A Follow Up to Building a Home Automation Alexa Skill with Pat

Michael Vincent Ymbong
Pat Inc
Published in
3 min readOct 12, 2017

Following up on my last article Building a Home Automation Alexa Skill with Pat , I’d like to answer some of the questions we’ve received about the meaning API.

“Right now we’ve developed a chatbot using a different nlp processing tool. It’s pretty good, but can’t deal with multiple intents in the same message. How does Pat deal with complicated sentences/language?”

Pat is a meaning API and because of its underlying linguistic framework, it breaks down human language for each word in the sentence with how it relates to the other words — context.

For example with a sentence like “The big bad wolf bit the fox that was sleeping at my backyard yesterday because it was hungry”, Pat will break that down into different elements.

Sentence breakdown

Some of these elements can even be broken down further. For example the:

  • Clause Periphery (why)
Breakdown for Clause Periphery “it was hungry”
  • Undergoer
Undergoer: “the fox that was sleeping”
  • Actor
Actor: “The big bad wolf”

*There is no existing capability to break down language by meaning at the moment and more complex sentences but the current NLP APIs are good for commands.

“The article mentions no training. Does this mean that Pat would just be used as-is with it’s general training? Or could it be trained to understand industry-specific words? Or, since it’s more about understanding the meaning of a sentence, this is unnecessary?”

Yes that’s right — it’s used ‘as is’ because we are training Pat on human language and the output is those semantic relationships shown above and in the previous article. Regarding adding industry specific words, yes again in the future there will be an LLI (language loading interface) for developers, that will handle new words/concepts. This will enable developers to load their application specific entities (Pat calls them referents), like movie titles, songs, actors.

“How is the same API suitable for lots of different use cases or domains?”

As Pat is being loaded with vocabulary and patterns, its knowledge increases. Pat can’t jump to understanding Wall Street Journal till it expands in its knowledge. So Pat’s early applications suit basic conversation.

When is the service available?”

Pat Inc is in the process of getting funding to load the system with vocabulary and patterns/concepts for language learning and home automation applications as per the above roadmap. Pat is working with early access developers for these applications to refine the API. So we don’t have a date yet for the public beta, but you can waitlist by going to the website.

If you want to waitlist for the Pat meaning API, visit www.pat.ai.

--

--