Content uploaded by Pascal Perez
Author content
All content in this area was uploaded by Pascal Perez
Content may be subject to copyright.
41A methodology for identifying and formalizing farmers’ representations...
During the last ten years, researchers have been working on natural resource manage-
ment (NRM) and ecosystem modeling using multi-agent systems (MAS) (Carpenter
et al 1999, Lansing 1991, Rouchier and Bousquet 1998). This research has focused
mainly on the interactions between biophysical and social dynamics as a means to un-
derstand the emergent behaviors of a system. A subset of experimental studies resulted
in the combination of MAS modeling and participatory approaches and demonstrated
the ability of participatory MAS modeling (PMASM) to promote discussions among
A methodology for identifying and for-
malizing farmers’ representations of
watershed management: a case study
from northern Thailand
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
Linking modeling tools and the participatory approach for development is not a
common combination. Participatory multi-agent system modeling (PMASM) is a
tool for sharing viewpoints among stakeholders and facilitating the negotiation
process. A key question of this approach is the acquisition and the modeling of
the various stakeholders’ representations. Our research team, whose Asian branch
is represented in this book, tries to formalize the passage from fieldwork to the
model by defining a methodology that can be implemented in the field. This
methodology adapts knowledge engineering acquisition techniques to in-field
stakeholders’ representations for PMASM. In a northern Thailand watershed, we
pursued implementation tests of this methodology. We first explored two ways
to tackle fieldwork (ethnographic and project surveys), both showing weaknesses
and strengths. We then built a first-version diagram syntax used for representing
individual farmers’ representations, and we considered options for analyzing
those diagrams. Finally, we tested the elicited representations by leading farmers,
through game-like sessions, to rebuild a model of their system structured by ele
-
ments and links. Results reveal a great heterogeneity of farmers’ representations,
which we intend to manage by establishing farmers’ synthetic profiles based
on their orientations toward specific elements and aspects of their social and
natural environment. Orientations of those profiles convey different conceptions
of the functioning of the system with which farmers interact. This also results in
decisions and reactions to issues that are different from one profile to another.
The identification and formalization will contribute to the implementation of a
computer model of farmers’ representations. Perspectives are drawn on two ways
to integrate representations into the modeling.
42
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
stakeholders involved in the participatory process and lead them to defining negotiated
scenarios (Bousquet et al 2002, Barreteau 2003, D’Aquino et al 2003, Etienne et al
2003). A key element of this approach is the construction of a shared representation
of the system among participants. It consists of taking into account stakeholders’
representations, emphasizing the differences among those representations, showing
some participants the differences and similarities of the others’ viewpoints, and fa-
cilitating a better understanding of the diverse views of the world. Thus, identifying
and integrating stakeholders’ representations is a necessary step of PMASM and the
question emerging is how to formalize this step. This is the aim of this paper, which
describes the setting up and application of methods for identifying and formalizing
how farmers represent watershed management. This formalization will enable us to
implement stakeholders’ representations into a model in a next phase.
The field site of this research is a catchment located in northern Thailand. A
great diversity of stakeholders intervenes in northern Thailand catchment manage-
ment. There are cultural differences among stakeholders (northern Thai villages in
the lowlands and various minority group settlers in the highlands), but also diversity
in terms of stakeholders’ involvement (farmers using resources, local extension and
development offices, state intervention). Those various stakeholders are all involved at
different levels in NRM and have recently been encouraged to interact more intensively
together. Stakeholders’ interactions in relation to NRM are not always smooth, and
have sometimes led to tensions and conflicts. We tackle the analysis of those interac-
tions from the angle of the representations that stakeholders have of catchment NRM.
Indeed, in northern Thailand, we found that stakeholders have various views about
the functioning of the social and natural system, about the issues to face, and the way
to handle them. Identifying stakeholders’ representations helps in understanding the
functioning of stakeholders’ interactions and the implications of their heterogeneous
points of view.
In an early stage of our research, we laid out the elements of a methodology
for identifying stakeholders’ representations (Becu et al 2003). This methodology is
based on the mutual use of knowledge engineering techniques (Gaines and Shaw 1993,
Menzies 2002) and PMASM. The application of this methodology to the northern
Thailand case study enabled us to identify and formalize a set of individual farmers’
representations. Classifications of farmers’ representations resulted in sets of farmers’
profiles that demonstrate the heterogeneity of farmers’ points of view. In this paper,
we focus on the construction, application, and preliminary assessment of this meth-
odology tested with local farmers. In the first part, we describe the field context and
the modeling background. Then, we present our methodology, its application, and the
results obtained. Finally, we assess the methods used and discuss the heterogeneity
of the individual representations.
Natural resource issues in northern Thailand
Three decades of agricultural transformation in northern Thailand have witnessed
increasing tension in relation to NRM. Permanent settlement of upland community
groups, farmers’ adaptations to market demand, and the increasing degree of state
intervention in the highlands have resulted in increasing interdependencies among
43A methodology for identifying and formalizing farmers’ representations...
stakeholders. The 1997 National Constitution provides members of local communities
the right to “use and preserve their local natural resources and environment” (sec-
tion 46). It also requires the state to “promote and encourage public participation in
the preservation, maintenance, and balanced exploitation of natural resources […]
in accordance with sustainable development principles” (section 79). These obliga-
tions are reinforced by the Eighth National Economic and Social Development Plan
(1997-2001), which calls for a “greater participation of local people and community
organizations in the management of natural resources” (Missingham 2000). Hence,
various local stakeholders are now strongly encouraged to interact and collaborate
on water management issues.
Meanwhile, tensions among stakeholders in relation to NRM are increasing and
open conflicts are sometimes erupting (Vorapien 1994, Kanwanich 1997). In particu-
lar, several governmental and nongovernmental organizations in northern Thailand
claimed that deforestation resulted in a dramatic decrease in water availability during
the dry season. This assumption was repeatedly mentioned by lowland farmers to ac-
cuse upland settlers of reducing downstream flow. However, issues as fundamental
as the relationship between upland agriculture and forest destruction or the impact of
upstream agricultural intensification on downstream agricultural viability are contested
by several experts (Alford 1992, Enters 1995, Schmidt-Vogt 1998). They argue that
the expansion of irrigated schemes and horticulture in the lowlands are responsible
for an increasing water demand. Supporters argue that these evolutions have increased
the demand for water during the dry season, which now has to face a fluctuating water
supply (Walker 2003, Waranoot and Bengtsson 1993). If deforestation and catchment
hydrological equilibrium are often driving social tensions, related issues such as soil
conservation, erosion, or irrigation infrastructure management are also sometimes
leading to highly contentious management options.
