Abstract
In this article, we evaluate approaches using logic reasoning applied to an ontology
for literary characters. The inference tool Racer and the programming language Prolog
were tested and compared to see if they can serve as a computer assisted approach in
this scenario.
Both offer options to draw inferences, but the usage requests a good comprehension of
logics. Intuitive and individual queries are also difficult to produce with solely
logic constructs. Furthermore, information in humanities is often vague, ambiguous,
or even contradictory. Solving such problems by logic reasoning which is simply based
on true or false conclusions will become challenging and might exceed the limits of
logic reasoning.
At the moment, to the author’s knowledge, only few such approaches, especially in
literature studies, have been published. Existing approaches show promising results
in modelling information in humanities. Therefore, further research should be
directed to make ontologies and further approaches like logic reasoning even more
popular and applicable in the humanities and literature studies.
Introduction
Information seems to be essential for our lives. Humans get information everywhere,
not only on the Internet, TV or radio channels and books, but also by talking to
people or looking at road signs. We are constantly faced with problems and by solving
these problems, our ability of using information is requested. But do we always find
a solution just in time? And do we know when and where we can use information
correctly? In daily life, we have to get information as quickly as possible. The
Internet is a “mine” of information, but, like in a real mine, we need to light
the darkness, to navigate by maps so that we are able to dig at the right point. The
Internet is a giant source of information but often completely unstructured, leading
many to the idea of structuring and grouping information presented on the Internet,
the
Semantic Web. Since the introduction of Semantic Web
technologies and standards, like
RDF and
OWL, ontologies and other related methods have entered many
research fields. In life science, they are already established, e.g. GeneOntology
[
Stevens et al. 2000]. Ontologies belong to approaches of information or
knowledge representation. Now, this technology has been also introduced into several
projects in the humanities, e.g. GOLD [
Farrar et al. 2002], WordNet [
Gomez 2004], TermNet [
Lüngen 2007], FRBR [
Renear et al 2006], and DISCOVERY [
Smith 2007]. GOLD is an
ontology for descriptive linguistics, e.g. syntax, morphology or linguistic data
structures. WordNet is a lexical database of English that is also presented as an
ontology. Thereby, TermNet, a terminological net for the domain of text technology,
takes the concepts of WordNet into account. While these are linguistic ontologies,
the FRBR ontology defines concepts of bibliographic cataloguing and indexing. Based
on domain ontologies and other data bases, the DISCOVERY project links different
philosophical projects together.
But does the obvious popularity of ontologies really point at the advantages? Or is
the popularity only a trend in research?
There are some criticisms of ontologies [
Veltman 2004], [
Shirky 2005]. Veltman points out that the definition of meaning is often
rather limited when using ontologies, especially in humanities. He argues that this
definition is adequate for machine-to-machine transactions in, for example, business,
but it does not address the needs of culture. There, meaning comprises much more and
needs to be handled by more suitable methods. In addition, Shirky criticizes the
incomplete methods of classifying things. Summarized, he argues that categorization
is always dependent on world views, which might not be accepted by other parties.
Both authors hint at very important points, which should be discussed critically. But
that goes beyond the scope of this contribution.
Nevertheless, the results of the approaches mentioned before show that ontologies and
supporting technologies like logic reasoning are useful ways to express structured
and hierarchical information in the humanities. In literature studies, the usage of
ontologies is not very common. A reason for this might be that in this field, the
contribution of an ontology to research topics is often rather difficult. Definitions
of literary phenomena are unstable and unclear. Furthermore, in humanities and
especially in literature studies, topics of research often cannot be measured like in
life science [
Zöllner-Weber & Apollon 2008]. As such it seems to be ever more challenging to
introduce ontologies to these research areas, and, one might ask, which kinds of
applications are offered and whether benefits might be drawn when applying such
technologies.
Since their initiation in philosophy, ontologies, with their methods of subsumption
related to semantics, have been expressed by logical expressions. Thereby,
information about an ontology can be obtained using logic reasoning [
Baader & Nutt 2003]. By formulating logic expressions, information that is
often implicit or hidden in the ontology can be queried and conclusions can be drawn.
This means that information about the structure and elements of ontologies can be
retrieved, which can be of interest for human users, i.e. manual processing and
automated processing, respectively. For example, even for human users, it might be
difficult to get a detailed overview of relations between the elements in large
ontologies without appropriate methods and utilities. Another interesting aspect
might arise if ontologies from different sources are merged or mapped [
Noy 2003, 18–9]. A merging or mapping may result in an enlarged
description of the ontology domain.
Such a method is also of interest for schemes and models of topics in the humanities.
Semantic relations that are essential in humanities can be expressed by creating an
ontology. After storing information in an ontology, consecutive processing, like
logic reasoning, creates further possibilities like consistency checks or support by
merging data.
