Training Prototype NLU

An overview on how to create smart prototypes using Voiceflow's built-in NLU.

Overview of Smart Prototyping

Today, Conversation Designers need to have a strong understanding of user experience design, as well as the underlying conversation logic and nuances, that supports their designs.

When prototyping and training your NLU/NLP, designers can dive a layer deeper than surface-level happy paths. In fact, when conversation designers train their experiences, they’re searching to understand how their flows will work in conversation in human-contexts; they’re looking to validate their user experience.

Now with the addition of NLU/NLP, designers can now conduct higher fidelity user-testing all from within the tool.

Voiceflow Designers can now easily access NLU/NLP powered tests more rapidly, unlock increased understanding of their experience, and enable more rapid design and prototyping cycles.

Training Your Prototype (NLU)

You can ensure that your Voiceflow conversation prototype is trained with the model (utterances, intents, entities) in your Voiceflow project coupled with the analysis, handling & intent/task fulfillment, or NLU.

Voiceflow will prompt you if your Assistant needs training to create a high-fidelity testing experience. In entering Test mode, you will be prompted or see a display message in the Training window drop-down (above the Dialog window drop-down).

Tip: Always train your model after adding or modifying your Intents, Utterances and Entities. This ensures you’re testing with the most up-to-date model, and is best practice.

To put it simply, we train the data so that your conversation understands how to deal with unknown data, input or user replies.

Training a model simply means learning good values or paths/intents to take in your conversation for all the weights and the bias from labeled examples, such as entities and utterances contained.

The more you build, the more you should be training – and training your most recent model will be the most accurate depiction of your real user experience. Training your NLU is a critical part of creating a high-fidelity testing experience.

With Smart Prototyping & our native NLU/NLP, we now support various new use cases and testing opportunities, in combination with our General Assistant projects. This will make it easier for designers to test at a higher level of intelligence and teammates/user testers to experience a more representative version of your designs.

Training (NLU) Duration

After you hit Train Assistant, it can take up to a few minutes (20 - 200 seconds) to train your assistant. Typically, the timing depends on the number of intents and other conversation logic (utterances & entities) in your project and Model Manager/NLU model.

Upon completion of the training of your project's model, you will notice that the Train Assistant button is greyed out and you are now using the highest-fidelity testing for your assistant.

Best Practices with NLU Prototyping

If you edit the overall dialogue model or Model Manager (addition / change / deletion of sample utterances, addition / change / deletion of slots/entities, etc.), you will need to Train Assistant again.

For optimal performance, you should have at least 5-10 Utterances for each Intent. For entities, at least 10 values & variations for your Entity types, synonyms, and mentions in other utterances.

Other docs in this section