AGENTS AND ONTOLOGIES FOR UNDERSTANDING

AND PRESERVING THE ROCK ART OF MOUNT BEGO

L. Papaleo

ICT Department, Province of Genova, Genova, Italy

G. Quercini

Institute for Advanced Computer Studies, University of Maryland, College Park, U.S.A.

V. Mascardi

, M. Ancona

, A. Traverso

CS Department,

DARFICLET, University of Genova, Genova, Italy

H. De Lumley

Laboratoire D

epartemental de Pr

ehistoire du Lazaret, ADEVREPAM, Nice, France

Keywords:

MAS for rock-art preservation, Ontology extraction, Sketch interpretation, Visual database for rock-art.

Abstract:

This paper describes the joint effort of computer scientists, archaeologists, and historians for designing a

multi-agent system that exploits ontologies for the semantic description of the Mount Bego petroglyphs, thus

moving a step forward their preservation. Most components of the MAS have already been developed and

tested, and their integration is under way.

1 INTRODUCTION

For decades the area around Mount Bego has been

deemed as a sort of a bewitched place, rocks being

carved with “thousand devils” (Pierre de Montfort,

XV century). On the other hand, archaeologists and

historians look at this place as an incredibly valu-

able source of knowledge, due to the up to 40,000

ﬁgurative petroglyphs and 60,000 non-ﬁgurative pet-

roglyphs scattered over a large area at an altitude of

2,000 to 2,700 meters.

The historical relevance of the Mount Bego pet-

roglyphs is unquestionable, as they date back to the

early Bronze Age, when humans left no written evi-

dences and the only witnesses of their existence are

their tools and, indeed, their “drawings”.

While learning from written sources is relatively

easy, even if this may actually depend on the source,

images generally lend themselves to a number of dif-

ferent and often conﬂicting interpretations, which is

the case of the Mount Bego carved rocks. Hence, the

coherence of any new interpretation of petroglyphs

must be checked against multiple sources.

Another major issue is that Mount Bego rocks are

not protected in a safe place such as a museum and

thus they are constantly exposed to rough weather as

well as vandalism of careless or malicious visitors.

The explorer who ﬁrst realized the importance of

Mount Bego carvings from an historical point of view

and started a systematic activity for preserving them,

was Clarence Bicknell. In 1897, he sketched 450

drawings on small sheets of paper. Between 1898

and 1910 he realized up to 13,000 drawings and re-

liefs, part of which were then published in (Bicknell,

1913). Bicknell’s collection is completed by inedited

drawings owned by the University of Genoa. These

amount to 16,000 drawings and reliefs on different

materials. At the foot of almost all sheets Bicknell

wrote personal notes on the depicted subject, the lo-

cation of the petroglyph, a name assigned to the rock

with the petroglyph and the date of the relief. The

legacy owned by the University of Genoa also in-

cludes nine notebooks, ﬁlled with notes in Victorian

English, which can be subdivided into ﬁve excavation

diaries and four note books. which cover a timespan

of 10 years (1902-1912).

Many years after Bicknell’s campaigns, several

teams led by Henry de Lumley have been surveying

and mapping this important archaeological area start-

ing from 1967.

To the purpose of preserving Mount Bego rock art,

we are moving along two directions: ﬁrst, we will in-

288

Papaleo L., Quercini G., Mascardi V., Ancona M., Traverso A. and De Lumley H..

AGENTS AND ONTOLOGIES FOR UNDERSTANDING AND PRESERVING THE ROCK ART OF MOUNT BEGO.

DOI: 10.5220/0003188002880295

In Proceedings of the 3rd International Conference on Agents and Artiﬁcial Intelligence (ICAART-2011), pages 288-295

ISBN: 978-989-8425-41-6

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

tegrate the wonderful collection of Bicknell into an

existing database of data relative to Mount Bego cur-

rently hosted by the Laboratoire D

epartemental de

ehistoire du Lazaret (Adevrepam), based in Nice,

France; then we have implemented, and we are cur-

rently integrating into our MAS, most of the meth-

ods and interfaces that will allow end users to add

other data that might become available in the future:

data will be recognized and semi-automatically cate-

gorized, according to speciﬁc ontology-driven crite-

ria.

