Conceptual Modelling of Narrative Texts : An Example
Centre National de la Recherche Scientifique
Abstract
NKRL (Narrative Knowledge Representation Language) is a conceptual language which intends to provide a normalised, description of the semantic contents (the `meaning') of NL narrative texts. We introduce firstly the general architecture of NKRL, and we give some examples of its characteristic features. We supply then the detailed NKRL coding of a section of an actual narrative document.
Introduction
In the so-called `narrative', the main part of the information content consists in the description of `facts' or `events' which relate the real or intended behaviour of some `actors' (characters, personages, etc.) : these try to attain a specific result, experience particular situations, manipulate some (concrete or abstract) materials, deliver or receive messages, etc. Please note that, in this sort of texts, the actors or personages are not necessarily human beings ; we can have narrative documents concerning the vicissitudes in the journey of a nuclear submarine (the `actor' or the `personage') or the various avatars in the life of a commercial product.
Being able to represent the semantic content of narrative information -- i.e., its key `meaning' -- in a general, accurate, and effective way is both conceptually relevant and economically interesting. In this paper, we suggest that at least some directions for formally establishing a standard vehicle for representing the semantic content of narrative information could be found in the solutions adopted in a conceptual language like NKRL (acronym of Narrative Knowledge Representation Language), expressly created for describing (and actually exploiting) narrative documents. The representation principles used in NKRL have, in, fact, a long history of successful, concrete applications which go back to the Seventies ; recent implementations have been realised in the framework of two European projects : NOMOS, Esprit P5330, and COBALT, LRE P61011.
The lower layer of NKRL consists of a set of general representation tools which are structured into several integrated components, four in our case.
The definitional component of NKRL supplies the tools for representing the `important notions' (concepts) of a given domain ; in NKRL, a concept is, therefore, a definitional data structure associated with a symbolic label like physical_entity, human_being, taxi_ (the general class including all the taxis, not a specific cab), city_, etc. These definitional data structures are, substantially, frame-like structures ; moreover, all the NKRL concepts are inserted into a generalisation/specialisation (tangled) hierarchy that, for historical reasons, is called H_CLASS(es), and which corresponds well to the usual `ontologies' of terms.
To give rise to well-formed ontologies of NKRL concepts, the original notions must be conceived in terms of sets and subsets -- where, as usual, asserting that a pertains to the set A, (a : A) is equivalent to assert that a is characterised by the type A. Moreover, no confusion is allowed between `subsets' -- giving rise to standard H_CLASS concepts, like european_city which is a specialisation of the concept city_ -- and `instances', like paris_. Instances are still, of course, directly associated as terminal symbols with particular H_CLASS terms, but they are in the domain of another NKRL component, the enumerative component, see below. Moreover, a fundamental assumption about the organisation of H_CLASS concerns the differentiation between `notions which can be realised (instantiated) directly into enumerable specimens', like `chair' (a `physical object') and `notions which cannot be instantiated directly into specimens', like `gold' (a `substance') -- please note that a notion like `white gold' is a specialisation (subset) of gold, not an instance. The two high-level branches of H_CLASS stem, therefore, from two concepts that -- adopting the terminology used in (Guarino, Carrara and Giaretta 1994) -- we have labelled as sortal_concepts and non_sortal_concepts, see also Figure 5 in subsection 3.2 below. The specialisations of the former, like chair_, city_ or european_city, can have direct (immediate) instances (chair_27, paris_), whereas the specialisations of the latter, like gold_, or colour_, can admit further specialisations, see white_gold or red_, but do not have direct instances.
The enumerative component of NKRL concerns then the formal representation of the instances (concrete examples, see lucy_, wardrobe_1, taxi_53, paris_, not subsets) of the sortal concepts of H_CLASS. In NKRL, their formal representations take the name of `individuals'. Individuals are characterised by the fact of being countable (enumerable), of being associated with a temporal dimension (and, often, with a spatial dimension), and of possessing unique symbolic labels (lucy_, wardrobe_1, taxi_53) ; see, below, subsection 3.1 for the representation of `collections' of instances. Throughout this paper, we will use the italic type style to represent a `concept_', the roman style to represent an `individual_'.
