# Footnotes

1 Thus the relation symbol |= is understood either as “satisfies” or as “entails” depending on whether its first operand is an interpretation or a set of formulas.

2 Note that equality is not a predicate constant, according to this definition. Although syntactically it is similar to binary predicate constants, it does not belong to the signature, and the semantics of first-order logic, defined below, treats equality in a special way.

3 ResearchCyc [169] contains the knowledge base open to the public for research; certain portions of Cyc itself are not open to the public. The knowledge base of ResearchCyc contains over a million assertions.

4 The term *logicism *generally refers to the school of thought that mathematics can be reduced to logic [270], *logicists *to the proponents of logicism. Within the artificial intelligence community, however, a *logicist *refers to a proponent of logicist AI, as defined in this section [257].

5 Although [111] was published in the 1980s, a preliminary version was first written in the late 1970s.

6 Each year the International Conference on Theory and Applications of Satisfiability Testing hosts a SAT competition or race that highlights a new group of “world’s fastest” SAT solvers, and presents detailed performance results on a wide range of solvers [141–143, 215]. In the 2006 competition, over 30 solvers competed on instances selected from thousands of benchmark problems. Most of these SAT solvers can be downloaded freely from the web. For a good source of solvers, benchmarks, and other topics relevant to SAT research, we refer the reader to the websites SAT Live! (http://www.satlive.org) and SATLIB (http: //www.satlib.org).

7 The contrast between typical- and worst-case complexity may appear rather obvious. However, note that the standard algorithmic approach in computer science is still largely based on avoiding any nonpolynomial complexity, thereby implicitly acceding to a worst-case complexity view of the world. Approaches based on SAT solvers provide the first serious alternative.

8 Due to space limitation, we cannot do justice to a large amount of recent work on complete SAT solvers, which consists of hundreds of publications. The aim of this section is to give the reader an overview of several techniques commonly employed by these solvers.

9 In some implementations involving lazy data structures, solvers do not keep track of the actual number of satisfied clauses. Instead, the formula is declared to be satisfiable when all variables have been assigned truth values and no conflict is created by this assignment.

10 As in our discussion of the complete SAT solvers, we cannot do justice to all recent research in local search solvers for SAT. We will again try to provide a brief overview and touch upon some interesting details.

11 It has been recently shown that by finely tuning the noise parameter, Walksat can also be made to scale well on hard random 3-SAT instances, well above the clause-to-variable ratio of 4.2 [208].

12 We use C[ν/x] to denote the simplified CSP obtained by setting the value of variable x to ν in C.

13 PSPACE-complete problems are generally believed to be significantly harder than NP-complete problems; cf. [179].

14 http://www.w3.org/2001/sw/WebOnt/.

15 The use of S is motivated by the close connection between this DL and the modal logic **S4**.

16 This is due to the algorithm searching a tree of worst-case exponential depth. By re-using previously computed search results, a similar algorithm can be made to run in exponential time [66], but this introduces a considerable overhead which turns out to be not always useful in practice.

17 In description logics, blocking was first employed in [8] in the context of an algorithm that can handle the transitive closure of roles, and was improved on in [13, 46, 12, 92].

18 _{For }ALC, we can always construct a finite cyclical model in this way; for more expressive DLs, we may need different blocking conditions, and we may need to unravel such cycles in order to construct an infinite model.

19 The cited papers actually use automata based approaches to show EXPTIME results for *extensions *of ALC^{.}

20 Sub-Boolean DLs are DLs that are not equipped with all Boolean operators.

21 Using the so-called pre-completion technique [88], this is also the case for inference problems involving ABoxes.

22 Strictly speaking, we obtain a finite automaton with word transitions, i.e., transitions that may be labelled with a word over Σ rather than a letter of Σ.

23 In fact there are 3 “species” of OWL: OWL Lite, OWL DL and OWL full, only the first two of which have DL based semantics. The semantics of OWL full is given by an extension of the RDF model theory

[83].

24 In fact, there are a few additional constructors provided as “syntactic sugar”, but all are trivially reducible to the ones described in Fig. 3.6.

25 http://www.topbraidcomposer.com/.

26 We note, however, that Reiter treats open defaults differently and uses a more complicated method to define extensions for them. A theory of open default theories was developed by [73]. Some problems with the existing treatments of open defaults are discussed in [5].

27 For an introduction to computational complexity theory and for basic definitions and results on polynomial hierarchy, see, for example, [46, 103].