Recent literature on environmental management, and catchment management
in particular, places a strong emphasis on achieving negotiated settlements to such
conflicts (Brown et al 1995, Crowfoot and Wondolleck 1990). In northern Thailand,
such approaches are often seen as an appropriate way forward in a social and political
climate that places increasing emphasis on participation. Understanding the interac-
tions among stakeholders having different interests and viewpoints is one step in such
an arrangement. In this perspective, multi-agent-based modeling used together with
knowledge engineering techniques may help explain these interactions.
Modeling representations with multi-agent systems
MAS focus on interactions between agents as a means to understand the emergent
behavior of a system (Ferber 1995). That is how a multi-agent model can simulate
the interactions between two agents gathering a resource, with each having a differ-
ent view about that resource (Epstein and Axtell 1996). The implicit assumption is
that individual behaviors are driven by their specific objectives and perceptions of
the system. Therefore, researchers working with MAS have become increasingly
interested in modeling individual representations.
44
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
Still, modeling the specific nature of representations is not an easy task as the
concept of representation itself is subject to several contrasting theories (Lauriol 1994,
Descola 1996, Hutchins 1999). Two main trends can be identified. The first approach,
known as cognitivist, states that representations are stabilized knowledge structures
that are mentally built using a set of symbols and logical inferences, and that they can
be stored in a long-term memory and reused (Craik 1943, Johnson-Laird 1983). On
the other hand, the constructivist approach states that individual representations are
temporary constructs elaborated through social interactions and communication and
they are highly context-dependent (Piaget 1971). In both cases, knowledge and deci-
sion-making are not fully conscious in the mind, either because some of the elements
and processes that constitute knowledge and representation are said to be unconscious
(Newell 1982) or because the nature of representation is said to be socially constructed
and continuously evolving (Röling 1996).
Hence, the modeling issue comes down to a choice between theoretically de-
signed knowledge or empirically elicited knowledge.
Modeling from theories
So far, there is no unified theory in the field of MAS. Coming from artificial intelli-
gence, the belief-desire-intention architecture has long been the most popular theoreti-
cal framework (Conte and Castelfranchi 1995). More recently, social scientists have
challenged this view and proposed alternative frameworks (Gilbert 1995). Some of
these models, such as the Consumat theory, have been tested against experimental
data (Jager and Janssen 2003).
Companion modeling approach
Companion modeling is a trend of PMASM dedicated to NRM
1
(Bousquet et al 1999).
It involves stakeholders in various phases of the modeling process. Stakeholders
provide feedback about the model structure and the simulations produced thanks
to iterative interactions with the designers. Several versions of the model might be
discussed as its construction evolves (Barreteau and Bousquet 2000). This approach
may also use workshops in which models are created in complete interaction with
stakeholders. During those working sessions, stakeholders design the model using
different model artifacts (computer model, role game) and researchers act as facilita-
tors in this process (Bousquet et al 2002). Two applications of companion modeling
are ongoing in northern Thailand, focusing on issues of deforestation (Promburom et
al, this volume) or soil erosion (Trébuil et al, this volume).
By building models of stakeholders’ representations in a participatory way, this
work serves to create a shared representation and to simulate scenarios. This process
is especially appropriate for taking into account the social construction of representa-
tions and for giving a relevant validation of the model. Now, when looking at previous
experiences, a variety of ways have been used for identifying and integrating stake-
holders’ representations in the models. They may present individual representations
either separately from each other or in an aggregated way. Moreover, and depending
1
http://cormas.cirad.fr/en/reseaux/ComMod/index.htm.
45A methodology for identifying and formalizing farmers’ representations...
on the goal aimed for, the emphasis is given either to the individuals’ representation
of their biophysical and social environment or, when individuals’ behaviors are highly
driven by others’ behaviors, to their representation of the others’ representations of the
environment. Nevertheless, in the community of PMASM users, a common trend is the
use of conceptual models (one may use a single model or a set of models) to express
the shared representation and therefore the individual representations. But, how do we
ensure that the conceptual model holds the individual representations? In some cases,
the identification and integration of individual representations are reached through the
researcher’s understanding of the system dynamics. In other cases, these are ensured
through participatory methods such as role-playing games or group discussions. But,
the identification and integration of individual representations are often not a formal
procedure. Now that PMASM has proven its usefulness in promoting discussions and
negotiated scenarios among stakeholders, we felt the need to reinforce the ways to
reach the creation of a shared representation.
Methodological assumptions
We adapted knowledge engineering techniques to our specific working context,
which deals with NRM and actors often performing ill-defined tasks. A methodol-
ogy based on seven elements that constitute its fundamentals was elaborated during
a study in the Orb Valley in southern France on wine farmers’ perceptions of runoff
and erosion processes (Becu et al 2003). The fundamentals of this methodology are
summarized below.
A constructivist perspective
We acknowledge the constructivist perspective and believe that the nature of rep
-
resentation is socially constructed through people’s interactions with their physical
environment and their social relations. We assume also that representations have a
psychological existence in people’s minds and thus may be elicited. But, we recognize
that these representations may evolve due to the elicitation process itself. Therefore, any
elicited representation should be used as a basis for discussion rather than decision.
The use of elicitation
Our methodology uses elicitation techniques coming from knowledge engineering
as a way to access individual representations. Elicitation consists of asking experts
to describe and give information about a system and to model that information.
Typically in knowledge engineering, experts are humans possessing special skill
or knowledge, derived from training or experience, in some particular field (Gaines
2000). Experts should show abilities in answering questions, explaining results, and
identifying issues. Elicitation focuses on the expert’s knowledge about a domain and
on the way he or she makes decisions. The implicit viewpoint on representation is
thus the cognitive approach.
In the field of knowledge engineering, there are different approaches. Within the
transfer view that we follow, elicitation and modeling of representations are treated as
two successive and independent phases. The eliciting process is composed of a direct
acquisition of information, followed by the interpretation of the collected information.
46
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
Acquisition may be achieved through semistructured interviews, process monitoring,
or ethnographic surveys. These tools are highly complementary as behavioral observa-
tion may help in solving communication shortcomings or misunderstandings (Trimble
2000). Although it is severely criticized by knowledge engineers, we consider that
individual semistructured interviewing is the most appropriate elicitation technique in
the context of our application. When dealing with stakeholders in the context of NRM,
interviews and meetings are common and well accepted by local actors. Moreover,
we believe that the weaknesses of interviews (interpretation biases and inability to
extract tacit knowledge) can be corrected by parallel techniques such as joint field
observations, anthropological surveys, or stakeholders’ zoning.
