Michael, Welcome to the dev list and NLPCraft! Short answer: ------------------ These two sentences get parsed into the similar set of tokens (based on the model [1]) which match the same intent - hence the same action for both sentences.
NOTE: if you want to play with NLPCraft make sure to run from 'master'. Project is actively being developed and official releases lag behind. Longer answer: -------------------- If you run LightSwitch example you can look at the data probe log out and see very detailed output for all different parsing variants. You will notice that NLPCraft automatically filters out stop words, detects user-defined named entities from [1] (often via multiple synonyms), and find the best matching intent. One of the key aspects of NLPCraft is the fact that it does not require any classic ML learning (corpus development, prep & training) - it only requires a simple, deterministic model [1] and thus providing the deterministic answers. You can also see at the start of the data probe that it reports over 13K unique synonyms - all auto-derived from the same model [1], so it provides very deep "comprehension" for the this subject domain (light switch operation). Hope it helps, 1. https://github.com/apache/incubator-nlpcraft/blob/master/nlpcraft-examples/lightswitch/src/main/resources/lightswitch_model.yaml On Fri, May 14, 2021 at 1:11 AM Michael Wechner < [email protected]> wrote: > Hi Together > > I have just noticed the NLPCraft Project and it sounds very interesting! > > I am an ASF member since 2004, whereas I have become very interested in > NLP/NLU recently and would be happy to contribute in the future if > possible. > > To start with, I would be curious to understand how NLPCraft is detecting > similar sentences, like for example > > "Turn the lights off in the entire house" > and > "Turn off all lights now" > > ? > > All the best > > Michael >
