Natural Language Interfaces

There are multiple interfaces in Pyramid that allow users to ask questions in natural language and get a response. This allows less technical people to access sophisticated functionality using normal language, greatly accelerating their use of the system. How you access the natural language functionality depends on the module you are using but the Chatbot is the primary conduit.

The use cases for natural language range from asking for background images and Python code scripts to generating new visualizations and entire storyboards. The latter examples are sometimes referred to as Natural Language Querying or NLQ, since they involve querying your data. However, 'NLQ' is a subset of the natural language functionality. In effect, NLQ offers a way to query data quickly and intuitively; without the need to navigate the complexities of data models or how to select and configure an appropriate visualization. For example, if you want to build a visual, Pyramid interprets your questions and automatically finds the relevant hierarchies, measures, and elements, places them into the appropriate drop zones, and selects and builds the optimal visualization given your data.

Note: This feature is available with the Enterprise Edition license only. For more information, see Pyramid Licensing.

Natural Language Capabilities

The following capabilities help you to understand the type of activities that Chatbot can help you with:

  • You can use the Chatbot to build and analyze visuals, presentations, and publications at design time, and to explore presentations at runtime.
  • You can use the Chatbot to interact with your data in a step-wise manner, creating and then changing your visuals to reveal more about your data.
  • You can use the Chatbot to modify visuals after they have been generated, changing a grid to a pie chart or redrawing your visual once you have added new data.
  • You can add filters or set slicers in visuals, just by asking for them using your prompts. For example, if you say "Create a sales report for the last 120 days," 120 days is interpreted as a date filter and added to your visual.
  • You can incorporate prompts where the intent is complex and the language is natural, and the Chatbot can break down the request and respond appropriately. For example, "Generate a report to show Sales by Team, County and State, for only 120 days." or "Compare sales to expenses by promotions and show a regression of sales to expenses."
  • You can add synonyms for your LLM, to make sure your reports contain all relevant content.
  • You can generate explanations or insights, to provide more information about your visuals or the slides or pages in your presentations or publications.

Note: If it is clear that your request isn't supported, the Chatbot will try to respond with information about why it couldn't handle the prompt. For example, if you are in the Discover app and you ask the Chatbot to "add a picture of a kitten," it responds with a message indicating that it isn't able to add a picture.

Language Models

Your natural language requests are processed by sophisticated engines in Pyramid which pass on the message for interpretation to an underlying language model.The result-set returned from that interpretation is used to script, build, and analyze your data based on your request.

Your administrator can select different language models, with the two main "types" being:

  • Pyramid's built-in Portable Language Model or PLM. This model is designed to provide a lightweight, fast engine for processing natural language requests in English.
  • Third-party Large Language Models or LLMs, which can vastly expand the interpretive capabilities of the system and operate in multiple spoken languages. One or more powerful LLMs are available for use.

This multi-pronged approach is the basis for Pyramid's natural language strategy.

Warning: If you are using one of the LLMs (not the PLM), your assets will be generated using public domain algorithms. This can produce erroneous and inconsistent or random results. Use at your own risk.

Multiple LLMs

As described above, there are a number of different LLM Engines that may be used to interpret your Natural Language Queries (NLQs), including the internal PLM. Your selection affects the behavior and reach of your Chatbot.

  • For more information about LLM engine selection and the associated effects, see Multiple LLM Engines.

Limitations of LLMs

Be aware that when you use the Chatbot to interpret your natural language queries, you may experience the following limitations:

  • Any query takes 30 - 40 seconds of generation time to produce a result.
  • Any query that includes reference to facts and figures will draw those facts and figures from the "domain of public information."

NLQ and Chatbot Functionality

NLQs are useful at two separate phases of the analysis process: Design Time, when you build and analyze your visuals, publications, and presentations. Runtime, when you interact with presentations as a reader.

Design time

At design time, when you are building and analyzing your visuals, publications, and presentations, you can access and make use of the Chatbot in Discover Pro or Lite, Present Pro or Lite, and Publish Pro or Lite. In Discover, you can also use NLQs to build your visuals when you "Ask a question" to begin your Discover session.

Runtime

At runtime, you can interact with presentations to focus on or transform particular aspects of the presentation that are important to you.