Associated with semistructured interviews, the interpretation is often made
using the protocol analysis technique, based on the knowledge-level theory (Newell
1982). The principle consists of identifying in the transcript of an interview all the
words and semantic expressions related to the elements and concepts that are relevant
to the project. The experience of knowledge engineers using protocol analysis has
refined and adapted Newell’s knowledge-type classification. Knowledge engineers
have identified different types of what they call knowledge objects and associated
typical words and semantic expressions for each of them (Ehret et al 2000, Gray and
Kirschenbaum 2000). That is how, using those classifications, we can extract the
knowledge objects of a transcript containing a stakeholder’s views on whatever topic
and, by combining those pieces of information, we can obtain a conceptual model of
the stakeholder’s representation. To integrate representations in a running MAS model,
we also adapted the knowledge objects classification to the unified modeling language
(UML) formalism often used in MAS (Le Page and Bommel, this volume) and that
greatly facilitates the implementation phase (Grady et al 1998, Graham 2001).
Taking situated cognition into account
Situated cognition theory considers that representations are context-dependent (Giger
-
enzer and Todd 1999, Menzies 1996). Thus, we try to place the interviewees in a context
that makes sense for the topic of the representation that is examined. In their transect
method, Ross and Abel (2000) make it possible to extract information concerning
spatially distributed processes by interviewing stakeholders during a walk across the
case study area. Similarly, in our methodology, (1) interviews should be done in the
field, at a location relevant to the interviewee’s actions, and (2) the interviewer’s first
question should be related to the interviewee’s main actions at this location.
Use of multi-agent systems
Our main reason for choosing MAS is that it is especially appropriate for taking into
account the heterogeneous social representations of a system and has been proven to
be highly useful in simulating agents with different viewpoints and behavior (Ferber
1995, Etienne et al 2003). Moreover, it can be used to explore stakeholders’ represen
-
tations in a dynamic way, which is useful for our methodology in two ways. On the
one hand, it allows us to check the model consistency according to the stakeholders’
authentication of its different components. On the other hand, simulations developed
with MAS are very efficient communication media as the model presented on a com-
puter screen displays the environment in a simple and synthetic way. One of the best
47A methodology for identifying and formalizing farmers’ representations...
pieces of evidence of this is the selfCormas application, for which Senegalese farmers
were able to discuss MAS results displayed on the screen of a laptop (Bousquet et al
2002, D’Aquino et al 2003).
Northern Thailand application
The northern Thailand application aims at modeling resource management for a small
catchment, with the integration of stakeholders’ representations using the above meth-
odology. Research work was divided into three phases: data collection, data analysis,
and validation. We will first describe the fieldwork context and then the three phases.
As this research is still ongoing, this paper will focus on the elicitation process. Per-
spectives about the modeling process will be presented in the next section.
The context in village highlands and lowlands
Within the framework of a broader collaboration with the Land Development Depart
-
ment of Thailand, we have selected the Pang Da catchment, which occupies 15 km
2
and is located about 30 km northwest of Chiang Mai City. As we were also interested
in lowland irrigated water management systems, we extended the study area to the
portion of the Samoeng River located downstream from its confluence with the Pang
Da River. Thus, this area is hydrologically dependent on the Pang Da catchment. A
rapid rural appraisal was carried out and two main case studies were selected for the
elicitation and modeling of farmers’ representations. Each of these case studies, an
upstream and a downstream village, has specific social, agricultural, and economic
contexts (Fig. 1).
The upstream case study (1,250 m) is a Hmong ethnic group village of 103
households and a population of approximately 700 individuals, called Buak Jan.
Agriculture is characterized by vegetable and flower production. There are two water
sources for irrigation: small streams and a spring in the middle of the village. Streams
are often private property and water sharing occurs principally among relatives, but
the spring is an open-access water source and no collective management rules are
defined at this level. The lo-
cal issue of this case study
is the lack of water for
irrigation during the dry
and warm season (March
to May). Indeed, by Febru-
ary, streams are usually dry
and farmers can count only
on the spring from which
they have to pump water
to irrigate. In addition to
consuming electricity and
money, this spring dries up
during average hydrologi-
cal years at the beginning
of May, resulting in an
Fig. 1. Map of the study area.
48
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
incapacity to irrigate for about 2 weeks. For flower production, which is dominant at
this time of the year, this drought results in a decrease in flower plant production and
sometimes in the death of a part of the flower plants. During dry years, drought can
last for 1 month or more. Free water access to the spring results in a heterogeneous
pattern of individual satisfaction, leading to indifference toward the global water
scarcity problem for farmers who always have water and to irritation for those who
are less successful than others in getting water first.
The downstream case study (540 m) is a northern Thai village of 102 households
and a population of approximately 500 individuals, called Sai Mun. Agriculture, which
is the activity of most villagers, relies mainly on paddy field cultivation. Those paddy
fields are irrigated areas belonging to three irrigated schemes with similar individual
and collective water management. Farmers grow rice for home consumption during
the rainy season and cash crops during the dry season (November to May). Contrary
to what we thought at first, the cropping pattern is not driven by water management
but highly depends on soil fertility. Most farmers had continuously grown garlic for
about 20 years during the cold season as a high-value cash crop. This factor, together
with others, has resulted in a decrease in soil fertility (decrease in soil nutrients, organic
matter, and pH; soil structure disintegration), which has affected garlic and other cash
crop yields for about three years. Farmers react to this problem in heterogeneous ways:
for example, some try new cropping patterns, other than garlic, that are supposed to
improve soil fertility, and others apparently ignore the problem and keep on cropping
as they always did.
As farmers act on their system and react to local issues in different ways, we
assume that these heterogeneous behaviors depend on each farmer’s representations
of the biophysical and social environ-
ment. Describing and modeling those
representations should then lead to a
better understanding of farmers’ actions
and therefore of the system dynamics.
To do this, we conducted an elicitation
process in both case studies that was di-
vided into three phases. Data produced
at each phase were used as inputs for
the next phase (Fig. 2).
Ethnographic and project surveys
Two different interviewing approaches
were used in two separate villages: in
-
dividual semistructured interviews and
individual discussions (some would call
them open interviews) combined with
observations.
In Buak Jan, 12 individual semi-
structured interviews were conducted
within the framework of a formal
research project about water manage-
Fig. 2. Phases of the methodology and informa-
tion transferred between each step.
49A methodology for identifying and formalizing farmers’ representations...
ment established with the village headman. Each interview was done in the farmer’s
field after having met him previously two or three times to ensure a good relationship
between the interviewee and interviewer. We tried to diminish Thai-English interpre-
tation biases by training the translator on the semistructured interviewing techniques
and by conducting the interviews as much as possible in Thai and recording them.
