ABSTRACT
In other words, beyond my specific assumptions regarding the time-con strained nature of memory and comprehension, I have tried to adopt wellaccepted hypotheses (e.g., memory possesses temporal partitions, STM has limited capacity, etc.) found in the psycholinguistic literature. Should a meth odology successfully prevent a reader from allocating more resources to the comprehension process due to the “strain” of the experiment, we could then hope that such assumptions could be confirmed or proven incorrect. On this topic, however, the computer scientist is obviously dependent on the psycho linguist, and we must remember the warnings of Dillon (1980) and Spiro (1980), reasserted more recently by Perfetti (1993) and Graesser, and Kreuz (1993): An experiment always has a context, and this context may signifi cantly affect the observed performance. The acknowledgment that perfor mance is difficult to predict (due to the multitude of factors that affect it, including social and cultural ones) should not, however, lead us to postulate some self-validating notion of “readerly competence” (that abstracts away
performance). Instead, we should pursue the characterization of an interpreta tion with respect to a performance context, and, from a pragmatic viewpoint, we should be ready to accept that, ultimately, it is the user of IDIoT that acts as the sole judge of the acceptability of an interpretation.
