INCIHUSA   20883
INSTITUTO DE CIENCIAS HUMANAS, SOCIALES Y AMBIENTALES
Unidad Ejecutora - UE
capítulos de libros
Título:
Genre-driven constraints on semantic choice
Autor/es:
VÍCTOR M. CASTEL
Libro:
Systemic Functional Linguistics: Exploring Choice
Editorial:
Cambridge University Press
Referencias:
Lugar: Cambridge; Año: 2012;
Resumen:
The need to enhance the CG generator Genesys so that it can generate sentences in their natural contexts, i.e. texts, lead us to the well-established tradition in SFL according to which genre structure significantly constrains the class of linguistic signs available for expression within a given text type. Based on Villar´s (2012, Forthcoming) descriptive generalizations on the narrowing effects that the GSP of OTG has on the semantic potential of EG, we addressed the problem of determining how to accomplish this crucial task for text generation in formal terms that are independent of computational implementation issues in the sense of Maxwell & David (2008) so that the "software lifetime problem" does not arise. The first step was to present the basic formal properties of the CGF semantic system network, focussing on two types of constraint, namely: same pass preference rule and lower unit preference operations. The task of same pass preference rules is to account for whatever co-ocurrence restrictions govern the probabilistic distribution of semantic features in a complete traversal of the network for a given semantic unit, e.g. ´situation´. The task of lower unit preference operations is to account for whatever co-occurrence restrictions govern the probabilistic distribution of semantic features in a traversal of the network for a given semantic unit, e.g. ´thing´, on the basis of a configuration of features existing in the selection expression resulting from a higher unit traversal of the network, e.g. ´situation´. The second step was to present Villar´s (2012, Forthcoming) descriptive generalizations on the constraining impact that genre properties of a given text-type, namely OTG´s, have on the probabilistic re-setting of the semantic potential of EG as this is defined in the CGF. This presentation was made in accordance with the CGF formal properties discussed in section 2, i.e. the genre-driven constraints imposed by OTG´s on the CG semantic system network of EG were defined using the CFG formalism. This in turn implied two specific moves: (i) defining the GSP of OTG´s as a system network of genre features, and thus as a network that can be constrained by resorting to the CGF same pass preference rule formalism, and (ii) defining OTG-driven constraints on the semantic system network of EG in terms of the lower unit preference formalism, i.e. GPR´s. The resulting formalisation of the probabilistic resetting of the semantics of EG on the basis of the OTG constraints was implemented so that now the CG generator generates not only isolated sentences but also sentences in their natural contexts. This implied incorporating a system network for genre features of OTG whose function is to define a class of possible OTG genre structure plans (cf. figure 1). Finally, an example of generation was given which illustrated how the implementation of the genre-driven constraints can account for the generation of a sentence like York is a fantastic city, as the sign associated with a terminal genre feature of a specific OTG genre plan instance. The GPR ´gp1´ in (1) narrowed down the scope of the semantic choices available for selection on the part of the user of GeneSys_IDE by resetting the probabilities of the semantics of EG into the probabilities of the semantics of the English of OTG. Similar GPR´s were defined for other genre features of the given OTG genre plan, for each terminal genre feature will be associated with the relevant preferences. The significance of this enhancement of Genesys is that sentence planning can now also be defined so that it is sensitive to the semantic potential of EG as this is restricted by the GPR´s of specific genres. This genre-based narrowing down of semantic choice appears to be an important contribution to the generation of sentences within their natural contexts. We foresee that we are closer to modelling the constraining effects that a given realization of a genre constituent, i.e. a sign, has on the semantic choices available for the following and/or preceding genre constituent(s), and thus on the planning of their realization. Genesys_IDE has now the formal resource to do so though the relevant descriptive generalizations are yet to be found.