We describe the OntoNotes methodology and its result, a large multilingual richly-annotated corpus constructed at 90% interannotator agreement. An initial portion (300K words of English newswire and 250K words of Chinese newswire) will be made available to the community during 2007.

OntoNotes: The 90% Solution

Information Content (IC) is an important dimension of word knowledge when assessing the similarity of two terms or word senses. The conventional way of measuring the IC of word senses is to combine knowledge of their hierarchical structure from an ontology like WordNet with statistics on their actual usage in text as derived from a large corpus. In this paper we present a wholly intrinsic measure of IC that relies on hierarchical structure alone. We report that this measure is consequently easier to calculate, yet when used as the basis of a similarity mechanism it yields judgments that correlate more closely with human assessments than other, extrinsic measures of IC that additionally employ corpus analysis.

/pdf/an-intrinsic-information-content-metric-for-semantic-4ssbcjw41z.pdf

An intrinsic information content metric for semantic similarity in WordNet

Individuals: An Essay in Descriptive Metaphysics

We present a method and a tool, OntoLearn, aimed at the extraction of domain ontologies from Web sites, and more generally from documents shared among the members of virtual organizations. OntoLearn first extracts a domain terminology from available documents. Then, complex domain terms are semantically interpreted and arranged in a hierarchical fashion. Finally, a general-purpose ontology, WordNet, is trimmed and enriched with the detected domain concepts. The major novel aspect of this approach is semantic interpretation, that is, the association of a complex concept with a complex term . This involves finding the appropriate WordNet concept for each word of a terminological string and the appropriate conceptual relations that hold among the concept components. Semantic interpretation is based on a new word sense disambiguation algorithm, called structural semantic interconnections.

/pdf/learning-domain-ontologies-from-document-warehouses-and-2ks0aqv50o.pdf

Learning Domain Ontologies from Document Warehouses and Dedicated Web Sites

This paper describes the organization of extracted risk-relevant data in a relational database created at Regulatory Data Corporation for use by security professionals. The initial effort involved creating sets of data-extraction variables around a set of risk relevant keywords. The keywords clustered around events rather than entities and the data extraction variables that were developed centered on semantic roles of event participants. To facilitate future data extraction efforts in this genre, we organized events, participants, keywords and grammatical forms into an ontology. This ontology of risk-relevance (henceforth known as the ORR) is a tool for both data extraction professionals and risk-assessment professionals that allows new entries to be added easily when the need for additional information arises.

Formal Ontology in Information Systems

In this paper we introduce the DOLCE upper level ontology, the first module of a Foundational Ontologies Library being developed within the WonderWeb project. DOLCE is presented here in an intuitive way; the reader should refer to the project deliverable for a detailed axiomatization. A comparison with WordNet's top-level taxonomy of nouns is also provided, which shows how DOLCE, used in addition to the OntoClean methodology, helps isolating and understanding some major WordNet?s semantic limitations. We suggest that such analysis could hopefully lead to an ?ontologically sweetened? WordNet, meant to be conceptually more rigorous, cognitively transparent, and efficiently exploitable in several applications.

/pdf/sweetening-ontologies-with-dolce-35idxeexkf.pdf

Sweetening Ontologies with DOLCE

entities The main characteristic of abstract entities is that they do not have spatial nor temporal qualities, and they are not qualities themselves. The only class of abstract entities we consider in the present version of DOLCE is that of quality regions (or simply regions). Quality spaces are special kinds of quality regions, being mereological sums of all the regions related to a certain quality type. The other examples of abstract entities reported in Figure 2 (sets and facts) are only indicative. 2.3 Basic functions and relations According to the general methodology introduced in [Gangemi et al. 2001], before discussing the DOLCE backbone properties, we have first to introduce a set of basic primitive relations, suitable to characterize our ontological commitments as neutrally as possible. We believe that these relations should be, as much as possible, • general enough to be applied to multiple domains; • such that they do not rest on questionable ontological assumptions about the ontological nature of their arguments; • sufficiently intuitive and well studied in the philosophical literature; • hold as soon as their relata are given, without mediating additional entities. In the past, we adopted the term formal relation (as opposite to material relation) for a relation that can be applied to all possible domains. Recently, however, [Degen et al. 2001] proposed a different notion of formal relation: “A relation is formal if it holds as soon as its relata are given. Formal relations are called equivalently immediate relations, since they hold of their relata without mediating additional individuals”. The notion of basic primitive relation proposed above combines together the two notions. Roughly, a basic primitive relation is an immediate relation that spans multiple application domains. The axioms constraining the arguments of primitive relations and functions are reported in Table 3, and summarized in Figure 4. 1 The notion of ‘immediate relation’ seems to be equivalent to what Johansson called ground relation [Johansson 1989]. According to Johansson, a ground relation “is derivable from its relata”. We understand that the very existence of the arguments is sufficient to conclude whether the relation holds or not. This notion seems also equivalent to that of “internal relation”. IST Project 2001-33052 WonderWeb: Ontology Infrastructure for the Semantic Web 14 Parthood: “x is part of y” P(x, y) → (AB(x) ∨ PD(x)) ∧ (AB(y) ∨ PD(y)) Temporary Parthood: “x is part of y during t” P(x, y, t) → (ED(x) ∧ ED(y) ∧ T(t)) Constitution: “x constitutes y during t” K(x, y, t) → ((ED(x) ∨ PD(x)) ∧ (ED(y) ∨ PD(y)) ∧ T(t)) Participation: “x participates in y during t” PC(x, y, t) → (ED(x) ∧ PD(y) ∧ T(t)) Quality: “x is a quality of y” qt(x, y) → (Q(x) ∧ (Q(y) ∨ ED(y) ∨ PD(y))) Quale: “x is the quale of y (during t)” ql(x, y) → (TR(x) ∧ TQ(y)) ql(x, y, t) → ((PR(x) ∨ AR(x)) ∧ (PQ(y) ∨ AQ(y)) ∧ T(t)) Table 3. Basic axioms on argument restrictions of primitives. Parthood and Temporary Parthood The endurants/perdurants distinction introduced in the previous section provides evidence for the general necessity of having two kinds of parthood relations: a-temporal and timeindexed parthood. The latter will hold for endurants, since for them it is necessary t o know when a specific parthood relationship holds. Consider for instance the classical example of Tibbles the cat [Simons 1987]: Tail is part of Tibbles before the cut but not after it. Formally, we can write P(Tail, Tibbles, before(cut)) and ¬P(Tail, Tibbles, after(cut)). Atemporal parthood, on the other hand, will be used for entities which do not properly change in time (occurrences and abstracts). In the present version, parthood will not be defined for qualities. With respect to time-indexed parthood, two useful notions can be defined. We shall say that an endurant is mereologically constant iff all its parts remains the same during its life, and mereologically invariant iff they remain the same across all possible worlds. For example, we usually take ordinary material objects as mereologically variable, because during their life they can lose or gain parts. On the other hand, amounts of matter are taken as mereologically invariant (all their parts are essential parts). Dependence and Spatial Dependence There are basically two approaches to characterizing the notion of ontological dependence: • non-modal accounts (cf. [Fine and Smith 1983] and [Simons 1987] , pp. 310-318) • modal accounts (cf. [Simons 1987]). Non-modal approaches treat the dependence relation as a quasi-mereological primitive whose formal properties are characterized by axioms. However, as Simons has justly observed, such axiomatizations cannot rule out non-intended interpretations that are purely topological in nature. The only way to save them is actually to link them with modal accounts. In a modal approach, dependence of an entity x on an entity y might