Collocations in a Learner Corpus (review)

If vocabulary was the flavour of the month in applied linguistics in the last decade, in this one it is the multiword unit (MWU). This refinement follows a certain logic, because, if the claims made for said unit are even half accurate, then all levels of the language teaching industry are in for a significant rethink. Ideas to be incorporated would be that grammars emerge from phrases, not vice versa; that main tasks in language acquisition are piecemeal, not rule based; and that functioning lexicons consist not in manageable handfuls of words but in vast array of combinations lexicalized to varying degrees and operating within mazes of apparently random restrictions. No surprise that the rethink has hardly begun, with progress somewhat held up, until recently, by the lack of clear terms and an empirical database. To contribute to ongoing work on both fronts is the purpose of Nadja Nesselhauf's book, which is based on her doctoral study of one type of multiword unit in the written production of advanced German-speaking ESL students.

Nesselhauf has assembled a corpus of advanced learner writing with a view to inspecting one of its MWUs, following procedures for learner corpus research established by Granger (1998). But Nesselhauf goes beyond anything published to date in her delimitation of phenomena and her generation of comparable data. In a detailed but (largely) readable account of her methodology, she carefully separates out collocation as the type of MWU she will look for, and, within that category, verb–noun collocations (i.e., ride a bike not *drive a bike), with the specification that the restriction (on ride) be fully arbitrary rather than meaningful. A catalogue of such collocations is hand-extracted from her learner corpus, and native collocations separated from learner deviations by native raters; collocations are counted in terms of frequency and range, and deviations are categorized by type and probable intended meaning. All data can be traced back through individual writers to a background questionnaire itemizing years of ESL study, extent of exposure to English abroad, and conditions of writing (e.g., timed or untimed; dictionary yes or no). To call this 'a lot of work' is an understatement, and, indeed, the amount of manual work involved raises the question of whether this approach can be scaled up to a larger corpus (than the 200,000 words used here), as Nesselhauf proposes.

But even a smallish corpus, carved with instruments so fine, can generate interesting information. A predictable finding is that collocation remains a serious problem well into advanced learning. Less predictable is that neither years of instruction, nor years abroad, nor writing with or without time pressure, with or without a dictionary, has any effect on the number of collocations employed or on the number of deviations. Particularly interesting are the deviations exposed by imputing intended meaning – as when a learner writes 'I don't take care of carrots,' which is a good collocation, except that he probably means 'I don't care for carrots' (which a computer match of learner strings against a standard corpus would have missed).

So, then, the problem is even worse than we thought; but what is the solution? Nesselhauf explores awareness versus learning as solutions. Learners are not (encouraged) in the habit of scanning language to become aware of restrictions on word combination. The collocation problem resembles one from the vocabulary research: that a word met in rich contexts can have a meaning so obvious that the word itself does not register in memory – 'ride a bike' paints a picture so clear that there is little motivation for the learner to notice that it was ride and not drive. A lengthy section on pedagogical implications suggests ways of promoting awareness as well of developing a collocational syllabus.

Interspersed in Nesselhauf's treatment are attempts to clarify unresolved issues in the MWU agenda. One concerns Kjellmer's (1991) idea that, while natives process language in prefabricated sequences, learners rely on grammars and lexicons, which leaves them 'sounding odd...