SOFTWARE ARCHITECTURE WITH EMERGENT SEMANTICS
How can systems be weakly coupled, but strongly referenced
Len Yabloko
Next Generation Software
Keywords: Semantic Web, Software Architecture, Grid Computing.
Abstract: This paper offers a unified modelling and computing pa
radigm with explicit (i.e. traceable to the facts)
epistemological support of semantic primitives. It builds on the method of logic stratification and alignment
first proposed by the author that allows to overlay the global computational grid with meta-space based on
distributed shared content addressable memory (a.k.a. Tuple Space). This paper shows a strong connection
between semantic, algebraic and topological properties of such meta-space, which makes it ideal conduit
directly relating semantics of primitives to grounding of enterprise processes in the global computational
grid.
1 INTRODUCTION
Emergent Semantics is a form of interoperability in
open and dynamic environments, such as global
computing grid, which is achieved incrementally
with negotiations conducted to reach agreements on
common interpretations within the context of a given
task (Aberer, 2004). Following is a brief outline of
its principles:
1) Agreements fulfill a protocol upon which shared
dy
nam
ic context can be constructed;
2) Assumptions must be constantly validated;
3) Global agreements result from aggregations of
local agreem
ents;
4)
Local interactions occur in the global context;
5) Agreements induce semantic self-organization;
Informally, self-organization
can be c
haracterized by
a complete distribution of control and by the
restriction to local interactions, information and
decisions. Global structures can then emerge from
such local interactions.
One particularly successful example of emergent
sem
a
ntics is link-based ranking as used in Google. A
global semantic agreement is obtained on “general
importance” of Web documents. We can compare
Google’s approach to the one taken by Web
directories, such as Yahoo. In web directories the
decision on importance of Web documents is taken
globally, manually and centrally. This clearly limits
the scalability of the approach.
A next natural step beyond ranki
ng-based m
ethods,
which ignore the structure of the content, would be
to apply the principle of emergent semantics to
obtain interpretations for structured data. One
possible avenue of how this might be achieved is
currently being opened in the area of peer-to-peer
(P2P) data management, where local schema
mappings are introduced in order to enable semantic
interoperability (Majkic, 2004). Such local schema
mappings perform as local communication
mechanisms for establishing consensus on the
interpretation of data. Once such infrastructures are
in place, the principles of emergent semantics
become directly applicable, and semantic structures
can build up from large numbers of purely local,
pair-wise interactions.
For practical reasons it is not possible to include in a
database
all the facts
pertaining to the objects in a
given application domain. The Closed World
Assumption (CWA) simply declares that all relevant
facts are stored in the database, so that any statement
that is true about the actual world can be deduced
from facts in the system. CWA in computer and
physical science have a lot in common, not only
because both are simplifying assumptions, but also
because in both cases the answers to some of the
most fundamental problems are deferred for
pragmatic reasons. In computers science such
problem is known as “Frame Problem”: under the
CWA any dynamic reasoning becomes non-
monotonic, requiring revision of a database (or more
broadly a belief base) for every new observation.
Simply not doing so is often (in error) considered a
monotonic reasoning. And this is why conventional
283
Yabloko L. (2005).
SOFTWARE ARCHITECTURE WITH EMERGENT SEMANTICS - How can systems be weakly coupled, but strongly referenced.
In Proceedings of the Seventh International Conference on Enterprise Information Systems, pages 283-288
DOI: 10.5220/0002529402830288
Copyright
c
SciTePress
information technology runs into the Frame Problem
when used on the Web. That fact resulted in current
attempts to develop so called “Semantic Web”,
which allows correct monotonic reasoning over
semantically fixed information, combined with non-
monotonic reasoning required by Web’s emergent
semantics. Semantic Web must provide following:
– The ability to compute the consensus semantics or
reality based on an analysis and aggregation of the
individual events observed.
– The ability to establish and maintain the scope if
each premise, which implies the ability to establish
the (global) identity of its source.
The rest of this paper is organized as follows.
Section 2 will focus on applying well known results
of research in non-monotonic reasoning to Emergent
Semantics. Section 3 will apply very recent results in
non-classical logic and Category Theory to
Emergent Semantics. Section 4 will introduce the
software architecture proposed by the author, and
Section 5 will discuss future possibilities.
2 LOGIC STRATIFICATION
In First-Order Logic (FOL), laws and facts are
propositions, and there is no special mark that
distinguishes a law from a fact. To distinguish them,
a context mechanism is necessary to separate first-
order reasoning with the propositions from meta-
level reasoning about the propositions and about the
distinctions between laws and facts. This kind of
separation in logic is called stratification.
