Overview

Baidu SmartLearn Doc Translation is a SaaS platform that combines NLP technology with the speed of machine translation and the accuracy of human expertise. We aim to refine machine translations by merging human translation with machine learning, enhancing domain-specific understanding for better accuracy and cost-efficiency.

63% usage for Terminology and Memory features
NPS Increase from 18 to 62

Impact

User Satisfaction 31%↑

Translation efficiency 26%↑

Translation accuracy increased by 28%
Average processing time decreased by 10%

Machine translations lack accuracy, while human translations are high in cost and time

Background

In the translation market, a clear divide exists between the speed of machine translations and the accuracy of human translators. While machines are fast, they often lack detail, particularly in specialized fields. Conversely, human translators achieve higher accuracy but at the cost of time and higher expenses due to the complexity of the job.

Why are Machine Translations often inaccurate?

Step1

Next, we conducted a secondary analysis and 3X stakeholder interviews with translation technology experts, professional translators, and clients with translation needs, aiming to identify key pain points and opportunities in machine translation.

How can we address the Inaccuracies in Machine Translation?

we explored various methods like human reviews and pre-loading terminology. Our experiments showed that combining machine learning with smart AI significantly improves translation accuracy and speed, while reducing costs. This approach effectively trains the system to blend human expertise with technological efficiency.

Step2

Step3

How Can Machine Learning Make Translations Better?

We integrate machine learning with human input to enhance machine translations, continuously refining their accuracy. This approach blends AI efficiency with human precision, improving translation quality and efficiency over time.

Simulating color blindness on the editing dashboard revealed color-only highlights were unclear, reducing efficiency. We added cues like borders and underlining, then retested, significantly enhancing visibility and accessibility for color blind users.

Enhancing Accessibility for Color Blind Users

Iteration# 01 Accessibility

Iteration# 02 Efficiency

The design iteration addresses the challenge of domain-specific vocabulary by mandating domain selection at project creation and aligning all terms with their appropriate domains, significantly enhancing translation accuracy.

Domain-Specific Vocabulary Precision

Embracing Learning in Unfamiliar Domains

Diving into data and machine learning was tough without any background knowledge. But, my eagerness to learn, constant questioning, quick learning, and working closely with the tech team not only helped me grasp the concepts but also played a big part in our project's success. This experience showed me how being adaptable and keen to learn can help overcome challenges and lead to great outcomes.

Putting user needs and diversity at the heart of our methodology shifted our approach significantly. By focusing on designing for a wide range of users, including those with specific accessibility needs, we were able to make more inclusive and effective solutions. This commitment to understanding and catering to the diverse needs of our users not only improved our product's usability but also reinforced the importance of a user-centric approach in creating meaningful and accessible technology for everyone.

Takeaways

Prioritizing User-Centric methodology and Diversity