The reminder of this paper is organized as fol-

lows. An overview of our approach is discussed in

Section 2; details on the design and partial implemen-

tation of a multi-agent system supporting sketch in-

terpretation algorithms are given in Section 3; semi-

automatic extraction of ontologies from texts is dis-

cussed in Section 4. In Section 5 we describe how

our framework will be used to integrate, analyze and

interpret data on Mount Bego. Related work and con-

cluding remarks are ﬁnally presented in Section 6.

2 OVERVIEW OF OUR

APPROACH

The cultural relevance of the Mount Bego area calls

for immediate action for preservation. For this reason

the domain experts involved in this paper (H. de Lum-

ley himself and A. Traverso, an historian working at

DARFICLET, the Department of Archaeology, Clas-

sical Philology and their Traditions in the Christian,

Medieval and Humanistic Ages, University of Genoa)

welcomed the computer scientists’ proposal to com-

plement the techniques they usually adopt to preserve

the area with digital preservation and restoration tech-

niques.

Our long-term goal is the creation of a repository

which may eventually become a reference at Euro-

pean level as a repository for Bronze Age petroglyphs.

To this aim the repository needs to come with tools

that ease the integration of new data and allow anal-

ysis, comparison, interpretation and search on them.

Although here we only focus on Mount Bego, our ap-

proach is general enough to handle scenarios other

than Mount Bego. Just to make an example, many

studies

demonstrate that there are strong similari-

ties between Mount Bego and Valcamonica’s petro-

glyphs. Hence, our approach should easily apply to

preserve and maintain in digital form Valcamonica’s

petroglyphs, in the same way as it applies to Mount

Bego’s ones, coherently with our long-term objective.

http://www.rupestre.net/pdf rtf/valca bego fr.pdf

In the current setting we have access to a database

managed by Adevrepam, whose data are the best can-

didates for starting to ﬁll the European Bronze Age

petroglyphs repository. It is a PostgreSQL database,

equipped with the PostGIS module to manage ge-

ographical objects and accessed through the Py-

GreSQL module. It includes up to 45,000 among im-

ages, texts and cards obtained from reliefs in Mount

Bego area. Each carved rock has a unique identiﬁer

number and precise GPS coordinates along with semi-

structured annotations about the petroglyphs.

Three aspects form the basis of our work:

1. Integration and Semantic Annotation. Our

main short-term goal is the integration of the Bicknell

legacy, which is the most valuable source of knowl-

edge on the petroglyphs that have been destroyed,

into the Adevrepam database. Bicknell did not limit

himself to faithfully depicting the petroglyphs, which

would have been a remarkable contribution itself, but

he also wrote semi-structured annotations, including

location, rough description of the represented subjects

and personal thoughts and interpretations. We remark

that even if the Bicknell legacy is relatively small,

a manual integration would be time-consuming and

error-prone, besides being not scalable to our long-

term goals. Therefore, we need to design a tool which

helps domain experts integrate the data in a semi-

automatic way. In this scenario duplicates are the ma-

jor issue; the Bicknell legacy, in fact, contains draw-

ings and annotations of petroglyphs which are already

in the Adevrepam database. If this is the case, the in-

tegration tool needs to recognize duplicates in order

to avoid the creation of two separate entries on the

same subject. Recognizing duplicates is not straight-

forward, especially in the ﬁeld of rock art, where dif-

ferent petroglyphs may share more or less the same

patterns while being different. As far as semantic

annotation is concerned, we aim at creating seman-

tic relations among similar petroglyphs (that is those

sharing the same patterns) thus allowing successive

full and partial retrieval according to ontology-driven

metadata and visual content.

2. Analysis and Normalization. The new inte-

grated data must be checked against the existing ones

for coherence. We recall that Bicknell was not an

archaeologist and, even if the scientiﬁc community

agrees on the importance of his work, some inaccu-

racies are still possible in his drawings. Moreover, to

the best of our knowledge, an in-depth analysis and

assessment of his work has not been done before. To

this extent, we propose the exploitation of techniques

based on both image similarity and ontology-driven

metadata matching.

