months back, we released a white paper on what global companies can expect to see in 2016 for Machine Translation (MT). Continue reading
KantanMT recently published a white paper on what global companies can expect to see in 2016 for Machine Translation (MT). The MT industry is rapidly changing and moulding itself to the technical needs and globalization requirements of the present day. Our white paper puts forward six major MT trends that all businesses need to heed in order to stay relevant and ahead of their competitors.
The first part of this two-part series examined the factors a multinational organization should consider before introducing MT into their business workflows. It also described how Machine Translation can help to break down barriers to communication within these organizations, improve internal communications and increase productivity as a result.
This second post looks at the challenges of implementing MT from a Language Service Provider’s (LSP) perspective, and examines how MT could be used to improve the LSPs competitive positioning and productivity, as part of the translation business model.
The decision for an LSP to adopt Machine Translation technology is almost always strategic. LSPs have been leveraging the use of translation memories since the widespread adoption of the technology in the 80’s. Translation memories or bilingual corpora are the primary assets for training customized Statistical Machine Translation (SMT) engines.
According to Joshua Duhl, localization is important for three main reasons:
- Effectively engage customers
- Maximize revenues
- Maintain global brand recognition
Why are LSPs using Machine Translation?
The key drivers for the introduction of Machine Translation are technology advancements and changes within the global communication dynamic – factors which are increasing the demand for real-time translations of multilingual content. In response to this demand, MT is becoming a necessity for many LSPs who want to offer their clients the full spectrum of translation and localization services.
Language Service Providers benefit from the ability to leverage their language assets (TMs) when introducing MT into their business. This greatly reduces the time it takes to get started with MT. Translation Memory has a firmly established positioning as the industry’s main translation tool, and now, because the two can be integrated, LSPs and translators can reap the benefits of combining both TM and MT technologies in a bid to increase efficiencies, and enhance their service offerings.
Before integrating MT, an LSP should take certain factors into consideration.
Rustin Gibbs, Senior Solutions Architect and Joe DiDamo, Senior Business Development Manager at Jonckers Translation and Engineering carried out a critical study on the challenges of integrating MT into the LSP workflow. I listed some challenges from the article below:
Challenges of implementing MT into the translation pipeline
- Perception of MT – Machine Translation is often considered a sensitive subject. Negative perceptions based on earlier versions of the technology still remain strong. Convincing those involved with language services, on both the client and translator side that MT can be a useful translation tool will take time.
- QE and Post-editing effort – Quality Estimation and calculating post-editing effort is the biggest factor hindering MT adoption. While theories and technologies for measuring quality and post-editing effort are available, there is still no widely accepted industry standard.
- Training Data – the quality of freely available training data can be unreliable and sourcing or creating high quality training data is costly and time-consuming.
- Language assets – even though LSPs have access to their client’s language assets, the client’s intellectual property rights mean LSPs may not be in a position to leverage these across different client projects.
- Language Pair – both the language pair and domain play a part in the quality of the MT output. Translating content to and from more complicated language combinations can be less successful, as LSPs face the challenge of collecting higher volumes of quality to make sure the engine performs at an optimum level.
How should LSPs integrate an MT system?
As with multinational organizations, the decision to integrate Machine Translation should be done with a well-defined implementation strategy in place and realistic expectations about the integration.
LSPs thinking of implementing MT should ask themselves:
- What are the potential costs? How will the costs and resources be managed while building and training the MT engines and which MT system best suits my needs. Do we have the resources and capacity to license the software and build our own systems or should we outsource to MT vendors?
- Can the technology be implemented with as little disruption as possible within our existing localization workflow? One of the biggest challenges of implementing a new technology is an increased risk of setbacks and issues while users adjust to using the new technology. The time required for training and fixing these issues should be factored into the implementation process.
- Project Managers should be able to fully understand MT technology, so they can make informed decisions when deciding on the most efficient MT solution for each individual project workflow (decisions based on domain, language pair, MT engine quality, project lead-time). PMs should also be familiar with the engine’s training data and its similarity to the writing style of the source text.
- How can we leverage language assets across clients and what possibilities are available for sharing assets? Being able to share language assets will cut the costs and time required to create and collect training data. Building partnerships with organizations and companies can reduce the burden of creating language assets.
This is not an exhaustive list of points for LSPs to consider before implementing MT, LSPs should also ask some of the questions outlined in the first part of this post ‘Leveraging MT to Improve Productivity’.
Once LSPs have sat with all these questions, they should have a clear picture about their MT needs and their capacity to implement it within their business. LSPs may want to gradually introduce MT across their key domains and language pairs, and over time, increase this as demand develops.
LSPs who integrate MT will quickly see the benefits from translation economies of scale – TMs can be leveraged across translation processes, which means that they can give more value to the company.
Thinking about introducing Machine Translation in your LSP? Email Niamh (firstname.lastname@example.org) to set up a demo and discuss a KantanMT solution that fits your business needs!