be defined as follows: x depends on y iff, necessarily, y is present whenever x is present. Such a definition seems to be in harmony both with commonsense intuition as well as philosophical tradition (Aristotle, Husserl), despite the fact that there are some cases where, as Kit Fine has shown, this characterization is vacuous. Indeed, according to the definition, everything is trivially dependent on necessarily existing or always present IST Project 2001-33052 WonderWeb: Ontology Infrastructure for the Semantic Web 15 objects. However, Simons has shown that it is possible to exclude such vacuous examples and while this move might be philosophically dubious, it makes perfect sense in an engineering approach to ontologies of everyday contingent objects. Our concept of dependence involves the notion of presence in time as well as modality. We mainly use two variants of dependence, adapted from [Thomasson 1999]: specific and generic constant dependence. The former is defined both for particulars and properties, while the latter only for properties. A particular x is specifically constantly dependent on another particular y iff, at any time t, x can't be present at t unless y is also present at t. For example, a person might be specifically constantly dependent on its brain. This notion is naturally extended to properties by defining that a property φ is specifically constantly dependent on a property ψ iff every φer is specifically constantly dependent on a ψer. A property φ is generically constantly dependent on a property ψ iff, for any instance x of φ, at any time t, x can't be present at t, unless a certain instance y of ψ is also present at t. For example, a person might be generically constantly dependent on having a heart. We define spatial dependence as a particular kind of dependence which is grounded not only in time (presence), but also in space. The definitions are as above with the further requirement that y has to be spatially co-localised with x in addition of being co-present. This notion is defined both for endurants and perdurants. Constitution Constitution has been extensively discussed in the philosophical literature: • Doepke (cit. in [Simons 1987] p.238) “x constitutes y at time t iff x could be a substratum of y’s destruction.” • Simons (cit. in [Simons 1987] p.239) “When x constitutes y, there are certain properties of x which are accidental to x, but essential to y. (...) Where the essential properties concern the type and disposition of parts, this is often a case of composition, but in other cases, such as that of body/person, it is not.” Constitution is not Identity – Consider the following classical example. I buy a portion of clay (LUMPL) at 9am. At 2pm I made a statue (GOLIATH) out of LUMPL and I put GOLIATH on a table. At 3pm I replace the left hand of GOLIATH with a new one and I throw the old hand in the dustbin. There are three reasons to support the claim that LUMPL is not GOLIATH: (i) Difference in histories LUMPL is present a 9am, but GOLIATH is not [Thomson 1998] (ii) Difference in persistence conditions At 3pm GOLIATH is wholly present on the table, but LUMPL is not wholly present on the table (a statue can undergo replacements of certain parts, but not an amount (portion) of matter, i.e. all parts of LUMPL are essential but not all parts of GOLIATH are essential [Thomson 1998]. LUMPL can survive a change of shape, GOLIATH not. (iii) Difference in essential relational properties It is metaphysically possible for LUMPL, but not for GOLIATH, to exist in the absence of an artworld or an artist or anybody's intentions [Baker 2000]. Participation The usual intuition about participation is that there are endurants “involved” in an occurrence. Linguistics has extensively investigated the relation between occurrences and their participants in order to classify verbs and verbal expressions. Fillmore's Case Grammar [Fillmore 1984] and its developments (Construction Grammar, FrameNet) is one of the best attempts at building a systematic model of language-oriented participants. On the other hand, the first systematic investigation goes back at least to Aristotle, that defined four “causes” (aitiai), expressing the initiator, the destination, the instrument, and IST Project 2001-33052 WonderWeb: Ontology Infrastructure for the Semantic Web 16 the substrate or host of an event. Sowa further specified subsets of aitiai on the basis of properties borrowed from linguistics (cfr. [Sowa 1999]). In an ontology based on a strict distinction between endurants and perdurants, participation cannot be simply parthood; the participating endurants are not parts of the occurrences: only occurrences can be parts of other occurrences. Moreover, the primitive participation we introduce is time-indexed, in order to account for the varieties of participation in time (temporary participation, constant participation). Quality inherence and quality value Finally, three primitive relations are introduced in order to account for qualities: a generalize

