Saturday, January 13, 2018

My little NLI-GO DBPedia demo

With some pride (but not very much ;) I present to you my natural language interaction demo:

http://patrickvanbergen.com/dbpedia/app/

 This is the first presentable result of many years work of trying to create a library that allows a user (you) to interact with a database in natural language.

The language in this example is DBPedia, a database that is based on Wikipedia. It is a triple store that contains information about the people in Wikipedia, among other things.

My library tokenizes and parses the sentence and it converts the parse tree into a relational representation. A solution is looked up and executed. This involves a series of simple database requests (in this case Sparql queries). The results are integrated to more relations. From these relations a response in natural language is generated.

The point of my library is not to create a little interface to DBPedia. I am looking to make the interface to databases in general as simple and powerful as possible. This means that more databases, types of databases, domains, natural languages, etc, should be addable with few configuration files. No coding should be necessary eventually. It's a long way to go.

Nevertheless, I invite you to try these sentences with the demo, and check out the productions that will be shown. These are some of the steps the library takes to find the answer.

  • How many children had Lord Byron?
  • How many children had Michael Jackson?
  • Who married Elvis Presley?
  • Who married Donald Trump?
I hope you find this interesting.

On SQLAlchemy

I've been using SQLAlchemy and reading about it for a few months now, and I don't get it. I don't mean I don't get SQLAlchem...