1

I am interested in exploring how chatbots work in understanding what the user is trying to say in the sentence they typed.

I have a very primitive idea that they split the sentence into a bunch of words that act as anchors in order to figure out what's purpose of the sentence and what is the user trying to say in it.

They have a set of words already in its database/vocabulary and it grows overtime.

What I am interested in is,

  1. What's this sentence synthesis method called? Is there some sort of name for this terminology?
  2. How can we grow the vocabulary of the chatbot?
  3. Are there any SDKs for sentence synthesis? SDK with online APIs are much preferred as they can provide updated vocabulary which grows overtime in their system
Umer Farooq
  • 7,356
  • 7
  • 42
  • 67

1 Answers1

1

An early idea for computergenerated speech is Chomsky hierarchy. The sentence is subdivided in subject verb object and a overall grammar generates new sentences. This was used in 1970s AI production systems like SOAR. A more recent way of producing artificial speech is discussed under the umbrella term "storytelling" Managing interaction between users and agents in a multiagent storytelling environment The idea in short: a domain is descriped as a textadventure. The player has different options like "take object" or "leave house". And the AI is programmed to solve such games. In contrast to chomsky grammars, there is meaning behind the words in form of the gamescore which can be reached. Understanding a story/sentence is equal to collect points in the game. The output of such systems is grounded on the domain. In the above paper is on page 5 the example "bank robbery" explained as an ontology.

Manuel Rodriguez
  • 734
  • 4
  • 18