Corpus-based and corpus-driven research, although certainly no longer new, is an exciting direction for researchers interested in the lexicogrammatical patterns in learner writing. This corpus provides researchers interested in conducting corpus research with a collection of 3.7 million words from English learners with 16 different mother tongues (Bulgarian, Chinese, Czech, Dutch, Finnish, French, German, Italian, Japanese, Norwegian, Polish, Russian, Spanish, Swedish, Turkish, and Tswana). Each mother tongue subcorpus consists of approximately 200,000 words, except for the Chinese subcorpus, which consists of nearly 500,000 words. The text type is 93% argumentative essays, with the remaining 7% made up of literary analyses and responses to articles. A majority of the texts (approximately 65%) were written under untimed circumstances, with unrestricted access to reference materials. However, the Chinese-Cantonese subcorpus, the largest of the mother tongue subcorpora, consists mainly of timed texts. The texts have been produced by high-intermediate to advanced learners (B2-C2 proficiency bands of the Common European Framework of Reference), although the editors acknowledge that the proficiency level of the writers is a fuzzy variable and that the proficiency designations should be taken as approximations only.

The corpus has been tagged for part of speech (POS) with the constituent likelihood automatic word-tagging system (CLAWS4; Garside & Smith, Reference Garside, Smith, Garside, Leech and McEnery1997), using the CLAWS version 7 (C7) tagset (see http://ucrel.lancs.ac.uk/claws/ for more information). This functionality represents an improvement over the first version of the corpus, released in 2002. A small pilot study conducted by the compilers found 95–99.1% accuracy for the POS tagging. The editors’ recommendation that a pilot study be done before any major analysis is conducted is well taken: For example, a sample query for demonstrative adjective + noun concordances included a number of that complementizers in the results. Queries can be entered for lexical units, POS tags, or combinations of the two. A user can also create queries based on inflectional and derivational morphology.

A second change from the earlier version of the corpus is that text profile information is stored in separate files and linked to the texts. Query results can be analyzed and sorted by a variety of variables associated with each text, such as the language and educational background of the writer and the task conditions under which the text was produced. Descriptive statistics for both texts and occurrences of query targets can be computed by the corpus software and exported in a variety of file formats. The software also generates bar graphs and pie charts for these statistics, although it does not seem possible to save or export these figures. The concordance lines can also be exported in a variety of formats, and the texts associated with a query can be compiled into a subcorpus and saved as a single text file.

The manual includes a description of the subcorpora, instructions for use and a description of the status of English in each of the contributing countries. The manual is clear and easy to read, and the detailed table of contents makes it easy to navigate, although the lack of an index requires some flipping back and forth when looking for a specific detail. The editors did an excellent job of indicating aspects of the corpus that may influence results (e.g., a search for the lexical item all will not return any tokens tagged as part of the multiword unit all of a sudden), but a researcher may want to make careful note of these aspects, as they are spread throughout the manual.

The directions for using the corpus software are straightforward, and it is possible to successfully perform corpus searches within 30 min of loading the software. There was one error in the manual—indicating that a capitalized lexical item should be entered to obtain both capitalized and lowercase tokens—which was otherwise clearly and accurately written. The software itself performed well: There were no freezes, and even searches with large result sets were completed in 1–2 min on a laptop computer. For researchers interested in examining lexicogrammatical patterns in learner writing, this is an excellent resource, providing texts from a variety of language backgrounds in an easily navigated interface.