The holistic view in linguistics

invisible feature that distinguishes the two noun stems. And, again, the atomistic view allows many equivalent solutions that we have no ground to choose from. The holistic approach, on the other hand, imposes us an explanation of the behaviour of híd and szív in terms of the different patterns they obey, and this is the single and necessary explanation within a holistic framework. We can explain the appearance of linking vowels in the same vein, without using the concept of linking vowels at all. In this case, too, it is not sufficient to take a single paradigm into account. For example, in order to understand the word form meaning ‘house (superessive)’, we have to look at all superessive forms in addition to forms of ház ‘house’. Not all word forms ending in -Vn (with a non-low vowel) can be interpreted as superessive, only those that can be contrasted with other (nonsuperessive) forms that start similarly. For example, the form házon ‘house (superessive)’ fits the patterns of word forms starting in ház(and bearing the meaning ‘house’), on the one hand, and the patterns of other superessive forms, like falon, boton etc. The reason why it takes the form házon rather than *házan is that the latter would not fit the pattern of (is not similar to) any other superessive form. Incidentally, this explanation entails a substantive prediction, namely: if Hungarian contained superessive forms ending in -an, then the form *házan would be at least marginally or sporadically acceptable. No such prediction is made by any possible atomistic model. To summarize, the holistic description of word forms relies on themultidimensional comparison of existing forms. This may remind one of the socalled optimality theoretic approach, which posits a kind of competition between conceivable candidates. The winner of the compatition is the form matching best the patterns that are functionally most adequate In my example, házon would be the form that matches the best the forms bearing the functions ‘house’ and ‘superessive’. But there are major differences between optimality theory and analogical theories. First, in standard optimality theories, it is not existing patterns that decides between competing forms, but violable constraints assumed to be universals (but the relative importance of which is language specific). Second, standard optimality theories do not evaluate surface forms in terms of fitness, but relation-