Toward a probabilistic automata model of some aspects of code-switching

As Kolers (1972) has pointed out, the study of bilingualism offers a distinctive way to study some of the mental activities underlying the way that people use language. Code-switching (the introduction into the context of one language stretches of speech which exhibit primarily the other language's phonological and morphological features) offers a unique opportunity for studying some of the more complicated aspects of bilingual speech, particularly regarding the momentary choice of language. Researchers whose work has influenced this paper include Haugen, Weinreich, Mackey, Fishman, Rayfield, Clyne, Hasselmo, Gumperz, Diebold, Lance, Valdes-Fallis, Pfaff, and Timm. Among the list of poorly understood phenomena in code-switching are (i) seemingly unpredictable switching, and (2) the possibility of formal, complex rules underlying different codeswitching patterns. It is these open questions which modelling might illuminate. During the past decade, probabilistic automata have been used to model many aspects of nondeterministic, intelligent behavior as developed pedagogically in Paz (I97I) and surveyed in Narendra & Thathachar (I974). Thus it is natural to try to model code-switching with probabilistic automata in an attempt to improve our understanding of code-switching and its relation to language use and intelligence. The attempt at modelling code-switching will provide guidance for quantification, and allow the resulting theory to be compared with actual code-switching patterns. As is usual in modelling, inconsistencies are expected to provide guidance for further refinement of the theory and development of the model. The purpose of the model is to select either Spanish or English as the language to be used, the selection to be made from input strings of pure English and pure Spanish, in a fashion acceptable to bilinguals. Goals of the model at this stage of development are (i) to model code-switching for lexical need, (2) to model apparently random code-switching, (3) to model the dependency of codeswitching upon sociolinguistic context as in Vald's-Fallis (1976), (4) to model code-switching within syntactic constraints as observed by Timm (I975 and Pfaff) (I975), and (5) to provide a vehicle for further refinement in the theoretical development of this complex process. Assumptions used in the present version of the model are (i) that context-invariant 'chunks' of information delineate import-