eHUMANITIES DESKTOP

An Architecture for Flexible Annotation in Iconographic Research

R¨udiger Gleim, Paul Warner

Dep. for Computing in the Humanities, Goethe University, Georg-Voigt Straße 4, Frankfurt, Germany

Alexander Mehler

Texttechnology/Applied Computer Science, Bielefeld University, Universit¨atsstraße 25, Bielefeld, Germany

Keywords:

Online resource annotation, Iconography, Crowd sourcing, Resource management, Image database system.

Abstract:

This article addresses challenges in maintaining and annotating image resources in the ﬁeld of iconographic

research. We focus on the task of bringing together generic and extensible techniques for resource and anno-

tation management with the highly speciﬁc demands in this area of research. Special emphasis is put on the

interrelation of images, image segements and textual contents. In addition, we describe the architecture, data

model and user interface of the open annotation system used in the image database application that is a part of

the eHumanities Desktop.

1 INTRODUCTION

Digital image archives play an important role as a

teaching aid in classrooms in many ﬁelds, includ-

ing biology, geography and chemistry. But image

archives also serve scientiﬁc research as well, for ex-

ample in the arts and in iconographic research in the

historical sciences. When speaking of a certain dig-

ital image archive the term is usually understood as

a combination of the images themselves and the soft-

ware system delivering them. This tight link is be-

cause the software systems are often specially de-

signed for a speciﬁc purpose or are an adaption of a

more general system for a speciﬁc purpose. In the

following we therefore distinguish between the Im-

age Database (Image DB) as a structured collection

of images and the Image Database Management Sys-

tem (Image DBMS) as the software system managing

the resources. Both Image DB and Image DBMS con-

stitute an Image Database System (Image DBS). We

focus on systems which are accessible online rather

than on ofﬂine variants.

Since image database systems are often developed

or deployed with a speciﬁc purpose in mind there are

a vast number of applications which can differ widely

in terms of functionality. Commonly this includes a

basic means to browse and explore images. Usually

some core meta data like name, artist or author, copy-

right information and keywords are included. In ad-

dition, open systems allow for uploading new images

and support tagging images with keywords. A pop-

ular example is Flickr

. There are also a number of

scientiﬁc systems like Medical Picture

or the Biol-

ogy Image Library

which fall under that category. In

many usage scenarios this functionality is sufﬁcient.

For scientiﬁc research however more elaborate

systems are required which offer extensive meta

data and allow for expressing image interrelations.

Prometheus (Dieckmann (2008)) is a distributed im-

age archive for art history, archeology and the cul-

tural sciences. The project was developed at the

University of Cologne starting in 2001. Its main

target and strength is offering integrated access to

distributed, heterogenous image databases for re-

search and education. Prometheus offers a modern

web-based user interface to organize and search im-

ages. ConedaKOR

is a commercial web-based image

database system for archiving, managing and search-

ing image collections and meta data. It puts special

emphasis on the annotation of image interrelations

and dependencies which can be used for exploring the

http://www.ﬂickr.com

http://bilddatenbank.medicalpicture.de

http://www.biologyimagelibrary.com

http://www.coneda.net/kor/overview

214

Gleim R., Warner P. and Mehler A.

eHUMANITIES DESKTOP - An Architecture for Flexible Annotation in Iconographic Research.

DOI: 10.5220/0002803902140221

In Proceedings of the 6th International Conference on Web Information Systems and Technology (WEBIST 2010), page

ISBN: 978-989-674-025-2

database.

Image database systems, especially very large

ones, require well-annotated images in order to sup-

port search and retrieval, and to avoid ”data graves”

(i.e. stored information that cannot be accessed or

discovered easily). The degree to which meta data

can be annotated depends on the underlying annota-

tion scheme. An annotation scheme deﬁnes the data

ﬁelds which can be used to describe a given image. In

most usage scenarios the annotation scheme is ﬁxed

and chosen to ﬁt a particular demand. In this article

we propose an approach which puts emphasis on ﬂex-

ibility regarding the annotation schemes. Instead of

promoting a ﬁxed annotation scheme the (authorized)

users of the system should be able to create custom

schemes. Furthermore the annotation schemes should

be extensible or editable as new requirements for an-

notating images arise. We also focus on resource

management. How can the degree of user access to

images, collections and annotations be managed?

