An important kind of tacit knowledge in the context of the Semantic Web are the social communication structures among heterogeneous knowledge sources and users. Communication structures heavily influence the way knowledge is generated and used, because in a context of distributed and autonomous information sources like in the Semantic Web, knowledge is constituted and adapted pragmatically through possibly conflictive communication processes. As a way to set social structures in relation to distributively acquired knowledge, this work proposes Open Ontologies and Open Knowledge Bases for the annotation of (first-level) knowledge with emergent social meta-data (social reification). Whereas traditional approaches to knowledge and ontology integration emphasize the consensus finding among the participants, Open Ontologies and Open Knowledge Bases explicitly model semantical heterogeneity in multiple levels of complexity reduction, and allow the probabilistic weighting of inconsistent knowledge resulting from their assertive weight in their communicative context.
The Semantic Web can be seen as the most important effort toward large scale
knowledge building and sharing in an open information environment. Decisive for the success
of this long-term task is the provision of formalisms and mechanisms for the
communication (i.e. symbolic interaction) of a very large number of distributed, autonomous
knowledge sources and users. Shared ontologies and knowledge bases play a crucial
role in this scenario, since they enable such communication, and knowledge acquisition
among autonomous information sources is basically a communicative act.
Traditional approaches to the modeling and acquisition of ontologies and instance
knowledge have several shortcomings in this respect as they seldom handle meaning
dynamics, they seldom consider knowledge as being contextualized with intentions, processes
and effects from the “outside world”, and they usually have no concept for the
treatment of semantic heterogeneity (e.g. resulting from contradictions) that does not result
in a loss of information. Whereas approaches like Emergent Semantics [
With this work we propose Open Ontologies and Open Knowledge Bases as a general approach to the social acquisition and annotation of knowledge for open environments like the Semantic Web (but also, e.g., for open P2P systems and Semantic Grids). It is primarily meant to introduce a fundamentally novel perspective rather than providing technical specifications. 2
The Semantic Web has several key characteristics that make the acquisition and representation of knowledge complicate in contrast to closed systems and applications: Openness Access, number and contributions of information sources are unrestricted for its major part.
Opaqueness of knowledge sources The intentions of knowledge providers are more or less unknown and their trustability and reliability cannot be guaranteed. Opaqueness of users The impact of a knowledge contribution to the Semantic Web on its users is often hard to predict.
High dynamics and complexity There are very large, heterogeneous and fluctuating amounts of knowledge sources, knowledge contributions and users.
Highly controversial Several domains of web knowledge are highly controversial, e.g. in regard to politics, culture and product assessments by consumers. It seems to be extremely unlikely that such fundamentally divergent world views can be homogenized even in regard to general ontological concepts in the foreseeable future. Thus, semantic inconsistency is a reality knowledge management must cope with. No authoritative background knowledge Decentralized structures and different background knowledge lead to a high diversity of individual knowledge. Missing process knowledge Currently, the representation of machine accessible knowledge focusses on “knowledge end-products”, not on the representation of processes that generate, modify or use knowledge.
