ALIGNING AGENT COMMUNICATION PROTOCOLS

A Pragmatic Approach

Maricela Bravo and Martha Coronel

Morelos State Polytechnic University, Jiutepec, Morelos, Mexico

Keywords: Communication protocols in multi-agent systems, ontologies.

Abstract: Nowadays, there is a clear trend in using common ontologies for supporting communication interoperability

between multiple heterogeneous agents over Internet. An important task that must be solved before

implementing ontology-based solutions is the identification of semantic relations to establish alignments

between communication primitives. A frequent methodology for aligning different communication

primitives consists of processing definitions provided by human developers based on syntactical

classification algorithms and semantic enhancement of concepts. We think that the information provided by

human developers represents an important source for classification. However, to obtain real semantics, we

believe that a better approach would analyze the usage of the primitive in the communication protocol. In

this paper we present a pragmatic approach for aligning communication primitives, considering their usage

in the protocol. To evaluate our solution we compare the resulting relations and show that our approach

provides more accuracy for relating communication primitives.

1 INTRODUCTION

The problem of communication heterogeneity in

multi-agent systems (MAS) has become a classical

research topic (Furlan 2001, Labrou 1999, Chaib-

Draa 2002). Although there are important research

communities developing standard agent

communication language (ACL) specifications like

KQML (Finning, 1994) or FIPA ACL (FIPA, 2003),

implementations of such specifications in MAS

differ in syntax and semantics, causing

heterogeneity. Ontologies have shown good results

for supporting communication interoperability

among multiple heterogeneous agents over Internet

(McGuinness 2000, Noy 2000, Guido 2007), but the

design and implementation of ontologies, requires a

human expert to analyze and compare among those

communication implementations to discover and

define relations among primitives. Then research

efforts to propose mechanisms for identifying

relations among multiple implementations of ACL

will benefit the automatic deployment of

autonomous agents over Internet.

In this paper we deal with the problem of

identifying semantic relations to establish

alignments between different implementations of

ACL, considering the message representation and

the protocol of conversation. Aligning is the task of

finding semantic relationships between primitives

from two or more communication protocols. Figure

1 shows an example of relations between two sets of

primitives, where a dashed line represents a “is

similar to” relationship and a continued line

represents a “is equal to” relationship.

Figure 1: Relations between two sets of primitives.

A frequent methodology for aligning different

communication primitives consists of processing

definitions provided by human developers, based on

classification algorithms and applying semantic

enhancement of concepts. We think that the

 CFP

 Propose

 Accept

 Reject

 Terminate

 Acknowledge

 Modify

 Withdraw

 RFQ

 Initial_offer

 Accept

 Reject

 Offer

 Counter_offer

Communication

protocol of agent A

Communication

protocol of agent B

200

Bravo M. and Coronel M. (2008).

ALIGNING AGENT COMMUNICATION PROTOCOLS - A Pragmatic Approach.

In Proceedings of the Third International Conference on Software and Data Technologies - PL/DPS/KE, pages 200-205

DOI: 10.5220/0001894702000205

 SciTePress

information provided by human developers is crucial

for classification and enhancement of concepts.

However, to obtain real semantics, we believe that a

better approach would also consider the pragmatic

usage of communication primitives in the

communication protocol.

In this paper we present a pragmatic approach for

aligning communication primitives, considering

their usage in the protocol. To evaluate our solution

we compare the resulting relations and show that our

approach provides more information for relating

communication primitives.

The rest of the paper is organized as follows. In

section two we present related work with the subject

of this research. In section three we describe the

general procedure for aligning communication

protocols. In section four we present a case study to

show the applicability of our approach. In section

five we evaluate the results and finally, in section six

we conclude.

