77 Natural-Language Semantics for Associations Joerg Evermann School of Information Management, Victoria University, Wellington, New Zealand joerg.evermann@mcs.vuw.ac.nz 1 Introduction Conceptual models describe an application domain to further communication and understanding, and serve as the basis for subsequent software design and implementation. For a language to be used for conceptual modelling, the seman- tics of its constructs must be well-defined w.r.t. the application domain. The semantics of the association construct1 , central to object-oriented mod- elling languages, are problematic from the software perspective [1, 2], as well as in conceptual modelling. The definitions in the literature often obscure, rather than clarify the meaning of the construct. Prior research interpreted associations ontologically as mutual properties [3], and classified them according to linguistic and cognitive considerations [4]. The ontological interpretation confounds properties and interaction, while the latter does not explain the meaning of associations. Relationships and associations have been interpreted as relations, i.e. sets of tuples [2, 5] and in terms of their meaning for subsequent system implementation and programming [1]. The semantics of a language construct are defined by its semantic mapping to an element of the semantic domain [6]. For purposes of conceptual modelling, the semantic domain consists of those concepts in which we perceive the application domain, with which we think and reason about the domain. These concepts are human cognitive concepts. Significance of Cognitive Linguistics Research in cognitive linguistics has demon- strated that the most fundamental cognitive concepts are those that are encoded syntactically or morphologically in natural language (e.g. [7, 8]). Cross-linguistic research shows that variations in syntactic features correspond to variations in cognition, confirming the close relationship between the two. Studies have shown evidence of such a relationship in a number of domains such as color categoriza- tion, spatial reasoning, gender systems, etc. Developmental research examines how the development of cognitive structures influence the development of lin- guistic competence, or vice-versa. Either direction of influence confirms the re- lationship between language and cognition. 1 Composition and aggregation associations are outside the scope of this paper. Proceedings of the CAiSE'05 Forum - O. Belo, J. Eder, J. Falcão e Cunha, O. Pastor (Eds.) © Faculdade de Engenharia da Universidade do Porto, Portugal 2005 - ISBN 972-752-078-2 78 Joerg Evermann 2 Natural Language Semantics for Associations Noting the structure of associations, Embley hints at the possible semantics of the association: ”Relationships associate one object with another, similar to the way verbs and verb phrases relate one noun or noun phrase to another” [9, p. 18]. Hence, identifying the semantics of verbs can be used to define and clarify the semantics of associations. A UML profile [10] is used to formalize the proposed semantics. Verb Semantics The most fundamental distinction made in cognitive linguistics is between spatial entities, such as things, places and paths, and temporal en- tities, such as events and states. The former are expressed by nouns and noun phrases, the latter are expressed by verbs2 [11–13]. The temporal domain consists of two concepts: states and events [11–16]. Consequently, we suggest that associations represent two types of concepts: states and events. For events, the main verb usually expresses dynamic action or activity, e.g. ’Customer has ordered product’, ’Supplier will ship product’. In contrast, a state expresses static conditions that hold between associated objects. No change occurs in the objects and states are not commonly associated with activity. In English, they are generally expressed by the verb ”be”, e.g. ’Professor is member of faculty’, ’Product is located in warehouse’. Properties of Events As events and states are expressed by verbs, they possess all of the semantic concepts that natural languages mark on verbs or verb phrases. The upper part of Table 3 summarizes the set of such concepts proposed by cross- linguistic research [14–19] and research in cognitive linguistics [11–13, 18, 19], and gives explanations and examples. The table also shows how these distinctions are formally realized in the proposed UML profile. Causation Beyond the semantic concepts for all events, natural languages mark a further set of semantic concepts for causal events [11]: Directness, Immedi- acy, Coextensiveness, and Resistance. They are shown, with explanation and examples, in the second part of Table 3. Event Participants Events are expressed by verbs, which in turn possess one or more arguments [14–16]. As verbal arguments play thematic roles, so the participating classes or objects in associations must play thematic roles. Table 1 shows the roles proposed by [11, 14–17]. 3 Example Consider an association without the proposed semantics attached: A Shipping Clerk participates in a ”shipping” association with a Customer and a Package. This model is ambiguous w.r.t. the semantic notions described in Sec. 2. For 2 But see ’temporalization’ and ’reification’ in [11]. 