Methodology for Integration of Fishers' Ecological Knowledge in Fisheries Biology and Management Using Knowledge Representation [artificial Intelligence]

The fisheries crisis of the last decades and the overexploitation of a great number of stocks (FAO 1995) have been due mainly to the inadequacy of scientific knowledge, uncertainties in assessments and/or failures of the management systems. These problems are critical when the management of coastal ecosystems and artisanal fisheries is involved. These systems possess great complexity due to the high number of human factors that influence their functioning and the fishing activity. Smallscale coastal fisheries have a much greater social significance than offshore industrial fisheries, despite the larger economical importance of the latter (only in macro-economic terms). The artisanal coastal fisheries in Galicia (NW Spain) are in a general state of overexploitation derived from the mismatch between management (derived implicitly from models designed for industrial finfisheries) and the biological and socioeconomic context. Freire & García-Allut (2000) proposed a new management policy (based on the establishment of territorial users’ rights, the involvement of fishers in the assessment and management process in collaboration with the government agencies, and the use of protected areas and minimum landing sizes as key regulations) to solve the above problems. As well as a new management system, research should pay special attention to the design and use of inexpensive and rapid methodologies to get relevant scientific data, and introduce local or traditional ecological knowledge of the fishers to the assessment and management process. In the case of artisanal fisheries in Galicia (NW Spain) there are also a number of indicators that reveal overfishing (Freire 1999; Freire 1000a): 1) the virtual depletion and collapse of several stocks (for example lobster, spiny lobster, sea bream) whose catches are irrelevant today but were important historically in the area, 2) the time series of catches that, despite problematic interpretation, show that there has been a decline in many cases from the 1940s-60s to the present time, e.g. crustaceans, and 3) specific assessments, such as on the spider crab in the Ría de Arousa (Freire 1000b) reveal exploitation rates greater than 90% per fishing session. As well as showing indicators of overfishing, the following differential characteristics of the Putting Fishers’ Knowledge to Work: Conference Proceedings, Page 228 artisanal sector complicate the design of successful management systems: Also, following this line of work, it is worth mentioning a fuzzy logic expert system whose knowledge base incorporates fishers' knowledge in the form of heuristic rules (Mackinson and Nottestad 1998). Consequently our approach complements the work in (Neis 1999) and (Mackinson 1998) both in content and methodological aspects. 1. From a biological standpoint, the species harvested by the artisanal coastal fleet of Galicia, and particularly the great majority of invertebrate species, present a number of characteristics which render useless the classical analytical models of finfish population dynamics used in the management of industrial fisheries. These species, sedentary benthic or mobile benthic/ demersal, have a strong and persistent spatial structure and are characterized by the following: 1) complex life cycles (planktonic dispersing larval stages and sedentary or low mobile benthic or demersal postlarval stages), 2) a spatial distribution characterized by the existence of aggregations which are evident on different scales, 3) a population structure that could be defined as meroplanktonic meta-populations in which the postlarval stages make up a chain of local populations along the coast with low migration and dispersal levels, interconnected by a planktonic larval stage, and 4) the aggregated stock-recruitment relationship is not applicable to a segment of a metapopulation. The remainder of the paper is organized as follows. The next section defines the concept of Fishers' Ecological Knowledge (FEK) which is rooted in ethnoscience and cultural ecology traditions. Section 3 argues that given the characteristics of FEK and what we want to do with it, Description Logics (DLs) are a good choice to represent FEK. In section 4 we describe our methodology. A visual terminological language which has been designed to facilitate knowledge input is described in section 5. The paper ends with some conclusions. FISHERS' ECOLOGICAL KNOWLEDGE FEK is a specialized branch of TEK (Traditional Ecological Knowledge). The concept of TEK appeared in the mid-1980s, and social scientists have argued that it represents at least a critical supplement to scientific understanding. Mailhot (1993) gave an explanatory definition of TEK: 2. In an industrial fishery, the relationships between the economic benefits obtained by the fishery and its biological and social complexity is high, which would make it possible to fund and develop intensive lines of research. In terms of the artisanal coastal fisheries of Galicia, the economic yield of each of the species harvested does not appear to be able to support specific lines of research which could complete our incomplete scientific knowledge. ”the sum of the data and ideas acquired by a human group on its environment as a result of the group's use and occupation of a region over many generations”. FEK (Neis 1999) typically