2 RELATED WORK

The problem of communication interoperability in

MAS represents a common topic, which has been

researched from different perspectives. (Guido et al.,

2006) presented a common ontology of agent

communication languages to bridge the gap between

two approaches of defining semantics: mental

attitudes and social commitments. (Fourlan de Souza

et al., 2001) defined the problem of interoperability

and presented Saci (Simple Agent Communication

Infrastructure), a tool for programming

communication among distributed agents and a

CORBA bridge to overcome the interoperability

problem. (Labrou et al., 1999) described the

interoperability problem between agents. They

stated that any solution should take into

consideration three aspects: a) various languages,

representing different programming paradigms, b)

different hardware platforms and operating systems,

and c) few assumptions about the internal structure

of agents. They also described two possible layers of

solutions: translation between languages and

guaranteeing that the semantic content is preserved

among applications. (Chaib-Draa, 2002) presented

the related work and trends on semantics of ACL, he

compared the semantics of KQML and FIPA-ACL,

and identified that in both cases communicative acts

are described in terms of beliefs, intentions, desires

and similar mental states. Finally he concluded that

agents are almost never programmed using such

mental states directly. Therefore it is almost

impossible to verify whether the messages are used

correctly by the agents and the link between theory

and practice in ACL use is still very big.

To solve the problem, the most common solution

reported in literature consists of defining and using a

shared ontology and a translation approach between

the different ACL implementations. But the

implementation of such a solution requires a human

expert to analyze and compare among those ACL

implementations to discover and define relations

among primitives. In this specific task many authors

have presented different techniques and algorithms

for aligning and matching vocabularies in various

research areas, such as data base schema integration,

knowledge engineering, natural language processing

and information systems integration. In data base

research (Rahm, 2001) argues that “match” is a

fundamental operation to manipulate data schemas,

which takes two schemas as input and produces a

mapping between elements that correspond

semantically each other. (Batini, 1986) presented

various data base schema integration methods and

established three integration phases: schema

comparison, schema conforming and schema

merging. In the area of knowledge engineering

various methods have been proposed. Chimaera

(McGuinness, 2000) is a semi-automatic merging

and diagnosis tool developed by the Stanford

University Knowledge Systems Laboratory. It

provides assistance in the task of merging

knowledge bases produced by multiple authors in

multiple scenarios. PROMPT (Noy, 2000) is an

algorithm that provides a semi-automatic approach

to ontology merging and alignment. PROMPT

determines possible inconsistencies in the ontology,

which result from the user actions, and suggests

ways to remedy these inconsistencies.

In the above related works we can see a common

concern in the problem of communication

heterogeneity in MAS. We can appreciate also that

there are many aspects in communications that can

cause heterogeneity. Various authors have reported

solutions based on the incorporation of ontologies,

and the need for aligning concepts from different

sources. Aligning is a task that depends on the

provided semantics of communication primitives,

but due to different implementations of agents, such

semantics may differ from one agent to another. In

this work we are dealing with communication

protocols, where documentation of primitives is

generated during development time. We take as

input those descriptions and extract keywords to

generate classifications according to the type of

primitive. After classification we compare such

ALIGNING AGENT COMMUNICATION PROTOCOLS - A Pragmatic Approach

201

classifications with the pragmatic usage of the

primitive in the communication protocol. In contrast

to reported works, in this paper we present a

combined approach, which considers the

information included in protocols.

3 ALIGNING COMMUNICATION

PROTOCOLS

Aligning is the task of finding semantic relations

among different concepts, in this work we are

dealing with different ACL implementations, an

ACL implementation consists of descriptions of

communication primitives and definitions of

protocols (sequences of messages). To find semantic

relations among those communication primitives,

most aligning solutions consider only the

descriptions provided by human developers of

agents; on the contrary, in this work we obtain the

meaning of a communication primitive by analyzing

its usage in the communication protocol. Therefore,

we evaluate the moment when the primitive is issued

in the communication protocol.

