Getting across information communities

Usefulness of non-textual information in the Web, such as photos or raw data, relies on textual descriptions. Traditional keyword-based search engines use these annotations for the indexing process. Annotations also enable the content’s evaluation regarding the searching user’s intended use. The descriptions should be flexible enough to take varying needs of different information communities into account. But they also have to be explicit enough to ensure correct and unambiguous interpretation. Using the example of data served by an OGC WFS, this section illustrates how Semantic Web techniques prove to be a suitable approach to meet both these requirements.

OGC standards¹ specify how to access and process spatial data served by OGC Web services. Functional parameters such as the operations and parameters are well-defined. With just the URL of the service as information, desktop clients automatically request service metadata to retrieve an interface description. The format of the provided data, like the Geography Markup Language (GML) for Web Feature Services, is predetermined. Once retrieved, the data can be directly processed and visualized without the need for understanding the encodings or the data models. Standards enable syntactic interoperability between the various components within an SDI. They do not, however, explain what it represents. A process like an interpolation may be invoked automatically on data. It may accept any point-based spatial data, even if attributes used to compute the interpolated values are missing. The computation will either simply fail. Or even worse, it will produce visually correct, but content wise useless results. Interoperability is not restricted to the syntax, the semantics must be considered as well. This section introduces shared models of domain knowledge as tool to capture the meaning of any GI source, for example raw sensor data from satellites. Spatial data is explicitly grounded in space through coordinates. Free text including implicit place names is considered here as GI source as well, since such georeferenced content may be as relevant as the sensor data. Semantic Annotations establish a link between the GI source and common domain vocabularies well accepted within the particular GI community. Finding ways to mediate between similar or closely related concepts from different domain vocabularies may help to realize an SDI for an Amazon spanning across the many different GI communities.

We have to consider three different types of actors (expert, citizen, and policy-makers) to understand the complex interactions between the different GI communities (Davis et al. 2009), and how Semantic Web technologies can help to facilitate information exchange across them. The expert, maybe scientist or journalist, is responsible for the observation, analysis, and representation of information. Scientists are responsible for observing and analyzing phenomena, e.g. using satellite images to detect recently cleared patches of forests, and to publish this information in form of reports or scientific articles. Journalist can be described in a similar fashion: they research and write articles about current events and publish them, for example, as articles on the Web. Though not always true, the expert has the responsibility to create such information in an objective manner. Target group for published information are the citizens and the policy makers. On the basis of facts within the information, the policy makers are entrusted to make decisions which respect the interests of the majority of the citizens. In the end it is this last group which is affected by decisions, and it mostly comprises the local population with their interest in matters like growing economy or a steady supply of food.

GI communities integrate the various kinds of actors. A journalist specialized in business affairs may write about the current plans of a livestock company to explore new farmland. Scientists study the consequences of economic development on the environment. Their discoveries may then again be used by journalists to illustrate the deforestation trends. In the following examples we focus on three GI communities, which we simply call the economics, ecology and ethnology community. In the case of deforestation, actors of these communities have their own (often conflicting) view on the matter. An economics researcher reporting on the economic development of a region most probably will not use the term “deforestation” as concept to describe his data. But since it is relevant, the environmental policy maker studying destruction of the ecosystems should be able to find it. Getting across GI communities requires techniques like ontologies and rules, which are able to mediate between the different vocabularies.

The usefulness of domain ontologies, and consequently their use for semantic annotations, depends in particular on the used vocabulary. Such vocabularies are controlled by trusted authorities, i.e. international organizations like the United Nations, governmental institutions, or established companies. Examples for controlled vocabularies used by the economics community are the International Standard Industrial Classifica tion of All Economic Activities United Nations Statistics Division (UNSD 2009) or the STW Thesaurus for Economy (German National Library of Economics (ZBW) 2009). On the other side, commonly used thesauri for the environmental domain are the GEMET General Multilingual Environmental Thesaurus or the AGROVOC Thesaurus (Food and Agricultural Organization of the United Nations (FAO) 2009). The last three are of particular interest for us: they already provide Web Service interfaces to retrieve the vocabulary in RDF. STW and GEMET make use of the W3C Standard SKOS (Simple Knowledge Organization) (Miles and Bechhofer 2009; Miles and Pérez-Agüera 2007). AGROVOC is currently implementing an interface to retrieve its content as OWL ontology with a certain number of well-defined relations (Liang et al. 2006).