79 Role Description Agent The performer of an action Patient To whom something is done, who undergoes an action Object To what something is done Theme The topic of the event Experiencer Who experiences (listens, sees, etc.) something Beneficiary Who undergoes an action with a benefit Locative The location of an event Perceiver The perceiver who sees, feels, etc. an action Instrumental The instrument by which the action is performed Source The source of the action (generally of a motion action) Goal The goal of the action Reason The reason for the action Purpose The purpose of the action Author The speaker or write (for communicative actions) Recipient Who receives something by means of the action Comitative Something that accompanies the action Table 1. Thematic roles (cases) marked on verbs Stereo type Base Class Parent Description (Additional Tags Semantics) State Associa- Association N/A A static condition involving Tense tion two or more objects Event Associ- Association N/A A dynamic interaction be- Tense, Aspect, Progres- ation tween two or more objects sivity, Iterativity, Punc- tuality, Telicity, Modal- ity, Volitionality, Oppo- sition, Success Causal Event Association Event As- An event association where Directness, Immediacy, Association sociation the dynamic interaction is Coextensiveness, Resis- caused by an object (or event) tance Event Partici- Association N/A An association end linked to Thematic Role pant End either a state or event as- sociation, and linked to an object or event participat- ing in this association Table 2. Stereotypes for the Natural Language Semantics Profile example, we don’t know whether the association represents planned shipments, shipments in progress, past shipments or recurring (standing) shipping orders. To explicate the intended semantics, we employ the proposed profile (Fig. 1). The model now shows the roles of the participants: The shipping clerk is the agent, the packages are the objects, and the customer plays the locative role. This indicates that the packages are shipped to or from the customer, rather than for the customer (i.e. at customer’s cost/on the customer’s account). The explicit tags show that the association expresses past (Tense), completed (Aspect), shipping events, not for example current, in-progress shipping. Ship- Name Explanation Examples Type Multi Values plicity Tense Relative temporal posi- Order was taken (past tense), order is taken (present Enumeration 1..1 Past, Present, Future tion of activity tense) Aspect State of completion of Order has been processed (imperfective), order had been Enumeration 1..1 Perfective, Imperfective activity processed (perfective) Progressivity Does the activity have a Shipper delivers product (final state), factory manufac- Boolean 1..1 80 Joerg Evermann final state? tures products (keeps manufacturing, no final state)) Iterativity Is the activity repeti- Customer picks up orders on Wednesday morning (re- Boolean 1..1 tious? peats every Wednesday), customer picks up the order next Wednesday morning (once only) Punctuality Temporal distribution or Product leaves assembly (punctual), product is being Enumeration 1..1 Punctual, Durative interval painted (durative) Telicity Does the activity have a Inventory is reduced (accidentally), inventory is cleared Boolean 1..1 goal? (purposefully, goal-driven) Modality Permission, ability, obli- Customer (can) pick up order (Possible), Customer Enumeration 1..* Actual, Desirable, Pre- gation, etc. (must) pick up order (Obligatory) dicted, Obligatory, Possi- ble, Impossible, Optional, Permissible, Forbidden Volitionality Is the activity willful? Machinist repaired the machine (neutral), machinist was Boolean 0..1 made to repair the machine (willful) Opposition Positive or negative ef- Customer defrauds business (negative), customer refunds Enumeration 0..1 Negative, Positive fects money owing (positive) Success Is success the effect or Staff enters area (prevention), staff enters area (effect) Enumeration 0..1 Effect, Prevent prevention of change? Directness Number of links in causal Machine damaged product (1), machine caused profits to Integer 1..1 chain drop (> 1) Immediacy Temporal continuity in Shipping product reduces inventory (continuous), Ship- Enumeration 1..1 Continuous, Discontinuous causal chain ping product increases profits (discontinuous, effect may be delayed) Coextensiveness Temporal overlap of Breaking the machine caused faulty products (Onset, Enumeration 1..1 Onset, Extended cause and effect cause does not need to be maintained), lowering the tem- perature to harden the product (Extended, cause must be maintained) Resistance Effectuating or enabling Using the forklift to unload goods (Effectuating), opening Enumeration 1..1 Effectuating, Enabling causation the valves to unload goods (Enabling, removing blockage) Thematic Role Enumeration 1..1 ref. Tab. 1 Table 3. Tag Definition and Tag Values for the Natural Language Semantics Profile 81 ping progressed towards a goal (Progressivity, the delivery of the package) and occurred once only, not repeatedly. Shipping was durative, i.e. it took some time, and was the effort of some agent (telicity). The association represents actual shipping events of the past, rather than past plans, abilities, etc. (modality). Stereotyping the association as