AGENTS AND ONTOLOGIES FOR UNDERSTANDING AND PRESERVING THE ROCK ART OF MOUNT BEGO

289

3. Interpretation and Semantic Enrichment.

Since giving an interpretation to petroglyphs is not

straightforward and two or more different interpre-

tations may coexist and complement each other, we

plan on developing a tool that helps domain experts

to assess their interpretations and theories. Petro-

glyphs are elementary drawings, with shapes that can

be found in many of them and so they are repeated.

Given a shape, different interpretations can be as-

signed to it. For example, a zigzag line is likely to

represent water (de Lumley and Echassoux, 2009), a

ﬁgure with two intersecting lines may represent a man

and so on. Therefore, the system we are going to im-

plement will be able to analyze all petroglyphs and,

using ad-hoc sketch interpretation techniques, to take

out all the “meaningful shapes” that occur frequently

and to propose them to the domain expert. A foreseen

usage scenario is that where the domain expert assigns

an interpretation to each shape and adds ontology-

driven metadata to that shape. Afterward, she can also

access the petroglyphs to check whether its interpreta-

tion ﬁts in the depicted scene, and the system provides

suggestions of interpretations, thus being a real sup-

port for the experts. Moreover, the tool will also able

to check for frequent co-occurrences of two shapes

in the same petroglyph which may denote a possible

joint interpretation of the two shapes (that is the two

shapes have different interpretations if they occur in

the same petroglyph).

In the following sections we will describe the

technology we will use to implement these three

steps.

3 MULTI-AGENT SYSTEM

ARCHITECTURE

As it is well-known MASs are an optimal solution

when it comes to manage and organize data from mul-

tiple sources, which is the case in most Cultural Her-

itage applications such as ours. That is why we turned

to a MAS as a basis of our framework.

The architecture of our MAS will extend the one

discussed in (Casella et al., 2008) by adding spe-

ciﬁc classes of agents to it and enriching the sys-

tem with ontology-driven knowledge, thus extending

the communication interfaces with Semantic Web fa-

cilities. More speciﬁcally, the framework presented

in (Casella et al., 2008) serves as a basis for a multi-

domain sketch interpretation system, called AgentS-

ketch, that recognizes and interprets symbols and sim-

ple shapes. AgentSketch uses either on-line or off-line

interpretation of symbols; in the ﬁrst case, the sys-

tem starts the interpretation process while the symbol

is being drawn, whereas in the latter the interpreta-

tion is carried out on a complete image. Sketch inter-

pretation is the linchpin to implement our framework,

as stated in Section 2 and AgentSketch works reason-

ably well on such tasks; therefore, we will use it while

adapting it to our speciﬁc case. In the following we

summarize the main features of Casella et al.’s agent

framework (Section 3.1) and we show how we will

extend it (Section 3.2). How we will use it is the sub-

ject of Section 5.

3.1 The Agent Framework

The main novelty of the agent framework in (Casella

et al., 2008) over existing approaches is its ﬂexibil-

ity, as it can be used in different contexts. Current

solutions (Kaiser et al., 2004; Juchmes et al., 2005;

Azar et al., 2006, just to cite some recent ones) either

borrow techniques from stroke recognition, therefore

limiting the set of symbols that can be recognized,

or restrict themselves to particular domains, or they

impose a-priori an usage mode (either on-line or off-

line).

The agent framework (Fig. 1) is composed of four

kinds of agents:

• Interface Agent (IA), that represents an interface be-

tween the agent-based framework and any input de-

vice used to draw a sketch. In the case of an online

interpretation, this may be a pen-based device.

• Input Pre-Processing Agent (IPPA), that processes

the input received from the Interface Agent and

sends the obtained results to the Symbol Recogni-

tion Agents described below, using a format compli-

ant with the interpretation approach they apply.

