Message-Id: <a05100e03ba02b33d04cb@[212.180.34.100]>
Date: Thu, 21 Nov 2002 17:03:03 +0100
From: Joost Breuker <breuker@swi.psy.uva.nl>
Subject: [ontoweb-sig1] case1 notion of ROLE
Dear all, here is an example of a `case'.
Best wishes, Joost Breuker
The notion of ROLE in ontologies
Problems about ROLEs in ontologies are not new (e.g. van Heijst,
Schreiber & Wielinga, `97b, do not consider roles to be first
class citizens (classes) in ontologia).
Problems with ROLE's in ontologies were signalled at least at two
occasions (`cases'), but the longer I have been reflecting and
modelling the more other kinds of problems arise. I will present
first these two cases, give my view on what a role is about, and
finally list a number of problematic issues.
CASE-1.
At the EKAW-2002 the first morning was devoted to
Ontologies and in three consecutive presentations the same
example came up. It is a classic: also to be found in the KA-2
ontology and it has almost a textbook flavor:
AGENT
PERSON
STUDENT
RESEARCHER
TEACHER
If we agree that STUDENT etc. is a ROLE, we get the following
problem. A PERSON is an INDIVIDUAL, ie s/he is unique, at least
during his lifetime and it is probably the most typical example
of individuality. Now if we assume that George Bos can take two
or even all three roles in the same period (not necessarily
executing all these all of the time. However, it is not
difficult to come up with roles that can be performed
simultaneously; see below), there is a problem when we also
create three instances of George Bos.
AGENT
PERSON
STUDENT
George Bos
RESEARCHER
George Bos
TEACHER
George Bos
By itself we can solve it by stating that these three instances
should be called: George-Bos-as-student,
George-Bos-as-researcher, etc. However, that is not what the
IS_A relation allows us to say. Although we may assume that this
IS_A is not an exclusive IS_A (which one would prefer it to be
anyway as it complies with Occam's razor), it does really
conflict with the notion of INDIVIDUAL to have more than one
instance.*) There may not be direct conflicts, because most
personal data of George Bos do not conflict with his teacher or
researcher data, so one obtains simply `additional' inheritance;
not further specialized inheritance. In fact this independence
also suggests that the role is not a further specification of
PERSON. Other problems may arise however if George Bos changes
roles (eg in a company), giving rise to the classical data-base
examples that are used to justify relational data-bases. I see
it as bad modelling: using short-cuts that satisfy specific use,
but it is not sufficiently generic (a classical problem in
knowledge bases; ontologies were supposed to remedy such
ad-hoc-ness and near-sightedness..).
In fact, in the SUMO ontology one finds a `correct' but utterly
ugly construction:
COGNITIVE_AGENT
PERSON_BY_SOCIAL_ROLE
PERSON_BY_OCCUPATIONAL_ROLE
RETIRED_PERSON
In this case a composite class (instead of instance) is
specified, so that again as many George-Boses can be
instantiated for each role. This example also may also show that
compounding terms should be avoided as much as possible. The
compounding, combining etc. occurs in modelling a situation rather than in
the elements (concepts, classes) with which we interpret the
situation.
[NB: this by itself may be an issue for discussion:
1) what are
`simple' classes and shouldn't we use only `simple' classes, at
least in a foundational ontology? and
2) there is not only the issue of compounding but also of
`modifying' (eg retired_person) and probably worse:
3) the use of negation in foundational ontologies (as in
`non-physical' etc. (even in DOLCE ;-))]
Therefore it seems advisable to separate ROLES from the PERSONs
that may perform a role, particularly if roles are incidental
(see below). There is a PERSON (INDIVIDUAL) view and a ROLE
view (to represent as multiple classification...)
My initial solution was to put also ROLES under AGENT.
However, this does not solve the problem. It may be true that in
modern jargon an individual may in fact consist of a `society
of agents' (etc.) but again a particular agent may be assigned
more than one role. One may say that any agent may again be decomposed
into a society of agents, but That may not be a sufficient reason, but
below I will give some other arguments.
CASE-2.
At the last SIG-1 meeting in Sardinia I presented an IS_A
hierarchy of roles in criminal law, excerpted as follows:
judicial_role
defendant
offender
convict
There was an objection that a convict was not a subclass of a
defendant **) I replied that convicts are subsets of defendants.
Which saved me for the moment, but I had overlooked a number of other
principles. My reply was correct from the point of view of
sets of individual persons (see above). However, as one can be
both a piece of furniture and a chair at the same time (that is
what abstraction is about) one cannot be a defendant and a
convict at the same time: worse the attributes of a defendant do
not inherit at all for a convict except his PERSON. The
set-subset relation I pointed to referred to the PERSONs taking
roles; not to the ROLES themselves (ie, in principle I made the
same shortcut as in Case 1...)
