If-conditionals: Corpus-based classification and frequency distribution

This paper discusses the frequency distribution of the types of if-conditionals recognised in the corpus-based classification developed in Gabrielatos (2010: 230-265). It is pertinent to mention at the outset that if-conditionals have been estimated to account for about 80 per cent of all conditional constructions in written British English (Gabrielatos 2010: 49). The classification was partly adapted from Quirk et al. (1985: 1072-1097), and was based on two interrelated criteria: a) the nature of the link between the two parts of a conditional, (henceforth, protasis and apodosis) and b) the modal nature of the apodosis. The quantitative analysis discussed here provides insights into the nature of each type, and the ways that the interaction of the type of link between protasis and apodosis, and the type of modality expressed by the apodosis gives rise to their potential for use in communication. The motivation for the development of a corpus-based classification of if-conditionals was the realisation that existing classifications have not been tested on representative samples of actual use, and, as a result, exhibit particular limitations (Gabrielatos 2010: 152-188). These limitations can be better understood when we consider the distinction between introspectioninformed, data-informed, and corpus-based classifications (adapted from Gabrielatos 2010: 10-13). Introspection-informed classifications, and the examples used to support them, are derived merely from the analyst’s introspections and informal observations. Data-informed classifications are supported by attested examples of use (e.g. taken from newspapers, novels, television, internet, overheard conversations, or corpora). However, these examples are selected ad hoc (even when the source is a corpus) to exemplify types that have been formulated on the basis of introspection or informal (i.e. unsystematic) observations, and can have no claim to being representative. Corpus-based classifications are based on an appropriate representative corpus, and adhere to the “principle of total accountability” to the data (Leech 1992: 112). That is, the analysis and resulting theoretical interpretations have to account for all relevant items in the corpus sample (in our case, if-conditionals) – no items are ignored or discounted, however inconvenient they may be for the proposed classification. In addition, corpus-based classifications can provide information on the frequency and distribution of particular types. Classifications that are not informed by the examination of representative samples of natural occurring language can be expected to reflect the analyst’s introspections rather than actual language use; that is, even if they use attested examples, they leave open the possibility that types of if-conditionals may have been left out, because they are not accessible via the analyst’s introspection, or have escaped the analyst’s attention, or, worse still, because they are incompatible with the proposed classification. More specifically, the detailed examination of existing classifications of conditionals in Gabrielatos (2010: 10-13, 152-188) identified the following interrelated shortcomings: