Machine Learning: Going Beyond Machine Translation (Webinar)

As the saying goes: the more you do, the more you know. Just like humans, machines can be trained over time to recognize patterns and respond with highly specific actions.

Jennifer Chew and Benjamin Loy walk through different ways content creators can use Machine Learning with Smartling to increase efficiency and create high quality translations in the platform.

Summer School | Machine Learning: Going Beyond Machine Translation from Smartling on Vimeo.

What's machine learning?

It's a way to model and perform desired actions at scale. According to the University of Washington, "Machine learning algorithms can figure out how to perform important tasks by generalizing from examples." Essentially, algorithms are built to recognize patterns within existing data points. Then, these patterns are applied to new data points to inform actions to take.

How Smartling uses machine learning:

Here are some examples of machine learning in the Smartling platform:

1. File Type Detection - When customers load their content into the Smartling platform, Smartling uses a support vector machine to identify file types and parse apart strings accordingly.
2. Quality Confidence Score - Smartling uses a decision tree to decide how different data points impact the overall evaluation of confidence in the quality of a translation.
3. Tag Alignment - HTML tags - like for italics - sometimes get misplaced when parts of speech appear in different places in a different language. Smartling's machine learning is trained on adjective placements in French to ensure that tags are moved to the correct part of sentence, from the English to French translation.
4. MT Auto Select - Smartling's newest tool, MT Auto Select, will be able to ingest words and group them thematically (e.g. Health, Restaurants, Legal), and then assign the best MT engine to translate that specific theme of words.

It's important to note, though, that while machine learning algorithms can uncover powerful insights at scale around your translations, they are never 100% perfect. Always spot-check your results and let us know what you find. We're always training our models to create ever more accurate insights.