As argued in the preceding chapter, the link between the explanans and the explanandum is weakened by the fact that students of language are free to take their explanation from any area that appears suitable to them. The connection between language and the extralinguistic world would be more compelling if this freedom were curtailed by restricting the pool of possible explanations to a single area, viz. psycholinguistics. This restriction should make it possible to gauge more precisely the contribution that language processing can make to the explanation of linguistic structure. While this procedure by no means negates the importance of other disciplines, it focuses upon only one in an attempt to ascertain how far this one takes us. Implicationally, interactions among various disciplines have to be ignored.
A second modification concerns the direction of analysis. Whereas previous investigations moved from linguistic patterns to extralinguistic explanations, the present study moves from extralinguistic (i.e. processing) patterns to language structure. This reversal of direction is reflected in a change of terminology. Linguistic facts are no longer explained post hoc but predicted. The basic assumption is that it is possible to derive predictions about language structure from processing principles. If these predictions are borne out, a particular linguistic structure may be said to be explainable in terms of these processing principles. Importantly, this explanation is different in kind from the conventional one. The explanation-by-prediction is not post hoc.
Three reasons have motivated this innovation, two objective and one subjective. Beginning with the latter, the direction from psycholinguistics to linguistics matches my personal background. My original research interest was in psycholinguistics. With time, it shifted to (or rather broadened to include) the more structural aspects of language. To give a rather drastic example, when I examined psycholinguistic data from Arabic, I discovered that the processing model had implications for the structure of poetic rhymes. I am not ashamed to confess that I did not have the slightest idea of Arabic verse when I formulated the relevant predictions (see section 6.3 for the outcome of this strand of research).
The second reason has already been hinted at above. The explanation-byprediction is more restrictive in that it reduces the risk that a given account fits a given linguistic pattern by chance. The probability of finding an explanation for a