Why your company needs a custom trained machine translation engine
Published on 04 May 09:31 by Erik Chan
What is a machine translation (MT) engine
In a previous article about neural machine translation, I described the AI technology behind machine translation in layman's terms. For the benefit of those who are less familiar, a machine translation engine is software that can translate a sentence from one language to another language.
Machine translation has been around for quite some time but historically the quality has been less than stellar. Nowadays the technology has come a long way from its origins. Modern machine learning have pushed the quality of machine translation into an entirely new realm and pushing human-level like accuracy. The cutting edge technology behind recent advancements in machine translation is called neural networks. Using neural networks to to offer machine translation is called neural machine translation (NMT). NMT is designed to learn language much like the human brain does, adapting to archives of translated sentence pairs you train it with.
Industry specific machine translation (MT) engines
As machine translation becomes more and more accurate, we are starting to see a need for industry specific machine translation engines. After all, a high quality translation always takes into account the context of the text it is translating. A simple word such as 'share' can be translated differently according to the context it is being used ('share' can be referring to a company share, Facebook share, the verb share, etc..) At TranslateFX, our base machine translation engines are trained specifically for translating financial and legal texts to mimic the correct vocabulary, syntax, and sentence structure used in finance and law. A great analogy are professional human translators, they also specialize based on particular verticals and industries to perform precise and quality translation that best fits the context. Often times, industry specific professional translators charge a premium for their experience and understanding of the terminology and knowledge in banking, technology, medical, engineering, etc..
Customized machine translation (MT) engines
While most companies can benefit from industry specific machine translation, neural machine translation technology allows us to develop software personalized to a client's requirements. This could be a machine translation engine developed for a specific type of document text and/or developed to mimic a certain brand, voice, or writing style. Without saying, being able to mimic the tone and style to keep translated texts consistent over a period of time is important to the original authors of the text.
The benefits of customized machine translation
The benefits of customized machine translation engines vs any typical machine translation are many fold. Most importantly, it improves the accuracy of the translation which in return decreases the cost of translation (spent on human post-editing and review). From our experience, the difference between google translate (a typical machine translator) and a customized machine translation engine is something between 50% and 90% accuracy. Meaning we have observed cases where google translate is ~50% accurate translating a particular document and a customized machine translator for that specific document is as high as ~90% accurate.
One other main benefit of personalized machine translation is translation style and usage of terminology. Custom machine translation can adhere to particular syntax usage and make sure the terms stay consistent in all translations. Company and people names are typically inaccurate and not-consistent when translated with generic machine translators while customized engines have been trained to adhere to particular glossary translation definitions.
One final and also very important benefit of customized machine translation is consistency. Because the machine translation engine is a piece of software, we can expect it to function the exact same way it was developed every second and day of the month. It ensures the translation will not change regardless of changes in human resources (employees leaving, getting sick, and other management factors.) The hidden costs of human resources is often overlooked when it comes to something that has historically been human and manual labor intensive.
How much training data is required
Machine translation engines require volumes of quality training data. Volumes are typically millions of translated sentence pairs (in both languages) and are usually in the double digit gigabytes range. The quality of this data is just as important as the quantity, there is a saying 'garbage in garbage out' and this is more true than ever for training neural machine translation engines. Training a machine translation engine with enough data allows it to learn from enough different scenarios, and the quality of the data (translation) allows the engine to mimic quality translation.
The great news is some custom machine translation engine providers like TranslateFX are already equipped with this huge volume of data in the respective vertical or industry, and can assist to develop personalized machine translation engines with much less data from a client or organization.
Do I need a custom developed machine translation engine?
Companies and institutions who hire one or more professional in-house translators will benefit from industry specific machine translation engines. I have often found that companies tend to select and pick the documents they choose to translate based on the capacity of their translation team and what is absolutely necessary due to the fact that it is labor and time intensive. The use of any machine translation engine will nearly always improve the capacity of an in-house team.
Custom developed machine translation is more suited for:
i) medium to large translation teams of at least three translators,
or ii) where the texts to be translated are specific to a particular document type or industry jargon,
or iii) the need for a specific language style, syntax, tone, or terminology is required in the final translation.
If you are interested in learning more or figuring out whether machine translation makes sense for your organization, feel free to ping me at erik[at]translatefx.com, I'm always happy to chat.
Key Takeaway to Share:
- Everything you need to know about Translation Memory
- Hiring post-edit translators
- AI Machine Translation and Terminology Consistency
- Best sites to hire professional translators
- Hiring an in-house Translation Team vs. External Agencies
- Private Banking Digest - 8 June 2020
- Add AI language capabilities to increase cross-border business transactions
- Equity Research Digest - 8 June 2020
- Top 3 challenges translating equity research reports
- What is Neural Machine Translation & How does it work?
- Private Banking Digest - 8 June 2020
- Equity Research Digest - 8 June 2020
- Importance of glossary in legal and financial translations
- Reshaping equity research business with AI
- With the rise of AI translation, why hire human translators
- Is AI Ready to Translate Financial News
- AI translation for Company Financial Reports
- Applying AI translation to Equity Research Reports
- Does my company need a translation management system
- Should my company implement AI translation tools
TranslateFX develops AI translation technology specifically for financial and legal institutions. The company develops AI models and workflow tools for clients of all sizes. We believe humans always play and important part of the process and our tools reduce the time and costs of translation by 60% or more.
- China Securities Regulatory Commission
- China Banking and Insurance Regulatory Commission
- China Banking Regulatory Commission
- People's Bank of China
- U.S. Securities and Exchange Commission
- U.S. Financial Industry Regulatory Authority
- U.S. Financial Accounting Standards Board
- Hong Kong Securities and Futures Commission