• Symbol Recognition Agents (SRAs), each one de-

voted to recognize a particular symbol of the do-

main by controlling one hand-drawn symbol recog-

nizer (HDSR). HDSR are not considered agents since

they are just passive providers of services within the

MAS. They must be programmed by the system de-

veloper in order to recognize the symbols of the

graphical language under consideration. SRAs may

collaborate with other SRAs in order to apply context

knowledge to the symbols they are recognizing, and

with the Sketch Interpretation Agent described below

that deals with the sketch interpretation activity. Of

course, we must assume that a professional with skills

on the computer science side speciﬁes the patterns of

symbols of interest, by deﬁning one hand-drawn sym-

bol recognizer for each of them. This activity is time

consuming and expensive, but the domain expert in-

volved in the activities described in this paper, namely

the end users of the designed MAS, believe that it

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

290

Figure 1: MAS architecture. Red components represent extensions to the original MAS architecture by Casella et al.. Red

lines identify new communication interfaces that will be developed to allow agents interact.

should be worth devoting time and resources to this

activity, given the beneﬁts that they could gain from

it.

• Sketch Interpretation Agent (SIA) that provides the

correct interpretation either of the sketch drawn so far

(in case of an on-line drawing process) or of the entire

sketch (in case of an off-line interpretation process) to

the IA.

AgentSketch instantiates the general agent frame-

work architecture and has been exploited for recog-

nizing UML Use Case Diagrams in both on-line and

off-line modes. It is implemented on top of Jade (Bel-

lifemine et al., 2007).

3.2 Our Extensions

We are working at the extension of Casella et al.’s

architecture by adding agents with new functional-

ities and by deﬁning ad-hoc communication inter-

faces among the agents and the data layer (namely the

knowledge base, the multimedia DBs and the signa-

tures DB in Figure 1). We have designed two agent

types that will be added to the existing agent frame-

work:

• Classiﬁcator Agents (CA), that will use both the in-

terpretation of the sketch or image processed by the

SIA and additional information that may be attached

to the image itself (see Section 5.1) to classify the im-

age (or sketch) according to the ontologies deﬁning

the MAS domain vocabulary.

• Accuracy Agents (AA), that will exploit CBIR tech-

niques (Veltkamp and Tanase, 2002; Datta et al.,

2008; Shishir K. Shandilya, 2010) by extracting a

characterization (signature) of the images content

(given ad-hoc heuristics, possibly formalized in the

ontology), used for indexing and searching according

to visual content similarity. This activity will be per-

formed also for measuring the accuracy of Bicknell’s

drawings with respect to the more recent reliefs made

by de Lumley’s team.

The domain vocabulary of agents populating our

MAS is provided by a set of ontologies (stored in the

Ontology Knowledge Base, which we already started

to ﬁll) that may be both developed by hand, or ex-

tracted in a semi-automatic way from existing infor-

mation sources (see Section 4).

4 ONTOLOGY EXTRACTION

The MAS described in Section 3 heavily relies on

the exploitation of ontologies for multimedia content

classiﬁcation and retrieval. We developed an ontology

describing the petroglyphs found in Mount Bego (Pa-

paleo et al., 2010, Figure 3). Our ontology was manu-

ally built and is based on the results of the archaeolog-

ical reliefs of de Lumley and his team (de Lumley and

Echassoux, 2009). We remark that our aim is to inte-

grate in our repository new data as they are available

and this may require updating the ontology or creat-

ing new ones: it is self-evident that manual updates

take time and are error-prone. Therefore we designed

and implemented a Role Ontology Extractor tool that

AGENTS AND ONTOLOGIES FOR UNDERSTANDING AND PRESERVING THE ROCK ART OF MOUNT BEGO

291

Figure 2: A schema describing the communication between a Classiﬁcator Agent and the Interface Agent in our MAS. The

integration task will be performed under the supervision of the domain expert who, on the basis of the information provided

by the Classiﬁcator Agent on tags, can decide whether the content must be added as “new item” in the multimedia DB or as

an “update” of an existing entry.

semi-automatically generates concepts and relations

from texts, thus easing the process of creating new on-

tologies from scratch and/or updating existing ones.