The recorded interviews were translated verbatim afterward, resulting in an English
transcript. However, this necessary translation phase definitely resulted in obvious
losses or misinterpretations of the farmers’ words and even more, considering that
Thai is not the native language of the Hmong people. Topics tackled in the interviews
were defined in collaboration with the headman’s village but more important than
the topics themselves for the focus of this paper is the type of questions asked of the
farmers. As our interest was in collecting the interviewee’s representation of his en-
vironment, topics were introduced through “How does this operate?” type questions
when talking about an environmental state, through “Why is this so?” type questions
when talking about an environmental dynamics, and through “What do you do about
X and why?” type questions when talking about an action. Prompting questions were
then used within each topic, either to invite the interviewee to develop his argument
or to talk about a predefined subtopic. After interviewing, we reread the transcripts
and prepared additional questions that we asked of the same farmers approximately
1 month after the first interview.
The Sai Mun study was conducted with a different fieldwork approach, which
has more to do with ethnographic work than an interviewing approach. As no formal
framework was defined for our presence in the village, the research team developed a
relationship with the villagers through a continuous presence among them for 7 months,
joining them in agricultural activities and discussing various topics about their lives.
As contact with farmers became closer, discussions were refined and more explicit
questions were asked. The types of questions asked were of the same nature as in
Buak Jan (“Why is this so?”, “What is happening here?”). When our understanding
of the system issues became more accurate, we used the soil fertility issue as a way to
structure a discussion guideline about farmers’ representations of their environment.
Fourteen individual discussions following this guideline were conducted in the field
with the help of a translator. As we didn’t want to use the tape-recorder because of
the informal relation of the research with the village, we developed with the translator
a note-taking technique to ensure a minimum loss of information during the process:
(1) during the interview, rapid note-taking; (2) just at the end of the interview, quickly
completing the missing parts of the notes; (3) in the following hours after the interview,
chronological rereading of notes to complete missing parts and, as far as we could
recall, rewriting the conversation in the way that the interviewee expressed it.
As ethnographic or project surveys may be considered as a method on their
own for identifying farmers’ representations, it was important to keep track of the
representations identified at this stage before starting the next research phase. By do-
ing so, we were able to compare the results of ethnographic and project surveys with
those of transcript analysis (which is another method for identifying representations)
and thus discuss the representations’ elicitation aspects of these two approaches. The
results were presented in the format of classifications of farmers’ representations. In
this paper, we present only the classification of Sai Mun farmers built from an ethno-
50
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
graphic-type fieldwork approach. The comparison with the transcript analysis results
will be presented later in this paper.
For Sai Mun, two classifications of farmers’ representations were produced
separately by different members of the research team (Table 1). In classification 1,
farmers are classified according to their behavior toward agricultural partners: work
alone or in contract with companies or institutions. Within the first group, farmers are
divided according to their open-mindedness (open-minded or not). In classification 2,
farmers are classified according to the “wideness” of their representations. The “wide
representations” category corresponds to farmers taking many elements into account
when making decisions about cropping or about resource management, whereas farm-
ers within the “narrow representations” category are analyzing the system in a simple
way (taking few elements into account for decision-making). We also found a parallel
with an initiator/follower classification assuming that initiators need many elements
when making decisions, whereas followers don’t because they base their decisions on
what others have experienced already. For the second group, we distinguish between
farmers who are followers because (1) they focus only on profit maximization and
they don’t want to spend time thinking about biophysical dynamics, and (2) they are
not self-confident for various reasons, mainly social reasons.
Transcript analysis
This phase aimed at extracting through a protocol analysis of the individual transcripts
the elements and relations that would form the individual representations of the farm
-
ers. As a matter of fact, even if knowledge objects used in protocol analysis can all
be classified in terms of elements and relations, their definition can be more precise
than two categories only. However, we chose this classification for simplification as
we intended to use the resulting conceptual model for further discussions with the
farmers. The classification used for the protocol analysis is shown in the last column
of Table 2.
The protocol analysis started with the preparatory phase of the transcripts. When
multiple interviews had been done with the same farmer (as in the case of Sai Mun),
transcripts were merged. The transcripts were then reread farmer per farmer to split
each transcript into various themes. Themes were chosen both according to the themes
Table 1. Classification of farmers in Sai Mun: classification 1 in columns and clas-
sification 2 in rows.
Classification 1 Work alone Work with Not
agricultural classified
Open- Not open- partners
Classification 2 minded minded
Wide representations (initiator) F11, F5 F12, F9, F3 F1
Narrow representations (follower)
Focus on profit maximization F4, F10 F8, F13 F14
Not self-confident F2 F7
In between F6
51A methodology for identifying and formalizing farmers’ representations...
defined before the interviews and discussions and the actual themes discussed by the
farmers (e.g., no specific theme for cropping was predefined before the interviews
and discussions but this theme appeared explicitly during the discussions). Themes
were identical for each farmer in each case study; however, when the information in
a transcript was quite limited, we didn’t feel the need to do the thematic classification.
It appeared to us that the thematic classification was rather more a way to organize the
protocol analysis when information was very rich than an analysis by itself. Moreover,
after their identification with protocol analysis, elements and relations extracted from
each farmer were combined in an individual diagram by whatever themes they were
belonging to at first. The exception to this aggregation of the different themes is the
case of Buak Jan, whose diagrams appeared to be very “wide” (numerous elements
and relations), and which we split into four individual thematic diagrams to make
them easier to understand. Examples of the resulting individual diagrams are given
in Figures 3 and 4.
As shown in these figures, the diagrams resulting from the elicitation process
are not easily readable at first. However, distinctions can be made. Figure 3 shows a
soil-oriented representation of a Sai Mun farmer, whereas other farmers’ diagrams from
the same village show a market- and selling-oriented representation. These orienta-
tions are shown by the type of elements found in the diagrams as well as by the great
number of converging relations going to the element “soil” in Figure 3, for example.
Figure 4 is an example of the higher quantity of information elicited in Buak Jan, as
we already stated above. Thus, the total number of elements for Buak Jan farmer 4 is
41, whereas there is a maximum of 19 elements per individual representation in Sai
Mun village.
Although the diagrams’ first reading could give information about their orienta-
tion, we conducted a qualitative analysis of them to extract more accurate results and
establish a farmers’ classification. Within the literature on qualitative data analysis,
the concept of grounded theory is used when talking about theories formulated from
Table 2. Correspondences among knowledge objects, UML formalism, semantic expressions, and
the classification used for protocol analysis in the northern Thailand case study.
Knowledge object UML
a
formalism Semantic expression Classification used
Concept Class Usually equivalent to nouns Element
(object, person, etc.)
Instance Instance Ex.: “my car” is an instance of “car” Element
Process (task, activity) Operations Ex.: “build a house,” Relation
“design the engine”
Attribute and value Class attribute and Attribute: ex.: “cost,” “age”
instance of attribute’s Value: ex.: “120 kg,” Attribute
value “heavy”
Rule Methods Ex.: “If…, then…”, “Do… until…” Relation
Relationship Association, Usually equivalent to Relation
aggregation, or passive verbs; ex.: “…is
inheritance a…”, “…is part of…”
a
UML = unified modeling language.