Section 2 discusses the challenges of annotating

images in iconographic research. Section 3 intro-

duces the eHumanities Desktop, a web-based system

for resource management and analysis in the human-

ities. We focus on the annotating mechanisms which

the system offers to annotate arbitrary resources and

how they can be used to address the challenges in-

troduced in section 2. Section 3.2 presents the cur-

rent state of the eHumanities Image Database System

which already implements important parts of the pro-

posed functionalities. Finally we give a conclusion

and prospect of future work.

2 CHALLENGES IN RESEARCH

ON ICONOGRAPHIC

ANNOTATION

From the point of view of linguistic annotation any

sign aggregate (e.g., a sentence, a paragraph or a

whole text) can be annotated regarding two dimen-

sions:

• We may segment the aggregate and annotate its

internal structure along various functional, se-

mantic or purely structure-oriented annotation

models. In this case, we deal with intra-aggregate

relations (e.g., intratextual relations as bridging

(Vieira and Poesio, 2000) or meronymic rela-

tions that span logical (e.g., document) structures

(Power et al., 2003)).

• Alternatively, we may interrelate the aggregate as

a whole with other aggregates and their segments.

Table 1: The combinatorics of mono- and multimodal rela-

tions speciﬁed by intra- and inter-(sign )aggregate relations.

intra-aggregate inter-aggregate

monomodal intratextual relations; intertextual relations;

intrapictorial relations interpictorial relations

multimodal intratextual-intrapictorial text-image relations

(text-segment-to-image-

segment relations)

In this case we deal with inter-aggregate rela-

tions (e.g., syntagmatic or paradigmatic relations

of lexical units (Hjelmslev, 1969; Barthes, 1977)

and, of course, intertextual relations that are es-

tablished between texts (Thibault, 1997)).

The small spectrum of sign relations that is fun-

damental for the build-up of sign systems (Hjelmslev,

1969) is complemented by the distinction of different

modalities of symbolic and iconographic signs (Kress

and Leeuwen, 1996). In this sense, we can distin-

guish, for example, intra-textual relations of a focal

text that relate its iconographic segments with its tex-

tual segments. Analogously, we may start with an

image to distinguish intra-pictorial relations that in-

terrelate textual segments of an image with some of

its pictorial segments. Finally, we have to distinguish

relations of sign aggregates as a whole that interre-

late signs of the same, or different, modality. In this

sense, we deal with text-text, text-image or image-

image relations (see Table 1 for a summary of the ma-

trix spanned by these options).

From this perspective we can distinguish the fol-

lowing challenges in research on iconographic anno-

tations:

• Text segmentation: how to automatically seg-

ment texts into their constituents as instances of

functionally or semantically demarcated, recur-

rent segment types (e.g., rhetorical relations, ar-

guments, propositions etc.)?

• Image segmentation: how to automatically seg-

ment images into their constituents as instances

of functionallyor semantically demarcated, recur-

rent segment types (e.g., emblems, logos, and less

conventional, but recurrent image components)?

• Text linkage: how to automatically interlink texts

(and their segments) with each other along intra-

or intertextual relations?

• Image linkage: how to automatically interlink

images (and their segments)with each other along

intra- or interpictorial relations?

• Multimodal linkage: ﬁnally, how to interlink texts

(and their segments) and images (and their seg-

eHUMANITIES DESKTOP - An Architecture for Flexible Annotation in Iconographic Research

215

ments) with each other along intermodal rela-

tions.

Note that we conceive of image segmentation in

semantic or functional terms in that the resulting seg-

ments are seen to be semantic or functional units.

These are signs in terms of semiotics as they recur in

different texts or images to contribute to their mean-

ing constitution (Peirce, 1934; Eco, 1986).