We can note immediately that, in NKRL, the concepts and their instances (individuals) are kept conceptually distinct even if they are represented using the same data structures. The main reason for adopting this type of organisation is linked with the very different epistemological status of concepts -- which define a generic, abstract mould for some useful notions to be taken into consideration in a given domain -- with respect to the individuals, which represent specific entities occurring in the context of some concrete events of the domain. In this respect, concepts are permanent, at least in the context of a given application ; individuals represent, on the contrary, unpredictable, randomly-occurring entities stored sequentially into the system according to their order of arrival in its environment. Please note that, see before, individuals are also linked with the concept hierarchy, where they appear as the `leaves' of particular sortal concepts ; they are, therefore, indexed by the H_CLASS hierarchical structure.
The `events' proper to a given domain -- i.e., the dynamic processes describing the interactions among the concepts and individuals that play a `role' in the contest of these events -- are represented by making use of the `descriptive' and `factual' tools.
The descriptive component concerns the tools used to produce the formal representations (called predicative templates) of general classes of narrative events, like `moving a generic object', `formulate a need', `having a negative attitude towards someone', `be present somewhere'. In the context of the descriptive component, the events taken into consideration must be `structured events', i.e., they must be characterised by the explicit mention of an actor, an object, an instrument, etc. Correspondingly -- and in opposition to the binary data structures used for concepts and individuals -- predicative templates are characterised by a threefold format where the central piece is a semantic predicate, i.e., a named relation that exists among one or more arguments introduced by means of roles :
(Pi (R1 a1) (R2 a2) ... (Rn an)) ,
see the examples in subsection 3.1. Presently, the predicates pertain to the set {BEHAVE, EXIST, EXPERIENCE, MOVE, OWN, PRODUCE, RECEIVE}, and the roles to the set {SUBJ(ect), OBJ(ect), SOURCE, DEST(ination), MODAL(ity), TOPIC, CONTEXT}. Please note that, in NKRL, predicates and roles express only some functional relationships strictly necessary for the full appraisal of the `meaning' to be represented. In this sense, our roles, e.g., are more similar to the `correlators' of Ceccato's operational linguistics, see (Ceccato 1967) -- where to each correlator corresponds a set of language-independent, mental operations -- than to the (syntactically constrained) Jackendoff's thematic roles (Jackendoff 1990), which are strongly influenced by the search for some form of optimal correspondence between the semantic contents and the surface form these contents can assume in a particular natural language. Templates are structured into an inheritance hierarchy, H_TEMP(lates), which corresponds, therefore, to a `taxonomy (ontology) of events'.
The instances (`predicative occurrences') of the predicative templates, i.e., the NKRL representation of single, specific events like "Tomorrow, I will move the wardrobe", "Lucy was looking for a taxi", "Mr. Smith has fired Mr. Brown", "Peter lives in Paris" are in the domain of the factual component. The reasons for distinguishing between descriptive and factual component are identical to those already examined in a definitional/enumerative context .
The upper layer of NKRL consists of two parts. The first is a `catalogue' which gives a complete description of the formal characteristics and the modalities of use of the well-formed, `basic templates' (like `moving a generic object' mentioned above) associated with the language. Presently, the basic templates are more than 150, pertaining mainly to a (very general) socio-economico-political context where the main characters are human beings or social bodies. By means of proper specialisation operations it is then possible to obtain from the basic templates the (specific) `derived' templates that could be concretely needed to implement a particular, practical application -- e.g., `move an industrial process' -- and the corresponding occurrences -- e.g., "move, in a well-defined spatio-temporal framework, this particular industrial production". In NKRL, the set of legal, basic templates included in the catalogue can be considered, at least in a first approach, as fixed (see the `Conclusion' for the advantages of this solution). Please note, however, that the possibility (if need be) of inserting new basic elements in the catalogue is not excluded, and new templates can (carefully) be created on the model of the existing ones.
The second part of the upper layer is given by the general concepts that pertain to the upper levels of H_CLASS -- such as sortal_concepts, non_sortal_concepts, physical_entity, modality_, event_, etc., see also Figure 5 below. They are, as the upper levels of H_TEMP, invariable, i.e., they are not subjected to change when another application in a different domain is taken into account. These concepts form a sort of upper-level, invariable ontology to be compared with Bateman's `Generalized Upper Model' (Bateman, Magnini and Fabris 1995), even if our selection criteria are, once again, functional more than linguistically motivated.
