Replicating the Bourdin and Fayol results on cognitive load in German children and adults

In an influential paper, Béatrice Bourdin and Michel Fayol (1994; and subsequently 2000, 2002) explained, in terms of cognitive load, that low-level processes of written language production are more difficult than those involved in oral language production due to the cognitive costs of grapho-motoric and orthographic behavior. Their results were obtained with mother-tongue speakers of French, a language with a rather little transparent orthography. The main aim of the present paper is to replicate the basic findings of Bourdin and Fayol in German speakers. These replications include the assessment of individual memory span in the written as opposed to the oral mode, as well as in elementary school children as opposed to university students. Moreover, we examine developmental changes of the cognitive costs of writing during elementary school. The French results replicate quite well for German participants. Elementary school children, but not university students show worse memory span performance in writing, compared to oral recall. This effect holds for second graders as well as for forth graders, although an age-related overall increase in the number of correctly recalled items occurs in either recall mode. Conclusions from these date relate, among other things, to the fact that until the end of primary school knowledge or cognitive abilities will not be assessed equally well in speaking and writing. Bourdin and Fayol’s findings: A summary Theoretical approach By posing the question: „Is written language production more difficult than oral language production?“ (Bourdin & Fayol, 1994), the authors started to report on a line of research that examined the cognitive costs of so called low-level activities in writing, i. e. formulation (lexical access and sentence generation), and graphic transcription and execution, as opposed to high-level processes that involve planning, i. e. the generation, selection and ordering of ideas. They based their expectations on a general resource model of the cognitive system, assuming that there is a limited capacity of working memory, that high-level as well as low-level components of writing put load on working memory, and that with increasing load of one component, performance based on the other components would suffer due to a decrease of the remaining resources. Bourdin and Fayol further assumed that children, but not (literate) adults, lack to some degree automation of low-level processes in writing (particularly spelling and the motor activity of handwriting) which should impair the performance of higher-level activities. (See Alamargot & Chanquoy, 2001; Grabowski, 1996; Kellogg, 1999; McCutchen, 1995; for more general accounts on cognitive processes in writing and writing development.) In order to specifically empirically address low-level processes of language production, they used a task which did not call for planning, idea generation, or sentence formulation, thus excluding high-level processes as far as possible. Rather, participants were acoustically provided with word lists of increasing lenghth and were, after the presentation of each list, asked to recall the list completely and in its original order, either in the oral or in the written mode of language production. Although this task Grabowski: Replicating the Bourdin and Fayol results ... 2 resembles a simple word span test (see, for example, Daneman & Carpenter, 1980), Bourdin and Fayol did not determine individual memory spans, but analysed the proportions of correctly recalled words within each list length. The key experiment In their basic experiment (= Experiment 1), 24 university students were presented with lists from four to nine words long, 24 second graders and 24 forth graders with lists from three to seven words long. There were always two specimens of each list length, one of which was to recall orally, while the other one was recalled in written form. Thus, participants constantly changed recall modes, while the list length steadily increased. Of course, mode order and assignment of list length specimens to recall modes were counterbalanced. Results include the absence of a mode effect for university students, i. e. serial recall perfomance remained consistent across written and oral recall; a significant mode effect for both groups of school children who show better recall in the spoken mode as compared to the written mode; and a significant effect of the children’s age, according to which fourth graders do generally better than second graders on the serial recall task. These results indicate that in second graders as well as in fourth graders, writing, which is not yet fully automated, involves higher cognitive load than speaking, thus reducing working memory capacity available for the serial recall task. Further experiments While Experiment 1 can be considered the key experiment of the Bourdin and Fayol (1994) research, the researchers added a series of experiments to support their interpretation against several possible objections. From these experiments, we learn that it is not the faster execution speed of speech as opposed to writing that facilitates the memory recall task as such, thus leading to better oral performance (Experiment 2); that children in a dictation mode, where the transcription of the recalled items was done by the experimenter, performed like in the oral mode (Experiment 3a), whereas performance of adults decreased in a mode of unpractised writing (Experiment 3b), suggesting that a lack of automation in handwriting could indeed be responsible for the overall result pattern. While all of the results discussed so far should, in principle, apply to the members of all linguistic communities (at least restricted to those who employ alphabetic script), because their interpretations relate to general cognitive processes, the authors continue the series of experiments by considering some aspects that could also play a role but are specific to French and the used verbal materials, like the more or less regular spelling of words and their familiarity (Experiment 4). In subsequent research, Bourdin and Fayol continued to address the issue of cognitive costs of graphic activity in writing. In Bourdin and Fayol (2000), the non-automated processes in children’s writing are considered a secondary task which is, within a dual-task paradigm, compared to other secondary