[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Semantic Primes
coi rodo
I think what Brandon is trying to do, and me too, with the Jorne Project
(http://jorne.org -- which I hope he will join), is to be able to
convert Semantic Web constructs to and from Lojban. I agree that the
Lojban gismu are not exactly semantic primes.
Some ontologies for semantic primes can be found in:
English Wordnet at http://wordnet.princeton.edu/
SUMO/MILO at http://suo.ieee.org/
OpenCyc at http://opencyc.org/
and others
But because gismu can be combined logically (and the language is
syntactically non-ambiguous) it is an appropriate prose language for
creating and describing semantic content. We would need to build
standard semantic hierarchies in order to do that.
For example they can be Lojban<->Lojban, Lojban<->Wordnet or Lojban<->SUMO.
I think it would be very powerful to have the gismu formally related to
concepts from each of the above (in a hierarchy/ontology such as OWL).
This forms a natural dictionary (think "semantic gloss") that is
accessible to machine reasoning and can also form the basis for
automated discovery of Lojban glosses in other languages. I started
looking at some of the gismu and trying to map them to SUMO concepts but
have not had much time to spend on it lately. English has a Wordnet
database, why shouldn't Lojban? ;-) And we can do it even better because
of the logical aspects of Lojban.
Imagine the following scenario:
-------
Last year, in 2015, the Semantic Web arrived in a big way. Half of the
world's web sites have converted their pages into webs of concepts
instead of text documents linked by "keywords" (which are what HTML
hyperlinks are). Software conversion agents are becoming autonomous, and
can now automatically discover and reason about knowledge gathered from
sites on the web. This can happen because the semantics of the content
is encoded in RDF tags. But a lot of the ancient 2005-era legacy data in
natural language format still needs to be properly converted. It's a
massive undertaking. Lojban is playing an important part in this.
By 2012, all the major natural languages had a wordnet and grammarnet.
But Lojban had the most machine-compatible of any of these, because it
didn't rely on a neural network or fuzzy logic to process the grammar.
So the World Semantic Council voted to make Lojban as the official
intermediate prose description language for semantic webs.
Once any natural language document is converted into a semantic (RDF)
description, software agents need to check the correctness of the
mapping. These quality testing agents convert the RDF structures into
Lojban. Lojban here is working like a hybrid of natural language and
data structure. The agents compare the Lojban against the original text
and its English, Chinese, and Hindi translations from RDF. Currently, a
human also checks the generated text for semantic correctness (according
to the StandardHuman2011 semantic standard). This is taking too much
time and money. There are rumors that the human QA checks will be
reduced by 50% next year.
Note: This document is a translation to English from Quechua and German
versions (via RDF and Lojban) and its semantic consistency has not yet
been verified.
-------
This stuff will become real, and it's already happening in some ways.
Witness the amazing growth of RSS site feeds lately. And XML is used
everywhere. Take a look at the W3C's vision for the semantic web and its
clear that Lojban would fit into the picture very well.
http://www.w3.org/2001/sw/
la iuban
blog at http://brian.mxdj.com
John E Clifford wrote:
This sounds like an interesting project and I am
glad you think that Lojban may help. But you
should note early on that Lojban gismu are not
meant to be semantic primes but rather higher
nodes in a web that will be more useful for
forming compounds (primes make for very long
compounds -- check out AUI). Notice, by way of
making the point, the current discussion about
{pilka} and {skapi}, which probably share some
primes and yet are clearly different and are both
gismu. Taking gismu as fixed points, your web
will have to go down as well as up to fit other
concepts in, if that is the overall scheme you
are following (I don't keep up, so my ideas are
very 1970's -- roughly old stone age, I suspect).
-