52
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
p
H
N
u
trie
n
ts
p
H
N
u
trie
n
ts
Fig. 3. Farmer 11 representation: example of soil-oriented representation in Sai Mun village.
53A methodology for identifying and formalizing farmers’ representations...
Farmer 4 representation of soil
Farmer 4 representation of cropping
Rainfed plot
Crop
Soil
Plot
Road
Not irrigated. Grows
lettuce, carrot, and
potato in RS
if the plot is far away
from the road, it's difficult
to transport the product
The soil is not
fertile so the crops
do not grow well
Gerbera
Vegetables
Terraces
Seedlings
Need to make terraces to grow
and apply manure and fertilizer
Need to prepare
seedlings
Bad soil
Good soil
Is black and
have a texture
with aggregates
Is red and
has a texture
like dust
Sprinkler
Sprinkler is the
best technique
(leaves can drink)
Market
Cash
Crops grown need to
have a wide market
Growing crops
requires a lot of cash
for everything
Soil
Crop
Soil type doesn’t
influence cropping
Irrigation
In winter, the soil
can keep water longer
(irrigate every 2 –
In DS, have to
irrigate every day
Erosion
Slope
On important slopes,
the soil flows a lot.
If the plot is flat, the
soil flows just a little
Chemical fertilizer
Made the
soil bad
In the future, soil should
be improved because it has
been used for a long time
(it has degenerated). So
it needs a lot of manure
Sprinkler
Soil can drink water
very deep but need to
irrigate by hand
for flowers' leaves
King
OHD
Our king said that vetiver
is a wall to protect from water
flow and erosion. I think
I’m going to grow vetiver
What they do is good but
it does not really work because
they built the hillside ditches
very far from each others, so,
when the water flows strongly,
it destroys the ditches.
Soil type doesn’t
influence erosion
Rainfed plot
Crop
Soil
Plot
Road
Not irrigated. Grows
lettuce, carrot, and
potato in RS
if the plot is far away
from the road, it's difficult
to transport the product
The soil is not
fertile so the crops
do not grow well
Gerbera
Vegetables
Terraces
Seedlings
Need to make terraces to grow
and apply manure and fertilizer
Need to prepare
seedlings
Bad soil
Good soil
Is black and
has a texture
with aggregates
Is red and
has a texture
like dust
Sprinkler
Sprinkler is the
best technique
(leaves can drink)
Market
Cash
Crops grown need to
have a wide market
Growing crops
requires a lot of cash
for everything
Soil
Crop
Soil type doesn’t
influence cropping
Irrigation
In winter, the soil
can keep water longer
(irrigate every 2–3 days).
In DS, have to
irrigate every day
Erosion
Slope
On important slopes,
the soil flows a lot.
If the plot is flat, the
soil flows just a little
Chemical fertilizer
Made the
soil bad
In the future, soil should
be improved because it has
been used for a long time
(it has degenerated). So
it needs a lot of manure
Sprinkler
Soil can drink water
very deep but need to
irrigate by hand
for flowers' leaves
King
OHD
Our king said that vetiver grass
is a wall to protect from water
flow and erosion. I think
I’m going to grow vetiver grass.
What they do is good but
it does not really work because
they built the hillside ditches
very far from each other, so,
when the water flows strongly,
it destroys the ditches.
Soil type doesn’t
influence erosion
Fig. 4. Parts of farmer 4 representation: representation of cropping and of soil in Buak Jan village. RS = rainy season, DS = dry season,
OHD = Office of Highland Development.
54
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
empirical observations. Practitioners develop theories through induction based on
observation of a phenomenon (Glaser and Strauss 1967). This bottom-up approach
uses some techniques that are especially relevant for our purpose. The formulation of
a new theory begins by coding the data and formulating relationships among the coded
objects, just as we have done with the elements of the farmers’ representations.
Then we define two criteria—groundedness and density—that refer to the code
frequency and to the relation frequency, respectively (Strauss and Corbin 1990). For
the analysis of the individual representation diagrams, we adapt the two previous
criteria to our data set. We defined two main indicators: the number of elements and
the number of relations within an individual representation. For Sai Mun village, we
have also calculated the number of relations with the element “crop” and the number of
relations with the element “soil.” Crop and soil are chosen because they are among the
main elements used by Sai Mun farmers. Those indicators were then used to establish
classifications of Sai Mun and Buak Jan farmers. In Table 3, for Sai Mun farmers, the
numbers of elements and relations have been expressed in terms of relative quantity
ranges. The same kind of table was made for Buak Jan farmers.
A comparison with the ethnographic approach classification shows that the
extreme groups correspond in both classifications (group 1 corresponds to the “wide”
representations group; groups 5 and 6 correspond to the “narrow” representations and
not self-confident group, except F3). However, the “narrow” representations focusing
on the profit maximization subgroup are distributed among very different groups of
the postanalysis classification. A main reason for this inconsistency is the preanalysis
classification inclination toward cropping and soil issues rather than toward economic
issues. That kind of comparison is interesting as it reveals weaknesses and strengths of
each type of classification. Indeed, the general convergence of the two classifications
for Sai Mun village confirms in some way the relevance of the fieldwork approach
adopted in Sai Mun village, as our preanalysis understanding of the system seemed
quite accurate. However, incoherence such as with the profit-oriented subgroup dem-
onstrates that a preanalysis classification is very dependent on our personal orientation
toward the topic studied. Moreover, when comparing the preanalysis and postanalysis
classifications for Buak Jan, many more mismatches were found and we are tempted
Table 3. Classification of Sai Mun farmers according to number of
elements and range of relations.
Group Farmer No. of No. of relations
elements
a
Total With crop With soil
1 1,9,12 + ++ ++ ++
2 13,14 + ++ ++ +
3 11 ++ ++ + ++
4 5,6 0 0 + +
5 2,8 + – + –
6 3,4,7,10 – – – – –
a
++ = very numerous, + = numerous, 0 = medium, – = few, – – = very few.
55A methodology for identifying and formalizing farmers’ representations...
to attribute this to the Buak Jan fieldwork approach, which resulted in a less accurate
understanding of the system than in Sai Mun.
If the use of groundedness and density criteria resulted in a first interesting
classification, it was unable to convey the orientations of the representations, such as
soil- or market-oriented. Therefore, we worked on a second classification based on
the type of elements embedded in each representation. This will be discussed later on
in this paper.
