identifier/identifier generation

[Melanie Courtot]

Hi,

Here is an attempt at defining the identifier generation process and
the resulting identifier based on this morning's meeting. I think it
ties in the historical/systematic approach used to generate the
identifier (we generate identifiers as part of a series) and the
purpose of the identifier which is denoting something and picking out
that something from a set.

Ideally, I would like to have the process achieving a specific
objective as discussed this morning, however I am not sure that saying
the identifier generation process achieves planned objective
generating identifiers really adds anything (but an extra level of
complexity :) )

Comments and suggestions are welcome.

Melanie



*** identifier generation

Definition: the planned process of generating an information content
entity used as identifier according to a system or following a
specific method.

Editor note:
1. the process of generating identifiers uses a systematic approach
2. there is an intent to generate multiple identifiers: even if when
we start the series and have only one identifier in the set, we expect
there will be additional elements in the series to be identified and
we will generate additional identifiers following the same systematic
approach

Examples of usage:
http://tools.ietf.org/html/rfc4122
http://docs.jboss.org/hibernate/core/3.3/reference/en/html/mapping.html#mapping-declaration-id-enhanced



*** identifier

Definition:
An information content entity that can be used repeatedly to denote
the same thing in different contexts and with the intent of picking
out that thing out of a larger set.

Examples of usage:
Lot number, SSN, Pubmed ID
http://en.wikipedia.org/wiki/Identifier
http://en.wikipedia.org/wiki/Universally_Unique_Identifier

Restriction: is_specified_output of identifier generation


---
Mélanie Courtot
TFL- BCCRC
675 West 10th Avenue
Vancouver, BC
V5Z 1L3, Canada

[Barry Smith]

I think that you are safe if to identify the
objective of an identifier generating process as
being that of generating objectives. Those
processes which are performed to yield outputs of
a certain sort are processes whose objective is
yielding outputs of that sort. Definitionally.
(In this they are distinct from processes such
as: going for a walk; doodling; bodyguarding; eating ...)
BS

>--
>informatio...@googlegroups.com
>To change settings, visit
>http://groups.google.com/group/information-ontology

[Michel Dumontier]

Hello,

  

> *** identifier

> 

> Definition:

> An information content entity that can be used repeatedly to denote

Why "used repeatedly"? once is sufficient. i really don't see this as a necessary condition

> the same thing in different contexts 

Why "different contexts"? are there identifiers that denote the same thing in one context, but not another?

> and with the intent of picking out that thing 

This seems redundant with the use of "denote" in the first part -> the relation denotes is defined as:

"denotes is a primitive, instance-level, relation obtaining between an information content entity and some portion of reality. Denotation is what happens when someone creates an information content entity E in order to specifically refer to something. The only relation between E and the thing is that E can be used to 'pick out' the thing."

So adding this bit just repeats the predicate definition.

> out of a larger set.

I object to the "larger" set. I would argue for "collection", as even one item can be part of a single-item collection.

So, i suggest in turn:

An information content entity that denotes an entity in some collection.

-=Michel=-

--
informatio...@googlegroups.com
To change settings, visit
http://groups.google.com/group/information-ontology

--
Michel Dumontier
Associate Professor of Bioinformatics
Carleton University
http://dumontierlab.com

[Melanie Courtot]

On 17-Nov-09, at 12:27 PM, Michel_Dumontier wrote:

> Hello,

>
>
>> *** identifier
>>
>> Definition:
>> An information content entity that can be used repeatedly to denote

> Why the "used repeatedly" - once is sufficient. i really don't see

> this as a necessary condition

Once is sufficient, but I wanted to reflect the fact that we usually
assign identifiers with the intent of using them several times. For
example a lot number is usually used multiple times, by different
persons. In my opinion there is some kind of idea of perdurance/
repeatability associated with an identifier. Saying it can be used
repeatedly is not wrong, even if we use it only once.

>
>> the same thing in different contexts

> Why "different contexts"? if it refers to the same thing, then this
> statement is irrelevant -> are there identifiers that denote the

> same thing in one context, but not another?

This is to address Jonathan's concern about the identifier being
contextual or not (http://code.google.com/p/information-artifact-ontology/issues/detail?id=65
).
I wanted to specify that the notion of identifier is attached to the
thing, not to the context.

>
>> and with the intent of picking out that thing

> This seems redundant with the use of "denote" in the first part ->
> the relation denotes is defined as:
>
> "denotes is a primitive, instance-level, relation obtaining between
> an information content entity and some portion of reality.
> Denotation is what happens when someone creates an information
> content entity E in order to specifically refer to something. The
> only relation between E and the thing is that E can be used to 'pick
> out' the thing."
>
> So adding this bit just repeats the predicate definition.

I agree - as mentioned I would like to associate an objective with the
process, I believe that what makes an identifier is the intent of
using it as an identifier, but as said, I haven't found a satisfactory
way to express this.

>
>
>> out of a larger set.

> I object to the "larger" set. I would argue to a "collection", as

> even one item can be part of a single-item collection.

I am happy to update to "out of a set" or "out of a collection".
Again, here I wanted to reflect the intent of picking out from a set:
we traditionally create a series of identifier to denote similar
elements in a set, and assign to each of them their own identifier to
be able to pick them out of that set.

>
> So, i suggest in turn:
>

> An information content entity that denotes an entity from some
> collection.

This seems a reasonable alternative - I am just a bit cautious about
relying on the "denotes" definition so completely, knowing it has just
been finalized. I would feel more comfortable updating the denotes
definition replacing "in order to" with "with the intent of" for
example. I also would prefer keeping extra bits of the definition in
editor notes for example. Ideally, I would really like to formalize
the objective "picking something out of a set" - I think that would
really help and would make it more obvious.

What about keeping your definition to start with, adding the intent/
context in notes?

If no objection I'll update the file tomorrow.

Melanie

>
>
> -=Michel=-

[Alan Ruttenberg]

On Tue, Nov 17, 2009 at 5:42 PM, Pat Hayes <pha...@ihmc.us> wrote:

"denotes is a primitive, instance-level, relation obtaining between an information content entity and some portion of reality. Denotation is what happens when someone creates an information content entity E in order to specifically refer to something. The only relation between E and the thing is that E can be used to 'pick out' the thing."

Ouch. I would take strong exception to this use of 'pick out' as part of the definition of denotation. Denoting names often cannot be used to "pick out" anything.  

That's a fair criticism. I suppose we could just drop this last sentence. The attempt was to emphasize that 'denotes' doesn't mean very much - to trim the expectations. 

 

They simply refer. I can use the string 'xy23' to denote my oldest pet cat, but it certainly cannot be used to pick her out, in any useful sense of 'pick'. In fact, some sense of 'picking out' seems to be what distinguishes the idea of an identifier from a mere referring name. Identifiers, unlike merely denoting names, must actually identify (in some sense). 

By "must actually identify" I guess you mean something like "can be used in some predefined process to discriminate one thing from another"? If not, could you unpack what you mean? Identifiers, by themselves, do nothing.

 

> out of a larger set.

I object to the "larger" set. I would argue for "collection", as even one item can be part of a single-item collection.

So, i suggest in turn:

An information content entity that denotes an entity in some collection.

Why does this set or collection need to even be mentioned? Many ontologies explicitly reject the idea of sets, yet they still support the ideas of denotation and reference. I think this whole idea of there being some containing set is a confusing and irrelevant distraction. If one wants to convey the identity-parade idea implicit in 'identify', then say that the identifier distinguishes the referent from other entities. No need to require them to all be in a set. 

I like that.

So  

Identifier: "An information content entity that can be used repeatedly to denote the same thing in different contexts and to distinguish the referent from other entities."

(but I have to still read and absorb Barry's comment)

-Alan

[Michel Dumontier]

On Tue, Nov 17, 2009 at 10:53 PM, Alan Ruttenberg <alanrut...@gmail.com> wrote:

On Tue, Nov 17, 2009 at 5:42 PM, Pat Hayes <pha...@ihmc.us> wrote:

"denotes is a primitive, instance-level, relation obtaining between an information content entity and some portion of reality. Denotation is what happens when someone creates an information content entity E in order to specifically refer to something. The only relation between E and the thing is that E can be used to 'pick out' the thing."

Ouch. I would take strong exception to this use of 'pick out' as part of the definition of denotation. Denoting names often cannot be used to "pick out" anything.  

That's a fair criticism. I suppose we could just drop this last sentence. The attempt was to emphasize that 'denotes' doesn't mean very much - to trim the expectations. 

 

They simply refer. I can use the string 'xy23' to denote my oldest pet cat, but it certainly cannot be used to pick her out, in any useful sense of 'pick'. In fact, some sense of 'picking out' seems to be what distinguishes the idea of an identifier from a mere referring name. Identifiers, unlike merely denoting names, must actually identify (in some sense). 

By "must actually identify" I guess you mean something like "can be used in some predefined process to discriminate one thing from another"? If not, could you unpack what you mean? Identifiers, by themselves, do nothing.

 

> out of a larger set.

I object to the "larger" set. I would argue for "collection", as even one item can be part of a single-item collection.

So, i suggest in turn:

An information content entity that denotes an entity in some collection.

Why does this set or collection need to even be mentioned? Many ontologies explicitly reject the idea of sets, yet they still support the ideas of denotation and reference. I think this whole idea of there being some containing set is a confusing and irrelevant distraction. If one wants to convey the identity-parade idea implicit in 'identify', then say that the identifier distinguishes the referent from other entities. No need to require them to all be in a set. 

I like that.

so do i.

 

So  

Identifier: "An information content entity that can be used repeatedly to denote the same thing in different contexts and to distinguish the referent from other entities."

again, i find "that can be used repeatedly to denote the same thing in different contexts" to be verbose and non-essential. Just because something can be said, doesn't mean it should, particularly with definitions.  Melanie suggested that it appears in an Editors note, and I agree.

identifier: An information content entity that distinguishes the referent from other entities.

simple, yet effective.

-=Michel=-

 

(but I have to still read and absorb Barry's comment)

-Alan

[Alan Ruttenberg]

On Tue, Nov 17, 2009 at 3:28 PM, Barry Smith <phis...@buffalo.edu> wrote:

I think that you are safe if to identify the
objective of an identifier generating process as
being that of generating objectives.

s/objectives/identifiers/ 

 

Those
processes which are performed to yield outputs of
a certain sort are processes whose objective is
yielding outputs of that sort. Definitionally.
(In this they are distinct from processes such
as: going for a walk; doodling; bodyguarding; eating ...)
BS

OK. I read this as it's ok to say something like:

identifier generation objective is_a objective

identifier generation achieves_planned_objective identifier generation objective

2 things:

1. We have objective specification in IAO versus objective. The distinction is between objectives (the realizable entity of some sort) and information artifact - some concretizable e.g. document articulating the objective. So the question: Are all identifiers generated according to an articulated spec?

2. I think what Melanie was getting at by mentioning the objective was more along the lines of: Objective to create something that can subsequently be used in a certain kinds of processes, those processes being the sorts that do things like

a) distinguish one entity from any other of the entities that have previously been assigned an identifier by the same method.

b) establish that a certain entity is the same one about which some information was recorded previously, in order to associate some historically collected information to the entity, or add further information about it

...

i.e. talk more about the processes in which identifiers are participants. Right now the realizable aspect of the identifier is strictly in the english "can be used...".

-Alan

[Melanie Courtot]

I think the repeatedly is important - as I mentioned I think there is
a notion of "stability" of an identifier: you generate an identifier
with the intent of using it several times. I don't think it is
controversial to keep it here, as we say "can be used".
I was unsure about the context part and suggested to keep it in an
editor note, but I think Pat has a valid example and based on that I
would rather keep it.

>
> identifier: An information content entity that distinguishes the
> referent from other entities.

IMO, to link to denote based on this would require to be very familiar
with the ontology, and I would rather be a bit more verbose than too
succinct - especially in our current development phase: I am a bit
worried that we end up loosing information by trying to be too concise.

Melanie

[Alan Ruttenberg]

On Tue, Nov 17, 2009 at 11:02 PM, Michel Dumontier <michel.d...@gmail.com> wrote:

Identifier: "An information content entity that can be used repeatedly to denote the same thing in different contexts and to distinguish the referent from other entities."

again, i find "that can be used repeatedly to denote the same thing in different contexts" to be verbose and non-essential. Just because something can be said, doesn't mean it should, particularly with definitions.  Melanie suggested that it appears in an Editors note, and I agree.

identifier: An information content entity that distinguishes the referent from other entities.

simple, yet effective.

I don't think it's effective. It suggests that the identifier does something on its own, rather than being something that is used to do something. However I think it would be better if made even more explicit, by pointing to the processes in which it is used, as by creating objective and realizing process types, as discussed in my note responding to Barry.

-Alan

 

[Michel Dumontier]

On Tue, Nov 17, 2009 at 11:37 PM, Melanie Courtot <mcou...@gmail.com> wrote:

On 17-Nov-09, at 8:02 PM, Michel Dumontier wrote:

On Tue, Nov 17, 2009 at 10:53 PM, Alan Ruttenberg <alanrut...@gmail.com> wrote:


On Tue, Nov 17, 2009 at 5:42 PM, Pat Hayes <pha...@ihmc.us> wrote:

"denotes is a primitive, instance-level, relation obtaining between an information content entity and some portion of reality. Denotation is what happens when someone creates an information content entity E in order to specifically refer to something. The only relation between E and the thing is that E can be used to 'pick out' the thing."

Ouch. I would take strong exception to this use of 'pick out' as part of the definition of denotation. Denoting names often cannot be used to "pick out" anything.

That's a fair criticism. I suppose we could just drop this last sentence. The attempt was to emphasize that 'denotes' doesn't mean very much - to trim the expectations.

They simply refer. I can use the string 'xy23' to denote my oldest pet cat, but it certainly cannot be used to pick her out, in any useful sense of 'pick'. In fact, some sense of 'picking out' seems to be what distinguishes the idea of an identifier from a mere referring name. Identifiers, unlike merely denoting names, must actually identify (in some sense).

By "must actually identify" I guess you mean something like "can be used in some predefined process to discriminate one thing from another"? If not, could you unpack what you mean? Identifiers, by themselves, do nothing.

> out of a larger set.
I object to the "larger" set. I would argue for "collection", as even one item can be part of a single-item collection.


So, i suggest in turn:

An information content entity that denotes an entity in some collection.

Why does this set or collection need to even be mentioned? Many ontologies explicitly reject the idea of sets, yet they still support the ideas of denotation and reference. I think this whole idea of there being some containing set is a confusing and irrelevant distraction. If one wants to convey the identity-parade idea implicit in 'identify', then say that the identifier distinguishes the referent from other entities. No need to require them to all be in a set.


I like that.


so do i.

So

Identifier: "An information content entity that can be used repeatedly to denote the same thing in different contexts and to distinguish the referent from other entities."


again, i find "that can be used repeatedly to denote the same thing in different contexts" to be verbose and non-essential. Just because something can be said, doesn't mean it should, particularly with definitions.  Melanie suggested that it appears in an Editors note, and I agree.

I think the repeatedly is important - as I mentioned I think there is a notion of "stability" of an identifier: you generate an identifier with the intent of using it several times.

i disagree. you generate an identifier to identify. the fact that you might use it more than once is not necessary to this definition, it is consequential.

 

I don't think it is controversial to keep it here, as we say "can be used".
I was unsure about the context part and suggested to keep it in an editor note, but I think Pat has a valid example and based on that I would rather keep it.

an identifier, by virtue that it identifies something, would be valid across contexts.  in the example provided, "carbon" is still an identifying information content entity, even when applied to isotopes (it distinguishes them from other elements). 

identifier: An information content entity that distinguishes the referent from other entities.

IMO, to link to denote based on this would require to be very familiar with the ontology,

mmm? the word "denotes" is not used. can you elaborate?

 

and I would rather be a bit more verbose than too succinct - especially in our current development phase: I am a bit worried that we end up loosing information by trying to be too concise.

guidelines for usage are useful additions to ontologies.

 

-=Michel=-

Melanie

simple, yet effective.

-=Michel=-


(but I have to still read and absorb Barry's comment)

-Alan



--
Michel Dumontier
Associate Professor of Bioinformatics
Carleton University
http://dumontierlab.com

[Michel Dumontier]

Indeed.  I'm keen to see the development of the other related entities involving the process of identification according to some specification.

-=Michel=-

[Barry Smith]

The challenge for Alan is to give an example of an ICE that
distinguishes but cannot be used repeatedly to denote the same thing.
An example (far-fetched?) might be a batch of machines each one of
which is programmed to generate a unique random number that is
visible for one second every day (the numbers would be different from
day to day). The number would serve to distinguish, but could not be
(usefully) used repeatedly.
I think we may need a third clause: An ICE ... and that is part of a
system of ICEs doing the same job...
BS

At 12:14 AM 11/18/2009, Michel Dumontier wrote:

><http://dumontierlab.com>http://dumontierlab.com

>
>--
>informatio...@googlegroups.com
>To change settings, visit

><http://groups.google.com/group/information-ontology>http://groups.google.com/group/information-ontology
>

[Pat Hayes]

On Nov 18, 2009, at 12:02 PM, Barry Smith wrote:

> The challenge for Alan is to give an example of an ICE that
> distinguishes but cannot be used repeatedly to denote the same
> thing. An example (far-fetched?) might be a batch of machines each
> one of which is programmed to generate a unique random number that
> is visible for one second every day (the numbers would be different
> from day to day). The number would serve to distinguish, but could
> not be (usefully) used repeatedly.

Hmm, I can't see any useful sense in which they can be said to
distinguish, either. Certainly having the number wouldn't enable you
to pick out one of the machines from the others. But in the same
spirit ...

> I think we may need a third clause: An ICE ... and that is part of a
> system of ICEs doing the same job...

... how about this example. There are huge mail-sorting systems now in
use which consist of fast conveyors running east-west with other
conveyors just above them running north/south, and gates which open
briefly to allow one package to drop from one of the latter to a
selected one of the former (which routes it to some further selection
process, or to a bin to go on a particular airplane.) Barcode readers
read the address on a package entering the system and send a message
to the appropriate gate to open at the time the package will arrive at
it. Arguably, these messages constitute 'identifiers' of the packages
on the belt while they are in transit from the reader to the gate
which will appropriately divert them. And if so, these identifiers are
unique, are used only once and cannot be re-used, and do in fact
identify in the sense of 'picking out' one from others, which is after
all the entire point of all this expensive machinery.

Pat

------------------------------------------------------------
IHMC (850)434 8903 or (650)494 3973
40 South Alcaniz St. (850)202 4416 office
Pensacola (850)202 4440 fax
FL 32502 (850)291 0667 mobile
phayesAT-SIGNihmc.us http://www.ihmc.us/users/phayes

[Barry Smith]

At 06:06 PM 11/18/2009, Pat Hayes wrote:

>On Nov 18, 2009, at 12:02 PM, Barry Smith wrote:
>
>>The challenge for Alan is to give an example of an ICE that
>>distinguishes but cannot be used repeatedly to denote the same
>>thing. An example (far-fetched?) might be a batch of machines each
>>one of which is programmed to generate a unique random number that
>>is visible for one second every day (the numbers would be different
>>from day to day). The number would serve to distinguish, but could
>>not be (usefully) used repeatedly.
>
>Hmm, I can't see any useful sense in which they can be said to
>distinguish, either. Certainly having the number wouldn't enable you
>to pick out one of the machines from the others. But in the same
>spirit ...

well, they would serve for a second
perhaps some would be quick enough to take a photograph
but the machines below serve to make the point

[Bjoern Peters]

I don't care particularly about the repeated/single use identifiers, but I think it is important to have the 'intent' clearly stated in the definition of identifier as we want to use it. My crack at it:

identifier: An ICE created to distinguish the referent from other entities in a 'refer by identifier process'.

'refer by identifier process' A planned process in which an ICE is generated which has as a part an identifier which indicates that the ICE is about the entity denoted by the identifier.

I think Pat's example is good, and illustrates what we should not consider an identifier. In that example, the message would be some computer encoding for 'open in 123 seconds'. None of that is an identifier. If in contrast there is a second barcode scanner before the gate, the message could have been 'open when package XYZ arrives', XYZ being the barcode, then yes, that message contains an identifier. 

In Barry's example, I am missing a description on how someone purposely programmed the display of those codes in order to identify those machines.

----- Original Message -----
From: "Pat Hayes" <pha...@ihmc.us>
To: "Barry Smith" <phis...@buffalo.edu>
Cc: "Michel Dumontier" <michel.d...@gmail.com>, "Alan Ruttenberg" <alanrut...@gmail.com>, "Melanie Courtot" <mcou...@gmail.com>, "Jonathan Rees" <j...@creativecommons.org>, "information-ontology Discuss" <informatio...@googlegroups.com>
Sent: Wednesday, November 18, 2009 3:06:57 PM GMT -08:00 US/Canada Pacific
Subject: Re: identifier/identifier generation


On Nov 18, 2009, at 12:02 PM, Barry Smith wrote:

> The challenge for Alan is to give an example of an ICE that  
> distinguishes but cannot be used repeatedly to denote the same  
> thing. An example (far-fetched?) might be a batch of machines each  
> one of which is programmed to generate a unique random number that  
> is visible for one second every day (the numbers would be different  
> from day to day). The number would serve to distinguish, but could  
> not be (usefully) used repeatedly.