These issues have in common that they rise mainly from the autonomy and proactivity of knowledge sources and users, being black- or gray-box actors with more or less opaque goals they pursue asserting or forming their individual world views. The way such autonomous entities (conceptually captured in the notation of information agents in this work) exchange information is communication. Although truly intelligent information agents are not expected to be widely spread on the internet in the foreseeable future, web knowledge can already be considered as communicative, because it is generated in order to influence its recipients and its intentionality and reliability is often unknown. This is even true if knowledge is communicated indirectly, tacitly or asynchronously using e.g. static web sites. Web knowledge is also contextualized with other web knowledge, and it can be agreed as well as denied by other knowledge facets (respectively their sources). Therefore, it appears to be reasonable to consider the Semantic Web as a very large, heterogeneous and hybrid system of interacting information agents (including humans), where information provided by humans and computationally generated knowledge co-exist. Due to the highly distributed character and the heterogeneity of this partially “wild grown” multiagent system, besides agreed protocols and formalisms, shared ontologies and knowledge bases are expected to be extremely useful to enable and improve mutual understanding and interactivity. Because knowledge on the Semantic Web is not only required in order to improve communication, but, maybe even more important, is an emergent outcome and constituent of communication, the key properties of communication need to be taken into account when it comes to building such ontologies and knowledge bases. Thus, viewing the Semantic Web as a system of directly or indirectly communicating information agents, we propose a communication-oriented paradigm, which has several implications for the retrieval and modeling of distributed knowledge. Most important, knowledge management for the Semantic Web needs to cope with the fact that the meaning of information on the web can never be determined for sure in general, might change, and might be constituted from the possibly conflicting opinions of large sets of knowledge sources. The primary goal of Open Ontologies and Open Knowledge Bases is to make the knowledge contributions of large, fluctuating and possibly conflicting sets of autonomous sources usable in a computational sense, i.e. to provide computationally accessible meta-data to the users even if such socially accumulated knowledge is inconsistent or unreliable (especially in the absence of trustability). For this purpose, the social layer of knowledge on the web needs to be found and made explicit by means of semantic annotation to the web users. In particular, the technical openness of shared knowledge like ontologies and the comparability of distributed, local knowledge needs to be improved, knowledge artifacts need to be interpretable as parts of communication processes (with induced relationships like assertion, agreement, contradiction, request, revision, specialization, generalization...), and the complexity of socially accumulated knowledge needs to be reduced without the need to come to a consent among the participants and with as less loss of information about social heterogeneity as possible.
Largely neglecting these aspects, most of the current efforts in order to build the
Semantic Web concentrate on the specification of languages and tools for the
modeling of agreed, homogeneous knowledge, and research is just beginning to take into
consideration phenomena like the social (i.e. communicative) impact of resource
descriptions, conflicting opinions, information biased by e.g. competing commercial or
political interests, and inconsistent or intentionally incorrect information. Bringing
information (e.g. via web sites or web services) into the web is in fact a social act,
and the relationship between informational artifacts on the web is communicative (i.e.
specifying, agreeing, contradicting...). This can of course produce intentional and
unavoidable inconsistencies (e.g. company interests versus customer interests or various
conceptualizations due to differences in culture). If these are ignored, or filtered out,
ranked/recommended or homogenized too early (e.g. applying trust), important
information for the user or the application might be lost. In order to make this important
information available, we propose the following:
– Knowledge facets on the web like meta-data annotating web pages must be seen as
subjective belief assertions of rational intelligent black-box agents (artificial agents
as well as human users). They are created with certain intentions which are more or
less hidden and are situated within action processes in order to make the
successful assertion of this particular “truth” more likely (with advertisement as the most
usual case, but also e.g. user recommendations regarding products and political
statements, and even lexicon entries).
– Knowledge heterogeneity needs to be made explicit. Since knowledge sources are
more or less opaque with hidden belief and goals, the need for instruments that
enable the comparison of different standpoints becomes more important for
knowledge users.
– Knowledge heterogeneity needs to be explained. Publication of knowledge on the
web is an assertive act that is embedded within a pragmatical context of reasons
and implications. In fact, the meaning of knowledge cannot be determined without
considering this pragmatical context [
Whereas it is already widely agreed that the statements of human individuals can only be transferred to machine understandability with a more or less degree of uncertainty, the need for the use of probabilistic and approximate representation formalisms in order to model collectively constituted knowledge on the web is still largely neglected.
Figure 1 shows the semantical levels proposed by Tim Berners-Lee for the structure of the forthcoming Semantic Web, with extensions (red/light gray font) we recommend for some aspects of this concept in response to the mentioned issues. In particular, it appears to be inevitable to us to provide formalisms and calculi that explicitly consider semantically heterogeneous meta-data like resource descriptions and ontologies created from the contributions of multiple sources that compete for the assertion of their individual “truths” and interests. Of course, the Semantic Web is already open, but for a broad acceptance and to provide value to its users, we strongly suppose that communicative (i.e. social) relationships among closed “islands” of knowledge like contradiction or agreement need to be made explicit formally and technically as part of the layers of a “socially-aware“ Semantic Web, using a concept called social reification (cf. next section). In this regard, the empirical derivation and stochastic modeling of open meta-data seems inevitable if the set of knowledge sources is either very large, or fluctuates, or generates indefinite information.