Therefore it seems that the relationship between roles - at
least these roles - is not an IS_A one but a DEPENDENCY one. The
roles are defined by criminal law procedures and these
procedures define the roles.
MY VIEW ON ROLES
These cases, but also a lot of other modeling problems I had
with roles, made me see ROLEs not as an active, agent-like
concept, but similar to a SET OF REQUIREMENTS ON BEHAVIOUR of an
INDIVIDUAL AGENT. So, I make a distinction between the
specification of a role (or the assignment of a role to a person)
and the actual execution of this role. Aldo Gangemi has pointed
me out that a similar distinction exists between a PLAN and the
actual execution of a plan (as is well known in robotics).
Particularly in legislation this distinction is evident. Norms
are generically specified, which means the do not address
individual persons, but role-players. The traffic regulation
addresses pedestrians and drivers of vehicles. Most norms
(articles) only concern drivers of motor vehicles. I may not
comply with a norm in the same way as I may play my role not as
assumed, specified etc. This distinction is necessary to be able
to state the difference between intended or prescribed behaviour
and actual behaviour. Also, when a person does not play his role
according to norms, in the legal world it means that an
individual person ***) is held responsible: not a role!
SOME OTHER ISSUES WITH ROLES
This solves only a part of the puzzle. Some observations:
1. Some roles are naturally related by subsumption (eg driver of
vehicle vs driver of car) but many others by dependencies created
by processes, actions or procedures (defendant vs convict) and
some others by part-of relations (
2. As roles specify behaviour they are closely related to
activities. In DOLCE/WonderWeb this is viewed as PARTICIPATION and
it covers roles as identified in case grammar
"The usual intuition about participation is that there are
endurants "involved" in an occurrence. Linguistics has
extensively investigated the relation between occurrences and
their participants in order to classify verbs and verbal
expressions. Fillmore's Case Grammar [Fillmore 1984] and its
developments (Construction Grammar, FrameNet) is one of the best
attempts at building a systematic model of language-oriented
participants." WonderWeb Deliverable D17, p 16
I am not sure whether these linguistic roles can be generalized,
i.e. that roles denote always participation in activities. Here
is what I wrote in an article (and when I believed that roles are
agents...):
"LRI-Core knows about the distinction between a person as a
lifetime identity and roles that a person may perform during this
life. Roles and persons are both agents, and agents are both
physical and mental objects. We need this perspective to be able
to understand what is meant by the generic statement that
``drivers of vehicles should keep to the right'': drivers are
roles that can perform actions~\footnote{To be precise, there are
two kinds of roles involved here: the role of a person to play
`driver' and the actor-role where the driver performs the drive
action. These latter roles are roles of actions, while the
former roles are roles of agents.}"
3. In my view roles are classes, but in many representations roles
are relations (In KL-ONE the term for attribute or relation is
role). I am not sure whether the distinction is of great
importance. Grammatical roles (actor,
recipient, object, etc) are represented as relations (slots) of
actions. Also, the defendant performs a role in a procedure (a
set of actions) but this kind of role is not only defined BY/IN
the procedure, but has also all kinds of rights and duties
associated, and probably many more attributes that suggest that
it is rather to be viewed as a class.
---------------
Footnotes
*) However, a system developed at the OU, also presented at this
EKAW (forgot the name; have no direct access here), that
generates webpages for a university department, will generate two
pages if George Bos is both a teacher and a researcher. The
personal data (age etc.) will be the same, but the aim was to
have webpages per person (with many roles) and not per class of
roles; of course it easy to adapt the web-page generator but I
view this rather as an ad-hoc remedy for a dubious
representation.
**) In fact, it was Bob Wielinga who made this objection; one of
the authors of `roles are not classes'...
***) Because the law is able to create virtual persons (judicial
persons) institutions may be viewed --under certain conditions --
as individual persons (and get for instance fines for polluting
the environment).
===========================================
References
@Article{ vanHeijst:97b,
author = "G. van Heijst and A. Th. Schreiber and B.J. Wielinga",
title = "Roles are not classes: a reply to Guarino",
journal = "International Journal of Human-Computer Studies",
year = 1997,
volume = 46,
number = "2/3",
pages = "311-318",
}
--
PLEASE REPLY to: breuker@swi.psy.uva.nl or
breuker@lri.jur.uva.nl
Joost Breuker
University of Amsterdam
SWI (Social Science Informatics) | LRI (Computer Science and Law)
Roeterstraat 15 | P-Box 1030
1018 WB Amsterdam | 1000 BA Amsterdam
tel: (+31) 20 525 3494/6798/6789
fax: (+31) 20 525 3495/6896
http://lri.jur.uva.nl:80/~breuker
This archive was generated by hypermail 2b30 : Mon 27 Feb 2006 - 15:50:02 EET