Re: peer-to-peer was Re: Distributed querying on the semantic web

 > Anyway, my apologies if this seems off-topic,

It seemed highly on-topic to me (and so I will drift off-topic ... I 
think I get back on topic with the in-line comments)

We can have a vision of tightly controlled meanings with authorities 
(based on DNS) having final say; or we can imagine every user having 
their view and the final authoritative definition being the social 
process with all the messiness that that involves.

Personally I just don't think the former works - it tries to freeze the 
ever-changing world. Maybe it's just that I prefer a messy, incoherent 
world to any one individual or group's clear vision of an ideal.

The trick, that Peter appears to see as achievable using owl:imports and 
related mechanisms, is, within this messiness, to allow clear enough 

We manage with natural language and other semiotically charged actions 
(such as a bunch of flowers). Highly ambiguous, but used in the right 
way, at the right time, ...

in-line comments follow

Jonathan Chetwynd wrote:

> oops,
> well jumping in as a novice :-)
> Our users prefer graphics symbols icons illustrations or photos with 
> their text.
> Many have learning difficulties, low literacy or other issues with text.
these symbols will have meaning for your users, which they (the 
'reader') will control. Any attempt to try and legislate an 
authoritative meaning in this sort of context is doomed to failure.

> Some bright sparks would like a centralised server using rdf that has a 
> concept and image database.
> then we can have local use for personal images, but more general images 
> can be 'translated' from one language to another, whether the languages 
> are spoken, text, symbolic, or sign...
I think whether to have a centralised server or a peer-to-peer system 
can be simply a question of distributed system architecture; but what 
seems on-topic is whether we have a *semantic architecture* that is 
centralized or peer-to-peer. My understanding of this thread is that 
even a semantic arcitecture which vests a distributed authority in 
domain name owners to make the meaning of terms in their namespaces 
suffers some weaknesses of a centralized architecture for meaning.

Thus we can identify three levels of centralization in a semantic 

1: fully centralized
   terms and their meaning are defined by a single world wide authority.
See many standard internet vocabularies for doing X or Y

2: terms decentralized, meanings per term centralized
   anyone can make up a term, but for each term there is a (potentially 
different) central authority that defines its meanings

3: fully decentralized
   anyone can make up a term, anyone can make up a meaning

We know that (1) can work for some applications.
We know that (3) works (our life as a society, both natural language and 
other modalities of meaning use this method)

There are many who argue that (2) is the right way for the SemWeb.
[Maybe I am going to argue myself into (2), I prefer the beauty of (3), 
but the engineer in me sees the attraction of (2)]

Note that the choice of semantic architecture is orthogonal to the 
choice of distributed system architecture - it is possible to implement 
SemArch (3) on a mainframe serving the planet; it is possible to 
implement SemArch (1) on thousands or millions of peers circling the globe.

> However I'd prefer a peer to peer system (BitTorrent?), because in this 
> way the users can have greater control over the meaning, interpretation 
> and relation of concepts.
> It also may mean that language can develop faster and more naturally...
> This does mean, much better authoring tools.
> Anyway, my apologies if this seems off-topic, but if anyone knows of a 
> simple explanation that refers to something similar, together with a 
> working example, that would be sweet.
> thanks
> Jonathan Chetwynd
> "It's easy to use"


Received on Thursday, 22 April 2004 15:00:56 UTC