Cory, (01)
you wrote: (02)
> If we take your example: "Dashboard:Z is:within Car:Y
> We can also express this as a "contains-within" relation with roles such as
>"container", "contained". If a stakeholder were to look at one statement about
>one dashboard the verb form may make sense. If they were looking a lot of
>"contains-within" relations (say, as a car manufacturer), they would look at a
>table - that table may have columns for "container" & "contained". (03)
> Verb phrases are, by their nature, directional and taken from the perspective
>of one of the roles - in this example the dashboard. That the car contains the
>dashboard is another verb phrase expressing the same fact from the perspective
>of the car. The "inverse" style used in OWL works somewhat for binary
>relations but falls apart with nAry. (04)
[EJB] Well, there may we be other 'readings' (Nijssen/Halpin) or 'synonymous
forms' (SBVR) of the same relationship. Semantically, 'inverse relations' are
just a special case, but it happens to have a supporting mathematical notion
that gives rise to viable implementation approaches. Further, of course, any
n-ary relationship can be cast as a class whose instances are states of
affairs, to which each of the roles has a binary relationship that is
characterized by the role, as Cory mentions below. This has additional
mathematical and implementation value. Note, however, than in that rendering,
the roles become the verbs. (05)
> Also, when we have a fact or context about one verb phrase the impact of such
>a statement on the other verb phrases can be difficult to compute. So if I say
>"Toyota said (Dashboard:Z is:within Car:Y)" then the "Toyota said" statement
>implicitly applies to Car:Y contains Dashboard:Z. But the underlying "fact"
>has now become distributed and lacks identity. (06)
[EJB] Be vewy, vewy caweful. A 'fact' is a statement that is taken to be
true, or a state of affairs that is taken to 'obtain' (be real). Once you
nominalize the statement, you are not taking it to be anything. The
nominalized statement may be considered to represent a class of states of
affairs (situations) that may or may not have any members. It does not lack
identity -- it is exactly the class described by that statement. The problem
is whether your formal language has any way of representing the transformed
statement. In RDF land, what was added is Named Graphs. In FOL land, you add
some kind of 'nominalization' function, e.g. the 'that' operator in IKL. If
alternatively you turn every nominalized relationship into a class of states of
affairs, then you can write Toyota asserts-actuality-of 'state-of-affairs'.
But you are now turning the ontology into microcode, or equivalently, operating
on a metamodel. (07)
> Consider that these multiple verb phrases are traversals across the roles and
>between roles and the containing relation. The relation is where the identity
>of the fact becomes consistent. The reified relation form is essentially the
>"fact" that provides context for the multiple verb phrases that may traverse
>it. In all cases we can infer all the verb phrases implied by a reified
>relation. In the case of a non-qualified binary relation we can infer the
>reified relation from the verb phrase, but not in other cases. (08)
[EJB] I understand the 'reified verb' as the class whose instances are states
of affairs, and every time you stuff a constant/individual into one of the
roles, you get a subclass -- the ones in which that role is played by that
individual. When you have filled all the role slots that way, you have the
subclass of state of affairs that is represented by the original statement.
And yes, you can attach all of the readings/verb-phrases to the 'reified verb'
class that you like. In SBVR terms, you are talking about multiple
'representations' of the same 'concept'. (09)
[EJB] But there is nothing special about binary relations in this regard. The
only thing special about binary relations is that from any role-player there
are only two meaningful traversals -- one to the class and one to the other
role. In a ternary relation, however, there are three such traversals -- one
to the class and one to each other role. (010)
[EJB] Whether in fact a given pair of role players in a binary relation
uniquely determines an instance of the class is quite another thing entirely.