Figure 2 shows the general architecture for

aligning communication primitives. The solution is a

semi-automatic approach; it has an automatic

extraction and classification of primitives, based on

the use of a Bayesian-based classification, and a

pragmatic analysis of primitives in protocols, based

on the use of Finite State Machines (FSM). The

overall process for helping agent developers and

system integrators in the task of deploying multiple

software agents over Internet is described next:

a). Extraction of Communication Primitives. The

first step consists of obtaining the

communication primitives and their

descriptions. This process is executed by

extracting the natural language descriptions of

communication primitives from the agent

communication interface. Currently we are

getting this information through a Web-based

environment.

b). Classification of Primitives. Once communication

primitives have been acquired and their

descriptions preprocessed, the next step consists

of classifying the primitives. According to the

classification scheme proposed by (Müller,

1996), communication messages are divided

into three groups: starters, if they initiate a

communication, reactors, if they react on a

given statement and completers, whether they

complete a communication. We took this

classification because it considers the moment

in the sequence of messages exchanged. We use

a previously implemented ontology, which

defines classes, subclasses and attributes. We

apply a Bayes classifier to identify to which

class each communication primitive belongs to.

c). Align Primitives. The next step consists of

aligning primitives using a pragmatic approach,

based on the use of finite state machines (FSM),

in order to compare the primitive’s real usage in

the communication protocol. We believe that a

combined strategy based on the analysis of

syntactical description and protocol pragmatics

provides more information of the relatedness of

the communication primitives.

d). Establish Relationships. Based on the previous

classification and aligning steps we can proceed

to define semantic relationships between

communication primitives, according to the

following rules:

 Two communication primitives are equal (EQ)

if they belong to the same class, and if they

have the same usage in the FSM.

 Two communication primitives are

similar_pragmatic (SP) if they do not belong to

the same class, but they have the same usage in

the FSM.

 Two communication primitives are

similar_semantic (SS) if they belong to the

same class, but they do not have the same usage

in the FSM.

 Two communication primitives are different

(DF) if they do not belong to the same class,

and if they do not have the same usage in the

FSM.

4 A CASE STUDY

In this section we present a case study to show the

applicability of our approach. The objective of this

case is to align a set of communication primitives

from three different agents using their natural

language descriptions and applying the process

described in Section 3.

1. Extraction of Communication Primitives. Let A, B

and C be the names of the agents that will

participate in bilateral communication processes.

In table 1 we present the set of communication

primitives and descriptions that agents use for

formulating messages.

ICSOFT 2008 - International Conference on Software and Data Technologies

202

Figure 2: Architecture for classification and relation generation.

Table 1: Descriptions of communication primitives.

A Communication primitives

{(CFP, “Initiate a communication process by calling for

proposals”),

(Propose, “Issue a proposal or a counterproposal”),

(Accept, “Accept the terms specified in a proposal

without further modifications”),

(Terminate, “Unilaterally terminate the current

communication process”),

(Reject, “Reject the current proposal with or without an

attached explanation”),

(Acknowledge, “Acknowledge the receipt of a

message”),

(Modify, “Modify the proposal that was sent last”),

(Withdraw, “Withdraw the last proposal”)}

B Communication primitives

{(Initial_offer, “Send initial offer”),

(RFQ, “Send request for quote”),

(Accept, “Accept offer”),

(Reject, “Reject offer”),

(Offer, “Send offer”),

(Counter-offer, “Send counter offer”)}

C Communication primitives

{(Call for proposal, “Initiate a call-for-proposal”),

(Propose proposal, “Send a proposal or a

counterproposal”),

(Reject proposal, “Reject the received proposal with or

without an attached explanation”),

(Withdraw proposal, “Withdraw the previous proposal

that was sent”),

(Accept proposal, “Accept the terms and conditions

specified in a proposal without further modifications”)

(Change proposal, “Change the proposal that was sent”)

(Inform proposal, “Inform the receipt of a proposal”)

(Terminate communication, “Unilaterally terminate the

communication process”)}

2. Classification of Primitives

According to the classes in the Ontology described

in section 3, any communication primitive can be a

starter, a reactor or a completer, depending on the

time during the communication protocol, when it

occurs. We preprocessed the descriptions of the

primitives to extract keywords, and then using the

definitions from the Ontology we applied a Bayes

classifier to identify to which class each

communication primitive belongs to. Table 2

presents the resulting classification of

communication primitives of agents A, B and C.

Table 2: Resulting classification of primitives.