Hmm, I can't see any useful sense in which they can be said to  
distinguish, either. Certainly having the number wouldn't enable you  
to pick out one of the machines from the others. But in the same  
spirit ...

--
Bjoern Peters
Assistant Member
La Jolla Institute for Allergy and Immunology
9420 Athena Circle
La Jolla, CA 92037, USA
Tel: 858/752-6914
Fax: 858/752-6987
http://www.liai.org/pages/faculty-peters

[Barry Smith]

At 07:36 PM 11/18/2009, Bjoern Peters wrote:
>I don't care particularly about the repeated/single use identifiers,
>but I think it is important to have the 'intent' clearly stated in
>the definition of identifier as we want to use it. My crack at it:
>
>identifier: An ICE created to distinguish the referent from other
>entities in a 'refer by identifier process'.
>
>'refer by identifier process' A planned process in which an ICE is
>generated which has as a part an identifier which indicates that the
>ICE is about the entity denoted by the identifier.

Won't do. I'm afraid. It is circular.
I think there are also problems with appeals to intent. Suppose a
clerk intends to produce an identifier system, presses the relevant
button, prints 1000 identifier labels and attaches them to 1000
pieces of equipment, without checking that the labels of have the same number.

[Bjoern Peters]

We don't do unsuccessful planned processes, so the clerk example doesn't apply.

As for circularity, I don't see it, could you help? I am trying to write this similar to the 'role' definition pattern, where 'identifier' is the something bearing the role, and 'refer by identifier' is the process in which it is 'realized'.  There is separate process of 'identifier creation', I hope that was obvious.

- Bjoern

[Adam M. Goldstein]

On Nov 18, 2009, at 8:09 PM, Bjoern Peters wrote:

<snip>

At 07:36 PM 11/18/2009, Bjoern Peters wrote:
>I don't care particularly about the repeated/single use identifiers, 
>but I think it is important to have the 'intent' clearly stated in 
>the definition of identifier as we want to use it. My crack at it:
>
>identifier: An ICE created to distinguish the referent from other 
>entities in a 'refer by identifier process'.
>
>'refer by identifier process' A planned process in which an ICE is 
>generated which has as a part an identifier which indicates that the 
>ICE is about the entity denoted by the identifier.

Won't do. I'm afraid. It is circular.
I think there are also problems with appeals to intent. Suppose a 
clerk intends to produce an identifier system, presses the relevant 
button, prints 1000 identifier labels and attaches them to 1000 
pieces of equipment, without checking that the labels of have the same number.

Barry's right about the problems with intent, but doesn't intent have to be in there somewhere? I take it that none of us have exactly the same DNA, i.e., our personal genomes all differ; so in principle we could use someone's genetic code as an identifier. This would even allow us to "pick out" the person, I think, because we could sample someone's DNA and then see if the sample matched anyone in our big database of genomes. But I would say that someone's DNA code isn't an identifier until someone wants to use it for that, to distinguish each person from others (or from anything else in the universe, I suppose).

I think Barry's example is not totally clear---wouldn't using a machine built for the purpose of creating unique identifiers be a good way for someone intending to create those identifiers to do so? Probably I am being too particular maybe---the example is that the would-be identifiers really do come out the same, so that the person intended to make identifiers, but didn't.

How about: "refer by identifier process" =df "A planned process in which an ICE is generated for the purpose of creating [whatever identifiers turn out to be], using a reliable process for creating [whatever identifiers turn out to be.], so that the resulting ICE's are [whatever identifiers turn out to be]."

I geuss a consequence of doing it this way is that "identifier" has to be defined without reference to the process by which identifiers are created, so that we can talk about the process as ending up with identifiers but not make explicit reference to identifiers in the definition of "refer by identifier process".

------------------
Adam M. Goldstein PhD, MSLIS
--

z_calif...@shiftingbalance.org
http://www.shiftingbalance.org
--

http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621

--
(914) 637-2717 (msg)
--
Dept of Philosophy
Iona College
715 North Avenue
New Rochelle NY 10801

[Adam M. Goldstein]

On Nov 18, 2009, at 8:09 PM, Bjoern Peters wrote:

<snip>

As for circularity, I don't see it, could you help? I am trying to write this similar to the 'role' definition pattern, where 'identifier' is the something bearing the role, and 'refer by identifier' is the process in which it is 'realized'.  There is separate process of 'identifier creation', I hope that was obvious. 

See below

M 11/18/2009, Bjoern Peters wrote:
>I don't care particularly about the repeated/single use identifiers, 
>but I think it is important to have the 'intent' clearly stated in 
>the definition of identifier as we want to use it. My crack at it:
>
>identifier: An ICE created to distinguish the referent from other 
>entities in a 'refer by identifier process'.
>
>'refer by identifier process' A planned process in which an ICE is 
>generated which has as a part an identifier which indicates that the 
>ICE is about the entity denoted by the identifier.

Won't do. I'm afraid. It is circular.

I think Barry is referring to "has as a part an identifier" which is part of the definition of "refer by identifier process." Presumably, if "refer by identifier process" is going to be a part of the definition of "identifier", then "identifier" can't be a part of the definition of "refer by identifier process." "Identifier" also appears in ". . . denoted by the identifier."

[Bjoern Peters]

Circular would be:
identifier = those things created in a identifier generation process,
and
identifier generation process = those processes creating an identifer.

We are routinely using the pattern of defining e.g. 'nursing role' as a role borne by a human realized in a nursing process, and then go on to define the nursing process as one where a human bearing the nursing role provides care to another human. Nothing circular there, rather 'nursing role' identifies one participant in the nursing process, similar to what 'identifier' does in the 'refer by identifier process'.

As for the need for intend: you have it exactly right. Without intend, lots of things will be an identifier, not only genetic code, but also your precise location right now.

- Bjoern
 


 



----- Original Message -----
From: "Adam M. Goldstein" <z_calif...@shiftingbalance.org>
To: "Adam M.Goldstein" <z_calif...@shiftingbalance.org>
Cc: "Bjoern Peters" <bpe...@liai.org>, "Barry Smith" <phis...@buffalo.edu>, "Michel Dumontier" <michel.d...@gmail.com>, "Alan Ruttenberg" <alanrut...@gmail.com>, "Melanie Courtot" <mcou...@gmail.com>, "Jonathan Rees" <j...@creativecommons.org>, "information-ontology Discuss" <informatio...@googlegroups.com>, "Pat Hayes" <pha...@ihmc.us>
Sent: Wednesday, November 18, 2009 5:39:31 PM GMT -08:00 US/Canada Pacific
Subject: Re: identifier/identifier generation

--

[Adam M. Goldstein]

On Nov 18, 2009, at 9:19 PM, Bjoern Peters wrote:

Circular would be: 
identifier = those things created in a identifier generation process, 
and 
identifier generation process = those processes creating an identifer. 

We are routinely using the pattern of defining e.g. 'nursing role' as a role borne by a human realized in a nursing process, and then go on to define the nursing process as one where a human bearing the nursing role provides care to another human. Nothing circular there, rather 'nursing role' identifies one participant in the nursing process, similar to what 'identifier' does in the 'refer by identifier process'.

I see what you mean about the role <-> realize in a process type of definition. I suppose Barry will have to clarify what he meant when he said your previous account of identifiers was circular. Presumably he is familiar with the type of description you are talking about with roles etc.

One thing that does strike me though is that with a nurse, we know roughly what that is. My sense is that, with the discussion of identifiers and related processes, we don't really know what it is, and so we are looking for something more like an analysis, rather than defining "nursing role" in terms of the realization of nursing processes.

Looking back at what you have above, is "identifier" analogous to "nursing role?" It doesn't seem to me that it is. We are not trying to say what an identifier role is, but we are trying to say what an identifier is.

As for the need for intend: you have it exactly right. Without intend, lots of things will be an identifier, not only genetic code, but also your precise location right now.

How nice to get something exactly right . . . :) Anyhow it does seem clear that intentions are required here.

[Pat Hayes]

On Nov 18, 2009, at 6:36 PM, Bjoern Peters wrote:

I don't care particularly about the repeated/single use identifiers, but I think it is important to have the 'intent' clearly stated in the definition of identifier as we want to use it. My crack at it: 

identifier: An ICE created to distinguish the referent from other entities in a 'refer by identifier process'.

'refer by identifier process' A planned process in which an ICE is generated which has as a part an identifier which indicates that the ICE is about the entity denoted by the identifier. 

I think Pat's example is good, and illustrates what we should not consider an identifier. In that example, the message would be some computer encoding for 'open in 123 seconds'. None of that is an identifier. If in contrast there is a second barcode scanner before the gate, the message could have been 'open when package XYZ arrives', XYZ being the barcode, then yes, that message contains an identifier.  

Actually the example was supposed to illustrate the opposite intuition. I see no clear difference between the timing case and the XYZ barcode case. They both are ways of identifying the package, under different  circumstances. Why does something which could be glossed as "the package which will arrive on this belt in 123 seconds" not count as an identifier, or at least an identifying description? It is sufficient to pick out that one package from all the others. Why is that not sufficient to make it count as an identifier? 

Pat

[Jonathan Rees]

My prototypical identifier system is bar codes on railroad cars.
Here's how they can be used:

- Have I seen this car before? (I have a list of the identifiers for
cars I've seen)
- How many different cars have I seen? (Same as number of different
identifiers on my list of car sightings)
- Which car on this train did I load that wheat into? (The one marked
with identifier XYZ, which I recorded when I loaded it; I can
communicate that to someone else and they can ask further questions)
- Whose car is that? (I look up its identifier XYZ in the train's
manifest / roster, which will have an 'owner' column)

Roughly: an identifier is a quality of a thing, a measurement (or
reading) of which acts a proxy for measuring other properties of the
thing that would distinguish that thing from things it might otherwise
be confused with.

E.g. in a set of Russian dolls height could be an identifier, since if
you know the height you know which doll it is. An identifier can only
play the role of identifying if you've had a chance previously to make
an association with it.

Jonathan

[Melanie Courtot]

I think it is an identifying description as you mention but not an
identifier in the sense we would like to use it. While saying "the
package which will arrive on this belt in 123 seconds" will allow you
to get one package, it can not be used repeatedly to allow you to pick
the same package each time.

Melanie

[Melanie Courtot]

Could we then say that our identifier generation process is actually
an identifier system generation process, which has output an
identification system (instances of which are the HLA nomenclature, a
bar codes system), and that this identification system in turn has
part identifiers?

I would avoid saying an identifier is a quality of a thing, as there
are probably identifiers in the system that will never refer to
anything (e.g. some kind of HLA allele that will never exist) I would
still consider them identifiers, they just never will realize their
identifying role/function.

In most cases however we would have a second process of identifier
assignment which has input an identifier and an entity and outputs
both associated, an identified entity (i.e. what you describe as "The
one marked with identifier XYZ, which I recorded when I loaded it")
and in which the identifying realizable entity is realized.

I am not sure I would say that in general height would be an
identifier. It may be applicable in the specific example of Russian
dolls, but there would have had been agreement on using height as an
identification system in that set (which system probably restricts the
set that can be chosen: we want only a "working" Russian doll set in
which no two dolls have the same size, it has to be a restricted set
etc), which brings us back to having an agreed upon system allowing
the identification to happen at a later stage.

Does that fit your use case?

Melanie

[Barry Smith]

At 09:33 AM 11/19/2009, Melanie Courtot wrote:
>Could we then say that our identifier generation process is actually
>an identifier system generation process, which has output an
>identification system (instances of which are the HLA nomenclature, a
>bar codes system), and that this identification system in turn has
>part identifiers?

The process of generating a system of identifiers is one thing, and
standardly will involve generating identifiers as parts.
Any generation of a single identifier will also presuppose the
generation of a system of identifiers.
But the two are different process types.
BS

[Pat Hayes]

OK, thanks. So that seems to nail down 'repeatability of
identification' as a key property of an identifier, right?

Pat

[Pat Hayes]

I am reminded of an observation made by Geoffrey Nunberg, who
overheard two waitresses chatting, and one said to the other, "Check
out the hamburger on table 3", referring of course to the customer who
ordered the hamburger. Apparently, using dishes as identifiers of
customers is common in the service industry.

Pat

[Barry Smith]

>
>OK, thanks. So that seems to nail down 'repeatability of
>identification' as a key property of an identifier, right?
>
>Pat

For the sake of clarification, I have been assuming that we are
attempting to define 'identifier' as an information artifact with
these key property; examples: device serial numbers, URIs, SSNs,
passport numbers, bank account numbers, credit card numbers, ...
Thus not every proper name and not every definition description will
be an identifier in this sense.

[Pat Hayes]

I understand, but Im trying to nail down the exact sense involved.
Melanie's rejection of the belt-timing example seems to be based
wholly on its lack of repeatability, and this helps (for me) to get
the intuitions clearer. So here's another, related, issue: does the
identifer (in your sense) have to be 'readable' (as all your examples
are, above) or can it be wholly opaque and used only inside some kind
of black-box machinery? Imagine for example a computer address inside
some compiled code running inside a computer chip inside a car engine,
or the like. Computing people would have no problem calling that an
'identifier', but I suspect y'all might not want to include it in this
category (?)

Pat

[Barry Smith]

At 12:08 PM 11/19/2009, Pat Hayes wrote:

>On Nov 19, 2009, at 9:28 AM, Barry Smith wrote:
>
>>
>>>
>>>OK, thanks. So that seems to nail down 'repeatability of
>>>identification' as a key property of an identifier, right?
>>>
>>>Pat
>>
>>For the sake of clarification, I have been assuming that we are
>>attempting to define 'identifier' as an information artifact with
>>these key property; examples: device serial numbers, URIs, SSNs,
>>passport numbers, bank account numbers, credit card numbers, ...
>>Thus not every proper name and not every definition description will
>>be an identifier in this sense.
>
>I understand, but Im trying to nail down the exact sense involved.
>Melanie's rejection of the belt-timing example seems to be based
>wholly on its lack of repeatability, and this helps (for me) to get
>the intuitions clearer. So here's another, related, issue: does the
>identifer (in your sense) have to be 'readable' (as all your examples
>are, above) or can it be wholly opaque and used only inside some kind
>of black-box machinery? Imagine for example a computer address inside
>some compiled code running inside a computer chip inside a car engine,
>or the like. Computing people would have no problem calling that an
>'identifier', but I suspect y'all might not want to include it in this
>category (?)

Couldn't one always, in such circumstances, program another machine
to flash the ID in a manner visible to humans?
BS

[Darren Natale]

This list of examples implies to me that an identifier cannot be
separated from the system to which it belongs. For example, PubMed IDs
and GenBank Identifiers (gi numbers) are both numerical identifiers, but
one serves solely to identify scientific publications while the other
serves to identify a sequence. Thus, the same number '1006569' actually
identifies different things in these different contexts.

[Michel Dumontier]

On Thu, Nov 19, 2009 at 10:22 AM, Pat Hayes <pha...@ihmc.us> wrote:

On Nov 19, 2009, at 6:02 AM, Melanie Courtot wrote:

On 18-Nov-09, at 10:39 PM, Pat Hayes wrote:

On Nov 18, 2009, at 6:36 PM, Bjoern Peters wrote:

I don't care particularly about the repeated/single use identifiers, but I think it is important to have the 'intent' clearly stated in the definition of identifier as we want to use it. My crack at it:

identifier: An ICE created to distinguish the referent from other entities in a 'refer by identifier process'.

'refer by identifier process' A planned process in which an ICE is generated which has as a part an identifier which indicates that the ICE is about the entity denoted by the identifier.

I think Pat's example is good, and illustrates what we should not consider an identifier. In that example, the message would be some computer encoding for 'open in 123 seconds'. None of that is an identifier. If in contrast there is a second barcode scanner before the gate, the message could have been 'open when package XYZ arrives', XYZ being the barcode, then yes, that message contains an identifier.

Actually the example was supposed to illustrate the opposite intuition. I see no clear difference between the timing case and the XYZ barcode case. They both are ways of identifying the package, under different  circumstances. Why does something which could be glossed as "the package which will arrive on this belt in 123 seconds" not count as an identifier, or at least an identifying description? It is sufficient to pick out that one package from all the others. Why is that not sufficient to make it count as an identifier?

I think it is an identifying description as you mention but not an identifier in the sense we would like to use it. While saying "the package which will arrive on this belt in 123 seconds" will allow you to get one package, it can not be used repeatedly to allow you to pick the same package each time.

OK, thanks. So that seems to nail down 'repeatability of identification' as a key property of an identifier, right?

Pat

What you've described is one system of identification, which involves a fixed interval of time, and it can indeed be used to distinguish one package from another. If you needed this identification to persist into other temporal intervals, you'd probably want to affix a label on the package with that information (perhaps related to absolute elapsed time or clock time). 

-=Michel=-

 

[Barry Smith]

It is not the same identifier -- even though, if it were a number, it
would be the same number.
Identifiers use numbers, but they are not numbers, inter alia because
they are sociohistorical creations (information artifacts).
BS

[Darren Natale]

These identifiers do indeed comprise numbers, but that is immaterial to
the point. My point will be made with the answer to this question:

*WHY* are these not the same identifier?

[Barry Smith]

At 01:19 PM 11/19/2009, Darren Natale wrote:
>These identifiers do indeed comprise numbers, but that is immaterial
>to the point. My point will be made with the answer to this question:
>
>*WHY* are these not the same identifier?

Why is my name not the same name as the name of this gentleman:
http://www.barrysmith.com/_/HOME.html ?
Let me count the ways:
the relevant naming acts (acts of baptism), in which these two names
were created, occurred at different times and on different continents.
BS

[Darren Natale]

Yes, this is what I was looking for. Indeed, it points to the idea that
the definition of an identifier must include some statement about
context or range, no? Just wondering if it is necessary. Anyway, this
is what I took away from Melanie's "identifier system" idea.

[Pat Hayes]

On Nov 19, 2009, at 12:44 PM, Barry Smith wrote:

> At 01:19 PM 11/19/2009, Darren Natale wrote:
>> These identifiers do indeed comprise numbers

Um... numerals, actually.

>> , but that is immaterial
>> to the point. My point will be made with the answer to this
>> question:
>>
>> *WHY* are these not the same identifier?
>
> Why is my name not the same name as the name of this gentleman:
> http://www.barrysmith.com/_/HOME.html ?

Hmm, seems to me that it is the same name. At any rate, I do not know
how to distinguish a name used in this sense from a name used in your
sense. In fact, I don't think there is any way to individuate a name
used in your sense. It certainly isn't a string of characters, it
cannot be written down, and Im not sure it can even be known. I know
you are the Barry Smith that I know, but I have no idea how you were
baptised. I can quote a name used in my sense: "Barry Smith" denotes
the string bee-ay-ar-ar-... etc. , but how would one indicate a name
used in your sense? Just quoting won't distinguish the Barry-Smith-
name-of-you from the Barry-Smith-name-of-him.

> Let me count the ways:
> the relevant naming acts (acts of baptism), in which these two names
> were created, occurred at different times and on different continents.

But that does not require that there are two different names involved,
only that a name got re-used. Happens all the time :-)

Pat

[Adam M. Goldstein]

On Nov 19, 2009, at 4:07 PM, Pat Hayes wrote:

>
> On Nov 19, 2009, at 12:44 PM, Barry Smith wrote:

<snip>

>>
>> Why is my name not the same name as the name of this gentleman:
>> http://www.barrysmith.com/_/HOME.html ?
>
> Hmm, seems to me that it is the same name. At any rate, I do not know
> how to distinguish a name used in this sense from a name used in your
> sense. In fact, I don't think there is any way to individuate a name
> used in your sense. It certainly isn't a string of characters, it
> cannot be written down, and Im not sure it can even be known. I know
> you are the Barry Smith that I know, but I have no idea how you were
> baptised. I can quote a name used in my sense: "Barry Smith" denotes
> the string bee-ay-ar-ar-... etc. , but how would one indicate a name
> used in your sense? Just quoting won't distinguish the Barry-Smith-
> name-of-you from the Barry-Smith-name-of-him.
>
>> Let me count the ways:
>> the relevant naming acts (acts of baptism), in which these two names
>> were created, occurred at different times and on different continents.

My sense is the same as Barry's. The graphical expressions of the names are the same, but they are different names. There are vocal expressions of it too, described by Pat above. The written doesn't denote the vocal; both denote (refer to) the person. Though you don't know how the baptism occurred, you do know that others you learned the name from know it because they learned it from others who learned it from others who . . . . who know how the baptism occurs. Some instances of "Barry Smith" will be correctly used to refer to the ontologist, others will not be. There are some necessary truths about the ontologist Barry Smith---born in a given place of certain parents, and probably a few other things.