28 We note however, that default logic also played a role in the development of the stable-model semantics [13] and, in fact, the default-logic connection of stable models ultimately turned out to be more direct [82, 15, 14].

29 To avoid problems with the treatment of quantifiers, we restrict our attention to the propositional case. Consequently, we have to list “normality” rules explicitly for each object in the domain rather than use schemata (formulas with variables) to represent concisely families of propositional rules, as it is possible in default logic. The “normality” rule in our example concerns Professor Jones only. If there were more professors in our domain, we would need rules of this type for each of them.

30 Proofs in a modal logic use as premises given assumptions (if any), instances of propositional tautologies in the languageponens and the *necessitation rule*L^{K}, and instances of modal axioms of the logic. As inference rules, they use modus, which allows one to conclude KA once A has been derived.

31 When applying the propositional consequence operator to *modal *theories, as we do here, we treat formulas KA as propositional variables.

32 A binary relation that is irreflexive and transitive.

33 Typically, W is assumed to be nonempty. This assumption is not necessary for our considerations here and so we do not adopt it.

34 In the original paper by Shoham, a stronger condition of well-foundedness was used.

35 For simplicity we consider a special case of the more general original definition which allows arbitrary countable ordinals.

36 An exception to a default is called *weak *if it stops application of the default without defeating its conclusion.

37 For simplicity we omit several less important features of the language.

38 We shall use the capitalized term “Belief Revision” to refer to the *research area*; the same term in lower case letters will be used for the *process *of belief change.

39 It should be noted that representing a belief set as a theory, presupposes that agents are *logically omniscient*. In this sense the AGM paradigm is tailored for *ideal reasoners*.

40 Although these postulate where first proposed by Gardenfors alone, they were extensively studied in collaboration with Alchourron and Makinson in [2]; thus their name.

41 The high priority of ϕ over previous beliefs may not always be related to its reliability. For example, in the context of the *Ramsey Test *for conditionals, ϕ is incorporated into a theory K as part of the process of evaluating the acceptability of a counterfactual conditional ϕ > ψ (see [30]).

42 Another interesting instance of contraction is during argumentation. Consider two agents A and B that argue about a certain issue for which they have opposite views. It is quite likely that *for the sake of argument *the two agents will (temporarily) contract their beliefs to reach some common ground from which they will then starting building their case.

43 Incidentally, like with the AGM postulates for belief revision, one can show that there exists more than one function_{10 }−˙ satisfying (K −˙ 1)–(K −˙ 8).

The result still holds even if −˙ does not satisfy (K −˙ 5) (i.e. the *recovery postulate*).

44 Consider, for example, the theory , where p and q are propositional variables, and suppose that we want to contract by . There are more that one maximal subsets of K failing to entail

p ∧^{12}qIn other words, a, one of which containsϕ-_{remainder is a subset}p but not q, while another contains such that q but not p., and (ii) for any , if

_{then }.

45 In the limiting case where ϕ is a tautology, K ⊥⊥ϕ is defined to be {K}.

46 In particular, they satisfy the *basic *postulates (K −˙ 1)–(K −˙ 6) but fail to satisfy the *supplementary *_{postulates (K }−˙ _{7) and (K }−˙ _{8).}

47 At first glance, condition (C-) may seem unnatural. Indeed there is an equivalent and much more intuitive way to relate epistemic entrenchments with contraction functions (see condition (C) in [31]). However, condition (C-) is more useful as a construction mechanism for contraction functions.

48 Recall that M^{L }is the set of all consistent complete theories of L, and for a theory K of L, [K] is the set of all consistent complete theories that contain K.

49 With one notable exception: Spohn’s work [93] on iterated revision which will be discussed in Section 8.6.

50 This is often called the *foundational approach *to belief revision. ^{19}Apart from the degenerate case where the two are identical.

51 Earlier publications by Hansson also report on similar results. However [41] gives a more detailed and uniform presentation of his work on belief base contraction.

52 To be more precise, gives only a *partial *definition of ∗; namely only its restriction to K. For a complete specification of ∗ we would need a whole family of epistemic entrenchments, one for every theory of L. This abuse in terminology occurs quite frequently in this chapter.

53 It turns out that this ensconcement-generated revision function ∗ has yet another interesting property. It can be constructed from following another route: gives rise to an epistemic entrenchment by means of (EN1), which in turn produces a revision function by means of (E∗), which turns out to be identical with ∗.