The aim of this work is to investigate probable applications for ontologies in the
humanities, especially for literature studies. As mentioned, it is rather challenging
to work with ontologies in humanities. In this contribution, an example ontology and
its supporting tools, here focused on logic reasoning, are reviewed on three levels:
a) the user’s view, b) the methodological limits and c) limits of the tools with
respect to literature studies.
In Section 2.1, an introduction to ontologies, especially the way they are used in
Artificial Intelligence, is given, and their connection to logic reasoning is
outlined. In Sections 2.2 and 2.3, an ontology for the description of literary
characters and the application of logic reasoning to this ontology is presented [
Zöllner-Weber 2006]. In Section 3, logics related to ontologies are introduced.
Afterwards, the ontology for literary characters is used to test different
logic-based methods; two tools,
Racer and
Prolog, are compared.
Racer
represents pure inference machines, whereas
Prolog has
been taken as an example for logic programming languages. In Section 4, the ontology
and the applied logical reasoning tools regarding the aforementioned points are
discussed. Finally, a conclusion and an outlook are given.
Materials and Methods
In the next sections, an introduction to ontologies and especially to the ontology
for literary characters is given.
Introduction to Ontologies
Briefly, ontologies are used to provide organisation and retrieval of information
semantically. Thereby, similarities between information provided by an ontology
are searched by means of a semantic relation rather than by matching search
strings or other similar measures. There are two comprehensions of the term of
ontology: the pure formalistic approach of classifying objects as in the field of
Artificial Intelligence and machine learning, and the more transcendental approach
in the humanities and especially philosophy.
The term ontology, which originates from philosophy and contains the study of
being or existence in general, was introduced by Aristotle [
Philosophie-Lexikon 1991]. Aristotle was interested in describing the existence of things in the world.
He asked fundamental questions like “What
is existence?” On the one hand, he discussed human existence as it is
still discussed in philosophy, and on the other hand, he developed a system
(universals and particulars) of how the existence can be described, a formalism.
In particular, this is based on descriptive logics. The system of these logics is
contained in an ontology, which inherits formalistic principles, and therefore is
a form of expressing logics. In the ancient world and maybe in Aristotle’s view,
the different sciences did not exist separately. In today’s view, he mingled
transcendental questions with pure formalism. Over the course of centuries, these
two parts have been split into different sciences. In a simplified way, formalism
and classification have been moved to natural science and mathematics while the
transcendental core of Aristotle’s questions is discussed in philosophy and
theology. By transferring the term of an ontology to the AI, it is more restricted
and focused on modelling of concepts of the real world in computer systems. Gruber
gives a definition: “An ontology is a formal, explicit
specification of a shared conceptualisation”
[
Gruber 1993, 199]. In this approach, Gruber refers to groups of people with a common
comprehension of the world. In that respect, the initial question of existence is
no longer in focus but still has to be described. Therefore, the formalism relied
upon in an ontology is used. An ontology in AI comprises hierarchically structured
concepts, also called classes, of a part of the world (a
domain). The
representation should be produced in a machine-readable language. Noy et al.
explain, “An ontology defines a common vocabulary
for researchers who need to share information in a domain”
[
Noy et al. 2001, 1]. An ontology consists of a set of objects that are divided into classes,
concepts, properties (also slots, or roles) and the restrictions of the roles [
Noy et al. 2001, 3]. By relating main and more specialized classes to
an ontology, a hierarchy can be created. Additionally, so-called instances of the
classes represent individual objects of the selected domain. Although, the
structure of an ontology is rather static, the included information can be queried
and manipulated in several ways, e.g. using W3C standards like the
Web Ontology Language (OWL) or
Resource Description Framework Schema (RDFS) [
Antoniou & Harmelen 2003]. OWL has been created to enlarge web sites with
semantic information and to make the Internet usable as a structured information
source [
Ziegler 2004, 126].
Discussion about ontologies, in this contribution, is focused on the formalistic
method. Therefore, the discussion of existence, which has a long tradition in
humanities, is not relevant here because an ontology is just seen as a very useful
method to structure phenomena of literature studies. In the following, the
ontology, criteria, and tools are presented.
An Ontology for the Description of Literary Characters
To investigate ontologies and logic reasoning as tools in humanities, a concrete
example is used. Here, the description of literary characters has been realised as
an ontology.
