Affix Productivity and Base Productivity

Morphological productivity is generally seen as affix-driven, with a given affix selecting base words satisfying a range of formal and semantic selectional restrictions. Plag (1999), however, has shown that there are base-driven selectional restrictions, and Hay (2000) and Hay & Baayen (2002) have shown that the frequency relation between derived and base word is correlated with affix productivity. This paper further investigates the role of the base with respect to the productivity of affixes, focusing on the possible role of the morphological productivity of the base itself. Base productivity has been studied recently in some detail from the perspective of language comprehension, where base productivity is referred to bymeans of the term family size (Schreuder & Baayen, 1997; De Jong, Schreuder, & Baayen, 2000). Some words, such as scythe or clown, have no or only a fewmorphological descendants, complex words in which they occur as a constituent. Such words have a small family size and no or very limited productivity as base words. By contrast, words such as man or coin occur as a constituent in many other complex words. They have a high degree of base productivity and a high family size. In comprehension, simplex words with a high family size and complex words with base words with a high family size are processed more quickly than matched controls with a small family size. This effect of Family Size has been observed for Dutch, German, English, Hebrew, and Finnish. What the family size effect shows is that, far from being isolated islands, words are connected in the mental lexicon with their morphological relatives. This raises the question whether the degree of morphological connectivity of a word or stem in the mental lexicon, its stem productivity, might in some way affect affix productivity. Experimental evidence that this may indeed be the case is provided by an experiment reported in De Jong et al. (2002). In this visual lexical decision experiment, participants were presentedwith novel but otherwise fully regular prefixed and suffixed Dutch words. Response latencies to these words emerged as depending on a number of factors such as age of acquisition, the frequency of the base word, and the family size of the base. Complex words coined from a stem with a high family size were responded to faster than complex words coined from a stem with a low family size. This shows that the interpretation of novel complex words is facilitated when the base of the complex word is well embedded in the network of morphological relations in the mental lexicon. Is there evidence from production that stem productivity, operationalized in terms of family size, affects affix productivity? To answer this question, we will focus on the patterns to be found in the complex words that speakers produce. We do so for English on the basis of the CELEX lexical database, (Baayen, Piepenbrock and Gulikers 1995), which is based on a corpus of some 18 million words. Hay & Baayen (2002) used the CELEX data to construct a morphological database of 112 morphological categories (sets of complex words sharing the same affix) of bi-morphemic derived words, for details see Hay & Baayen (these proceedings). We enriched this morphological database with counts of the family size of the (simplex) base words of these derivedwords. Let the ‘categorial family size’ of a morphological category be the sum of the (logarithms of the) family sizes of the base words of themembers of this category. More precisely, given a morphological category of a given affix with V (affix) different (bi-morphemic) derived words, its categorial family