Names aren't physical, and I don't know how you could indicate one, indexically, at least. They aren't character strings, etc. They can be used incorrectly or correctly.

This seems pretty straightforward to me, which I don't say to suggest that it's obvious, but it's just how I think.

>
> But that does not require that there are two different names involved,
> only that a name got re-used. Happens all the time :-)

If you mean, similar expressions in writing or speech, yes; if you mean, the singular referring term, no.

>
> Pat
>



<- snip ->

[Pat Hayes]

Absolutely not. This idea of there being some kind of chain of
reference linking every name use to a single source 'baptism' event
which attaches the name to iits referent, though often cited, is
complete hokum. And even if it were true, it cannot possibly explain
HOW the name usage at the end can be referential, since there is
nothing causal in this chain, even when it exists.

> Some instances of "Barry Smith" will be correctly used to refer to
> the ontologist, others will not be. There are some necessary truths
> about the ontologist Barry Smith---born in a given place of certain
> parents, and probably a few other things.
>
> Names aren't physical, and I don't know how you could indicate one,
> indexically, at least. They aren't character strings, etc. They can
> be used incorrectly or correctly.

Well, then, it falls to you (or others who hold this view) to give an
account of just what kind of thing these non-physical, non-lexical,
non-symbolic entities actually are. To me they seem like a fiction.
Seems to me that all the other Pat Hayeses all simply *have the same
name as* me. This is of course a problem with names, but one doesn't
solve the problem by inventing a fantasy where the names aren't
"really" the same after all. What we actually are presented with are
(let me call them) name-tokens, things like "Barry Smith" and "Pat
Hayes", actual character strings. And then we get from those to actual
people. This is a many-many correspondence, which isn't helped by
inventing something in between which is 1:1 with the people but
ontologically mysterious: not physical, not symbol, not lexical, not
linguistic, not ...

> This seems pretty straightforward to me, which I don't say to
> suggest that it's obvious, but it's just how I think.

To me it seems like the most rampant mysterianism. A name is something
that can be written on an envelope or a name-tag or a legal document.
If you ask me to write my name, I will do so. I will not protest that
my actual name is something ineffable, unique to me, that cannot be
rendered using the Latin alphabet or indeed in any character string of
any possible alphabet. Thats not going to cut it with the Government,
for a start.

>> But that does not require that there are two different names
>> involved,
>> only that a name got re-used. Happens all the time :-)
>
> If you mean, similar expressions in writing or speech, yes; if you
> mean, the singular referring term, no.

Then I have absolutely no idea what you mean by a 'referring term'.
Referring terms ARE expressions in writing or speech. How do you
distinguish "Pat Hayes" from "Pat Hayes" ? Or if neither of these are
my true name, how do you refer to my name *at all*?

Pat

>
>>
>> Pat
>>
>
>
>
> <- snip ->
> ------------------
> Adam M. Goldstein PhD, MSLIS
> --
> z_calif...@shiftingbalance.org
> http://www.shiftingbalance.org
> --
> http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
> --
> (914) 637-2717 (msg)
> --
> Dept of Philosophy
> Iona College
> 715 North Avenue
> New Rochelle NY 10801
>

[Barry Smith]

If you have a credit card number, and a social security number, and
they are numeral for numeral the same (string), do you have two
numbers, or one?
BS

[Barry Smith]

> > My sense is the same as Barry's. The graphical expressions of the
> > names are the same, but they are different names. There are vocal
> > expressions of it too, described by Pat above. The written doesn't
> > denote the vocal; both denote (refer to) the person. Though you
> > don't know how the baptism occurred, you do know that others you
> > learned the name from know it because they learned it from others
> > who learned it from others who . . . . who know how the baptism
> > occurs.
>
>Absolutely not. This idea of there being some kind of chain of
>reference linking every name use to a single source 'baptism' event
>which attaches the name to iits referent, though often cited, is
>complete hokum

Both I, and the other Barry Smith referred to, were in fact baptized.
I agree that there is not in every case a baptism.

>. And even if it were true, it cannot possibly explain
>HOW the name usage at the end can be referential, since there is
>nothing causal in this chain, even when it exists.

This was not what it was used to explain. It was used to try to help
string-reductionists such as yourself, to see that
string-reductionism may not be quite the whole story.

> > Some instances of "Barry Smith" will be correctly used to refer to
> > the ontologist, others will not be. There are some necessary truths
> > about the ontologist Barry Smith---born in a given place of certain
> > parents, and probably a few other things.
> >
> > Names aren't physical, and I don't know how you could indicate one,
> > indexically, at least. They aren't character strings, etc. They can
> > be used incorrectly or correctly.
>
>Well, then, it falls to you (or others who hold this view) to give an
>account of just what kind of thing these non-physical, non-lexical,
>non-symbolic entities actually are. To me they seem like a fiction.
>Seems to me that all the other Pat Hayeses all simply *have the same
>name as* me. This is of course a problem with names, but one doesn't
>solve the problem by inventing a fantasy where the names aren't
>"really" the same after all.

This does not sound like an argument. More like a: names are only
strings, therefore those who say they are something else are fantasizing.

> What we actually are presented with are
>(let me call them) name-tokens, things like "Barry Smith" and "Pat
>Hayes", actual character strings.

More begging the question. You assume that the conclusion you favor
is true, and use this conclusion as an argument against those who
deny the conclusion.

>And then we get from those to actual
>people. This is a many-many correspondence, which isn't helped by
>inventing something in between which is 1:1 with the people but
>ontologically mysterious: not physical, not symbol, not lexical, not
>linguistic, not ...

Again: If you have a credit card number, and a social security

number, and they are numeral for numeral the same (string), do you
have two numbers, or one?

> > This seems pretty straightforward to me, which I don't say to
> > suggest that it's obvious, but it's just how I think.
>
>To me it seems like the most rampant mysterianism. A name is something
>that can be written on an envelope or a name-tag or a legal document.

A numeral is something that can be written on an envelope.
A name (in the sense here at issue) is more like a number, than like a numeral.

>If you ask me to write my name, I will do so. I will not protest that
>my actual name is something ineffable, unique to me, that cannot be
>rendered using the Latin alphabet or indeed in any character string of
>any possible alphabet.

Discombobulating again, aren't we?

>Thats not going to cut it with the Government,
>for a start.
>
> >> But that does not require that there are two different names
> >> involved,
> >> only that a name got re-used. Happens all the time :-)
> >
> > If you mean, similar expressions in writing or speech, yes; if you
> > mean, the singular referring term, no.
>
>Then I have absolutely no idea what you mean by a 'referring term'.
>Referring terms ARE expressions in writing or speech. How do you
>distinguish "Pat Hayes" from "Pat Hayes" ? Or if neither of these are
>my true name, how do you refer to my name *at all*?

I guess that, if you have namesakes, then this is, in fact, quite a
difficult job; but not impossible. Your name is, for example: the
name of the Pat Hayes who was born in 1944 and wrote the "Naive
Physics Manifesto"

[Bjoern Peters]

As OBI development is driving the initial use case of 'identifier', I want to propose a more limited definition of 'centrally registered identifier' (CRID for short in what follows). I believe the approach outlined below will address all that OBI needs. It does not however attempt to cover all ways of identification. I am re-using multiple ideas raised by others on this thread before.

Definitions:

creating CRID system =def: a planned process with the objective to establish a system that allows to refer to specific entities of a certain kind and store information about them, by establishing a CRID registry and plan specifications for the process of 1) assigning a CRID and 2) looking up a CRID.

Example: Introduction of a personal identity card system in the UK. OBI agreeing on using a sourceforge for issue tracking, and registering a sourceforge account for that purpose. Agreeing on an alphanumeric format for strings which will be tracked in an Excel sheet to track blood samples received for use in bjoern's swine flu study.

assigning a CRID =def: a planned process in which an a new CRID is created and stored in the CRID registry along with information about an entity associated with the CRID.

example: A journal publisher receiving a Pubmed ID for a newly published article. Me applying for license plates for a new car. A manufacturer assigning a batch number to the antibodies produced today.

looking up a CRID=def: a planned process in which a request to the CRID registry is made to return the information associated with the CRID.

example: Searching for a pubmed ID on http://pubmed.org. Looking up the date in an Excel sheet which a blood sample arrived based on a donation ID. A policeman looking up the registration information associated with a vehicle license plate.

CRID registry =def: a bearer of an ICE comprising a list entries pairing a CRID with an associated ICE

Example: Piece of paper, Excel sheet, MySQL database

CRID =def: an ICE which specifies a CRID registry and allows obtain information about an entity associated with the CRID.

Example: 'Pubmed ID:12345'. 'ISBN:12345'.
Counter example: '12345'

Some notes:
- A CRID system implicitly defines a 'denotation relation' between the CRID and the entity about which information is stored. This denotation requires to specify both the CRID and the CRID system in which it was created.

- CRIDs can be used outside without looking up the CRID in the registry, e.g. in JR's example for railroad car barcodes used by a machine to see if it already loaded a certain car. In such secondary uses, it is assumed though that the CRIDs are well formed (e.g. unique, readable in a certain format), which enables their use to 'denote'.

- No claim is made that the information stored about the entity associated with the CRID allows to identify it (e.g. a batch number won't help you find where a produced antibody is now).

- The CRID approach may be useful to less formal identifiers such as names: 'Birth certificates' are CRIDs. Normal use of human names can be taken as a shorthand for a CRID.

[Pat Hayes]

On Nov 19, 2009, at 6:20 PM, Barry Smith wrote:

>
>>> My sense is the same as Barry's. The graphical expressions of the
>>> names are the same, but they are different names. There are vocal
>>> expressions of it too, described by Pat above. The written doesn't
>>> denote the vocal; both denote (refer to) the person. Though you
>>> don't know how the baptism occurred, you do know that others you
>>> learned the name from know it because they learned it from others
>>> who learned it from others who . . . . who know how the baptism
>>> occurs.
>>
>> Absolutely not. This idea of there being some kind of chain of
>> reference linking every name use to a single source 'baptism' event
>> which attaches the name to iits referent, though often cited, is
>> complete hokum
>
> Both I, and the other Barry Smith referred to, were in fact baptized.
> I agree that there is not in every case a baptism.

My point is rather that the idea that reference is somehow determined
by a kind of chain-of-implicit-agreement with its source in an act of
baptism, perhaps one in the distant past, is wrong. Its a seductive
idea, and I think Quine uses it somewhere, but I still think it simply
does not stand up to scrutiny. The 'chains' one has to imagine just
become too long and fantastical to be convincing. What ur-act of
naming caused "Paris" to refer to the capital city of France?

>> . And even if it were true, it cannot possibly explain
>> HOW the name usage at the end can be referential, since there is
>> nothing causal in this chain, even when it exists.
>
> This was not what it was used to explain. It was used to try to help
> string-reductionists such as yourself, to see that
> string-reductionism may not be quite the whole story.

I don't accept that Im being 'reductionist' in any way. Strings exist
(we agree, I hope?) and can be used as names. This doesn't seem like
reductionism to me, it seems like stark common sense. What is there to
reduce?

>>> Some instances of "Barry Smith" will be correctly used to refer to
>>> the ontologist, others will not be. There are some necessary truths
>>> about the ontologist Barry Smith---born in a given place of certain
>>> parents, and probably a few other things.
>>>
>>> Names aren't physical, and I don't know how you could indicate one,
>>> indexically, at least. They aren't character strings, etc. They can
>>> be used incorrectly or correctly.
>>
>> Well, then, it falls to you (or others who hold this view) to give an
>> account of just what kind of thing these non-physical, non-lexical,
>> non-symbolic entities actually are. To me they seem like a fiction.
>> Seems to me that all the other Pat Hayeses all simply *have the same
>> name as* me. This is of course a problem with names, but one doesn't
>> solve the problem by inventing a fantasy where the names aren't
>> "really" the same after all.
>
> This does not sound like an argument. More like a: names are only
> strings, therefore those who say they are something else are
> fantasizing.

Im not making an argument here. I am issuing a challenge. It seems
obvious to me, and obviously in conformity to common usage in human
society, to say that a person's name is a word or words, a string of
characters in a language, something that can be written down, copied,
typed or spoken. Certainly these things - let me call them name-words
- do indeed exist, and are commonly used to at least somehow
*indicate* names, and are *referred to* as names. (I take it we agree
on this, at least?) You apparently wish to claim, however, that they
are not names; but that names, properly understood, are something
else, something more abstract or platonic, perhaps? At any rate, not
something *lexical* in nature. In your ontology, therefore, there must
be several entities where I can see only one. To me, these extra, non-
string, entities seem peculiar, un-natural. I do not know how to tell
them apart, or to individuate them, or to count them. I do not know
their essential nature, or their necessary properties (they are not
physical nor symbolic in nature, for example. What *kind* of thing are
they?) It seems to me, to repeat my challenge, that it now falls to
you to explain to me why I should even believe that these (to me)
mysterious things should exist at all. They remind me of God. I see
nothing to suggest that they do exist, and every reason to suppose
that they do not. To repeat, I am not mounting an argument against
their existence: I simply am asking for reasons to believe in them.
Why are my Occamist intuitions wrong, in this case?

>
>> What we actually are presented with are
>> (let me call them) name-tokens, things like "Barry Smith" and "Pat
>> Hayes", actual character strings.
>
> More begging the question. You assume that the conclusion you favor
> is true, and use this conclusion as an argument against those who
> deny the conclusion.

Not at all. I was merely making a simple observation. Surely you will
agree that the things on a page, or written on the screen you are
reading this message on, are themselves mere character strings, so
that "Pat Hayes" and "Pat Hayes" are two tokens of the same character
string. And that these strings are what actually get written and sent
from place to place on the Internet, for example. Right? If not, then
we have a deeper misunderstanding than I would have thought possible.

My point was only that if I read the text "Pat Hayes" then it might
refer to me (be a - what? a rendering? - of my True Name) or it might
refer to someone else (be a rendering of their True Name), and there
is no way to know which, just by looking at what is actually written.

>
>> And then we get from those to actual
>> people. This is a many-many correspondence, which isn't helped by
>> inventing something in between which is 1:1 with the people but
>> ontologically mysterious: not physical, not symbol, not lexical, not
>> linguistic, not ...
>
> Again: If you have a credit card number, and a social security
> number, and they are numeral for numeral the same (string), do you
> have two numbers, or one?

Well, to speak strictly, you never actually 'have' a number. To 'have'
an SS number is bear a certain relationship to a number, one which
comes with a lot of social baggage in the USA. Similarly for CC
number, and banks. The relationships are very different, of course.
But I would indeed say that in this strange coincidence case that you
bear these relationships to the same number, yes.

>
>>> This seems pretty straightforward to me, which I don't say to
>>> suggest that it's obvious, but it's just how I think.
>>
>> To me it seems like the most rampant mysterianism. A name is
>> something
>> that can be written on an envelope or a name-tag or a legal document.
>
> A numeral is something that can be written on an envelope.
> A name (in the sense here at issue) is more like a number, than like
> a numeral.

So names cannot be written on envelopes? I really, really find this an
extraordinary idea, one I have never met before. (I presume that SS
numbers also cannot be written, on similar grounds?)

>
>> If you ask me to write my name, I will do so. I will not protest that
>> my actual name is something ineffable, unique to me, that cannot be
>> rendered using the Latin alphabet or indeed in any character string
>> of
>> any possible alphabet.
>
> Discombobulating again, aren't we?

No, just trying to get it absolutely clear what is being said. I take
it as obvious (?) that things that can be written down in, say, ASCII,
are character strings. So if a name is not a character string, then it
follows that names cannot be written down in, say, ASCII. But the same
argument will apply to any other alphabet, of course.

>
>> Thats not going to cut it with the Government,
>> for a start.
>>
>>>> But that does not require that there are two different names
>>>> involved,
>>>> only that a name got re-used. Happens all the time :-)
>>>
>>> If you mean, similar expressions in writing or speech, yes; if you
>>> mean, the singular referring term, no.
>>
>> Then I have absolutely no idea what you mean by a 'referring term'.
>> Referring terms ARE expressions in writing or speech. How do you
>> distinguish "Pat Hayes" from "Pat Hayes" ? Or if neither of these are
>> my true name, how do you refer to my name *at all*?
>
> I guess that, if you have namesakes, then this is, in fact, quite a
> difficult job; but not impossible. Your name is, for example: the
> name of the Pat Hayes who was born in 1944 and wrote the "Naive
> Physics Manifesto"

So the only way to refer to my name is by referring to me (by a
suitably detailed description, or by ostention if I am nearby) and
then saying, in effect, "his name". Yes, I can see that that does
work, in a strange way, but it seems to me to be a reductio ad
absurdum of the very idea of using names to refer to people. We have
names in order that people can refer to us, in language, by *using*
our names. They are our identifiers for communication. If the only way
to access the name is via the person named, then the name is
(literally) useless. The only time you can use it is when you don't
need to.

I have a friend who, due to a head injury, is unable to remember names
at all. (I suffer from a mild form of the same infirmity, but his is
extreme.) When speaking to him, it is necessary to refer to third
parties by description, and the descriptions become abbreviated to
what are in effect names. Thus, tallwhitehair and sexyofficeblonde
began as descriptors, became abbreviated and now play the role of
names when talking to, I'll call him Joe. Sounds to me that your
theory of names is rather like an assumption that we all, like Joe,
have damaged frontal lobes.

Pat

[Pat Hayes]

One number, playing two roles (so the simple 'have' is misleading.)
But I'm perfectly happy to say that SS number and CC number are
distinct entities even though they are the same numerical string. In
this case, the (one) string stands in two different relationships,
which are clearly distinguishable. But the same line of thinking
doesn't lead me to distinguish "Pat Hayes" being my name from "Pat
Hayes" being the name of http://www.deaconpat.com/ . There, the roles
are identical.

Pat

[Bjoern Peters]

Pat, I would be curious what you think of my proposal which I think addresses the points in both of your follow up emails (although in a more restrictive context).

[Barry Smith]

At 01:09 AM 11/20/2009, Pat Hayes wrote:
>One number, playing two roles (so the simple 'have' is misleading.)
>But I'm perfectly happy to say that SS number and CC number are
>distinct entities even though they are the same numerical string. In
>this case, the (one) string stands in two different relationships,
>which are clearly distinguishable. But the same line of thinking
>doesn't lead me to distinguish "Pat Hayes" being my name from "Pat
>Hayes" being the name of http://www.deaconpat.com/ . There, the roles
>are identical.

So if you have a visa card with number n, and 10 years later someone
else has a visa card with the string-wise same number n, then these
two credit card numbers are the same number with the same role?
BS

[Jonathan Rees]

In my wiki page I tried to take a stance somewhere between Barry's and
stringism. The idea is that the question of the relation of a given
"Pat Hayes" string (quality) to another could be based on copying.
Whether something is a copy (transitively) of something else has no
reference to meaning, mental state, baptism, or even role. Rather it
is an issue of history or provenance. In any particular case it may be
unknowable, but sometimes there is good evidence one way or another.
For example, when a large random number is seen twice, you can be
pretty sure that the second sighting derives from the first via
copying and other information-preserving processes (double reversal,
encryption/decryption, packaging with other stuff, etc.). Another kind
of evidence would come from knowledge of the procedures, methods, and
actions applied in the handling of the entity and its copies, such as
perimeter security. Or you could simply watch to see what happens
starting with the point of origin - track physical movement of copies
and all copying and transformation processes.

At least that's my attempt to make the ICE idea more rigorous, without
going all the way to stringism.

I agree with Pat that trying to convince anyone that "Pat Hayes" (the
name of ...) from "Pat Hayes" (the name of ,,,) are different names is
going to be an uphill struggle, since T.C.Mits would just say we have
two people with the same name. Even if the idea is sound, it would be
better to coin a new term for name-in-role (or name-as-pedigree if you
buy my stuff).

Jonathan

[Bjoern Peters]

I want to suggest once more to discuss these questions in the context of centrally registered IDs (a few emails back), which cover all of the examples we were trying to cover

(Barry's list)

>>examples: device serial numbers, URIs, SSNs,
>>passport numbers, bank account numbers, credit card numbers, ...
>>Thus not every proper name and not every definition description
>>will be an identifier in this sense.

Limiting to those better defined types of identifiers will help with the more complex issues discussed. For example, I would argue that a credit card number is only a CRID together with issuer information and expiration date (if they are indeed re-used). Also, Jonathan's question on how a generically dependent continuant 'spreads' is going to be straightforward to discuss in that context, as there clear 'issuing' and 'copying' processes.