Briefly, several theories of literary characters [
Jannidis 2004],
[
Nieragden 1995] are combined to create a base of a formal
description using an OWL ontology [
Zöllner-Weber & Witt 2006]. Categories which
describe features and actions of characters mentioned in stories had to be adapted
and applied. It is aimed at representing the mental information structure of a
reader, which (s)he has in mind when reading a book [
Jannidis 2004, 185]. Categories describing general aspects of literary characters form
the main classes of the ontology, e.g. inner and outer features, actions on other
characters and objects. In this contribution, characters are regarded as
individual objects, which are not linked to roles or archetypes [
Propp 1968], [
Greimas 1971]. Asking what a character
means to a reader means not what kind of characters we have, but what general
structure and attributes a character has. We regard a character as a complex
cognitive entity in the reader’s mind, rather than emphasizing metaphors or
archetypes. In this contribution, approaches that separate characters by roles,
genres, etc., are fazed out, because they might superimpose literary criticism on
the daily phenomenon of reading literature. Reading literature and understanding
characters does not necessarily require that everyone be a scholar in literature
studies. It is of interest how a normal person as well as a scholar describes a
character. By using a hierarchy, individual descriptions of characters should get
a common ground to comparing these descriptions. By using sub classes, categories
can subsume features of special characters or groups of characters. In addition,
so-called instances of the classes represent individual and explicit objects of
the domain of literary characters. Here, direct information about a character
given in a text is assigned to an instance. Together with the information of the
class hierarchy and other instances, a single mental representation of a character
is modelled (cf.
Figure 1). In this approach,
individual description, the pre-step of interpretation, is focused. Finding the
same patterns revealed in different descriptions might be regarded as a common
sense. By analysing these, similarities based on the cultural background of
readers or of writing and reading traditions might be found. A more detailed
description of this ontology is given in [
Zöllner-Weber 2006], [
Zöllner-Weber & Witt 2006].
Applications using the Ontology for Literary Characters
Having explained the concrete implementation of the ontology of literary
characters, we focus this section on logic reasoning, which is a central principle
when dealing with ontologies.
In general, conclusions drawn from logic reasoning can be inferred from given
information. This means implicit information is derived from explicit information.
In formal logic, given information is called assumption; the operation is called
conclusion. The methods induction [
Charniak & McDermott 1985, 22] and
deduction [
Charniak & McDermott 1985, 14] are subsumed under the term of
logic reasoning. Induction means inferring from special concepts to general
concepts whereas deduction constitutes the opposite process from general concepts
to special ones. Here, the conclusions can only receive the values true or false.
The assumptions of the induction have to be true so that a conclusion can be
constituted as true as well. The conclusion is called an inference.
Logic reasoning is one possible application for ontologies. It is probably helpful
(i) to check consistency during ontology development, (ii) to enable
semi-automatic merging of (domain) ontologies as well as (iii) to deduce hidden
information contained in the ontology. These three tasks can be applied to all
elements of ontologies, classes as well as instances. As Baader et al. mention,
logic reasoning can fulfill different purposes in the phase of creating an
ontology and in the phase of using it [
Baader & Nutt 2003, 4], e.g.
investigating the structure of categories/ concepts, or testing if every object is
used in the intended and not contradictory way. In different situations of a work
process, logic reasoning can be used to avoid or to solve problems: If several
persons build together an ontology, new included elements can be checked for
inconsistency or redundant information can be detected.
Coming back to the mentioned mapping processes, one can state that for the
ontology of literary characters, it might be useful, if two persons add categories
with the same or similar meaning to the ontology, to produce suggestions for
mapping. Applications like mapping or merging are useful for the development and
usage of ontologies. But recently, to the author's knowledge, there have been no
other ontologies in literature studies which deal with literary characters or
narratology and which could be related to this ontology, i.e. by merging or
mapping. To be successful, synonymy has to be taken into account and a relating
mechanism has to be implemented, because processing of synonymy is not included
directly in an ontology so far. Therefore, these tasks cannot be applied
completely to the ontology for literary characters at the moment.
Especially in the humanities, where textual data is often crucial, applications
can be used which combine ontologies with text annotations or information
retrieval. Ontologies consist of information, ideas, and/or facts, which also
occur or are repeated in (text) sources that are not included in an ontology. This
might be especially interesting when dealing with ontology learning. For example,
terms contained in both sources, in text annotations/web pages and an ontology
[
Baader & Nutt 2003, 4] can be queried for. But here arises the
question of whether arranging data that both sources can match would be a
redundant and time-consuming work. In addition, setting up more elaborate match
processes could require a complex system including lexica or external rules [
Puppe et al. 2000, 635–6], [
Benjamins et al. 1999].
Nevertheless, logic reasoning is not only useful for such processes. Logic
reasoning can also help during users' orientation period. Hierarchical relations
can be highlighted so users can get a structured overview about all elements and
their relations. Another aspect is obtaining information that is implicitly hidden
in the data.
These examples show only some of the imaginable methods which might be performed
by logic reasoning, but already they outline possibilities for an ontology in the
humanities. It does not matter for which discipline ontologies are developed; they
all have to pass a development, test, and deployment phase and thus, the methods
of logic reasoning might be helpful.
However, mapping or merging is yet not possible because of the singularity of the
ontology for literary characters. Such operations might become possible if further
ontologies in this field will become available.
