terms will cause interleaving of unrelated hierarchies
and produce spurious inferences. For example, given
that HAMMER is a TOOL and HAMMER is also an
OPERATION, if POWER-HAMMER is a HAMMER,
then POWER-HAMMER becomes a TOOL as well as
an OPERATION. The latter inference is spurious.
Homonyms can cause incorrect descriptions; for
example, a concept can be either primitive or defined;
if HAMMER as a tool is a primitive concept, and as an
operation it is a defined concept, then choosing either
type will lead to incorrect description.
Homonyms can also cause punning. OWL-DL
requires the identifiers of objects, classes, and properties to be mutually disjoint. Punning is the result of
violating this constraint. For example, prepositions
like USING and WITH occur as concepts in the language ontology and as properties in the manufacturing ontology.
The new ontology adopts the one term, one meaning (OOM) principle, where a new term will carry
only one meaning. Therefore, each sense of a homonym will be independently represented. Thus, HAMMER will split into three terms, each with a single
meaning and a distinct namespace.
lex:HAMMER opr:HAMMER tool:HAMMER
This solves the homonym problem. Now, homonyms
will have matching labels but different IRIs and will
not cause spurious inferences.
In the GSPAS ontology, name variations (like synonyms, acronyms, abbreviations, misspellings,
regional variations, names given by external sources,
and others) are treated as synonyms (call them GSPAS
synonyms). GSPAS synonyms are stored as data values in the associated term and so the classifier does
not process them. The same approach is used in the
new ontology where GSPAS synonyms are stored in
OWL annotation property. Next, we present an alternative approach and give reasons for rejecting it.
GSPAS synonyms of classes and objects can be
modeled using the predefined properties owl:equivalentClass and owl:sameAs, respectively. Now, GSPAS
synonyms become logical terms and the classifier will
process them. This has some side effects. First, we
cannot tell apart a term and its synonym because
both are first-class terms; this is not wrong, but the
synonym relation goes out of sight. Second, the synonym relation is neither symmetric nor transitive,
but owl:equivalentClass and owl:sameAs are both
symmetric and transitive and so will induce spurious
synonym relationships. Third, the GSPAS synonyms
become new terms and may cause homonym problems. This can be solved at the expense of introducing spurious homonyms (matching labels but different IRIs). For these reasons we reject this approach
and treat synonyms as data values.
In GSPAS ontology, part-of-speech (POS) information
is modeled in two ways: POS tags (like noun, verb,
and others) appear as concepts in the taxonomy (so
words in Standard Language can specialize them),
and POS tags are stored as data values in a nondefin-
itional attribute. In the new ontology, we model POS
tags as concepts in the taxonomy. The tags stored in
the attributes are remodeled into the taxonomy by
creating suitable POS concepts and subsumption
HAMMER has three senses: As an OPERATION it
operates on an OBJECT restricted to HAMMERABLE
type, and as a TOOL its SIZE is restricted to HAM-MER-SIZE. In the interest of space we will ignore the
lexical sense of HAMMER.
HAMMER ⊑ OPERATION ⊓ TOOL ⊓
(∀OBJECT.HAMMERABLE ⊓ ∃OBJECT) ⊓
(∀SIZE.HAMMER-SIZE ⊓ ∃SIZE)
Conceptually, ontology conversion takes a GSPAS
term description and creates one or more new
descriptions after resolving homonyms and imple-
menting the various design choices. For the case of
hammer, our goal is to split its description into two
HAMMERopr ⊑ OPERATIONopr ⊓ ∃OBJECTopr.HAMMERABLEobj
HAMMERtool ⊑ TOOLtool ⊓ ∃SIZEtool.HAMMER-SIZEtool
where each new term is assigned a single namespace
that is denoted by its subscript, the left side of a
description is a name, and the right side is an expres-
sion that refers to other term descriptions in the
Technically, the GSPAS ontology conversion
reduces to the problem of assigning one or more
Figure 6. Reengineered Ontology.
Language Ontology Manufacturing Ontology