[Pat Hayes]

On Nov 20, 2009, at 7:45 AM, Barry Smith wrote:

> At 01:09 AM 11/20/2009, Pat Hayes wrote:
>> One number, playing two roles (so the simple 'have' is misleading.)
>> But I'm perfectly happy to say that SS number and CC number are
>> distinct entities even though they are the same numerical string. In
>> this case, the (one) string stands in two different relationships,
>> which are clearly distinguishable. But the same line of thinking
>> doesn't lead me to distinguish "Pat Hayes" being my name from "Pat
>> Hayes" being the name of http://www.deaconpat.com/ . There, the
>> roles
>> are identical.
>
> So if you have a visa card with number n, and 10 years later someone
> else has a visa card with the string-wise same number n, then these
> two credit card numbers are the same number with the same role?

Different person, but yes, same number, same role.

This seems absolutely obvious to me, by the way, Im amazed we are even
having this conversation. You just said yourself they are the same
numeral, so surely these identical numerals denote the same number. I
guess it could be that we read one of the numerals as written in
hexadecimal, or something, but that hardly seems plausible....

Hmm, that triggered a thought. Maybe we could say that a numeral *on a
credit card* denotes something other than a simple number. That might
work. We would need to distinguish 'credit card contexts' or something
like that, which systematically re-interpret numerals to denote
special 'credit card numbers', these being different from ordinary
numbers in some special financial way. They probably don't have
arithmetic operators defined on them, for example. But this seems like
a lot of work to no real purpose, as AFAICS there is no harm in taking
the obvious line that a credit card number is simply a number, or at
any rate that a CC numeral (speaking strictly now) is simply a numeral.

Here's an argument that might begin to persuade me. Suppose some bank
decides to start using CC numbers with letters in them. After all, why
not? Then we would likely still call them 'credit card numbers' even
though they were clearly no longer numerals. And we wouldn't feel any
strong cognitive dissonance. Which seems to argue that these numbers
are in fact just *codes*, not really being interpreted as numerals at
all. The fact that it makes no sense to add or multiply them also
argues in this direction. But all that this argument would make me
want to do is to retreat from 'number' to 'numeral', ie treat these
things like product codes or bank routing 'numbers' (which also have
letters in them), being simply unique character strings which should
not be thought of as denoting anything (except possibly themselves).

Pat

[Barry Smith]

At 11:08 AM 11/20/2009, Pat Hayes wrote:

>On Nov 20, 2009, at 7:45 AM, Barry Smith wrote:
>
>>At 01:09 AM 11/20/2009, Pat Hayes wrote:
>>>One number, playing two roles (so the simple 'have' is misleading.)
>>>But I'm perfectly happy to say that SS number and CC number are
>>>distinct entities even though they are the same numerical string. In
>>>this case, the (one) string stands in two different relationships,
>>>which are clearly distinguishable. But the same line of thinking
>>>doesn't lead me to distinguish "Pat Hayes" being my name from "Pat
>>>Hayes" being the name of http://www.deaconpat.com/ . There, the
>>>roles
>>>are identical.
>>
>>So if you have a visa card with number n, and 10 years later someone
>>else has a visa card with the string-wise same number n, then these
>>two credit card numbers are the same number with the same role?
>
>Different person, but yes, same number, same role.
>
>This seems absolutely obvious to me, by the way, Im amazed we are even
>having this conversation. You just said yourself they are the same
>numeral, so surely these identical numerals denote the same number. I
>guess it could be that we read one of the numerals as written in
>hexadecimal, or something, but that hardly seems plausible....
>Hmm, that triggered a thought. Maybe we could say that a numeral *on a
>credit card* denotes something other than a simple number. That might
>work.

Finally, light.

>We would need to distinguish 'credit card contexts' or something
>like that, which systematically re-interpret numerals to denote
>special 'credit card numbers', these being different from ordinary
>numbers in some special financial way.

Indeed.

>They probably don't have
>arithmetic operators defined on them, for example. But this seems like
>a lot of work to no real purpose, as AFAICS there is no harm in taking
>the obvious line that a credit card number is simply a number, or at
>any rate that a CC numeral (speaking strictly now) is simply a numeral.

There is no harm in treating as a mere string a demand note from your
bank requesting that you pay them the $1000 you owe them before
Tuesday. But it is not a mere string.

>Here's an argument that might begin to persuade me. Suppose some bank
>decides to start using CC numbers with letters in them. After all, why
>not? Then we would likely still call them 'credit card numbers' even
>though they were clearly no longer numerals. And we wouldn't feel any
>strong cognitive dissonance. Which seems to argue that these numbers
>are in fact just *codes*, not really being interpreted as numerals at
>all. The fact that it makes no sense to add or multiply them also
>argues in this direction. But all that this argument would make me
>want to do is to retreat from 'number' to 'numeral', ie treat these
>things like product codes or bank routing 'numbers' (which also have
>letters in them), being simply unique character strings which should
>not be thought of as denoting anything (except possibly themselves).

At no stage did I suggest that credit card numbers denote anything.
My point is that they are not mere strings, but something (as you are
beginning to glimpse above) very different, inter alia because of
their special historical origin, and their embeddedness within a
special globally extended socio-electronic system.
BS

[Melanie Courtot]

Hi Bjoern,
I like the general idea. I have a few concerns about CRID, I guess
linked to the label "central registry". However as you mention a piece
of paper could be a CRID registry, so I am happy with the idea behind
the label.

I would add a restriction CRID system has part CRID registry and CRIDs.

Some more specific comments in line.

On 2009-11-19, at 8:53 PM, Bjoern Peters <bpe...@liai.org> wrote:

> As OBI development is driving the initial use case of 'identifier',
> I want to propose a more limited definition of 'centrally registered
> identifier' (CRID for short in what follows). I believe the approach
> outlined below will address all that OBI needs. It does not however
> attempt to cover all ways of identification. I am re-using multiple
> ideas raised by others on this thread before.
>
> Definitions:
>
> creating CRID system =def: a planned process with the objective to
> establish a system that allows to refer to specific entities of a
> certain kind and store information about them, by establishing a
> CRID registry and plan specifications for the process of 1)
> assigning a CRID and 2) looking up a CRID.
>

While I believe in most cases we will indeed have entities of the same
kind I am not sure this is a requirement for the system.
I like the system generation process as it reflects the idea that
identifiers are historically linked, and creating an "instant
identifier" wouldn't work as its usage wouldn't be repeatable (because
not recorded, there would be no way to ensure it always refer to the
same thing)
I also believe we refer to entities or collection of entities, e.g.
for lot number.

> Example: Introduction of a personal identity card system in the UK.
> OBI agreeing on using a sourceforge for issue tracking, and
> registering a sourceforge account for that purpose. Agreeing on an
> alphanumeric format for strings which will be tracked in an Excel
> sheet to track blood samples received for use in bjoern's swine flu
> study.

Maybe even add nomenclatures, classifications...

>
> assigning a CRID =def: a planned process in which an a new CRID is
> created and stored in the CRID registry along with information about
> an entity associated with the CRID.

I think the CRID can be created prior to the assignement. I can create
a barcode system, print out 10 000 tags and stick them on my products
only later on. I would put CRID creation within the system creation.
Similarly for storage - assigning a CRID is really only associating
the existing identifier to the entity it denotes.
I thought maybe that is where the role/function of the CRID is realized?

>
>
> example: A journal publisher receiving a Pubmed ID for a newly
> published article. Me applying for license plates for a new car. A
> manufacturer assigning a batch number to the antibodies produced
> today.
>
> looking up a CRID=def: a planned process in which a request to the
> CRID registry is made to return the information associated with the
> CRID.
>
> example: Searching for a pubmed ID on http://pubmed.org. Looking up
> the date in an Excel sheet which a blood sample arrived based on a
> donation ID. A policeman looking up the registration information
> associated with a vehicle license plate.

I didn't think about that one and it makes complete sense to me -
CRIDs are indeed somehow more useful if we can retrieve the
information :)

>
> CRID registry =def: a bearer of an ICE comprising a list entries
> pairing a CRID with an associated ICE
>
> Example: Piece of paper, Excel sheet, MySQL database

Again, I am not 100% happy with the label (maybe something like
"identifier association system"?) but I agree with the idea (and
labels/definitions can be refined later)

>
> CRID =def: an ICE which specifies a CRID registry and allows obtain
> information about an entity associated with the CRID.
>
> Example: 'Pubmed ID:12345'. 'ISBN:12345'.
> Counter example: '12345'

I like this examples, as they clearly show that without the system
there is no identifier (at least in our OBI driving cases).
If others think this is an acceptable first step, maybe we could try
formalizing/implementing this - knowing that there are other
distinguishing entities ("the hamburger on table 3") that we don't
address here.

Thanks,
Melanie

[Bjoern Peters]

Thanks for your comments Melanie! Obviously this was build on some of the things you said before. responses inline.

----- "Melanie Courtot" <mcou...@gmail.com> wrote:


> > Definitions:
> >
> > creating CRID system =def: a planned process with the objective to
>
> > establish a system that allows to refer to specific entities of a
> > certain kind and store information about them, by establishing a
> > CRID registry and plan specifications for the process of 1)
> > assigning a CRID and 2) looking up a CRID.
> >
>
> While I believe in most cases we will indeed have entities of the same
> kind I am not sure this is a requirement for the system.

True, there could be CRID systems where limitations to certain entities is not useful, and it can be removed from the definition. It will be a useful way to distinguish different subclasses of CRID systems, e.g. those identifying individual human beings vs. those identifying railcars etc.

> I like the system generation process as it reflects the idea that
> identifiers are historically linked, and creating an "instant
> identifier" wouldn't work as its usage wouldn't be repeatable (because
> not recorded, there would be no way to ensure it always refer to the
> same thing)
> I also believe we refer to entities or collection of entities, e.g.
> for lot number.

I was assuming a 'collection of entities' is_a entity.

>
> > Example: Introduction of a personal identity card system in the UK.
>
> > OBI agreeing on using a sourceforge for issue tracking, and
> > registering a sourceforge account for that purpose. Agreeing on an
>
> > alphanumeric format for strings which will be tracked in an Excel
> > sheet to track blood samples received for use in bjoern's swine flu
>
> > study.
>
> Maybe even add nomenclatures, classifications...

I was thinking about that, and agree that they fit, but hesitated putting them in as examples, as they sound like 'ontologies', and that would alert Alan's instincts against self-reflection.

> >
> > assigning a CRID =def: a planned process in which an a new CRID is
> > created and stored in the CRID registry along with information about
> > an entity associated with the CRID.

> I think the CRID can be created prior to the assignement. I can create
> a barcode system, print out 10 000 tags and stick them on my products
> only later on. I would put CRID creation within the system creation.
> Similarly for storage - assigning a CRID is really only associating
> the existing identifier to the entity it denotes.
> I thought maybe that is where the role/function of the CRID is
> realized?
>

I thought we would only call something a CRID that is recorded in the registry with an association to an entity. In that view, the printed out tag is not a CRID yet. Similarly, a number that confirms to the credit card format is not a credit card number until it is registered and can be used for transactions.

> >
> >
> > example: A journal publisher receiving a Pubmed ID for a newly
> > published article. Me applying for license plates for a new car. A
> > manufacturer assigning a batch number to the antibodies produced
> > today.
> >
> > looking up a CRID=def: a planned process in which a request to the
> > CRID registry is made to return the information associated with the
> > CRID.
> >
> > example: Searching for a pubmed ID on http://pubmed.org. Looking up
> > the date in an Excel sheet which a blood sample arrived based on a
> > donation ID. A policeman looking up the registration information
> > associated with a vehicle license plate.
>
> I didn't think about that one and it makes complete sense to me -
> CRIDs are indeed somehow more useful if we can retrieve the
> information :)
>
> >
> > CRID registry =def: a bearer of an ICE comprising a list entries
> > pairing a CRID with an associated ICE
> >
> > Example: Piece of paper, Excel sheet, MySQL database
>
> Again, I am not 100% happy with the label (maybe something like
> "identifier association system"?) but I agree with the idea (and
> labels/definitions can be refined later)

I am happy with changing the label. I just want to be careful about distinguishing the piece of paper on which the associations are stored with the system that also includes the plans specifications on how to use it.

> >
> > CRID =def: an ICE which specifies a CRID registry and allows obtain
>
> > information about an entity associated with the CRID.
> >
> > Example: 'Pubmed ID:12345'. 'ISBN:12345'.
> > Counter example: '12345'
>
> I like this examples, as they clearly show that without the system
> there is no identifier (at least in our OBI driving cases).
> If others think this is an acceptable first step, maybe we could try
>
> formalizing/implementing this - knowing that there are other
> distinguishing entities ("the hamburger on table 3") that we don't
> address here.
>

Great! That would be my proposal as well. I also believe this will help defining other means of identification, most of which I think function essentially the same way except that the explicit processes / identifiers happen in heads of people rather then being explicitly trackable.

[Barry Smith]

Some of this, it seems, is relevant to this work:
http://wings.buffalo.edu/cogsci/Activities/Colloquium/CLLQf05/FinalCeustersRome4BS.pdf
See also further references here:
http://www.referent-tracking.com/RTU/?page=papers.phtml
BS

[Bjoern Peters]

Thanks for the link, Barry. I read through the first 28 pages of the manuscript, and it is clearly related to this. To summarize (Barry correct me if this is wrong): The paper proposes a referent tracking system (RTS) to assign unique identifiers to instances encountered in medical practice such as instances of human being, fracture of elbow, X-ray film. Everything I wrote about a CRID system applies to an RTS. In addition, the proposed RTS is defined more specifically to serve its intended purpose. For example, identifiers for data are explicitly excluded, the desire to avoid two identifiers pointing to the same entity is stated, and additional processes such as the ability to reserve identifiers for potential use are described.

Overall, I think that this further confirms that the CRID proposal identifies a useful set of identification systems (An RTS is_a CRID system). Unless there are objections, I would like to go ahead and implement the proposal on Tuesdays call.

The one addition I would have to my previous email is to add a relation

domain: CRID,
range: entity
label: CRID_refers_to
definition: a relation between a CRID and an entity about which information is stored in the CRID registry associated with the CRID.

- Bjoern








- Bjoern






processes identified in the paper (identifier creation, assignment, and terminology chosen is similar to my email, which I take as a conformation

[Robert Hoehndorf]

>>>>> "PH" == Pat Hayes <pha...@ihmc.us> writes:

Hi,

>> This does not sound like an argument. More like a: names are
>> only strings, therefore those who say they are something else
>> are fantasizing.

PH> Im not making an argument here. I am issuing a challenge. It
PH> seems obvious to me, and obviously in conformity to common
PH> usage in human society, to say that a person's name is a word
PH> or words, a string of characters in a language, something that
PH> can be written down, copied, typed or spoken. Certainly these
PH> things - let me call them name-words - do indeed exist, and
PH> are commonly used to at least somehow *indicate* names, and
PH> are *referred to* as names. (I take it we agree on this, at
PH> least?) You apparently wish to claim, however, that they are
PH> not names; but that names, properly understood, are something
PH> else, something more abstract or platonic, perhaps? At any
PH> rate, not something *lexical* in nature. In your ontology,
PH> therefore, there must be several entities where I can see only
PH> one. To me, these extra, non- string, entities seem peculiar,
PH> un-natural. I do not know how to tell them apart, or to
PH> individuate them, or to count them. I do not know their
PH> essential nature, or their necessary properties (they are not
PH> physical nor symbolic in nature, for example. What *kind* of
PH> thing are they?) It seems to me, to repeat my challenge, that
PH> it now falls to you to explain to me why I should even believe
PH> that these (to me) mysterious things should exist at all. They
PH> remind me of God. I see nothing to suggest that they do exist,
PH> and every reason to suppose that they do not. To repeat, I am
PH> not mounting an argument against their existence: I simply am
PH> asking for reasons to believe in them. Why are my Occamist
PH> intuitions wrong, in this case?

Let me try to give a non-BFO answer to the challenge. Let there be a
naming-relation den(S,E), saying that the string-token S names (or
denotes) the entity E. So, on a letter you send to BS, the
string-token "Barry Smith" on the letter will name BS, the
ontologist.
You observe, correctly, that the same string type (and maybe token)
can refer to different entities. In one relation (instance) of the type
den, such as den("Barry Smith",BS), the token "Barry Smith" will refer
to one entity, to BS the ontologist. The same token "Barry Smith" may
refer to another entity in a different context (i.e., in a different
relation), but in the relation-instance above, the string-token "Barry
Smith" refers to only one entity.
Now, take the relation-instance (let's call it R) apart. There is the string "Barry
Smith", what we may call the name.
There is "Barry Smith"-as-being-the-first-argument-in-R, or
"Barry Smith"-as-being-the-name-in-R.
There is the relation-instance (individual, or n-ary property).
There is BS, the human being and ontologist.
There is BS-as-being-the-second-argument-in-R, or
BS-as-being-the-referent-in-R.

I think Barry would like to call "Barry
Smith"-as-being-the-first-argument-in-R the name. You (and me, and
most people) would call the string (type) "Barry Smith" the name. I
think there is no problem, we have ontologies so that we can specify
what we mean: either the string type is called "name", or the role it
plays within a specific relation instance (R).

Now, in this analysis, there is a problem: BFO has no relational roles
(afaik), and the roles of BFO cannot serve the function required
here. Maybe (but I am no BFO expert) the relation "den" can be
replaced by naming processes, and a string-token can play the role of
being-the-name in these processes, while BS plays the role of
being-the-referent. I find this strange, but it may provide a BFO
compatible solution to your puzzle while still maintaining Barry's
claim that names are the roles string play in naming something.

PH> My point was only that if I read the text "Pat Hayes" then it
PH> might refer to me (be a - what? a rendering? - of my True
PH> Name) or it might refer to someone else (be a rendering of
PH> their True Name), and there is no way to know which, just by
PH> looking at what is actually written.

You are right, of course. Whenever "Pay Hayes" refers to you, the
token (and, arguably, the string-type in a different, lifted relation)
play different roles in different relation-instances.

>> number, and they are numeral for numeral the same (string), do
>> you have two numbers, or one?

PH> Well, to speak strictly, you never actually 'have' a
PH> number. To 'have' an SS number is bear a certain relationship
PH> to a number, one which comes with a lot of social baggage in
PH> the USA. Similarly for CC number, and banks. The relationships
PH> are very different, of course. But I would indeed say that in
PH> this strange coincidence case that you bear these
PH> relationships to the same number, yes.

Here I disagree: you bear different relations to the same number, and
the same number plays different roles (role-types) in these relations.
There is a binary has-SS relation, with role-types SS-Number and SS-Bearer
(for lack of better name), and similarily has-CC with roles CC-Number
and CC-Bearer. The same number plays different roles in different
relations: has-SS(123,X) and has-CC(123,X), with different roles for
123 (assuming it is the number).

>> A numeral is something that can be written on an envelope. A
>> name (in the sense here at issue) is more like a number, than
>> like a numeral.

PH> So names cannot be written on envelopes? I really, really find
PH> this an extraordinary idea, one I have never met before. (I
PH> presume that SS numbers also cannot be written, on similar
PH> grounds?)

Names can be written on envelopes, if by "name" you mean the
string. BS seems to be talking about the role the string plays in a
certain relation (that you establish by writing the string on the
envelope), and the role itself cannot be written down (roles are
similar to properties).

Rob.

[Pat Hayes]

But read on: this COULD be done, but there is no POINT in doing it. It
achieves nothing except needless complication and a host of
unnecessary entities, cluttering up the ontology to no purpose.

>> They probably don't have
>> arithmetic operators defined on them, for example. But this seems
>> like
>> a lot of work to no real purpose, as AFAICS there is no harm in
>> taking
>> the obvious line that a credit card number is simply a number, or at
>> any rate that a CC numeral (speaking strictly now) is simply a
>> numeral.
>
> There is no harm in treating as a mere string a demand note from your
> bank requesting that you pay them the $1000 you owe them before
> Tuesday. But it is not a mere string.

I don't know why you use the "mere". It is, however, a string. Of
course, like any other character string, it may have a meaning (as
this one does, for example). That however does not make it into
something **other than** a character string. Having properties (such
as being meaningful) is one thing; being something other than what it
in fact is, is another. You seem to be confusing these notions. The
string "chat" in English means one thing, in French something
different. Nevertheless, it is the character string that it is, in
both cases. There is only one of it. It is created by hitting four
keys C H A T in succession.

>
>> Here's an argument that might begin to persuade me. Suppose some bank
>> decides to start using CC numbers with letters in them. After all,
>> why
>> not? Then we would likely still call them 'credit card numbers' even
>> though they were clearly no longer numerals. And we wouldn't feel any
>> strong cognitive dissonance. Which seems to argue that these numbers
>> are in fact just *codes*, not really being interpreted as numerals at
>> all. The fact that it makes no sense to add or multiply them also
>> argues in this direction. But all that this argument would make me
>> want to do is to retreat from 'number' to 'numeral', ie treat these
>> things like product codes or bank routing 'numbers' (which also have
>> letters in them), being simply unique character strings which should
>> not be thought of as denoting anything (except possibly themselves).
>
> At no stage did I suggest that credit card numbers denote anything.
> My point is that they are not mere strings, but something (as you are
> beginning to glimpse above) very different, inter alia because of
> their special historical origin, and their embeddedness within a
> special globally extended socio-electronic system.

