Geospatial semantics: beyond ontologies, towards an enactive approach

Geospatial semantics: beyond ontologies, towards an enactive approach Pasquale Di Donato LABSITA – Sapienza University of Rome, Piazza Borghese 9, 00186 Rome, Italy. pasquale.didonato@uniroma1.it Abstract . Current approaches to semantics in the geospatial domain are mainly based on ontologies, but ontologies, since c ontinue to build entirely on the symbolic methodolog y, suffers from the classical problems, e.g. the symbol grounding problem, affecting representational theories. We claim for an enactive approach to semantics, where meaning is considered to be an emergent feature arising cont ext-dependently in action. Since representational theories are unable to d eal with context, a new formalis m is required toward a contextual theory of concepts. SCOP is considered a promising formalism in this sense and is briefly described. Keywords . Semantics, enactive cognition, quantum like, SCOP 1 Introduction The current scene of Geographic Inform ation (GI) is characterised by the provision of services, in a distributed information systems environment, that enable to integrate distributed information resources Dealing with data integration basically implies addressing two main types of heterogeneity: data heter ogeneity and semantic heterogeneity. Data heterogeneity refers to differences in data in terms of data type and data formats, while semantic heteroge neity applies to the meaning of the data (Hakimpour 2001); se mantic heterogeneity ma y consists of naming 2 Pasqual e Di Donat o heterogeneities -when different words/expressions are used for the same (semantically alike) concept- a nd conceptual heterogeneities -when different people and or disciplines have a different interpretation, conceptualisation of the same “thing”- (Bishr 1998). The Open Geospatial Consortium and the ISO TC 211 provide specifications and standards supporting the deployment of geospatial web services. These specifications and st andards address the interoperability issue at syntactic level, but are limited in terms of semantics and do not provide a consistent model for the semantics integration/composition of geospatial services (Einspanier 2003). Coping with semantic interoperability is a challenging task, since it has more to do with how people perceive and give m eaning to “things”, rather then with integrating software comp onents through standard interfaces (Harvey 1999). Semantics deals with aspects of meaning as expressed in a language, either natural or technical such as a computer language, and is complementary to syntax which deals with the structure of signs (focusing on the form) used by the language itself. In the area of distributed data sources and services, semantic interope rability refers to the ability of systems to exchange data and functionalities in a meaningful way . Semantic heterogeneity occurs when there is no agreement of the meaning of the same data and/or service functionality. Data creation happens in a context or in an application doma in where concepts and semantics are clear to the data creator, either because they are explicitly formalised either they are naturally applied due to a yearly experience. But with distributed data resources this context is missed and unknown to the end user. This means th at, in order to achieve semantic interoperability, semantics should be formally and explicitly represented (Kuhn 2005). Current approaches to overcome sema ntic heterogeneities rely on the use of ontologies and reasoning engines for concepts matching am ong different ontologies. The main drawback is that ontologies, being forms of a priori agreements, are decontextualised and decontextualise experience; instead our assumption is that semantics reconciliation depends on contextual human sense-making. This claims for a new formalism for geospatial semantics. The rest of this paper is organised as follows: section 2 deals with the semantics issue in the broader context of cognitive sciences; section 3 briefly summarises current approach es to semantics in distributed information systems; section 4 summarises curr ent approaches to semantics in the GI arena; secti on 5 introduces the SCOP formalism Geospatial semantics: beyond ontologies, towards an enactive appr oach 3 2 Semantics There is a general agreement on considering that semantics deals with relationships between linguistic expressions and thei r meaning; but when it turns do define such relationships opinions highl y diverge (Gärdenfors 2000) in a dispute which is mainly philos ophical. Dealing with formal semantics means opting for one of the two following paradigms, and the choice is mostly philosophical: • Realistic semantics, which co mes in two flavours: - Extensional: in extensional se ma ntics terms of a language L are mapped onto a “world” W . The main aim is to determine truth conditions for sentences in L against W . Extensional semantics is rooted in Tarski’s model theory for first order logic, where sentences from a language get their meaning via a correspondence to a model assumed to be a representation of the world: this meaning is independent of how people understand it; - Intentional: in intentional semantics the language L is mapped onto a set of possible worlds , and the aim continues to be that of providing truth conditions for sentences in L ; • Cognitive semantics: in cognitive seman tics the meanings of sentences are “mental”, and linguistic expressions are mapped onto cognitive structures. According to Gärdenfors (2000): (i) meaning is a conceptualisation in a cognitive model and is independent of truth; (ii) cognitive models are mainly perceptual; (iii) semantic elemen ts are based on spatial and topological objects, and are not symbols; (iv) cognitive models are image sche matic and not propositional; (v) semantics is primary to syntax; (vi) concepts show prototype effects. In order to analyse the semantics issu e in a broader context we need to shift our focus to developments in the field of cognit ive sciences. A critical review of cognitive sciences evolution can be found, for example, in (Anderson 2003), (Froese 2007), (Steels 2007), (Dreyfus 2007), (Gärdenfors 1999), (Licata 2008). Cognitive Sciences is an interdisciplin ary field of investigation, with ideas coming from several disciplines such as philosophy, psychology, neurosciences, linguistics, computer science, anthropology, biology, and physics. The main aim of cognitive sci ences is trying to answer questions such as “What is reason? How do we make sense of our experience? What 4 Pasqual e Di Donat o is a conceptual system and how is organised? Do a ll people use the same conceptual system?” (Lakoff 1987, xi) Two main different approaches try to answer these questions in different ways. The traditional approach claims that reasoning/cognition is essentially a form of information processing oriented to problem solving. It comes in two flavours: cognitivism , which sees the human brain as a deterministic machine manipulating symbols in an algorithmic way; connectionism , which uses a sub-symbolic representation and considers cognition as emergent from a network of atomic com ponents (Artificial Neural Networks). The new approach is characterised by the so-called embodied-embedded mind hypothesis and its variants and extensions, such as situated cognition , and enactivism . Two terms are borrowed from Lakoff (1987) to indicate the two approaches; objectivism for the traditional approach, and experiential realism for the new approach: 2.1 Objectivism Objectivism assumes that reason and cognition consist of symbols manipulation, where symbols get meaning through a correspondence to the real