2 Examples of characteristic NKRL features
2.1 Descriptive/factual components
Figure 1 supplies a simple example of factual NKRL code. It translates a fragment of fictitious, COBALT-like news story : "Yesterday, the spokesman said in a newspaper interview that his company has bought three factories abroad", which is represented according to the rules for encoding `plural situations' in NKRL, see (Zarri 1997).
Two occurrences (factual component), identified by the symbolic labels c1 and c2 and instances of basic NKRL templates, bring out the main characteristics of the event. The arguments human_being_1, company_1, newspaper_1, interview_1, purchase_1, factory_99, abroad_ are individuals ; yesterday_ is a fictitious individual introduced here, for simplicity's sake, in place of real or approximate dates, see, e.g., (Zarri 1992) for some details about the representation of temporal information in NKRL. spokesman_ and cardinality_, that pertain both to the property_ subtree of the H_CLASS hierarchy, are concepts. The attributive operator, SPECIF(ication), is one of the NKRL operators used to build up structured arguments (or `expansions'), see (Zarri and Gilardoni 1996) ; the SPECIF lists, with syntax (SPECIF e1 p1 ... pn), are used to represent some of the properties which can be asserted about the first element e1, concept or individual, of the list, e.g., human_being and chairman_ in c1. The arguments, and the templates / occurrences as a whole, may be characterised by the presence of particular codes, `determiners' or `attributes', which give further details about their significant aspects. For example, the `location attributes', represented as lists, are linked with the arguments by using the colon (`:') operator, see c2.
c1) MOVE SUBJ (SPECIF human_being_1 (SPECIF spokesman_ company_1)) OBJ #c2
OBJ (SPECIF purchase_1 (SPECIF factory_99
(SPECIF cardinality_ 3))): (abroad_)
Figure 1 - An example of NKRL code.
The non-empty HasMember slot in the data structure explicitly associated with the individual factory_99 makes it clear that this last, mentioned in c2, is referring in reality to several instances of factory_. Individuals like factory_99 are `collections' rather then `sets', given that the extensionality axiom (two sets are equal iff they have the same elements) does not hold here. In our framework, two collections, say factory_99 and factory_100, can be co-extensional, i.e., they can include exactly the same elements, without being necessarily considered as identical if created at different moments in time in the context of totally different events, see also (Franconi 1993). In Figure 1, we have supposed that the three factories were, a priori, not sufficiently important in the context of the news story to justify their explicit representation as specific individuals, e.g., factory_1, factory_2, factory_3 ; please note that, if not expressly required by the characteristics of the application, a basic NKRL principle suggests that we should try to avoid any unnecessary proliferation of individuals.
Going on now with some information about the descriptive/factual structures, we reproduce in Figure 2 the (tripartite), basic template of H_TEMP that gives rise to the occurrence c2 of Figure 1. In this figure, optional elements are in round brackets (see, e.g., the role SOURCE which introduces the possible `origin' of the situation). In the corresponding occurrences, see c2, the variables (x, y, u, etc.) are replaced by concepts or individuals according to the associated constraints, expressed as combinations of concepts pertaining to the H_CLASS upper levels.
IsA : PRODUCE3 ; `perform a task or action'
PRODUCE SUBJ x : [<location_>]
OBJ (SPECIF y e | { e }+) (SOURCE u : [<location_>])
[date-1: <date_>, `initial date' | `observed date']
[date-2: <date_>, `final date']
x = <human_being_or_social_body>
y = <purchase_> ; e = <valuable_entity>
u = <human_being_or_social_body>
v = <human_being_or_social_body>
w = <modality_> | `symbolic label of predica tive occurrence'
Figure 2 - An example of basic template.
Figure 3 is a schematic representation of the H_TEMP hierarchy, where only the branches BEHAVE, MOVE and PRODUCE have been developed to some, very limited extent. In this figure, the syntactic description of the templates is particularly sketchy : e.g., all the symbolic labels have been eliminated, and the templates are discriminated only through the associated natural language comments. For a more realistic picture of a template, see Figure 2 above. The codes `!' and `_' mean, respectively, `mandatory' and `forbidden' (e.g., in the `transmit an information to someone' template, MOVE sub hierarchy, the presence of a DEST(ination) role is expressly required). We can conclude the discussion of Figure 1 by noticing that the MOVE construction at the origin of c1 is necessarily used to translate any event concerning the transmission of an information ("the spokesman said ..."). The corresponding template makes use of what is called a `completive construction'. Accordingly, the filler of the OBJ(ect) slot in the occurrences (here, c1) which instantiates the transmission template is always a symbolic label (#c2) which refers to another predicative occurrence, i.e., the occurrence bearing the information content to be spread out ("the company has bought three factories abroad ..."), see also Figure 3.
As a second example of descriptive/factual structures, we give now, Figure 4, an NKRL interpretation of the narrative sentence : "We have to make orange juice" which, according to Hwang and Schubert (Hwang and Schubert 1993: 1298), exemplifies several interesting semantic phenomena. To translate the general idea of `acting in order to obtain a given result', we use :
i) A (predicative) occurrence (c3 in Figure 4), instance of a basic template pertaining to the BEHAVE branch of H_TEMP, and corresponding to the general meaning of "focusing on a result". This occurrence is used to express the "acting" component, i.e., it allows us to identify the SUBJ of the action, the temporal co-ordinates, possibly the MODAL(ity) or the instigator (SOURCE), etc.
ii) A second predicative occurrence, c4 in Figure 4, which contains a different NKRL predicate (e.g., PRODUCE in Figure 4) and which is used to express the `intended result' component. Please note that, for all the `focus on...' constructions, the reference time is the timestamp (`observed date' in Figure 4) linked with the first occurrence (BEHAVE). The second occurrence, which happens `in the future', is necessarily marked as hypothetical, i.e., it is always characterised by the presence of an `uncertainty validity attribute', code `*'.
iii) A `binding occurrence', c5, linking together the previous predicative occurrences and labelled with GOAL, an operator pertaining to the `taxonomy of causality' of NKRL. Binding structures, templates or occurrences -- i.e., lists where the elements are symbolic labels, c3 and c4 in Figure 4 -- are second-order structures used to represent the logico-semantic links which can exist between (predicative) templates or occurrences.
H_TEMP (hierarchy of predicative templates)
.. BEHAVE templates ; predicate : BEHAVE
... `external manifestation of the subject' ; _OBJ
.... `acting in a particular role' ; MODAL < role_ [ex: rugby_player] >
.... `manifest a particular quality' ; MODAL < quality_ >
... `focus on a result' ; _ OBJ, DEST, TOPIC ; _modulator `against, for' ; !GOAL binding structure
.... `act explicitly to obtain the result' ; _modulator `ment'
.... `wishes and intentions' ; ! modulator `ment'
... `concrete attitude toward someone/something' ; !OBJ, MODAL ; _DEST, GOAL ; _mod. `ment'
.. EXIST templates ; predicate : EXIST ; !location of the SUBJ ; _DEST
.. EXPERIENCE templates ; predicate : EXPERIENCE ; !OBJ ; _DEST
.. MOVE templates ; predicate : MOVE
... `moving a generic entity' ; OBJ < entity_>
.... `move a material thing' ; OBJ < physical_entity >
..... `change the position of something"' ; _DEST (ex: "move the wardrobe")
..... `transfer something to someone"' ; !DEST (ex: "send a letter to Lucy")
... `generic person displacement' ; SUBJ = OBJ = <human_being> ; !location of SUBJ, OBJ
... `transmit an information to someone"' ; !DEST
.... `transmit a generic information' ; OBJ < type_of_information [ex: message_] >
.... `transmit a structured information' ; OBJ `label of binding/predicative occurrence'
.. OWN templates ; predicate : OWN ; !OBJ ; _DEST
.. PRODUCE templates ; predicate : PRODUCE ; !OBJ
... `conceive a plan or idea' ; !modulator `ment'
... `creation of material things' ; OBJ < physical_entity > ; _modulator `ment'
... `perform a task or action"' ; OBJ < action_name >
.... `acquire, buy' ; OBJ < purchase_ >
... `relation involvement' ; SUBJ (COORD) ; OBJ mutual_agreement MODAL <relationship_>
... `production of events by an active cause' ; SUBJ < active_cause > ; OBJ < event_>
.. RECEIVE templates ; predicate : RECEIVE ; !OBJ ; _DEST
Figure 3 - Schematic H_TEMP hierarchy (ontology of events).
The general schema for representing the `focusing on an intended result' domain in NKRL, see also Figure 3, is then :
ca) BEHAVE SUBJ <human_being_or
cb) * <predicative_occurrence>, with any syntax
In Figure 4, `oblig' and `ment' are `modulators', see (Zarri 1995), i.e., particular determiners used to refine or modify the primary interpretation of a template or occurrence as given by the basic `predicate -- roles -- argument' association. `Ment(al)' pertains to the `modality' modulators. `Oblig(atory)' suggests that "someone is obliged to do or to endure something, e.g., by authority", and pertains to the `deontic modulators' series. Other modulators are the `temporal modulators', `begin', `end', `obs(erve)', see again (Zarri 1992). Modulators work as global operators which take as their argument the whole (predicative) template or occurrence. When a list of modulators is present, as in the occurrence c3 of Figure 4, they apply successively to the template/occurrence in a polish notation style to avoid any possibility of scope ambiguity. In the constructions for expressing `focus on ...', see Figure 3, the absence of the `ment(al)' modulator in the BEHAVE occurrence means that the SUBJ(ect) of BEHAVE takes some concrete initiative (acts explicitly) in order to fulfil the result ; if `ment' is present, as in Figure 4, no concrete action is undertaken, and the `result' reflects only the wishes and desires of the SUBJ(ect). several_, see c3 and c4, is used conventionally in NKRL to represent the cardinality of sets of totally undefined size, like those corresponding to a generic plural referent, as in `men' or `books', see (Zarri and Gilardoni 1996). In c4, `i' is conventionally used to denote an unknown time increment with respect to the timestamp that characterises c3.
c3) BEHAVE SUBJ (SPECIF informant_1
human_being (SPECIF cardinality_ several_))
c4) * PRODUCE SUBJ (SPECIF informant_1 human_being
(SPECIF cardinality_ several_)))
OBJ (SPECIF orange_juice (SPECIF amount_)) date1: observed date + i
Figure 4 - Representation of `wishes and intentions'.
2.2 Definitional/enumerative com- ponents
Figure 5 gives a very sketchy representation of the upper level of H_CLASS (hierarchy of concepts, definitional component). Note, in this figure, the presence of a sortal concept event_. As already stated, the events represented by means of templates and occurrences must be structured events, i.e., they must be characterised by the specific indication of an actor, an object, etc. If we only want to explicitly mention an event (e.g., president_clinton_interview_49), without giving all its structural details, we will simply make use of a particular individual, i.e., of an instance of event_. substance_ and colour_ are regarded as examples of non sortal concepts. For their generic terms, pseudo_sortal_concepts and characterising_concepts, we have adopted again the terminology of (Guarino, Carrara and Giaretta 1994).
The data structures used for concepts and individuals are essentially frame-based structures, and their design is relatively traditional. These structures are composed of an OID (the generic symbolic label of the concept to be defined, or the specific name of the individual to be created), and of a set of characteristic features (slots). Three different types of slots are used, `relations', `attributes', and `procedures', see, e.g., (Zarri 1995a, Zarri 1997) for some details.
3 A detailed example of NKRL representation
Consider the following section of the Maria Otero's text titled "Latin America : Accion Speaks Louder then Words" and proposed for the Interlinguas Workshop :
Banco Solidario, S.A., or Bancosol, grew out of a non-profit joint venture created in 1986 by prominent members of the Bolivian business community and ACCION International. The latter brought with them leadership and seed capital, while the former provided technology and methodology. PRODEM, as the programme was named, provided credit and training to broaden employment opportunities for the very poor self-employed, encourage investment in microbusiness, and increase the income generated by this sector. PRODEM used the group lending technique of `solidarity groups' and began making small working capital loans. In its first five years of operation, PRODEM financed loans to over 13,300 microentrepreneurs, 77 per cent of whom were women, disbursing over $27 million in loans averaging $273
Two general remarks apply here :
· All the occurrences of the translation are instances of standard `basic templates' pertaining to the NKRL `catalogue', see above. No new template has been created specially for the purpose in hand.