They are of course embedded in such a system, but that does not
prevent them being numerals. They *are* numerals. If you add one to my
Visa card number, the result is even.

Pat

[Pat Hayes]

In which case, we have an immediate problem, since by writing den(S,E)
you have effectively already denied this possibility. A name can
denote only one thing (in a particular interpretation). There is only
one denotation relationship. (I am assuming that we are talking within
the usual semantic framework of formalized notations such as FO logic.)

> In one relation (instance) of the type
> den, such as den("Barry Smith",BS), the token "Barry Smith" will refer
> to one entity, to BS the ontologist. The same token "Barry Smith" may
> refer to another entity in a different context (i.e., in a different
> relation), but in the relation-instance above, the string-token "Barry
> Smith" refers to only one entity.

I do not know what you mean here by "context". Unfortunately, the C-
word has been used to mean virtually anything, so it is itself,
ironically, one of the most context-sensitive words ever used in
technical discourse. But if you mean that each relation imposes or
defines its own context, so that names (strings) can change their
denotation depending upon the relation they are said to stand in, then
I must object, since that would make all formalization impossible.

> Now, take the relation-instance (let's call it R) apart. There is
> the string "Barry
> Smith", what we may call the name.

Right.

> There is "Barry Smith"-as-being-the-first-argument-in-R, or
> "Barry Smith"-as-being-the-name-in-R.

? No, there is "Barry Smith", the name. It is what it is, no matter
what relationships it may stand in to other entities. In this
instance, it is in fact a string of eleven characters, as indicated by
the quotation marks.

> There is the relation-instance (individual, or n-ary property).
> There is BS, the human being and ontologist.
> There is BS-as-being-the-second-argument-in-R, or
> BS-as-being-the-referent-in-R.

Again, these are just longer ways of saying, there is BS.

> I think Barry would like to call "Barry
> Smith"-as-being-the-first-argument-in-R the name. You (and me, and
> most people) would call the string (type) "Barry Smith" the name.

But these are the same thing. They are in fact *necessarily* the same
thing, if we are here talking within a normal logical framework of
denotation and reference. (And if we are not, then we have larger work
to do in order to discover what is appropriate, including re-building
formal logic.)

> I
> think there is no problem, we have ontologies so that we can specify
> what we mean: either the string type is called "name", or the role it
> plays within a specific relation instance (R).

Be careful. There is the role itself, a kind of relation, and there is
the thing - in this case, the name - which stands in that role. My
point, above, is that a thing does not change its actual nature - does
not become something different - when it stands in a role. So the
string-as-first-argument (or, for that matter, the string-as-a-
decoration-on-my-aunties-hat) is still a string, and indeed is still
the very same string, as it was when you think of it as a string
simpliciteur.

Apparently we do agree, after all. This is exactly what I have been
saying: it is the SAME number.

> There is a binary has-SS relation, with role-types SS-Number and SS-
> Bearer
> (for lack of better name), and similarily has-CC with roles CC-Number
> and CC-Bearer. The same number plays different roles in different
> relations

Exactly. The SAME number. So, we have numbers, some of which can stand
in a role or relationship to other things, like banks and people and
credit cards. But our universe of discourse does not have to contain
CC numbers *as well as* non-CC, ordinary, mathematical numbers. It
does not have various *kinds* of number in it; it just has numbers,
some of which are used for other purposes than arithmetic.

> : has-SS(123,X) and has-CC(123,X), with different roles for
> 123 (assuming it is the number).
>
>>> A numeral is something that can be written on an envelope. A
>>> name (in the sense here at issue) is more like a number, than
>>> like a numeral.
> PH> So names cannot be written on envelopes? I really, really find
> PH> this an extraordinary idea, one I have never met before. (I
> PH> presume that SS numbers also cannot be written, on similar
> PH> grounds?)
>
> Names can be written on envelopes, if by "name" you mean the
> string. BS seems to be talking about the role the string plays in a
> certain relation (that you establish by writing the string on the
> envelope), and the role itself cannot be written down (roles are
> similar to properties).

Right, so a BS-name cannot be written down. To repeat, I find this an
extraordinary idea. To me it is close to being a refutation of the
argument that names are not character strings.

Pat

>
> Rob.

[Robert Hoehndorf]

>>>>> "PH" == Pat Hayes <pha...@ihmc.us> writes:

Hi,

>> Let me try to give a non-BFO answer to the challenge. Let
>> there be a naming-relation den(S,E), saying that the
>> string-token S names (or denotes) the entity E. So, on a letter
>> you send to BS, the string-token "Barry Smith" on the letter
>> will name BS, the ontologist. You observe, correctly, that the
>> same string type (and maybe token) can refer to different
>> entities.

PH> In which case, we have an immediate problem, since by writing
PH> den(S,E) you have effectively already denied this
PH> possibility. A name can denote only one thing (in a particular
PH> interpretation). There is only one denotation relationship. (I
PH> am assuming that we are talking within the usual semantic
PH> framework of formalized notations such as FO logic.)

I am not sure if I understand how this refers to FOL semantics. With
"den", I intended to refer to an ontological relation, which will
appear in the signature of a FOL theory and for which we have to
develop axioms. We can write axioms requiring that one string-token
may denote at most one entity, but this seems to be wrong, as shown in
a number of examples so far.

The "den" I meant does not refer to the relation mapping variable and
term symbols to elements in the universe of the model. So I do not
think there is a problem with FOL semantics here.

>> In one relation (instance) of the type den, such as den("Barry
>> Smith",BS), the token "Barry Smith" will refer to one entity,
>> to BS the ontologist. The same token "Barry Smith" may refer to
>> another entity in a different context (i.e., in a different
>> relation), but in the relation-instance above, the string-token
>> "Barry Smith" refers to only one entity.

PH> I do not know what you mean here by "context". Unfortunately,

The context is the individual relation (in BFO, read "n-ary property"
instead of relation).

PH> the C- word has been used to mean virtually anything, so it is
PH> itself, ironically, one of the most context-sensitive words
PH> ever used in technical discourse. But if you mean that each
PH> relation imposes or defines its own context, so that names
PH> (strings) can change their denotation depending upon the
PH> relation they are said to stand in, then I must object, since
PH> that would make all formalization impossible.

Please explain why this makes formalization impossible.
Some questions to see if we agree:
Do you agree that strings can denote entities?
Do you agree that some string denote entities?
Can one string denote two different entities?

>> Now, take the relation-instance (let's call it R) apart. There
>> is the string "Barry Smith", what we may call the name.

PH> Right.

>> There is "Barry Smith"-as-being-the-first-argument-in-R, or
>> "Barry Smith"-as-being-the-name-in-R.

PH> ? No, there is "Barry Smith", the name. It is what it is, no
PH> matter what relationships it may stand in to other
PH> entities. In this instance, it is in fact a string of eleven
PH> characters, as indicated by the quotation marks.

Yes, the string is the string, it does not change what it is. However,
there may be another entity, let us call it a qua-individual (taken from
DOLCE) or relational role (from GFO), which is "Barry
Smith"-qua-name-in-R. In particular, the role the string "Barry Smith"
plays within a particular denotation relation gives rise to this
qua-individual or role. Similarily, there is BS-the-ontologist, and
the qua-individual BS-the-ontologist-qua-referent-in-R.

This assumes that relations (not the relations in FOL, but the
real-world relations that are instances of some kind of "Relation"
class in an ontology) give rise to different modes of participation: a
part-of relation holds between a /part/ and a /whole/, and an entity
can participate in one part-of relation qua /part/, in another qua
/whole/. The "context" (or distinguishing feature) is the relation (an
individual), to which both qua-entities belong (and depend on).

>> There is the relation-instance (individual, or n-ary property).
>> There is BS, the human being and ontologist. There is
>> BS-as-being-the-second-argument-in-R, or
>> BS-as-being-the-referent-in-R.

PH> Again, these are just longer ways of saying, there is BS.

No. There are more things needed here than just the entity BS: there
are qua-entities within a relation, BS-qua-referent-in-R. This entity
is different from BS, it is dependent (both on BS and on R!), while BS
is independent.

>> I think Barry would like to call "Barry
>> Smith"-as-being-the-first-argument-in-R the name. You (and me,
>> and most people) would call the string (type) "Barry Smith" the
>> name.

PH> But these are the same thing. They are in fact *necessarily*
PH> the same thing, if we are here talking within a normal logical
PH> framework of denotation and reference. (And if we are not,
PH> then we have larger work to do in order to discover what is
PH> appropriate, including re-building formal logic.)

What is a "normal logical framework of denotation and reference"? I am
talking about an ontological denotation relation for which we can give
axioms in FOL, but which is not a part of the logic itself. So I think
we are talking about different things here.

>> I think there is no problem, we have ontologies so that we can
>> specify what we mean: either the string type is called "name",
>> or the role it plays within a specific relation instance (R).

PH> Be careful. There is the role itself, a kind of relation, and
PH> there is the thing - in this case, the name - which stands in
PH> that role. My point, above, is that a thing does not change
PH> its actual nature - does not become something different - when
PH> it stands in a role. So the string-as-first-argument (or, for

We agree completely so far.

PH> that matter, the string-as-a- decoration-on-my-aunties-hat) is
PH> still a string, and indeed is still the very same string, as
PH> it was when you think of it as a string simpliciteur.

Here I disagree: you are saying that the thing (the string) and the
role it plays are the same thing. But this cannot be so: the string
(token) is an independent entity, but the role it plays within a
relation is both existentially dependent on the player (the string)
and the relation (of the denotation type) in which it plays the
role. This qua-entity or role has different properties than the
string: the role is dependent, the role has no shape, the role has no
length, etc.

PH> Well, to speak strictly, you never actually 'have' a
PH> number. To 'have' an SS number is bear a certain relationship
PH> to a number, one which comes with a lot of social baggage in
PH> the USA. Similarly for CC number, and banks. The relationships
PH> are very different, of course. But I would indeed say that in
PH> this strange coincidence case that you bear these
PH> relationships to the same number, yes.
>> Here I disagree: you bear different relations to the same
>> number, and the same number plays different roles (role-types)
>> in these relations.

PH> Apparently we do agree, after all. This is exactly what I have
PH> been saying: it is the SAME number.

>> There is a binary has-SS relation, with role-types SS-Number
>> and SS- Bearer (for lack of better name), and similarily has-CC
>> with roles CC-Number and CC-Bearer. The same number plays
>> different roles in different relations

PH> Exactly. The SAME number. So, we have numbers, some of which
PH> can stand in a role or relationship to other things, like
PH> banks and people and credit cards. But our universe of
PH> discourse does not have to contain CC numbers *as well as*
PH> non-CC, ordinary, mathematical numbers. It does not have
PH> various *kinds* of number in it; it just has numbers, some of
PH> which are used for other purposes than arithmetic.

Of course. And we can use roles (or qua-individuals) to distinguish
how the numbers are used within different relations.

>> Names can be written on envelopes, if by "name" you mean the
>> string. BS seems to be talking about the role the string plays
>> in a certain relation (that you establish by writing the string
>> on the envelope), and the role itself cannot be written down
>> (roles are similar to properties).

PH> Right, so a BS-name cannot be written down. To repeat, I find
PH> this an extraordinary idea. To me it is close to being a
PH> refutation of the argument that names are not character
PH> strings.

If we agree that there is both the string and the role/qua-individual
that the string plays in a denotation-relation, I do not see a
problem: you want to call the string "name", while Barry would like
to call the role/qua-individual "name". Anyway there are two different
ontological categories for which different axioms hold, and how they
are named seems secondary, in my opinion.

The important task that I see ahead is to get the distinguishing
features (axioms) for the two categories right, together with axioms
for the denotation relation. It may help to shed some light on the
argument we are having here.

Rob.

[Pat Hayes]

Another thought:

On Nov 20, 2009, at 11:23 AM, Barry Smith wrote:

> At 11:08 AM 11/20/2009, Pat Hayes wrote:
>
>> On Nov 20, 2009, at 7:45 AM, Barry Smith wrote:
>>
>>> At 01:09 AM 11/20/2009, Pat Hayes wrote:
>>

....

>> . Maybe we could say that a numeral *on a
>> credit card* denotes something other than a simple number. That might
>> work.
>
> Finally, light.
>
>> We would need to distinguish 'credit card contexts' or something
>> like that, which systematically re-interpret numerals to denote
>> special 'credit card numbers', these being different from ordinary
>> numbers in some special financial way.
>
> Indeed.

....

> At no stage did I suggest that credit card numbers denote anything.

Well, I did that (above) and you, apparently, agreed. My 'thought',
above, was that a numeral, when printed or embossed on a credit card
(or typed into 'write card number' box on a web page, etc.) , should
be understood as *denoting* something other than a natural number,
something called a 'credit card number', which would be a different
category of entity. So, is this also what your notion amounts to, that
the CC context changes the **denotation** of the string of numerals?
Because this still leaves the actual numerals embossed on the card as
being a character string, indeed a numeral. Or do you want to say that
**the actual symbols on the card** are not numerals or characters any
more, by virtue of their credit-card-ish location? That would be like
treating them as being written in a different language. We would, I
think, need to distinguish the symbol "3" considered as a numeral
denoting three, from "3" the symbol used as part of a credit card
'number' (scare quotes because we should say 'numeral', strictly),
these having the same kind of accidental-same-shape relation as
uppercase C in the English alphabet and the Cyrillic syllabic 's'
character. Perhaps there should be a Unicode block reserved for credit
card symbols.

I have to say that either of these possibilities seem to me to be
baroque and unnecessary. It is much simpler and closer to normal
informal usage and just common sense, to simply say that strings are
strings (wherever they occur or however they are used) and that
numerical strings are numerals (even if they are not being used for
arithmetic or 'numbering' purposes.)

Pat

[Darren Natale]

Hi Bjoern,

I like the definition of CRID, especially as it speaks to the point that
an identifier cannot uniquely identify something without knowing the
system within which it works. My reading of the CRID definition is that
it does imply an identifier points to a unique entity. Is this correct?

[Bjoern Peters]

Hi Darren,

I can see three possible senses of 'unique'. My thoughts:

- the same CRID cannot be assigned twice. So there is a unique entity that it is associated with it during assignment

- that entity can be a set of things. For example, a CRID can point to a batch of antibodies produced that is split up into multiple aliquots which have one batch number.

- it is possible that the same entity gets two CRIDs assigned. In many CRID systems, it is desired to avoid this, and it is part of the plan specification not to assign a new CRID to an entity that has already been assigned one. At the same time, many CRID system realize that duplication errors happen quite a bit, and have explicit ways of 'merging' two CRIDs that are found to be duplicates (e.g. most issue tracking system; we even found several duplicate pubmed IDs for the same journal article; Barry's RTS also mentions this specifically). As in OBI the plan specification is always right, we can distinguish CRID systems that guarantee single IDs per entity, those that have formal processes to remove them, and those that don't care.

Hope this helps.

Best,

- Bjoern

[Darren Natale]

I already assumed the second point, so agreed there. In fact, I agree
with everything you wrote. I see no problem in having two CRIDs
assigned to a single entity, only in having two entities assigned to the
same CRID.

[Pat Hayes]

On Nov 22, 2009, at 12:14 PM, Robert Hoehndorf wrote:

>>>>>> "PH" == Pat Hayes <pha...@ihmc.us> writes:
>
> Hi,
>
>>> Let me try to give a non-BFO answer to the challenge. Let
>>> there be a naming-relation den(S,E), saying that the
>>> string-token S names (or denotes) the entity E. So, on a letter
>>> you send to BS, the string-token "Barry Smith" on the letter
>>> will name BS, the ontologist. You observe, correctly, that the
>>> same string type (and maybe token) can refer to different
>>> entities.
> PH> In which case, we have an immediate problem, since by writing
> PH> den(S,E) you have effectively already denied this
> PH> possibility. A name can denote only one thing (in a particular
> PH> interpretation). There is only one denotation relationship. (I
> PH> am assuming that we are talking within the usual semantic
> PH> framework of formalized notations such as FO logic.)
>
> I am not sure if I understand how this refers to FOL semantics. With
> "den", I intended to refer to an ontological relation, which will
> appear in the signature of a FOL theory and for which we have to
> develop axioms.

Why do you call this 'denotation', then? In using FOL we are already
assuming a denotation relationship. I took you to be referring to
that. BUt in any case, my point stands. If you want your 'den' to be
sensitive to context in some way, then you should have given it more
arguments.

> We can write axioms requiring that one string-token
> may denote at most one entity, but this seems to be wrong, as shown in
> a number of examples so far.

That one name denotes one entity is built into the logic. If we are
talking about denotation in English then the situation is more
complicated, agreed. My own ontology-writing instincts suggest that in
cases like this, the best way to approach the matter is to try to
elucidate the necessary complications explicitly, rather than tinker
with the underlying logic. So, if names denote differently in various
"contexts", then let us set out to identify these contexts and
axiomatize their properties. But then we should begin by writing
den(S,C,E) rather than den(S, E), giving an explicit formal hook for
the relevant context C.

>
> The "den" I meant does not refer to the relation mapping variable and
> term symbols to elements in the universe of the model.

Oh, I see. Or rather, I stand corrected. But I do not quite see what
is going on here. You wrote "den(S,E)" which looks to me like a formal
assertion, and I was assuming that it was in (some variety of) formal
logic or mathematical notation, and so was intended to be interpreted
using the usual semantic conventions for such notations, which
typically involve denotation. (Right?) And under those, there is no
provision for 'contextually' changing the denotation of whatever
expression you substitute for 'S' in this formula.

> So I do not
> think there is a problem with FOL semantics here.

I think there is, see above.

>
>>> In one relation (instance) of the type den, such as den("Barry
>>> Smith",BS), the token "Barry Smith" will refer to one entity,
>>> to BS the ontologist. The same token "Barry Smith" may refer to
>>> another entity in a different context (i.e., in a different
>>> relation), but in the relation-instance above, the string-token
>>> "Barry Smith" refers to only one entity.
> PH> I do not know what you mean here by "context". Unfortunately,
>
> The context is the individual relation (in BFO, read "n-ary property"
> instead of relation).

Hmm. I do not follow in what sense a relation can be said to be a
context. From your reasoning in this example, it would appear that you
intend a name occurring in an argument position of a relational
sentence to be considered to be a "context" defined by that use, so
possibly denoting something different from the same name occurring in
a different relational sentence. Is that more or less correct?
(Subsequent responses assume that it is.)

>
> PH> the C- word has been used to mean virtually anything, so it is
> PH> itself, ironically, one of the most context-sensitive words
> PH> ever used in technical discourse. But if you mean that each
> PH> relation imposes or defines its own context, so that names
> PH> (strings) can change their denotation depending upon the
> PH> relation they are said to stand in, then I must object, since
> PH> that would make all formalization impossible.
>
> Please explain why this makes formalization impossible.

Because if the same name can mean something different every time it
occurs in a relational term (being in a different "context") , then in
effect, all name tokens are different names from one another. This
renders all the usual quantificational inference rules immediately
invalid, and also makes equality meaningless. We would be left with a
purely propositional logic. For example, it would not be the case that

P[a] (an expression containing one or more occurrences of the name 'a')
entails
(exists (x) P[x/a] )

> Some questions to see if we agree:
> Do you agree that strings can denote entities?

When used as names, yes.

> Do you agree that some string denote entities?

Yes

> Can one string denote two different entities?

Not when the string is used as a name, no. That is, a name denotes one
entity (in a given interpretation.)

>
>>> Now, take the relation-instance (let's call it R) apart. There
>>> is the string "Barry Smith", what we may call the name.
> PH> Right.
>
>>> There is "Barry Smith"-as-being-the-first-argument-in-R, or
>>> "Barry Smith"-as-being-the-name-in-R.
> PH> ? No, there is "Barry Smith", the name. It is what it is, no
> PH> matter what relationships it may stand in to other
> PH> entities. In this instance, it is in fact a string of eleven
> PH> characters, as indicated by the quotation marks.
>
> Yes, the string is the string, it does not change what it is. However,
> there may be another entity, let us call it a qua-individual (taken
> from
> DOLCE) or relational role (from GFO), which is "Barry
> Smith"-qua-name-in-R.

Of course there may be many entities. But you are here **quoting** the
string, and by the usual conventions, this means that you are
referring, unambiguously, to the actual character string. Call this
string FOO, for now: FOO = "Barry Smith", a string of eleven
characters. Now, you seem to also be referring to FOO-qua-name-in-R,
and I do not know how to interpret this. What does it mean for a
character string to be understood 'qua' anything? If I send this
character string across a computer network, it is still the same
character string. Can I send a role across a computer network?

> In particular, the role the string "Barry Smith"
> plays within a particular denotation relation gives rise to this
> qua-individual or role.

