OWL: the original

In the early 1970's, Prof. William A. (Bill) Martin of MIT's Project MAC (later the Lab for Computer Science) began development of a powerful knowledge representation language called OWL.  Although the acronym was never officially defined, many of us who participated in the project believed it to stand for "One World Language." Throughout several incarnations, OWL was used to support research on

The heart of the system is a formal language for knowledge representation that took a Wittgenstein-like approach to knowledge.  Approximately, this says that the meaning of any concept in the language is the totality of all the other concepts linked to it. Concepts were formed by specialization, OWL included a derivative subclassifier that made certain taxonomic inferences, and a very general notion of characterization allowed great flexibility in expressing assertional knowledge.  Critically, all characterizations could be reified as concepts, allowing the system to use meta-level descriptions of its own features.  Inference methods could be specified within the language itself, or could be implemented as Lisp procedures for efficiency.

Bill unfortunately became gravely ill in 1980, and died in 1981.  As a result, the OWL group dispersed to other institutions and research efforts, and development on OWL itself ceased.  Details of OWL were not fully published, but [Szol87] gives a comprehensive overview of the system and [Hawk85] describes the classifer.  [Mart81] is an incomplete draft of a book Bill was writing describing his insights into the subtle but profound relationships between knowledge representation and structure and metaphors in natural language.  The other references below show uses of OWL in various research projects and thus shed additional light on its capabilities.

OWL for the Semantic Web

In 2002, I was asked if the name OWL could be re-used for a new formal knowledge representation language in support of the Semantic Web, based on a modern implementation of a description logic such as DAML+OIL.  I believe this is a highly fitting tribute to Bill's pioneering work in knowledge representation.

OWL-related Bibliography

... to be completed...

Brown, G. P. (1975). Instances in OWL, MIT: 33.

This memo reflects a time slice in OWL development. Some of the constructs and concept classes assumed to exist show sign of obsolescence, and some new classes and constructs will probably be introduced. References to <Martin 1975> are to a August 4, 1975 draft (along with WWORLD > of August 15) and do not necessarlly reflect hls current thinking or thc for~ of the final vcrsion. Simllarly, references to the OWL i~ple~entation reflect ny understanding of its state prlor to August 1, 1975. I a~ puttlng thls me~o out now because the sorts of OWL changes that look imminent would change the detall, but not thc spSrit, of the papcr. A revised version -- based on feedback from this version and OWL changes that occur o~er the next few months -- will be forthconing.

Haimowitz, I. J., R. S. Patil, et al. (1988). Representing medical knowledge in a terminological language is difficult. Proc. Twelfth Annual Symposium on Computer Applications in Medical Care: 101-105.

Hawkinson, L. B. (1975). The Representation of Concepts in OWL. Proceedings of the Fourth International Joint Conference on Artificial Intelligence.

Hawkinson, L. B. (1980). Xlms: A Linguistic Memory System. 545 Technology Square, Cambridge, MA, 02139.

Long, W. J. (1977). A Program Writer, Laboratory for Computer Science, Massachusetts Institute of Technology.

Martin, W. A. Automatic Programming and Mathematical Software.

Martin, W. A. III. Syntactic Structures.

Martin, W. A. Interactive Systems -- Theories of Implementation.

Martin, W. A. Introduction to OWL II.

OWL II is a language for knowledge representation under development at the MlT Laboratory for Computer Science [Szolovits, et al., 1977]. OWL II is based on the Linguistic Memory System (LMS) developed by Hawkinson (1975). A guiding principle of the language is that it should mimic the knowledge representation conventions of English speakers to the extent that theories of these conventions can be developed. In [Martin, 1978], I established basic conventions for the construction of representation heirarchies in OWL II. In this paper, I go on to establish conventions for the construction of descriptions which correspond to English words, phrases, and sentences.

In OWL II, we take the view [Vygotsky (1962), Whorf (1956)] that language provides a framework for the organization of knowledge. This paper discusses properties which are frequently attributed to such frameworks. A set of primitives and operations by which such frameworks can be generated is suggested.

Further embellished, such frameworks underly the proposed representation of English words, phrases, and sentences.

Martin, W. A. Problems and Possibilities for Near Term Natural Language Data Base Query Systems.

Martin, W. A. Semantic Structures.