54 There are in fact some subtle differences between the definition of ⊕ presented herein and the one given by Lindstrom in [58], which however are only superficial; the essence remains the same.

55 A synonym for multiple revision is *infinitary *revision. In fact this is the term used by Zhang and Foo in [107].

56 It is not hard to see that (SF) entails (SD). Simply notice that from (SF) it follows that all spheres in S are elementary, and consequently, the intersection of [Γ ] (for any set of sentences Γ ) with *any *sphere of S is also elementary.

57 Similarly to sentence revision, in this section we shall use the term *sentence contraction *to refer to the original contraction functions whose inputs are sentences rather than sets of sentences.

58 In fact, all we can deduce is that K ∗ ϕ ∗ ψ is a theory containing ψ.

59 In this sense an ordinal conditional function κ is quite similar to a system of spheres S: both are formal devices for ranking possible worlds in terms of plausibility. However κ not only tells us which of any two worlds is more plausible; it also tells us by *how much *is one world more plausible than the other.

The left subtraction of two ordinals α,β such that, is defined as the unique ordinal γ such that

Mα =_{L}^{33}:That is, given an OCFκ+∗ &ϕ,d. '(r) = 0} satisfies the AGM postulates (κ and any d > 0, the functionK ∗ 1)–(K ∗ ∗_{8).}defined as

60 This is the case where the new information ϕ contradicts the original belief set (since κ(¬ϕ) > 0, the agent originally believes ¬ϕ).

61 Although it should be noted that Darwiche and Pearl argue that this shift is not necessitated by technical reasons alone; conceptual considerations also point the same way.

62 _{If }<· is required to be a strict order (i.e., transitive and antisymmetric), then things are not as simple but it is still possible (in principle) to reduce screened revision to relationally screened revision.

63 To be precise, in [43] the function ∗ does not have to satisfy the AGM postulates; when it does, the induced operator is called a *credibility-limited AGM revision*. Herein we focus only on such operators and therefore, for the sake of readability, we have dropped the AGM advert from the title of .

64 This constructive model is slightly different from the one discussed in [43] but it is nevertheless equivalent to it.

65 Nevertheless, many would still classify Belief Merging as a sub-area of Belief Revision.

66 Ignoring high pressure situations at which both the freezing point and the boiling point are the same, known as the triple point in thermodynamics.

67 It might even be possible to construct the reals over an interval using landmark introduction, via a method analogous to Dedekind cuts.

68 This is known as the principle of probability kinematics [88].

69 A variable Z is a nondescendant of X if Z /∈ X**U **and there is no directed path from X to Z.

70 From a complexity viewpoint, all posterior marginals can be computed using a number of PR queries that is linear in the number of network variables.

71 Recall, that ∼ represents the compatibility relation among variable instantiations.

72 Jointrees correspond to tree-decompositions [144] in the graph theoretic literature.

73 For a simple distinction between observing and intervening, note that observing D leads us to increase our belief in its direct causes, A and B. Yet, our beliefs will not undergo this increase when intervening to set D.

74 However, describing time as *flowing *might even be an assumption too far! Several authors have considered time with *gaps *in it [112, 28].

75 As one might expect, quantification in temporal logics is related quite closely to quantification in modal logics, though quantified modal logics are not without difficulties [120, 195].

76 In certain domains, containing rectangular objects which are uniformly aligned, this can still be a useful representation, see, for example, [208] where the layout of text blocks on envelopes is learned. A theoretical analysis into the n-dimensional generalisation of the Allen calculus can be found in [9].

77 Much relevant material is published in the proceedings of COSIT (the Conference on Spatial Information Theory), GIScience (the International Conference on Geographical Information Science), the journal Spatial Cognition and Computation, as well as regular AI outlets such as the AI journal, the Journal of Artificial Intelligence Research (JAIR) the International Journal of Geographical Information Science, and the proceedings of such conferences as KR, AAAI, IJCAI, PRICAI and ECAI.

78 Simons [189] says: “No one has ever perceived a point, or ever will do so, whereas people have perceived individuals of finite extent”.

79 Mereotopology is defined and discussed in detail in Section 13.2.4 below.

80 See [59] for a review of the use of relation algebras in spatial and temporal reasoning.

81 For the sake of uniformity, in a number of cases we have renamed predicate and other symbols in this chapter from the original formulation.