The author of present paper had proposed (Yabloko,
2003) a method and algorithm for logic stratification
and alignment that is very much in line with (Sowa,
2002) proposal of stratifying non-monotonic
reasoning into FOL reasoning over the beliefs
represented in ontologies with graph-based meta-
theory of contexts. It’s major novelty, however, is in
direct epistemological support for semantic
primitives.
For illustration of its basic principle consider an
open set of possible worlds w
o,
w
1
w
2 …
with
accessibility relation R and semantic primitives a,b,
…, which form the alphabet shared among different
worlds. Although the meaning of each symbol is
grounded differently in each world, that is the
symbol can be used for very different purposes and
have very different significance associated with it,
all symbols relate to each other in a very specific
way in each world. Now, let’s assume that the only
requirement for a “safe passage” from one world to
another is that relations among some symbols (of
our choice) do not contradict those in other worlds.
This situation is not very different from the one that
take place when boarding an airplane: each
passenger carries his items of significance by the
screening machine at the gate, which determines if
those items are safe for flying. If certain item is
marked unsafe, then the passenger has a choice to
leave it at the gate or perhaps to travel by car. The
transport mechanism between the worlds, proposed
by the author, is called “causal stream” because the
relations in question are causal relations between the
events represented by each symbol. Note that these
relations do not have to be direct. The requirement is
that of all possible event chains in each world - none
violates causal relations between the given subset of
events. This requirement easily translates into
mathematical notion of partial order. In fact, the
entire idea is a variant of more general idea,
developed in Domain Theory (Karazeris, 2001), of
representing a process as a characteristic function
measuring the extent to which the event
(computation path) shapes can be realized by the
process.
The idea of “causal stream” has additional
epistemological significance and can be used for
consensus derivation in the emergent semantics
scenario. There it provides a direct connection
between semantic grounding of a process and its
physical (e.g. computational) properties. So
negotiating agents can adequately access the
consequences of their decisions, and not merely its
semantics. Dually, it reinforces the semantic
primitives with epistemological support in a form of
concrete physical properties of the word, which they
can express. That epistemological link defines the
architectural term “strong reference”, coined by the
author of present paper to emphasise its effect on
systems architecture, which becomes less brittle as a
result of adding plasticity of emergent semantics. It
is important to note, however, that the fundamental
idea of linking semantics of an expression with its
pragmatics (ie. usefulness) was introduced by Frege
in late 19
th
century as a sense of the expression or its
“mode of presentation”.
The above example shows how the belief revision in
non-monotonic reasoning can deal with uncertainty
resulting from incomplete knowledge. Indeed, the
transition from one possible world to another did not
require a complete knowledge revision, but only
affected that knowledge relying on a limited subset
of semantic terms. Moreover, the scope of such
revision is easily computable for as long as the
knowledge base is terminologically indexed, which
is a common technique, known as faceted taxonomy.
Thus, agents in emergent semantics scenario can
perform tractable reasoning and focused belief
revision under CWA.
ICEIS 2005 - SOFTWARE AGENTS AND INTERNET COMPUTING
284
3 SEMANTIC AGGREGATION
Essentially, aggregation is a form of meta-level
reasoning. In computer science meta-level reasoning
was introduced by the Type Theory, which is based
on the earliest form of logic stratification - called
ramification. The notion of logical context, or rather
of its philosophical predecessor – sense, introduced
by Frege, is totally absent in Russell’s theory. It is
replaced by the notion of the logical order of
proposition, and that has very profound implications
for semantics. Tarski’s Model Theory with its
declarative semantics can be seen as an attempt to
re-introduce the real world context into the
propositional logic.
It is worth recalling that classical (propositional)
logic came out of original “naïve” set theory as a
solution to a logical paradox in Frege’s
“Foundations of Arithmetic”, which Russell had
discovered and solved by means of ramification. The
paradox, however, was not exactly solved, but rather
deferred. In essence, the paradox, which Russell
himself called "vicious-circle principle," that is, "no
totality can contain members defined in terms of
itself", is a basic paradox of self-reference. Russell’s
solution transforms the totality into a logical
equivalent of mirrors hanging on both sides of a long
corridor (infinitely long in general case): each mirror
showing the picture of another mirror along and
across the corridor (such system of mirrors was
invented by Egyptians to bring the light deep inside
the famous piramids.) If each mirror is a set of
propositions, then its logical order corresponds to
the order of the mirror relative to the source of light.