It is the GI community’s responsibility to create and serve its specific domain ontology. But one GI community alone cannot create the mediation rules mapping towards other domains. Sophisticated knowledge is needed in both, the domain of discourse and logics. Global community-independent knowledge bases like Wikipedia or Freebase⁵ prove that, as long as its usefulness is apparent, experts from varying fields are motivated to collaborate to create high-quality content. Here we have to assume that these mappings exist, and that repositories on the web serve these rules. Figure 7 illustrates how rules can establish a link between the economics and the ethnology community.

A GI source, e.g. a WFS embedded in an SDI, has been semantically annotated with concepts coming from a domain ontology build on top of the STW thesaurus. On the right side, another source has been semantically annotated, this time with concepts from an ethnology thesaurus. The GI source ontology in this case may not only represent spatial data. For semantic data integration, it may also refer to the user’s request.

The Semantic Web is seamlessly embedded in the traditional Web. The same applies for the Geospatial Semantic Web or an SDI integrating varying communities. The proposed SDI is not explained by its Web Services or a catalog which serves as entry point. All spatial content on the Web which yields information about one region or theme of interest is implicitly part of the SDI. The semantic annotation linking between one SDI’s relevant domain vocabulary and the content can establish an explicit relation. After all, it is not one user’s task to identify where to publish data. Once it is somewhere on the Web, crawling algorithms, e.g. deployed by Google, index the data and make them accessible to a broader public. Hence, the semantic-enabled SDI for the Amazon should be defined though its purpose. Its various components including Web Services, RDF repositories for the rules and ontologies, portals for searching content, and more, are simply linked with each other. Figure 7 illustrates the open aspect of the architecture, in this case for the mappings between various domain ontologies.

In “Semantic conflicts within GI source descriptions”, hierarchical knowledge (e.g. too application- or too community-specific knowledge) has been identified as one potential source for semantic conflicts. In the given example, we argued that a given name of the remote sensor of a satellite yields information about the data resolution. This expert knowledge unnecessarily restricts the potential audience of this particular information. Users simply searching for high-resolution data without explicitly specifying the remote sensor will either fail to discover it, or will not be able to evaluate its fitness for use. He might use the following rule to specify a query: Coverage(?x) ∧ hasResolution(?x,high) → sqwrl:select(?x). The built-in predicate sqwrl: select in the head (consequent) returns all ?x which match the criteria in the body (antecedent). The user only requests coverages ?x with a high resolution, the content semantics are not defined here. If the data has been semantically annotated using the rules from Fig. 6, a reasoner supporting SWRL can re-classify the content. The rule RemoteSensor(?a) ∧ hasIdentifier(?a,?b) ∧ swrlb:matches(“CCD”) → hasResolution (?a,high) defines, if the identifier of a remote sensor has the value “CCD”, then its resolution can be classified as high.

The feature type named desmatamentoType has been semantically annotated with the domain concept Deforestation. The concent serves information about recently detected patches of cleared forests in the Amazon region. In a second example we detail how rules can help to span across the individual community vocabularies. The rule gemet: Deforestation(?x) → annot:domainReference (desmatamentoType, ?x) creates an explicit link to the GEMET (European Environment Agency (EEA) 2009) vocabulary from the environmental community. But researchers studying influencing factors for the region’s economic development would rather use terms like “logging” or “forestry” for describing (and searching) such content. The following rule models an equality relation: stw:Logging(?x) ∧ stw : Forest (?y) ∧ stw : operatesIn ( ?x,?y ) → gemet:Deforestation(?x). It basically states that a forest participates in the process Deforestation if it (the same individual) also participates in the industrial activity Logging. If a user requests GI semantically annotated with gemet:Deforestation, the reasoner-supported query processing algorithm will then also consider GI described with stw:Logging to be potentially relevant. The rule is obviously over-simplifying the relation between the two terms. Logging as industrial activity is performed in all parts of the world, and in many case in a sustainable manner. Deforestation as process is a long-term degradation of woodlands driven by illegal logging, forest fires, or forest clearing to acquire farmland. But a perfect representation of the explicit and implicit relations between the terms of different vocabularies cannot be the goal of this approach. An automatic discovery, evaluation, and integration of information, is not the intent of this approach. It should rather enlarge the number of potentially relevant GI (the recall); it remains in the responsibility of the user to evaluate its suitability. To reach semantically correct rules, if possible at all, requires experts from the various communities to collaborate. This is a challenging task, which may exclude non IT experts, including the majority of citizens and policy makers, from the process. But to realize our vision to deliver information to all interest groups, regardless their cultural and educational background, we have to acknowledge that inconsistencies and imprecise definitions will be part of the domain ontologies. Hence, semantic components in the Amazon SDI have to deal with uncertain and inconsistent knowledge.