The Role Ontology Extractor tool extracts from a tex-

tual document the most relevant concepts that may

be used in the MAS as well as relationships among

them and their generalized super-concepts (Bozzano

et al., 2010). The selection of meaningful relation-

ships among the ones output by the extractor must be

supervised by the domain expert: it is well known that

Word Sense Disambiguation (Agirre and Edmonds,

2006) that the extractor applies in order to give each

word its correct meaning within the context provided

by the text, is an AI-complete problem, and a software

tool can only support, but not substitute, the human

user. Once the selection of meaningful relationships

has been completed, the Role Ontology can be created

from them: concepts are mapped into OWL classes,

relationships among concepts are modeled as OWL

properties, and the taxonomic relationships between

concepts and their generalization is translated into the

OWL subClassOf relation. The Role Ontology gen-

erated in this way can be used inside the MAS as a

reference vocabulary among agents.

To show how the Role Ontology Extractor works,

we run it on the following sentence from (de Lum-

ley and Echassoux, 2009): “These petroglyphs, [...]

,translate not just the daily preoccupations of these

populations who needed rain, sources and lakes in

order to fertilize their ﬁelds, but also their cosmo-

logical myths. At the center of these myths are the

bull god, brandishing lightning, master of the storm

and provider of fertilizing rain, and the high goddess,

mother goddess or goddess earth, who needs to be

fertilized herself by rain from the sky in order to bring

abundance to humans.”

The extracted meaningful relationships are listed

below, and their ontological representation can be

found in (Papaleo et al., 2010, Figure 4).

Quality → concept relationships

bull → god ; pastoral → population; agricultural →

population; southern → alps; ancient → bronze age;

daily → preoccupation; cosmological → myth; high

→ goddess; mother → goddess; goddess → earth.

Concept → action → concept

bull → brandish → lightning; god → brandish →

lightning; population → need → rain; population →

need → source; population → need → lake; source

→ fertilize → ﬁeld; lake → fertilize →ﬁeld; rain →

bring → abundance.

The selection of meaningful generalizations and

relationships from all those produced by the Role On-

tology Extractor has been made by hand. Once this

hand-made ﬁltering stage has been completed, the

OWL ontology can be generated in an automatic way.

We are currently working at making the ontology

compliant to CIDOC-CRM, a high-level ontology

that enables information integration for Cultural

Heritage data (Doerr, 2003) which is also known

as standard ISO 21127:2006. The code of the Role

Ontology Extractor, developed by Michele Bozzano

as part of his Bachelor Thesis at the CS Department

of Genoa University, is available under GPLv2 license

(http://www.disi.unige.it/person/MascardiV/Software/

roleExtractor.html). It was implemented using

SWI Prolog extended with the ProNTo Morph

library for natural language processing

(http://www.ai.uga.edu/mc/pronto/Schlachter.pdf).

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

292

5 THE FRAMEWORK AT WORK

This section shows how our framework will be used to

integrate new data into the Adevrepam database and

to retrieve content based on sketch interpretation.

5.1 Data Integration

Classiﬁcator Agents (CAs) are responsible for inte-

grating new data into the existing Adevrepam reposi-

tory while updating the knowledge base and keeping

the repository consistent. CAs work at the semantic

level of images (photographs of Bicknell’s drawings

like those shown in the right hand side of Figures 3

and 4, new photographs or sketches of Mount Bego

petroglyphs, as well as of petroglyphs dating back to

the same period, but discovered in other places) and

may take annotations from existing sources or speciﬁc

metadata attached to images into account. A CA is

activated when a new image is ready to be integrated.

The CA sends the image to the Interface Agent (IA)

responsible for the interpretation process. When the

Sketch Interpretation Agent (SIA) completes the in-

terpretation task, the IA sends the result back to the

CA that triggered the interaction. This result is a set

of concepts taken from the domain ontologies, stat-

ing the interpreted meaning of the set of recognized

symbols, and used to tag the image. If metadata or

annotations were attached to the image, the CA ex-

ploits them to provide an even more reﬁned tagging,

by identifying those concepts in the ontologies clos-

est to the expected meaning of metadata. To carry out

this task we developed and experimented with success

an ontology matching algorithm that heavily exploits

natural language processing (Mascardi et al., 2009).