Uncovering elements of the individual’s representation through “playable
stories”
Establishing a methodology. When it came to the phase of validation of our findings
in terms of the actual individual representations of farmers in both case studies, it
appeared clearly that validating both elements and interactions of our diagrams was
too much to do all at once. Indeed, the total amount of different elements found in
Sai Mun, for example, was more than 90 for all farmers. When we started to count
the number of different relations, we quickly arrived at more than 100 types of inter
-
actions, after which we stopped counting. For the Buak Jan case study, the numbers
were even larger. Validating such great diversity, element by element and relation by
relation, was unrealistic; thus, we decided to focus on validating the important elements
of the individual representations. Therefore, we identified around 60 main elements
for each case study from the analysis of the individual representation diagrams and
used them during individual sessions we conducted with each farmer of the sample.
Those sessions are halfway between gaming sessions and story telling; we thus called
them “playable stories.”
Playable stories aimed to lead farmers to rebuild their world by selecting and
organizing the elements of their world that were dominant in their representational
system. The elements selected by each farmer during those sessions were then com-
pared with the ones in their representation diagram as a means of validation. Each
of the 60 elements mentioned above was therefore transcribed onto a card on which
the name of the element was written in Thai and English (e.g., one card for weir, one
card for trader, one card for rice, etc.). The 60 resulting cards were then placed on a
panel in such a way that farmers could have an overview of the different elements at
a glance (see Fig. 5).
Fig. 5. Cards for each element are placed on a panel for farmers to see them all at a glance.
56
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
To invite farmers to choose cards, as well as with regard to the situated cogni-
tion assumption, a story giving broad elements of the surrounding environment was
recounted during the session so that farmers could locate themselves in a real-world
context. The story told was the same for each farmer of a village and included dif-
ferent periods in which one topic at a time was emphasized. For example, for the Sai
Mun case study, the first period focused on water management and the second and
third focused on soil and market, respectively. Within this virtual world, farmers were
invited to act, make decisions, choose cards that they thought were important for their
way of life, and organize them if they wanted to.
To make those sessions a bit more entertaining, we added features such as
virtual bank notes, which were used to pay and earn money, and meetings within the
story with different stakeholders of the system with which the farmer was invited
to converse (such as a soil scientist, a canal manager, a banker, a trader). Our story
became a kind of gaming session and it was presented to farmers as one.
As a way to combine different approaches for identifying the important elements
of the farmers’ representations with this methodology, we organized the sessions in
three separate and consecutive phases. In phase 1, we presented the board on which
the cards were placed and asked the player to pick the one that he thought was im-
portant for his occupation. During phase 2, we recounted our story, step by step and
year by year (one year is divided into six steps), and asked the farmer to “act” within
this story as explained above. During this phase, the panel with the cards was hidden
from the farmer and, while he was explaining what he was doing within the story, one
of the interviewers was choosing the cards corresponding to the elements mentioned.
When an element mentioned wasn’t already available in the panel, a new card was
added. All cards chosen, in all phases, were placed on a central board visible to all.
During phase 3, we presented the panel to the farmer a second time with all the cards
that he didn’t mention or choose yet and we asked him to pick some new elements
if he wanted to. Then, we discussed the different cards or groups of cards that were
placed on the central board as a way to enrich the discussion. We also used cords to
represent the interactions mentioned by the farmers (see Fig. 6).
Fig. 6. Cards of the board are linked with cords corresponding
to interactions between elements.
57A methodology for identifying and formalizing farmers’ representations...
Assessing the interest of “playable stories” for revealing elements. The research
on and use of those playable stories to validate the elements of farmers’ representations
is still ongoing and only preliminary outputs can be mentioned. One primary output
is that those sessions were able to reveal tendencies of the farmers’ behavior in the
game; for example, some farmers focused more on the market and earning aspects
whereas others were oriented toward soil management. Identification of those main
farmer representation orientations first came from our general impression at the end
of each session; for example, some farmers spoke much more about elements related
to the soil, whereas others were always arguing about prices, incomes, and markets.
But, much more important than these subjective impressions, we were able to identify
and describe those orientations by analyzing the set of cards that were chosen during
the session. Indeed, as we recorded all the cards picked during the session, we could
quantify and analyze objectively what happened during the sessions and this analysis
confirmed the ability of the playable story to reveal the orientations of farmers’ repre-
sentations. In more details, with some cards being used or mentioned by the players
more than other cards, we could weight the relative importance of the different cards.
To do so, during the game we recorded players’ reactions about the cards by arguing,
giving comments, or asking information about a card. We then summed the number
of times a farmer used, mentioned, or reacted to a card and assimilated the sum total
to the weight of the element. We are currently combining these quantitative results
with qualitative data extracted from the game (the verbatim transcriptions of farmers’
reactions about cards) and preliminary results show that they contribute well to the
definition of farmers’ orientations toward specific interests.
Discussion
This paper focuses on the test of various methods for identifying and formalizing
farmers’ representations. We therefore presented the construction, application, and
some preliminary results of those methods. Although this research is still ongoing,
it is possible at this stage to assess the methodology from our experience. Moreover,
an interesting specificity of this methodology is that it was applied at an individual
level. This enabled us to demonstrate the heterogeneity of farmers’ points of view,
which we discuss in the second part of this section.
Methodology assessment
During the course of this research, we have tried diverse approaches for collecting
information. Our aim was to find methodological elements that would tend to more
accurately reflect the stakeholders’ representations collected. An important source
of bias when collecting this kind of information is the relationship between the
interviewer and the interviewee (Portmann and Easterbrook 1992). Factors such as
mistrust between the two persons may lead the interviewee to distort his answers.
Knowledge engineers practicing elicitation techniques also demonstrate that meet-
ing each interviewee several times is useful for creating a trustful relationship (Lépy
1997). The ethnographic and project survey approach used in Sai Mun and Buak Jan
villages, respectively, resulted in a different nature of the relationship between the
interviewer and interviewees. Although it is difficult to assess this relationship objec-
58
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
tively, we believe that the ethnographic approach enabled a more trustful relationship
that resulted in more accurate responses from the interviewees. In contrast, the project
survey approach was much less time-consuming. Moreover, as audio-recording was
used with this approach, it resulted in more information per transcript than with the
ethnographic approach.
The representation diagrams completed in phase 2 (Figs. 3, 4) show interesting
results as they show the elements of the system and the relations among those elements,
but they also carry in their structure the orientations of the person’s representation and
his strategies (that may intervene in the decision-making process). Thus, when looking
at a single individual diagram, one can follow the train of thought that demonstrates
some logic of thinking or strategies. Still, those diagrams have limitations when one
tries to analyze them collectively. We were able to use the different types of elements
and relations to define classes of representations among the individual diagrams, but
we were lacking methods to compare them according to elements such as train of
thought or strategies. This difficulty may also partly be explained by the fact that each
strategy or logic of thinking also contains series of elements and relations. This also
demonstrates that tendencies and similarities can be found among several individuals
in terms of elements referring to a specific topic, but that these individuals organize
those similar elements in different ways.
