Changing the viewpoint of error analysis : from local to textual
Résumé
Debates continue about the best way to analyse and interpret second language writing errors. Some advocate a traditional approach using surface descriptions in terms of deviation from the target language (cf. Corder 1967; Dulay, Burt & Krashen 1982). Others recommend a more lexico-grammatical approach using formal linguistics categories (Granger 2003; Nicholls 2003; Milton & Cheng 2010). Another thorny issue concerns the context in which individual errors are analysed. Often items may seem appropriate when considered only in relation to the surrounding clause or sentence, but inappropriate when considered in relation to the preceding or upcoming text. In other words, once local errors have been corrected (orthographic, lexical, and grammatical), textual problems may remain and lead to text that is incoherent, or difficult to understand.
Using sample corpus data, taken from a corpus of over 120,000 tokens of French University students writing in English, we propose a method to take into account both local and textual errors. This approach is inspired by Halliday's (1994) theory of minimum bracketing. Using an annotation software package, UAM CorpusTool (O’Donnel 2010), we will give details of the multiple viewpoints possible in error tagging. This method compels us to focus not only on the level of grammaticality in learner errors but also on the level of meaning and textual construction as a whole.
In this paper particular attention will be paid to the use of adverbs and to pronominal references with it, this and that. A preliminary study of our data has shown that the analysis of such items can greatly benefit from the “local to textual” approach to error analysis