This section gave an overview over the different actors and GI communities representing the users of an SDI. The different backgrounds and interests of the various GI communities can result in different, sometimes conflicting domain vocabularies. GI community-specific organizations control vocabularies which have to be used to build the domain ontologies used for the semantic annotations. Two examples are presented which show how rules can bride the gap between these different domain ontologies.

The approach to integrate different communities has to consider the plethora of sources for potentially relevant information as well. The selected GI format usually depends on the author’s community. Scientists often distribute discoveries as raw observation data, VGI is usually published as OGC KML files, and by far the most prevalent form in the Web is implicitly referenced GI such as news articles. In this section the problems of heterogeneity among the different forms of information sources are discussed. Semantic interoperability is mainly regarded as a problem of heterogeneous content. In the following we explain why this view is not sufficient and why we additionally have to consider the semantics of content and form.

For the discovery of GI representing deforestation on the Amazon it makes a difference whether you find a report that describes the development in the last decade, a diagram that visualizes this development, or an OGC Web Mapping Service presenting the annual progress of deforestation as a map. The different representations serve different needs. A written report may highlight certain aspects or certain problematic areas. A map provides an overview of the mapped area without putting emphasize on certain details. A diagram is suitable for stressing the temporal aspect, i.e. presenting a development over time. We argue the variety of GI forms has to be taken as seriously as the already addressed variety in describing GI content.

Linguists and communication theorists have discussed the duality between form and content for a long time. Platon reflected on the duality of form and content of single signs already more than one and a half millennia ago (Nöth 2000). Modern semiotics usually refers to the studies of content and form of De Saussure (2001) and Peirce (1966). In communication science the duality of content and form is analyzed at the scale of whole messages. The message requires an expression, the content, and a medium, the form (Maletzke 1963). An idea in your mind, coming for example from observing an event in your environment, can be regarded as the content of your message. Communicating this idea requires to have it expressed according to a language. The language is a semiotic code that is shared across a community and especially between the sender and the recipients of the message (Eco 1991). The chosen language determines the form. Expressions in a natural language usually result in a text which is read and interpreted by the recipient to gain information. Formal languages result in other forms of information sources, formulas and equations are for example expressed in a certain mathematical notation. Graphic expressions like pictures, drawings, diagrams and graphs also use a certain semiotic code to convey information (Nöth 2000).

GI sources have always been available in a wide variety of forms, simply because maps provide an additional, rich and expressive way to represent spatial information. Furthermore, maps can be easily combined with other forms like diagrams, illustrations and texts. The implications of different forms of GI sources where analyzed by Bertin (1967) more than 40 years ago. The Web gave rise to many new tools facilitating the creation of information. In the same time it also fostered rapid dissemination of ideas. Public email lists, blogs, wikis and personal websites are only some examples of the many new forms of information sources. Publishing is no longer a formal act restricted to a limited circle of people. Publishing information to communicate your ideas to a global audience is nowadays as normal as talking about it with your neighbor. New tools fostering social interaction and collaborative work resulted in a Social Web (Dickson 2001) which has the potential to enable public engagement on far larger scale as it has been anticipated in the beginnings of PPGIS research. The Brazilian environmentalist’s blog describing recent developments in the Amazon may be read by people coming from all over the globe. The same accounts for the articles of a German biodiversity expert writing about the impact of these developments on the wildlife. Only ten years ago this local knowledge was nearly inaccessible, participation was limited to experts with immediate access to the information. In the Social Web, collaboration across information communities is slowly becoming reality. Merging Social Web and Semantic Web will even accelerate this phenomenon (Gruber 2008).