Preliminary results of playable stories show some elements of interest regarding
the nature of the representation extracted with this method. When compared with the
elements of the representation diagrams, the type of elements extracted with playable
stories are much more oriented toward actions and decision-making. Elements such
as forest, mountain, or underground are never mentioned by the farmers during the
playable stories. Once again, situated cognition theory contains elements of discus-
sion that can explain these differences. Indeed, the context in which the interviewees
are placed during the interviewing phase and the playable story is different. During
the interviews, farmers were asked to discuss their environment in a general way,
explaining processes of various elements and reasons for their thoughts and actions.
In contrast, during the playable stories, interviewees were asked to act in their environ-
ment and eventually to comment on it as well as the reasons for acting in such a way.
Therefore, farmers expressed their representation of the environment oriented toward
action within the playable stories, whereas they had revealed their representation of
the environment in a generic way during the interviews. Thus, current results tend
to show that farmers use some parts of their representation of the environment when
making decisions and performing tasks. Reasons for such behavior may be arising
from simplifications, which are often made during a decision-making process, made
to restrict a choice to its core (Gigerenzer and Todd 1999).
Importance of heterogeneity of representations and modeling perspectives
All the way through the identification process of our methodology, we found specific
perspectives that farmers have for different aspects of their system. Elements and
relations of the diagrams reveal that some farmers are more oriented toward soil,
some toward market and selling aspects, and some toward partnership with private
companies or institutions. Orientations revealed by playable stories complement the
previous and help refine the profile of each farmer. These results are very demonstrative
59A methodology for identifying and formalizing farmers’ representations...
of the heterogeneity of farmers’ perceptions of their social and natural environment
and how they react to specific issues. In the case of Sai Mun, for example, farmers
whose representation is oriented to soil aspects perceive the decrease in soil fertility as
the result of the intensive use of chemical fertilizer associated with garlic production.
They developed a thinking process about the relations among chemical fertilizer (as
well as other inputs), soil, and plant. This process is based on their own experience,
on comparisons with other farmers’ practices, and on technical information they
acquired from the radio or from technicians from local institutions. The conceptions
that resulted from their thinking process as well as the source of information used are
reflected in the diagrams. For example, Figure 3 shows the conception that the farmer
has of the benefit of manure for soil fertility and how he perceives the pH as being
dependent on soil nutrients. These conceptions result in specific decisions. That is
why the farmer of Figure 3 will not grow garlic on a plot in which he thinks the soil
is acidic but will grow soybean instead, or will use manure. Now, in the same vil-
lage, the representation diagrams oriented toward profit aspects reveal a completely
different view and reaction to the soil fertility issue. Those farmers explicitly refer to
the soil as a resource used for production that can be managed. Investment is then the
means to improve soil fertility. Here, the use of chemical fertilizer is not reappraised
and should be completed by additional inputs such as bioorganic fertilizer. Similarly,
those farmers will stop growing garlic if they consider that it is not profitable given
the additional inputs required.
Throughout this example about farmers’ conceptions and reactions to soil fertil-
ity, we showed that the representation diagrams can explain farmers’ way of thinking
and how different conceptions of a system result in different decision-making. The
next step of this research now consists of integrating those different representations
into the modeling. The coauthors of this paper foresee two main possibilities for
integrating representations into the modeling. On the one hand, the model is a direct
transcription of the stakeholder’s representation and all objects of the model cor-
respond to an element elicited within at least one farmer’s transcript. On the other
hand, agents introduced in the model use the representation and those agents interact
with other objects that are coming from scientific knowledge and not from elicited
elements. Figure 7 gives a schematic representation of these two forms of stakehold-
ers’ representation models.
Our aim is now to apply our reflections on the integration of individual rep-
resentations into the modeling to the northern Thailand data set and to submit the
resulting models to the stakeholders. Even if the representations elicited were only
the ones of the farmers of the catchment, after the playable story phase, most farm-
ers spontaneously asked us to organize meetings with this type of playable stories
grouping together diverse types of stakeholders (government institutions, the Land
Development Department, the Royal Project Foundation, etc.). Our perspective is
thus to organize feedback working sessions with all stakeholders’ groups, present
our models, and use them as a way to discuss the diverse representations and ways of
thinking present in the catchment.
60
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
References
Alford D. 1992. Streamflow and sediment transport from mountain watersheds of the Chao
Phraya Basin, northern Thailand: a reconnaissance study. Mountain Res. Dev. 12(3):237-
268.
Barreteau O, Bousquet F. 2000. SHADOC: a multi-agent model to tackle viability of irrigated
systems. Ann. Operations Res. 94:139-162.
Barreteau O. 2003. The joint use of role-playing games and models regarding negotiation
processes: characterization of associations. J. Artif. Societies Social Simul. 6(2).
Becu N, Bousquet F, Barreteau O, Perez P, Walker A. 2003. A methodology for eliciting and
modeling stakeholders’ representations with agent-based modeling. In: Hales D, Edmonds
B, Norling E, Rouchier J, editors. Multi-agent-based simulation iii. 4th international
workshop, MABS 2003, Melbourne, Australia, July 2003, revised papers. Lecture Notes
Artif. Intell. 2927:131-148.
Bousquet F, Barreteau O, Le Page C, Mullon C, Weber J. 1999. An environmental modelling
approach: the use of multi-agent simulations. In: Blasco F, Weill A, editors. Advances in
environmental and ecological modelling. Amsterdam (Netherlands): Elsevier. 219 p.
Bousquet F, Barreteau O, D’Aquino P, Etienne M, Boissau S, Aubert S, Le Page C, Babin D,
Castella J-C. 2002. Multi-agent systems and role games: collective learning processes
for ecosystem management. In: Janssen MA, editor. Complexity and ecosystem manage
-
ment: the theory and practice of multi-agent approaches. Cheltenham (UK)/Northampton
(USA): Edward Elgar Publishers. p 248-285.
Brown V, Smith DI, Wiseman R, Handmer J. 1995. Risks and opportunities: managing envi
-
ronmental conflict and change. London (UK): Earthscan Publications Ltd.
Actual representation elicited
The objects of the model
are representations
An introduced agent uses representations
Elicited representations can be modeled in two different waysElicited representations can be modeled in two different ways
Fig. 7. Two types of stakeholders’ representation models.
61A methodology for identifying and formalizing farmers’ representations...
Carpenter S, Brock W, Hanson P. 1999. Ecological and social dynamics in simple models of
ecosystem management. Conserv. Ecol. 3(2).
Conte R, Castelfranchi C. 1995. Norms as mental objects, from normative beliefs to norma
-
tive goals. In: Castelfranchi C, Muller JP, editors. From reaction to cognition. Berlin
(Germany): Springer Verlag. p 186-199.
