Professional Documents
Culture Documents
for Verbs
April 2, 2014
1 Introduction
Our purpose in this paper is to look at the case of the representation of verbs,
or more specically verbs that represent states or events. Distinguish between
is a and hyponymy.
2 Lexicons and Ontologies
Assume the existence of a lexicon L in a language L that consists of a set of
lexical items (words or phrases) {l
1
, ..., l
k
} in the language L. We can provide
dierent kinds of morphosyntactic information by mapping the lexical entries in
L to phonetic representations, category information, subcategorisation frames,
etc. We will also assume that there exists an ontology O with vocabulary items
consisting of relation symbols and constants. We use the ontology to provide the
semantics for the lexicon by associating concepts in O with the lexical entries
in L.
We call the pairing (l, o) of a lexical entry l with a vocabulary item o belong-
ing to V with respect to the ontology O an ontological word sense, or simply
sense for short. We can represent this word sense coupling with a reied sense
object s S, where S L V , this is the approach taken by lemon, give ref
and explain.
The sense s = (l, o) should be understood as referring to the word l when
taken to mean o. We assume that the ontology here is language independent
in that the dierent relations between items in the ontology arent based solely
on how the corresponding words are used in language, but instead that the
ontology is based on additional information derived from world knowledge.
For example, given the lexical item dog we can pair it with the vocabulary
item dog where dog represents the extension of dog, this can be represented as
a sense (dog, dog).
The same lexical entry can of course have dierent meanings, so that we
cannot simply dene a function from lexical entries to ontological elements in
order to reprsent meaning, instead we need a relation between lexical entries and
1
ontological elements. For example the word bank has at least two distinct senses:
in one of these it refers to a building where people go to deposit money, and in
another it refers to the (sloping) shore of a river. The lexical item bank can be
paired with both the concepts bank as building and bank as sloping ground,
say, where our ontology restricts the meaning of these concepts based on a
semantic network of relations, e.g., our ontology might contain the following
formulae:
x.(bank as building(x) building(x))
x.(bank as sloping ground(x) r.(borders(x, r) river(r)))
If we restrict ourselves to lexical items (like a great number of noun phrases)
whose meaning can be represented by concepts (by which we mean unary pred-
icates or one place relations), then we can give a very straightforward denition
of what it means for two lexical entries l
1
, l
2
to be hyponyms (with respect to
our ontology O): we dene the binary relation hyp between senses so that hyp
holds between two senses s
1
= (l
1
, C
1
) and s
2
= (l
2
, C
2
) where C
1
, C
2
are unary,
if we can show that the formula below follows from our ontology:
x.(C
1
(x) C
2
(x)).
Similarly we can dene a relation syn representing synonymy as above but with
the following formula
x.(C
1
(x) C
2
(x)).
We can generalise this to lexical entries that have a meaning that we would
wish to represent using n-ary for n 2, relations for example with nouns such
asmother or friendship.
This denition is of course over-simplistic and fails to capture a number
of important dicult cases. Indeed there will be many situations in which
two senses for two dierent lexical entries are synonymous according to our
denition (based on our ontology) but which arent quite synonymous in actual
language use. Perhaps these two senses are only regarded as near synonyms
(as is frequently noted its extremely dicult to nd true synonyms in natural
language) or it could be that there isnt a ne enough distinction made in the
ontology (perhaps the distinction which is made in actual language use is one
which we wouldnt want to make in a well designed, ecient ontology anyway).
It may be a matter of register (think of all the slang words referring to the
male and female genitalia, which apparently dier only in terms of the contexts
in which it is proper to use them), or more generally of connotation versus
denotation.
The denition of hyponymy in lexical semantics is usually given in terms of
sentence frames and textual entailment, that is on the basis of how the word is
actually used. Cruse gives the following denition of hyponymy.
A sense
1
s = (l, c) is a hyponym of the lexical item s
= (l
, c
) if the sentence
1
Cruse uses the term lexical unit to denote the form meaning complex which we refer to
as sense.
2
A is f(l) unilaterally entails (i.e., entails but is not entailed in turn by) A is
f(l); where l is to be understood as c and l
as c
.
Here as in Cruse f(X) is an indenite expression, and represents the mini-
mum syntactic elaboration of a lexical item X for it to function as a complement
of the verb to be.
By this denition student is a hyponym of person, red apple is a hyponym
of apple, and apple is a hyponym of food.
However, adhering to certain ontology design principles such as OntoClean
that put a strong emphasis on metaphysical or cognitive soundness in conceptual
modelling, could result in an ontology in which for example, the lexical entry
student is represented as a role and therefore cannot properly be regarded as
referring to the same kind of thing as the word person does, so there should
be no subsumption relation between the concepts student and person, nor,
to take another example, between the concepts group and organisation.
Thus our denitions of hyponymy as the relation hyp, as well of syn require
a great deal of tweaking and it is clear that we should specify a number of ex-
ceptions to the statements expressed in the formulae given above; even if this is
heavily dependent on what is actually in our ontology some general principles
may be formulated. It seems that in most cases the formula we gave above for
hyp above, is a sucient but not necessary condition for when two lexical items
are hyponyms in Cruses denition. It is also likely that many of these excep-
tions can be systematically dened for example certain classes of roles including
for students or professionals can be related via a natural language hyponymy
relation to the class of persons. We can dene these as special relations that
are in a sense imposed upon the ontology and are specically language related
but which are systematically based on ontological categories.
Cruse denes the relation of taxonymy (a subrelation of hyponymy) as hold-
ing between two senses s = (l, c), s
= (l
, c