world (or possible worlds ) objectively defined and independent of an y interaction with human beings: incidentally this means that cognition is substantially context-free . Since the approach involves computation, it is also known as computationalism (Licata 2008), (Gär denfors 2000). Objectivism is rooted in the logical positivism wave of analytical philosophy as formalised at the beginning of the 20 th century by the Vienna Circle. The assumption is that scientific reasoning is based on observational data derived from expe riments: new knowledge is acquired from data through logically valid infe rences. Only hypothesis grounded in first-order logic with model-theoretic interpretations – or som e equivalent formalism – have a scientific validity. Objectivism is also reductionist since it assumes that a system can be totally analysed and defined in terms of its components, in a kind o f divide et impera process. The main fallacy of the reductionist hypothesis is to give for granted the reversibility of the process, but the hypothesis is not necessarily constructionist : “… the more the elementary partic le physicists tell us about nature of the fundamental laws, the less rele vance they seem to have [...] to the rest of science. [...] The behaviour of large and co mplex aggregates of elementary particles [...] is not to be understood in Geospatial semantics: beyond ontologies, towards an enactive appr oach 5 terms of a simple extrapolation of the properties of few particles. [...] at each level of complexity entirely new properties appear ...” (Anderson 1972, 393). We may summarise the fundamental views of objectivism as follows (Lakoff 1987): • the mind can be considered as a computer (a Turing machine); • sy mbols get their meaning in relation to “things” in the real world, thus they are internal representations of an external objective reality independent of human being; • thought is context-free and disembodied; • categorisatio n is the way we make sense of experience and categories are defined via sharing necessary and sufficient memberships properties; • category symbols are grounded (get their m eaning) in categories existing in the world independent of hu man being. Dreyfus, especially with his often cited book What Computers Can't Do: A Critique of Artificial Reason , has strongly criticised computationalism; his criticism is inspired by the Heideggerian criticism to the reductionist position, rooted in the Cartesian philosoph y, of computationalism. According to Heidegger cognition is the result of our experience in being-in-the-world , and is grounded i n our disposition to react in a flexible way as required by a specific context (Froese 2007). Guided by this idea, Dreyfus (2007) claims that the representation of significance and relevance is the main problem of computationalism: assuming that a computer stores facts about the world, how can it manage to know which facts are relevant in any given situation?. A version of the relevance problem is the well known frame pr oblem (Dennet 1984), (McCarthy 1969), i.e. the problem for a computer, running a representation of the world, in ma naging world changes: which changes are relevant for the new situation? Which have to be retained, since relevant? How to determine what is relevant and wha t is not? The frame problem may be considered as a manifestation of sy mptoms of the symbol grounding problem (Harnad 19 90). Harnad questions the claim of computationalism that semantic interpretation of a formal symbolic system is intrinsic to the sy stem itself; instead, he clai ms that meaningless symbols, manipulated on the basis of their shape, are grounded in anything but other meaningless symbols. Imagine we have a calculator and type 2+2=4 : it is undoubted that this makes sense, but it makes sense in our head and not in the calculator. Furthermore, a symbolic system is vul nerable to the Searle's Ch inese Room argument (Harnad 1994), discussed in (Searle 1980) as a criticism 6 Pasqual e Di Donat o to the computationalism position accord ing to which an appropriately programmed computer is a mind and has cognitive states. Another questioned point is related to the traditional view of categories: on the objectivist view a category is defined as a set of “things” that share certain properties, which are necessary and sufficient conditions for defining the category itself. Implications of this position are: (i) members of a category are all equivalent, there is not a better example; (ii) categories are independent of any peculiarity of people doing the categorisation; (iii) new categories ar e derived through the composition of existing categories on a set-theoretic base. According to Rosch (1978) most categories do not have clear-cut boundaries and exhibit a prototype effect, i.e. some members are better examples of a category than others (“apple” is a better example of the category “fruit” than “fig”). Rosch refers to a perceived world, rather than a metaphysical one without a knower: an object is put in a category on the basis of a similarity judgement to the category prototype as perceived by a knower. This means that human capacities to perceive, to form mental images, to organise “things” play their role in categorisation; categories are culture-dependent conventions shared by a given group of people, and become accepted at global level th rough a communicative process (Licata 2008), (Gärdenfors 2000). As a first alternative approach to co mputationalism, connectionism is based on the assumption that we need to simulate the brain structure and functioning in order to understand c ognition. Connectionism claims that cognition is the result of the emergence of global states from a network of simple connected components: the focus is primarily on learning, rather than on problem solving. Several authors have debated if this is actually a new paradigm or rather a different approach to the implementation of classical systems. What differentiates connectionism fro m computationalism is basically the functional architecture of the computation (parallel vs. serial), a nd the nature of the representation (sub- symbolic vs. symbolic); for both connectionism and computationalism cognition is basically a form of information processing. Both approaches lack of embodiment and autonomy (Searle 1980). 2.2 Experiential realism Different studies in anthropology, linguistics, psychology show results in conflict with the objectivist view of the mind; the evidence suggests a different view of human cognition, whose characteristics are briefly summarized: Geospatial semantics: beyond ontologies, towards an enactive appr oach 7 • Mind is embodied, meaning that our cognitive system is determined by our body. Our thoughts, ideas, concepts, and other aspects of our mind are shaped by our body, by our perceptual system, by our activity and experience in the world (Lakoff 1987); • Thought is imaginative, and employ s metaphors, metonymies and image schemas. This imaginative capacity is embodied since metaphors, metonymies and images are often based on bodily experience (Lakoff 1987); • Thought has gestalt properties, i.e. the way our brain ope rates is holistic, parallel, and analogue with self-organ ising tendencies, thus it is not atomistic (Lakoff 1987); • Cognitive models are mainly image- schematic and not propositional. Metaphors and metonymies, which are considered exceptional features on the traditional view, play a foundational role since they are transformations of image-schemas (Gärdenfors 2000); • Concepts show prototype effects (Gärdenfors 2000); • Meaning is a conceptual structure in a cognitive sy stem (Gärdenfors 2000). Experiential realism claims that cognitive