jeudi 12 juillet 2012

Trying FRED, a tool for producing RDF/OWL ontologies from natural language sentences

FRED - - A tool for automatically producing RDF/OWL ontologies and linked data from natural language sentences, currently limited to English.
FRED is based on C&C [1] and Boxer [2], a NLP tool that transforms natural language text into a logical form compliant to Discourse Representation Theory. We process Boxer output and apply a set of heuristics and semantic transformations in order to obtain RDF designed for the Semantic Web. In this process, we emphasize the relation to linguistic frames, supporting FrameNet and VerbNet vocabularies, and to ontology design patterns.

So I tried first some examples from ATTEMPTO ACE 6.6 in a Nutshell :
A customer inserts a card and the machine checks the code.
Not bad ! FRED uses this class and predicates :
respectively for the sentence, the grammatical subject, and the direct complement.
NOTE: The Options: "FrameNet roles NER Tìpalo WSD" do not have tooltips, so it's not easy to know what they mean.

A customer inserts a VisaCard or inserts a MasterCard, and inserts a code.
The disjunction (or) is not reflected at all in the output. Neither is the negation correctly represented:
It is false that a customer inserts a card.

Now I'll try an example of mine that works with EulerGUI:
If a user U has-as-friend a person X, and has-as-friend a person Y, and X is not Y,
and X has-as-friend Z, and Y has-as-friend Z, and Z is not U
then Z is-a-proposed-friend-for U.
It's clear that the if-then structure is not understood. Even a simple sentence like this is wrong :
If the code is valid then the ATM accepts the card.
FRED tries to be smart, and "thinks" that the ATM is a foaf:Organisation !

1 commentaire:

  1. Hi Jean-Marc I see this after a while ;) thanks for the feedback, but please notice that
    FRED has improved substantially in the meantime. Please check your three examples: