4th International Conference on Writing Analytics: Writing Analytics, Data Minin

This paper reports on an ongoing exploratory project that aims to specify parameters, requirements and resources for a collaborative, large-scale European project that (i) builds a multilingual corpus of student writing in Dutch, English and other European languages and (ii) develops corresponding applications for the digital analysis and evaluation of written language at word, sentence and discourse levels. In order to reach that aim, the project follows a six-pronged approach: (1) establishing a European project group; (2) defining the parameters of the corpus; (3) reviewing the international research literature on written language development; (4) consulting panels of experts in writing education; (5) testing and adapting available digital tools for the analysis and evaluation of written language; and (6) setting an agenda for tool development. The project proposed is innovative in two respects: it addresses a largely neglected area in writing research and writing education, and it promotes the use, adaptation and development of digital tools for the analysis and evaluation of written language by beginning and intermediate student writers, students whose texts present serious challenges for software than has been developed to deal with mature, professional and error-free writing.

The presentation will focus on the preliminary outcomes of the review (3 above), and on the achievements of work in annotating errors in student writing. Texts of almost 400 Dutch students from grades 8, 9 and 10 in two genres (descriptive and argumentative) have been annotated manually for errors in spelling, punctuation, use of vocabulary, sentence grammar, referential and relational coherence, and topic shifts. From the point of view of writing analytics these annotations are necessary, since they help establish corpora of student writing that are suitable for automatic analysis. The manual annotations also establish a standard for machine learning that is directed toward detection and correction of errors.