The form of GI is often considered to be an indicator for certain characteristics. VGI as most prevalent content in the Social Web is usually created by non-experts, which has a significant impact on potential applications for such content. Aspects like reliability, uncertainty, timeliness, or the author’s credibility play an important role if the usefulness of VGI for a certain application has to be assessed. But in most case explicit information about these quality properties is missing, either due to the lack of support by the format, or because the author was not aware of its importance. These parameters are hard to assess and rarely communicated. Consequently, the value of VGI for location-aware applications is disputed by many. But projects like Wikipedia and OpenStreetMap or the wide-spread usage of open source software illustrates that volunteered contributions can be of high-quality and reliable enough for even professional applications. The problem of evaluating quality can actually be extended to every kind of GI source. Well-documented GI compliant to OGC standards is still useless if no information about the publisher is included. But if crucial information such as the author’s background is included into the GI source ontology (which is independent from the GI’s form), the assessment of the GI’s usefulness regarding its fitness for use is separated from its original form.

The quality may be an aspect of the content, but information about the form is still needed to understand how to use the information. As described in “Semantic web for SDIs”, domain ontologies are community-specific and are built on the community’s vocabulary. The vocabulary used to describe the special characteristics of the form, such as encoding or language, as well as the quality aspects are also described in domain ontologies. But in contrary to domain ontologies representing a certain view on reality, a domain reference as separation between information and reality is not needed here. Figure 8 illustrates how the GI source ontology extends a specific domain ontology capturing the semantics of form and content, e.g. a GML ontology to describe the encoding. Taking the example of the GI source ontology in Fig. 4, a user might take the following rule to specify desired content: annot: domainReference ( ?y, ?x ) ∧ Deforestation (?x) → sqwrl:select(?y). By simply adding some details, he can also constrain the form and a certain quality aspect: hasResolution(?y, high) ∧ annot:domainReference(?y,?x) ∧ Deforestation(?x) ∧ gml:Coverage(?y) → sqwrl:select(?y).

In this section we introduced the need for a clear separation between form and content. The domain reference acts as bridge between these two aspects of GI. The form has to be described in the GI source ontology as well, since information such as encoding or quality are important to assess the GI’s usefulness. But assessment requires a shared vocabulary again. Specific domain ontologies capturing information-specific properties have therefore to be used to create the GI source ontology. If we are able to embed this approach (including the semantic annotations) into the SDI for the Amazon, the vision of discovery and integration of heterogeneous content and heterogeneous form may become reality.

Already the selection of appropriate concepts from the domain vocabularies makes semantic annotations a challenging task, which often depends on expert knowledge in the particular domain as well as in formal logics. Even more complex is the specification of the rules in SWRL. The content provider, though expert in his particular field, may find such a task too demanding. Easy creation of semantic annotations and rules depends on tools which can help to find correct concepts and accordingly to specify the rules. In this section we introduce how a Recommender System (RS) can support users.

RS render suggestions for items, in this case the concepts from a domain ontology, based on certain similarity metrics. Collaborative filtering (CF) (Manning et al. 2008) techniques analyze user habits, for example by computing similar user profiles and recommending items from a similar user’s collection. The similarity of the (geospatial) data is considered for Content-based (CB) recommendation techniques (Ramezani et al. 2008). In the case of rules for semantic annotations, a RS can assume that concepts are appropriate if very similar content refers to them. For the specification of the mappings in between domain ontologies, already existing annotations and rules can provide the basis. In “Reasoning across information communities: walking through the scenario” we illustrate how the suggested methods can be embedded in the Amazon SDI. The RS module is part of the component which supports the GI publication, e.g. as news article on a newspaper’s website. The search engine also has to include the RS to create the rules describing the required content.

According to the discussed scenario, GI sources such as raw spatial data or unstructured, but georeferenced content have to be annotated. The rules of the semantic annotations are part of the GI source ontology, which itself is automatically created from the GI source metadata. In the case of unstructured, but textual content such as a news article, the structured metadata has to be created first. Text mining (Tan et al. 2006) techniques (e.g. based on Formal Concept Analysis (Cimiano et al. 2005)) can analyze free text and extract key concepts. The result is a list of terms (often already with assigned categories such as Person, Place, Company); an example is the viewer of the Calais Web Service⁶ maintained by Thomson Reuters. The resulting collection of terms is obviously incomplete and inconsistent, e.g. only few of the extracted place names really refer to place covered by the text, and even the best mining algorithm can not extract all relevant concepts. In the end manual input by the author is required to complete and clean the list as preparation for the transformation into the according GI source ontology. In the case of spatial data such as satellite images, structured metadata is usually automatically created and extended by the providing organizations.

With the support of the author, this process can be performed without directly interacting with the ontology. Selecting, deleting, and adding key terms is not necessarily different from the tagging process which is commonly applied to content in the Web, such as bookmarks on Delicious or photos on Flickr.⁷ The extracted word lists represent an excerpt of the author’s vocabulary with all its inconsistencies, ambiguities, and bias towards his personal background. In the next step, the author needs to link the key terms to the concepts from the domain vocabulary. For this step, a recommender system analyzes the existing GI source metadata, and retrieves similar content in the catalog. In the case of GI, the similarity measurement has to consider all three dimensions: space, time, and theme (Kuhn 2005). Spatial or temporal similarity, e.g. both GI sources cover the Amazonas region, is not easy to compute. We assume spatial or temporal similarity if two GI source roughly (with a scale-dependent degree of error) occupy the same region on the spatial or temporal dimension. Thematic similarity, e.g. both GI sources show locations of ongoing deforestation, cannot be computed automatically due to the mentioned semantic conflicts. But the co-occurrence of terms in different word lists can indicate similarity, which again may indicate that both sources can be annotated with the same or closely related domain concepts. The nature of recommendations depends on the dimensions the similarity has been computed for. Computing spatial similarity, but no thematic similarity, should for example only render recommendation for place names.

In most case a direct link as in the first rule of Fig. 6 may be sufficient. For more complex rules which include relations between the concepts, a more sophisticated solution is required. User interfaces can, for example, represent rules in a controlled, but easily understandable language. In the SWING project we have studied the use of graphs⁸ in combination with intuitive interfaces to perform this task (Grčar 2008). But discussing how innovative user interfaces supported by recommendations can help users to create even complex rules (and consequently support and encourage even non-IT experts to semantically annotate their content) remains subject for future research. In this section we briefly discussed how a RS module can support even IT novices to find and publish GI in the Amazon SDI. The expressiveness of Semantic Web technologies can facilitate collaboration across information communities, but we need to find ways to abstract from the underlying complexity to make the benefits accessible to all parts of the population.

In the introducing scenario the cattle company’s decision to clear the rain forests on the Ayoreo-Totobiegosode tribe’s land was prevented by public authorities. The Paraguayan government withdrew the working permit for the company on this land (Survival International 2009a). Just recently (Survival International 2009b), the company requested again the license to clear the land to turn it into pasture land. The request’s approval is endangering the ecosystem, including the wildlife as well as the Indian tribe. We mentioned in “Introduction” that an SDI for the Amazon could better support a sustainable development of the environment, if we expand our understanding on its original purpose. An SDI should not only include components serving GI, it should also be able to actively disseminate it to all potential interest groups. Such an active delivery has to span across the various information communities, regardless of the individual’s cultural or educational background. In the remainder of this section we illustrate how we envision an early warning system embedded into the SDI build on top of the introduces rules. The presented rules approximate chains of cause and effect, and can help to understand how processes eventually endanger the tribe’s existence.

Catalogs enable the registration and discovery of OGC Web Services. Most GI, including implicitly georeferenced content like news articles as well as volunteered GI, is not included in such registries. A journalist from a local newspaper may just have written about the company’s request, the information is then published on the newspaper’s website. The website’s target audience is restricted to the local population. But perhaps it is the newspapers policy to use semantic annotations to better categorize their own content, including explicit references to place names and concepts from an economics domain ontology. Semantic annotations are embedded into the content’s metadata, a crawler (Manning et al. 2008) which is continuously searching for websites with model references does eventually index the article.

On the other side, a computer expert working for the NGO Survival International has setup a news feed which continuously publishes information about timely events which may be potentially endangering the indigenous tribes. The feed is based on GeoRSS (OGC) 2009), the events are restricted to the territory of the Ayoreo-Totobiegosode tribe. The feed, used as early warning system for the organization, is created automatically. The computer expert has defined SWRL rules which states that any published GI source with content which can be considered as threat for the Indians should appear as item in the feed. Once the crawler has indexed the mentioned semantically annotated news article, the rules presented in the following section are triggered. Only seconds later, the just published article has been disseminated to all people who have subscribed to the news feed.

The traditional components of an SDI can obviously not perform these tasks. SWRL rules, domain ontologies, and GI source ontologies are encoded in RDF and can be served by traditional RDF repositories.⁹ The RDF encoding enforces URIs as identifiers for the rules or concepts. Constraining this further to URLs enables inference algorithms working with the rules to simply move along the graph of RDF elements coming from different repositories (the URL of the next element can be resolved to the actual RDF code). The interfaces including the recommender system for semantic annotations and decidable rules can be integrated as modules into existing community websites. Created in a loosely-coupled, decentralized fashion, this network of RDF repositories and community catalogs can be seamlessly integrated into an existing SDI. The rules, which may trigger the information flow across the communities, are processed by reasoner-supported components integrated into existing applications. Hence, from a technological perspective the idea of embedding the proposed methodology into an existing SDI for the Amazon is feasible. Creating the intuitive interfaces for creating the rules and semantic annotations is far more challenging and remains subject for future research.

Still, the process of creating a rule to approximate chains of cause and effect is straightforward. The original fact is defined in the antecedent of our rule and the inferred statement is specified in the consequent. The facts in our case are the individuals modeled in the GI source ontologies. Once registered in the knowledge base, the rule engine is invoked and processes the new rules. The individuals from the local ontologies are related via rules to the domain ontologies. For the scenario we assume the existence of three community ontologies representing the domains of economics, ecology, and ethnology. The following concept map in Fig. 9 presents some of their concepts as well as the concept Threat which can be considered domain-independent. All concepts are aligned to the foundational ontology DOLCE¹⁰ (Lehmann et al. 2004), concepts from this ontology are marked with the prefix dolce:. One of the cattle company’s industrial activities is the production of cattle, which relies on sufficient supply on food available on pasture land. The meadows have to be part of the company’s estate, and relies on the industrial development of the existing land into pasture land. In the ecology ontology, the concept Deforestation is modeled as a kind of the concept ecological destruction with forest as affected ecosystem. For the scenario, only the relation between a tribe and its territory has to be specified.

The geographical region is one common denominator which can be used to map in between the domain ontologies. In particular this means that we consider a particular instance representing the geometry of the economy’s Estate to be also an instance of ecosystem from the ecology ontology. This rule is obviously oversimplifying the relation between real property and ecosystem. The rule should actually include means to check if the geometry of the estate is spatially within or overlapping with existing ecosystems like the Great Chaco. As already mentioned, SWRL supports built-in predicates which can be easily extended (O’connor et al. 2007). Spatial extensions, which delegate the comparison of GI source geometries to spatial algorithms, is therefore feasible, but unfortunately not yet available. As depicted in Fig. 10, the mapping rules should (if appropriate) be modeled in both directions. The second rule states that an ecosystem is also the tribe’s territory. Again, this rule does not model all needed aspects: the tribe’s territory is probably only a small part of the ecosystem. Note that completeness (if it can be even reached) is not necessary to reach to intended goal of a warning system. The rules should only approximate causal relations, they should (and can) not model all aspects of the complex relationships between different information communities.

The first two rules in Fig. 11 model that the need for pastures may require industrial development (e.g. clearing primeval forest) of the company’s estate. The intended meaning of the rules is included. In the third rule, we jump from the economics to the ecology ontology. An early warning system for deforestation threats may already issue the first messages to potential subscribers. Here, another rule is triggered to bridge the gap between the ecology and ethnology vocabularies. A Deforestation instance has been asserted to affect an ecosystem which is also home to the Indian tribe. This is modeled as direct threat to the Indians. The consequent of the last rule is just what the NGO’s computer expert is interested in. Since he specified something like Tribe(?a) ∧ endangers(?c,?a) → sqwrl:select(?c) for the creation of the news feed, the rule engine can add the GI source which triggered the whole chain as new item to his news feed.

Our use case has shown how we can reason across domains. If the local news article (usually restricted to a local audience) has been semantically annotated, it can be immediately disseminated to a global readership. All sorts of GI, including professional sources like raw sensor data, volunteered GI and simply georeferenced content, can benefit from the approach as long as structured metadata is made available and GI source ontologies can be generated and semantically annotated. In this section, we walked through the scenario discussed in the introduction. Altogether, we presented three examples where we believe rule-base semantic annotations can be applied to avoid some of the more typical semantic conflicts. The lack of semantic “interoperability” between different information communities is one important factor for the conflicting views on reality. The presented methods help to close the vocabulary gaps between the communities, and could contribute to the SDI for the sustainable development of the Amazon.