In addition, the human user should be focused when testing logic reasoning. In
order to test logic based applications, several tasks are defined, - which are
similar to the queries given in linguistic approaches [
Lüngen 2007]:
Retrieval of
-
Individuals/instances that belong to a given class. The
results should give information about the relationship between special
classes and their instances so that it is possible to obtain information
about a character, which is disseminated in a text, e.g. that characters
consist of special features whose information is expressed as
instances
-
Individuals/instances that contain a given property (and maybe a
given value). The results should give information about the
relationship between instances and special properties and to which class the
instance belongs so that the apportionment of additional information can be
focused, thereby it might be possible to examine which special kinds of
properties only belong to some classes
-
Bottom up/top down. The result should show the relationship
of classes and its super classes, it should also show the arrangement of the
features/actions of characters and the structure of the mental
representations
Testing Applications of Logic Reasoning
OWL and its Relation to Description Logic
The ontology used in this contribution was implemented in
OWL
Description Logic
[
Antoniou & Harmelen 2003]. The standard OWL consists of three sub languages
(OWL Light, OWL DL, OWL Full) created on different levels of information
representation. Before going on, a short summary of the Description Logic whose
ideas are included in OWL DL: OWL is often used in its sub language OWL DL, as it
inherits concepts of a logic formalism called
Description
Logic (DL). In DL, concepts are grouped in a so-called
TBox and individuals belong to an
ABox. The structure of a TBox is represented as subsumption so that
more general concepts contain special ones, which correspond to the
concepts/classes of an ontology. Regarding the ontology for literary characters,
all classes that have been defined (cf.
Figure 2)
are collected in the TBox. An ABox subsumes all instances of the modelled domain.
This can be seen as a description of a domain. Coming back to the ontology for
literary characters, this means that the ABox is composed of the instances
contained in the ontology.
As Nardi and Brachman explain, these methods are usually a variant of first-order
predicate calculus (cf. [
Nardi & Brachman 2003, 2]). Therefore, it is
possible to use the mechanisms of logic formalism like reasoning or drawing
inferences. To operate on data that is formatted or transformed into Description
Logic, inference algorithms have been developed. These algorithms are implemented
in different programmes like FaCT++ [
Tsarkov & Horrocks 2006], Pellet [
Sirin et al. 2007], and Racer [
Haarslev et al. 2004], which are DL
reasoners, or Prolog
http://www.swi-prolog.org/ which is also able to query OWL
ontologies.
Racer and Prolog
In the next sections, the two already-mentioned tools, the inference machine Racer
and the logic programming language Prolog, will be described.
Introduction to Racer
Here, we investigate whether Racer (
Renamed ABox and
Concept Expression Reasoner Professional) is a good tool for drawing
inferences on ontologies in the humanities. Racer was developed for approaches
that are based on OWL/ DL and offers inference functions as well as consistency
checks (cf. [
Racer Systems 2005, 1]).
According to the concepts of an ABox and TBox in DL, the classes and
individuals are presented and inferred separately (cf.
Figure 3). To infer in OWL, Racer uses the query language
nRQL
[
Racer Systems 2005, 87]. This means that the notion in the XML
standard of OWL has to be translated to this query language. Queries in nRQL
consist of a head and a body similar to predicate calculus [
Haarslev et al. 2004].
The syntax for a TBox and ABox differs from each other. By an ABox query, for
example possible individuals of a special class can be retrieved. It is also
possible to query for individuals with special properties. So, information
nested in individuals and properties can be related to each other. In order to
retrieve individuals with a data type property, so-called head projection
operators have to be used. In a TBox, classes that do or do not have
a given name can be also queried for.
Figure 3 shows how Racer is working in
conjunction with an ontology. Except for the graphical notion of TBox and ABox,
all queries in nRQL have to be inserted manually. The results are also shown in
nRQL. The user has no graphical interface when querying in the ontology.
Another helpful option is to use a special construct that can be used in
combination with the classes that should be retrieved. This means that direct
hierarchical relations between classes like sub or super relations can be
retrieved. But one should note that Racer can only offer a result if the
classes are in a hierarchical relation to each other.
In general, when querying an ontology by Racer, different results can be
obtained. First, a Boolean is returned (true or false), i.e. an assumption
about the ontology can be verified. Second, a literal value is returned,
depicting certain information queried for.
All these efforts are reasonable when dealing with huge or (partly) unknown
ontologies. Logic reasoning using Racer might help a human to orientate
himself/herself and to work easily with such an ontology. Figure 4 shows the
processing of possible queries of an ABox or TBox.
Introduction to Prolog
Another possibility for drawing inferences is to use a programming language
that is specialized for logics, e.g. Prolog. It is based on first-order
predicate calculus [
Fisher 2006] and provides unification,
backtracking, and tail recursion: operations that are especially useful for
ontologies. Inferences can be drawn by using a parser that operates on a
so-called
Prolog fact base. This means that, similar to Racer and
its query language, OWL ontologies have to be transferred to a notion in
Prolog.