This gets very pointedly into the semantics of the reified relation. Are the
instances of the reified relation 'states of affairs' or merely 'tuples of
participants'? (In OWL, it is the latter.) A given pair may correspond to no
instance of the class if the relation does not hold for them. Or, it may hold
for more than one instance of the class of states, seen as states, as distinct
from tuples. But again there is nothing special about binary relations -- a
given tuple of the proper arity for the relation will have exactly the same
properties. This is exactly the point of relational algebra. (Thank you, Ted
Codd.) Further the relational idea of 'projection' allows one to reduce a
complex state of affairs to meaningful tuples involving fewer roles. In so
doing, a given tuple can appear more than once, representing different states
that involve the same players in those roles (while ignoring some other
distinguishing role players). For example, you can represent the relationship
X has some dog in fight F by a projection of (person, dog, fight) that
eliminates the dog. If you do this to a binary relation, the semantics of the
result is whether X ever plays the role, and whether you can tell how many
different partners X has depends on the tuple/state distinction in interpreting
the reified relation. If you do this to an n-ary relation, the equivalent
existentialized role semantics (some dog) can be inferred, and at heart the
binary case has the same semantics. (011)
> Verb phrases also contain 2 essential concepts. In your example this is "is"
>and "within". The "is" part qualifies the verb with a timeframe (current). If
>I have was:within" and "should be:within" I start to see a whole lot of verb
>phrases where the connection between them is not apparent. The "within"
>semantic is not explicit. With a reified relation I can qualify the entire
>relation with "currently" or "in the past" or "as a plan for the future", etc.
>So one reified relation and a "tense" can provide context and a semantic for a
>large number of verb phrases that traverse that relation, qualified by such a
>tense. (012)
[EJB] This is "whole nother" can of worms. Verbs have intent, tense, mode and
aspect (I think). 'Tense' has to do with time: present, past, future.
'Aspect' is progressive vs. perfective. 'Mode' is where the 'must', 'may',
'can', 'should' modifiers come in. If you 'qualify the relation' with any of
them, you get a different relation from the 'present simple' one. If you add a
property or properties to the relation (an additional role, column), you get a
n+m-ary relation that can be reduced to the original n-ary relation by
'projecting out' the tense, aspect, mode columns you added. But, for example,
a relation in which you add a timestamp role/column has the same property:
(person) is-located-in (location) at (time) is similar in almost every way to
(person) (tense) located-in (location). (013)
[EJB] Importantly, this "modified verb" stuff is where linguistic structure,
relational algebra, and formal logics start parting company, and it is all
about the semantic loads of the stuff you are adding. In relational algebras,
the user who writes the query imputes the semantics to the projection. In
formal logic, I have to have interesting axioms to do that kind of thing. But
in natural language, I recognize those particular kinds of modifiers as having
common semantics over many/all verbs. In formal logic, I have to work to
provide the semantics for (tense=past), and the RDB sees no fundamental
distinction between (tense=past) and (delivery-mode=rail). (014)
[EJB] Bottom line: There there be dragons. Please don't take us there. (015)
> Where we want reusability and the ability to federate across independently
>conceived ontologies (and schema) we need a way to manage these different
>potential ways to express the same or related facts. We should also be able to
>represent the relationship between a verb form and a noun/role form as both
>will invariably be used. (016)
[EJB] Fully agree! (017)
> Since the reified form can provide context for multiple verb phrases I see it
>as essential for any reuse strategy. Also, since stakeholders will use both
>forms in their day to day business we must support both (and their
>relationship) to capture their conceptual framework in a reusable way. (018)
[EJB] I agree that the reified form is useful and must be supported. Making
it the cornerstone for representing verbs is less defensible. We have to
invent new verbs for the roles in reified relations, because the participations
become binary facts. So, languages like RDF and OWL only reify non-binary
relations. But RDBs reify many binary relations as well. Further, it is not
clear that what one view sees as an n-ary relation is necessarily an n-ary
relation, with the same value for n, in other views. We can understand
relationships that are really the same verb with different adverbial modifiers,
but we would have to be able to model that. And we must distinguish that case
from something like (machine) requires (time) for (operation) on (material),
which is intrinsically quaternary in concept. (It can be reduced to
(implemented operation) requires (time), where (implemented operation)
conceptualizes the interaction of the other roles, but the table still has four
columns.) (019)
Regards,
Cory Casanave (020)
From: ontology-summit-bounces@xxxxxxxxxxxxxxxx
[mailto:ontology-summit-bounces@xxxxxxxxxxxxxxxx] On Behalf Of John McClure
Sent: Tuesday, February 04, 2014 1:19 AM
To: Ontology Summit 2014 discussion
Subject: Re: [ontology-summit] The tools are not the problem (yet) (021)
Thanks Ed - I've reviewed my email and yes I see a need to change a few items
but I don't think those affect the thrust of my remarks particularly that verbs
are indicated "for designations for fact types - usually a verb, preposition,
or combination thereof". I value this document because it provides a normative
semantic for certain prepositions when used with certain verbs. What this
provides to me is a very practical demonstration that an effective
representation can be had that does not involve nouns within/as a predicate.
