16.1596, Review: Semantics/Comp Ling: Nirenburg & Raskin (2004)

[LINGUIST List]

LINGUIST List: Vol-16-1596. Wed May 18 2005. ISSN: 1068 - 4875.

Subject: 16.1596, Review: Semantics/Comp Ling: Nirenburg & Raskin (2004)

Moderators: Anthony Aristar, Wayne State U <aristar at linguistlist.org>
            Helen Aristar-Dry, Eastern Michigan U <hdry at linguistlist.org>

Reviews (reviews at linguistlist.org)
        Sheila Dooley, U of Arizona
        Terry Langendoen, U of Arizona

Homepage: http://linguistlist.org/

The LINGUIST List is funded by Eastern Michigan University, Wayne
State University, and donations from subscribers and publishers.

Editor for this issue: Naomi Ogasawara <naomi at linguistlist.org>
================================================================

What follows is a review or discussion note contributed to our
Book Discussion Forum. We expect discussions to be informal and
interactive; and the author of the book discussed is cordially
invited to join in. If you are interested in leading a book
discussion, look for books announced on LINGUIST as "available
for review." Then contact Sheila Dooley at collberg at linguistlist.org.

===========================Directory==============================

1)
Date: 17-May-2005
From: Petra Gieselmann < petra at ira.uka.de >
Subject: Ontological Semantics

	
-------------------------Message 1 ----------------------------------
Date: Wed, 18 May 2005 21:58:31
From: Petra Gieselmann < petra at ira.uka.de >
Subject: Ontological Semantics


AUTHORS: Nirenburg, Sergei; Raskin, Victor
TITLE: Ontological Semantics
SERIES: Language, Speech, and Communication
PUBLISHER: MIT Press
YEAR: 2004
Announced at http://linguistlist.org/issues/16/16-527.html


Petra Gieselmann, Interactive Systems Lab, University of Karlsruhe

DESCRIPTION / SUMMARY

This book aims to describe a comprehensive approach called ontological
semantics to the treatment of text meaning by various NLP-applications,
such as machine translation, information extraction, etc. Ontological
semantics consists of an integrated complex of different theories and
methodologies. It is built on microtheories covering such diverse areas as
specific language phenomena, processing heuristics, and implementation
system architecture. All these theories are coordinated at the level of
knowledge acquisition and runtime system architecture implementation.

The book is divided in two parts: The first part describes ontological
semantics and explains its main theoretical points in relation to other
fields, such as cognitive science and the AI paradigm, the philosophy of
science, linguistic semantics and the philosophy of language,
computational lexical semantics, and studies in formal ontology. The
second part deals with the content of ontological semantics. It discusses
text-meaning representation, static knowledge sources, the processes
involved in text analysis, and the acquisition of static knowledge.

Chapter 1 gives an introduction to ontological semantics. The approach is
centered around the metaphor of an intelligent agent. Therefore, two
intelligent agents are necessary at least: The discourse producer and the
discourse consumer. The model consists of the following dynamic knowledge
sources: an analyzer, a generator, and a module for world knowledge
maintenance and reasoning. Furthermore, the static knowledge sources
consist of an ontology, a fact repository, a lexicon, an onomasticon (a
lexicon of proper names), a text-meaning representation formalism and some
knowledge for semantic processing (structural mappings, knowledge
supporting treatment of reference, etc.). Since a single comprehensive
theory covering all these different aspects seems not feasible, the
authors introduce the concept of microtheories which are bunched according
to the phenomena they can deal with. In addition, the authors explain
different architectures, such as the stratified model, the flat model and
the constraint-satisfaction model; the latter being largely adopted in
ontological semantics. Also the relations of ontological semantics to the
non-semantic components of an NLP system and the development of
ontological semantics over the last years are briefly outlined.

Chapter 2 is a very theoretical and philosophical description of
ontological semantics compared to other theories. The first section
explains the need for philosophical discussions in general and in the
field of computational linguistics in particular. Then the authors give
definitions for the main components in scientific theories, such as
purview, premises, body, and justification. The conceptual space within
which all linguistic semantic theories can be positioned is composed of
diverse parameters. The authors also explain the relation between theories
and methodologies associated with them. In addition, this chapter
discusses practical applications of theories and their influence on
relations between theories and methodologies. Finally, the authors
describe how this philosophical approach can be used to characterize and
analyze ontological semantics. Therefore, they concentrate on one example
parameter: Explicitness.

