Andrea C. Schalley - A meaning representation framework for decompositional and compositional semantics

In this paper, we will present a meaning representation system that was originally developed for decompositional verbal semantics (Schalley 2004), and which is currently being extended to integrate compositional meaning (or, more broadly, discourse representation).
   This meaning representation system, the Unified Eventity Representation (UER), is based on the understanding that linguistic meanings are conceivable as conceptual configurations and that semantic descriptions amount to structured representations or ‘modellings’ of such underlying configurations and their components. The distinction into different componential types is essential for faithful representations of conceptual configurations. The UER iconically reflects this: it distinguishes different modelling elements for different conceptual types (for instance, attribute, role, or class [for sortal concepts]). Modelling elements can, if their respective well-formedness rules allow this, be nested and related to each other.
   What is important from a theory-building point of view is the explicit specification or ‘definition’ of modelling elements. This is done via the metamodel, where the general semantics for modelling elements is defined. For instance, the modelling element ‘meronomy’ (used for the representation of part-whole relations) is specified via the metamodel element ‘Meronomy’, which defines the general semantics for part-whole relations. This definition includes a range of attributes that are relevant for part-whole relations (such as the attribute ‘isSeparable’) and to which a specific part-whole instance attaches values (such as the ‘yes’ value for the attribute ‘isSeparable’, indicating that the part can be separated from the whole [cf. Winston, Chaffin, Hermann 1987]). Additionally, constraints that apply to all meronomies are listed, such as that a whole has to have more than one proper part.
   Definitions of modelling elements at the metamodel level thus make extensive use of attribute-value pairs (cf. Barsalou 1992, 1993a, 1993b). Although it would be exaggerated to say that the system, as it currently stands, is fully frame-based, we believe it could be converted into a frame-based system. The crucial step would be to iterate definition levels so often as to arrive at a purely frame-based level. Or, taking frames as starting point: to define configurations of attribute-value pairs which can be recursively used for modelling and further specification. This results in a highly sophisticated representation system at the model level, with cognitively relevant concept types being readily available for semantic representation. The paper will elaborate this idea further.
   Examples of UER representations of decompositional semantics (for instance, the meaning of the verb wecken ‘wake up’) and compositional semantics (for instance, the meaning of Sandra weckte Niels ‘Sandra woke up Niels’) will be discussed. Here again, different representation levels – namely model level and object level – will come into play, further demonstrating the importance of the multi-layered system architecture just indicated. (Currently, the UER adapts the standard modelling language for the design of object-oriented software systems in computer science, the Unified Modeling Language (UML) [cf. Object Management Group 1997-2007], including its design and four-layered architecture.) A discussion of other aspects of UER attribute-value structures will conclude the paper.