Data in Prolog consists of facts and rules. The facts
can be considered a vocabulary on which Prolog works. Facts, called
predicates, are defined with arguments included to describe the
knowledge, and given to Prolog. A predicate is defined by giving a head and
arguments (in brackets). By using rules, the fact base can be enlarged so that
different relations and combinations between the modelled elements can be
achieved.
Using Prolog as an inference tool for ontologies, the
SWI-Prolog implementation offers several useful modules for
manipulating RDFS and OWL data. In this approach, the RDFS module for
SWI-Prolog has been used
http://www.swi-prolog.org/pldoc/package/semweb.html. By loading
an ontology, the module represents the data as triples in a Prolog fact base.
The RDFS module can query for the relations between classes, properties, and
individuals by using predefined predicates, e.g. relation between classes and
instances. But one can also query for instances with a given property.
It is also possible to retrieve instances with a given literal, e.g. special,
property value. For comparison reasons, the usage of literals like character
names included in the ontology for literary characters is very useful. Thus,
all the instances or pieces of information of two literary characters can be
retrieved. But it might also be interesting to query for characters created by
a special author. For example, a user wants to retrieve every instance with the
name of a given author in its property. Information can be filtered.
Hierarchical relations of classes, e.g. super-sub class relations can also be
queried top down or vice versa. By combining the predicates of the module, a
wide range of different queries can be supported.
Discussion
Having outlined the technical principles, in the following, we discuss the
application of logic reasoning to an ontology in literature studies on three levels:
a) the user’s view, b) the methodological limits including ontologies, and c) limits
of the tools especially for literature studies.
The task defined in Section 2.2 could be realised by queries formulated in nRQL and
Prolog.
Table 1 in the appendix summaries all types of queries
implemented and executed so far. The query in nRQL depicted in Figure 5, as an
example, as well as in
Table 2 in the appendix,
demonstrates the possibilities of using logic reasoning.
Literally, it could be asked if a certain description of a character is featured by a
class (instance-class relation), combination of content of the primary text and
narrative/ additional information (instance-property relation), and the hierarchical
structure of a character (top-down relation).
Racer and Prolog differ in their basic concept of logics: Prolog is based on
predicate calculus whereas Racer uses the concept of description logic (ABox, TBox
syntax). Both tools have been evaluated in this work. In comparing Prolog and Racer,
a huge difference between them is that SWI-Prolog cannot only be used for drawing
inferences — the parser can handle all kind of data represented as a Prolog fact
base. In contrast, Racer is a solely inference machine.
Both have their own representation and syntax so that a user who is querying an
ontology needs detailed knowledge on handling both programmes, i.e. to draw
conclusion with these tools, the plain representation of the ontology in OWL, an XML
standard, has to be transformed into the programmes’ own representation in advance.
Additionally, a possible user has to be familiar with the concepts of an ontology to
formulate reasonable queries. This assumes certain knowledge in theory in ontologies
and logic but also some technical “hands on” experience for daily use. It is
questionable if these skills can be a requirement for a scholar with traditional
education in literature studies without proper training. In summary, both programs
are reasonable for inference tasks with ontologies in humanities. Both have
advantages and disadvantages, especially formulating queries seem to be difficult.
SWI-Prolog might be a choice since complete scripts can be executed. On the other
hand, Racer provides a graphical representation of ontologies.
As mentioned, an ontology consists of a set of objects, concepts, and further
entities, which are related to each other [
Noy et al. 1997, 53]. These
objects are divided into classes, properties [
Noy et al. 2001, 3].
Additionally, instances of the classes represent individual objects of the selected
domain. By relating main classes and more specialised ones of an ontology, a
hierarchy can be created. Therefore, only information in a hierarchical order can be
modelled using an ontology. Similarities between concepts or classes commonly
occurring when modelling certain information from literature cannot be expressed
directly by an ontology. However, there exist certain structures within the OWL
languages that allow for such relations. But these relations have to be modelled
explicitly during the creation of the ontology. Concerning the description of
literary characters as in this work, theories which tend to have a hierarchical
structure for describing the characters as Nieragden [
Nieragden 1995]
should be included. Theories that are based more on network-like structures, however,
should be modeled using topic maps or similar techniques, for example.
Besides this disadvantage, one has also to note that vague information can be handled
by the ontology through the instantiation of a class. While the class describes the
more general and common agreed on concept, the instance refers to a concrete example,
also including possible individual interpretations. This kind of modelling has been
used for the ontology of literary characters.