includes not only categories of fishes, but also information on behavior, ecology, meteorology and oceanography, and references to time and space that can complement scientific knowledge. Moreover, FEK is an updated understanding that includes the latest changes occurring in the local marine environment. However, those who plan management policies are usually politicians who work unilaterally in collaboration with technicians from the administrations, and disregard entirely the knowledge of the fishers within their field of experience. Some examples that occurred in Galicia in recent years may serve as an illustration. Artisanal fishers used the traditional fish trap (cylindrical and closed) to fish velvet swimming crab and octopus. In order to regulate these resources, the administration required fishers to employ a more selective type of trap (square and open) designed by its technicians to fish exclusively octopus. The fishers bought these new traps and soon discovered that they were inefficient. They Faced with these scenarios, some argue that finding ways to incorporate fishers' participation would improve our capacity to manage fisheries sustainable. Neis (1999) presents a methodology for collecting and integrating fishers' ecological knowledge into resource management, but the formal representation of this knowledge is not addressed. We believe that formal representation using AI (specifically Knowledge Representation) techniques could not only assist in the acquisition and refinement of this knowledge, but could also facilitate comparison with other knowledge systems (scientific knowledge), the observation of possible changes in these over time, and the impact of both knowledge systems on management initiatives. The aim of this paper is to a) show that Description Logics and Terminological Systems are good candidates for this task, b) describe the methodology designed to carry out this task, c) develop a case study implementing this and d) document the evaluation by biologists. Page 229, Garcia-Allut et al: Integrating Fishers’ Knowledge using Artificial Intelligence required more work and produced less. The response of fishers was to replace the new traps with the traditional ones behind the back of the administration. This process went on for several years before the administration recognized its error which had resulted in an economical setback for the artisanal fisheries. The government, in opposition to an important sector of fishers, also opened the fishing season for velvet swimming crab at a critical time of its reproduction, thus putting the stock in danger. This latter situation example continued for several years. Therefore, our main objective is to acquire new knowledge that can be applied to the sciences involved in designing management models for artisanal fisheries in Galicia. The generic scope of knowledge that we will need to achieve the above goals will be centred, in turn, on acquiring knowledge and information on coastal ecosystems, population dynamics, descriptions of habitats and bottom types, interactions and relationships between species, behavior and feeding habits, reproductive zones and seasons, climate (atmospheric and oceanic) influences on the species, stock assessment of fishes, crustaceans and molluscs, reconstruction of the history of marine ecosystems in relatively short periods, etc. After filtering, systemizing and formalizing fishers' ecological knowledge, it can contribute to broaden our understanding of many of these topics. METHODOLOGICAL CHOICE: DESCRIPTION LOGICS It has been recognized by Neis (1999) that the main hurdle associated with combining science and FEK is methodological: finding ways to combine these two knowledge systems. In (Neis 1999) and other works, methodologies and research techniques to acquire traditional knowledge are described. These include: analysis of discourse, selection of information, semiguided open interviews, surveys on specific points of knowledge, analysis of the distribution maps of the resources and habitats drawn up by the fishers (Ames, this volume), and other documents of a functional nature that they may have, such as notebooks and graph interpretations (depth sounder, radar), etc. This work is being done almost exclusively by anthropologists and this knowledge circulates mostly through channels of dissemination of maritime anthropology. If this knowledge could be represented in a formal manner, it could be refined, reused, shared with others or integrated with biological knowledge in a principled way. Therefore Knowledge Representation (KR) plays an important role in improving the knowledge of biologists, technicians, anthropologists and fishers, with the ultimate goal of designing better fisheries policies. Two main properties of FEK are that it is a very large body of knowledge and it is subject to continuous changes. Up to now, anthropologists have seen the work of formalizing FEK as part of their research area. This situation motivated us to seek a methodology where the anthropologist is not only an end-user of the resulting knowledge-based system, but he/she is involved in the knowledge engineering process from the beginning. Anthropologists can certainly break down the domain into its characteristic elements, even possibly express them in a computer language. However these tasks must be accomplished in the framework of a formal model, since the lack of a formal semantic foundation could lead to several