The question deferred by this solution is: how to
transform logical propositions (corresponding to
observed light) into a theory (corresponding to the
World outside.) No such transformation is known!
But if we take what is arguably the most important
application of Set Theory: Relational Algebra, and
look at its most basic construct: binary relation from
the functional point of view provided by Category
Theory, we will discover a natural transformation
called pullback, which can be applied inductively to
transform any set of binary relations into a theory.
Scope does not allow us to provide a detailed
explanation of that fact, but a theory constructed this
way is intentional in that it applies to the extension
(i.e. the World) up to isomorphism (Vigna, 2002).
The inverse is also true: when we obtain the
extensional knowledge by making and observation,
the intentional a-priory cause is not knowable
because the observation itself is its causal
component. Pullback transformation corresponds to
irreversible process of semantic aggregation.
Aristotle in “The Categories” includes what we see
as effects (the results or consequences of the event)
as causes. His metaphysical framework included
four types of cause or explanation:
E efficient - that which makes a change happen,
M material - what the change happens to,
F formal - what the change results in, and
N final - the end or purpose of the change.
Figure 1 represents a categorical pullback of binary
multi-relations corresponding to four-way causality:
Aggregation
N Intentional
E Extentional E Time
M F M F
Strong Reference
Figure 1: Causal stream
Conjecture 1: For any process, represented as
causal strem, there is a group of natural
transformations which does not violate the stream.
This group is called “ strong reference”.
4 SOFTWARE ARCHITECTURE
Global Grid protocol architecture has five layers:
fabric, connectivity, resource, collective and
application (Foster, 2001). Only “collective” layer
includes some level of coordination between
services and resources. This situation is sometimes
called “hour-glass” architecture, in which the impact
of application and fabric on each other is not
controlled. This is the root cause of brittleness.
This paper proposes an alternative architecture that
removes the “most fragile” point of “hourglass” by
reinforcing collective layer with external “feedback”
loop (Fig. 2) corresponding to the strong reference.
Such feedback guaranties accurate cross-reference
from different domains in a virtual organization.
SOFTWARE ARCHITECTURE WITH EMERGENT SEMANTICS - How can systems be weakly coupled, but strongly
referenced
285
Logical
Tuples
Computation path Primitive Tuples
Tuple Space
Strong Reference
Collective Upper ontology (M3)
Application Ontology (M2)
Resource
(M1)
A
B
Connectivity
Tuple Space
Fabric (M0)
Class1
Class2
Class3
Class4
Class5
Class6
SU O _C lass
1
SU O _C lass
2
SU O _C lass
3
SU O _C lass
4
SU O _C lass
5
SU O _C lass
6
Aggregation
Typed Typed Typed
Tuple Tuple Tuple
Upper ontology
Ontology Stereotypes
Service
Model
Host Semiotic Fusion Host
Figure 2: Software architecture with emergent semantics. Bottom left (a): Logical organization of the Grid using M0-
M1 stratification (4L-MDA). Bottom right (b): Software stratification within a single domain (ie. single side of the Grid
Pyramid). Top left (c): Tuple Space serving as strong reference between the fabric and collective layer of the Grid. Top
right (d): Logical grounding of the resource and application layers of the Grid in federated Tuple Space with typed
tuples born by the connectivity layer of the Grid
ICEIS 2005 - SOFTWARE AGENTS AND INTERNET COMPUTING
286
It serves as reinforcing “fibre” that reduces system
brittleness. Classic four-side pyramid is shown only for
simplicity. In fact the number of domain is not limited.
Domains are separated by vertical edges of pyramid that
correspond to accessibility relations between domains.
The cylinders supporting the pyramid illustrate
connectivity to the fabric.
4.1 Service Grounding
Grounding connects the service model to
communication-level protocols, such as the message
description of Web Services Description Language
(WSDL). Although, the binding of atomic process to
WSDL operation is limited by XML-based
communication protocol (such as SOAP), this
binding does not preclude any additional bindings of
service to context (Ankolekar, 2001).
Figure 3: DAML Service grounding
Onto-Space provides additional binding of operation
to the context, and then uses Tuple Space
coordination mechanisms to enable strong reference.