82 For further discussion of this issue see [27, 58].

83 In fact, in [49] a type is quadruple, but we ignore the final component here.

84 Galton [92] coined this name.

85 A simpler, purely mereological calculus (usually called RCC-5), in which the distinctions between TPP and NTPP, TPPi and NTPPi, and DC and EC are collapsed has also been defined and investigated [127, 117].

86 Actually, a simpler 2 × 2 matrix [65] known as the 4-intersection featuring just the interior and the boundary is sufficient to describe the eight RCC relations. However the 3×3 matrix allows more expressive sets of relations to be defined as noted below since it takes into account the relationship between the regions and its embedding space.

87 The RCC-8 relations have different names in the 9-intersection model, in fact English words such as “overlap” instead of PO.

88 Actually it is straightforward to specify relative measurements given an “absolute” calculus: to say that x > y, one may simply write x − y = +.

89 Section 13.3.2 introduces composition tables.

90 E.g., because distances are sometimes measured by time taken to travel, and an uphill journey may take longer than a return downhill journey [113].

91 Of course, orientation and distance primitives as discussed above already add something to pure topology, but as already mentioned these are largely point based and thus not directly applicable to describing shape of a region.

92 Due to the missing regularity conditions in the intuitionistic encoding, Nebel’s result turned out to be incomplete.

93 Claim 24 in this paper is subsequently admitted not to hold [28]; further work on this problem, generally known as the “map graph” recognition problem can be found in [29, 30, 197, 31].

94 Note that one can lift this notion of closeness from individual relations to entire scenes via the set of relations between the common objects and thus gain some measure of their conceptual similarity as suggested by [23].

95 A closely related notion is that of “closest topological distance” [67]—two predicates are neighbours if their respective n-intersection matrices differ by fewer entries than any other predicates; however the resulting neighbourhood graph is not identical to the true conceptual neighbourhood or continuity graph— some links are missing.

96 Some existing research on this problem includes [82, 80, 192].

97 In principle, high-level physical reasoning could enter into visual recognition, either by providing constraints or measures of likelihood for possible scenes [44] or by relating physical conditions of the image formation process to qualities of the image—e.g., if the lens cap is left on, the image will be black. In practice, the former has been rarely attempted in vision research, and the latter, as far as I know, has never been attempted.

98 One reflection of the cognitive salience of this category is the persistent attempt in eighteenth- and nineteenth-century physics to reduce all physics to mechanical interactions of small solid objects; e.g., the kinetic theory of heat, or Maxwell’s mechanical model of electrodynamics.

99 A *closed *region is one that includes its boundary. The decision to use a closed rather than an open region is arbitrary, but it simplifies description to specify one or the other. A closed region is *regular *if it is equal to the closure of its interior, and thus is “thick” everywhere and does not have any one- or twodimensional pieces.

100 A displacement is a composition of a rotation around the origin and a translation. A translation in k dimensions is characterized by a vector t-; any point x is mapped into x + t-. A rotation in two dimensions (relative to a fixed origin) is characterized by an angle φ. A rotation in three dimensions is characterized by three angles; there are a number of different systems of angles that can be used for this purpose, such as the Euler angles. Alternatively, a k-dimensional rotation can be characterized by a k × k orthonormal matrix.

101 Man-made mechanisms tend to rely on kinematic constraints when possible, because they are extremely robust. A large external force or impact is generally required to make solid objects significantly bend or break, and there is no way to cause two solid objects to spatially overlap.

102 The desiderata for such a theory are that it corresponds to experiment; that it satisfies global constraints, such as conservation of energy, momentum, and angular momentum; that it yields a solution for all wellposed initial-value problems; that numerical calculations converge; and that it can be justified in terms of a more detailed elastic model of solid objects.

103 A similar language, L^{B}_{n }, can be defined for reasoning about belief if we substitute the modal knowledge operators K_{i }by analogous belief operators B_{i}, for all i ∈ [n]. We will speak in terms of knowledge and make explicit mention of belief when this is warranted.

104 It is often convenient to think of E as a state of knowledge of the group of all agents. Indeed, there are cases where a fact can be known to all members of a given set G of agents, but not to the rest. For example, we may be interested in whether all students of a given class know that the exam date has been changed. Within a larger framework, it might not be of interest to ensure that, say, everyone in the university knows this. When analyzing such scenarios, it is customary to use operators such as EG, which stand for *everyone in *G *knows*. Similar restriction to groups of agents will be applicable to other states of knowledge that we discuss below.