Chapter 3 deals with the history of semantics and different philosophical
and linguistic traditions. First, the chapter briefly describes the roots
of linguistic semantics starting already with Plato. The authors explain
the different semantic traditions covering also diachronic semantics and
examinations on the historical meaning change. They explain different
semantic approaches from Ogden, Richards and Bar Hillel up to contemporary
approaches in detail. In addition, the chapter summarizes the ideas on
compositional semantics and their influence on ontological semantics. The
authors also discuss key ideas from other semantic traditions and evaluate
them against ontological semantics.

Chapter 4, together with chapter 3, relates ontological semantics to other
important semantic approaches and issues. Chapter 4 concentrates on
lexical semantics and focusses on four central issues already raised in
the lexical semantics of the late 80's to early 90's. First, the
advantages and disadvantages of generative vs. enumerative lexicons are
discussed and the authors explain why they think that every good lexicon,
including ontological semantic ones, should be capable of accommodating
novel meanings and therefore be generative. The next section discusses the
complicated relationship between semantics and syntax and explains why the
authors do not believe that there is a complete isomorphism between the
two and that such a simplifying assumption cannot be hold in practice.
Furthermore, the chapter discusses sentential meaning and the relation to
the meaning of the words in detail. In ontological semantics, semantic
meaning is defined as a text-meaning expression obtained through the
application of rules for syntactic analysis, for linking syntactic and
semantic dependencies and for establishing the meaning of lexical units.
Therefore, a formal world model, namely the ontology, is crucial.

Chapter 5 discusses the differences between ontological semantics and
other ontological efforts. The first section places ontology in the
context of metaphysics. The authors discuss formal ontology and its
contributions to ontological semantics as far as theoretical and also
practical issues are concerned. As an example of practical issues the
semantic web initiative is explained. The chapter explains the differences
between ontology and natural language giving some examples of
crosslinguistic semantic divergences which result in problems for
multilingual ontologies.

Chapter 6 is the first chapter of the second part. It deals with meaning
representation in ontological semantics.  The first section discusses the
problems of meaning proper and possible inferences. Then the authors
explain the text meaning representation (TMR) giving different examples.
The TMR includes the lexical information and the results of morphological
and syntactic analysis of the input text. Therefore, the text meaning
representation uses two basic means: instantiation of ontological concepts
and instantiation of semantic parameters not connected to the ontology.
The Backus Naur Form specifies the syntax of TMR which consists of a set
of propositions connected through text-level discourse relations, such as
modality, coreference, time dependencies and style. The propositions are
units of semantic representation corresponding to single predications in
context (typically realized as clauses).  This means that the TMR results
above all from the process of disambiguation by the analyzer. Therefore,
ontological semantics uses semantic selectional restrictions stored in the
lexicon and the ontology described in detail in chapter 8. The authors
give different examples of TMR specifications. In addition, they discuss
possibilities to represent synonyms and paraphrases in TMR.

Chapter 7 explains the different static knowledge sources, such as the
ontology, the fact repository and the lexicons and their relations. The
ontology consists of definitions of concepts representing classes of
objects or events in the real world. It is language-independent and uses a
collection of property-value pairs. Inheritance mechanisms are available
and also multiple inheritance is allowed, although seldom used in real
applications until now. Semantic properties describe the nature of objects
and events in the ontology, such as physical properties, inherent
properties, is-a-properties, and also properties specifying the semantic
arguments (i.e. case roles for predicates). In addition, the ontology also
contains complex events instantiated from the text input to provide
expectations for further sentence processing in the text (similar
to 'scripts' explained by Schank and Abelson 1977 for example.)  Last but
not least, an axiomatic definition of the ontology is given. The fact
repository includes records of past experience. Therefore, you can find in
the fact repository instances of ontological concepts. The lexicon
contains a collection of entries indexed with their citation form in the
languages available in the system. Every entry includes all the lexemes
with the same base form, regardless of pronunciation, sense or syntactic
information. Each entry contains information on the lexical category, the
orthography, phonology, morphology, syntax, etc. In the onomasticon,
proper nouns can be found.