Admittedly I have to study pgs 35+ to be more proficient with their vocabulary,
but I think my points still stand. (022)
I also see the issue of ontology modularization being facilitated by
predicators, as I believe a simple approach is to use namespaces for tensal
verbs. For example, the 'is' namespace would be a different from a 'was'
namespace; each tensal namespace identifies applicable direct and indirect
object properties via specification of particular prepositions, indicatives,
articles and adverbs relevant to the tensal. One's ontology merely imports the
predicate vocabulary on a tensal by tensal basis. This approach does not
prevent -- in fact it facilitates -- mixing tensal namespaces from different
suppliers with a chief benefit being that one's property ontology is only as
large as necessary for a given application, no larger. (023)
Permit me to address the 'is part of' relation, because partitives have an
obvious importance to us all. If one wants to say a certain tire is part of a
car, why not use what the English language gives us: "Wheel:X is:on Car:Y"
which most certainly has a different semantic than "Dashboard:Z is:within
Car:Y". Under the manufactured term 'isPartOf' regime, eg "Wheel:X isPartOf
Car:Y" plus "Dashboard:Z isPartOf Car:Y" you've lost semantics about the part'd
relationship to the car, as a part eg is it removeable without taking the car
apart, can it fall off the car and so on.... this is what prepositions provide
us that is elided by isPartOf. For the inverse relations, "Car:Y has:within
Dashboard:Z" and "Car:Y has:this Wheel:X". (024)
And thanks for the noting ORM/DAMA tie-ins. OMG has a number of interesting
factions.
/jmc
On 2/3/2014 4:38 PM, Barkmeyer, Edward J wrote:
John,
I think you got an earlier version of the SBVR spec. The "preferred terms"
were changed in v1.1, because many people in someone's intended audience for
SBVR recognized fact-type, etc., as terms in the ORM language, whose primary
use is database design, and these folk were all about vocabularies, and wanted
to distance themselves from that. (But v1.0 in 2008 was heavily influenced by
the "ORM faction" out of DAMA.) (But now that I think of it, the publication
of v1.1 was delayed by nearly 2 years by tooling problems.)
The current version is v1.2, from last fall: http://www.omg.org/spec/SBVR/1.2.