If the set of concepts that tags the image is iden-

tical to the set that tags another image already stored

in the database, the CA suggests to the user that the

image might refer to content already present in the

database and hence should not be considered as a new

entry but as an update of existing information. On the

other hand, if no similar tags are associated with any

entry, the CA suggests that the image should be added

as a new entry. Indeed, the CA must always interact

with the domain expert in order to correctly add new

content in the database. However, the CA provides

suggestions to the human user that lighten the burden

upon her.

After a decision has been made, the CA classi-

ﬁes the image according to the ontology concepts that

tag it and stores it in the multimedia database with all

the necessary metadata. Figure 2 illustrates the set of

tasks and the communication among the agents in the

MAS and the user.

After the integration step, the framework must en-

sure that the new data are correct and coherent with

the existing data. In our speciﬁc case this means that

we assess the accuracy of the drawings of Clarence

Bicknell and to do so we need to compare his draw-

ings with the data already in the repository. To this

aim, the exploitation of techniques based on image

similarity measures have been explored. An Accuracy

Agent (AA) is activated by the user in order to estab-

lish if the selected image is part of one or more al-

ready existing classes of signatures (stored in the sig-

natures DB) or if a new similarity class must be deter-

mined. If necessary, the new signature class is created

and the signatures DB updated accordingly. Thus, in-

clusion, exclusion and intersection predicates are im-

plemented to ensure partial similarity measures and

ranking list of result answers. Once stored, the user

can ask to an Accuracy Agent the similarity measure

between two images. For example, it should make

sense to measure the similarity between the images

shown in Figure 3 and 4. Accuracy Agents operate

at the image level by running algorithms that compute

the similarity between images, without knowing noth-

ing of their meaning.

Figure 3: Petroglyph identiﬁed by id. ZIIGIR3: de Lumley

team’s relief (left; private collection owned by Adevrepam)

and Bicknell’s relief (right; private collection owned by

University of Genoa).

Figure 4: Petroglyph identiﬁed by id. ZVIIGIIR7: de

Lumley team’s relief (left; private collection owned by

Adevrepam) and Bicknell’s relief (right; private collection

owned by University of Genoa).

AGENTS AND ONTOLOGIES FOR UNDERSTANDING AND PRESERVING THE ROCK ART OF MOUNT BEGO

293

Figure 5: Petroglyphs of corniculates of the Mont Bego region. From de Lumley, H. and Echassoux, A. (2009).

The assessment of Bicknell’s drawings has never

been done and is very important for the preservation

of Mount Bego. Suppose that after this assessment we

ﬁnd out that Bicknell’s drawings are accurate for ev-

ery ﬁgure but the horned one; this would imply that

we could digitally recreate all missing petroglyphs

with high accuracy, except for those representing the

horned ﬁgures.

5.2 Content-based Image Retrieval

Since Bicknell, the way archaeologists work today

has dramatically changed. Instead of botanical paper

sheets and pencil, they bring portable devices in their

excavation campaigns, often equipped with sketch-

based interfaces, as well as cameras and technical in-

struments for documenting with high precision their

discoveries. Consider the scenario where an archaeol-

ogist making excavations either in the Mount Bego re-

gion or in any region where similar petroglyphs have

been found (for example, the Valcamonica Valley

)

discovers a new petroglyph. She may wonder if sim-

ilar petroglyphs have been already recorded in the

“Bronze Age petroglyphs” repository. To this aim,

she may either take a picture of the petroglyph (Fig-

ure 6a) or draw a sketch by means of her PDA’s sketch

based interface (Figure 6b).

In both cases, the agents devoted to sketch inter-

pretation may start their interpretation task (in on-line

mode if the archaeologist sketched the petroglyph, in

off-line mode otherwise) and discover that the pattern

of the sketch (or of the picture given in input) respects

the pattern of a known symbol, namely that of cor-

niculates, based on the results of previous campaigns

(Figure 5). The new sketches (or images) can be up-

loaded in the repository and structured according to

the semantic data provided (ontology-driven and vi-

sual) thus enriching the repository with new multime-

dia content and new knowledge.

