Quantitative approach to morphological productivity: some recent developments

This article aims at presenting the quantitative approach to morphological productivity based mainly on Baayen's work. The discussion starts out from the widely accepted distinction between a qualitative and a quantitative approach. It is argued that there are two main quantitative approaches: one based on dictionaries, the other on large text corpora. While the dictionary-based investigation is limited to measures based on type frequency (V), the corpus-based research requires another variable: the token frequency (N). The main idea behind the relation of type frequency and token frequency is that the former (V) can be viewed as a function of the latter (N). The increasing value of N (given by the corpus size) will lead to the increasing value of V. This relation gives rise to the definition of vocabulary growth curve (BAAYEN, 1992 ; 2008). Two additional measures are also presented. The rate at which the vocabulary grows can in fact be captured by the proportion of hapax legomena (V1), the types that occur precisely once, to the overall number of tokens of the formations with a given affix. The notion of vocabulary growth rate (P = V1 / N) (BAAYEN, 1992) is thus introduced. Finally, a third statistical tool of modelling the relation of V, N and V1, put forward by BARONI, EVERT, 2006, is presented. It is the frequency spectrum, which is a specific object that views the number of types (V) as a function of a frequency rank (m) assigned to every type according to its token frequency. Some problems typical of this quantitative approach are also discussed, namely the difficult relation between hapax legomena and neologisms, and the role of the number of tokens for the assessment of the productivity. As far as the role of the number of tokens is concerned, it is shown that - in the light of the evident fact that the measure depends directly on the corpus size - it is not possible to compare corpora of different sizes using this measure (cfr. BAAYEN, 1992 : 117). In order to overcome this problem, two main techniques are presented: binomial interpolation and extrapolation. Especially, three modes of extrapolation are introduced. In conclusion, something is said about the research and particular studies being conducted within this framework.

Keywords: morphological productivity; quantitative approach; hapax legomenon; neologism; binomial interpolation; extrapolation