-Ed
From: ontology-summit-bounces@xxxxxxxxxxxxxxxx
[mailto:ontology-summit-bounces@xxxxxxxxxxxxxxxx] On Behalf Of John McClure
Sent: Monday, February 03, 2014 4:23 PM
To: ontology-summit@xxxxxxxxxxxxxxxx
Subject: Re: [ontology-summit] The tools are not the problem (yet)
Ed,
Thank you for raising SBVR as a source of wisdom; this substantial body of work
directly supports my critique of most ontologies on the semantic web whether
used for Big Data exercises or not. You mentioned SBVR's 'verb concepts' (well,
you called them 'verb concept symbols') -- this term is a formal synonym in
SBVR for 'fact types'. Its Annex C, SBVR Structured English states clearly that
verbs are used "for designations for fact types - usually a verb, preposition,
or combination thereof". Even a cursory examination of SBVR's axioms shows
their rigorous use of verbs, auxiliaries and prepositions for predicates. No
nouns, though, please note. (025)
Again, my diagnosis is that an unbridled use of nouns within/as predicates
causes (a) modelling problems (b) reuse problems (c) SME stakeholdership
problems. This disease ("nomine praedicato") is present in many RDF ontologies
and in certain FOL ontologies like Cyc Dolce et al, several referenced as
exemplary ("serving as a desirable model; representing the best of its kind")
by wise ones in this forum. I've tried to present this diagnosis in terms I
thought would directly resonate with FOL/CL folks who live here, but the
pushback feels to me furious! (026)
As you know, I have (shared) some solutions I'm finding work well in a semantic
wiki environment, but we need to agree on what the problems are first to really
get into evaluating alternative solutions. If everyone thinks my analysis is
way off-base, fine, I'm invested in hearing others' diagnoses (concerning why
there is so little reuse/development of ontologies) including "poor tools"
which, quite possibly to me alone, are merely band-aid treatments for the
disorder nomine praedicato. (027)
/jmc
PS "is part of" is not a defined SBVR term -- I found it several times used in
textual descriptions however.
On 2/3/2014 8:32 AM, Barkmeyer, Edward J wrote:
John,
The fact that is-part-of is ill-defined is not a consequence of its failing to
appear in a dictionary. The inverse relationship may be called 'contains' or
'comprises' or 'includes' or (!@#$%) 'has', and those are in English language
dictionaries and they are not well-defined, either. The problem there is just
that speakers often fail to realize that there are many possible
interpretations of these terms and the only way to be clear about what you mean
is to provide the axioms you intend. Of course, if your intention is vague,
then the absence of clarity of the expression matches the absence of clarity
in the thinking.
The second problem here is about whether we are talking about linguistic
structure, and that of the English (natural) language in particular, or we are
talking about 'semantic structure'. For knowledge engineering, it is really
important to get past the linguistic structure and recognize the semantic
structure of utterances. In particular, the noun 'part' is a role-noun. It is
not possible for a thing to be a 'part' (except by extension). It is only
possible for a thing to be a 'part OF' something in particular. (And it is
similar in other European languages, at least, that a thing 'is-a-part'
of/at/to/in.) But that is an enforced linguistic structure; the semantic
intent is subject is-part-of object.
Teaching people the English language and its grammar is a trade; and all the
stuff we learned about formal models of English we learned from people who
practice that trade. Now we are talking about knowledge-engineering. It is a
different trade; and it has different formal models of 'language'. In this
trade, we use the English words to assist in conveying intent, but the
"semantic grammar" is different.
We will continue to talk past one another if we use similar grammar terms with
importantly different meanings. If 'verb phrase' can only mean what your
English teacher taught you, then we will have to adopt something like the SBVR
'verb symbol' or 'verb concept wording' to make it clear that we mean something
different. (In SBVR, 'is (a) part of' is a "verb symbol"; and 'subject is a
part of object' is a "verb concept wording", where 'subject' and 'object' are
somehow marked as "placeholders" for "verb concept roles". The idea is that
these are the grammatical parts of semantic patterns for speech. Noam Chomsky
used different terms for the same concepts in describing the "deep structure"
of natural language 50 years ago (and John Sowa will undoubtedly find similar
passages in the work of Charles Peirce 100 years before Chomsky).
-Ed
From: ontology-summit-bounces@xxxxxxxxxxxxxxxx
[mailto:ontology-summit-bounces@xxxxxxxxxxxxxxxx] On Behalf Of John McClure
Sent: Monday, February 03, 2014 6:57 AM
To: ontology-summit@xxxxxxxxxxxxxxxx
Subject: Re: [ontology-summit] The tools are not the problem (yet)
On 2/2/2014 2:23 PM, doug foxvog wrote:
On Wed, January 22, 2014 02:51, John McClure wrote:
John,
Please look at the predicates referenced in slides
<From%20Textual%20Entailment%20to%20Knowledgeable%20Machines,Invited%20presentation%20at%20the%20Joint%20Symposium%20on%20Semantic%20Processing%20%28JSSP%2713%29,%202013,Download%20PowerPoint%20Slides>
in several of the presentations you cited. You will see that they all
use a verb and often a preposition, with one exception (part of, which
can equally be expressed as is-within).