Craik K. 1943. The nature of explanation. Cambridge (UK): Cambridge University Press.
Crowfoot JE, Wondolleck JM. 1990. Environmental disputes: community involvement in
conflict resolution. Washington, D.C. (USA): Island Press.
D’Aquino P, Le Page C, Bousquet F, Bah A. 2003. Using self-designed role-playing games and
a multi-agent system to empower a local decision-making process for land use manage
-
ment: the SelfCormas experiment in Senegal. J. Artif. Societies Social Simul. 6(3).
Descola P. 1996. Constructing natures: symbolic ecology and social practice. In: Descola P,
Palsson G, editors. Nature and society: anthropological perspectives. London (UK):
Routledge. p 82-102.
Ehret BD, Gray WD, Kirschenbaum SS. 2000. Contending with complexity: developing and
using a scaled world in applied cognitive research. Human Factors 42(1):8-23.
Enters T. 1995. The economics of land degradation and resource conservation in northern
Thailand: challenging the assumptions. In: Rigg J, editor. Counting the costs: economic
growth and environmental change in Thailand. Singapore: Institute of South East Asian
Studies.
Epstein JM, Axtell R. 1996. Growing artificial societies: social science from the bottom up.
Brookings Press and MIT Press.
Etienne M, Le Page C, Cohen M. 2003. A step-by-step approach to building land management
scenarios based on multiple viewpoints on multi-agent system simulations. J. Artif.
Societies Social Simul. 6(2).
Ferber J. 1995. Les systèmes multi-agents: vers une intelligence collective. Paris (France):
InterEditions.
Gaines BR, Shaw MLG. 1993. Eliciting knowledge and transferring it effectively to a knowl
-
edge-based system. IEEE Trans. Knowl. Data Eng. 5(1):4-14.
Gaines BR. 2000. Knowledge science and technology: operationalizing the enlightenment.
Lecture notes of PKAW2000, Pacific Rim Knowledge Acquisition Workshop, Sydney,
Australia, December 2000.
Gigerenzer G, Todd PT. 1999. Simple heuristics that make us smart. Evolution and Cognition
Series, ABC Research Group, Oxford University Press. 384 p.
Gilbert N. 1995. Simulation: an emergent perspective. Conference on New Technologies in
the Social Sciences, Bournemouth, UK.
Glaser BG, Strauss AL. 1967. The discovery of grounded theory: strategies for qualitative
research. Chicago, Ill. (USA): Aldine.
Grady B, James R, Ivar J. 1998. The unified modeling language user guide. Addison-Wes
-
ley.
Graham I. 2001. Object-oriented methods: principles and practice. Object Technology Series,
Addison-Wesley, 3rd edition.
Gray WD, Kirschenbaum SS. 2000. Analyzing a novel expertise: an unmarked road. In:
Schraagen JMC, Chipman SF, Shalin VL, editors. Cognitive task analysis. Mahwah,
N.J. (USA): Erlbaum. p 275-290.
Hutchins E. 1999. Mental models as an instrument for bounded rationality. Proceedings of the
Dahlem workshop on bounded rationality: the adaptive toolbox, 18 May 1999.
Jager W, Janssen M. 2003. The need for and development of behaviorally realistic agents.
Lecture Notes Artif. Intell. 2581:36-49.
Johnson-Laird PN. 1983. Mental models: towards a cognitive science of language, inference,
62
N. Becu, O. Barreteau, P. Perez, J. Saising, and S. Sungted
and consciousness. Cambridge (UK): Cambridge University Press; Cambridge, Mass.
(USA): Harvard University Press.
Kanwanich S. 1997. Agricultural war: the case of the drying rivers. Bangkok Post, 20 July
1997.
Lansing JS. 1991. Priests and programmers. Princeton, N.J (USA): Princeton University
Press.
Lauriol J. 1994. Approches cognitives de la décision et représentation sociale. Rev. Int. Systém.
8(2):139-166.
Lépy N. 1997. Expertise et acquisition des connaissances en intelligence artificielle. In: SPI’97,
Rennes, France.
Menzies TJ. 1996. Assessing responses to situated cognition. Proceedings of the KAW ‘96:
Banff Knowledge Acquisition Workshop, 21 September 1996.
Menzies TJ. 2002. Knowledge elicitation: the state of the art. In: Chang SK, editor. Handbook
of software engineering and knowledge engineering. Volume II. Singapore: World-
Scientific.
Missingham B. 2000. Participatory development in Thailand: a review of some relevant litera
-
ture. Integrated Catchment Assessment and Management Centre, Australian National
University, Canberra, Australia.
Newell A. 1982. The knowledge level. Artif. Intell. 18:87-127.
Piaget J. 1971. Science of education and the psychology of the child. New York (USA): Viking
Press. (In French: Psychologie et pédagogie, 1969.)
Portmann M-M, Easterbrook SM. 1992. PMI: knowledge elicitation and De Bono’s thinking
tools. Proceedings of EKAW-92, Sixth European Workshop on Knowledge Acquisition
for Knowledge Based Systems, Heidelberg, Germany, May 1992.
Röling N. 1996. Towards an interactive agricultural science. Eur. J. Agric. Educ. Ext. 2:35-
48.
Ross E, Abel N. 2000. Eliciting mental models of landscape processes: the transect method. In:
Moore GT, Hunt J, Trevillion L, editors. Environment-behaviour research on the Pacific
Rim. Sydney (Australia): Faculty of Architecture, University of Sydney. p 295-310.
Rouchier J, Bousquet F. 1998. Non-merchant economy and multi-agent systems, an analysis
of structuring exchanges. Lecture Notes Artif. Intell. 1534:111-123.
Schmidt-Vogt D. 1998. Defining degradation: the impacts of swidden on forests in northern
Thailand. Mountain Res. Dev. 18(12):135-149.
Strauss AL, Corbin J. 1990. Basics of qualitative research: grounded theory procedures and
techniques. Newbury Park, Calif. (USA): Sage.
Trimble JA. 2000. Structuring knowledge acquisition in software development projects. South
African Comp. J. 26:172-180.
Vorapien P. 1994. Villagers say fires set deliberately. Bangkok Post, 13 March 1994.
Walker A. 2003. Agricultural transformation and the politics of hydrology in northern Thailand.
Dev. Change 34(5):941-964.
Waranoot T, Bengtsson BE. 1993. A conflict over natural resources between highland and
lowland populations in Thailand. Paper read at Tropical Rainforest Research—Current
Issues, April 1993, at Bander Seri Begawan.
Notes
Authors’ address: CIRAD-Ca, 73 rue J.F. Breton, 34398 Montpellier Cedex 5, France, email:
nicolas_becu@yahoo.com.