· The concepts (and their instances) used in the translation are, in their great majority, standard H_CLASS concepts, common to many possible applications. A few ad hoc concepts, like group_lending_technique or working_capital have been created, however, to keep the translation as much as simple and comprehensible as possible. Note that, as already remarked above, when the arguments of the semantic predicates represent general properties, or when there is no precise need for creating specific individuals to be used again, e.g., in other occurrences (coreference), the arguments are represented as generic concepts instead of individuals.
Figure 5 - An abridged view of the `upper level' of H_CLASS .
The coding of the textual fragment examined includes two main elements : the creation of Bancosol from the PRODEM non-profit joint venture (2), and the description of the characteristics and aims of PRODEM (3). REFER is the NKRL operator used to represent a `weak causality' relationship : 3 is necessary, but not sufficient, to explain 2, see, e.g., (Zarri 1992).
2) EXIST SUBJ bancosol_: (bolivia_)
SOURCE (SPECIF prodem_ (SPECIF
EXIST + [abs, begin] is the NKRL formula (template) used to express the origin (the `birth') of living beings or social bodies.
The description of PRODEM's characteristics concerns the details of its creation (4), the aims (5), and the description of the first five years of operation (6).
7) PRODUCE SUBJ (COORD1 (SPECIF busi- nessman_1 bolivian_ important_ (SPECIF cardinality_ several_)) accion_international): (bolivia_)
OBJ (SPECIF prodem_ (SPECIF joint_venture non_profit))
8) MOVE SUBJ (SPECIF businessman_1 boliv- ian_ important_ (SPECIF cardinality_ several_))
OBJ (COORD1 technology_ method- ology_
The Bolivian businessmen provided PRODEM with technology and methodology.
9) MOVE SUBJ accor_international
OBJ (COORD1 leadership_ ini- tial_capital)
PRODEM's aim was to provide credit and training for the very poor self-employed etc. (10) ; to do this, PRODEM used the group lending technique ... (11).
10) (GOAL 12 *(COORD 13 14 15))
The provision of credit and training (12) was intended to broaden the employment opportunities (13), to encourage investment (13), and to increase the income (14).
OBJ (COORD1 credit_ training_)
(SPECIF cardinality_ sev- eral_))
13) EXPERIENCE SUBJ (SPECIF self_employed_1 (SPECIF poor_ very_) (SPECIF cardinality_ several_))
OBJ (SPECIF employ- ment_opportunity
The EXPERIENCE templates are typically used to translate the fact of `being in a given state', `be affected b y positive/negative events', etc.
OBJ (SPECIF investment_ mi- crobusiness_)
The above NKRL structure is typically used to express a concrete, favourable attitude to someone or something, see also Figure 3 above.
OBJ (SPECIF expansion_ (SPECIF income_ mi- crobusiness_)
To fulfil its aims, PRODEM used a group technique (16) and began making small loans (17).
MODAL (SPECIF user_ (SPECIF group_lending_technique solidarity_groups))
The BEHAVE template corresponding to this occurrence is used to represent the idea of `acting in a particular role' (here, role = user_ ), see again Figure 3 above.
OBJ (SPECIF loan_ work- ing_capital small_ (SPECIF cardi- nality_ several_))
OBJ (SPECIF loan_1 dollar_usa (SPECIF cardinality_ sev- eral_) (SPECIF global_amount (SPECIF million_27)) (SPECIF
DEST (SPECIF microentre- preneur_1 (SPECIF cardi- nality_ (SPECIF more_than 13,300)) (SPECIF subset_ microentrepreneur_1 (SPECIF cardinality_ (SPECIF percent_ 77 micro entrepreneur_1)) female_))
The DEST(ination) role is sometimes called `benefactive role' in the literature.
Several (at least partly) implemented formalisms exist that claim to be able to represent extensive chunks of natural language semantics -- see Conceptual Graphs (Sowa 1984), SNePS (Shapiro and Rapaport 1992), CYC (Lenat and Guha 1990), LILOG (Herzog and Rollinger 1991), Episodic Logic (Hwang and Schubert 1993), etc. Moreover, since Schank's work (Schank 1973), several authors have tackled the specific problem of representing structured events see, e.g., the recent Park's proposal (Park 1995), where a set of primitives for modelling the dynamic aspects (`events') of a domain is provided.