A role is simply another relationship (between a fact or trope or
proposition or whatever one wants to call them, and one of the things
it is about.) The string FOO is what it is, and is denoted by the
quoted expression used above. The various relationships it stands in
have no bearing on its nature or identity. Now, you may want to say
that the quoted expression, repeated displayed on the next line:

"Barry Smith"

should be understood as denoting not a string but something else, when
used in certain relational contexts. To this proposal I would make the
same objections I give above against allowing names to change their
denotation according to how they are used, but with the added force
that the quotation convention itself is a valuable part of a practical
logic, so yet more of the inferential machinery would be wrecked by
this proposal. And, to go back to the original theme, there is
absolutely no need to do this, and it would achieve no extra clarity
or insight.

> Similarily, there is BS-the-ontologist, and
> the qua-individual BS-the-ontologist-qua-referent-in-R.
>
> This assumes that relations (not the relations in FOL, but the
> real-world relations that are instances of some kind of "Relation"
> class in an ontology

?? What?? Why would one not use FOL relation symbols to denote actual
relations? This seems crazy. FOL provides us with a powerful,
expressive framework for writing ontologies. In a very real sense it
is as expressive as any formalism can possibly be. Why not use it for
what it was designed to be used for?

> ) give rise to different modes of participation: a
> part-of relation holds between a /part/ and a /whole/, and an entity
> can participate in one part-of relation qua /part/, in another qua
> /whole/. The "context" (or distinguishing feature) is the relation (an
> individual), to which both qua-entities belong (and depend on).

So every assertion of a relation holding between entities is a
different "context"? See above for my response to this notion.

>
>>> There is the relation-instance (individual, or n-ary property).
>>> There is BS, the human being and ontologist. There is
>>> BS-as-being-the-second-argument-in-R, or
>>> BS-as-being-the-referent-in-R.
> PH> Again, these are just longer ways of saying, there is BS.
>
> No. There are more things needed here than just the entity BS: there
> are qua-entities within a relation, BS-qua-referent-in-R. This entity
> is different from BS, it is dependent (both on BS and on R!), while BS
> is independent.

I disagree. Multiplying entities without cause is the worst sin in
ontology design. I hereby assert that (1) there is no actual NEED to
do any of this, and (2) that the 'qua' usage is ontologically
meaningless. A foo is a foo. A foo qua baz is also a foo, And of
course, I invite you to prove me wrong :-).

>
>>> I think Barry would like to call "Barry
>>> Smith"-as-being-the-first-argument-in-R the name. You (and me,
>>> and most people) would call the string (type) "Barry Smith" the
>>> name.
> PH> But these are the same thing. They are in fact *necessarily*
> PH> the same thing, if we are here talking within a normal logical
> PH> framework of denotation and reference. (And if we are not,
> PH> then we have larger work to do in order to discover what is
> PH> appropriate, including re-building formal logic.)
>
> What is a "normal logical framework of denotation and reference"? I am
> talking about an ontological denotation relation for which we can give
> axioms in FOL, but which is not a part of the logic itself. So I think
> we are talking about different things here.

I am talking about an ontology, written in FOL or some subset of FOL.
You appear to be talking about an ontology written in a different
formalism altogether, one which itself is described in FOL axioms used
as a metatheory (but which does not satisfy FOL inference rules, for
example.) It may be related to some of the 'context logics' developed
in AI/KR work during the last decade or so.(?)

>
>>> I think there is no problem, we have ontologies so that we can
>>> specify what we mean: either the string type is called "name",
>>> or the role it plays within a specific relation instance (R).
> PH> Be careful. There is the role itself, a kind of relation, and
> PH> there is the thing - in this case, the name - which stands in
> PH> that role. My point, above, is that a thing does not change
> PH> its actual nature - does not become something different - when
> PH> it stands in a role. So the string-as-first-argument (or, for
>
> We agree completely so far.
>
> PH> that matter, the string-as-a- decoration-on-my-aunties-hat) is
> PH> still a string, and indeed is still the very same string, as
> PH> it was when you think of it as a string simpliciteur.
>
> Here I disagree: you are saying that the thing (the string) and the
> role it plays are the same thing. But this cannot be so: the string
> (token) is an independent entity, but the role it plays within a
> relation is both existentially dependent on the player (the string)
> and the relation (of the denotation type) in which it plays the
> role. This qua-entity or role has different properties than the
> string: the role is dependent, the role has no shape, the role has no
> length, etc.

I simply do not believe in these things you call 'entities-in-a-role',
if they are intended to be separate entities in the universe. My
ontology, it would seem, is much sparser than yours. I consider this
to be an advantage, but you may of course very well not agree, this
being an essentially aesthetic judgement.

Not with your assumptions, since there is then no guarantee that a
numeral shall even denote a number. Apparently the same numeral, say
"547", can denote all kinds of things: in one place, a natural number;
in another, an SS number; in another, a soldier; in another, a house
in a street, and so on. So we cannot even know when we are referring
to a number. In the worst case (which I fear you will slide into
fairly quickly), *every* occurrence of *every* name will denote
something distinct from every other occurrence, since it will be
understood as 'qua' some context unique to that token.

>>> Names can be written on envelopes, if by "name" you mean the
>>> string. BS seems to be talking about the role the string plays
>>> in a certain relation (that you establish by writing the string
>>> on the envelope), and the role itself cannot be written down
>>> (roles are similar to properties).
> PH> Right, so a BS-name cannot be written down. To repeat, I find
> PH> this an extraordinary idea. To me it is close to being a
> PH> refutation of the argument that names are not character
> PH> strings.
>
> If we agree that there is both the string and the role/qua-individual
> that the string plays in a denotation-relation

I do not agree with this, no. Or at any rate, only if we understand
'role' to mean 'relation of a particular kind'. The relation certainly
exists, but it does not change the nature of the entities that it
relates. Put another way: the A-qua-B construction is meaningless (or
trivial: Nec(A qua B = A) ), in my view.

> , I do not see a
> problem: you want to call the string "name", while Barry would like
> to call the role/qua-individual "name". Anyway there are two different
> ontological categories for which different axioms hold, and how they
> are named seems secondary, in my opinion.
>
> The important task that I see ahead is to get the distinguishing
> features (axioms) for the two categories right, together with axioms
> for the denotation relation. It may help to shed some light on the
> argument we are having here.

I am afraid that I see this 'task' as completely pointless, since
there is nothing to distinguish.

[Robert Hoehndorf]

>>>>> "PH" == Pat Hayes <pha...@ihmc.us> writes:

Hi,

>> I am not sure if I understand how this refers to FOL
>> semantics. With "den", I intended to refer to an ontological
>> relation, which will appear in the signature of a FOL theory
>> and for which we have to develop axioms.

PH> Why do you call this 'denotation', then? In using FOL we are
PH> already assuming a denotation relationship. I took you to be
PH> referring to that. BUt in any case, my point stands. If you
PH> want your 'den' to be sensitive to context in some way, then
PH> you should have given it more arguments.

Bad choice of words on my side. Let's call the relation "name",
then. A string-token names some entity. This may be more adequate for
the discussion of identifiers here.

>> We can write axioms requiring that one string-token may denote
>> at most one entity, but this seems to be wrong, as shown in a
>> number of examples so far.

PH> That one name denotes one entity is built into the logic. If
PH> we are talking about denotation in English then the situation
PH> is more complicated, agreed. My own ontology-writing instincts
PH> suggest that in cases like this, the best way to approach the
PH> matter is to try to elucidate the necessary complications
PH> explicitly, rather than tinker with the underlying logic. So,

Nobody wants to tinker with the logic (I hope). I am assuming
classical predicate logics here.

PH> if names denote differently in various "contexts", then let us
PH> set out to identify these contexts and axiomatize their
PH> properties. But then we should begin by writing den(S,C,E)
PH> rather than den(S, E), giving an explicit formal hook for the
PH> relevant context C.

Ok, I think I was a bit quick with writing down the example. After a
bit more thought, I think I would like a setting like this (assuming
FOL, and calling the relation "name" instead of "den", and ignoring
time for instantiation):

instanceOf(Relation, Category).
instanceOf(Name, Category).
is-a(Name,Relation).
instanceOf(R,Name).
instanceOf("Barry Smith",String).
instanceOf(BS,Human).
is-a(NameR, Role).
is-a(RefR, Role).
instanceOf(R1,Name).
instanceOf(R2,RefR).
plays("Barry Smith",R1).
plays(BS,R2).
role-in(R1,R).
role-in(R2,R).

Plus appropriate axioms for the primitive relations.

>> The "den" I meant does not refer to the relation mapping
>> variable and term symbols to elements in the universe of the
>> model.

PH> Oh, I see. Or rather, I stand corrected. But I do not quite
PH> see what is going on here. You wrote "den(S,E)" which looks to
PH> me like a formal assertion, and I was assuming that it was in
PH> (some variety of) formal logic or mathematical notation, and
PH> so was intended to be interpreted using the usual semantic
PH> conventions for such notations, which typically involve
PH> denotation. (Right?) And under those, there is no provision
PH> for 'contextually' changing the denotation of whatever
PH> expression you substitute for 'S' in this formula.

Again, my fault for a badly chosen name. I did not intend to suggest
changes to FOL semantics.

>>>> In one relation (instance) of the type den, such as
>>>> den("Barry Smith",BS), the token "Barry Smith" will refer to
>>>> one entity, to BS the ontologist. The same token "Barry
>>>> Smith" may refer to another entity in a different context
>>>> (i.e., in a different relation), but in the relation-instance
>>>> above, the string-token "Barry Smith" refers to only one
>>>> entity.
PH> I do not know what you mean here by "context". Unfortunately,
>> The context is the individual relation (in BFO, read "n-ary
>> property" instead of relation).

PH> Hmm. I do not follow in what sense a relation can be said to
PH> be a context. From your reasoning in this example, it would
PH> appear that you intend a name occurring in an argument
PH> position of a relational sentence to be considered to be a
PH> "context" defined by that use, so possibly denoting something
PH> different from the same name occurring in a different
PH> relational sentence. Is that more or less correct?
PH> (Subsequent responses assume that it is.)

I assume a "relation" to be a real-world entity, "the glue that holds
things together" in Barwise' words. They are not equivalent to
relational sentences in any formalism. I believe that relations are
entities in their own right, and they are quite different from what is
called a "relation" in FOL (in particular, they are not sets). A
relation is something we can encounter in the world and describe in
our ontology (the PATO contains, as part, an ontology of relations,
modelled as n-ary properties).
To answer your question, a name in a relational sentence will always
denote the same thing, and when it occurs in a different sentence it
still denotes the same thing. Similarily, an entity in the world can
take part in many different relations
(glue-that-holds-things-together-relations) and still be the same
thing.

PH> (strings) can change their denotation depending upon the
PH> relation they are said to stand in, then I must object, since
PH> that would make all formalization impossible.
>> Please explain why this makes formalization impossible.

PH> Because if the same name can mean something different every
PH> time it occurs in a relational term (being in a different
PH> "context") , then in effect, all name tokens are different
PH> names from one another. This renders all the usual
PH> quantificational inference rules immediately invalid, and also
PH> makes equality meaningless. We would be left with a purely

Replace "denotation" by "name" (or any other relation name). I meant
to talk about a relation in the signature of a FOL theory and axioms
for this relation, not any changes to semantics.

>> Some questions to see if we agree: Do you agree that strings
>> can denote entities?

PH> When used as names, yes.

Interesting. So a string-as-string (or string-qua-string) cannot be
used to denote an entity, but a string-qua-name can? What is the
difference between those if you do not admit qua-entities?

>> Do you agree that some string denote entities?

PH> Yes

>> Can one string denote two different entities?

PH> Not when the string is used as a name, no. That is, a name
PH> denotes one entity (in a given interpretation.)

Ok, but what about the world, not in logical semantics? I thought we
are trying to analyze an ontological relation that is used in reality
by people and machines, not a formal relation in FOL semantics. It
seems that in real world, one and the same string may name different
entities?

PH> ? No, there is "Barry Smith", the name. It is what it is, no
PH> matter what relationships it may stand in to other
PH> entities. In this instance, it is in fact a string of eleven
PH> characters, as indicated by the quotation marks.
>> Yes, the string is the string, it does not change what it
>> is. However, there may be another entity, let us call it a
>> qua-individual (taken from DOLCE) or relational role (from
>> GFO), which is "Barry Smith"-qua-name-in-R.

PH> Of course there may be many entities. But you are here
PH> **quoting** the string, and by the usual conventions, this
PH> means that you are referring, unambiguously, to the actual
PH> character string. Call this string FOO, for now: FOO = "Barry
PH> Smith", a string of eleven characters. Now, you seem to also
PH> be referring to FOO-qua-name-in-R, and I do not know how to
PH> interpret this. What does it mean for a character string to be
PH> understood 'qua' anything? If I send this character string
PH> across a computer network, it is still the same character
PH> string. Can I send a role across a computer network?

A string is just a string, and no matter in what relations it stands,
it will not change. But it gains new properties by standing in
particular relations.

>> In particular, the role the string "Barry Smith" plays within a
>> particular denotation relation gives rise to this
>> qua-individual or role.

PH> A role is simply another relationship (between a fact or trope
PH> or proposition or whatever one wants to call them, and one of
PH> the things it is about.) The string FOO is what it is, and is
PH> denoted by the quoted expression used above. The various
PH> relationships it stands in have no bearing on its nature or
PH> identity. Now, you may want to say that the quoted expression,
PH> repeated displayed on the next line:
PH> "Barry Smith"
PH> should be understood as denoting not a string but something
PH> else, when used in certain relational contexts. To this
PH> proposal I would make the same objections I give above against
PH> allowing names to change their denotation according to how
PH> they are used, but with the added force that the quotation
PH> convention itself is a valuable part of a practical logic, so
PH> yet more of the inferential machinery would be wrecked by this
PH> proposal. And, to go back to the original theme, there is
PH> absolutely no need to do this, and it would achieve no extra
PH> clarity or insight.

>> Similarily, there is BS-the-ontologist, and the qua-individual
>> BS-the-ontologist-qua-referent-in-R.
>> This assumes that relations (not the relations in FOL, but the
>> real-world relations that are instances of some kind of
>> "Relation" class in an ontology

PH> ?? What?? Why would one not use FOL relation symbols to denote
PH> actual relations? This seems crazy. FOL provides us with a
PH> powerful, expressive framework for writing ontologies. In a
PH> very real sense it is as expressive as any formalism can
PH> possibly be. Why not use it for what it was designed to be
PH> used for?

Because the ontologies of FOL and the ontologies we want to build do
not match. It seems to me that you want to use FOL /as/ an ontology,
by assuming that relations-in-the-world correspond to FOL-relations,
that predication corresponds to instantiation, etc. But it does not
work with the ontologies we are building, and certainly not with BFO
or DOLCE (already because instantiation is time-indexed and
ternary). Instead, we use FOL for what it is built for: make the
properties and interrelations of the primitive predicates in our
ontology precise through axioms; and among the primitives, there seem
to be things like instance-of(x,y), is-a(x,y), Universal(x),
Particular(x), Relation(x), etc. If you use FOL /as/ the ontology,
your ontology will be rather sparse, and quite far detached from
the real world, I fear.

>> ) give rise to different modes of participation: a part-of
>> relation holds between a /part/ and a /whole/, and an entity
>> can participate in one part-of relation qua /part/, in another
>> qua /whole/. The "context" (or distinguishing feature) is the
>> relation (an individual), to which both qua-entities belong
>> (and depend on).

PH> So every assertion of a relation holding between entities is a
PH> different "context"? See above for my response to this notion.

Not assertion, no. The context is the relation, the thing in the
world.

>>>> There is the relation-instance (individual, or n-ary
>>>> property). There is BS, the human being and ontologist.
>>>> There is BS-as-being-the-second-argument-in-R, or
>>>> BS-as-being-the-referent-in-R.
PH> Again, these are just longer ways of saying, there is BS.
>> No. There are more things needed here than just the entity BS:
>> there are qua-entities within a relation,
>> BS-qua-referent-in-R. This entity is different from BS, it is
>> dependent (both on BS and on R!), while BS is independent.

PH> I disagree. Multiplying entities without cause is the worst
PH> sin in ontology design. I hereby assert that (1) there is no
PH> actual NEED to do any of this, and (2) that the 'qua' usage is
PH> ontologically meaningless. A foo is a foo. A foo qua baz is
PH> also a foo, And of course, I invite you to prove me wrong :-).

Ok, I give the standard example:
1. Airline A transports 1000 passengers a day:
transport(A,P1). Passenger(P1).
transport(A,P2). Passenger(P2).
,,,
transport(A,P1000). Passenger(P1000).

2. Every passenger is a human: forall X(Passenger(X) -> Human(X)).

Therefore:
3. Airline X transports 1000 humans a day. (clearly false, as soon as one
human travels twice a day with A).

Now, you say that a human-qua-passenger (second argument in the
"transport" relation) is-a human, so when you count passengers, you
count humans and (2) is true. How do you solve the puzzle without
qua-entities or another form of individual roles? In particular, what
do you count when you do not admit roles/qua-entities?

>> What is a "normal logical framework of denotation and
>> reference"? I am talking about an ontological denotation
>> relation for which we can give axioms in FOL, but which is not
>> a part of the logic itself. So I think we are talking about
>> different things here.

PH> I am talking about an ontology, written in FOL or some subset
PH> of FOL. You appear to be talking about an ontology written in
PH> a different formalism altogether, one which itself is
PH> described in FOL axioms used as a metatheory (but which does
PH> not satisfy FOL inference rules, for example.) It may be
PH> related to some of the 'context logics' developed in AI/KR
PH> work during the last decade or so.(?)

Not at all, I am talking about classical FOL (or higher order, but
predicate logics) theories in which we write the axioms for our
ontology. But I object to identifying FOL with the ontology, or even
parts of the ontology (like instantiation).

>> it plays within a relation is both existentially dependent on
>> the player (the string) and the relation (of the denotation
>> type) in which it plays the role. This qua-entity or role has
>> different properties than the string: the role is dependent,
>> the role has no shape, the role has no length, etc.

PH> I simply do not believe in these things you call
PH> 'entities-in-a-role', if they are intended to be separate
PH> entities in the universe. My ontology, it would seem, is much
PH> sparser than yours. I consider this to be an advantage, but
PH> you may of course very well not agree, this being an
PH> essentially aesthetic judgement.

I believe it helps in modelling, e.g., see the counting problem
above. It is not the "entity-in-a-role" thing, it is the role itself
(the individual role, an instance of some Role category). X-qua-Y is
just my clumsy way of referring to them. Qua-entities may have very
different properties from the entities on which they depend.
A string in a naming relation to entity A could be a particularily
good name, because it may convey certain properties about A, while the
same string may be a rather bad name for entity B.
A road from U to V, where U is a valley and V a mountain top, may go
uphill, and the same road, qua road-from-V-to-U, downhill (example
from "Qua Theories". R. Poli, 1998).

>> Of course. And we can use roles (or qua-individuals) to
>> distinguish how the numbers are used within different
>> relations.

PH> Not with your assumptions, since there is then no guarantee
PH> that a numeral shall even denote a number. Apparently the same
PH> numeral, say "547", can denote all kinds of things: in one
PH> place, a natural number; in another, an SS number; in another,
PH> a soldier; in another, a house in a street, and so on. So we

No, you are right, of course. It denotes the number. It can name many
different things.

PH> cannot even know when we are referring to a number. In the
PH> worst case (which I fear you will slide into fairly quickly),
PH> *every* occurrence of *every* name will denote something
PH> distinct from every other occurrence, since it will be
PH> understood as 'qua' some context unique to that token.

I think this won't happen. Although a string could potentially be used
to name a different thing every time, we are pretty good at choosing
names which are rather unique, at least within one domain of
application (including social groups).

R

[Melanie Courtot]

Few more questions:

1. what is the difference between "CRID_refers_to" and using the
"denotes" relation?

denotes =def: denotes is a primitive, instance-level, relation
obtaining between an information content entity and some portion of
reality. Denotation is what happens when someone creates an
information content entity E in order to specifically refer to
something. The only relation between E and the thing is that E can be
used to 'pick out' the thing. This relation connects those two
together. Freedictionary.com sense 3: To signify directly; refer to
specifically

I think we agreed on removing the sentence about "picking out", so the
updated definition would read:
"denotes is a primitive, instance-level, relation obtaining between an
information content entity and some portion of reality. Denotation is
what happens when someone creates an information content entity E in
order to specifically refer to something. This relation connects those
two together. Freedictionary.com sense 3: To signify directly; refer
to specifically "

2. regarding CRID

Current proposal:

CRID =def: an ICE which specifies a CRID registry and allows obtain
information about an entity associated with the CRID.
Example: 'Pubmed ID:12345'. 'ISBN:12345'.
Counter example: '12345'

The more I think about it, the more there seems to be some
"identifying" objective. It would maybe be safer if we were to try and
build upon those, I guess lots of things can be used as identifiers. I
liked the very initial proposal which said identifiers are ICE with a
specific objective (making it a defined class)- and I don't think it
is contradictory with the current proposal?