tasks during serial recall of word lists of increasing length. Again, oral recall proved to be superior to writing in primary school ages. In terms of secondary task load, written performance compared to a condition in which a sound categorization task was to be handled in parallel to the oral recall of the word lists, which clearly consumed attentional resources from working memory’s central executive. Through the careful dissociation between motor, graphic and lexical aspects by appropriately designed secondary tasks beyond the very oral and written recall, the authors also showed that the cognitive load of a simple drawing task appeared to significantly decrease from second to fourth grade of primary school, which indicates that a mere lack of automation in graphomotoric skills can not be (solely) responsible for the apparent cognitive costs of written recall, that impair serial recall performance when compared to the oral mode. Eventually, Bourdin and Fayol (2002) again compare the cognitive costs of oral and written language production. Here, however, they directed their interest to adults as well as to a more complex text production task. Although the overall research question of identifying the cognitive load of low-level processes in writing is maintained, the study goes beyond the scope of replications discussed in this paper. Grabowski: Replicating the Bourdin and Fayol results ... 3 Estimating Bourdin and Fayol’s research There are several aspects according to which the above-reported line of research deserves particular estimation, as seen from the perspective of the present paper’s author who as well is interested in the functioning of the diverse sub-processes of oral and written language prodcution in terms of working memory load. First, language production is a process the complexity of which is often beyond systematic experimental control. Research on language comprehension processes is much more advanced, at least historically as well as in terms of research output, because the materials are easier to control (Bock, 1996) and measures of understanding are easier to obtain in form of closed-class variables (which are, in turn, easier to analyse). Consider, for example, the Handbook of Psycholinguistics (Gernsbacher, 1994) that contains only one chapter among 34 chapters that is explicitely devoted to language production (and still restricted to grammatical aspects)! On the contrary, Bourdin and Fayol found appropriate tasks and experimental variations to decompose the language production process such that the interesting aspects remain under full experimental control. Language production involves hierarchically ordered sub-processes, from planning to motor execution, which can, therefore, only be controlled from the top downwards or from the bottom upwards. By employing a serial recall task, Bourdin and Fayol nicely eliminate any variation due to planning processes, syntactic processes, and processes of lexical selection. The remaining processes and their explanation, however, still appear difficult and complicated enough to be pursued for years. Second, after an early period of investigation in which speaking and writing have been mainly compared with respect to their products (Blass & Siegman, 1975; Hidi & Hildyard, 1983; Horowitz & Newman; Portnoy, 1973) or their temporal economy (Gould, 1978), both modes of language production spent their scientific lifes mostly separately and moved in different theoretical circles (Alamargot & Chanquoy, 2001). Particularly with respect to educational aspects, however, it would be helpful as well to directly compare the functioning of both verbal modes on all involved processing levels. For example, it is far from clear whether speaking and writing offer equivalent possibilites for students to express their cognitive achievements. Here, Bourdin and Fayol offer a theoretically uniform approach to the comparison of oral and written language production on low-level processes; similiar proposals for high-level processes came from Herrmann and Grabowski (1995). Reading Bourdin and Fayol’s results from another educational perspective, they point to a broadly neglected problem, namely the role of language production for diagnostic processes in primary school. If, as the above-mentioned results show, speaking and writing differ with respect to the loads they impose on working memory resources and, therefore, lead to significantly different performances on tasks as simple as serial recall of word lists, there is no reason to assume that both verbal modes would provide equivalent indications of more complexe cognitive processes or knowledge in long-term memory, where much more processes are involved that could differ between speaking and writing in relevant ways. Together, the obtained results appear relevant enough to show whether they would also emerge in a population different from French primary school children. If the authors’ interpretation is correct, the findings should allow for successful replication in the young speakers of another language, because the observed differences between the oral and the written mode of recall relate to general cognitive principles. Experiment A: Replication of the key experiment The first experiment is aimed at the hypothesis of Bourdin and Fayol (1994, Experiment 1), according to which „in children, but not in adults, there is a greater load on working memory in the written mode than in the oral mode, due to the fact that the latter is mastered better than the former“ (p. 597). We used a serial recall task quite similar to the French original. With respect to the mode of analysis, however, we will not adopt the original procedure for three reasons. First, the way Bourdin and Fayol (1994) obtain their results from a a two-step analysis of proportions of recalled words appears, in statistical respect, neither smooth nor convincing. Second, it does not connect to memory span measures, although the applied method and task is pretty much the same. Third, it takes a standardized measure like memory span (or more complexe spans: cf. Ransdell & Levy, 1999) to compare results on serial recall tasks across materials and procedures. Therefore, we will describe Grabowski: Replicating the Bourdin and Fayol results ... 4 the obtained results in terms of simple word spans, in the sense of Daneman and Carpenter (1980), as the number of words that, after hearing them once, can be recalled in the correct order.