Decomposition and Delexicalisation in Learners’ Collocational (mis)behaviour

This paper presents some of the theoretical considerations which have arisen from the author’s learner corpus research over the past couple of years (Philip, 2005a, 2005b, 2006, 2007, in press). In particular, it highlights a mismatch between the language of native speakers and that of learners, not simply at the textual level, but at a more abstract, conceptual level. The findings of this extended piece of research suggest that it is time to shift the focus of investigation away from learner errors per se, and concentrate on what barriers seem to be preventing learners from achieving fluency in a second language. This means not only that the causes of errors, often dismissed as interlanguage, have to be investigated more thoroughly, but also that the notion of avoidance has to be taken into consideration. While it is acknowledged that most empirical linguists are uncomfortable with the idea of measuring avoidance, it will be shown during the course of this paper that focusing on error alone is an inadequate means of identifying problems in language acquisition. Learner error must be considered as part of a trio which also includes avoidance, and calquing from the L1 which happens to match L2 patterns (which I will call here fortuitous accuracy) both of which can be measured contrastively: avoidance is determined by its low frequency (or absence) in relation to equivalent native speaker texts in the L2 (cf. Martelli, 2006), while fortuitous accuracy can be identified through the comparison of back-translated forms in a general reference corpus in the students’ L1. The trio of error, avoidance and fortuitous accuracy paints a much more detailed picture of the strategies that advancing learners use in their language production than can be achieved by studying error alone, as is shown in the discussion below. By examining both what is and is not present, and comparing this to L1 and L2 native norms, it possible to get inside one of the major problems affecting language learning at higher levels: collocations in conventionalised abstract and figurative language. Some collocations are easier to learn than others. Many noun+noun and adjective+noun collocates reflect linguistically what can be observed in the real world, and so their lexical co-occurrence seems natural and relatively unproblematic. However, as language moves away from the concrete and observable towards the abstract and, apparently, arbitrary, the likelihood of a collocate to be deemed ‘logical’ by a learner diminishes, as does its accurate recycling in free production. Although there is a growing body of research based on learner-corpus data, it is somewhat limited by its focus on error alone. In particular, while it is possible to state, for example, that “the total number of collocational errors is 105. Of these 0 are adverb +