It's a phrase that has a different range of meanings. The name
"isPartOf" would be a verb phrase.
[JMc] No, a 'verb phrase' by common definition includes direct and indirect
objects, so "isPartOf" is not a verb phrase under that definition. There is a
narrower definition (see footnote 4) but "isPartOf" fails to qualify there as
well. You see "isPartOf" is a concept that cannot be found in any dictionary.
It becomes anything you want it to be, depending on the day and weather. Thus
freed of any anchor to an OED-based reality, you sanction artificial constructs
which imo leads to modelling and interchange issues. (028)
It can be useful within a single system to have a naming standard.
[JMc] It is more useful to have the same design approach within both
FOL/non-FOL worlds, with the latter conforming to the former since data
scientists' worktops & toolsets are -- from your view at least -- the ultimate
destination for this information. (029)
All RDF properties that I've seen
-- okay, there's the exception is-a in a few ontologies -- use nouns.
This system is often for predicates that mean "has <Noun>" (with
different meanings of "have" in different cases) but with the name
also indicating the intended range of the predicate.
[JMc] Sure that intention of the property's author is clear - but the 'name'
ends up being a non-word, something not in a dictionary, a concept whose
meaning is merely mechanical, a programmer's concept. I say this is why
ontologies are unnecessarily large, having a set of properties ill-designed
because they are ill-defined. (030)
The naming system has no bearing on the semantics -- although
maybe with the ease of a human reader to come to some level
of understanding of what was intended.
As long as the system of nomenclature is consistent for a given
ontology, that's fine.
It's no wonder that FOL'ers need to transform these models into
something more in accordance with what I cited originally in this thread
One may choose to change names from an adopted system to fit the
local nomenclature. Or are you referring to transformations that
add semantics?
[JMc] I am referring to translations of noun-oriented properties crafted in an
RDF world to verb-oriented properties crafted in the FOL world. To the extent
the FOL world adopts RDF properties (principally to avoid the work of said
translation) is the extent to which the FOL world is not following its own rule
that predicates are verb-oriented. This is "impedance mismatch" in action
(definition: a measure of the opposition caused by differences between two
paradigms, especially between object-orien ted development and relational
databases ....
. 1997, Bhavani M. Thuraisingham, Data Management Systems: Evolution and
Interoperation (ISBN 0849394937), CRC Press, page 33:
Some argue that having impedance mismatch is difficult for programming
intensive applications. (031)
-- which you deleted in your original response (the guts of the original
note) to my irritation:
There are two competing notions of the predicate in theories of
grammar <https://en.wikipedia.org/wiki/Grammar>.^[1]
<https://en.wikipedia.org/wiki/Predicate_%28grammar%29#cite_note-1>
...
The second derives from work in predicate calculus (predicate logic
<https://en.wikipedia.org/wiki/Predicate_logic>, first order logic)
and is prominent in modern theories of syntax and grammar. _/*In
this approach, the predicate of a sentence corresponds mainly to the
main verb and any auxiliaries*/_ that accompany the main verb...
Note that the predicate *corresponds* to a verb. That does not mean
that it *is* a verb or *must be labelled with* a verb.
[JMc] Thank you for responding to a central part of my critique. Unfortunately
you're using a manufactured definition for "correspond" ... The usual
definition for "correspond" is to *have a close similarity; match or agree
almost exactly* -- yours is closer to "may/might correspond", that is, you're
seeing words there that don't exist. (032)
The matter of negation is a different, side issue in my mind. Please
comment on the fundamental criticism I am making here about the design
pattern for RDF properties: FOL's verbs vs RDF's nouns.
My comment here is "to each her own". It is good to have a consistent
system of naming.
However, that does not mean that one's own fantastic
system of nomenclature given by the deities and any other system is
blasphemy.