A Sketch Interpretation Agent can be activated

also in the case in which a user wants simply to search

for content in the repository “similar” to an input data.

In this case, after the interpretation has taken place,

a Classiﬁcation Agent must compare the tags result-

ing from the interpretation process with those tagging

http://www.rupestre.net/alps/valcamonica.html

content in the database, in the same way described in

Section 5.1.

The Classiﬁcation Agent will return a ranked list

of multimedia content according to tag similarity cri-

teria that can be different depending on users’s needs

(which can be deﬁned using speciﬁc ontology-driven

parameters). Also, users can decide to browse the

repository according to the structure of the knowl-

edge base, using speciﬁc ontology-driven paths and

constraints, similarly to the approach presented in

(Vrochidis et al., 2008).

Figure 6: Picture (from http://www.cg06.fr/cms/annexes/

merveilles/w musee merveilles/) and sketch of an petro-

glyph.

6 RELATED AND FUTURE

WORK

As it is well known, a lot of research has been carried

out the wide context of Cultural Heritage, leading to

the development of many projects and tools. The most

notable example is the “Epoch” Network of Excel-

lence (http://www.epoch-net.org/, contract IST-2002-

507382) recently concluded, which aimed at pro-

viding a clear organizational and disciplinary frame-

work for increasing the effectiveness of work on the

interface between technology and the cultural her-

itage of human experience. The Epoch NoE col-

lected several tools for managing and organizing CH

multimedia content, as for example AMA (Archive

Mapper for Archaeology, http://ama.ilbello.com/),

a web tool for mapping archaeological datasets

to a CIDOC-CRM compliant format, or MAD

(Managing Archaeological Data, http://www.epoch-

net.org/index.php?option=com

content&task=view&

id=216&Itemid=332) an application designed to

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

294

store, manage and browse structured and unstruc-

tured archaeological datasets encoded in a seman-

tic format. Other recent projects that shared sim-

ilar goals with the Epoch NoE are Europeana,

http://www.europeana.eu/portal/ and 3D COFORM,

Tools and expertise for 3D collection formation,

http://www.3d-coform.eu/.

In (Vrochidis et al., 2008) a hybrid multimedia re-

trieval model is presented which provides a search en-

gine capable to perform similar tasks to those we pre-

sented in this paper, without explicit use of a MAS.

Within this wide scenario, our key application is

related to the preservation of Bronze age petroglyphs,

by also taking advantage of the incredible valuable

Bicknell’s collection; we aim, as our ultimate goal, at

designing a semantically annotated multimedia repos-

itory as a reference at European level as a thorough

database of Bronze Age petroglyphs, which would

be a deﬁnite contribution for domain experts, as rock

art sites are spread all over Europe. To this pur-

pose, we proposed a framework able to formalize the

knowledge related to the available multimedia content

through ontologies and to access and query the reposi-

tory by using ad-hoc sketch interpretation algorithms.

Besides the completion of the missing MAS compo-

nents, their integration in the MAS and a careful test-

ing, other future activities include, on the one hand

to interact with the actors of Epoch in order to under-

stand if our goal can be part of the more wide research

community and on the other hand, to compare our ap-

proach with the one in (Vrochidis et al., 2008) in or-

der to evaluate possible improvements. Finally, we

point out that our framework focuses on the manage-

ment, structuring and organization of multimedia con-

tent related to the Bronze Age but a lot of work could

be done also in the “presentation” of this content to

non-expert ﬁnal users. We have already obtained pre-

liminary results (Ancona et al., 2010) on interactions

among 3D virtual worlds, living autonomous agents

and semantic enriched multimedia content. The inte-

gration of those achievements on the presentation side

within this framework will allow us to provide the

domain experts, but also the potential virtual tourists

and curious, with immersive experiences engaging on

both the educational and ludic sides.