Starter Reactor Completer

CFP

Propose

Modify

Withdraw

Acknowledge

Reject

Terminate

RFQ

Initial_Offer

Offer

Counter_Offer

Reject

Call for

proposal

Propose

proposal

Withdraw

proposal

Change proposal

Inform proposal

Accept proposal

Reject proposal

Terminate

communication

3. Align Primitives

To align communication protocols we need first to

calculate the total number of communication links

and identify the pairs of agents that may participate.

a) Number of Communication links

Considering a set of n agents, the possible

number of peer to peer communication links

Agent

communication

interface

Agent

communication

interface

Agent

communication

interface

Agent

communication

interface

Extraction of

descriptions

AUTOMATIC EXTRACTION AND CLASSIFICATION

PRAGMATICAL ANALYSIS

Text

Descri

tions

Text

processing

List of

words

Classification

rimitives

Ontology

Rules

Relation

generation

FSM

generation

Calculus of

Number of

Possible

Communication

Links.

Primitives

usa

Comparison

Among each

Pair of

Communication

Link using FSM.

ALIGNING AGENT COMMUNICATION PROTOCOLS - A Pragmatic Approach

203

among them is n

. However, as we are

evaluating heterogeneity, we need to extract

the number of communication links where

agents are equal, which is n. We also

considered that a communication link between

agents (a, b) has the same heterogeneity as a

communication link of agents (b, a), thus we

reduced the number of different

communication links dividing by 2.

CL = (n

-n)/2

CL = (3

-3)/2 = 3

(1)

B) Set of Different Communication links

Considering a set of agents, the set of different

communication links is given by:

DCL = { (a

, a

), (a

, a

), … , (a

, a

) }

DCL = { (A, B), (A, C), (B, C) }

(2)

For each different communication link, we used a

FSM to compare the real implementation of the

primitives in the communication protocol. Figure 3

shows the aligned protocols of the different

communication links:

a). FSM of communication protocol between agents

A and B, which share the neccessary

communication primitives to support

interoperability;

b). FSM of communication protocol between agents

A and C, both protocols share the three classes of

communication primitives and other specific

primitives: Terminate, Acknowledge and Inform

that both agents use; and

c). FSM of communication protocol between agents

B and C, with the neccessary communication

primitives to support interoperability.

4. Establish Relationships

Taking the classification of primitives shown in

Table 4 and the aligment of protocols shown in

Figure 2, we finally established semantical

relationships between primitives following the set of

rules presented in Section 3. We defined only equal

and similar relations for primitives that are

syntactically different.

We did not established differences, because this kind

of relations will not support communication

interoperability. However, they are important to

measure heterogeneity and to propose another

solution based on a learning approach. Results of

this process are shown in Table 3. To define

relations we used the form:

Figure 3: Relations between two sets of primitives.

Table 3: Semantic relations primitives.

CL(A, B) CL(A, C) CL(B, C)

EQ(A, CFP, B,

RFQ)

EQ(A, CFP, C, Call

for proposals)

EQ(B, RFQ, C,

Call for proposals)

EQ(A,

Propose, B,

Offer)

EQ(A, Propose, C,

Propose proposal)

EQ(B, Offer, C,

Propose proposal)

EQ(A, Modify,

Counter_offer)

EQ(A, Modify, C,

Change proposal)

EQ(B,

Counter_offer, C,

Change proposal)

SP(A, Propose,

Initial_Offer)

EQ(A, Withdraw, C,

Withdraw proposal)

EQ(B, Accept, C,

Accept proposal)

SS(A, CFP, B,

Initial_Offer)

EQ(A,

Acknowledge, C,

Inform proposal)

EQ(B, Reject, C,

Reject proposal)

EQ(A, Accept, C,

Accept proposal)

SP(B,

Initial_Offer, C,

Propose proposal)

EQ(A, Reject, C,

Reject proposal)

SS(B,

Initial_Offer, C,

Call for proposals)

EQ(A, Terminate, C,

Terminate

communication)