To the best of our knowledge, these tasks are

still unsolved. That is, we cannot yet automati-

cally segment texts and images, nor can we auto-

matically identify and type their relations. Inter-

estingly, this especially holds for textual units (in

spite of recent advancements in text mining (Feldman

and Sanger, 2007) and text-technology). Although

there are promising approaches to text segmentation

(e.g., (Marcu, 2000)), the automatization of this ef-

fort beyond logical or layout structures is still out of

reach (Stede, 2007) (see Teufel and Moens 2002 for

a promising approach in this area). However, this is

what we need if we want to explore images and their

recurrent components by analogy to semantic text re-

trieval. Consequently, we conceive a kind of image-

related usage semantics, by analogy to a linguistic se-

mantics, in which signiﬁcant co-occurrences of pic-

torial elements are explored as a reference point of

their paradigmatic relations of mutual substitutability.

Such an approach has been implemented successfully

on the level of words and multi-word units (Miller and

Charles, 1991; Landauer and Dumais, 1997; Rieger,

2001; Heyer et al., 2006). What we plan to do is to

transpose this approach onto the level of images and

text-image relations. That is, we conceive an image

semantics in which pictorial elements serve as types

whose syntagmatic contiguity (or neighborhood) as-

sociations and whose paradigmatic similarity associ-

ations are automatically computed. These are indis-

pensable ingredients of a semantics that can assign a

picture as a whole a meaning which, in turn, is the

starting point for interlinking this picture with a text

that has a similar (paradigmatically associated) or re-

lated (syntagmatically associated) meaning.

At the present time, the only way we see to mas-

ter this challenge is to support the human (i.e., non-

automatic) annotation by text-technological means in

order to gain test data by which machine learning al-

gorithms can be trained to foster the semi-automatic

annotation of these units. On the one hand, this re-

minds one of the paradigm of human computation

(von Ahn, 2006; von Ahn et al., 2006; von Ahn, 2008)

and games with the purpose of image segmentation.

On the other hand, we depart from this paradigm

in that we do not design games, but aim to enable

users to annotate any of the relations mentioned so

far, whether mono- or multimodal, reported in Table

1. The subsequent sections describe our current status

in preparing and supporting these efforts.

3 eHUMANITIES DESKTOP

Even though computer-based methods are well estab-

lished in many areas of research in the humanities it is

still challenging to bring together resources for anal-

ysis with elaborate tools for information processing.

Often tools are ﬁxed on an input format, are too com-

plex to be used by a broader public, or address a very

speciﬁc research problem and cannot be extended.

The eHumanities Desktop (Gleim et al. (2009))

aims at integrating an elaborate resource management

system with easy to use application modules for work

directly on the data. Its scope principally targets all

ﬁelds of research in the humanities including for ex-

ample linguistics, the social sciences, and the histor-

ical sciences. The design principle of the system is

to offer the full functionality of both established and

cutting edge processing and analysis methods while

keeping it usable for a broad public. The eHumanities

Desktop is in the line of systems which aim at the in-

tegration of resources and methods in different areas.

Chiarcos et al. (2008) presented ANNIS, a general

framework for integrating annotations from different

tools and tagsets. GATE (Cunningham (2002)) is a

system for ﬂexible text categorization and engineer-

ing. Clarin V´aradi et al. (2008) aim at a large-scale

European research infrastructure to establish an inte-

grated and interoperableinfrastructure of language re-

sources and technologies.

The primary objective during the development of

the system was to achieve a light-weight core system

which is ﬂexible enough to be adapted to virtually any

application, yet powerful enough to allow for mass

data and distributed, concurrent usage. The core func-

tions include user and group management, repository

and document handling and ﬁnally their interrelations

in terms of access permissions. The core layer of the

eHumanities Desktop provides a programmatic API

which offers means to control the authorities (users

and groups) and the resources they work on. It is

designed to abstract from internal representation and

storage details. All other components of the system

are designed around this core API. This allows for a

modular design and integration of new features into

the system which stem from various collaborations

and research projects.

Typical usage scenarios include distributed re-

search groups that work collaboratively on shared re-

sources. This requires ﬁne-grained access manage-

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies

216

Figure 1: Diagram depicting the master data model of the

eHumanities Desktop.

ment on both resources and applications. Further-

more, sophisticated means are required to organize

documents and repositories. Figure 1 illustrates the

data model. The principal concept is that Authori-

ties have a certain level of Access Permission on Re-

sources. The access permission can be either read,

write, delete or grant, as commonly understood from

widely-used relational database systems or ﬁle sys-

tems. An authority can either be a user or a group.