105 In a Stanford Technical Report that was later published as [25].

106 Reiter [34] defined regression for a more general class of formulas that can contain *Poss *atoms.

107 Alternatively, one could also view the body of a rule as the precondition of the corresponding action.

108 If a predicate does not occur as the head of a rule, then add an axiom to say that this relation does not hold for any situation.

109 Strictly speaking, it is not a basic action theory as the right side of a successor state axiom may contain ∃s.Q(s) when there is a negation in front of Q in a rule, thus is not a uniform formula.

110 We do not treat modal logic versions of the event calculus [9].

111 Kowalski and Sergot [45] use implication (⊃) in OEC1 and OEC2. Sadri [87, p. 134] points out that bi-implication (≡) was intended by Kowalski and Sergot but not used in order to prevent looping when running the axioms in Prolog.

112 Kowalski [40] uses implication, whereas Sadri [87] uses bi-implication.

113 Events with duration may also be represented as fluents that are initiated and terminated by instantaneous events. For example, a moving event with duration can be represented using the axioms *Initiates*(*StartMoving*,*Moving*,t) and *Terminates*(*StopMoving*,*Moving*,t). See also the discussion of continuous change in Section 17.5.7.

114 Preferential entailment reduces the set of classical models of a theory by only retaining those models that are minimal according to a given preference relation, a strict partial order over logical interpretations [66].

115 Essentially, an order-sorted language allows the use of sub-sorts; for example, CAR and BICYCLE may be sub-sorts of the VEHICLE sort.

116 This will initially result in a scenario where it is assumed that any attempt to board a plane always succeeds, regardless of whether a person carries a gun or is drunk. In addition, ramifications of action effects will be included in action specifications rather than being specified separately.

117 This scenario is an elaboration and concretization of a sketch for a scenario proposed by Vladimir Lifschitz in on-line discussions in the Electronic Transactions on Artificial Intelligence (ETAI/ENAI), and was previously published in [11, 42].

118 Assume that Boris, Erik and Dimiter own the combs*comb1*,*comb2 *and*comb3*, respectively.

119 A number is often suffixed to each label, as in **per3**. These numbers are used to disambiguate references in the text and have no semantic meaning.

120 In some earlier versions of TAL, an explicit *Observe *predicate was introduced in the base logical language L(FL) to which observation statements are translated. Distinguishing sensor-generated facts about the world from other facts is useful when interfacing such logics to robotic systems. One might choose to view observation statements as perception statements, although this is not done in the current version of TAL.

121 Note that due to the regularity of the solution, such extensions could be implicit in an action macro, thus avoiding unneeded clutter in the representation and delegating representation responsibility to the system rather than the knowledge engineer.

122 Note that this does not rule out the generation of plans with concurrent actions and one version of TALplanner does generate actions concurrently.

123 We appeal to the use of *semantic attachment *[71] techniques in the implementation of TAL and TALplanner by liberal use and invocation of built in mathematical and other functions associated with value domains for features.

124 http://www.languagecomputer.com.

125 Recall that a word is a hypernym of another if the former is more generic or has broader meaning than the latter.

126 Further information can be found at: http://www.hlt.utdallas.edu/~moldovan/CS6373.06/IS_Knowledge_Representation_from_Text.pdf, http://www.hlt.utdallas.edu/~moldovan/CS6373.06/IS_SC.pdf, and http://www5.languagecomputer.com/demo/polaris/PolarisDefinitions.pdf.

127 To complete the connection, axioms for “travel” and “go” are also needed.

128 As this sense of verb “to be” does not admit a predicative complement, constant x27 is unused.

129 Yes/no questions have a simpler structure and are not discussed here to save space. The translation of the LLFs of *Where- *and *When-*queries that do not rely on verb “to be” (e.g., “where did John pack the laptop”) has not yet been fully investigated.

130 Fluents are relevant properties of the domain whose truth value may change over time [37, 38].

131 Recall that, in more complex situations, the definition of relation *time*_*point *can involve the use of defaults, to allow the assignment of time points to be refined during the mapping.

132 Since in the travel domain the origin and destination of trips do not change over time, the formalization is designed to allow to specify the origin using a static relation rather than a fluent. This simplification is not essential and can be easily lifted.

133 http://www.tcs.hut.fi/Software/smodels/.

134 The issue of translating the answer back into natural language will be addressed in future versions of the system.

135 Obviously these defaults are meaningful only in the context of travel-related stories, and can be suitably qualified in AnsProlog . We omit the qualification to simplify the presentation.