Cheers,
Melanie

---
Mélanie Courtot
TFL- BCCRC
675 West 10th Avenue
Vancouver, BC
V5Z 1L3, Canada

[Bjoern Peters]

I wanted to avoid the use of 'denote', and make an explicit relation for this purpose, to focus on what we truly understand. The discussion by PH et al is a nice example of what I want to avoid. It may well turn out that this is a sub-relation of denote, but for now I am not sure. Take for example anonymized donor IDs reported in a manuscript. Does the donor id D01 denote an individual human being? Can I 'pick out' that human being?

The only claim made by the 'CRID_refers_to' relation is that during CRID assignment and recorded in the CRID registry, there is a relation between the CRID and the entity information is recorded about.

- Bjoern

[Alan Ruttenberg]

On Thu, Nov 19, 2009 at 11:53 PM, Bjoern Peters <bpe...@liai.org> wrote:
> As OBI development is driving the initial use case of 'identifier', I want to propose a more limited definition of 'centrally registered identifier' (CRID for short in what follows). I believe the approach outlined below will address all that OBI needs. It does not however attempt to cover all ways of identification. I am re-using multiple ideas raised by others on this thread before.

The biggest problem i see in this is the "centrally".
Serial numbers don't need this.
Other kinds of identifiers, such as hashes, don't need this - see
http://en.wikipedia.org/wiki/Distributed_hash_table

I was muddling over Pat's problem with me using "distinguishing" as a
criteria. What I think I meant was that they can be used to determine
sameness and differentness. A serial number is stamped on an
instrument at a factor, it's number is on the invoice. You receive the
instrument and invoice and check that the serial numbers are the same.

Or: You want to count something that moves around (e.g. fish). You tag
them with an identifier and then record which identifiers you see -
not counting times when you see the same identifier twice. I'm sure
there are analogs with various dna bar code experiments - will look.

-Alan

[Adam M. Goldstein]

On Nov 24, 2009, at 12:01 PM, Alan Ruttenberg wrote:

> On Thu, Nov 19, 2009 at 11:53 PM, Bjoern Peters <bpe...@liai.org> wrote:
>> As OBI development is driving the initial use case of 'identifier', I want to propose a more limited definition of 'centrally registered identifier' (CRID for short in what follows). I believe the approach outlined below will address all that OBI needs. It does not however attempt to cover all ways of identification. I am re-using multiple ideas raised by others on this thread before.
>
> The biggest problem i see in this is the "centrally".
> Serial numbers don't need this.
> Other kinds of identifiers, such as hashes, don't need this - see
> http://en.wikipedia.org/wiki/Distributed_hash_table

But I think that the idea was to try to provide an account of a concept with a narrower scope, CRID, in contrast with identifiers considered as a whole? So we are in principle only dealing with centrally registered things?

[Bill Hogan]

To emphasize Alan's point, there is no central registry for universal product codes (now known as global trade item numbers).  That is, manufacturers create new GTINs all the time without registering them (except with their trading partners but that is not central), but in a way that ensures their uniqueness for a particular type of product.

Bill

[Bjoern Peters]

> > The biggest problem i see in this is the "centrally".
> > Serial numbers don't need this.
> > Other kinds of identifiers, such as hashes, don't need this - see
> > http://en.wikipedia.org/wiki/Distributed_hash_table
>
> But I think that the idea was to try to provide an account of a
> concept with a narrower scope, CRID, in contrast with identifiers
> considered as a whole? So we are in principle only dealing with
> centrally registered things?
>
> ------------------
> Adam M. Goldstein PhD, MSLIS

Exactly right, the purpose of focusing on CRIDs was to pick examples for identifiers that will be very easy to model, and cover much (not all as Alan pointed out) of what OBI will need.

> --
> z_calif...@shiftingbalance.org
> http://www.shiftingbalance.org
> --
> http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
> --
> (914) 637-2717 (msg)
> --
> Dept of Philosophy
> Iona College
> 715 North Avenue
> New Rochelle NY 10801

[Pat Hayes]

Hi Robert, thanks for your extended and careful explanations. It seems
clear that you are taking an overall approach to ontology design which
is in almost every way at odds with mine. We are unlikely to agree,
and had probably best stop arguing, and agree to disagree. Axioms that
I write will be incomprehensible to you, and vice versa. Such
disagreements about ontological style are, of course, inevitable. But
in the interests of possible dialog, comments are interjected below.

Pat

On Nov 23, 2009, at 3:41 PM, Robert Hoehndorf wrote:

>>>>>> "PH" == Pat Hayes <pha...@ihmc.us> writes:
>
> Hi,
>
>>> I am not sure if I understand how this refers to FOL
>>> semantics. With "den", I intended to refer to an ontological
>>> relation, which will appear in the signature of a FOL theory
>>> and for which we have to develop axioms.
> PH> Why do you call this 'denotation', then? In using FOL we are
> PH> already assuming a denotation relationship. I took you to be
> PH> referring to that. BUt in any case, my point stands. If you
> PH> want your 'den' to be sensitive to context in some way, then
> PH> you should have given it more arguments.
>
> Bad choice of words on my side. Let's call the relation "name",
> then. A string-token names some entity. This may be more adequate for
> the discussion of identifiers here.

OK, but as I say, my point stands, whatever you call the relationship.
Its first argument is a string, was my original point, and not a
'string-qua-...' entity. Its being a string is not affected by its
role in this or any other relational assertion. And this, I claim, is
indeed a consequence of our using normal FO logic.

However, I have a deeper problem with this strategy. Apparently, you
believe that there are two distinct relationships between names and
the things they name: the denotation relationship on which formal
logic is based, and which we are using when we ourselves wrote axioms,
and this other one, which you call 'name', which is a different
relationship. I have to say, I find this rather extraordinary. Why do
you think these are not the same relationship, or at least the same
kind of relationship? They seem so to me. To be sure, the logical case
is a very simple, basic, stripped-down account of all the various
namings that occur in everyday human life, but it is not different *in
kind*. After all, Tarski's own examples ("Snow is white") were taken
from natural language, and the basic framework has been widely applied
to quite complex "natural" human language.

As an aside, all this is much easier to write in the CLIF syntax for
FOL, which allows unrestricted quantification over relations named by
relation symbols.
Here is a transcription (CLIF uses infix, LISP-style syntax, so that P
applied to A and B is written (P A B) rather than P(A, B) ). I have
used subClass because is-a is also widely used to mean the
instantiation relationship, so is horribly ambiguous. Cyc uses
'genls', with the arguments reversed, for this relationship.)

(Category Relation)
(Category Name)
(subClass Name Relation)
(Name R)
(String "Barry Smith")
(Human BS)
(subClass NameR, Role) ;;is NameR a typo?
(subClass RefR Role)
(Name R1)
(RefR R2)
(plays "Barry Smith" R1)
(plays BS R2)
(role-In R1 R)
(role-in R2 R)

However, for the record, I have absolutely no idea what Role, role-in
or plays are supposed to mean, or what kind of entities R1 and R2 are,
in the above.

> Plus appropriate axioms for the primitive relations.
>
>>> The "den" I meant does not refer to the relation mapping
>>> variable and term symbols to elements in the universe of the
>>> model.
> PH> Oh, I see. Or rather, I stand corrected. But I do not quite
> PH> see what is going on here. You wrote "den(S,E)" which looks to
> PH> me like a formal assertion, and I was assuming that it was in
> PH> (some variety of) formal logic or mathematical notation, and
> PH> so was intended to be interpreted using the usual semantic
> PH> conventions for such notations, which typically involve
> PH> denotation. (Right?) And under those, there is no provision
> PH> for 'contextually' changing the denotation of whatever
> PH> expression you substitute for 'S' in this formula.
>
> Again, my fault for a badly chosen name. I did not intend to suggest
> changes to FOL semantics.

OK.

PATO?

> contains, as part, an ontology of relations,
> modelled as n-ary properties).

Relations are real entities, they are not sets (though they have sets
closely associated with them), and they can be described by an
ontology. Yes to all that; but none of that suggests that they are
"quite different from what is called a "relation" in FOL", or at least
the version of FOL that is embodied in Common Logic. That is in large
part why we developed CL, to be a genuinely ontology-neutral FO
framework that treats relations as first-class entities. In fact, in
CL, there is no logical distinction between individuals and relations:
all relations (and all functions) are first-class entities in the
universe of discourse.

However, I still do not understand how a relation, howsoever 'real
world' it may be, defines a context.

> To answer your question, a name in a relational sentence will always
> denote the same thing, and when it occurs in a different sentence it
> still denotes the same thing.

Well, OK, then what does the construction <name>-qua-<context> mean?
Surely your point has been that a name, when used in one relational
'context', denotes one thing, but when used in another may denote
another. So "Barry Smith" denotes Barry the ontologist in one context
and some other Barry in a different context, or a quoted expression
denotes a mere string in one context but something altogether less
mere in a different one. And if so, then this directly contradicts
what you say, above. And if you have not been saying this, I have
apparently failed to grasp your point.

> Similarily, an entity in the world can
> take part in many different relations
> (glue-that-holds-things-together-relations) and still be the same
> thing.

Indeed.

>
> PH> (strings) can change their denotation depending upon the
> PH> relation they are said to stand in, then I must object, since
> PH> that would make all formalization impossible.
>>> Please explain why this makes formalization impossible.
> PH> Because if the same name can mean something different every
> PH> time it occurs in a relational term (being in a different
> PH> "context") , then in effect, all name tokens are different
> PH> names from one another. This renders all the usual
> PH> quantificational inference rules immediately invalid, and also
> PH> makes equality meaningless. We would be left with a purely
>
> Replace "denotation" by "name" (or any other relation name). I meant
> to talk about a relation in the signature of a FOL theory and axioms
> for this relation, not any changes to semantics.
>
>>> Some questions to see if we agree: Do you agree that strings
>>> can denote entities?
> PH> When used as names, yes.
>
> Interesting. So a string-as-string (or string-qua-string) cannot be
> used to denote an entity, but a string-qua-name can? What is the
> difference between those if you do not admit qua-entities?

I did not say "qua" :-) When *used* as a name, the string denotes
something. But not all strings get used as names, and those that do
not may not denote anything. This is no more mysterious than saying
that a stick may be used as a club, and when it is so used, it hits
something. But every stick need hit anything.

Look, take this example. A stick may be used as a club, to hit
something with. When it is thus used, it may be said to *be* a club.
But this is simply to accord it, the stick, some new status. It is the
very same stick that it was before it was used as a club. Using a
stick as a club does not create a new individual, a stick-qua-club,
distinct from the stick itself and having properties that the stick
does not have. Rather, the correct way to describe this state of
affairs is to say that the stick *is* the club. When I hold the stick
and use it as a club, I do not suddenly hold two things, or cease to
hold the stick. Some changes to things do indeed create new objects,
or change things to the point that it is best to say that a new object
has been created; but these are rare, significant, events which are
typically the subject of much description, e.g. manufacturing
processes, not mere accidental ripples in the causal plenum. In fact,
methodologies such as OntoClean have been largely devoted to keeping
track of such identity-changing events and treating them with the
seriousness that they deserve. It seems to me that your qua-things
arise with much too frequency and with far too little excuse, and that
any universe will become so cluttered with them that reasoning about
it will be impossible. When does a stick-qua-club cease to exist? If I
use the stick as a club for a second time, do I re-use the stick-qua-
club, or create a new one? I am sure you can see where my worries are
going. And, to repeat, I am not looking for answers to questions like
these: rather, the need to answer them seems like a reductio of the
hypothesis that it would be constructive to admit them into existence.

>>> Do you agree that some string denote entities?
> PH> Yes
>>> Can one string denote two different entities?
> PH> Not when the string is used as a name, no. That is, a name
> PH> denotes one entity (in a given interpretation.)
>
> Ok, but what about the world, not in logical semantics?

Logical (better, Tarskian) semantics describes denotation in the
world. It is a particular, simple, basic case, indeed, but none the
less real for all that.

> I thought we
> are trying to analyze an ontological relation that is used in reality
> by people and machines, not a formal relation in FOL semantics. It
> seems that in real world, one and the same string may name different
> entities?

It may seem so, but I would prefer to say that this is only an
appearance. That is, the "real world" (BTW, I detest this phrase,
suggesting as it does that there are things which are not 'real' or
not in the world) cases of ambiguous denotation are always cases of a
complexity in the world not being properly described. The missing
context parameter for your 'den' might be an example.

> PH> ? No, there is "Barry Smith", the name. It is what it is, no
> PH> matter what relationships it may stand in to other
> PH> entities. In this instance, it is in fact a string of eleven
> PH> characters, as indicated by the quotation marks.
>>> Yes, the string is the string, it does not change what it
>>> is. However, there may be another entity, let us call it a
>>> qua-individual (taken from DOLCE) or relational role (from
>>> GFO), which is "Barry Smith"-qua-name-in-R.
> PH> Of course there may be many entities. But you are here
> PH> **quoting** the string, and by the usual conventions, this
> PH> means that you are referring, unambiguously, to the actual
> PH> character string. Call this string FOO, for now: FOO = "Barry
> PH> Smith", a string of eleven characters. Now, you seem to also
> PH> be referring to FOO-qua-name-in-R, and I do not know how to
> PH> interpret this. What does it mean for a character string to be
> PH> understood 'qua' anything? If I send this character string
> PH> across a computer network, it is still the same character
> PH> string. Can I send a role across a computer network?
>
> A string is just a string, and no matter in what relations it stands,
> it will not change. But it gains new properties by standing in
> particular relations.

I entirely agree. This is exactly what I have been arguing in this
thread. What I remain confused by is your agreeing to this, yet also,
apparently, wanting some strings to be other than strings: to be, in
fact, "qua-individuals", whatever those can possibly be understood to
be.

FOL has virtually no ontology, by itself. The only ontological
assumptions made by FOL are that
(1) entities can be distinguished from one another (so that they may
be said to be elements of a set) and
(2) stand in relations to one another; and
(3) at least one entity must exist.

To which one can add, in the CL version of FOL:
(4) anything that is named, is an entity

thus including relations, functions, character strings (if the
language permits quotation) and natural numbers (if the language
admits numerals), etc., into the universe of discourse.
Now, one might want to call this an 'ontology', but it is about as
minimal an ontology as one could hope to reduce anything to.

> It seems to me that you want to use FOL /as/ an ontology,
> by assuming that relations-in-the-world correspond to FOL-relations,
> that predication corresponds to instantiation, etc.

I want to use logic honestly, and work within the assumptions it
provides. CL assumes that relations exist, may themselves stand in
other relations, have properties, etc.., It also assumes that
relations have relational extensions, which are purely set-theoretic
in nature: the extension of a relation is the set of tuples of
individuals that stand in that relation to one another. It seems to me
that *any* notion of relation must admit this construction to be
meaningful, so I do not see this as being any kind of restriction or
limitation. Given all this, I see no reason at all to assert that
'real-world' relations are not best modeled in logic, as 'logical'
relations. Note my use of scare quotes, as I feel that this
distinction is misguided and indeed unsustainable. Every relation
which can be described by using FOL axioms is a FOL-relation, when FOL
is properly understood.

> But it does not
> work with the ontologies we are building, and certainly not with BFO
> or DOLCE (already because instantiation is time-indexed and
> ternary).

I am afraid that I have never taken DOLCE seriously as a foundation
for any kind of useful ontological activity, so this observation does
not carry very much weight for me. Being time-indexed is no barrier to
being first-order.

> Instead, we use FOL for what it is built for: make the
> properties and interrelations of the primitive predicates in our
> ontology precise through axioms; and among the primitives, there seem
> to be things like instance-of(x,y), is-a(x,y), Universal(x),
> Particular(x), Relation(x), etc.

Really? I am honestly amazed by this claim. Do you really, when you
look around and consider the actual world, see such things as
universals, particulars and relations? I confess I have never, ever,
in close to 60 years of wakefulness, seen or encountered anything that
I would have described as a Universal. I would not know how to
recognize one if I met it. I see things which I classify as instances
of categories, and similarly of relations, and I generalize from
those, of course. But I have no idea what a Universal would be. And as
for instance-of, this is just what I have called a 'dongle' : a piece
of meaningless syntax, the only purpose of which is to sneak around a
pointless syntactic restriction. Here is a CL axiom defining it:

(forall (A B)(iff (instance-of A B)(B A)))

That is, to say that x is an instance of y is simply to say that y
holds of x, and we already have a syntactic form for saying this, in
the very logic itself. (Bear in mind, before objecting, that in CL
semantics, (B A) does not simply assert that A is an element of the
set B, and that B is itself an individual entity which can have
properties, stand in relations, etc..)

BTW, here is the (usual) axiom for is-a:

(forall (A B)(iff (is-a A B)(forall (C)(if (A C)(B C))) ))

though some would prefer to weaken this to an 'if' and add other
axioms defining such properties as transitivity (or not). etc.

> If you use FOL /as/ the ontology,
> your ontology will be rather sparse, and quite far detached from
> the real world, I fear.

I profoundly disagree. I have never yet found anything that can be
precisely described, that cannot be described using FOL.

>>> ) give rise to different modes of participation: a part-of
>>> relation holds between a /part/ and a /whole/, and an entity
>>> can participate in one part-of relation qua /part/, in another
>>> qua /whole/. The "context" (or distinguishing feature) is the
>>> relation (an individual), to which both qua-entities belong
>>> (and depend on).
> PH> So every assertion of a relation holding between entities is a
> PH> different "context"? See above for my response to this notion.
>
> Not assertion, no. The context is the relation, the thing in the
> world.

But this does not seem to work. Take the relation of 'naming': it, the
relation, exists, let us agree. Then there is a string, "Barry Smith".
It also exists. On the face of it, these seem to have nothing to do
with one another, at least not by virtue simply of existing. They
become - I cannot say related - connected, let us say, by virtue of
the relation *holding* of the string and something else, in this case
BS, the ontologist. And this holding is exactly what is asserted by
the atomic sentence. So surely it is not the relation itself which can
provide a 'context' for reinterpreting the string. It can only do so
when it holds between the string and something else. But then this
holding is itself a relation, and we are in an infinite regress.

This is so simple that it is almost embarrassing to point out the
obvious flaw in the reasoning. The conclusion does not follow from the
premises given unless one also assumes that all those 1000 passengers
are distinct. If they are, and if a passenger is a human (as 2 says)
then it would follow that the same person cannot travel twice. (Since
in that case, that same person would be identical to itself when
counted as a passenger.) If on the other hand, one does allow a
passenger to fly twice, and one of them does, then some of the
passengers listed are identical, so the airline does not transport
1000 passengers a day. If, on yet a third hand, we take passengers to
be something like a pair <person, flight>, so that I am a different
passenger when I change aircraft - I guess these would be people-qua-
passengers or some such? - then your assumption (2) is simply false:
these entities are not human beings, but something more like time-
slices of human beings, episodes in the life of the human being rather
than the complete human being. (Proof: I become a different passenger
when I change planes, but I am the same human being. Ergo, passengers
are not human beings.) Either way, your ontology is simply wrong as
stated. And none of my suggested three repairs involve 'qua'.

>>> What is a "normal logical framework of denotation and
>>> reference"? I am talking about an ontological denotation
>>> relation for which we can give axioms in FOL, but which is not
>>> a part of the logic itself. So I think we are talking about
>>> different things here.
> PH> I am talking about an ontology, written in FOL or some subset
> PH> of FOL. You appear to be talking about an ontology written in
> PH> a different formalism altogether, one which itself is
> PH> described in FOL axioms used as a metatheory (but which does
> PH> not satisfy FOL inference rules, for example.) It may be
> PH> related to some of the 'context logics' developed in AI/KR
> PH> work during the last decade or so.(?)
>
> Not at all, I am talking about classical FOL (or higher order, but
> predicate logics) theories in which we write the axioms for our
> ontology. But I object to identifying FOL with the ontology, or even
> parts of the ontology (like instantiation).

FOL does not presuppose much, but one thing that it does, is the
notion of predication/instantiation. If you insist that you are using
a different notion of this than the one provided by FOL, then it seems
to me that you are not using FOL in any way other than as a meta-
theory of a different logic. (I strongly suspect, by the way, that FOL
is in fact perfectly adequate for your needs with regard to
instantiation, if you can just free yourself from the evil grip of the
usual textbook syntax for FOL, which imposes an artificial and
completely unnecessary distinction between relations and individuals.)

>
>>> it plays within a relation is both existentially dependent on
>>> the player (the string) and the relation (of the denotation
>>> type) in which it plays the role. This qua-entity or role has
>>> different properties than the string: the role is dependent,
>>> the role has no shape, the role has no length, etc.
> PH> I simply do not believe in these things you call
> PH> 'entities-in-a-role', if they are intended to be separate
> PH> entities in the universe. My ontology, it would seem, is much
> PH> sparser than yours. I consider this to be an advantage, but
> PH> you may of course very well not agree, this being an
> PH> essentially aesthetic judgement.
>
> I believe it helps in modelling, e.g., see the counting problem
> above.