WonderWeb Deliverable D17. The WonderWeb Library of Foundational Ontologies and the DOLCE ontology

entities The main characteristic of abstract entities is that they do not have spatial nor temporal qualities, and they are not qualities themselves. The only class of abstract entities we consider in the present version of DOLCE is that of quality regions (or simply regions). Quality spaces are special kinds of quality regions, being mereological sums of all the regions related to a certain quality type. The other examples of abstract entities reported in Figure 2 (sets and facts) are only indicative. 2.3 Basic functions and relations According to the general methodology introduced in [Gangemi et al. 2001], before discussing the DOLCE backbone properties, we have first to introduce a set of basic primitive relations, suitable to characterize our ontological commitments as neutrally as possible. We believe that these relations should be, as much as possible, • general enough to be applied to multiple domains; • such that they do not rest on questionable ontological assumptions about the ontological nature of their arguments; • sufficiently intuitive and well studied in the philosophical literature; • hold as soon as their relata are given, without mediating additional entities. In the past, we adopted the term formal relation (as opposite to material relation) for a relation that can be applied to all possible domains. Recently, however, [Degen et al. 2001] proposed a different notion of formal relation: “A relation is formal if it holds as soon as its relata are given. Formal relations are called equivalently immediate relations, since they hold of their relata without mediating additional individuals”. The notion of basic primitive relation proposed above combines together the two notions. Roughly, a basic primitive relation is an immediate relation that spans multiple application domains. The axioms constraining the arguments of primitive relations and functions are reported in Table 3, and summarized in Figure 4. 1 The notion of ‘immediate relation’ seems to be equivalent to what Johansson called ground relation [Johansson 1989]. According to Johansson, a ground relation “is derivable from its relata”. We understand that the very existence of the arguments is sufficient to conclude whether the relation holds or not. This notion seems also equivalent to that of “internal relation”. IST Project 2001-33052 WonderWeb: Ontology Infrastructure for the Semantic Web 14 Parthood: “x is part of y” P(x, y) → (AB(x) ∨ PD(x)) ∧ (AB(y) ∨ PD(y)) Temporary Parthood: “x is part of y during t” P(x, y, t) → (ED(x) ∧ ED(y) ∧ T(t)) Constitution: “x constitutes y during t” K(x, y, t) → ((ED(x) ∨ PD(x)) ∧ (ED(y) ∨ PD(y)) ∧ T(t)) Participation: “x participates in y during t” PC(x, y, t) → (ED(x) ∧ PD(y) ∧ T(t)) Quality: “x is a quality of y” qt(x, y) → (Q(x) ∧ (Q(y) ∨ ED(y) ∨ PD(y))) Quale: “x is the quale of y (during t)” ql(x, y) → (TR(x) ∧ TQ(y)) ql(x, y, t) → ((PR(x) ∨ AR(x)) ∧ (PQ(y) ∨ AQ(y)) ∧ T(t)) Table 3. Basic axioms on argument restrictions of primitives. Parthood and Temporary Parthood The endurants/perdurants distinction introduced in the previous section provides evidence for the general necessity of having two kinds of parthood relations: a-temporal and time-indexed parthood. The latter will hold for endurants, since for them it is necessary to know when a specific parthood relationship holds. Consider for instance the classical example of Tibbles the cat [Simons 1987]: Tail is part of Tibbles before the cut but not after it. Formally, we can write P(Tail, Tibbles, before(cut)) and ¬P(Tail, Tibbles, after(cut)). Atemporal parthood, on the other hand, will be used for entities which do not properly change in time (occurrences and abstracts). In the present version, parthood will not be defined for qualities. With respect to time-indexed parthood, two useful notions can be defined. We shall say that an endurant is mereologically constant iff all its parts remains the same during its life, and mereologically invariant iff they remain the same across all possible worlds. For example, we usually take ordinary material objects as mereologically variable, because during their life they can lose or gain parts. On the other hand, amounts of matter are taken as mereologically invariant (all their parts are essential parts). Dependence and Spatial Dependence There are basically two approaches to characterizing the notion of ontological dependence: • non-modal accounts (cf. [Fine and Smith 1983] and [Simons 1987] , pp. 310-318) • modal accounts (cf. [Simons 1987]). Non-modal approaches treat the dependence relation as a quasi-mereological primitive whose formal properties are characterized by axioms. However, as Simons has justly observed, such axiomatizations cannot rule out non-intended interpretations that are purely topological in nature. The only way to save them is actually to link them with modal accounts. In a modal approach, dependence of an entity x on an entity y might be defined as follows: x depends on y iff, necessarily, y is present whenever x is present. Such a definition seems to be in harmony both with commonsense intuition as well as philosophical tradition (Aristotle, Husserl), despite the fact that there are some cases where, as Kit Fine has shown, this characterization is vacuous. Indeed, according to the definition, everything is trivially dependent on necessarily existing or always present objects. However, Simons has shown that it is possible to exclude such vacuous examples and while this move might be philosophically dubious, it makes perfect sense in an engineering approach to ontologies of everyday IST Project 2001-33052 WonderWeb: Ontology Infrastructure for the Semantic Web 15 contingent objects. Our concept of dependence involves the notion of presence in time as well as modality. We mainly use two variants of dependence, adapted from [Thomasson 1999]: specific and generic constant dependence. The former is defined both for particulars and properties, while the latter only for properties. A particular x is specifically constantly dependent on another particular y iff, at any time t, x can't be present at t unless y is also present at t. For example, a person might be specifically constantly dependent on its brain. This notion is naturally extended to properties by defining that a property φ is specifically constantly dependent on a property ψ iff every φer is specifically constantly dependent on a ψ . A property φ is generically constantly dependent on a property ψ iff, for any instance x of φ, at any time t, x can't be present at t, unless a certain instance y of ψ is also present at t. For example, a person might be generically constantly dependent on having a heart. We define spatial dependence as a particular kind of dependence which is grounded not only in time (presence), but also in space. The definitions are as above with the further requirement that y has to be spatially co-localised with x in addition of being co-present. This notion is defined both for endurants and perdurants. Constitution Constitution has been extensively discussed in the philosophical literature: • Doepke (cit. in [Simons 1987] p.238) “x constitutes y at time t iff x could be a substratum of y’s destruction.” • Simons (cit. in [Simons 1987] p.239) “When x constitutes y, there are certain properties of x which are accidental to x, but essential to y. (...) Where the essential properties concern the type and disposition of parts, this is often a case of composition, but in other cases, such as that of body/person, it is not.” Constitution is not Identity – Consider the following classical example. I buy a portion of clay (LUMPL) at 9am. At 2pm I made a statue (GOLIATH) out of LUMPL and I put GOLIATH on a table. At 3pm I replace the left hand of GOLIATH with a new one and I throw the old hand in the dustbin. There are three reasons to support the claim that LUMPL is not GOLIATH: (i) Difference in histories LUMPL is present a 9am, but GOLIATH is not [Thomson 1998] (ii) Difference in persistence conditions At 3pm GOLIATH is wholly present on the table, but LUMPL is not wholly present on the table (a statue can undergo replacements of certain parts, but not an amount (portion) of matter, i.e. all parts of LUMPL are essential but not all parts of GOLIATH are essential [Thomson 1998]. LUMPL can survive a change of shape, GOLIATH not. (iii) Difference in essential relational properties It is metaphysically possible for LUMPL, but not for GOLIATH, to exist in the absence of an artworld or an artist or anybody's intentions [Baker 2000]. Participation The usual intuition about participation is that there are endurants “involved” in an occurrence. Linguistics has extensively investigated the relation between occurrences and their participants in order to classify verbs and verbal expressions. Fillmore's Case Grammar [Fillmore 1984] and its developments (Construction Grammar, FrameNet) is one of the best attempts at building a systematic model of language-oriented participants. On the other hand, the first systematic investigation goes back at least to Aristotle, that defined four “causes” (aitiai), expressing the initiator, the destination, the instrument, and the substrate or host of an event. Sowa further specified subsets of aitiai on the basis of properties borrowed from linguistics (cfr. [Sowa 1999]). In an ontology based on a strict distinction between endurants and perdurants, participation cannot be simply parthood; the participating endurants are not parts of the occurrences: only occurrences can be parts of other occurrences. Moreover, the primitive participation we IST Project 2001-33052 WonderWeb: Ontology Infrastructure for the Semantic Web 16 introduce is time-indexed, in order to account for the varieties of participation in time (temporary participation, constant participation). Quality inherence and quality value Finally, three primitive relations are introduced in order to account for qualities: a generalized