OWL II is a language for knowledge representation under development at the MIT Laboratory for Computer Science [Szolovits, et al., 1977]. OWL II is based on the Linguistic Memory System (LMS) developed by Hawkinson (1975). A guiding principle of the language is that it should mimic the knowledge representation conventions of English speakers to the extent that theories of these conventions can be developed. In [Martin, 1978] I established basic conventions for the construction of representation heirarchies in OWL II. In this paper, I go on to establish conventions for the construction of descriptions which correspond to English words, phrases, and sentences.

Martin, W. A. (1971). "Determining the Equivalence of Algebraic Expressions by Hash Coding." Journal of the Association for Computing Machinery 18(4): 549-558.

Let S be the set of rational exponential expressions with complex rational coefficients, single level exponentiation, and no division or element i in the exponents. Let p be a prime integer of the form 4q+1, where q is also prime. The expressions in S can be matched for algebraic equivalence by substituting random integer values for the variables, evaluating the exponents mod (p1), and evaluating the rational expressions mod p. When this is done equivalent expressions will evaluate to the same result; while, in typical situations, two expressions selected at random will evaluate to the same result with probability about 1/q. Thus, if p is taken close to the word size of a 36-bit machine, the probability of matching equivalent expressions is 1 and the probability of random match is about 10^-9. The problems of extending this scheme to allow division in the exponents and preserve the relation e^i-{pi}=1 are studied. No solutions to these problems were found, but a scheme which handles some cases by special case checks is presented. This scheme was implemented in a program for algebraic simplification.

Martin, W. A. (1971). "Sorting." Computing Surveys 3(4): 147-174.

The bibliography appearing at the end of this article lists 37 sorting algorithms and 100 books and papers on sorting published in the last 20 years. The basic ideas presented here have been abstracted from this body of work, and the best algorithms known are given as examples. As the algorithms are explained, references to related algorithms and mathematical or experimental analyses are given. Suggestions are then made for choosing the algorithm best suited to a given situation.

Martin, W. A. and R. Krumland (1972). MAPL: A Language for Describing Models of the World. Cambridge, MA, MIT.

Martin, W. A. (1972). A Data Set Language and Its Translation into IBM 370 PL/I (F), Automatic Programming Group.

Martin, W. A., R. Krumland, et al. (1973). More MAPL: Specifications and Basic Structures. Cambridge, MA, MIT.

Martin, W. A. (1973). Translation of English into MAPL Using Winograd's Syntax, State Transition Networks, and a Semantic Case Grammar. Cambridge, MA, MIT.

Martin, W. A. (1973). MAPL 2. Cambridge, MA, MIT.

Martin, W. A. (1974). OWL Notes Chapter 1: The OWL data structure and the OWL data base, Massachusetts Institute of Technology.

Martin, W. A. (1974). OWL Notes, Introduction: A System for Building Expert Problem Solving Systems Involving Verbal Reasoning, Massachusetts Institute of Technology.

Martin, W. A. (1975). Conceptual Grammar.

Martin, W. A. (1976). A Computational Approach to Modern Linguistics: Theory and Implementation.

Martin, W. A. (1976). Notes on Sentence Structure.

In these notes we are concerned with how an English sentence should be parsed as a first step in computer comprehension of English. We explore some of the structure of English sentences, produce a model of them, and structure the computer's semantic memory to exploit it.

Martin, W. A. (1976). A Theory of English Grammar.

The constructs specialization, modification, nexus, naming, productivity, and slot shift are defined and a theory of English grammar advanced in terms of these and related ideas. The theory provides a basis for the computer representation and manipulation of knowledge which can be conveyed explicitly by English sentences.

Martin, W. A. (1977). Interactive Systems---Theories of Implementation.

Martin, W. A. (1978). Descriptions and the Specialization of Concepts, Massachusetts Institute of Technology.

The OWL II system computes with expressions which describe an object from a particular viewpoint. These partial descriptions form a tree structure under the specialization operation, which preserves intensional properties. The descriptions are also related in terms of their extensions by characterization and exemplar links. Descriptions of individuals must always specify a context of the description. Eight ways in which one description can be a specialization of another are distinguished.

Martin, W. A. (1978). Metaphysical and Epistemological Foundations for a Linguistically Oriented Semantic Network.

