Re: SUO: Universal Time, other universals, and cultural contextof SUO.
Pat,
You keep citing them all the time. For granularity, you cited
Jerry Hobbs. WordNet is perhaps the most widely used collection
of categories, and it was done by a psychologist, George Miller,
who has been hanging out with linguists for the past 40+ years.
And there are quite a few logicians who hang out with computational
linguists, especially when working on theories of modality,
tense logics, etc.
> Adam, can you cite any? I havn't come across many such insights from
> computational linguistics in KR.
Following is a book review I wrote (which is scheduled for the
March 2001 issue of _Computational Linguistics_) of a book by
a linguist, M. A. K. Halliday, who has a lot of interesting ideas
related to topics in ontology. (I had to make some minor adjustments
to let it slip through your filter.)
John Sowa
______________________________________________________________________
Construing Experience through Meaning:
A language-based approach to cognition
by M. A. K. Halliday and Christian M. I. M. Matthiessen
London and New York: Cassell, 1999
xiii+657 pp; paperback, ISBN 0-304-70490-3, $102.00
reviewed by John F. Sowa
Michael Alexander Kirkwood Halliday has been actively analyzing and
documenting the interactions between syntax and semantics for over forty
years, and his systemic-functional theory has been a foundation for
important work in computational linguistics for at least thirty years.
The first major application of systemic theory was the SHRDLU system
by Winograd (1972). The largest ongoing series of applications has
been developed at the USC Information Sciences Institute: language
generation (Mann 1982; Hovy 1988); discourse analysis and rhetorical
structure (Mann & Thompson 1992); and the interface between the lexicon
and world knowledge (Bateman et al. 1990; Matthiessen 1995).
In this book, Halliday and Matthiessen present a comprehensive survey
of semantics and its relationships to syntax and cognition. Although
they present their subject from a systemic-functional point of view,
they show how their approach is related to a wide range of work in both
computational and theoretical linguistics. One notable omission from
their 23-page bibliography is Noam Chomsky, whose period of active
research almost exactly coincides with Halliday's. They do, however,
give a fair summary of semantic theories based on Chomsky's approach,
ranging from the early work of Katz and Fodor to the more recent work
by Jackendoff.
The book consists of fifteen chapters organized in five parts.
In Part I, the authors contrast the systemic approach with a view
of knowledge representation as a "piecemeal accumulation" of concepts
with "no overall organization." Instead of treating language "as a
kind of code in which pre-existing conceptual structures are more or
less distortedly expressed," they view language as a semiotic system
that serves "as the foundation of human experience." The goal of
systemic theory is to present a comprehensive view of how humans
construe experience through language. Unlike Chomsky, they do not
consider grammar as "autonomous" but as an integral part of the
_lexicogrammar_, which _realizes_ meaning in words, phrases, sentences,
and paragraphs.
Part II, comprising chapters 2 through 7, presents the _meaning base_,
which corresponds to what many authors would call an ontology. The
meaning base, however, represents categories of experience with a
topmost node called _phenomenon_ instead of categories of existence
with a topmost node called _entity_. The first subdivision of phenomena
is a three-way partitioning according to levels of complexity:
1. Elementary ideas or _elements_ are realized by the lexicogrammar
as words or short groups of words, such as _rain_, _from the west_,
or _8 hard-boiled eggs_. Computationally, elements may be
represented by slots in a frame, nodes in a graph, or typed
variables in logic.
2. Configurations of elements or _figures_ are realized by phrases
or clauses, such as _rain ending from the west_ or _chop finely_.
Computationally, a figure may be represented by various data
structures, such as a frame, list, or graph.
3. Complexes of figures or _sequences_ are realized by complex
sentences or paragraphs, such as _Take 8 hard-boiled eggs, chop
finely, mash with 3 tablespoons of soft butter, and add salt and
pepper_. Computationally, a sequence could be represented by
a network of frames, a list of lists, a graph of graphs, or
structures of objects in an object-oriented language.
Each of these categories is further divided and subdivided by various
distinctions, some dyadic and some triadic. Element is classified as
participant, circumstance, or process. Figure is classified by another
triad of relational (being or having), material (doing or happening),
and mental (sensing or saying). These categories are further elaborated
and illustrated with numerous examples. To demonstrate the generality
of the approach, Chapter 7 shows how the semantic categories realized
in English can also be realized in Chinese and other languages.