problems such as inconsistencies or circular definitions. Therefore, to be successful the Knowledge Representation Language (KRL) must be carefully selected. Epistemological adequacy must derive from the nature of FEK. Note that one of the major components of FEK is the categorization used by fishers to classify components of the environment and the organization of these categories into a system of representation. From a technological perspective we need a language that is both expressive and easy to learn. Implementations of DLs seem to be the right choice. From a logical and formal view, DLs integrate research done in semantic networks, frame systems and other object-oriented representations, and constitute the formal successor of the family of KL-ONE languages (Brachman 1985). During the last fifteen years the main issue of research in Description Logics has been the identification of the sources of intractability. The results of this research allow us to depart from a very basic language and to increase expressiveness while ensuring computational tractability. The primary aim of DLs is to express knowledge about concepts and hierarchies of concepts. DLs have declarative tarskian semantics and can be identified as sublanguages of First Order Logic (FOL). A concept expression is a general description of a class of objects in the target domain. Concept expressions are formed using various constructors, some of them expressing relations with other concepts (roles). Relations expressed by means of roles, can be qualified in several ways (type restrictions, value Putting Fishers’ Knowledge to Work: Conference Proceedings, Page 230 restrictions, number restrictions, etc.). Just by analyzing concept expressions, a taxonomy of concepts following generality-specificity criteria can be built. The efficient implementation of reasoning services is based on this hierarchical structure. The basic blocks of the descriptive languages are atomic concepts and roles. Atomic concepts can be considered as unary predicates, and atomic roles can be considered as binary predicates. Atomic concepts and roles are combined to build complex concepts and roles. Semantics allows the interpretation of concepts as subsets of objects (here called individuals) of the domain and the interpretation of roles as binary relations between objects of the domain. Therefore the extension of a concept is a set of individuals, and the extension of a role is a binary relation between individuals. Also following the semantics of language constructors, the equivalent in FOL of any concept or role expression can be obtained. Satisfiability and subsumption are the basic inferences in DLs. A concept is satisfiable if it can have a on-empty extension. A concept C is subsumed by a concept D if the extension of C is always a subset of the extension of D. Other inference tasks of great utility such as equivalence or classification can be reduced to satisfiability and subsumption. Reasoning about individuals is also provided with these logics. Since the seminal works in the field (Levesque 19 and 1987), reasoning in DLs and the tradeoff between expressiveness and tractability have been deeply studied, leading to important results see Donini (1997) for a survey. Terminological languages (also called concept languages) are implementations of DLs. Classic (Patel-Schneider 1991) and Fact (Horrocks 1998) are examples of well-known terminological languages. These languages allow us to define concepts and roles, to organize them by means of taxonomies, to define individuals and to make inferences on these elements and structures. Practical applications of description logics (terminological systems) using these and other terminological languages exist in a wide variety of domains: data and knowledge management systems (Borgida 1993 and 1995), global information systems (Levy 1995), clinical information systems (Rector 1997), software engineering (Devanbu 1991), etc. In our project, we have chosen to use Classic for several reasons. The language is expressive enough to be useful and limited enough to assure tractable reasoning. The language is simple and small enough to be really usable because it can be learned by non-experts in computer science. Even a methodology for using Classic has been published (Brachman 1991). This knowledge engineering methodology has been elaborated, emphasizing the modeling choices that arise in the process of describing a domain and the key difficulties encountered by new users. The language has additional features that increase usability such as a limited forward-chaining rule system and the possibility of concept definitions written as test functions in a procedural programming language. However, these additional features are designed following the principle that user code cannot subvert the knowledge representation system, that is, these additional features have to be kept opaque and should not destroy the correspondence between the reasoning subsystem and the formal semantics Lisp, C and C++ implementations of Classic exist, and an API (Application Programmer's Interface) is available. The distribution is now being handled by Bell Labs and licenses for research and commercial use can be obtained (ATT 1999). Putting it into practice This section shows our methodology from the following points of view: 1) interdisciplinarity, 2) description of the case study, 3) formulation of the case study and 4) evaluation of the results by a biologist.