On the contrary. That 'puzzle' is not the least puzzling once one
thinks clearly about the entities involved and their identity
conditions. Qua-thingies make counting impossible, as they amount to
having multiple identity conditions for the same entity. And notice,
the puzzle is resolved not by inventing a mysterious person-qua-
passenger, associated with a 'role', but simply by focussing on the
ontological issue that the puzzle highlights, that of identity through
time.

> It is not the "entity-in-a-role" thing, it is the role itself
> (the individual role, an instance of some Role category). X-qua-Y is
> just my clumsy way of referring to them. Qua-entities may have very
> different properties from the entities on which they depend.

Sorry, but I find this entire discussion too surreal to be of
interest. There is no such thing as a qua-entity, nor any need to
presume or invent them. They serve no useful purpose, and their very
existence is at best extremely doubtful, and they completely confuse
any kind of extensionalist, counting reasoning.

> A string in a naming relation to entity A could be a particularly

> good name, because it may convey certain properties about A, while the
> same string may be a rather bad name for entity B.

These are all various relationships between the string and other
entities. They are not entities in themselves. (Well, since relations
are entities, I guess they are indeed entities; but they are not new
kinds of name or new kinds of string, is the main point.)

> A road from U to V, where U is a valley and V a mountain top, may go
> uphill, and the same road, qua road-from-V-to-U, downhill (example
> from "Qua Theories". R. Poli, 1998).

If I follow this last example, it seems to be an even more trivial
matter: that the terms 'uphill'; and 'downhill' must be understood as
involving a direction, and are not simply properties of roads or
surfaces. We actually indicate directions by saying things like
'facing downhill', which would be meaningless if this were simply a
property of the road surface.

>
>>> Of course. And we can use roles (or qua-individuals) to
>>> distinguish how the numbers are used within different
>>> relations.
> PH> Not with your assumptions, since there is then no guarantee
> PH> that a numeral shall even denote a number. Apparently the same
> PH> numeral, say "547", can denote all kinds of things: in one
> PH> place, a natural number; in another, an SS number; in another,
> PH> a soldier; in another, a house in a street, and so on. So we
>
> No, you are right, of course. It denotes the number. It can name many
> different things.

>
> PH> cannot even know when we are referring to a number. In the
> PH> worst case (which I fear you will slide into fairly quickly),
> PH> *every* occurrence of *every* name will denote something
> PH> distinct from every other occurrence, since it will be
> PH> understood as 'qua' some context unique to that token.
>
> I think this won't happen. Although a string could potentially be used
> to name a different thing every time, we are pretty good at choosing
> names which are rather unique, at least within one domain of
> application (including social groups).

Ah, but I was not talking about the use of strings in human social
life, but about YOUR use of strings in the actual formalism. But OK, I
think I can now understand your main point, that strings of symbols do
get used in various ways to identify, name and refer in the social
world. Which is of course true. I still see absolutely no reason to
introduce qua-entities, and every reason to reject them.

Pat

>
> R

[Pat Hayes]

On Nov 24, 2009, at 3:43 PM, Adam M. Goldstein wrote:

>
> On Nov 24, 2009, at 12:01 PM, Alan Ruttenberg wrote:
>
>> On Thu, Nov 19, 2009 at 11:53 PM, Bjoern Peters <bpe...@liai.org>
>> wrote:
>>> As OBI development is driving the initial use case of
>>> 'identifier', I want to propose a more limited definition of
>>> 'centrally registered identifier' (CRID for short in what
>>> follows). I believe the approach outlined below will address all
>>> that OBI needs. It does not however attempt to cover all ways of
>>> identification. I am re-using multiple ideas raised by others on
>>> this thread before.
>>
>> The biggest problem i see in this is the "centrally".
>> Serial numbers don't need this.
>> Other kinds of identifiers, such as hashes, don't need this - see
>> http://en.wikipedia.org/wiki/Distributed_hash_table
>
> But I think that the idea was to try to provide an account of a
> concept with a narrower scope, CRID, in contrast with identifiers
> considered as a whole? So we are in principle only dealing with
> centrally registered things?

But the centrally registered aspect does not seem to be very relevant
to the issue of what makes it into an identifier. If a CRID is simply
an ID which is CR, we have not made much progress.

Pat

PS. On a procedural point, I have removed all the duplicate CCs , as I
believe they are redundant. Please someone correct me if this was a
mistake.

>
> ------------------
> Adam M. Goldstein PhD, MSLIS
> --
> z_calif...@shiftingbalance.org
> http://www.shiftingbalance.org
> --
> http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
> --
> (914) 637-2717 (msg)
> --
> Dept of Philosophy
> Iona College
> 715 North Avenue
> New Rochelle NY 10801
>

[Pat Hayes]

On Nov 24, 2009, at 11:01 AM, Alan Ruttenberg wrote:

> On Thu, Nov 19, 2009 at 11:53 PM, Bjoern Peters <bpe...@liai.org>
> wrote:
>> As OBI development is driving the initial use case of 'identifier',
>> I want to propose a more limited definition of 'centrally
>> registered identifier' (CRID for short in what follows). I believe
>> the approach outlined below will address all that OBI needs. It
>> does not however attempt to cover all ways of identification. I am
>> re-using multiple ideas raised by others on this thread before.
>
> The biggest problem i see in this is the "centrally".
> Serial numbers don't need this.
> Other kinds of identifiers, such as hashes, don't need this - see
> http://en.wikipedia.org/wiki/Distributed_hash_table
>
> I was muddling over Pat's problem with me using "distinguishing" as a
> criteria. What I think I meant was that they can be used to determine
> sameness and differentness. A serial number is stamped on an
> instrument at a factor, it's number is on the invoice. You receive the
> instrument and invoice and check that the serial numbers are the same.
>
> Or: You want to count something that moves around (e.g. fish). You tag
> them with an identifier and then record which identifiers you see -
> not counting times when you see the same identifier twice.

Interesting example. People actually do this with whales, but they
don't tag them: they use the unique patterns of their tails and
forelimbs to identify them. Similarly, I believe, with elephants and
their skin markings. And of course we do this with faces for people.
So are these 'natural' distinguishing features also identifiers?

Pat

[Pat Hayes]

On Nov 24, 2009, at 5:13 PM, Bjoern Peters wrote:

>>> The biggest problem i see in this is the "centrally".
>>> Serial numbers don't need this.
>>> Other kinds of identifiers, such as hashes, don't need this - see
>>> http://en.wikipedia.org/wiki/Distributed_hash_table
>>
>> But I think that the idea was to try to provide an account of a
>> concept with a narrower scope, CRID, in contrast with identifiers
>> considered as a whole? So we are in principle only dealing with
>> centrally registered things?
>>
>> ------------------
>> Adam M. Goldstein PhD, MSLIS
>
> Exactly right, the purpose of focusing on CRIDs was to pick examples
> for identifiers that will be very easy to model, and cover much (not
> all as Alan pointed out) of what OBI will need.

Let me suggest a slight weakening to a notion of a 'community accepted
ID' (CAID). CRIDs are CAIDs, but not necessarily the reverse. A CAID
is an identifier which is accepted by a community as referring to or
identifying a thing, so that the use of the identifier for certain
community-sanctioned purposes is understood by members of the
community to indicate the identified thing. The community may be large
and distributed and 'permanent' (all users of modern credit banking
facilities) or small, idiosyncratic and temporary (one person guiding
a driver backing a truck into a narrow driveway, using hand signals.)
If the community has a central registry defining its identifiers, then
they are CRIDs, but this is not required. The key point is that the
"meaning" of the ID is held in common by users of the ID within the
community.

One issue that I can see is that we should not identify a community
with the set of its members, because the same members might be in
several communities with different ID conventions.

I think this might be fairly easily adaptable from the CRID
intuitions, and be usefully broader in scope.

Pat Hayes

>> --
>> z_calif...@shiftingbalance.org
>> http://www.shiftingbalance.org
>> --
>> http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
>> --
>> (914) 637-2717 (msg)
>> --
>> Dept of Philosophy
>> Iona College
>> 715 North Avenue
>> New Rochelle NY 10801
>
> --
> Bjoern Peters
> Assistant Member
> La Jolla Institute for Allergy and Immunology
> 9420 Athena Circle
> La Jolla, CA 92037, USA
> Tel: 858/752-6914
> Fax: 858/752-6987
> http://www.liai.org/pages/faculty-peters
>

[Pat Hayes]

On Nov 23, 2009, at 5:58 PM, Bjoern Peters wrote:

> I wanted to avoid the use of 'denote', and make an explicit relation
> for this purpose, to focus on what we truly understand. The
> discussion by PH et al is a nice example of what I want to avoid. It
> may well turn out that this is a sub-relation of denote, but for now
> I am not sure. Take for example anonymized donor IDs reported in a
> manuscript. Does the donor id D01 denote an individual human being?

You better hope so. Or, better, there is a mutually accepted
presumption (between the writers and readers of the manuscript), in
the use of such an ID, that it does so denote an individual. If this
turns out to be wrong, such as when the data were faked or two records
were conflated due to a copying error, then that failure to denote is
sufficient to treat the use of the ID to have been wrong: illicit or
inappropriate or maybe illegal, but certainly in any case to then be
treated as incorrect.

> Can I 'pick out' that human being?

You can't, but someone (or something) can.

>
> The only claim made by the 'CRID_refers_to' relation is that during
> CRID assignment and recorded in the CRID registry, there is a
> relation between the CRID and the entity information is recorded
> about.

TO me, this reads as saying that *any* relation is sufficient to
establish the truth of CRID_refers_to, which cannot be right. Surely
we need to say something more about the nature of the relationship.

Seems to me that most (all?) examples of IDs, and in fact of referring
names more generally, involve the use of the ID/name in communication
of some kind. There is an act involving the transfer of symbolic
information from one place or agent to another, and the ID is used
successfully to refer to the thing identified (is there a word for
this? The identifiee?) when the recipient of the information
understands it to be about the identified thing. They key notion here
is a symbolic communication being "about" a thing, which is of course
denotation/reference in disguise, but applied to communicative acts,
and the states of comprehension they produce, rather than to the
symbols themselves.

Y'all will not be surprised that I cut my semantic teeth on Ogden &
Richards.

Pat

[Jonathan Rees]

Denotation is always just an interpretation made by some interpreting
agent. That is, "denotes" is a three-place relation, not a two-place
relation. But it is so useful to say a binary "x denotes y" without
mentioning the agent that we try to bind the agent somehow, either
into x (this is what happens when you try attach y to x through
originality, baptism, or provenance, which is the ICE idea) or into
"denotes" (perhaps we assume that "denotation" is deterministic within
all the uses cases for IAO, sensitive at worst to the process in which
it happens).

(Time is a fourth variable but I'll ignore it.)

One reason we think of denotation as binary is that in many cases
(among the ones we care about) there are so many agents x that will
all interpret x in the same way, and we are among them. We might call
the population of such agents a "linguistic community". But the only
assumption we really *need* is that the agent saying x and those
consuming x all interpret x in the same way. The "community" might
consist of only two agents.

The other reason is that when a message has multiple possible
interpretations, you can usually figure out which one is to be invoked
according to the process in which it's used.

Would it be so terrible if we formalized expression and interpretation
processes, that are more or less inverses of one another, in which an
agent and an uninterpreted information artifact participated? Then
"is_about" and other denoting and interpreting relationships would be
defined on such processes, or on entities derived from them (e.g. an
ICE could be isoontic to the class of all processes in which x denotes
y to *some* agent a).

(Well, really you don't need for senders and receivers to have a
common interpretation of an information thing; you only need a for the
receiver to interpret it so as to achieve the result the sender
"intends".)

Limiting analysis to particular scenarios such as intentional, orderly
identifying systems is a worthwhile exercise. However I am with Pat in
doubting that in the long run it will make the problem any easier.

Jonathan

[Bjoern Peters]

Great points Jonathan.

It was my understanding that we are limiting IAO to successful communication. I agree it would be useful to state what we mean by that explicitly, and represent the message expression & interpretation processes. Then, the 'is about' relations and all its children only hold for successfully communicated ICEs. Below is what I thought we are implying. I am writing it down to invite corrections. ICE = information content entity. GDC = Generically dependent continuant.

ICE creation = a process initiated by an agent with the objective to create a concretization of an ICE that is about something and which can be interpreted by other agents

ICE interpretation = a process in which an agent examines a concretization of an ICE and determines what it is about as intended by the ICE's creator

ICE copying = a process in which a new concretization of an existing ICE is created which retains the ability of agents to interpret what it is about.

ICE = a GDC whose concretizations are the specified output of a ICE creation or copying process, and which is about something.

'is about' = a relation that holds between an ICE and something that its creator intends the ICE to be about.

- Bjoern

[Jonathan Rees]

Looks like a good start. Is an ICE concretization (ICEC?) also 'about'
some things, such that all of the concretizations of an ICEC are about
the same things, namely the things that the ICE is about, right?
Otherwise how we would know the ICE is about - don't all the
properties of a GDC come from its concretizations? Or is it the
"aboutees" (and their surrounding social history) that unite all of
the qualities (or whatever concretizations are) into one ICE? Would
punning be explained by the same quality concretizing two different
ICEs, or would we necessarily have two different qualities in that
case?

Jonathan

[Pat Hayes]

On Nov 30, 2009, at 1:16 PM, Bjoern Peters wrote:

> Great points Jonathan.
>
> It was my understanding that we are limiting IAO to successful
> communication.

Oh. I think that this may be harmful, because it is question-begging:
it does not allow us to focus on what it is that makes a successful
communication actually be successful. But having stated this as a
worry, I will try to ignore it.

> I agree it would be useful to state what we mean by that explicitly,
> and represent the message expression & interpretation processes.
> Then, the 'is about' relations and all its children only hold for
> successfully communicated ICEs. Below is what I thought we are
> implying. I am writing it down to invite corrections.

> ICE = information content entity.

An example would help. Suppose I write a note to my wife: "Gone to get
sugar" and leave it stuck on the phone, for her to read when she gets
home, which I expect her to do before I get back from the store. The
note is written in pencil on a yellow stickynote. In this scenario, is
the ICE (1) the physical note, marks in graphite on the yellow paper
surface; (2) the English sentence "Gone to get sugar"; (3) the
particular token of 2 encoded in 1; or (4) something else entirely?
I'm guessing that the ICE is the sentence 2, and that either 1 or 3,
or maybe both, is a concretization of it. (BTW, what does the
terminology "concretization" gain us over the more traditional notion
of a "token"?)

> GDC = Generically dependent continuant.

Is that important?

> ICE creation = a process initiated by an agent with the objective to
> create a concretization of an ICE that is about something and which
> can be interpreted by other agents

I think mere objective isn't enough: it has to succeed in the
creation.

This seems to say that the concretization is about something, rather
than the ICE itself. Is that intended?

This seems to assume the basic idea that needs (?) to be analyzed.
What makes an ICE *be* about something? After all, if I am an agent
who 'speaks' a purely private language, then it seems like I am
creating ICEs according to this definition, but they aren't going to
work for anyone else. I think purely private intentions to assign
aboutness should not count as ICEs: the public, shared acceptance of
the aboutness is surely part of the concept, no?

>
> ICE interpretation = a process in which an agent examines a
> concretization of an ICE and determines what it is about as intended
> by the ICE's creator

Does the interpreting agent have to be different from the creator?

(See above). This process seems extremely magical. How can anyone or
anything infer the creator's *intent* from looking at a created product?

Why do we need to refer to this as a 'process'? (Perhaps I am reading
more into this than was intended: to me, 'process' suggests something
extended in time, non-instantaneous.)

Why is it necessary to infer intent? If I read a road sign saying
"Caution: falling rocks" I do not need to know anything about the
intent of the author, or even that there was an author with an
intention. All that matters is what the sign itself tells me about the
local circumstances, relevant to my driving on the road. (It may be
objected that my treating the sign as a warning, as opposed say to a
joke or a mere decoration, is itself an inference of intent; a point I
concede, but would respond that this intent is not that of any author
of the sign itself.

> ICE copying = a process in which a new concretization of an existing
> ICE is created which retains the ability of agents to interpret what
> it is about.

Does the last clause imply that there could be things like copyings
but which do not retain this ability? Surely copying is a 'mechanical'
process which can be done without reference to any meanings of the
thing copied? Why not just say that is a creation of a new
concretization of the same ICE? The importance is that copying is a
process that can be done without doing any interpretation at all: it
requires only (at most) a grasp of the syntactic rules of the ICE, not
of the semantic rules.

>
> ICE = a GDC whose concretizations are the specified output of a ICE
> creation or copying process, and which is about something.

This is extremely circular.

>
> 'is about' = a relation that holds between an ICE and something that
> its creator intends the ICE to be about.

Again, I do not think that putting this much ontological burden on the
intentions of the creator is a good idea. A great deal of
communication takes place without the intentions of the creator being
involved in any way, and the *full* intentions of a communicator are
almost never made public, and indeed may not be accessible even to the
agent himself. No doubt you did not intend the definitions to be
understood in such a psychoanalytical way, but I think that trying to
nail down how to avoid a more Freudian interpretation of 'intention'
would be a distraction. I suggest that it would be more productive to
focus on the essentially public nature of reference used in
communication, and the necessary fact of having a shared common ground
of interpretation in order for communication to succeed, rather than
the essentially private and unknowable world of inner intentions.

Surely the central point of an ICE is that there is a community (a set
with more than one member) of agents all of whom will understand a
token of the ICE to refer to the same thing. Intentions are largely
irrelevant to this, and I suggest that the asymmetry between creator
and interpreter is also largely a distraction.

Pat Hayes

[Barry Smith]

ICEs (like organisms, cells, molecules, cellular
processes, etc.) exist at the level of types and
the level of instances. There are, e.g., the
types: journal article, database, lab note,
electronic health record; and then the instances
of these types: this journal article; that lab note; etc.

The latter then have concretizations: this
journal article on this particular hard drive on this particular laptop now.

Hence there are three items, not two.

> > GDC = Generically dependent continuant.
>
>Is that important?

GDC = the journal article that is concretized in
many different places: on your hard drive and in
the bound volume on my bookshelf, etc.
SDC (Specifically Dependent Continuant) = the
journal article on your hard drive

> > ICE creation = a process initiated by an agent with the objective to
> > create a concretization of an ICE that is about something and which
> > can be interpreted by other agents
>
> I think mere objective isn't enough: it has to succeed in the
>creation.
>
>This seems to say that the concretization is about something, rather
>than the ICE itself. Is that intended?

I think both, but I would be happy to consider arguments against this.

>This seems to assume the basic idea that needs (?) to be analyzed.
>What makes an ICE *be* about something? After all, if I am an agent
>who 'speaks' a purely private language, then it seems like I am
>creating ICEs according to this definition, but they aren't going to
>work for anyone else. I think purely private intentions to assign
>aboutness should not count as ICEs: the public, shared acceptance of
>the aboutness is surely part of the concept, no?

I agree.

> >
> > ICE interpretation = a process in which an agent examines a
> > concretization of an ICE and determines what it is about as intended
> > by the ICE's creator
>
>Does the interpreting agent have to be different from the creator?

No

>(See above). This process seems extremely magical. How can anyone or
>anything infer the creator's *intent* from looking at a created product?

There is something magical about aboutness, I'm afraid.

>Why do we need to refer to this as a 'process'? (Perhaps I am reading
>more into this than was intended: to me, 'process' suggests something
>extended in time, non-instantaneous.)

Yes.

>Why is it necessary to infer intent? If I read a road sign saying
>"Caution: falling rocks" I do not need to know anything about the
>intent of the author, or even that there was an author with an
>intention.

It got there somehow, that sign.

> All that matters is what the sign itself tells me about the
>local circumstances, relevant to my driving on the road. (It may be
>objected that my treating the sign as a warning, as opposed say to a
>joke or a mere decoration, is itself an inference of intent; a point I
>concede, but would respond that this intent is not that of any author
>of the sign itself.

No; could be of the placer of the sign.

> > ICE copying = a process in which a new concretization of an existing
> > ICE is created which retains the ability of agents to interpret what
> > it is about.
>
>Does the last clause imply that there could be things like copyings
>but which do not retain this ability? Surely copying is a 'mechanical'
>process which can be done without reference to any meanings of the
>thing copied? Why not just say that is a creation of a new
>concretization of the same ICE? The importance is that copying is a
>process that can be done without doing any interpretation at all: it
>requires only (at most) a grasp of the syntactic rules of the ICE, not
>of the semantic rules.
>
> >
> > ICE = a GDC whose concretizations are the specified output of a ICE
> > creation or copying process, and which is about something.
>
>This is extremely circular.

I agree.
BS