Users can be members of an arbitrary number of

groups and inherit all permissions of those groups.

This allows the system to reﬂect the demands of com-

plex research teams which access and share resources

in different roles. Resources on the other hand can be

documents, repositories (which help to structure doc-

uments), or functions.

3.1 Architecture for Resource

Annotation

Elaborate means for resource annotation and queries

are crucial for managing and retrieving resources. A

major challenge when conceiving an annotation sys-

tem for a general purpose platform is the wide range

of applications. Offering a static annotation scheme

like the well known Dublin Core

may be convenient

for speciﬁc solutions but not when the system is sup-

posed to be extensible to new requirements. Section

2 discussed the demands of annotating images for use

in iconographic research, which include, for exam-

ple, typed key-value pairs, hierarchical structures, or

typed image interrelations. This section addresses the

technical perspective of how a general purpose anno-

tation architecture could be designed to be extensible,

yet expressive and performant enough to be applied

to speciﬁc scenarios like image annotation. The con-

cepts being described are implemented as part of the

eHumanities Desktop.

http://dublincore.org

In the following we distinguish between annota-

tion schemes deﬁning the data ﬁelds which can be

used for annotation, and annotation documents which

are instances of speciﬁc schemes. The ﬁrst question

regarding the deﬁnition of annotation schemes is what

level of expressiveness should be offered. The more

expressive the data model, the better the general us-

ability for different purposes. But expressiveness is

costly, with a drop in performance when doing re-

trieval, and it also makes it more difﬁcult to offer the

user intuitive means to manage the annotations. A

self-evident option would be to rely on XML Schema

to deﬁne the structure of valid annotation documents.

XML Schema is a well-established standard and as

the annotation documents would be represented in

XML they could be processed by a wide range of

standard tools. The drawback of that solution is that

query performance of existing XML database sys-

tems is still not comparable to relational systems- at

least when it comes to data oriented structures. Thus

internally managing annotations as XML documents

would without question be elegant but would slow

down performance. Furthermore it would hardly be

possible to offer an intuitive user interface for deﬁning

annotation schemes to a user who is not accustomed

to XML schemas and their complexity.

We decided on a compromise. The user is of-

fered a graphical interface to deﬁne typed data ﬁelds,

can organize them hierarchically into trees if desired

and store them as an annotation scheme. Annotation

schemes, which deﬁne what data ﬁelds are available

for annotating resources, are represented in XML and

based on the Annotation Document Deﬁnition

lan-

guage. When actually annotating a resource the user

ﬁrst picks which annotation scheme to use and then

uses an automatically generated form to ﬁll in the

data. Internally both annotation schema and anno-

tation documents are stored in a relational database

which offers good performance for retrieval. How-

ever it is also possible to export both into XML and

process them with other tools.

Now what level of annotation expressiveness is ef-

fectively being offered by the system? An annotation

schema in the eHumanities Desktop is composed of

data ﬁelds. A data ﬁeld has a unique name, descrip-

tion and data type. The data types include strings,

numeric and boolean values, dates, and references to

other resources. Data ﬁelds may be ﬁlled with mul-

tiple values which can be used for representing key-

words for example. The multiplicity can be restricted

by specifying the lower and upper bound. In some

cases it may be desirable to deﬁne a value domain

for a data ﬁeld. A value domain can be speciﬁed by

http://xsd.hucompute.org/add.xsd

eHUMANITIES DESKTOP - An Architecture for Flexible Annotation in Iconographic Research

217

enumerating the possible values. Data ﬁelds can be

structured and organized into trees. Note that the pos-

sibility for multiple values for a given data ﬁeld also

includes root nodes of a data ﬁeld tree. In that case an

annotation document can contain multiple indepen-

dent instances of that data ﬁeld tree.

Figure 2: Example of collaborative image annotation.

The master data model of the eHumanities Desk-

top (see ﬁgure 1) regards both annotation schemes

and annotation documents as resources. As a con-

sequence they are also subject to access management.