Finally going one step higher, we might consider networks whose primitive elements are language-specific. The only formalism that I know of at the current time that embodies this view is OWL, whose elements are expressions based on English. In such a formalism, one would presumably "...take seriously the Worfian hypothesis that a person's language plays a key role in determining his model of the world and thus in structuring his thought" [Martin 1977, p. 985]. In OWL, there is a basic concept-structuring scheme (see [Hawkinson 1975]) which is used to build expressions, and strictly speaking, the principles of "specialization", "attachment", and "reference" are the primitives of the language. However, these primitives are neutral enough to be considered implementational, and thus the knowledge itself can be considered to form the structure of the data base. This seems operationally reasonable when OWL is looked at in detail -- the two expressions, (HYDRANT FIRE) and (MAN FIRE), while both specialized by FIRE, can have the specializations "mean" different things based on the rest of the network structure. This linguistic level represents perhaps the most radical view of semantic nets, in that the "primitives" are language-dependent, and are expected to change in meaning as the network grows. Links in linguistic level networks stand for arbitrary relationships that exist in the world being represented.

Martin, W. A. (1978). "Some Comments on EQS, a Near Term Natural Language Data Base Query System." ACM: 156-164.

Problems and possibilities for near term natural language query systems are discussed, with emphasis on the author's own system, EQS. First, the general objectives for near term systems in the areas of syntax, world knowledge, discourse, and problem solving are considered. Next, a comparison is made between the ATN parsing strategies In LADDER, ROBOT, PLANES, and EQS. Evidence for the importance of giving answers to queries not directly available In the data base is given together with some speculation on how the knowledge necessary for this might be acquired. After some general discussion of semantic data models, the semantic data model used in EQS is described, and the method used for building a semantic data model of an existing CODASYL data base is sketched. Lastly, the use of multiple levels of representation as a way to control complexity of such query systems is discussed.

Martin, W. A. (1979). Descriptions and the Specialization of Concepts. Artificial Intelligence: An MIT Perspective. P. H. Winston and R. H. Brown. Cambridge, MA, MIT Press: 379-419.

This section by William A. Martin describes OWL II, a system that computes with expressions that describe an object from a particular viewpoint. These partial descriptions form a tree structure under the specialization operation, which preserves intensional properties. The descriptions are also related in terms of their extensions by characterization and exemplar links. Descriptions of individuals must always specify a context of the individual. Nine ways in which one description can be a specialization of another are distinguished.

Martin, W. A. (1979). How to Use the Syntactic Analyser.

Martin, W. A. (1979). Philosophical Foundations for a Linguistically Oriented Semantic Network.

Martin, W. A. (1980). A Logical Form Based on the Structural Description of Events.

Martin, W. A. (1980). Natural Language and Representation of Knowledge.

Martin, W. A. (1981). Proposal to the Defense Advanced Research Projects Agency; For Research in Natural Language Processing and Computer Representation of Knowledge.

Martin, W. A. (1981). "Roles, Co-Descriptors, and the Formal Representation of Quantified English Expressions." American Journal of Computational Linguistics 7(3): 137--148.

A scheme is proposed for representing the logical form of English sentences, wherein the meaning of a network node depends on how it is related through "role-in" links to nodes representing more aggregate entities. It is argued that roles are a natural device for capturing many linguistic and philosophic distinctions, and that they are convenient for computational processing. In particular, it is shown how role-in links may be advantageously used in lieu of quantifier scope to represent quantificational dependencies.

Martin, W. A. and M. Bosyj (1976). Requirements Derivation in Automatic Programming. Symposium on Computer Software Engineering, Polytechnic Institute of New York.

Once a consultant has decided on an approach to the solution of a problem, he performs a feasibility analysis to verify that he has indeed found a solution. A good analysis usually requires the complete investigation of the client's firm. In practice, however, this investigation is seldom complete: only a few experts have a detailed knowledge of the factors that must be considered for an acceptable analysis.

We have implemented a program to assist the Operations Management consultant in performing such feasibility analyses for procurement problems that do not involve distribution or production activities. The approach to the solution of a procurement problem is assumed to be the hierarchical framework for planning and control systems advocated by Hax and Meal. The program is based on a questionnaire (written by Arnoldo C. Hax) which is essentially a summary of the kinds of questions that should be asked when deciding whether the hierarchical framework is applicable.