In principle, drawing inferences can retrieve implicit and hidden information. The
structure of OWL DL allows logic and formal applications, but the scope of the
logic-based queries is often restricted. Since the logic theorem is based on the
closed world assumption, only information already included in a fact base can be
returned. Mostly the implicit information that possibly exists in an ontology is
anchored in the hierarchical relations between the classes and between classes and
instances. Furthermore, the employed logical concept in description logics only can
give back true or false conclusions (or the elements of the fact base that match a
true result). However, when it is queried for vague information (not modelled via
certain similarity constructs as explained before) this approach will fail since they
cannot be expressed in logical terms [
Berthold 2003]. Furthermore, such
information is often ambiguous, or even contradictory [
Zöllner-Weber & Apollon 2008]. Here,
fuzzy logics or statistically methods might be superior. In computer science, many
approaches combine logic reasoning with other methods of information retrieval to
overcome the limits of logics. Logical queries are often embedded in systems, which
contain further sources like lexica to extract more information e.g. ontology
learning [
Cimiano et al. 2006].
More generally, when comparing literature studies and computer systems, one has to
take into account that semantics in literature studies often do not match semantics
in machine-to-machine learning [
Veltman 2004]. Often, technical, e.g.
machine-to-machine, semantics are more flat and not too complex, whereas in
literature semantics could be highly complex. As an example in the present work, the
description of the literary characters is dependent on the reader’s interpretation.
Therefore, the modelling of the ontology must allow for several interpretations
side-by-side in the ontology (via instantiations). This is also covered in [
McCarty 2005], where he points out that proper modelling in humanities
is important and that the model has to be kept open or flexible, i.e. to be able to
include the thinking itself. Furthermore, one has to note that the foundations i.e.
the literary theories of character descriptions, on which the ontology is built on,
mainly have hierarchical character and thus, are more advantageous for the proposed
task. Nevertheless, also the structure of a hierarchy is critical. Shirky notes that
hierarchies are usually built on a single world-view [
Shirky 2005]. But
mostly the weighting or the level of detail within different branches in the
hierarchy is influenced by the modeller or by economical aspects. Therefore, the
hierarchy might not be objective. This might not be a problem for machine-based
processing; however, a user of this ontology might not agree on the presented
hierarchy. This risks that the provided system will be abandoned by the user.
Conclusion and Outlook
In this article, we evaluated approaches using logic reasoning applied to an ontology
for literary characters. The inference tool Racer and the programming language Prolog
were tested and compared to see if they can serve as a computer assisted approach in
this scenario.
Both offer options to draw inferences, but the usage requests a good comprehension of
logics. In addition, intuitive and individual queries are also difficult to produce
with solely logic constructs. Furthermore, the hierarchy model of an ontology might
often not be suitable for certain topics in literature studies. For example, strands
of plots that evolve over time cannot be represented by using a hierarchy. For such
analysis, other approaches are more suitable (for example [
Meister 1999]). One should also keep in mind that information in humanities is often vague,
ambiguous, or even contradictory. An example of such a situation is given by the
relations between language and the world in Ludwig Wittgenstein’s philosophical work
Tractatus. If trying to model these by an ontology
(hierarchy), the author was faced by the fact that the hierarchy changes during the
progress of Wittgenstein’s approach [
Zöllner-Weber & Pichler 2007]. However, a model
represented by an ontology remains static. Solving such problems by logic reasoning
which is simply based on true or false conclusions will become challenging and might
exceed the limits of logic reasoning.
Nevertheless, the potential of logic reasoning for literature should be explored. It
is helpful in the different phases of the creation and deployment of ontologies in
the humanities and its sub disciplines like literature studies. For example, when
different experts are working on the same ontology, its consistency can be checked
and errors can be minimised in an early state of an ontology.
Probably, when dealing with such information and logic systems, Fuzzy Logic should be
considered [
Halpern 2005]. It offers a range of values that weight a
conclusion rather than a binary decision as the logics described in this
contribution. Also, combinations with other types of systems, like additional
knowledge bases, might support current ontology approaches in humanities and
literature studies.
At the moment, to the author’s knowledge, only few such approaches, especially in
literature studies, have been published. Therefore, a general judgement about the
value of computer-aided approaches as outlined in this contribution could not be
demonstrated to its limits. Existing approaches (for example [
Farrar et al. 2002], [
Lüngen 2007], [
Zöllner-Weber & Witt 2006])
show promising results in modelling information in humanities. Therefore, further
research should be directed to make ontologies and further approaches like logic
reasoning even more popular and applicable in the humanities and literature
studies.
