Next-Generation Conversational AI for Database Access:
The Technology Behind AutoQL
How innovation in conversational AI is revolutionizing the way today's leading businesses access and leverage their data
Chapter 9
Solving for Human Language Complexity
The custom language model not only encompasses an understanding of the structure of the database and the unique data present in said database, it’s also equipped with intelligence about the ways the humans that interact with that database are likely to ask questions about it.
To handle the nuances and complexity of human language, auxiliary machine learning models must be trained to support the conversational experience elements at play
Every industry has specific jargon, and individual businesses tend to use acronyms or terms that other businesses may not.
For example, a warehouse might use the term “work order” to describe data about a job to be completed for a client, but another warehouse might use
“customer request” to describe the equivalent.
To handle the nuances and complexity of human language (and field any inconsistencies in spelling or discern meaning from terminology that the computer isn’t familiar with) there also needs to be auxiliary machine learning models dedicated to supporting the conversational experience elements at play, beyond the intended capacity of the core language model.
Auxillary machine learning models work together to ensure that by the time the query reaches the translation portion of the journey, the AI fully understands what the user is asking and can return an accurate response to that query
These models enable more seamless conversational experiences and can be implemented
When users are not sure what kinds of questions to ask of their data.
When a user asks for data and the system needs to verify that it’s correctly matching the words used in the NL query with the unique data the user is seeking from the database.
To expedite the data exploration process with auto-populated suggested queries that relate to value labels specific to the database.
To verify that what a user is asking for is what the user is actually looking for when their initial input is too ambiguous for the system to understand off the bat.
A successful conversational AI system needs to have several machine learning models trained to handle each of these circumstances as they arise.
Leveraging the training data that has been generated and working in tandem with the core language model that has been built, these auxiliary machine learning models can be trained to recognize when a user’s initial input needs to be augmented in the moment to effectively generate the optimal SQL statement and return a helpful response.
