Knowledge in Formation
Summary
The aim of this project is the development of a cognitively based,
semiotical model of information processing with applications in
(cognitive) modeling, language parsing, meaningful summarization
and ontology design.
Information is conceived by us as the knowledge increment brought about on the occasion that an actor of
whatever nature receives input. Thus interpreted it presupposes an interpretational process in which data eventually result in meaningful responsive behavior, but at least add something to the stock of knowledge
present in the interpreting agent.
The primary goal of this research project is to give an account of the non-accidental properties of those
interpretation processes. A secondary goal is the design of a combinatorial engine fit for computational
interpretation. The later in order to use the engine for modeling cognitive behavior in different domains.
Since the machinery itself is simple (it offers nine types of combinations recursively), it enables
the systematic specification of domains by explication of the combinatorial properties of the signs that
make up the interpretation processes in the domain to be specified.
Theoretical background
At the heart of our research program reside three basic assumptions:
- Cognition is relational. On the one hand this means that mind is part of the reality it tries
to understand on the other this means that reality cannot be reduced to matter.
- An account of cognitive processes must be
teleological.
Any agent interprets data offered from the perspective of the goals, inscribed in it through past experience,
that happen to be operative in the process of their interpretation.
- Habitual behavior is rule governed, thus it is, in principle, capable of a computational
interpretation. Due to assumptions 1. and 2. this does not only mean that the interpreting mind must be
modeled (the combinatorial machine), but also that the conditions that give rise to the execution of
the habits must be modeled (the stock of knowledge and operational goals required).
A process model of cognitive activity
Even a quick scan of the intellectual history of man reveals that there is a close connection between
informational progress
and sign use. Just think about the sequence: artifacts and speech; handwriting; print and digital media,
and consider the effects each sign-technique has on the intellectual habits of its users [BR04].
The intimate relation between signs and thought can be generalized in the notion that every idea is only manifest
by being of the nature of a sign, so every thought is a sign and only on account of this feat an
interpreter is able to think.
But then it must be possible to study thought processes by studying the ways in which signs give rise to new signs
and by doing so realize a knowledge increment.
These assumptions place our program in the American pragmatist tradition of the
Peircean flavor in which the concept of sign plays a central role.
In conformity with the above we assume that knowledge arises from the cognition of phenomena by means of signs.
This primarily requires an interaction between the (input) phenomenon
functioning as a sign, and the observer as an interpreting system.
As a result of this interaction the phenomenon effects the observer
by changing its
sensory state. This change
again can be interpreted as a sign, for example by the brain,
by linking it with existing knowledge about similar changes and experience with
earlier response strategies, in order to generate an adequate response to
the observed phenomenon. This way, the phenomenon, which itself is
a sign of some interaction between phenomena, is re-presented through
a sequence of interactions between signs, starting with the input sign that
is offering itself for interpretation (response) and the interpreting
or dynamical sign that is being affected (interpreting system).
In virtue of the goal oriented nature of cognitive
activity, the interaction events define a process, which is the
(re)cognition of the input phenomenon as a meaningful sign to the end of generating a response.
Our model describes this process (see
For a definition of the model see:
The Machine in the Ghost
(to be found in Section "Articles 2009").
First exploration
Since the process model of cognitive activity explicates what must be assumed in order to account for
any cognitive process whatever, the model is general and turns out to be simple. Thus a heavy burden
is laid on the available knowledge or the lexicon, the items of which must contain the combinatorial
properties that enables them to get combined with the input data in order to derive a correct interpretation.
In the brain, so we assume, the way in which the neural tissue gets organized in the course of growing
experience and the activation patterns that get associated with the different occasions for input processing,
account for the lexicon and the combinatorial machine alike. Or, to put it in other words, in our nervous
system there is no identifiable place where a combinatorial machinery is located that crunches data that
are stored somewhere else and retrieved upon the proper occasion. Instead, the machinery is
coexistent
with the instances of processes of cognitive activity. In our semiotical process model we present an abstraction,
it specifies the moments present in all informational processes without going into the details of its physical
realization. From the point of view of the applicability of the model to specific domains this semiotical model
functions as a schema from which domain and/or actor specific (human, IT-based, organizational actors) models
can be derived.
For computational purposes we developed a 'naive' logical interpretation of the process model
in which each interaction discerned in the process is provided with a logical operation, see
[SF06], [FS00]. Through a set theoretical interpretation this enables a model in which signs can be
defined as sets and sign interactions as operations on sets and so it offers the promise of
the realization of applications. A drawback of this logical interpretation however consists
in the limitations it imposes on the scope of the model. In the logical interpretation the model
is only of use for what already is
habitual.
But, what already is habitual, goal governed behavior, is feasible for specification.
