HOO-2012 Shared Task Preliminary Call for Participation: ESL Preposition and Determiner Error Correction
Preliminary Call for Participation
The HOO (Helping Our Own) Exercise (see [Dale and Kilgarriff 2010]) is concerned with correcting textual errors. The HOO Pilot Shared Task (see [Dale and Kilgarriff 2011]) looked at a diverse range of error types in a small set of documents. HOO 2012, which will be hosted by the Building Educational Applications Workshop at NAACL 2012, focusses on the correction of preposition and determiner errors in a large collection of non-native speaker texts. These are widely recognized to be amongst the most challenging aspects of English lexico-syntax for non-native speakers to deal with: see [Leacock et al. 2010] for a review.
The goal of this task is to provide a forum for the comparative evaluation of approaches to the correction of errors in the use of prepositions and determiners by non-native speakers of English. Although these have already been the focus of a considerable body of research in natural language processing, so far it has been hard to compare the results delivered by different teams as a consequence of different data sets and slightly different task descriptions. This shared task provides a common dataset and a shared evaluation framework as a means of overcoming these problems.
The HOO-2012 Preps and Dets Shared Task follows on from the HOO-2011 Shared Task Pilot Round held in 2011 as part of the 2011 European Natural Language Generation Workshop. That task had a much broader focus on all kinds of errors in non-native speaker writing, and use a much smaller dataset. The evaluation framework for HOO-2012 is an enhancement of the scheme developed for HOO-2011, taking advantage of what was learned in that exercise.
The data to be used for the task is drawn from a subset of the Cambridge Learner Corpus (CLC) described in [Yannakoudakis et al 2011]. The data, which contains exam scripts written by students undertaking the First Certificate in English (FCE) exams, is jointly provided by Cambridge ESOL and Cambridge University Press.
The data we are using has been converted from the mark-up provided in the released version of the CLC FCE data to use the HOO annotation scheme (described here).
The data to be released for training consists of 1000 exam scripts extracted from the FCE dataset.
A further subset of 100 exam scripts will be released for testing and evaluation at the appropriate point in the schedule. We are endeavouring to obtain fresh data for this stage of the exercise, but in the event that this turns out not to be possible, we will use part of the published FCE dataset that was held back from the training data. For this reason, we would ask participants not to use the originally-released FCE dataset for training purposes, but only the HOO-formatted subset that we release independently.
The evaluation methodology is essentially the same as that used in the HOO Pilot Round, but limited to preposition and determiner errors only. Tools will be provided which compute detection (lenient recognition, with at least one character overlap), recognition (exact extent recognition of an error), and correction (provision of an appropriate replacement string in addition to recognition).
What You Should Do Now
Register your interesting by sending your email address to email@example.com. You'll be added to the HOO Google Groups list, where over the next few weeks some fine-tuning of the shared task will be discussed. Formal registration for the task will be as indicated in the schedule below.
The current schedule for HOO-2012 is as follows.
- Friday 27th January: Website registration for participation in HOO-2012 opens; development data for the Shared Task released.
- Friday 6th April: Test data for evaluation released.
- Friday 13th April: Deadline for submissions from teams for evaluation.
- Monday 23rd April: Results of evaluation released.
- Friday May 4th: Final versions of team reports for proceedings due.
This site will be frequently updated over the month of December 2011; last updated Friday 23rd December 2011.
Robert Dale, Macquarie University
R. Dale and A. Kilgarriff  Helping Our Own: Text massaging for computational linguistics as a new shared task. In Proceedings of the 6th International Natural Language Generation Conference, pages 261–265, Dublin, Ireland, 7th–9th July 2010.
R. Dale and A. Kilgarriff  Helping Our Own: The HOO 2011 Pilot Shared Task. In Proceedings of the 13th European Workshop on Natural Language Generation, Nancy, France, 28th–30th September 2011.
C. Leacock, M. Chodorow, M. Gamon, and J. Tetreault  Automated Grammatical Error Detection for Language Learners. Synthesis Lectures on Human Language Technologies. Morgan and Claypool.
H. Yannakoudakis, T. Briscoe and B. Medlock  A New Dataset and Method for Automatically Grading ESOL Texts. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, Portland, Oregon, USA, 19th–24th June 2011. .