Formal ontologies and knowledge bases are traditionally defined as agreed descriptions
of certain domains which serve as common ground for distributed tasks like
knowledge exchange, modeling and user information. This understanding leads to
difficulties if the informational input these media are build from is likely to be intentionally
inconsistent, and there either does not yet exist enough meta-knowledge like trust to
identify and filter out “inappropriate” or “wrong” data a priori, or there does not even
exist a concept of global inappropriateness or correctness at all. On the other hand,
sound and agreed ontologies are doubtless an inevitable prerequisite for efficient
knowledge creation, representation and exchange, whereby we consider implicit and emerged
ontologies and schemata (e.g. in the context of semi-structured data modeling) to be
such ontologies too. Of course, ontological heterogeneity can be overcome by means
of techniques like the renaming of inconsistent concepts, and in general, inconsistent
knowledge can be made consistent providing appropriate truth contexts [
Communication is here not so much to be understood as the exchange of symbols with a fixed meaning, but the other way round as a means to generate supra-individual meaning from interrelated interactions among black- or gray-box agents (i.e., agents with more or less unknown internal states, cognition and goals). The practical consequences arising from this are that OO&OKB need to be continuously adapted to new information, and the processes of creation, contextualization and interpretation of knowledge are integral aspects of OO&OKB themselves. In addition, communication among multiple agents likely requires mechanisms for the generalization of emergent meaning, since otherwise the complexity would grow too large due to the sheer number of individual knowledge contributions. Generalization is also a way to make OO&OKB look like homogeneous ontologies or knowledge bases if necessary, because at its highest level, generalization causes semantical homogenization among contradicting knowledge sources. Summing it up, Open Ontologies and Open Knowledge Bases have the following characteristics: Openness No (or as few as possible) initial assumptions are made regarding the benevolence, trustworthiness, relevance, informedness and cooperativeness of its sources. Nevertheless, information about e.g. (dis-)trust and knowledge (un-)reliability is likely derivable from Open Ontologies and Open Knowledge Bases, since these are special cases of social structures.
Dynamical derivation from communication OO&OKB are emergent from and evolving with ongoing communication (e.g. agent interaction, but also asynchronous, indirect or tacit communication e.g. via the semantically interrelated contents of web sites) of knowledge sources and knowledge users to assert (deny, specify...) information and to express and specify informational needs and expectations. Social background knowledge (existing social structures like laws) can be included in the derivation process.
semantical inconsistencies arising from contradictions and conflicts, and contain (consistent) annotations of (conceptual or instance) knowledge with meta-information about its social meaning within the course of communication.
This concept is related to context logic [
OO&OKB contain as first-order objects knowledge facets that have the form 1st-level
knowledge ← 2nd-level knowledge, where 1st-level knowledge partially describes a
domain concept in the same way as within usual ontologies (or instances of such
concepts, respectively, for Open Knowledge Bases), but probably in an inconsistent way
regarding other 1st-level knowledge in the same ontology. Since Open Ontologies are
primarily an abstract meta-concept build upon conventional approaches for the
representation of conceptual knowledge, we do not constrain or specify the sort of concrete
entities that are to be “wrapped” within an Open Ontology (Open Knowledge Base)
or at the content level of agent messages, like first-order logical statements, classes or
frames. For the same reason, we do also not make any assumptions relating to ontology
domains or concrete areas of application here. In contrast to 1st-level knowledge,
2ndlevel knowledge (also called social knowledge) depicts the social context of 1st-level
knowledge, the latter taken as generated from a communication act of an autonomous
source of knowledge. This kind of annotation of 1st-level knowledge with 2nd-level
knowledge we call social reification. A quite trivial kind of social reification is quoting
(e.g., ’Sue says: “...”’), but in general, all kind of information which describes how and
to what effect certain data is produced within a process of communication can be
informally understood as 2nd-level knowledge (and, of course, we can apply social
reification recursively, i.e. annotate 2nd-level knowledge with 3rd-level knowledge as in ’Sue
says: ’Tom says: “...”’ and so on). The most elementary forms of such social meta-data
are considered agent speech act types like assertion, denial or query, inducing relations
among single communication like ’Sue contradicts Tom’s statement saying “...”’ and
rich 2nd-level knowledge types such as knowledge source and user profiles and even
complex social systems like organizations. In an empirical communication model [
Open Ontologies and Open Knowledge Bases need to be learned from the observation
of communication processes. The technical requirements for this learning process are:
– information agents or other knowledge sources (e.g. peers in a P2P network, or
passive resources like web documents) able to communicate and query 1st-level
knowledge facets. In case of software agents, this can be done by means of a
formal agent communication language (since OO&OKB do not require agent
cooperativeness, speech act performatives used for collaboration like negotiation are not
required, although they would be useful).