The program is implemented as an "unstructured" questionnaire that guides the user in investigating various aspects of a problem while giving him complete freedom in deciding how and when to supply answers to questions. It allows him to change and skip answers whenever he desires. The program makes use of the OWL data base, a linguistically based conceptual hierarchy, to represent procedures for the questionnaire and to store data accumulated during the interaction. This data representation makes possible the presentation in English of a problem description, various evaluations and the reasons for the evaluations.

The system should prove useful to a consultant as a data gathering device, As an educational tool, it serves to illustrate the kind of information required for the successful analysis of a procurement problem.

Martin, W. A. and G. P. Brown (1977). A Preliminary Design for Data Base Modeling in OWL.

This report discusses the OWL Data Base Modeling Project being carried out in the LCS Knowledge Based Systems Group. This is a first pass at a description of the project to clarify its scope, goals, and the resources that it will need. Special attention is given to the interface with the user, in particular to the range of natural language forms that must be handled.

Martin, W. A., K. W. Church, et al. (1981). Preliminary Analysis of a Breadth-First Parsing Algorithm: Theoretical and Experimental Results, Massachusetts Institute of Technology.

We will trace a brief history of context-free parsing algorithms and then describe some representation issues. The purpose of this paper is to share our philosophy and experience in adapting a well-known context free parsing algorithm (Earley's algorithm [8, 9l and variations thereof [28, 13, 26, 27]) to the parsing of a difficult and wide ranging corpus of sentences. The sentences were gathered by Malhotra [22] in an experiment which fooled businessmen users into thinking they were interacting with a computer, when they were actually interacting with Malhotra in another room. The sentences are given in Appendix I. The Malhotra corpus is considerably more difficult than a second collection given in Appendix II (originally published in [15]). Section 4 compares empirical results obtained from these collections against theoretical predictions.

Martin, W. A., K. W. Church, et al. (1981). Preliminary Analysis of a Breadth-First Parsing Algorithm: Theoretical and Experimental Results. 545 Technology Square, Cambridge, MA, 02139.

Martin, W. A. and J. Moses. Why Programmers Make Bad Models of the Real World.

Martin, W. A. and P. Szolovits (1981). Semantic Networks in LISP: Fundamental Concepts and A Specific Implementation.

Patil, R. S. (1979). Design of a Program for Expert Diagnosis of Acid Base and Electrolyte Disturbances, MIT.

Patil, R. S. (1981). Causal Representation of Patient Illness for Electrolyte and Acid-Base Diagnosis. Cambridge, MA, MIT Laboratory for Computer Science.

Patil, R. S., P. Szolovits, et al. (1981). Causal Understanding of Patient Illness in Medical Diagnosis. Proc. Seventh International Joint Conference on Artificial Intelligence: 893-899.

Patil, R. S., P. Szolovits, et al. (1982). Information Acquisition in Diagnosis. Proc. National Conference on Artificial Intelligence. Pittsburgh, Pennsylvania, American Association for Artificial Intelligence: 345-348.

Patil, R. S., P. Szolovits, et al. (1982). Modeling Knowledge of the Patient in Acid-Base and Electrolyte Disorders. Artificial Intelligence in Medicine. P. Szolovits. Boulder, Colorado, Westview Press. 51: 187-222.

Swartout, W. R. (1977). A Digitalis Therapy Advisor with Explanations.

Swartout, W. R. (1981). Explaining and Justifying Expert Consulting Programs. Proceedings of the Seventh International Joint Conference on Artificial Intelligence.

Swartout, W. R. (1981). Producing Explanations and Justifications of Expert Consulting Programs.

Swartout, W. R. (1983). "XPLAIN: A System for Creating and Explaining Expert Consulting Programs." Artificial Intelligence 21: 285-325.

Szolovits, P., L. Hawkinson, W. A. Martin. (1977). An Overview of the OWL Language for Knowledge Representation. Proceedings of the Workshop on Natural Language Interaction with Databases, Schloss Laxenburg, Austria.

Szolovits, P. and W. A. Martin (1981). Brand X: Lisp Support for Semantic Networks. Proceedings of the Seventh International Joint Conference on Artificial Intelligence: 940-946.


Created by Peter Szolovits (psz@mit.edu), July 7, 2003.