Interpreting input - Natural Language Understanding

Understanding free text input is the basis for every AI Assistant. The transformation of text messages into data that can be processed by computers, including the interpretation of what is being talked about and what is desired, makes natural language interfaces possible in the first place.

While the NLU performance today has reached a good level with many providers, decisive quality differences lie in the training efficiency. Mercury.ai offers you a proprietary, hybrid NLU technology that combines different technological approaches to understanding natural language. Machine Learning-based and symbolic approaches offer complementary strengths. This approach makes it possible to achieve higher automated language performance with a much smaller amount of training data.

Annotation and bot training

To permanently improve the performance of speech comprehension, you have the possibility in the inbox to simply annotate messages that are not understood and thus inform the bot of the meaning of the message. The annotations flow continuously into the training of the AI Assistants and effectively increase the robustness of the NLU performance. Annotations can not only originate from the inbox but they can also be entered manually. Furthermore, annotations can also include information about your content and erroneous annotations can also be corrected or removed later.

Annotation Format

Multi-Language

Each AI Assistant can communicate in multiple languages. This allows you to use one underlying dialog structure across multiple languages. For effective internationalization, where you retain brand and content control without having to create a new bot for each new language.

NLP Debugger

Test and debug the language understanding of your AI Assistant in the NLP debugger. It provides deep insight into the interpetation structure of input and enables you to further develop your AI Assistant.

NLU Model

Each AI Assistant on the Mercury.ai platform has its own NLU model. This way, language comprehension is adapted to the quality and performance requirements of your company. The machine learning components and models are automatically optimized and managed.

Training Data Manager

The central place for the overview and editing of language comprehension data is the Training Data Manager. Here you can find an exact breakdown of all the content that trains the language comprehension of the AI Assistant. The search function allows you to identify specific training content and the edit function enables you to work on training items.

Edit Mode