TOWARDS A USER-FRIENDLY
SEMANTIC FORMALISM FOR NATURAL
by Craig Thomas
Computational semantics has become an interesting and important branch of computational linguistics. Born
from the fusion of formal semantics and computer science, it is concerned with the automated processing of
meaning associated with natural language expressions [ 2]. Systems of semantic representation, hereafter
referred to as semantic formalisms, exist to describe meaning underlying natural language expressions. To date,
several formalisms have been defined by researchers from a number of diverse disciplines including philosophy,
logic, psychology and linguistics. These formalisms have a number of different applications in the realm of computer
science. For example, in machine translation a sentence could be parsed and translated into a series of semantic
expressions, which could then be used to generate an utterance with the same meaning in a different language [ 14].
This paper presents two existing formalisms and examines their user-friendliness. Additionally, a new form of
semantic representation is proposed with wide coverage and user-friendliness suitable for a computational linguist.
Semantic formalisms are becoming an important part of many applications in computational linguistics. Regardless of the task, a semantic formalism needs to be sufficiently expressive to capture subtle
variations in meaning. Expressiveness is evaluated based on a number of different criteria. For example, to be useful for a diverse number of applications, a formalism must have wide linguistic coverage
and be precise. Wide coverage means the formalism can express a
large number of linguistic phenomena, from simple predicates to
more complex items including modality, tense, and quantification.
Precision means the formalism can provide a distinction between
apparently similar, but semantically different phenomena. For example, the differences between an inclusive and exclusive or. The question “Would you like cream or sugar?” demonstrates an inclusive or;
it implies none, one, or both cream or sugar as an option. The question “Would you like coffee or tea?” is exclusive; it implies either one
or the other, but not both. In this case, precision may be accomplished by using different symbols or operators.
Unfortunately, one major aspect of usability is missing from this
evaluation, namely the degree of user-friendliness. While user
friendliness is not important for machines or automated tasks, it is
crucial for humans. In these instances, it is vital to have a readable,
understandable, and intuitive formalism.
Consider three different situations. Imagine that a reporter is
attending a hockey tournament and wishes to record game details in
a clear and precise manner. Imagine also a children’s book author is
writing a narrative for a fairy tale. Finally, imagine a university student
is writing a report for a class. As an additional constraint, imagine
that all three of these publications need to be printed in both English
and French. Ideally, the information need only be recorded once in a
language independent fashion using a semantic formalism. Once
recorded, the information could be fed to a natural language generation system that would print the same story in both English and
French. The flow of this process is presented graphically in Figure 1.
It will be best if the authors can use an intuitive formalism without
becoming a linguistic expert or logician. The question is whether
such user-friendly semantic representation system already exists
The purpose of this paper is two-fold. Firstly, to present two existing semantic formalisms: the first order predicate calculus and Montague’s intensional logic. The basic machinery of each are presented
with simple linguistic examples, while more complex examples are used
to demonstrate their limited degree of user-friendliness. Secondly, it
will present and discuss some preliminary results from a new system
of semantic representation. Few of the goals of this new system are to
provide wide coverage, and to be user-friendly, therefore suitable for
(Speech, Text, etc.)