structural principles are assumed to be operant in all languages,
with only the "vocabulary", i.e. the lexical items and specific grammar, differing across languages. In this model, structures are preferred or disfavored based on their relative syntactic simplicity alone. A quite different proposal has been put forward by advocates of a frequency-based parser, where particular analyses would be adopted by the adult reader due to his/her prior exposure to a given language and its idiosyncrasies. Two distinct frequency models have been proposed. The one is corpus-based, taking into account the number of occurrences of particular syntactic structures in written corpuses and proposing an immediate effect of this factor on parsing (Brysbaert & Mitchell, 1996; Cuetos & Mitchell, 1988; Mitchell, 1994). The other focuses on lexical statistics, under the assumption that syntactic and lexical ambiguity are tightly linked (MacDonald et aI., 1994). As stated by MacDonald et aI. (1994, p. 682) "To the extent that information required by the syntactic component is stored with individual lexical items, it will be difficult to find a boundary between the two systems." These frequency-based approaches resemble the structural approach in that they assume a common underlying mechanism across languages. However, this assumption is based on statistical properties of the language rather than on syntactic complexity, and these statistics-- concerning the frequency of structures and/or of lexical properties--are by definition variable across languages (and within a language if one heeds individual differences or maturational changes). The strength of the frequency-based account lies indeed in its ability to accommodate cross-language differences, even for structures which are identical across very similar languages. The corpus-based, or "linguistic tuning" hypothesis has been disdained by some on the grounds that it does not specify how frequency counts should be tallied on the one hand, and that it