Appendix
Query for instance class relation: |
rdf(INSTANCE,'http://www.w3.org/1999/02/22-rdf-syntax-ns# type',CLASS),
rdf(CLASS,'http://www.w3.org/1999/02/22-rdf-syntax-ns#type',
'http://www.w3.org/2002/07/owl#Class'). |
Query for instances that belong to one class: |
rdf_has(X,Y,'http://www.figurenontologie.de/
#on27_action_and_behaviour'). |
Query for top down relation: |
rdf_has(Y,Z,X),rdf_has(A,B,Y). |
Querying hierarchies: |
rdf_has(SUBCLASS,'http://www.w3.org/2000/01/rdf-schema#subClassOf',SUPERCLASS),
rdf_has(SUPERCLASS,'http://www.w3.org/2000/01/rdf-schema#subClassOf',SUPERSUPER),
rdf_has(SUPERSUPER,'http://www.w3.org/2000/01/rdf-schema#subClassOf',SUPER). |
Query for sibling classes: |
rdf_has(SUBCLASS1,'http://www.w3.org/2000/01/rdf-schema#subClassOf',SUPERCLASS),
rdf_has(SUBCLASS2,'http://www.w3.org/2000/01/rdf-schema#subClassOf',SUPERCLASS),SUBCLASS1\=SUBCLASS2. |
Table 1.
Table 1 depicts several PROLOG queries formulated in this work for assessing
the ontology of literary characters.
An ABox query: |
(RETRIEVE(?X)(?X |http://www.figurenontologie.de/unnamed.owl#
on4_general_notes|)) |
A TBox query with negation: |
(TBOX-RETRIEVE(?X) (NEG (?X |http://www.figurenontologie.de/
unnamed.owl#on43_statement_about_the_analised_character|))) |
Query with head projection operator: |
(RETRIEVE(?X(TOLD-VALUE (|http://www.figurenontologie.de/unnamed.owl#
storytitle|?X))) (?X|http://www.figurenontologie.de/unnamed.owl#
on43_statement_about_the_analised_character|)) |
Table 2.
Table 2 depicts several Racer queries in nRQL formulated in this work for
assessing the ontology of literary characters.
Works Cited
Antoniou & Harmelen 2003 Antoniou, G., and F. Van Harmelen. “Web Ontology Language: OWL”. In S. Staab and R. Studer, Handbook on Ontologies. Berlin: Springer, 2003. pp. 67-92.
Baader & Nutt 2003 Baader, F., and W. Nutt. “Basic Description Logics”. In F. Baader D. Calvanese, D.L. McGuinness, P.F. Patel-Schneider and D. Nardi, The Description Logic Handbook. Cambridge: Cambridge University Press, 2003. pp. 43-95.
Baader et al. 2003 Baader, F., I. Horrocks and U. Sattler. “Description Logics”. In Steffen Staab and Rudi Studer, eds., Handbook on Ontologies. Berlin: Springer, 2003. pp. 3-28.
Benjamins et al. 1999 Benjamins, V.R., D. Fensel, S. Decker and A. Gómez Pérez. “(KA)2: Building Ontologies for the Internet: A Mid-Term Report”. International Journal of Human-Computer Studies 51 (1999), pp. 687-712.
Berthold 2003 BertholdM. “Fuzzy Logic ”. In M. Berthold and D.J. Hand, eds., Intelligent Data Analysis. Berlin: Springer, 2003. pp. 321-350.
Charniak & McDermott 1985 Charniak, E., and D. McDermott. Introduction to Artificial Intelligence. Reading MA: Addison-Wesley, 1985.
Cimiano et al. 2006 Cimiano, Philipp, Johanna Völker and Rudi Studer. “Ontologies on Demand? A Description of the State-of-the-Art, Applications, Challenges and
Trends for Ontology Learning from Text”. Information, Wissenschaft und Praxis 57: 6-7 (2006), pp. 315-320.
Farrar et al. 2002 FarrarS., Lewis W. D. and Langendoen D. T. “An Ontology for Linguistic Annotation”. In Semantic Web Meets Language Resources: Papers from the AAAI Workshop, Technical Report WS-02-16. Menlo Park, CA: 2002. pp. 11-19.
Gomez 2004 Gomez, Fernando. “Grounding the Ontology on the Semantic Interpretation Algorithm”. Presented at GWC 2004. Proceedings of the Second Global WordNet Conference (2004), pp. 124-129.
Greimas 1971 Greimas, A.J. Strukturale Semantik: Methodologische Untersuchungen. Braunschweig: 1971.
Gruber 1993 Gruber, T.R. “A translation approach to portable ontologies”. Knowledge Acquisition 5: 2 (1993), pp. 199-220.
Haarslev et al. 2004 Haarslev, V., R. Möller and M. Wessel. “Querying the Semantic Web with Racer + nRQL”. Presented at ADL '04. International Workshop on Applications of Description Logics (2004).
Halpern 2005 Halpern, Joseph Y. Reasoning about Uncertainty. Cambridge, MA: MIT Press, 2005.
Jannidis 2004 JannidisFotis. Figur und Person - Beitrag zur historischen Narratologie. Berlin: Walter de Gruyter, 2004.