Thus it is possible to control who is able to access

annotations and who is allowed to edit them. Further-

more the data model allows for multiple annotation of

a given resource. That way it is generally possible for

each user to create a separate annotation document.

This approach generally allows for crowd sourcing

image annotations (for example in classroom) and

later on, users could combine or rate the annotations

using measures like inter-annotator agreement for ex-

ample. Figure 2 shows an example of an image being

annotated independently by two different users using

the same annotation document deﬁnition. In this ex-

ample the user ’Alice’ has also given ‘Clara’ access

permissions to her annotation.

Section 2 has put emphasis on the annotation of

image sections for subsequent analysis. How could

this be implemented using the annotation mechanisms

of the eHumanities Desktop? An image may contain

an arbitrary number of sections to be annotated. This

would be mapped onto a data ﬁeld which can occur

multiple times and serve as a root node for all data

ﬁelds which are speciﬁc for a section. Possible data

ﬁelds for a section may include the coordinates deﬁn-

ing the section as well as keywords and other ﬁelds

to describe the content. Another approach would be

to regard image sections as independent (sub)images

which can be annotated separately. The sub-image re-

lation would then be represented as a typed reference

between the two images which can also be realized

using the annotation system.

In order to search and retrieve annotated docu-

ments the user is offered a data grid in which he can

choose how to organize, sort and ﬁlter resources. An

application which offers managing images and an-

notations as part of the eHumanities Desktop is pre-

sented in the next section.

3.2 Image Database System

The eHumanities Image Database System (or Image

DB for short) is a means for organizing, viewing,

and annotating images. As a part of the eHumani-

ties Desktop, the Image DB beneﬁts from a number

of other modules in the Desktop, such as the sys-

tem, user, and permissions control modules, as well as

the independent corpus management and annotation

modules. Since the eHumanities Desktop is struc-

tured in this way, with independent modules, these

features are then available to other modules as well.

The user and permissions control systems in partic-

ular make the Image DB ideal for use by groups of

collaborators, or in a classroom. Currently, the Im-

age DB is used for research in the ﬁeld of historical

sciences and as a teaching aid in a university setting.

Images are organized in collections, which are

similar to computer ﬁle system directories. This

structure allows for a simple, easy-to-use applica-

tion interface, consisting of a window with three tabs:

start, collections, and images. The start page, viewed

by default when starting the application, displays sys-

tem or group messages, a list of featured collections,

and a display of featured images for the user. These

messages and lists can, in the case of groups, be set

by the administrator or instructor. There is also a list

of recently-viewed collections, so the user can return

easily to ongoing projects. Double-clicking one of the

thumbnail featured images brings up a larger view of

the image in a popup window for better study. The

popup can be resized to enlarge or shrink the image.

Clicking the collections tab opens a window in

two columns. On the left is a list of collections the

user has permission to view, in multiple columns, in-

cluding name, description and any number of anno-

tation attributes. The collections can be sorted or

ﬁltered, or the attributes displayed/hidden, by click-

ing the attribute column header and accessing a drop-

down menu (see ﬁgure 3 below). Viewing and ﬁl-

tering by a particular annotation can be done using a

drop down menu at the bottom of the collection list,

where the available annotations are listed. When a

user has many collections, the display is limited by

default to 25, and the rest are available via a pag-

ing toolbar at the bottom. The user can quickly cy-

cle through the complete list, viewing 25 at a time,

by clicking the next or previous buttons on the tool-

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies

218

Figure 3: Screenshot of the images panel, showing the annotations tab and the sort/ﬁlter menu.

bar. It is also possible, via another drop down menu in

the pager, for the user to deﬁne the number of collec-

tions that are displayed at one time. Menu items, such

as ’new’, ’delete’, and ’share’ are available in a menu

bar at the top and via a contextmenu (opened by right-

clicking the mouse). Since permissions are graded in

the typical way (read, write, delete, and grant), what a

user sees or can edit is subject to ﬁne-grained control.

A user with grant privileges can permit other users

to see or edit collections or images, including select-

ing some (and not necessarily all) images within col-

lections for sharing. Annotations are also subject to

permissions, and can be shared with, or hidden from,

other users.