In order to derive a specification one first has to identify the goal and the input
that triggers the associated response or interpretation process. In a second phase the lexicon
is built by specifying the elements needed for each of the steps that collectively constitute
the habit. If sub-goals are involved, the associated sub-processes are specified along
the same lines recursively. The specification ends when a
lexicon
is specified that is sufficiently rich to reach the goal(s) on the proper occasions.
Research
We explored the corners of the envelope in different domains: syntactic parsing, problem specification,
modeling and text summarization. Since our main aims consisted in the first phases of the project in
an enhancement of our understanding of the processing schema and an assessment of the value of
the sign theoretical paradigm we work with, we preferred breadth over depth. Some words about
the results so far:
- Syntactic parsing
We developed a parser for natural language syntax conform the processing schema and tested it with
a simple lexicon. From the linear complexity of the parser, and the syntactic nature of computational
processes in general, it follows that all computational interpretations of our schema can be linearly
complex.
As a spin off we gained a greater understanding of the different kinds of combinatorial needs of
the lexical items.
- Problem specification
The processing schema specifies the distinctions pertinent to all sign processing, the
way
in which they are ordered and the ways in which they can be related . Since the schema professes to be general, it is to be expected to be present in the conceptualization of any problem. But, as it is an abstraction, it is devoid of content. This offers the possibility to use the empty schema as a heuristic device. In order to systematically specify a problem one has to formulate a goal, vague as it may be at the outset, and fill in the slots of the schema in such a way that the wished for result can be generated. In the course of specifying a problem sub-goals will pop up and as a consequence iterations are needed. If everything needed to reach the goals is specified, the problem is specified.
As a spin off we gained better insight in the way in which to deal with iterations of the schema.
- Organizational modeling
At the present time there are some pretty good business modeling techniques. Communication oriented methods like NIAM/ORM excel in retrieving and validating knowledge from domain experts and transforming the results into a consistent model (modeling how it is). A notable problem however with those approaches is the measurement of the semantic quality of a model in cases of business re-design (modeling how it ought to be or must become). It is with respect to this problem that the problem specification capabilities of the process model can be beneficial by providing a top down specification of the processes to be modeled, thus complementing the bottom up technique of NIAM/ORM.
As a spin off we refined our process notion, we gained more understanding of the different types of actor (IT-based, human and organizational) and in their interrelations.
- Meaningful summarization
Meaningful summarization as distinguished form statistical text summarization is very demanding since it involves the processing of text from multiple perspectives. It involves morpho-syntactical, syntactical, semantical, logical and discourse analysis of the text, to name just a few of the perspectives needed. To be frank, we only scratched the surface see [SF04]. However, since our processing schema entails the promise of a uniform representation for all domains pertinent to sign processing, it becomes feasible that the complex task of merging knowledge from different domains, can be reduced to the more simple task of structural co-ordination of domain specific instances of the processing schema.
We feel that the assumption that there exists a uniform mechanism for all domains of sign processing, is strengthened by
experimental evidence
of cognitive research. The results point out that at least in the syntactical and semantical domain, the brain performs a simultaneous analysis of the data offered. But, as the (full) meaning of the input may require an interpretation in both domains and the temporal difference between syntactic and semantic analysis is small (making a translation between the representations in the domains unlikely), the use of a uniform representation, at least in these domains, might be inevitable.
Looked at in this way meaning summarization is not different from interpretation. Hence, our highest goal is to reveal the calculational properties of semiosis or Knowledge in Formation.
Besides domain oriented research a theory oriented line of research has been pursued. Three interrelated topics surfaced.
- Semiotic foundations
Our work builds on the foundations laid by Ch. S. Peirce (1839 -1914). Many of the valuable sign theoretical manuscripts are still not easily accessible. Currently we are working on a derivation of our process model from those foundations [BS07].
- Modes of realization, its consequences and the possibilities offered
Sign interpretation processes are executed by dynamical signs or actors. Different types of dynamical signs can be distinguished. The order below is from more encompassing to more specific.
- Organizations as actors
Since the organization considered as an actor contains human and IT-based actors, this is the most encompassing mode of sign processing. It calls for an investigation of the interactions between the acting agents. Besides that, it is the domain in which the increasingly important question of architecture in information sciences finds its place.
- The Human actor
Studying the ways in which humans generate meaningful responses is a key to a deeper understanding of informational processes. On the one hand research in the cognitive sciences offers a possibility to
test
the model. On the other hand this kind of research offers material for attempts to rebuild the processes identified, but in a computational way.
- IT-based actors
The most interesting part of the development of IT-based actors consists in the urge it contains to be specific. Success and failure will be informative with regard to what can be modeled and what can not at the moment. Thus enhancing our understanding of ourselves.
- Domain specific applications
See the first part of this section for the domains we probed. An important function of domain specific research from the point of view of theory consists in the effects it has on the work on semiotic foundations.