/pdf/the-wonderweb-library-of-foundational-ontologies-preliminary-256ymr46tm.pdf

The WonderWeb Library of Foundational Ontologies Preliminary Report

In the last decade, ontologies have become a hot issue in the conceptual modeling community, which is chiefly due to the emergence of new application areas such as Electronic Commerce and the Semantic Web. The implementation of large-scale information systems, as well as their semantic interoperability, seem to be only tractable on the basis of ontologies as abtract domain models.

Ontology-Driven Conceptual Modelling

Foundational ontologies are axiomatic theories about high-level domain-independent categories of the real world. They constitute toolboxes of eminently reusable information modelling primitives for building application ontologies about specific domains. The design of foundational ontologies confronts the conceptual modeller with completely new challenges in respect of their content as well as their formalisation. As a case study, this article outlines the axiomatisation of the Object-Centered High-level REference ontology OCHRE, comparing its basic conceptual modelling choices with those underlying DOLCE, a foundational ontology developed in the Wonderweb project.

Luc Schneider

Papers

Sweetening Ontologies with DOLCE

WonderWeb Deliverable D17. The WonderWeb Library of Foundational Ontologies and the DOLCE ontology

The WonderWeb Library of Foundational Ontologies Preliminary Report

Ontology-Driven Conceptual Modelling

Designing foundational ontologies: The Object-Centered high-level reference ontology ochre as a case study