Machine and professional translation service Lingo24 has opened up its services via API. Speaking with ProgrammableWeb at last week’s APIcon UK, CTO David Meikle explained how the service can provide greater accuracy than big-name translators and how the API can learn as it goes to improve accuracy in any given industry vertical.
Lingo24 allows customers to choose the level of quality required for each translation text: from machine learning translation done automatically, to text reviews conducted by professional translators, to trans-creation services where a block of text in one language is completely copy rewritten in the destination language.
The service is used by international clients including the United Nations, the World Bank, Save the Children Fund and the telco Orange.
Now Lingo24 has opened up its services via API, allowing customers to integrate the translation service into their apps, products and websites directly, and to choose a relevant level of service as needed.
“Developers can sign up and get an API key straight away,” said Meikle.
Currently we have two APIs:
- An API for business documents translation. This gives you access to Lingo24's professional human translation service.
- An API for premium machine translation. It’s free for up to 100,000 words in pairs of English, French and Spanish. And then paid plans offer more words and access to more languages.
Creating a Context for Machine Translation
To make predictive machine learning products function well, algorithms need to work with a conceptual map that helps them better understand the context in which the text is relevant. In the last week, ProgrammableWeb has reported how machine learning service MonkeyLearn has built in this technique as part of its service, as has speech-to-recognition service MindMeld API. Both offer a way for customers to quickly create the knowledge base — or ontology — that can help algorithms make more accurate predictions. Louis Dorard, author of Bootstrapping Machine Learning, refers to this as the first task of predictive analytics APIs. First, train the model, then you can predict with the model, he told a fascinated audience at APIcon UK last week. (Dorard is also involved in organizing the upcoming international conference on predictive analytics APIs.)
For translation services, a similar-but-different process is used to build the ontology. Here, machine learning algorithms need to use phrase models for specific industry verticals to help align words and phrases in one language with the best translated terms in another language.
Lingo24 has focused on this by using Statistical Machine Translation techniques, said Meikle. "We have very focused vertical engines — banking, IT. The API selects the one that is best. We are looking more at semantic information, but if you send some text in off the fly, the API is looking at which engine to use to translate," he said.
Meikle sees regular, ongoing customers starting to build their own custom engines — similar to an ontology but in a phrase model — that helps regular machine translation improve its accuracy for clients:
We can help you build more custom engines and really drive that quality. We’ve got baseline vertical models and we can use use your existing translation memory, or other such assets as glossaries, to build an engine focused to you.
The challenge with machine translation is you need high quality pair-wise data. I need to understand that this segment in English equals that segment in Spanish. I need enough of that data in paired languages to be able to translate accurately.
The bit that really interests me about our APIs is the range of quality levels we make available, so you can target your spend on your content from machine translation to trans-creation.
Consistent Business Branding Across Regions
This is where ontologies come into play as a custom engine for Lingo24’s regular customers, Meikle said. As customers continue using the machine language services, the machine learning algorithm can build a database of most commonly used terms. Over time, this means that this information can be used either within a machine translation service or by professional translators to ensure translations match the sort of business language that the customer most commonly uses. In this way, the brand personality of the business can be preserved across languages. Said Meikle:
We do a lot of work on building up glossaries and dictionaries for different businesses. This is how one business wants to describe their work. And another business will build something slightly different. To further improve the glossary and terms you use, the translation memory over time is built up. We begin to know what terms are valuable to use.
Other Benefits: Security and Spend
Customers are already gaining multiple benefits from using Lingo24’s services and its newly released API:
- Security and privacy: Enterprises, law firms and startups appreciate the privacy of the Lingo24 machine translation service. “You may have concerns around using a popular service like Google for your translation because they reserve the right to re-use that content. We let you choose whether we store your translations. We can use that data to improve seed engines, but we don’t do that by default, you need to ask. We won’t track any of your content unless you want us to," Meikle said.
- Managing translation spend: “Often, those using our API are customers with high volumes of content," he said. "They can adapt their spend. For instance, they could try out their business presence in new markets and start with the machine translation API. As sales build in that market, they can move up in quality towards professional translation. We can help you pick the level of quality of the translation based on the value of your content. Once you are getting a bit of conversion from machine translation, then it may be worth increasing the quality of the translation.”
Lingo 24’s API is available in more than 30 language combinations, including the main European languages and Chinese, Arabic and Japanese. Users can access 100,000 translated words free of charge translating between any pairs of phrases in English, French or Spanish. Paid plans are available for those integrating the API with higher volumes or requiring more language combinations and technical support.