processes emerge from real- time, goal-directed interactions betw een agents and their environment. Experiential realism and the embodied hypothesis root their basis in the Heideggerian philosophy. In Being and Tim e (Heidegger 1927) He idegger claims that the world is experienced as a significant whole and cognition is grounded in our skilful di sposition to respond in a fl exible and appropriate way to the significance of the contextual situation. The world is made up of possibilities for action that require appropriate responses. Things in the world are not experienced in terms of entities with functional characteristics; rather our experience when pressed into possibilities deals directly responding to a “what-for”: thus, a hammer is “for” hammering and our action of hammering discovers the readiness-to- hand of the hammer itself. The readiness-to-hand is not a fixed functionality or characteristic encountered in a specific situation; rather it is experienced as a solicitation that requires a flexible response to the significance of a specific context (Drey fus 2007). This is similar to what Gibson calls “affordances” (Gibson 1977). Affordance is what matter when we are confronting an environment; we experience entities of the environment that solicit us to act in a sk ilful way, rather than their physical features, which do not influence our action directly. The notion of “image sche mas” has been jointly introduced by Lakoff and Johnson as a fundamental pillar of experiential realism (Hampe 2005). Image schemas are recurring and dynamic patterns of out perceptual 8 Pasqual e Di Donat o interactions with the world that give coherence to our experience; they are pre-conceptual structures directly meaningful since they are grounded in our bodily experience, and have an in herent spatial structure constructed from basic topological and geometrical structures, i.e. “ container ”, “ source-path-goal ”, “ link ” (Gärdenfors 2000). Metaphors and m etonymies are considered as cognitive operations that transform image schemas from a source to a target mental space. Fauconnier and Turner call this process “conceptual blending” (Fauconnier 1994). Since image schemas ar e conceptual structures, they pertain to a particular individual: the question is how individual mental spaces become shared conventions (Gärdenfors 2006).Recent studies demonstrate that conventions emerge out of a communicative process between agents (Steels 2006), (Loula 2005), (Puglisi 2008). Puglisi (2008) shows how a common language may emerge as a result of a communicative dialogue. A simulation with an assembly of agents demonstrates that a simple negotiati on scheme, based on game theory, may guarantee the emergence of a self-organised communication sy stem capable of discriminating and categorising objects in the world with few linguistic categories: in the simulation individual agents are endowed with the ability to form perceptual categories, while interaction/communication among agents produces the emergence and alignment of the linguistic categories. As Gärdenfors (2006) puts it, semantic s, thus meaning, is a “meeting of mind” where a communicative interac tion enables to reach a semantic equilibrium. 2.2 Enactivism Several authors (Dreyfus 2007), (Di Paolo 2003, 2007), (Froese 2007), (De Jaegher 2007) argue that paradoxically th is meeting of mind is what is actually missing in empirical and theoretical investigation of the embodied hypothesis, where the focus is ra ther on agent’s individual cognitive mechanisms as a form of closed sensorimotor feedback loops. Enactivism is an attempt to move embodied practices beyond their current focus (Froese 2007). Enactivis m is a term used by Maturana, Varela, Thomson, and Rosch to name their theories and is closely related to experiential realism; it is not a radi cally new idea, rather it is a synthesis of different ideas and approaches (Maturana 1980). Enactivism is characterised by five main ideas (Di Paolo 2007): • Autonomy: cognising organisms are autonomous b y virtue of their self- generated identity. A system whose identity is specified by a designer Geospatial semantics: beyond ontologies, towards an enactive appr oach 9 cannot exhibit autonomy since it can only “obe y” to rules imposed in the design. Autonomous agency emph asises the role of the cogniser in determining the rules of the “game” being played; • Sense-making: in our being-in-the-world we try to preserve our self- generated identity through inter actions and exchanges with the environment which are significant for us; we actively participate in the creation of meaning via our action, rather than passively receive stimulus from the environment and create internal “images” of it. De Jaegher (2007) further extends the notion of sense-making into the realm of social cognition, where the active coupling of an agent is with another agent. This is what Barsalou (2003) calls social-embodiment ; • Emergence: the notions of autonom y and sense-making invoke emergence. Autonomy is not a property of something, but the result of a new identity that emerges out of dynamical processes. Emergence means the formation of a new property or process with its own autonomous identity out of the interaction of existing processes; • Embodiment: for enactivism cognition is embodied action, temporally and spatially embedded. Reasoning, problem solving, and mental images manipulation depend on bodily structures • Experience: experience is a skilful aspect of embodied activity. As we progress from beginners to experts our performance improves, but experience also changes. The following table is a short synopsis of the different approaches to cognition: Computationalism Connectionism Embodiment/ Enactivism Metaphor for the mind Mind as com puter (Turing machine) Mind as pa rallel distribute d network Mind inseparable from experie nce and world Metaphor for cognition Rule-based manipulat ion of symbols Emergence of global states in a network of sim ple component s Ongoing interaction with the world a nd with other agents The world Separate and objective. Re-presentable via symbols Separate and objective Re-presentable via patterns on network activa tion Engaged Presentable through action Mind/ body Separable Separable Inseparable Table 1: synopsis of diffe rent approaches to cognition 10 Pasqual e Di Do nato 3 Dealing with semantics in information systems In the domain of distributed informati on systems the role of semantics for the automatic/semi-automatic exploitation of distributed resources is particularly emphasised. This process requires semantic interoperability, i.e. the capability of an information sy stem to understand the semantic of a user request against that of an information source and mediate among them (Sheth 1999). But how can semantics be specified? Uschold (2003) proposes the following classification: (i) implicit semantics; (ii) informally expressed semantics; (iii) formally expressed se mantics for human consumption; (iv) formally expressed semantics for machine processing. In (Sheth 2005) the following classification is proposed: • Implicit semantics: is the semantic implicit in patterns in data and not explicitly represented, i.e. co-occurrence of documents or terms in clusters, hyperlinked documents; • Formal se mantics in the form of ontologies: in order to be machine readable and processable, semantics need to be represented in some sort of formalism. Formal languages are based on the n otion of Model and Model Theory: expressions in the lang uage are interpreted in models assumed to be representations of the world or of possible worlds. Description logics (DLs) is the current dominant form alism: based on sets theory, it has the drawback of being not able to represent graded concept membership and uncertainty; • Powerful (soft) semantics: i mplies the use of fuzzy or probabilistic mechanisms to overcome the rigid interpretat ions of set-based mechanisms, and enables to represent degrees of memberships and certainty. In an environment of distributed, heterogeneous data sources and information systems, semantic inter operability refers to the ability of systems to exchange data and soft ware functionalities in a meaningful way; semantic heterogeneity, i.e. na ming and conceptual conflicts arises when there is no agreement on the meaning of the same data and/or software functionality. Explicit and formal semantic is seen as a solution to the problem, and this has motivated several authors to apply formal ontologies (Guarino 1998). Current practices, therefore, rely on ontologies creation and automated resources annotation, coupled with appropriate computational approaches, such as reasoning and query processing, for concept matching among different ontologie s and agai nst user queries (Sheth 2005). Geospatial semantics: beyond ontologies, towards an en active approach 11 Notwithstanding ontologies are seen as a solution to semantic heterogeneity, the irony is that a clear understanding of ontology itself is far to be achieved (Agarwal 2005), and this unders tanding varies across disciplines. As philosophical discipline Ontology (with a capital “o”) is the study of the “ being qua being ” (Guarino 1995), i.e. the explanation of t he reality via concepts, relations, and rules; th e term ontology (with a lowercase “o”) in the philosophical sense refers to a specific system of categories accounting for a specific idea of the wo rld (e.g. Aristotle’s ontology); in computer science an ontology refers to “An engineering artifact, constituted by a specific vocabulary used to describe a certain reality, plus a set of explicit assumptions regarding the intended meaning of the vocabulary words” (Guarino 1998, 4). Ontologies enable to capture in an explicit and formal w ay the semantics of information sources. In a distributed environment such as the Web, resources are distributed and there is often the need to integrate different information in order to satisfy a user request. Semantic integration of information relies, cu rrently, on ontologies integration: the process requires the identification of concepts similarity between ontologies and against user requests. There are different ways of employing ontologies for information integration; Wache (2001) identifi es three framework architectures: • Single ontology approach: all information sources are related to one common global (top-level) ontology. The approach assumes that all information sources have nearly the same conceptualisation; • Multiple ontologies approach: each i nformation source has its own ontology. Since it cannot be assumed that these ontologies share the same conceptualisation, the lack of a co mmon vocabulary makes the integration process difficult. In this case an inter-ontology mapping is required: this mapping tries to identif y semantically correlated terms via semantic similarity measurement; • Hy brid approach; hybrid approaches try to overcome the drawbacks of the two previous approaches. Each information source is described via its own ontology, but all source ontologies are built from the same global (top-level) ontology. Semantic similarity measurements play a crucial role in this process, since they provide mechanisms for comparing concepts from different ontologies, and are, thus, the basis of semantic interoperability (a survey of different approaches to measuring sim ilarity is provided in (Goldstone 2005)). 12 Pasqual e Di Do nato Since different ontologies may commit to different conceptualizations and may be implemented with differe nt formalisms, inconsistency is inevitable. Ontology mismatches may be classified in two broad categories (Visser 1997): • Conceptualisation mismatches: are inconsistencies between conceptualisations of a domain, which may differ in terms of ontological concepts or in the way these concepts are rel ated (i.e. in a subsumption hierarchy); • Explication mismatches: are related to the way the conceptualisation is specified, and occur when two ontologies have different definitions, but their terms, and concepts are the same. The information integration process would be straightforward, at a certain degree, if all ontologies would be similar in terms of vocabulary, intended meaning, background assumptions , and logical formalism, but in a distributed environment this situation is hard, if not impossible, to achieve since different users have different preferences and assumptions tailored to their specific requirements in specific domains. Notwithstanding the efforts to establish standards for ontology languages and basic top-level ontologies, there are still different approaches and heterogeneity between ontologies is inevitable (Krotzsch 2005). Several authors (Goguen 2004, 2005, 2005a), (Krotzsch 2005), (Zimmermann 2006), are investigating the application of categor y theory to this issue. Goguen (2005) presents a theory of concepts (Unified Concept Theory – UCT) that integrates different approaches – Lattice of Theories (LOT), Formal Concept Analys is (FCA), Information Flow (IF), Conceptual Spaces, Conceptual Integr ation (Blending), Ontologies – while preserving their underlying conceptualizations. UCT approach to semantic integration uses category theory tools to unify all these approaches and generalises them to arbitrary logics based on the theory of institutions (Goguen 2004a). Ontology-based approach to semantics is receiv ing also more foundational criticisms. Gärdenfors ( 2004) advocates, for example, that this approach is not very semantic; at the best it is ontological. Since it continues to build entirely on the symbolic methodology, it suffers from the symbol grounding problem; the question is how expressions in ontology languages may get any meaning beyond the formal language itself: ontologies are not grounded. Another difficulty with ontologies is that they are decontextualised and decontextualise experience; instead we claim that semantics reconciliation depends on contextual human sense-making . Ontologies are forms of a Geospatial semantics: beyond ontologies, towards an en active approach 13 priori agreements on a shared conceptualisation of a domain of interest, but meaning is an emergent feature arising context-dependently in action and acquired via participatory sense-making of socially coupled agents, rather than defined as symbolic rules (Di Paolo 2007), (De Jaegher 2007), (Flender 2008), (Dourish 2004). Therefore, the use of ontologies is insufficient in a dynamic situation, where all interpretations may not be anticipated and on-the-fly integration may be needed (Ouksel 2003). Several authors have proposed em ergent sem antics (Aberer 2004) as a solution; emergent semantics envisi ons a community of self-organising, autonomous agents interacting in a dynamic environment and negotiating meaning as required: this means that meaning emerges in context, but context itself is an emergent property, a feature of interaction of a community of practice (Dourish 2004).Emergent semantics is dynamic and self-referential, as a result of a self-organisation process; this requires some autonomous behaviour (cf. 2.2). Collaborative tagging is a new paradigm of the web, where users are enabled to manage, share and brow se collection of resources and to describe them with semantically meaningful freely chosen key words (tags). These tags cannot even be consid ered as vocabularies, since there is no fixed set of tags nor explicit agreement on their use. Nevertheless, this set of unstructured, not explicitly coor dinated tags evolves and leads to the emergence of a loose categorisation sy stem (folksonomy) shared and used by a community of practice (Cattuto 2007). Collaborative tagging falls within th e scope of semiotic dynamics, i.e. the study of how a population of agents establish a shared semiotic system. Semiotic dynamics has been defined as “[...] the process whereby groups of people or artificial agents collectively invent and negotiate shared semiotic dynamics, which they u se for communication or information integration” (Steels 2006, 32). Semiotic dynamics builds on diff erent AI techniques, borrowing also ideas from th e embodied hypothesis: the focus, however, is on social, collective, dynamic sense-making. Computer simulations (Loula 2005) , (Baronchelli 2006), (Puglisi 2008) have demonstrated that a population of embodied agents can self-organise a semiotic system. 4 Semantics and geographic information Traditionally GIScience has relied on an objectivistic approach to knowledge creation (Schuurman 2006): this view assumes that GISystems 14 Pasqual e Di Do nato represent the “real world” as it is independent of human cognition and perception of it (Schuurman 2002). The focus, therefore, has been ontological in nature more than epistemological; in determining geo- spatial ontologies, the question of “wha t exists” has gained much attention versus the question of “how what exists is identified and defined” (Agarwal 2005). Starting from early 1990s, however, several researchers have focused their attention on the epistemological aspects of GIS, accounting for human cognition and perception. These authors borrow ideas from the experiential realism and the epistemo logical model introduced by (Lakoff 1987), arguing that cognition structures the perception and representation of reality: their work builds on image schemata, conceptual blending, conceptual spaces, and affordances. Semantic issues have always been a key concern in GIS, since semantic interoperability plays a crucial role for the sharing and integration of geographic information (Harvey 1999). The use of ontologies is the most applied means to support semantic inte roperabilit y, and ontology has been recognized as a major research theme in GIScience (Mark 2000) It is possible to individuate two main approaches to ontology in GIS: • Philosophical approach: deals w ith top-level ontologies for the geographic domain, and takes an objectivistic view, i.e. reality as objectively existent independent of human cognition and perception. Works on this approach are, for example, (Mark 1999, 20 01), (Smith 2001, 2004), (Frank 2001, 200 7), (Kuhn 2003, 2005),(Galton 2003). Some of these authors highlight some issues, e.g. vagueness as well as cultural and subjective discrepancies th at are diffic ult to solve: Mark (2003) has shown that people from different places and cultures use different categories for geographic features; • Knowledge engineering approach: deals with ontologies as application- specific and purpose-driven engineering artifacts. Works on this approach are, for example, from (Kuhn 2001), (Camara 2000), (Hakimpour 2001), (Fonseca 2002), (Bernard 2003), (Klien 2004), (Raubal 2004), (Lutz 2006). Frank (2001) suggests that an ontology for GIS should be b uilt as a coordinated set of tiers of ontol ogies, allowing different ontological approaches to be integrated in a unified system constituted of the following tiers: (i) Tier 0 – human-independent reality; (ii) Tier 1 – observation of the physical world; (iii) Tier 2 – objects with properties; (iv) Tier 3 – social reality; (v) Tier 4 – subjective knowledge Frank’s five-tier architecture has a lot o f commonalities with the four- universes paradigm (Gomes 1998) app lied to geographic information by Geospatial semantics: beyond ontologies, towards an en active approach 15 (Camara 2000a). Inspired by this last work, and based on a realistic view of the world, Fonseca (2002a), introduces a five-universes parad igm: (i) physical universe; (ii) cognitive universe; (iii) logical universe; (iv) representation universe; (v) implementation universe. Representing the reality involves the conceptualisation of elements of the physical world via a collective agreement of a community sharing common perceptions; concepts are defined within a community of experts and are organised in a logical framework (ontologies). Halfway between a philosophical and an engineering approach, Kuhn (2003) proposes the Semantic Reference System as a framework for solving semantic interoperability pr oblems. In analogy with spatial reference systems, semanti c reference systems are co mposed of a semantic reference frame and a semantic datum: as the geometric co mponent of geographic information refers to spatial reference systems, the thematic component refers to semantic refere nce systems. A semantic reference frame acts like a coordinate system, as a framewor k to which terms can refer to get meaning: this reference frame is a formally defined top-level ontology. As a datum in spatial reference sy stems anchors the geometry model to the real world, a semantic datum grounds the terms of semantic reference frame: Kuhn (2005) suggests using image schemas as grounding mechanism. From a knowledge engineering point of view, ontologies have been applied for (i) geographic informatio n discovery and retrieval (Bernard 2003), (Klien 2004), (Lutz 2006); (ii) ge ographic information integration (Sheth 1999), (Visser 2002), (Wache 2001), (Hakimpour 2001); (iii) GISystems (Fonseca 1999, 2002); (iv) modelling user activity (Camara 2000), (Kuhn 2001), (Raubal 2004), (Ti mpf 2002). Application-specific, task-oriented, purpose-driven ontologies are aimed at information systems development: these ontologies emerge from requirements and contain knowledge li m ited to a specific area of application. The main issue with engineering ontologies is grounding: according to the ontology hierarchy proposed in (Guarino 1998), dom ain and task ontologies are grounded in top-level ontologies. The question is how these top-level ontologies are grounded themselves: this infinite regress should end at some point to an ontology, bu t this is objectivism /reductionism and we are not in sympathy with this approach. Borrowing ideas from embodied cognition and cognitive semantics, some authors propose image schemas as grounding mechanism and conceptual spaces as a new representation paradigm. Kuhn (2003), for example, proposes image schemas for grounding Semantic Reference Systems. Image schemas have been introduced in the geospatial domain by 16 Pasqual e Di Do nato (Mark 1989), have received formal spec ifications in (Rodriguez 1997) and (Frank 1999), and have been applied, for exam ple, to way finding in (Raubal 1998). However, the concept of image schemas re main controversial and ambiguous, and disagreements exist in image schemas research between two broadly contrasting approaches: the first approach, located in t he context of cognitive psychology and neurosciences, considers image schemas as expression of universal principles; the second approach, located in the context of anthropol ogy and cognitive-cultural linguistics, has a more relativistic view of image schemas and emphasises that cognition is situated in socio-cultura lly determ ined context (Hampe 2005). Cross-linguistic and cross-cultural studies (Choi 1999) show that image schemas, while operating in many languages, are not universal, instead they are culturally situated (Correa 2005); on the other hand, the tendency to make universalistic statements is based on few languages, above all English (Zlatev 2007). Some major drawbacks exist in the ontology approach to semantics in terms of dynamicity and context. Ontologies are a priori agreements on a shared conceptualisation of a domain of interest, a form of pre-given, decontextualised knowledge; this is problema tic as long as we consider the temporal extent on the know ledge. There is an empirical evidence of the fact that human manipulation of concepts is facilitated by considering them in relevant context (Barsalou 1993). Current approaches to context, also in the GI field (Rogriguez 1999), (Keßler 2007), are representational, i.e. they assume that context is stable, delimited information that can be known and encoded in just another information layer or another ontolog y in an information system . Meaning, instead, is an emergent f eature arising context-dependently in action and acquired via participatory sense-making of socially coupled agents, rather than as symbolic rules, a nd context itself is an emergent feature (Di Paolo 2007), (De Jaegher 2007), (Dourish 2004). 5 Toward a new formalism for geospatial semantics We start from the assumption that meaning and context are dy namically emergent from activity and interaction, determined in the moment and in the doing (Dourish 2004): there is no pre-given knowledge and no fixed properties that can a priori determine what is relevant. Current approaches to semantics, main ly ontology-based, are not able to satisfy and manage this assumption, and a new formalism is required. Geospatial semantics: beyond ontologies, towards an en active approach 17 The State-Context-Property (SCOP) formalism se ems promising in this sense: a detailed description of the fo rmalism can be found in (Aerts 2005, 2005a), (Gabora 2002, 2008), here a brie f description is provided. SCOP is a formalism based on a generalisation of mathematics developed for quantum mechanics that provides a means for dealing with context, concept combination and similarity judg ments, toward a contextual theory of concepts (Gabora 2008). At conceptual level SCOP falls within the enactive approach to cognition: the main ideas behind it may be summarised as follows: • Concepts as participatory (ecological view): concepts and categories are participating parts of the mind-world whole; • Contextuality: context influences the meaning of concepts and needs to be given a place in the description of concepts; • Concept co mbination: concept conjunctions exhibit emergent features that traditional theories (representational), are not able to “predict”; • Si milarity: similarity judgments are context-dependent. The description of a concept in SCOP consists of five elements: • A set of states the concept may assume; • A set of relev ant contexts; • A set of relev ant properties; • A function describing the applicability of certain properties in a specific state and context; • A function describing the probability fo r one state to collapse into an other under the influence of a specific context. For any concept we may have a number (infinite) of “possible” states, and each state is characterised by a unique typicality for instances and properties. A state of a concept that is not influenced by a specific context is said to be an eigenstate 1 for that context, otherwise it is said to be a potential state (superimposition state). A potential state may collapse to another state under the influence of a specific context: for example consider the concept Tree that under the context “ Desert island ” might collapse to state Palm Tree . The properties of a concept are themselves potential: they are not definite except in a specific c ontext. If a concept is in an eigenstate for a context, then the latter just detects what the concept is in acto (Gabora 2008), but if the concept is in a superposition state, then the context change this state: properties of the concepts may change under the influence of the 1 An eigenstate is a state asso ciated with definite properties 18 Pasqual e Di Do nato context from actual to potential and vice versa. Therefore each state of a concept is an eigenstate or a superposition state: if it is an eigenstate the properties are actual, otherwise most of the properties are potential: a context has the “power” to change a superposition state in an eigenstate . The various states of a concept can be described as a Hilbert Space , while the conjunction of two concepts can be modelled as an entanglement 2 through the tensor product of the two Hilbert Spaces describing those concepts. Similarity judgments between two co ncepts are only possible if their respective properties are compatible, i. e. they refer to the same context (refer to Gabora 2002 for a detailed description of context-sensitive measure of conceptual distance). Future work This work has dealt with semantics and how it is addressed in distributed information system and in the GI domain. Drawbacks of current practices, mainly based on ontologies, have been highlighted. Ontology-based approach to seman tics is problematic in ter ms of dynamicity and context, since ontologies, being forms of a priori agreements, are decontextualised. Instead we claim for an enactive a pproach to cognition and semantics where meaning is an emergent feat ure arising context-dependently in action. Since representational theories are unable to deal with these issues, a new formalism is required. The SCOP formalism is considered promising and has been briefly described. Future work will deal with an in depth analysis of the form alism in order to investigate its applicability to GI, possibly toward practical applications. References Aerts, D. Gabora, L. A theo ry of concep ts and their c ombination I; The structure of the sets of contex ts and properties. Kybernetes, 34( 1), 167-191, 2005 Aerts, D. Gabora, L. A theo ry of con cepts and their combination II; A Hilbert space representation. Kybernet es, 34(1), 192-221, 2005a 2 Entanglement means that two states are link ed together and it is not possib le to describe one state without mention of its counterpart Geospatial semantics: beyond ontologies, towards an en active approach 19 Aberer, K. Cudré  -M auroux, P. Ouksel, A.M. (editors), Catarci, T. Hacid, M.S . Illarrame ndi, A. Kashyap, V . Mecella, M. Mena, E. Neuhold, E.J . De Troyer, O. Risse, T. Scanna pieco, M. Saltor, F. de Santis, L. Spacc apietra, S. Staab, S. Studer, R. Em ergent Semant ics Principles and Issues. Da tabase Systems f or Advanced Ap plications, Lec ture Noted i n Comput er Science, 297 3, 25-38, 2004 Agarwal, P. Ontological considerations in GIScience. International Jo urnal of Geographical Infor mation Science, 19, 501–536, 2005 Anderson P.W. More Is Different. Scien ce, 177(4047), 393-396 , 1972 Anderson, M.L. Embodie d Cognition: a fiel d guide. Artificial Intelligence, 149, 91-130, 2003 Baronchelli, A. Felici, M. Caglioti, E. Loreto, V. Steels, L. Sharp transition towards shared voc abularies in multi-ag ent syste ms. Journal of Statistical Mechanics: Theory and E xperiments, 2006. Barsalou, L.W. Niedenthal, P.M. Barbey, A. Ruppert, J. Social embodim ent. In Ross, B. (Ed.): The Psychology of Learn ing and Motivation, 43, 43-92, 2003 Bernard, L. Einspanier, U ., Haubrock, S., Hübner, S. Kuhn, W., Lessing, R., Lutz, M. & Visser, U. Ontologies for In telligen t Search and Semantic Translation in Spatial Data Infrastru ctures. Photogrammetrie - Fernerkund ung – Geoinform ation, 6, 45 1-462, 200 3 Bishr, Y. Overcoming the semantic and oth er barriers in GIS interoperability. International Jo urnal of Geograph ical Information Science, 12(4) , 299-314, 1998 Camara, G. M onteiro, A.M.V. Paiva, J.A. Gom es, J. Velho, L. Towards a unified framework for geographical data m