[JMc] It is NOT "MY" system of naming -- it is "YOURS". This prevailing
definition of a predicate, specifically in the context of FOL and predicate
calculus -- THAT IS "YOURS" my friend. Predicate calculus and FOL are the
"deities" here not me that is for sure. (033)
Blasphemy? Ok here's an analogy. Predicate calculus is built upon the Triple
Model; undoubtedly you'd cry blasphemy when presented data that accords to some
weird non-conforming Unary Model. In the same way, predicate calculus is (said)
built upon predicators, as I've demonstrated by reference to published
definitions. I'm saying it's the predicate calculus community that should be
crying "blasphemy" at the non-conforming predicates developed within the RDF
community. Yet you're not, and consequently we end up with massive ontologies
that SMEs (and others) find mystifying. (034)
regards/jmc
On 1/21/2014 1:04 PM, John F Sowa wrote:
1. See the slides and publications by the Aristo Project at AI2:
http://www.allenai.org/TemplateGeneric.aspx?contentId=12
------------------------------------------------------------------------
On 1/21/2014 11:16 PM, John F Sowa wrote:
On 1/22/2014 12:01 AM, John McClure wrote:
How can FOL'ers not be implicitly derisive of the work RDF'ers are
diligently about, when the first reaction is to THROW IT AWAY?
That's not the point I was trying to make. I'm sorry that I used
the phrase 'throw it away' because it was not clear what I was
rejecting.
First point: FOL is a small subset of English and other NLs.
Any language that has the words 'and', 'or', 'not', 'some',
and 'every' can express full FOL. We all speak FOL every day.
Second point: I wasn't rejecting what can be expressed in RDF.
You can use RDF to describe anything that you see, hear, or feel.
Every observation in science can be described in RDF.
But RDF can't express negation. You can't say 'not'. And if you
take RDF and add negation, you get -- guess what -- full FOL.
Some things you can't say in RDF:
1. Options: you can't say 'or' in RDF, because (p or q) is
defined as not(not p and not q) -- and RDF can't say 'not'.
2. Rules: you can't express an if-then rule in RDF, because
(if p then q) is defined as not(p and not q).
3. Generalizations: you can express 'all' or 'every' in RDF
because 'every cat is an animal' is defined as
'it's false that some cat is not an animal'.
My major complaint about RDF is that it makes simple things difficult.
John
(035)
_________________________________________________________________
Msg Archives: http://ontolog.cim3.net/forum/ontology-summit/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontology-summit/
Unsubscribe: mailto:ontology-summit-leave@xxxxxxxxxxxxxxxx
Community Files: http://ontolog.cim3.net/file/work/OntologySummit2014/
Community Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OntologySummit2014
Community Portal: http://ontolog.cim3.net/wiki/
(036)
_________________________________________________________________
Msg Archives: http://ontolog.cim3.net/forum/ontology-summit/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontology-summit/
Unsubscribe: mailto:ontology-summit-leave@xxxxxxxxxxxxxxxx
Community Files: http://ontolog.cim3.net/file/work/OntologySummit2014/
Community Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OntologySummit2014
Community Portal: http://ontolog.cim3.net/wiki/
(037)
_________________________________________________________________
Msg Archives: http://ontolog.cim3.net/forum/ontology-summit/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontology-summit/
Unsubscribe: mailto:ontology-summit-leave@xxxxxxxxxxxxxxxx
Community Files: http://ontolog.cim3.net/file/work/OntologySummit2014/
Community Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OntologySummit2014
Community Portal: http://ontolog.cim3.net/wiki/ (038)
_________________________________________________________________
Msg Archives: http://ontolog.cim3.net/forum/ontology-summit/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontology-summit/
Unsubscribe: mailto:ontology-summit-leave@xxxxxxxxxxxxxxxx
Community Files: http://ontolog.cim3.net/file/work/OntologySummit2014/
Community Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OntologySummit2014
Community Portal: http://ontolog.cim3.net/wiki/ (039)
|