In the right column, the user can view or edit

information or annotations for a selected collection.

Certain attributes, such as name, description, creation

date, and owner, are available for all collections or im-

ages, regardless of annotation. There are two tabs in

this column, ’Collection Details’ and ’Annotations’,

for ease of viewing and editing. Selecting a collection

in the left column causes its details to appear in the

right, so one can quickly scan through the collection

list viewing details or annotations, as desired. With

adequate permissions, the user can edit attributes in

these windows, as well.

Annotations are the means of applying search-

able meta-data to the collections (or images), deﬁn-

ing them in user-deﬁned terms or according to pre-

existing deﬁnitions. Annotations, as noted above, are

handled by an independent module in the eHumani-

ties Desktop, and are therefore available to any type

of document collection. Annotations are deﬁned by

an annotation deﬁnition document (ADD), and mul-

tiple annotations, using one or more ADD’s, can be

applied to a collection or image. This way collabora-

tors or students can view the ’shared’ annotations of

others as well as their own, and use them for search-

ing and analyzing collections or images.

Searching, or ﬁltering, is available via ﬁlters in

the columns in the collection list on the left side of

the page. As mentioned above, one can search by an-

notation attributes by selecting the desired annotation

from the drop down menu at the bottom of the list.

This then makes the attributes of the annotation avail-

able in the attribute columns. Not all are displayed by

default, since some annotations have many attributes;

however, the user can simply select which columns to

display or hide, and use those for searching/ﬁltering.

Filters in multiple columns can be combined, and are

applied in the order selected, to quickly provide a re-

sult set for analysis. If the user selects ’none’ in the

annotations drop down menu, then only the general

attributes, such as name, description, owner, etc., are

available for display, sorting, or ﬁltering.

The interface for the ﬁnal tab, images, has the

same structure and facilities as the collections tab. Ev-

erything the user learns and becomes accustomed to

in the interface is then transferable between images

and collections. Images of course have the additional

component that they can be viewed and studied, and

they appear in the right side column as one of the

’attributes’ so to speak. Double-clicking the image

opens a popup window, as in the selected images on

the start page, so the user can adjust the image size

for analysis. Images that are too large for viewing in

eHUMANITIES DESKTOP - An Architecture for Flexible Annotation in Iconographic Research

219

the browser window are automatically resized to ﬁt

the available window.

Other important facilities include easy creation of

collections with ﬁles uploaded from the user’s com-

puter, or with selected ﬁles from other collections. It

is also a simple matter to add uploaded or transferred

images to a collection, to remove images from collec-

tions (without deleting them from other collections),

or to delete images entirely from the database. A user

can also easily download images to his or her com-

puter, for analysis or advanced editing.

The eHumanities Desktop, within which the Im-

age DB ’lives’, so to speak, is currently used by 110

users organized into 9 groups. The focus of the appli-

cation lies in linguistic applications like PoS Tagging,

lexical chaining and text classiﬁcation on the one

hand and iconographic research on the other. Users

include researchers (often working in groups), stu-

dents and classroom teachers. The system currently

manages about 10,000 documents of which about

1,700 are fully annotated images. The integration of

a larger image collection of about 50,000 annotated

images is planned in near future.

4 CONCLUSIONS

This article discussed the challenges of annotating im-

ages in the ﬁeld of iconographic research and how

such requirements could be met using the annotation

system of the eHumanities Desktop. Furthermore the

Image DB has been presented as a snapshot of ongo-

ing work which already implements a good part of the

requirements. Finally we provided information about

how the system is currently used by researchers and

students. Future work will address the implementa-

tion of the graphical user interface for image section

annotation and the development of positional queries.

REFERENCES

Barthes, R. (1977). Image, Music, Text. The Noonday Press,

New York.

Chiarcos, C., Dipper, S., G¨otze, M., Leser, U., L¨udeling, A.,

Ritz, J., and Stede, M. (2008). A ﬂexible framework for

integrating annotations from different tools and tagsets.

Traitement Automatique des Langues, 49(2):271–293.