odels. In Figueired o, L. (Ed): Procee dings of Geolnfo 2000 Workshop Brasileiro de Geoinformatica, 37 -44, 2000 Camara, G. Monteiro, A.M.V. Paiva, J.A . Souza, R.C.M. Action-Driven Ontologies of the Geographic Space: Beyond the Field-Object Debate. GIScience 2000: The Fir st Interna tional Conference on Geog raphic Information Sc ience, 2000a Cattuto, C. Loreto, V: Pietronero, L. Semiotic d ynamics and Collaborative Tag- ging. PNAS, 104(5), 1461-1464, 2007 Choi, S. McDonough, L. Bowerman, M. Mandler, J. Early sensitivity to language- specific spatial terms in English and Korean. Cognitive Development, 14, 241-268, 1999 Correa-Beningfield, M. Kris tiansen, G. Nava rro-Ferna n do, I. Candeloise, C. Image schemas vs “Com plex Promitives” in cross-c ultural spatial cognition. In Dirven, R. Langacker, R. Tay lor, J. (Ed s.): From perc eption to m eaning: image schemas in cognitive linguistics. Cognitive Linguis tic Research, 29, 343-366, 2005 De Jaegher, H. Di Paolo, E. Participatory s ense-making: An enactive appr oach to social cognition. Ph enomenology and the Cognitiv e Sciences, 6(4), 485-507, 2007 Dennet, D.C. Cognitive Wheels: The Frame Problem of Artificial Intelligence. In Hookaway, C. (Ed.): Minds, Machi nes and Evoluti on, Cambridge U niversity. Reprinted in Pylyshyn (1987). On-line 20 Pasqual e Di Do nato Di Paolo, E. A. Organi smically -inspired ro botics: hom eostatic ada ptation a nd teleology bey ond the closed sens orimot or loop. In Muras e, K. Asakura, T. (Eds.): Dynamical Systems Approach to Embodiment and Sociality, 2003, 19- 42 Di Paolo, E. R ohde, M. De J aegher, H. Horizons for the enactive mind: Values, social interacti on and play. I n Stewart, J. Gape nne, O. Di Paol o, E. (Eds): Enaction: Towards a New Paradigm fo r Cognitive Science. MA: MIT Press. On-line Dourish, P. W hat We Talk A bout When We Talk About Context. Personal and Ubiquito us Com puting, 8(1), 19 -30, 200 4 Dreyfus, H.L. Why Heideggeri an AI faile d and how fixi ng it woul d require making it more Heideggerian. Artificial Intellig ence, 171(18), 1137-1160, 2007 Einspanier, U. Lutz, M. Senkl er, K. Sim o nis, I. Sliwinski, A. Towards a process model for GI service composition. On-lin e http://ifgiweb.un i- muenster.de/~lutzm/einspan ier_et_al_gi-tage_ 2003.pdf Fauconnier, G. Turner, M. Conceptual Projection an d Middle Spaces. Report 9401 Department of Cognitive Science, University of San Diego, 1994 Flender, C. Kitto, K. Bruza, P. Beyond Ontology in Information Systems. QUT Technical report, FIT-TR-2008 -05. 2008. Fonseca, F.T. Egenhofer, M. J. Ontology-driven ge ograph ic information systems. In Medeir os, C.B. (Ed. ): 7 th ACM Sym posium on A dvences in Ge ographic Information Sy stems, 14-19 , 1999 Fonseca, F.T. Egenhofer, M. J. Agouris, P. Camara, G. Using ontologies for integrated ge ographic i nformation sy stems. Transactions i n GIS, 6, 231- 257, 2002 Fonseca, F.T. Egenhofer, M.J. Sem antic granularity in ontolo gy-driven geogra phic information systems. Annals of Mathematics and Artificial Intelligence – Special issue on Spatial and Temporal granularity, 36 (1/2), 121-151, 2002a Frank A.U. Raubal, M . Form al specification of im age schemata – a step towards interoperability in geographic inform ation systems. Spatial Cognition and Computat ion, 1, 67-101 , 1999 Frank A.U. Tiers on ontology and consistency co nstraints in ge ographical information systems. International Journal of Geographical Info rmation Science, 15(7), 667-678, 200 1 Frank A.U. Toward a mathematical theo ry for shapsho t and tempora l formal ontologies. In Fabrikant, S.I. W achowicz, M. (Eds.): The European Information Society. Leading the way w ith Geo-Information, Lecture Notes in Geoinform ation a nd Carto graphy, 317-334 , 2007 Froese, T. On the Role of AI in the O ngoing Paradigm Shift within the Cogn itive Sciences. In Lungarel la, M. Iida, F. Bonga rd, J. Pfeifer, R. (Eds.): 50 Years of Artificial Intelligence, Essays Dedicated to the 50th Anniversary of Artificial Intelligence. Lecture Notes in Co mputer Science, 4850, 63-75, 2007 Galton, A. Desiderata for a Spatio-temporal Geo-ontology. In Kuhn, W. Worboys, M.F. Timpf, S. (Eds.): COSIT2003, L ect ure Notes in Computer Science, 2825, 1-12, 2003 Geospatial semantics: beyond ontologies, towards an en active approach 21 Gabora, L. Aerts, D. Contextualiz ing concepts using a mathematical generalization of the quan tum form alism. Journal of Experim ental and Theoretical Artificial Inte lligence, 14 (4), 327-358 Gabora, L. Rosch, E. Aerts, D. Towar d an Ecologica l Theory of Conce pts. Ecological Psy chology, 2 0(1), 84- 116 Gärdenfors, P. Cognitive Scien ce: from computers to anthills as models of hu man thought. 1999 On-line http://etjan st.hb.se/bhs/ith//2-99/pg .htm Gärdenfors, P. Conceptual Spaces: The Ge ometry of Thought. The MIT Press, 2000 Gärdenfors, P. How to make the Sem antic Web more s emantic. In Vieu, A.C. Varzi, L. (Eds.): Form al Ontology in Inform ation Systems, IOS Press, 19-3 6, 2004 Gärdenfors, P. Warglien, M. C ooperation, Conce ptual Spaces and the E volution of Semantic s. In Vogt, P. Su gita, Y. Tuci, E. Nehaniv, C. (Eds.): Symbol Grounding and Beyo nd: Proceedings of the Third Internationa l Workshop on the Emergence and Evol ution of Lingui stic Comm unication, Lecture Note s in Computer Science, 4211, 16-30, 2006 Gibson, J.J. The theory of affordances. In Shaw, R . Bransford, J. (E ds.): Perceiving, Acting and Knowing. Hillsd ale, NJ, Erlbaum, 127-143, 1977 Goguen, J. Dat a, schema and ontol ogy integrat ion. Procee dings of Works hop on Combinat ion of Logics, 21 -31, 2004 Goguen, J. Information integration in in stitutions. On-line, 2004a http://www.cs.ucsd .edu/~goguen/pps/ifi0 4.pdf Goguen, J. What is a Concept? In Dau, F . M u g n i e r , M . L . S t u m m e G . ( E d s . ) : ICCS2005, Lecture Notes in Computer Scien ce, 3596, 52-77, 2005 Goguen, J. Ontology, Ontotheology, a nd Society. On-line, 2005a http://www- cse.ucsd.edu/~gogu en/pps/onto5.pdf Goldstone, R.L. Son, J. Y. Similarity. In Holyoak, K .J. Morrison R.G. The Cambridge Handbook of Thin king and R easoning, Ca mbridge U niversity Press, 13-36, 200 5 Gomes, J. Velho, L. A bstraction Para digms for Com puter Grap hics. The Visual Computer, 11 (5), 227-2 39, 1998 Guarino, N. Formal Ontology, Conceptual Analysis and Knowledge Representati on. Internatio nal Journal of Human an d Comput er Studies, 43(5/6), 625- 640, 1995 Guarino, N. Formal Ontology in I nformation Systems. In Guarino, N. (Ed.): Formal Ontology in Inform ation Systems. Proceedi ngs of FOIS'98, IOS Press, Amsterdam , 3-15, 19 98 Hakimpour, F. Geppert, A. Resolving Semantic Heterogeneity in Schema Integration: an Ontology Based Approach . In Proceedi ngs of the international conference on Formal Onto logy in Informati on System s (FOIS’01), 2 97-308, 2001 Hampe, B. Image Schemas in cognitive linguistics: Introd uction. In Dirven, R. Langacker, R. Taylor, J. (Eds .): From perception to m eaning: image schemas in cognitive linguistics. Cognitive Linguistic Research, 29, 1-12, 2005 Harnad, S. T he Sym bol Groundi ng Proble m. Physica D , 42, 335- 346, 199 0 22 Pasqual e Di Do nato Harnad, S. Computation Is Just Interpretable Symbol Man ipulation: Cognition Isn't. Minds and Machin es, 4, 379-390, 1994 Harvey, F. Kuhn, W. Pundt, H. Bishr, Y. Riedmann, C. Semantic Interoperability: a central issue for sharin g geogra phic info rmation. The Annals of Regional Science, 33, 213-232, 1999 Heidegger, M. Essere e Tempo. Longanesi , 2005. Itali an translation of Sein und Zeit, 1927 Keßler, C. Raubal, M. Janowicz, K. The Ef f ect of Context on Semant ic Simila rity Measurement. In Meersman, R. Ta ri, Z. Herrero, P. et al. (Eds .): On The Move - OTM 