Acceso usuarios
Juan Andrés Villena Ponsoda 
, Antonio Manuel Ávila Muñoz , José María Sánchez Sáez
The aim of this paper is to propose a multinomial probabilistic model able to account for the variation of /th s/ in Southern Spanish varieties. The current analyses carried up till now are based on frequency counts of mergers between /s/ and /th/, which are labeled "ceceo" and "seseo", as two social and regional norms of pronunciation. Standard Spanish distinction between /s/ and /th/ is therefore to be deduced from the total body of cases, counting every [s] corresponding to an etymological or canonical s as an occurrence of /s/, including the tokens of those individuals whose pronunciation pattern of /th s/ is overtly "seseo" (exactly the same for the contrary case of "ceceo"). Since both types (s > th and th > s) currently coexist in Southern speech communities as in Málaga city, it is not obvious to decide about the high proportion of cases where /s/ and /th/ are pronounced in a canonical way. Therefore we need a procedure to decide the rate of cases of canonical pronunciation due to chance. Two related multinomial models have been developed, implemented and tested on a sociological body of data from a social network of 31 individuals from Málaga city, extracted from a sociolinguistic investigation of vernacular varieties in Málaga (M.U.V.-Project). The first model calculates the single probability of seseo (s > s and th > s), ceceo (th > th and s > th) and distinction (s > s and th > th) for every individual studied. The second model calculates the single probability of distinction (s > s and th > th) vs. the single probability of reduction or "confusión" (s > th and th > s). To measure the degree of fit of both models, a maximum likelihood estimator has been developed with consistent and hopeful results. Comparison between relative frequency and probabilistic results demonstrates, finally, the more realistic approach lo this problem on the basis presented here.