REFERENCES

Agirre, E. and Edmonds, P. (2006). Word Sense Disam-

biguation: Algorithms and Applications (Text, Speech

and Language Technology). Springer-Verlag.

Ancona, M., Mascardi, V., Quercini, G., Bogdanovych, A.,

de Lumley, H., Papaleo, L., Simoff, S., and Traverso,

A. (2010). Virtual institutions for preserving and sim-

ulating the culture of Mount Bego’s ancient people. In

VAST 2010.

Azar, S., Couvreury, L., Delfosse, V., Jaspartz, B., and

Boulanger, C. (2006). An Agent-based Multimodal

Interface for Sketch Interpretation. In MMSP-06.

Bellifemine, F. L., Caire, G., and Greenwood, D. (2007).

Developing Multi-Agent Systems with JADE (Wiley

Series in Agent Technology). John Wiley & Sons.

Bicknell, C. M. (1913). A Guide to the Prehistoric Rock

Engravings in the Italian Maritime Alps. Tip. G.

Bessone.

Bozzano, M., Locoro, A., Martelli, M., and Mascardi, V.

(2010). Trattamento del linguaggio naturale tramite

Prolog: un approccio promettente per generare isti-

tuzioni virtuali da testi scritti. In CILC’10. In Italian.

Casella, G., Deufemia, V., Mascardi, V., Costagliola, G.,

and Martelli, M. (2008). An Agent-based Frame-

work for Sketched Symbol Interpretation. J. Vis. Lang.

Comput., 19(2):225–257.

Datta, R., Joshi, D., Li, J., and Wang, J. Z. (2008). Image

Retrieval: Ideas, Inﬂuences, and Trends of the New

Age. ACM Computing Surveys, 40(1).

de Lumley, H. and Echassoux, A. (2009). The Rock

Carvings of the Chalcolithic and Ancient Bronze Age

from the Mont Bego Area. The Cosmogonic Myths

of the Early Metallurgic Settlers in the Southern Alps.

L’Anthropologie, 113(5P2):969–1004.

Doerr, M. (2003). The CIDOC CRM An Ontological Ap-

proach to Semantic Interoperability of Metadata. AI

Magazine, 24(3):75–92.

Juchmes, R., Leclercq, P., and Azar, S. (2005). A Freehand-

sketch Environment for Architectural Design Sup-

ported by a Multi-agent System. Computers & Graph-

ics, 29(6):905–915.

Kaiser, E., Demirdjian, D., Gruenstein, A., Li, X., Niekrasz,

J., Wesson, M., and Kumar, S. (2004). A Multimodal

Learning Interface for Sketch, Speak and Point Cre-

ation of a Schedule Chart. In ICMI’04, pages 329–

330, New York, NY, USA. ACM.

Mascardi, V., Locoro, A., and Larosa, F. (2009). Exploiting

Prolog and NLP techniques for matching ontologies

and for repairing correspondences. In CILC’09.

Papaleo, L., Quercini, G., Mascardi, V., Ancona, M.,

Traverso, A., and de Lumley, H. (2010). Agents and

ontologies for understanding and preserving the rock

art of Mount Bego (extended version). Technical

Report DISI-TR-10-11, Genova University, Italy.

http://www.disi.unige.it/person/MascardiV/Download/

ICAART-tech-report.pdf.

Shishir K. Shandilya, N. S. (2010). A Survey on: Content

Based Image Retrieval Systems. Int. J. of Computer

Applications, 4(2):22–26.

Veltkamp, R. C. and Tanase, M. (2002). Content-Based Im-

age Retrieval Systems: A Survey. Technical report,

Department of Computing Science, Utrecht Univer-

sity.

Vrochidis, S., Doulaverakis, C., Gounaris, A., Nidelkou, E.,

Makris, L., and Kompatsiaris, I. (2008). A hybrid On-

tology and Visual-based Retrieval Model for Cultural

Heritage Multimedia Collections. Int. J. Metadata Se-

mant. Ontologies, 3(3):167–182.

AGENTS AND ONTOLOGIES FOR UNDERSTANDING AND PRESERVING THE ROCK ART OF MOUNT BEGO

295