Classical claims about the psychological reality of words (and paradigms) mostly predate the advent of experimental methodologies for probing the psychological status of theoretical constructs. But the classical perspective receives a measure of confirmation from contemporary experimental studies. One source of empirical support for a word-based conception of the mental lexicon comes from studies of frequency effects on lexical processing. To describe a lexicon as ‘word-based’ does not necessarily entail a literal repository of word forms. A word-based lexicon might be represented instead as a network structure in which paths or sequences corresponding to words have a particular salience or integrity. Frequency effects also need not (and almost certainly should not) be thought of in terms of what Baayen (2010) terms a ‘counter in the head’, based on pure repetition, but are more likely to reflect local syntactic and morphological co-occurrence probabilities. Thus although many initial studies and results are framed in terms of frequency effects, a discussion of these effects does not carry a commitment to a repetition- based explanation.
A number of these studies showed that the frequency of inflected forms and the size of derivational ‘families’ have a robust effect on lexical processing. One line of research investigated correlations between response latencies in visual decision tasks and various frequency measures related to inflected forms. The earliest studies established that the surface frequency of an inflected form in a corpus was negatively correlated with response latencies (Taft 1979). The base frequency of a word (the summed frequency of its inflected variants) was later found to exhibit a positive correlation with response latencies (Baayen et al. 1997). The logarithm of the ratio between these measures (surface frequency and base frequency) again exhibited a negative correlation (Hay 2001). These and other studies served to confirm the effect of token frequencies on the processing of inflected forms. A second line of research has demonstrated that the processing of an item is influenced by the size of its ‘morphological family’. Following Schreuder and Baayen (1997), this research investigated the effect of type frequency on response latencies in visual decision. A range of studies found that an increase in number of semantically transparent items in the morphological family of a form facilitated processing of the form (de Jong 2002; Moscoso del Prado Martin 2003; Moscoso del Prado Martin et al. 2004a).
A third line of research grew out of attempts to measure the ‘inflectional information’ or ‘morphological information’ expressed by a form. A series of initial studies measured the information carried by inflected noun forms in Serbian (KostiC 1991, 1995; KostiC et al. 2003). These studies developed a surprisal-based perspective in which the information carried by an item corresponds to the negative log of its probability (i.e., the less likely, the more informative an item is). This general measure was refined by weighting the probabilities of items for the number of functions and meanings they express. Kostic et al. (2003) showed that the resulting notion of inflectional information correlates positively with response latencies and that the processing cost of an inflected variant is predicted by its frequency and functional load.
These studies of information load led in turn to efforts to obtain a unified measure of the effects attributed to token frequency counts in the inflectional domain and type frequency counts in the derivation domain. In the process of pursuing this goal, Moscoso del Prado Martin et al. (2004b) adopt a notion of ‘paradigm’ that encompassed derivational families as well as inflectional paradigms and employ a standard entropy measure (Shannon 1948) to calculate a frequency-weighted measure of morphological information. These refinements permit Moscoso del Prado Martin et al. (2004b) to subsume the family size information that correlates with type-frequency effects and the inflectional information that correlates with token-frequency effects under a single measure, which they term the ‘information residual’ of a word.
Subsequent studies, including Baayen et al. (2006) and Baayen et al. (2008), suggested that inflectional and derivation effects should in fact be kept apart. However, the usefulness of an information-theoretic perspective was strikingly confirmed in the investigation of paradigmatic effects in Serbian declensions. As initially reported in Milin et al. (2009a), response latencies in a visual decision task were positively correlated with the degree of divergence between the probability distribution of an inflected variant in the paradigm of an item and the distribution of the variant within the inflection class to which the item belongs. The greater the divergence between these distributions, the longer the response latencies and the higher the error rates in lexical decision tasks. The models under current development within this tradition define a unified perspective on language processing by combining information theory and discriminative learning without consolidating separate morphological domains. The study of Balling and Baayen (2012) demonstrates the relevance of information theory to the study of ‘uniquenenss points’ in auditory comprehension, Mulder et al. (2014) provide an analysis of derivational family effects in terms of discriminative learning, and Baayen et al. (2011) and Baayen et al. (2013) outline a general model of paradigmatic and frequency effects.
The design of these studies was not guided by the assumptions of a classical model, though the results are explicable within this model. From a classical WP perspective, it is not altogether surprising that token frequency effects would be more significant for inflection, and type frequency more significant for derivation. Given the relatively uniform structure of inflectional systems, the type count of inflected variants should be comparable for (non-defective) items belonging to a given word or inflection class. Hence token frequency is expected to be the primary locus of variation. In contrast, as shown by resources such as CELEX (Baayen et al. 1995), the ‘families’ of forms associated with distinct derivational bases can vary by orders of magnitude. Hence type frequency effects are expected to be stronger for derivation.
Yet the same sources that highlight the contrast between inflectional paradigms and derivational families also reveal critical idealizations in the classical WP perspective. The productivity of regular inflectional processes is traditionally assumed to determine uniform paradigms for items within a given word or inflection class. Paradigm size is thus not expected to vary, except where forms are unavailable due to paradigm ‘gaps’ or ‘defectiveness’. However, as discussed in more detail in Chapter 8, this a priori expectation is not borne out. Many potentially available inflected forms are unattested in corpora, because corpora do not converge on uniformly populated paradigms as they increase in size. Instead, the forms of a corpus obey Zipf’s law at all sample sizes and exhibit a distribution in which the frequency of a form is inversely proportional to its rank. Hence larger corpora reinforce patterns exhibited in smaller corpora, while introducing progressively fewer new forms.
Insofar as corpora provide the best available descriptions of the input encountered by speakers they indicate that speakers are not, as often assumed, exposed to full paradigms for all relatively frequent open-class items. Rather, as Hockett (1967) suggests, the speaker’s lexicon comprises a collection of partial paradigms that collectively exhaust the form variation in the language:
in his analogizing... [t]he native user of the language... operates in terms of all sorts of internally stored paradigms, many of them doubtless only partial and he may first encounter a new basic verb in any of its inflected forms. (Hockett 1967:221)
The discussion in Chapter 8 proposes that lexical neighbourhoods are a central part of the creative engine of the morphological system, permitting the extrapolation of the full system of variants from partially attested patterns.