136 We use a complex sentence processer that processes complex sentences to a set of simple sentences. Thus we assume that there is one event in each sentence. We assign event numbers sequentially from the start of the text. This is a simplistic view and there have been some recent work on more sophisticated event analysis, such as in [47].

137 http://www.cogsci.ed.ac.uk/~jbos/RTE/.

138 http://svn.ask.it.usyd.edu.au/trac/candc/wiki/.

139 In fact such an axiom has a flaw. Consider a following pair *Text*: “Abraham is the husband of Sarah. Abraham is the father of Isaac. Isaac is the husband of Rebecca.” and *Hypothesis*: “Abraham is the husband of Rebecca.” Given a first-order logic representation of the pair and this axiom, *Text *entails *Hypothesis*. Resolving such issues is the problem of farther investigation.

140 http://www.math.chalmers.se/~koen/paradox/.

141 rdf: and owl: are common abbreviations for actual deferencable URIs that link to the standards documents that define RDF and OWL, respectively.

142 http://www.w3.org/2001/sw/hcls/.

143 It is worth noting that a number of large OWL ontologies also exist without instances, for example, the National Cancer Institute maintains a metathesaurus that is released in OWL. At the time of this writing it has over 50,000 class definitions [10].

144 We make length a string for simplicity of this example, we leave defining and enumerated class of appropriate lengths as an exercise to the reader.

145 The interested reader is also directed to SCL [20] an attempt to provide a unifying logic for the Web which allows some higher-order-like reasoning within the constricts of FOL.

146 http://www.w3.org/2001/sw.

147 Here we chose to identify each state of the domain with a basic Boolean proposition of CTL formulas. Actually, we would need only 9log_{2 }|S|< basic propositions, using a boolean encoding of the states.

148 It is easy to extend the algorithm to the case of more than one initial state.

149 To the best of our knowledge, the term was first used publicly by Reiter at his lecture on receiving the IJCAI Award for Research Excellence in 1993.

150 While planning languages like STRIPS [28] or PDDL [53] also qualify and have been used to control robots [51, 57, 20], they are more limited in that they only specify planning problems, but do not lend themselves to a general representation and reasoning framework for cognitive robots as advocated by Reiter.

151 Thinking of *all *actions as instantaneous in this way also has the advantage of reducing the need for true action parallelism, allowing us to use the much simpler variant of interleaved concurrency [17].

152 Modeling knowledge using possible worlds is due to Hintikka [35].

153 Instead of a single tree rooted at S0, we now have a forest of trees each with their own initial situation.

154 This is perhaps a difference of emphasis only: cognitive robotics tends to emphasize the robotic interaction with the world, whereas AOP tends to emphasize the mental state of the agent executing the program.

155 In many applications, we can preserve the effectiveness of an essentially deterministic situation calculus by pushing the nondeterminism into the programming.

156 Much of the work with *Trans *and *Final *requires quantifying over and therefore reifying programs. Some care is required here to ensure consistency since programs may contain formulas in them. See [17] for details.

157 To talk about reflexitivity and transitivity, we can consider *Trans *to be a binary relation over configurations, where a configuration is a pair consisting of a program and a situation.

158 In this description of the PRS, we have modified the original terminology somewhat, to be more in line with contemporary usage; we have also simplified the control cycle of the PRS slightly.

159 A more detailed exposition on epistemic logic is given in Chapter 15 of this Handbook.

160 Many of these assessment systems are aimed at reducing administrative workload and are not designed to solve the standard cases and leave atypical ones for manual assessment.

161 Problem-solving methods would be called “strategy patterns” in the terminology of Gamma et al. [13].

162 At a preparation meeting for a DARPA program in this area in 1995, the rumors were that DARPA management talked about the O-word.

163 http://ontolingua.stanford.ed.

164 http://ontology.teknowledge.com/.

165 http://www.loa-cnr.it/dolce.html.

166 http://wordnet.princeton.edu/.

167 http://www.nlm.nih.gov/pubs/factsheets/umls.html.

168 http://www.cs.man.ac.uk/~rector/ontologies/simple-top-bio/.

169 http://www.geneontology.org/.

170 http://www.w3.org/2004/02/skos/.

171 http://www.getty.edu/research/conducting_research/vocabularies/.

172 Material in this section has been taken from the CommonKads book [37], the CommonKADS website at http://www.commonkads.uva.nl and the website of Epistemics, http://www.epistemics.co.uk.