Similarities and differences between NKRL and ICKMOD

It’s nearly impossible to trace the sources of one’s own ideas without (1) a near-photographic memory, (2) relentless and scrupulous note-taking, and (3) a great desktop search engine … and consistently give credit where it is due. I have only the third benefit (dtSearch Corporation’s dtSearch Desktop) … and it is not nearly sufficient to achieve my goal. So I will apologize once again for not giving adequate credit to the many wonderful thinkers and technologists whose ideas I have leveraged in my writings. And I earnestly solicit your pointers to my errors, to sources that I have missed, and to experts to whom I have given insufficient credit.

So, in the spirit of giving credit where credit is due …

The fundamentals of the Insight-Centric Knowledge Model (See The Insight-Centric Knowledge Model — ICKMOD.) described in this resource are more closely aligned with Gian Pierro Zarri’s Narrative Knowledge Representation Language (NKRL) than with any other single model, school of thought, or implementation. Zarri’s work is described at length in his book, Representation and Management of Narrative Information: Theoretical Principles and Implementation (2009), and more recently. “Advanced computational reasoning based on the NKRL conceptual model,” which was published in the June 15, 2013, issue (Volume 40, Issue 8) of Expert Systems with Applications. Prof. Zarri was kind enough to send me a draft copy of the paper in March, 2013.

I strongly recommend both of these publications, but the first chapter of Zarri’s book merits attention by itself. It is the best-researched and most thorough review of relevant literature that I have ever encountered.

Although I developed the basic principles of ICKMOD before I encountered Zarri’s work, I am ecstatic that I have found so much confirmation of my thinking in Zarri’s work. That’s the good news. The bad news (for me, anyway), is that I must now be very careful not to think that my ideas are original when they already exist in Zarri’s work.

Differences in terminology

My interpretation of the differences between ICKMOD and NKRL may be incorrect at times because of differences in the vocabulary we use and in differences in our goals and emphasis.  Among the differences in vocabulary, perhaps the most important are:

  1. Zarri’s functional roles are equivalent to the controlling syntax of representations of meaning in ICKMOD. And I use the term Function as a type of Fact.
  2. Zarri uses the term Event to describe elementary (and often non-static) statements of the meaning of conditions and changes in the real world. I use the terms Fact and Observation, among others.
  3. In my terminology, Facts and Insights are combined into larger structures (which I have labeled Theses).

I do not claim that my terminology is better, and I think that the word syntax is often used with very different meanings. (For me, it’s close to the notion of grammar in natural languages.) I just want everyone to use the same terminology with the same meaning. That, ironically, is one of the big problems facing all disciplines devoted to making better use of the meaning behind communications.

High-level similarities

  • Both Zarri and I emphasize the complementary roles of static, controlled vocabularies (ontologies, in many cases) and formal representations of complex conditions and events in the real world — what I call Facts and others may refer to as propositions — in which the Concepts from those vocabularies participate.
  • We both believe, I think, that the N-ary relationships expressed in such propositions cannot be deconstructed into binary relationships and successfully reconstructed without loss of some of the original meaning. Zarri’s explicit predicate-based template approach implies a high-level semantics that binary relationships cannot adequately represent. Deconstruction of such propositions into binary relationships is possible but not particularly useful for capturing knowledge.
  • Direct support for reification (referencing prior representations of propositions) is essential to usable, effective representation of complex assertions about reality. Reification is essential for many reasons, including constructing “narratives” (in Zarri’s terminology).
  • Both NKRL and ICKMOD are fundamentally about aligning deep-structure representations of meaning with real-world conditions and events. Predicates, in Zarri’s words, are “deep predicates,” totally independent from “surface” syntactic considerations of, for example, the active or passive type. Meaning expressed in this way transcends the boundaries of natural languages.(In “Chomskian Linguistics” (, Colin John Holcombe observes, “Deep structure is the abstract underlying form, which determines the meaning of a sentence. Surface structure is what we write or speak. The two are connected by transformations like combination, addition and deletion. Or so Chomsky first argued. But in his Reflections on Language, Chomsky drew up something much more complicated. There were two structures or trees: one for deep and one for surface sentences. Transformation rules linked the two. Ambiguous sentences had two deep structures. Now the sequence was: The base tree was constructed with building rules and a lexicon. The transformation component mapped deep structures onto surface structures. A phonological component intervened to convert surface structures to surface sentences.)

Some high-level differences between NKRL and ICKMOD

Zarri’s NKRL is a much more mature, complete, tested model than ICKMOD, and it is implemented in a working product. There are many more differences between NKRL and ICKMOD than those described below, in part because many details of the ideas and models, in Zarri’s work simply have no counterparts in ICKMOD. You will need to read his publications to grasp the full extent of the differences. I strongly recommend doing that, even if you have no interest in ICMKOD.




A fully-formed language and a “fully implemented computer science environment.” An evolving “abstract reference model” designed to accommodate complementary software applications that help people work better.
Top-down specification of controlled vocabulary and hierarchical “templates” for predicative occurrences (instances of deep-structure propositions) Incremental formalization — starting from existing natural-language resources and adding structure and connections as that becomes possible.
Formal representation of [non-fictional] narratives, which Zarri defines as spatio-temporally bounded streams of elementary events.
“Pre-formal” representation of ad hoc, open-ended, multi-purpose knowledgebases. The boundaries of the knowledgebase are established by judgments of relevance by participants.
Hierarchical predicative occurrence structure with seven top-level templates (BEHAVE, EXIST, EXPERIENCE, MOVE, OWN, PRODUCE, RECEIVE) which govern the concepts that may participate in (play roles in) in formal statements about the real world.

Specialization of these templates creates an inheritance hierarchy of (in a sense, an ontology of) predicative occurrence types. Also referred to as “generalized predicates” by Zarri.

Assumes that predicate hierarchies will evolve from the collective communications of the individual or group concerned with the knowledge developed.

Does not postulate a set of universal high-level predicates from which other predicates may be derived by specialization. However, ICKMOD does not assert that this is wrong or impossible.ICKMOD is neutral with regard to theories of grammar. For example, the “controlling” element in any assertion — that is, an element that limits what other elements might meaningfully be incorporated into an assertion — might be a particular thing, not a particular predicate.

In the world of semantics, theorists focus on predicates “controlling” other elements in an assertion. But in the real world, we rarely remember meaningful events based on predicates; more likely specific things … and only then the predicates that connect them with other things.

All representations form a single stream of data. Higher-level representations should distinguish among “Dimensions” of (1) meaning, (2) evaluation of representations of meaning by participants, and (3) occurrences of Insights and Facts in documents.

The distinction between (1) and (2) is an important characteristic of argumentation/discussion systems (see

I also add an “Adminstrative” Dimension in my recent sample implementations.

A narrative can be represented as an NKRL application. Strong bias toward arc-and-node (arrows and circles) paradigm — a form of Directed Acyclic Graph. All nodes and relationships should have unique, persistent IDs and may have metadata — information that may be useful in many applications designed to represent meaning.
Does not specify broad categories of elementary Events by communication purpose. “Fact Types” that correlate with common types of statements about reality, including Function, Observation, Circumstance, Definition, Characterization, Set Definition, and Event. Facts can participate in multiple Insights. ICKMOD Fact Types could be viewed as complementary to NKRL templates.
Supports rules, inferencing, and queries. No direct support for rules, inferencing, or queries. Evaluation of “truth” is done by humans, supported where possible by complementary semantic applications. The formalization relative to natural language should assist in the development of many different applications.

© Copyright 2017 Philip C. Murray

This entry was posted in ICKMOD, model for representing knowledge. Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *