Good Machine Learning Practices (GMLP) in machine learning / AI/MT processes.
powered by human intelligence
At CQ fluency, we believe that the quality of machine translation is only as good as a the human intelligence behind its implementation. Our team of solutions architects do not only develop MT solutions, they fully customize them to your needs to maximize efficiency and optimize final output.
We leverage our deep understanding of healthcare, life science and medical documents and processes to ensure we implement custom, human-empowered MT solutions to solve a variety of use-cases. We offer fully automated or semi-automated document machine translation (MT) solutions, a web-based text translator, API implementations and custom connectors.
Our MT engines are secure and are fully managed by our team – removing the need for clients to involve IT teams on their side to implement a solution.
machine translation 101
rule based
these systems rely on algorithms based on grammar, syntax & phraseology of a language
statistical
train faster than RbMT systems as they rely on identifying parallel texts made available to the engine
neural
uses machine learning to teach the system how to produce the best results. this is becoming the standard MT engine but requires more CPU processing power
hybrid
a combination of two different methods of MT
adaptive
learns from real-time input and is similar to a more advanced version of translation memory
CQ fluency’s MT solutions
ease of use
– Web-based, on demand MT solution with user-friendly interfaces for all translation services.
– Optimized for short internal communications needing instant translation.
– MT can be applied to numerous file types (XML, Photoshop etc.) and integrations to client-specified systems.
security
– Transient – no request storage, no sharing.
– Single Sign On (SSO) with SAML 2.0.
– Optional use of VPN.
transparency
– Oversight on translation activity and spend.
– Audit capabilities to track usage.
– Reporting and visualizations to help empower MT usage/spend.
quality
– neural MT engines customized to your TM assets, glossaries and preferences
-increased automation and continual training of MT engine
our know-how
services include
– raw document/text MT for gisting
– light or full linguistic post-editing
– trained engines and continual training
– edit distance reporting to evaluate current MT solutions
– access through the CQ portal for added efficiency, security and on-demand service
Is machine translation with post-editing efficient?
Post-editing is the process in which experienced translators modify machine-generated results to achieve an acceptable final product.
Recent studies indicate that post-editing high-quality machine translations can indeed increase productivity in terms of turnaround times and reduce overall costs of translation. Editing poor machine translations can be an unproductive task and be more time-consuming and costly based on the hours of revision needed to create a quality translation.
At CQ fluency we make sure the use of MT is the right approach to a particular use case. We just do not implement MT blindly without doing a comprehensive analysis – making sure our results are production-ready before implementing the step into our localization workflow. This also impacts our results. The goal is to be more efficient by reducing the translation-editing time while preserving the preferred terminology and style that our clients require.
While other LSPs indiscriminately apply MT to any type of content for the sake of their revenue, at CQ fluency we are mindful about this approach, and we ensure our results are up to our customer’s standards.
one size does not fit all
Levels of MT vary widely depending on your specific use-case. For example, internal communications and emails probably do not require post-editing, while global purchasing policies, compliance documents, and clinical trial protocols require full post-editing.
Additionally, any translated materials that need to be submitted to regulatory agencies or patient-facing marketing materials are usually unsuitable for MT.
When choosing an MT solution, we understand that each client has different needs based on content and file type. That is why we employ a unique approach and consult with you to determine the best MT use for you, based on your quality expectations, deadlines, and budget.
avoiding common MT pitfalls
offering a one-size-fits-all service
At CQ fluency, before we begin the MT process, we consult with our clients to determine their needs and expectations. We then verify the use case, content, and document type(s) in order to offer the best solution that aligns with their project requirements.
Our technology experts, in collaboration with our sales teams have a portfolio of solutions adaptable to our clients’ needs. Even if the solution does not yet exist, our technology team can develop the solution accordingly.
assuming any language pair would get similar results
There are certain groups of languages that create more challenges than others and require further evaluation when comes to offering MT services. Assuming that English into a Foreign language is the same as Foreign into English when comes to MT results can prove detrimental to the final output. At CQ fluency we assess, test and evaluate our MT results in order to provide, “Production ready MT”.
failing to verify parsing rules and filters
A big part of a successful project is verifying segmentation and parsing of the content.
Very often this is also omitted, and linguists are left with a missing puzzle piece since MT (latest generation) works best when there is context. If tagging is separating ideas or line breaks are in the middle of sentences, results won’t be optimal.
ignoring feedback from linguists
Very often, a linguist is left with an MT result that is sub-optimal, hence why ignoring errors, such as incorrect words or mistranslations that occur repeatedly, can make a significant difference.
In many cases, these mistakes can be remedied quickly once the linguists flag any issues they detect. At CQ fluency we have a proprietary way to collect that information and improve results.
assuming that generic MT is as good as a trained MT
Language Service Providers commonly offer MT assuming that Generic results are going to be sufficient without prior evaluation. The result is an unhappy linguist that is left with more work than what was originally intended. This has a knock on effect on turnaround times and cost.
At CQ fluency we do not cut corners; the evaluation and assessment steps are well established to make sure the MT results are optimal, not only for the specific use case and language pair but also for the type of content. With the support of corresponding engineering steps and assets, a custom MT engine will produce the desired quality meeting deadlines and budget.
assuming linguists already know what to do in terms of post-editing MT
Mistakes that some providers make is assuming that all linguists know how to treat and work with MT results. Our Vendor Management team ensures that the linguists get the proper training when onboarding them for a particular job, providing the corresponding instructions and reference materials for a successful translation project.
future-proofing MT
When developing MT solutions, we leverage a proprietary methodology for reviewing datasets prior to engine training and evaluating the effectiveness of the training process.
We use state of the art technology, not only for MT related results but also for connectivity needs with our client systems and content processes. We are constantly evaluating all the possible solutions available, and our team of expert solutions architects continuously develop new solutions according to the specific client needs.
providing culturally relevant language services in your industry
MT solutions customized to your needs
Schedule a call with our innovative solutions architects to learn more about our MT and AI services, and discover how automation can elevate your translation and localization workflows.