Implications for Psycholinguistic Theory
Although the major thrust of this work is a linguistic one, it may not be off the mark to touch briefly upon the psycholinguistic significance of the empirical findings. This excursion is justified by the assumed interactivity between linguistic structure and language processing. It was argued in Chapter 3 that in order to derive the 'best' predictions it is essential to rely upon the 'best' psycholinguistic theory. Which theory is the 'best' is, of course, a matter of dispute. The standard method of seeking confirmation (or disconfirmation) of a scientific theory is to adduce arguments from the domain to which the theory appertains. For instance, a psycholinguistic theory is built upon psycholinguistic considerations. This methodology is perfectly acceptable; however, there is no reason not to extend the argumentative basis to areas beyond the home domain and in so doing, broaden the coverage of the theory. Linguistic findings may thus be used to adjudicate between alternative psycholinguistic theories. Clearly, one psycholinguistic theory is superior to another when the linguistic data are compatible with the one but incompatible with the other.
For the past twenty years or so, psycholinguistic theorizing has been dominated by the controversy about the serial modular vs. the parallel interactive approach (see Garrett, 1990 for a review). Both theories make radically different predictions about what is possible in language structure and what is not. The serial approach is more restrictive than the parallel one and can therefore be more easily tested. Let us consider the two backbones of the serial model, 'vertical' serialness and 'horizontal' serialness, the former referring to the information flow in the hierarchical organization of language and the latter to the linear order of smaller units within larger ones (e.g. segments within words).1 Because of the strictly top-down information flow in production, the serial model predicts an independence between the higher non-phonological and the lower phonological levels. Similarly, because of the strictly left-to-right strategy, processing on 'later' units can start only after the processing of 'earlier' units has been completed.
These two predictions are undermined by the structure of natural languages. If the phonological level was blind to what goes on at the lexico-semantic level, the relationship between the 'signifiant' and the 'signifié' of the linguistic sign would____________________