a ’Causal Event Association’ makes it clear that the shipping clerk caused the object to the be shipped. The causation is indirect: The shipping clerk is twice removed (directness of degree 3), she did not ship the packages herself, nor did she herself cause the courier to ship the packages. Instead, she had the courier ship the packages. The immediacy indicates that is a discontinuous causation, i.e. there is a time lag between the cause and the effect. Perhaps the shipping order takes some time to be processed by the courier. The event is a type of onset causation, as the shipping clerk does not have to maintain any action to sustain the shipping activity. Finally, the event is caused by enabling it, rather than effecting it. For example, shipping orders may already be issued but need to be approved by the shipping clerk. The approval removes the blockage and the event can proceed. In contrast, for an effectuating causation, the agent issues the shipping orders, rather than remove a hindrance. <> {Tense=Past, Aspect=Perfective, Progressivity=True, Iterativity=False, Punctuality=Durative, Telicity=True, Modality=Actual, Directness=3, Immediacy=Discontinuous,Coextensiveness=Onset, Resistance=Enabling} Agent Object Shipping Clerk Shipping Package Locative Customer Fig. 1. Example association representing causation using the proposed profile Without the proposed profile, the example could be interpreted in many different ways. Semantic distinctions are often implicit and based on domain or background knowledge. When this knowledge is not shared among modeller and model interpreter, the model may be interpreted incorrectly. The proposed profile forces the modeller to explicate the possible semantic distinctions and rely less on assumed background knowledge. Hence, it leads to more accurate model interpretations. 4 Discussion and Conclusion Especially in the context of MDA, we need to consider not only conceptual mod- elling, but also implementation concerns. This proposal does not introduce new 82 Joerg Evermann constructs, nor does it constrain the use of constructs. It has therefore no con- sequences for IS implementation. We believe that disambiguating the semantics of associations is a valuable contribution by itself. The fact that some distinctions may appear to be not applicable in some situations does not indicate a shortcoming of the present proposal. The cogni- tive linguistics research on which this proposal is based, suggests that, while not all languages make all distinctions, every distinction is grammaticized in some natural languages. Instead of dismissing concepts such as ’opposition’ or ’suc- cess’ as not relevant, they can offer insights into the application domain and its dynamics which may be hidden and require further exploration. They may also have significance in cross-cultural or cross-linguistic IS development contexts. Finally, the fact that events may be represented as classes instead of as- sociations, e.g. ’Shipment’, ’Enrollment’, ’Use’, etc. shows the need for further exploration of this research. The present paper is intended to clarify the seman- tics of associations, rather than the representation of events and states. References 1. Stevens, P.: On the interpretation of binary associations with the unified modelling language. Software and Systems Modeling 1 (2002) 68–79 2. Genova, G., Llorens, J., Martinez, P.: The meaning of multiplicity of n-ary associ- ations in UML. Software and Systems Modeling 1 (2002) 86–97 3. Wand, Y., Storey, V.C., Weber, R.: An ontological analysis of the relationship construct in conceptual modeling. ACM TODS 24 (1999) 494–528 4. Storey, V.: Understanding semantic relationships. VLDB J 2 (1993) 455–488 5. Dey, D., Storey, V.C., Barron, T.M.: Improving database design through the anal- ysis of relationships. ACM TODS 24 (1999) 453–486 6. Harel, D., Rumpe, B.: Meaningful modeling: What’s the semantics of ”semantics”? IEEE Computer (2004) 64–72 7. Bowerman, M., Levinson, S.C., eds.: Language acquisition and conceptual devel- opment. CUP, Cambridge, UK (2001) 8. Gentner, D., Goldin-Meadow, S., eds.: Language in Mind: Advances in the Study of Language and Thought. The MIT Press, Cambridge, MA (2003) 9. Embley, D.W.: Object-oriented systems analysis: a model-driven approach. Pren- tice Hall, Inc., Englewood Cliffs, NJ (1992) 10. OMG: The Unified Modelling Language Specification. Version 1.5. (2003) 11. Talmy, L.: Toward a cognitive semantics : Concept Structuring Systems. Volume 1. The MIT Press, Cambridge, MA (2000) 12. Jackendoff, R.: Semantics and Cognition. The MIT Press, Cambridge, MA (1983) 13. Jackendoff, R.: Semantic Structures. The MIT Press, Cambridge, MA (1990) 14. Frawley, W.: Linguistic Semantics. Lawrence Erlbaum, Mahwah, NJ (1992) 15. Whaley, L.J.: Introduction to Typology. The Unity and Diversity of Language. Sage Publications, Thousand Oaks (1997) 16. Palmer, F.R.: Grammatical Roles and Relations. CUP, Cambridge, UK (1994) 17. Cook, W.A.: Case Grammar Applied. The Summer Institute of Linguistic and The University of Texas at Arlington, Dallas, TX (1998) 18. Croft, W.: Typology and Universals. CUP, Cambridge, UK (1990) 19. Falk, Y.N.: Lexical Functional Grammar. An Introduction to Parallel constraint- based syntax. CSLI Publications, Stanford, CA (2001)