Cunningham, H. (2002). GATE, a general architecture

for text engineering. Computing and the Humanities,

36:223–254.

Dieckmann, L. (2008). Prometheus - the distributed digital

image archive for research and education goes interna-

tional. In Dunn, S., Keene, S., Mallen, G., and Bowen, J.,

editors, Proceedings of the EVA Conference, pages 61–

67, London, UK.

Eco, U. (1986). Semiotics and the Philosophy of Lan-

guage (Advances in Semiotics). Indiana University Press,

Bloomington.

Feldman, R. and Sanger, J. (2007). The Text Mining Hand-

book. Advanced Approaches in Analyzing Unstructured

Data. Cambridge University Press, Cambridge.

Gleim, R., Mehler, A., Waltinger, U., and Menke, P. (2009).

ehumanities desktop - an extensible system for corpus

management and analysis. In Proceedings of the Corpus

Linguistics Conference, Liverpool, UK.

Heyer, G., Quasthoff, U., and Wittig, T. (2006). Text Min-

ing: Wissensrohstoff Text. W3L, Herdecke.

Hjelmslev, L. (1969). Prolegomena to a Theory of Lan-

guage. University of Wisconsin Press, Madison.

Kress, G. and Leeuwen, T. v. (1996). Reading images: the

grammar of visual design. Routledge, London.

Landauer, T. K. and Dumais, S. T. (1997). A solution to

Plato’s problem: The latent semantic analysis theory of

acquisition, induction, and representation of knowledge.

Psychological Review, 104(2):211–240.

Marcu, D. (2000). The Theory and Practice of Discourse

Parsing and Summarization. MIT Press, Cambridge,

Massachusetts.

Miller, G. A. and Charles, W. G. (1991). Contextual cor-

relates of semantic similarity. Language and Cognitive

Processes, 6(1):1–28.

Peirce, C. S. (1934). Pragmatism and Pragmaticism, vol-

ume V of Collected Papers of Charles Sanders Peirce.

Harvard University Press, Cambridge.

Power, R., Scott, D., and Bouayad-Agha, N. (2003). Doc-

ument structure. Computational Linguistics, 29(2):211–

260.

Rieger, B. B. (2001). Computing granular word meanings.

A fuzzy linguistic approach in computational semiotics.

In Wang, P., editor, Computing with Words, pages 147–

208. Wiley, New York.

Stede, M. (2007). Korpusgest¨utzte Textanalyse. Grundz¨uge

der Ebenen-orientierten Textlinguistik. Narr, T¨ubingen.

Teufel, S. and Moens, M. (2002). Summarizing scientiﬁc

articles: experiments with relevance and rhetorical sta-

tus. Computational Linguistics, 28(4):409–445.

Thibault, P. J. (1997). Intertextuality. In Asher, R. E., edi-

tor, The encyclopedia of language and linguistics, pages

1751–1754. Pergamon Press, Oxford.

Vieira, R. and Poesio, M. (2000). An empirically-based sys-

tem for processing deﬁnite descriptions. Computational

Linguistics, 26(4):539–593.

von Ahn, L. (2006). Games witha purpose. IEEE Computer

Magazine, 39(6):92–94.

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies

220

von Ahn, L. (2008). Human computation. In IEEE 24th

International Conference on Data Engineering (ICDE

2008), pages 1–2.

von Ahn, L., Liu, R., and Blum, M. (2006). Peekaboom:

a game for locating objects in images. In CHI ’06: Pro-

ceedings of the SIGCHI conference on Human Factors in

computing systems, pages 55–64, New York. ACM Press.

V´aradi, T., Krauwer, S., Wittenburg, P., Wynne, M., and

Koskenniemi, K. (2008). Clarin: Common language

resources and technology infrastructure. In Calzolari,

N., Choukri, K., Maegaard, B., Mariani, J., Odjik, J.,

Piperidis, S., and Tapias, D., editors, Proceedings of the

Sixth International Language Resources and Evaluation

(LREC’08), Marrakech, Morocco. European Language

Resources Association (ELRA).

eHUMANITIES DESKTOP - An Architecture for Flexible Annotation in Iconographic Research

221