Lüngen 2007 Lüngen, H., and A. Storrer. “Domain Ontologies and Wordnets in OWL: Modeling Options”. Zeitschrift für Computerlinguistik und Sprachtechnologie 22: 2 (2007), pp. 1-17.
McCarty 2005 McCarty, Willard. Humanities
Computing. New York and Basingstoke: Palgrave Macmillan, 2005.
Meister 1999 MeisterJan Christoph. “Jenseits von Lagado. Literaturwissenschaftliches Programmieren am Beispiel des Kodierungsprogramms Move Parser 3.1”. Jahrbuch für Computerphilologie 1 (1999), pp. 71-83.
Nardi & Brachman 2003 Nardi, D., and R.J. Brachman. “An Introduction to Description Logics”. In Franz Baader Diego Calvanese, Deborah McGuinness, Daniele Nardi and Peter Patel-Schneider, eds., The Description Logic Handbook. Cambridge: Cambridge University PRess, 2003.
Nieragden 1995 NieragdenG. Figurendarstellung im Roman: Eine narratologische Systematik am Beispiel von David Lodges Changing Places und Ian McEwans The Child in Time
. Trier: Wissenschaftlicher Verlag Trier, 1995.
Noy 2003 Noy, N.F. “Tools for Mapping and Merging Ontologies”. In Steffen Staab and Rudi Studer, eds., Handbook on Ontologies. Berlin: Springer, 2003. pp. 365-384.
Noy et al. 1997 Noy, N.F., and D.L. McGuinness. “The State of the Art in Ontology Design”. AI Magazine 18: 3 (1997), pp. 53-74.
Philosophie-Lexikon 1991 Hügli, A., and P. Lübcke, eds. Philosophie-Lexikon. Reinbek bei Hamburg: 1991.
Propp 1968 Propp, Vladimir. Morphology of the Folktale. Austin and London: University of Texas Press, 1968.
Puppe et al. 2000 PuppeF., Stoyan H. and Studer R. “Kapitel 15: Knowledge Engineering”. In G. Görz C.-L. Rollinger and J. Schneeberger, eds., Handbuch der Künstlichen Intelligenz. München: Oldenbourg Verlag, 2000. pp. 599-641.
Shirky 2005 Shirky, Clay. “Ontology is Overrated: Categories, Links, and Tags”. Presented at O'Reilly ETech conference (2005).
Sirin et al. 2007 Sirin, Evren, Bijan Parsia, Bernardo Cuenca Grau, Aditya Kalyanpur and Yarden Katz. “Pellet: A practical OWL-DL reasoner”. Journal of Web Semantics 5: 2 (2007), pp. 51-53.
Smith 2007 Smith, D.C.P. “Re-Discovering Wittgenstein”. Proceedings of the 30th International Wittgenstein Symposium (2007), pp. 208-210.
Stevens et al. 2000 Stevens, R., C.A. Goble and S. Bechhofer. “Ontology-Based Knowledge Representation for Bioinformatics”. Brief Bioinform 1 (2000), pp. 398-414.
Tsarkov & Horrocks 2006 Tsarkov, D., and I. Horrocks. “FaCT++ Description Logic Reasoner: System Description”. Presented at IJCAR 2006. Proceedings of the International Joint Conference on Automated Reasoning (2006), pp. 292-297.
Ziegler 2004 ZieglerC. “Surfende Maschinen”. In Das XML-Kompendium. Hannover: 2004. pp. 126-130.
Zöllner-Weber & Apollon 2008 Zöllner-WeberAmélie, and Apollon Daniel. “The Challenge of Modelling Information and Data in the
Humanities”. Proceedings of the International Conference on Explorations of New Spaces, Relations and Dynamics in Digital Media Ecologies (2008).
Zöllner-Weber & Pichler 2007 Zöllner-WeberAmélie, and Pichler Alois. “Utilizing OWL for Wittgenstein's Tractatus”. 30th International Wittgenstein Symposium (2007), pp. 248-250.
Zöllner-Weber & Witt 2006 Zöllner-WeberAmélie, and Witt Andreas. “Ontology for a Formal Description of Literary Characters”. Presented at Digital Humanities 2006 (2006). Proceedings of Digital Humanities 2006 (2006), pp. 350-352.
Zöllner-Weber 2006 Zöllner-Weber, A.
“Formale Repräsentation und Beschreibung von literarischen
Figuren.” In G. Braungart, P. Gendolla, and F. Jannidis (eds), Jahrbuch für Computerphilologie 7. mentis Verlag, Paderborn
(2006), pp. 187–203.
Zöllner-Weber 2007 Zöllner-WeberAmélie. “Noctua literaria - A System for a Formal Description of Literary Characters”. In G. Rehm A. Witt and L. Lemnitzer, eds., Data Structures for Linguistic Resources and Applications. Tübingen: 2007. pp. 113-121.