In defence of the introduction of the NKRL formalism, we can however put forward (at least) the two following arguments :
· Making use of a relatively simple, intuitive and easily manageable formalism, NKRL offers some interesting solutions to very hard problems concerning the `practical' aspects of the knowledge representation endeavour, like modalities, temporal representation, representation of the wishes and intention domain, completive constructions, coding of second order logical structures, etc.
· A fundamental (and apparently unique) characteristic of NKRL is represented by the fact that the catalogue of basic templates, see subsection 3.1 above and Figure 3, can be considered as part and parcel of the definition of the language. This approach is particularly important for practical applications, and it implies, in particular, that : i) a system-builder does not have to create himself the structural knowledge needed to describe
the events proper to a (sufficiently) large class of narrative texts and documents ; ii) it becomes easier to secure the reproduction or the sharing of previous results.
Bateman, J. A., Magnini, B., and Fabris, G.: The Generalized Upper Model Knowledge Base : Organization and Use, in Towards Very Large Knowledge Bases - Knowledge Building and Knowledge Sharing. Amsterdam: IOS Press, 1995.
Ceccato, S.: Concepts for a New Systematics, Inf. Storage and Retriev. 3 (1967) 193-214.
Franconi, E.: A Treatment of Plurals and Plural Quantifications Based on a Theory of Collections, Minds and Machines 3 (1993) 453-474.
Gruber, T.R.: A Translation Approach to Portable Ontology Specifications, Knowledge Acquisition 5 (1993) 199-220.
Guarino, N., Carrara, M., and Giaretta, P.: An Ontology of Meta-Level Categories, in Proc. of the 4th Int. Conf. on Principles of Knowledge Repres. and Reasoning. San Francisco: Morgan Kaufmann, 1994.
Herzog, O., and Rollinger, C.-R., eds.: Text Understanding in LILOG - Integrating Computational Linguistics and Artificial Intelligence. Berlin: Springer-Verlag, 1991.
Hwang, C.H., and Schubert, L.K.: Meeting the Interlocking Needs of LF-Computation, Deindexing and Inference : An Organic Approach to General NLU, in Proc. of the 13th Int. Joint Conf. on Art. Intelligence. San Francsico: Morgan Kaufmann, 1993.
Jackendoff, R.: Semantic Structures. Cambridge (MA): The MIT Press, 1990.
Lenat, D.B., and Guha, R.V.: Building Large Knowledge Based Systems. Reading (MA): Addison-Wesley, 1990.
Park, B.J.: A Language for Ontologies Based on Objects and Events, in Proc. of the IJCAI'95 Workshop on Basic Ontological Issues in Knowledge Sharing, 1995.
Schank, R.C.: Identification of Conceptualizations Underlying Natural Language, in Computer Models of Thought and Language. San Francisco: W.H. Freeman, 1973.
Schank, R.C., and Abelson, R.P.: Scripts, Plans, Goals and Understanding. Hillsdale (NJ): Lawrence Erlbaum, 1977.
Shapiro, S.C., and Rapaport, W.J.: The SNePS Family, in Semantic Networks in Artificial Intelligence. Oxford: Pergamon Press.
Sowa, J.F.: Conceptual Structures : Information Processing in Mind and Machine. Reading (MA): Addison-Wesley, 1984.
Zarri, G.P.: Encoding the Temporal Characteristics of the Natural Language Descriptions of (Legal) Situations, in Expert Systems in Law. Amsterdam: Elsevier Science, 1992.
Zarri, G.P.: Representing and Querying Complex Conceptual Structures in the Framework of NKRL, in Suppl. Proc. of the 3rd Int. Conf. on Conceptual Structures. Berlin: Springer-Verlag, 1995
Zarri, G.P., and Gilardoni, L.: Structuring and Retrieval of the Complex Predicate Arguments Proper to the NKRL Conceptual Language, in Foundations of Intelligent Systems, ISMIS'96. Berlin: Springer-Verlag.
Zarri, G.P.: The `Descriptive' Component of a Hybrid Knowledge Representation Language, in Semantic Networks in Artificial Intelligence, Lehmann, F., ed. Oxford: Pergamon Press, 1992.
Zarri, G.P.: NKRL, a Knowledge Representation Tool for Encoding the `Meaning' of Complex Narrative Texts, Natural Language Engineering -- Special Issue on Knowledge Representation for Natural Language Processing in Implemented Systems, 3, 231-253.