Natural language understanding
One of the guiding principles behind Mercury's natural language understanding (NLU) components is that everything in a bot can be done conversationally. Visual elements like buttons support the conversation, but everything can also be achieved by talking about it. This makes a bot independent of the particular channel, so that it can be ported from a messenger to a purely speech-based channel, where visual elements like buttons are not available.
All bots run a hybrid natural language understanding pipeline that combines machine learning based NLU with rule-based computational grammars. Grammars are used to jump start your bot's NLU, yielding a decent and accurate performance without the need to provide training examples. The machine learning component, on the other hand, is used to ensure robustness and make the system quickly adaptable in a pay-as-you-go fashion. The combination of rule-based grammars and machine learning technology represents a sweet spot that ensures a quick start, lowers efforts in porting solutions to new domains and languages, at the same time providing robustness and allowing to fine-tune the system while going by the targeted provision of training examples.
Last but not least, bots are designed to speak several languages. The platform currently covers:
We will soon also support:
- Brazilian Portuguese
If you want to use your bot in a language we do not yet cover, please get in touch at
email@example.com. The above list is just a starting point.