CFP/RFQ

Initial_Offer/

Offer/

Propose

Modify/

Counter_Offer

Reject

CFP/

Call for

proposal

Propose

proposal/

Propose

Modify/

Change

Terminate

proposal

Reject

proposal

Terminate

communication

Inform/

Acknowledge

RFQ/

Call for proposal

Propose proposal/

Offer

Counter_Offer/

Change

Reject

proposal

Reject

oposa

ICSOFT 2008 - International Conference on Software and Data Technologies

204

REL(

, A

)

where

is the agent issuer of primitive P

is the agent issuer of primitive

5 EVALUATION

To evaluate our approach we considered important

to measure heterogeneity among the set of

heterogeneous agents. For this porpouse we defined

heterogeneity as a numerical measure to compare the

results before and after obtaining relations. The level

of heterogeneity results from dividing number of

different communication primitives (DCPT, without

relationship) by the total number of communication

primitives (CPT), see Formula 3.

Level of heterogeneity = DCPT / CPT (3)

Figure 4 shows the level of heterogeneity for

each different communication link, this measure was

calculated before and after defining relations

between primitives.

Figure 4: Level of heterogeneity before and after obtaining

semantical relations.

6 CONCLUSIONS

In this paper we have presented a pragmatical

approach for aligning communication primitives

between multiple heterogeneous agents. Our

approach is based on classification of human

provided descriptions of primitives and the analysis

of their usage in the FSM. The result of this analysis

helps the developer to identify equality and

similarity relationships between primitives by

comparing the provided descriptions with the real

usage of the primitive in the protocol.

Le level of heterogeneity measure was used to

evaluate how the resulting semantical relations

reduce the level of heterogeneity in communication

scenarios.

REFERENCES

Guido Boella, Leendert W. N. van der Torre. 2006. A

Foundational Ontology of Organizations and Roles. In

Proceedings of 4th International Workshop, (DALT

2006), Hakodate.

Furlan de Souza, M. A., Hübner, F. J., Sichman, J. S.,

Varella, M. A., 2001. Interoperability in Multi-Agent

Systems: Lessons Learned. In Proceedings of 2

Workshop on Multi-Agent Systems: Theory and

Applications (MASTA 2001), Portugal.

Labrou, Y., Finin, T., Peng, Y., 1999. The Interoperability

Problem: Bringing together Mobile Agents and Agent

Communication Languages. In Proceedings of th 3

Hawaii International Conference on Systems Science.

Chaib-Draa, B., 2002. Dignum, F., Trends in Agent

Communication Langugage. Computational

Intelligence, Vol. 18, No. 2.

Finning, T., Fritzon, R., and McEntire, R., 1994. KQML

as an agent communication language. In Proceedings

of the 3rd International Conference on Information

and Knowledge Management, Maryland.

FIPA, 2003. Foundation for Intelligent Physical Agents.

FIPA Specifications, available at http://

www.fipa.org/specifications/index.html.

Rahm, E., and Bernstein, P., 2001. A Survey of

Approaches to Automatic Schema Matching. The

VLDB Journal, Vol 10, pp. 334-350.

Batini, C., Lenxerini, M., and Navathe, S., 1986. A

Comparative Analysis of Methodologies for Database

Schema Integration. ACM Computing Surveys, Vol.

18, pp. 323-364.

McGuinness, D., Fikes, R., Rice, J. And Wilder, S., 2000.

The Chimaera Ontology Environment. In Proceedings

of the 7th National Conference on Artificial

Intelligence (AAAI 2000), Texas.

Noy, F. N., and Musen, M. A., 2000. PROMPT,

Algorithm and Tool for Automated Ontology Merging

and Alignment. In Proceedings of 17th National

Conference on Artificial Intelligence (AAAI 2000),

Texas.

Müller, H. J., 1996. Negotiation Principles, Foundations of

Distributed Artificial Intelligence. In G.M.P. O´Hare,

and N.R. Jennings, New York: John Wiley & Sons.

ALIGNING AGENT COMMUNICATION PROTOCOLS - A Pragmatic Approach

205