Part III consists of two chapters that show how the theory can be
implemented in a computational system for language generation, with
examples of weather reports and cooking recipes. Part IV consists of
three chapters that compare the theoretical and descriptive techniques
of systemic-functional theory to other approaches. The concluding Part
V consists of three chapters that apply systemic theory to an analysis
of how humans construe experience through language. Chapter 14 has
an intriguing analysis of the evolution of linguistic expressions
from folk theories to scientific theories. Instead of drawing a sharp
dichotomy between commonsense and scientific ways of thinking, the
authors show how the basic linguistic mechanisms of abstraction and
metaphor are used to systematize and formalize scientific language.
Metaphor is fundamental to both science and poetry. The primary
difference is that poets constantly strive to create novel metaphors,
while scientists recycle, revise, and elaborate the most successful
of their colleagues' metaphors.
In summary, this book makes a strong case for the systemic approach as
a fruitful alternative to Chomsky's view of autonomous syntax. The
authors demonstrate that semantics has important structures that are
cross-linguistic and formalizable. Although they present their data
with the terminology, notation, and viewpoint of the systemic-functional
approach, their analyses, distinctions, and categories can be adapted
to semantic theories based on other approaches.
The authors criticize the logic-based, model-theoretic approaches for
their limited ontologies and neglect of important aspects of language,
such as metaphor. Yet logicians recognize the need for richer
ontologies, and many, if not most would agree that semantics is the
proper starting point for a study of natural language. The authors
try to draw a sharp distinction between the deductive methods of logic
and the method of inheritance used in frame-based systems. A logician,
however, would reply that inheritance is the oldest of all rules of
inference; it was introduced by Aristotle for syllogisms, and it is
a derived rule in every modern system of logic. The methods of
unification used in many logic-based systems implement inheritance
in ways that are equivalent to or more general than frame systems.
Rather than being a competitor, the systemic approach can be
a valuable complement to the logic-based approaches.
The authors consider language as a semiotic system, but they only
mention the dyadic view of semiotics developed by Saussure and
linguists influenced by Saussure, such as Hjelmslev and Firth. CSP
analyzed the sign relation in greater depth than Saussure and emphasized
its irreducible triadic nature. Although Halliday and Mattheissen never
mention CSP, they have rediscovered many of CSP's triads in their
systemic analysis. Their choice of _phenomenon_ as the most general
category is an unconscious endorsement of CSP's point that his
categories were primarily phenomenological rather than ontological.
The systemic triad of being-having, doing-happening, and sensing-saying
corresponds to CSP's fundamental triad of Quality, Reaction, and
Representation. Most of the other triads in the systemic meaning base
also have a strong CSP flavor, and a more conscious application
of CSP's version of semiotics might help clarify and refine many
of the triadic distinctions in the systemic approach.
Perhaps the least attractive feature of the book is its formatting.
The authors used a conventional word processor to print camera-ready copy
on A4 paper, which the publisher reproduced without change. The result
is a heavy, unwieldy tome with a great deal of wasted paper, a generally
unfinished appearance, but a price tag of $102. With that price and
format, the book is destined to sell very few copies, the authors will
get little or nothing in royalties, the publisher's high price will seem
to be justified, and a potentially important book will never be read
by students who might profit from it. The book would get better
distribution if the authors had simply put the electronic version on
their web site; better yet, professional societies such as the ACL
should put books such as this on their web sites.
Bibliography
Bateman, J. A., R. Kasper, J. Moore, and R. Whitney. 1990.
_A general organization of knowledge for natural language
processing: the Penman upper model_. Research report.
Information Sciences Institute, University of Southern California,
Santa Monica.
Hovy, E. 1988. _Generating natural language under pragmatic
constraints_. Lawrence Erlbaum, Hillsdale, NJ.
Mann, W. C. 1982. _An overview of the Penman text generation
system_. Research report 83-114. Information Sciences Institute,
University of Southern California, Santa Monica.
Mann, W. C., and S. A. Thompson, eds. 1992. _Discourse Description:
Diverse Linguistic Analysis of a Fund-Raising Text_. Benjamins,
Amsterdam.
Matthiessen, C. M. I. M. 1995. _Lexicogrammatical Cartography:
English Systems_. International Language Sciences Publishers, Tokyo.
CSP, 1991-1998. _The Essential CSP_, vols. 1 and 2,
ed. by N. Houser and C. Kloesel, Indiana University Press, Bloomington.
Winograd, T. 1972. _Understanding Natural Language_. Academic Press,
New York.