Màster de Lingüística Aplicada i Adquisició de Llengües en Contextos Multilingües, Departament de Filologia Anglesa i Alemanya, Universitat de Barcelona, Curs: 2015, Tutor: Roger Gilabert Guerrero
Propositional complexity is a dimension of L2 performance that refers to the amount of
information that a person conveys in a given message and, according to Ellis and
Barkhuizen (2005), it can be measured in terms of idea units (IUs). This study does not
only aim at developing some guidelines as to how to segment oral and written data into IUs
in order to operationalize a measurement of propositional complexity, but it also aims at
investigating the relative impact of mode, discourse type, task type and task complexity on
participants’ production of IUs. In order to achieve these objectives, the study analysed
data that was generated by participants out of performing tasks that differed in mode,
discourse type, task type and task complexity. After segmenting this data following the
guidelines that were designed, it was considered that the guidelines might constitute a
reliable means of operationalizing propositional complexity, as a considerably high
agreement between raters was obtained. As regards the relative influence of mode
discourse type, task type and task complexity on the number of IUs conveyed, after
conducting a standard and a hierarchical multiple regression, the results showed that 38.5%
of the variability in production of IUs can be significantly explained by these independent
variables and that all of these variables made a significant unique contribution on the
number of IUs that can be produced. Nonetheless, the amount of variance in the dependent
variable explained by each of the predictors was different. In the standard regression, mode
appeared to be the best predictor, uniquely explaining 9.9% of the variance in production
of IUs, while the rest of the predictors independently explained between 1.7 and 4.4% of
the variance. In the hierarchical regression, nonetheless, the results (which were supported
by further analyses that were conducted), indicated that both mode and discourse type were
the best predictors of number of IUs conveyed. Task type and task complexity were found
to account for a similar amount of variance to the one obtained in the standard regression.