Chapter 8 outlines the basic processing mechanisms in ontological semantic
text analysis. The workflow consists of the following steps:
Preprocessing, building semantic dependencies, processing meaning beyond
basic semantic dependencies which includes the treatment of phenomena such
as aspect, modality and time and finally processing at the suprapositional
level. Thereby, processes at the suprapositional level consists of
reference and coreference phenomena, temporal ordering within TMRs and
discourse relations. The preprocessing is again divided into different
phases: tokenization and morphological analysis, lexical lookup, syntactic
analysis. After that, basic semantic dependencies are created based on a
propositional structure which is established first by means of a basic
selectional-restriction matching procedure. Sometimes, the basic
selectional-restriction matching procedure cannot completely disambiguate
the word senses of a given lexeme. In this case, the authors present
different back-up strategies, such as dynamic tightening of selectional
restrictions or comparing distances in ontological space. On the other
hand, it is also possible that a selectional restriction fails to find any
candidate for filling the value of property. The reasons for this case can
be twofold: There is no recognizable candidate in the input which might be
due to ellipsis or unexpected words or phrases in the input; on the other
hand, the given lexeme is available in the lexicon, but has no sense which
matches the selectional restriction which can be caused by non-literal
language for example. Therefore, relaxation of selectional restrictions in
the ontology is possible so that sentences such as 'The baby ate a piece
of paper.' or non-literal meanings such as 'The pianist played Bach.' can
be resolved. In addition, ontological semantics can also process
unattested inputs (i.e. words or phrases which are not covered by the
lexicons or onomasticons at the moment). Therefore, proper names are
already discovered by a preprocessing component of the analyser. Other
unattested input is processed by the available morphological, syntactic
and semantic analyzers to assign as many features to it  as possible given
the fact that no lexicon entry is available. Out of that and by means of
the semantic dependencies and the selectional restrictions of the other
concepts in the input, a new lexicon entry can be generated on the fly.

Chapter 9 deals with the acquisition of static knowledge sources for
ontological semantics. The authors explain the immense effort necessary to
develop such natural language resources, both in time and in trained human
resources. Therefore, automatic knowledge acquisition is desirable. The
acquisition of the ontology involves the decisions, whether a new concept
is necessary, where it should be integrated exactly and which properties
are crucial for this concept. This cannot be done automatically at the
moment, but requires trained people to work on this task. The lexicon can
at least be acquired by some automatic support. Therefore, the authors
explain the rapid propagation: A single sample entry for a whole class of
lexical entries is used to copy most of its properties to the whole class
with only slight changes (i.e. English adjectives of size can be rapidly
propagated in this way.) In addition, ontological semantics uses lexical
rules to derive the properties of some deverbal adjectives from their
corresponding verbs for example, such as 'abhorrent' from 'abhor'. The
authors stress the importance of the grain size of the lexicon and its
practical effability. Furthermore, they discuss the relationship between
ontological matching and lexical constrains. The fact repository can be
semi-automatically acquired by means of information extraction techniques
used on web pages.

Finally Chapter 10 gives a short conclusion and an outlook on future work
areas.

CRITICAL EVALUATION

"Ontological semantics" is an interesting and valuable contribution to the
NLP community. It offers a clear outline of the theory of ontological
semantics and explains the differences to other theories and traditions.
Especially the first part of the book is very theoretical and might not be
easy to read, especially for beginners in semantics. In my opinion, it
would be interesting to see how this ontological semantics is used in some
more practical examples, such as an MT application for example. The book
only briefly mentions some examples, but a complete workflow is missing
and would be interesting to get a general impression of the power of this
approach. Another interesting application for ontological semantics might
be dialogue processing which is only very briefly mentioned in this book.
Furthermore, the acquisition of language resources is very expensive in
time and human resources as already mentioned by the authors. Still they
did not explain any completely automatic methods for acquiring them, but
only some tools to facilitate the whole process for the linguist a little
bit. It might be interesting to see what other methods could fit here into
the theory of ontological semantics to gain an easier acquisition of new
resources and save time and money for new applications.

REFERENCES

Raskin, V. and S. Nirenburg. 1996. Adjectival Modification in Text Meaning
Representation. Proceedings of COLING 96. 842-847

Schank, R. and R. Abelson. 1977. Scripts, Plans, Goals and Understanding.
Hillsdale. NJ: Erlbaum.

Searle, J. (1969) Speech Acts: An Essay in the Philosophy of Language.
Cambridge: Cambridge University Press.

ABOUT THE REVIEWER

Petra Gieselmann has an M.A. in Computational Linguistics. Currently, she
works at the university of Karlsruhe in the Interacitve Systems Lab
towards her PhD. Her research interest lies in the field of dialogue
management and speech understanding. She is especially interested in
semantics and pragmatics in dialogue systems and error recovery by means
of resolution of anaphora and elliptical expressions.





-----------------------------------------------------------
LINGUIST List: Vol-16-1596