2007 Workshops, Part II. Le cture Notes in Com puter Science , 4806, 1274-128 4, 2007 Klien, E. Eins panier, U. Lutz , M. Hubner, S. An arc hitecture for ont ology-based discovery and retr ieval of georgaph ic information. In 7 th Confer ence on Geographic Information Science AG ILE2004, 179-188, 2004 Krötzsch, M. Hitzler, P. Ehrig, M. Su re, Y. Category Theory in Ontology Research: Concrete Gain from an Ab stract Approach. Technical Report, AIFB, University of Karlsruhe, 2005. On -line http://www.aifb.uni- karlsruhe.d e/WBS/phi/pub/pos_catmerg e.pdf Kuhn, W. Ontologies in su pport of activities in geogra phical space. International Journal of Geogr aphic Information Science, 15, 613-633 , 2001 Kuhn, W. Sem antic Reference System . In ternational Journal of Geograph ical Informat ion Science, 17, 40 5-409, 2003 Kuhn, W. Geospatial Semantics: Wh y, of What, and How? In Spaccap ietra, S. Zimányi, E. (Eds.): Journal on Data Semantics III. Lecture Notes in Compu ter Science, 3534 (3) , 1-24, 200 5 Lakoff, G. Wo men, Fire, and Dangerous T hings. The U niversity of C hicago Pres s, 1987 Licata, I. La logica aperta della m ente. Codice Edizioni, 2008 Loula, A. Gudwin, R. El-Han i, N.C. Queiroz, J. The emergence of symbol-based communicat ion in a complex sy stem of artificial creatures. Procee dings 5th International Conferen ce Integration of Knowledge Inten sive Multi-Agent Systems KIMAS'05: Modeling, Evolu tion and Engineering, 279 - 284, 2005 Lutz, M. Klien, E. Ontology-based retrie val of geographi c informat ion. Internatio- nal Journ al of Geograph ical Information Science, 20(3), 233-26 0, 2006 Maturana, H.R. Varela, F.J. Autopoiesi e cognizione. La realizzazione del vivente. Marsilio, 2001. Italian transalation of Au topoiesis and Cognition, 1980 Mark, D.M. Cognitive Image-Schemata for Geographic Inform ation: relations to user views and GIS interfaces. Procee di ngs of GIS/LIS’89 , 2, 1989, 551-560 Mark, D.M. Sm ith, B. Tversky , B. Ontolo gy and Geog raphic Obje cts: An empirical study of cognitive categorization . In Freksa, C. Mark, D.M. (Eds): Proceedings of Conference on Spatial Information Theory: A theoretical basis for GIS, COSIT1999, Lectur e Notes in Computer Science, 1661, 283-2 98, 1999 Mark, D.M. Ege nhofer, M. Hirtle, S. S mith, B. UCGIS Emerging Research Theme: Ontological Foundations for Ge ographic In formati on Science. 2000. On-line Geospatial semantics: beyond ontologies, towards an en active approach 23 http://www.ucgis.o rg/prior ities/research/research_wh ite/2000%20Pap ers/emer ging/ontolog y_new.pdf Mark, D.M. Skupin, A. Smitch, B. Featur es , objects and other t hings: ontol ogical distinctions in the geo graphic dom ain. In Montello, D.R. (Ed.): Procee dings of Conference on Spatia l Inform ation Theory : Foundati ons on Ge pgraphic Information Science, COSIT’01, 488-502, 2001 Mark, D.M. Turk, A. Landscape categories in Yindjibarndi: On tology, environm ent, and langua ge. In Ku hn, W. Worboys, M .F. Timpf, S . (Eds.): COSIT 2003, Lecture No tes in Computer Scien ce, 2825, 28-45, 20 03 McCarthy, J. Hayes, P. Som e philosophical problems from the standpoi nt of artificial intelligence. In Meltzer, B. Michie, D (Eds.): Machine Intellig ence 4. On-line http://www-formal.stanford.edu /jmc/mcchay69.pdf Ouksel, A.M. In-context peer-to-p eer in formation filtering on the web. SIGMOD Records, 32(3), 65-70 , 2003 Puglisi, A. Baronchelli, A. Loreto, V. Cultural route to the emergence of linguistic categories. PNAS, 105(23 ), 7936-7940, 2008 Raubal, M. Egenhofer, M. Comparing the co mplexity of wayfinding tasks in built environm ents. Environm ent & Planni ng B, 25( 6), 895-9 13, 1998 Raubal, M. Formalizing Conceptual Spaces. In Varzi, A Vieu, L. (Eds.): Formal Ontology in Information Systems, Proceedings of th e Third International Conference (FOIS 2004). Frontiers in Ar tificial Intelligence and Application s, 114, 53-164, 2004 Rodriguez, A. Eg enhofer, M. Image-Schemata-Based Spatial Inferences: The Container-Surface Algebra. Proceedings of the Inte rnational Conference on Spatial Information Theory: A Theo retical Basis for GIS. Lecture Notes in Computer Science, 1329, 35-52, 1997 Rodriguez, A. Egenhofer, M. Rugg, R. Assessing Semantic Similarity amont geospatial entity class definitions. In Vc kovski, A. Brassel, K. Schek, H.J . (Eds.): Inter operating Geo graphic In formation Systems IN TEROP99, Lect ure Notes in Computer Science, 1580 , 189-202, 1999 Rosch, E. Principles of categorization. In Rosch, E. Lloyd, B.B. (Eds.): Cognition and categorization. Hillsdale, NJ: Er lbaum. Repr inted in: Margolis, E. Laurence, S. (Eds. ) (1999). Concept s: Core readings. C ambridge, M A: MIT Press. Sheth, A.P. Changing focus Intero perability in Information Systems: Fro m System, Syntax, Structure to Sema ntics. In Goodchil d, M.F. Egen hofer, M.J . Fegeats, R. Kottman, C.A. (Eds.): Interoperating Geog raphic Information Systems. Kluwer, 5-29, 1999 Sheth, A.P. Ramakrishnan, C. Thomas, C. Semantics for the Semantic Web : The implicit, the Formal and the Powerful. In ternational Journal o n Semantic Web & Informati on Systems, 1(1) , 1-18, 2005 Schuurman, N. Reconciling So cial Constructivism and Realis m in GIS. ACME: An International E-Jo urnal for Critical Geo graphies, 1(1), 75-90, 2002 Schuurman, N. Formalization m atters: cr itical GIScience and ontology research. The Annals of the Association of Am erican Geographers, 96(4), 726 -739, 2006 24 Pasqual e Di Do nato Searle, J.R. M inds, Brains, a nd Program s. Behavioral a nd Brain Scie nces, 3 (3), 417-457, 1980 Smith, B. Mark , D.M. Ge ographic cate gories: an o ntological in vestigati on. International Jo urnal of Geograph ical Information Science, 15(7) , 591-612, 2001 Smith, B. Beyond Concepts: Ontology as Reality Represen tation. In Varzi, A. Vieu, L. (Eds): Proceed ings of FOIS2004. In ternational Confer ence on Formal Ont ology and Inf ormation Syst ems, 2004 Steels, L. Semiotic dynamics for embo died agents. Intellig ent Systems, IEEE, 21(3), 32- 3 8, 2006 Steels, L. Fifty Years of AI: From Symbols t o Embodim ent - and Back. In Lungarella, M. Iida, F. Bongard, J. P feifer, R. (Eds.): 50 Years of Artificial Intelligence, Essays Dedicated to th e 50th Anniversary of Artificial Intelligence. Lecture Notes in Co mputer Science, 4850, 18-28, 2007 Timpf, S. On tologies of W ayfinding: a traveler's pe rspective. Net works and Spatial Economics, 2(1), 9-33, 2002 Uschold, M. Where Are the Semanti cs in the Semanti c Web? AI Magazi ne, 24(3), 25-36, 2003 V i s s e r , P . R . S . J o n e s , D .M . B e n c h - c a p o n , T . J . M . S h a v e , M . J . R . A n A n a l y s i s o f Ontology Mismatches; Heterogeneity ve rsus Intero perability. AAAI 1997 Spring Sym posium on Ontologi cal Engineeri ng, 1997. O n-line Visser, U. St uckenschm idt, H. Schlie der, C. Ont ologies for ge ographic information processing . Computers & Geosciences, 28(1), 103- 117, 2002 Wache, H. V ogele, T. Vi sser, U. Stucke nschmi dt, H. Schuster, G . Neumann, H. Huber, S. Ontol ogy-based Int egration of In formation – A survey of existi ng approaches. Proceedings of the workshop on Ontologies and Informatio n Sharing at the International Joint Conference on Artificial Intelligence (IJCAI), 108–117, 2 001. Zlatev, J. Spatial Semantics. In Cuyckens , H. Geeraerts, D. (Eds): Hand book of Cognitive Linguistics. Ch apter 2 Zimmermann, A. Krotzsch, M. Euzenat, J. Hitzler, P. Formalizing Ontology Alignment and its Operations with Catego ry Theory. In Bennett, B. Fellbaum , C. (Eds): Proceedings of the Fourth Int ernational Conference on Form al Ontology in In formation System s (F OIS 2006), Frontie rs in Artificial Intelligence and Applications, 150, 2 77-288, November 2006