– a facility for the acquisition of OO&OKB from the observation of above
communications, e.g., a dedicated middle agent within the infrastructures of the respective
application, called a semantics observer (cf. figure 2).
– optionally, a pre-defined content of the Open Ontology or Open Knowledge Base,
in order to speed up the learning process of the semantics observer, and to avoid
the bootstrapping problem known from e.g. recommender systems, or to set static
social structures like norms
– a facility for the low-level storage and querying of persistent knowledge (e.g., a
database management system).
– optionally, a facility for the social reasoning upon the 2nd-level knowledge within
the Open Ontology or Open Knowledge Base. respectively (to deduce new facts
like “Sue is likely to contradict or specify Toms information”, but also to derive
trust relationships among the participants subsequently. Here, known techniques as
described in e.g. [
The acquisition of OO&OKB comprises the following main tasks, which have to be
performed in a loop as a continuous, incremental learning process for the whole period
of agent communication (please find details in [
Homogenity
… Heterogeneous
knowledge annotates/weights
Higher-level communication structures
of roles, organizations…
Heterogeneous knowledge annotates/weights Low-level communication structures of agents and clients (e,g, emergent profiles)
S ocial complexity /degree of heterogeneity
S emantics Observer (middle agent)
C ommunication processes
G e n e lir a z a it o n 2. Derivation and/or adaptation of 2nd-level knowledge according to the respective semantical model (e.g. empirically) 3. Stochastic generalization of 2nd-level knowledge 4. Social reification and generalization of 1st-level knowledge 5. Alignment with given, obligatory 1st-level knowledge (e.g. a normative top-level ontology) or normative 2nd-level knowledge (e.g. laws preventing certain utterance of certain information), if necessary.
As mentioned earlier, OO&OKB also require the generalization of meaning in
order to reduce their complexity (cf. figure 2). Generalization as a task in this sense has
two steps: 1) the merging of 2nd-level knowledge, 2) the subsequent merging of related
1st-level knowledge facets. Typically, 1) comprises the merging of similar social
processes to interactions patterns, and the combination of multiple similar behaving agents
to social groups or social roles. After applying such generalization rules to 2nd-level
knowledge, the annotated 1st-level knowledge needs to be merged accordingly. If, for
example, multiple agents forming a single social group make inconsistent assertions,
within the Open Ontology (Open Knowledge Base) each of these assertions obtains
a probabilistic weight expressing the degree of expected approval this assertions gets
from the role or group as a whole (calculated, e.g., from the frequency this assertion
has been uttered by different agents within this role or group) [
There is an obvious and rapidly growing need for knowledge-based systems capable of running in open environments like the Semantic Web with autonomous knowledge sources and users, given the increasing inter-operability and inter-connectivity among computing platforms. On the one hand, knowledge bases and ontologies should provide a stable ground for user information, agent and user communication and subsequent knowledge modeling, on the other hand, in open environments concept descriptions tend to be semantically inconsistent, they emerges from a possibly very large number of competing subjective beliefs and goals, and a priori there might be no such thing as a commonly agreed “truth” (in the “real world”, not even a discursive trend towards such a thing can be assumed). To cope with these two contradictory aspects must be a core concern of the communication-oriented paradigm of knowledge modeling and management, and is the basic motivation underlying the work described here. To this end, we have proposed Open Ontologies and Open Knowledge Bases as a fundamental step towards the modeling and representation of socially-induced knowledge heterogeneity for the Semantic Web.