Multi-Language Online Handwriting Recognition

Area of Science:

• Computer Science
• Artificial Intelligence
• Natural Language Processing

Background:

• Online handwriting recognition is crucial for mobile text input.
• Existing systems often have limitations in language support and adaptability.
• Google's system addresses the need for a versatile and accurate solution.

Purpose of the Study:

• To present Google's comprehensive online handwriting recognition system.
• To highlight its support for a wide range of languages and scripts.
• To detail its architecture and novel components for high-accuracy text entry.

Main Methods:

• Utilizing a combination of state-of-the-art and novel components in a flexible framework.
• Implementing unified time- and position-based input interpretation.
• Employing trainable segmentation and minimum-error rate training for feature combination.
• Using a cascade of pruning strategies for efficiency.

Main Results:

• The system supports 22 scripts and 97 languages, including those not handled by other systems.
• Achieved fast, high-accuracy text entry suitable for mobile, touch-enabled devices.
• Demonstrated adaptability for both cloud-based and on-device recognition.

Conclusions:

• The developed architecture allows for easy transfer of improvements across languages and scripts.
• The system's flexibility enables deployment on diverse computational platforms.
• Google's handwriting recognition is publicly available in products like Google Translate and Android.