Figure 4: OntoSpace Service grounding
The binding of operation to the context is provided
by type containers implemented by connectivity
layer. The same or different communication protocol
can be used, but the type systems used for this
purpose must provide an accessibility relation to the
service model. Semantic Web technologies seem to
be ideal fit for that purpose allowing DL-based and
other powerful type system (Horrocks, 2003) while
using XML for communication. In the OntoSpace
connected tuple spaces result in “streams” formed by
multirelation spans across the different domains.
For example, a frame-based ontology can support a
type constraints and inference required to create
simple Ontospace. Such system was, in fact,
developed by the author to represent a relational
database as a type container. It uses Protégé frame-
based environment (Noy, 2001) to implement Tuple
Space with logical tuples restricted to binary
relations. In it both primitive and logical tuples are
implemented as frames (Figure 11), Prolog
interpreter is used to implement type constraints, and
DL reasoner is used for type inference.
5 FUTURE WORK
It is an opinion of the author that future
computational grid must serve as a medium for
emergent semantics. As such it will need to
implement its principles in universal and consistent
manner. This paper introduced the idea of tuple type
systems that can be build using frame-based
SOFTWARE ARCHITECTURE WITH EMERGENT SEMANTICS - How can systems be weakly coupled, but strongly
referenced
287
containers (such as Protégé) in order to allow
tractable semantic aggregation. The effective
application of such semantic type containers will
require standardisation of tools around the idea of
strongly typed tuple space, similar to present
standardisation efforts in software application
(component) containers. Onto-Space project intends
to bring together presently isolated standards from
MDA and Semantic Web to that effect.
REFERENCES
Aberer, K., Catarci, T., Cudre-Mauroux, P., Dillon, T.,
Grimm, S., Hacid, M.S., Illarramendi, A., Jarrar, M.,
Kashyap, V., Mecella, M., Mena, E., Neuhold, E. J.,
Ouksel, A. M. , Risse, T., Scannapieco, M., Saltor, F.,
De Santis, L., Spaccapietra, S., Staab, S., Studer, R.,
and De Troyer O. (2004). Emergent Semantics. In
Proc. of the 1st International Conference on
Semantics of a Networked World; Lecture Notes in
Computer Science;3226; pp.14--43; Springer;
Heidelberg; ISBN: 3-540-23609-0.
Ankolekar, A., Hobbs, J. R., Burstein, M., Lassila, O.,
Martin, D., McDermott, D., McIlraith, S. A.,
Narayanan, S., Payne, T., Sycara, K.. (2001). DAML-
S: Service Description for the Semantic Web, In
Proc.of ISWC 2002, volume 2342 of LNCS, pages
348–363. Springer, 2002.
Foster, I., Kesselman, C., Tuecke, S. (2001). The Anatomy
of the Grid: Enabling Scalable Virtual Organizations
[Electronic version]. Intl. Journal of Supercomputer
Applications, 15(3), 2001
Horrocks, I., Patel Schneider, P. F. (2003, May). Three
Theses of Representation in the Semantic Web. In
Porc. of the 12th International World Wide Web
Conference: WWW2003
Karazeris P. (2001). Categorical Domain Theory: Scott
Topology, Powercategories, Coherent Categories
[Electronic version]. Theory and Applications of
Categories, Vol. 9, No. 6, 2001, pp. 106–120.
Majkic, Z. (2004, August).
Weakly-coupled ontology
integration for P2P systems.
In Proc of 1st
Workshop on Peer-to-Peer Knowledge Management:
P2PKM’04
Noy, N. F., Sintek, M., Decker, S., Crubezy, M.,
Fergerson, R. W. & Musen. M. A. (2001). Creating
Semantic Web Contents with Protege-2000. IEEE
Intelligent Systems 16(2):60-71.
Sowa, J. (2002, May). Laws, Facts, and Contexts:
Foundations for Multimodal Reasoning.
Knowledge
Contributors
, edited by V. F. Hendricks, K. F.
Jørgensen, and S. A. Pedersen, Synthèse Library vol.
322, Kluwer Academic Publishers, Dordrecht, 2003.
Retrieved from
http://www.jfsowa.com/pubs/laws.htm
Vigna, S. (2002). Multirelational Semantics for Extended
Entity-Relationship Schemata With Applications
[Electronic version]. In Proc. of the 21st International
Conference on Conceptual Modeling - ER 2002
Lecture Notes in Computer Science;2503; pp.35--49;
Springer; Heidelberg; Retrieved July, 2004 from
http://vigna.dsi.unimi.it/ftp/papers/Multirelations.pdf
ICEIS 2005 - SOFTWARE AGENTS AND INTERNET COMPUTING
288