Article Structure
Abstract
This work explores methods of automatically detecting corrections of individual mistakes in sentence revisions for ESL students.
Introduction
Quality feedback from language tutors can help English-as-a—Second-Language (ESL) students improve their writing skills.
Correction Detection
Comparing a student-written sentence with its revision, we observe that each correction can be decomposed into a set of more basic edits such as word insertions, word deletions and word substitutions.
A Classifier for Merging Basic-Edits
Figures 4 highlights the problems with indiscriminantly merging basic-edits that are adjacent.
Experimental Setup
We combine Levenshtein algorithm with different merging algorithms for correction detection.
Experiments We design experiments to answer two questions:
1.
Conclusions
A revision often contains multiple corrections that address different writing mistakes.
Topics
end-to-end
- In comparison, phrase extraction systems aim to improve the end-to-end MT or paraphrasing systems.Page 3, “Correction Detection”
- In addition to evaluating the merging algorithms on the standalone task of correction detection, we have also plugged in the merging algorithms into an end-to-end system in which every automatically detected correction is further classified into an error type.Page 4, “Experimental Setup”
- 5 We currently do not evaluate the end-to-end system over different corpora.Page 5, “Conclusions”
- Table 4: Extrinsic evaluation, where we plugged the two merging models into an end-to-end feedback detection system by Swanson and Yamangil.Page 5, “Conclusions”
See all papers in Proc. ACL 2014 that mention end-to-end.
See all papers in Proc. ACL that mention end-to-end.
Back to top.
gold standard
- Unlike evaluating grammar error correction systems (Dahlmeier and Ng, 2012), correction detection cannot refer to a gold standard .Page 3, “Correction Detection”
- However, existing revision corpora do not have the fine-grained annotations necessary for our experimental gold standard .Page 3, “Experimental Setup”
- The teachers’ annotations are treated as gold standard for the detailed corrections.Page 3, “Experimental Setup”
- comparing the boundaries of system’s detected corrections with the gold standard .Page 4, “Experimental Setup”
See all papers in Proc. ACL 2014 that mention gold standard.
See all papers in Proc. ACL that mention gold standard.
Back to top.
cross validation
- One part is used as the development dataset; the rest are used for 10-fold cross validation .Page 4, “Experimental Setup”
- The reported figures come from 10-fold cross validations on different corpora.Page 4, “Experimental Setup”
- When training and testing on the same corpus, we run a 10-fold cross validation .Page 5, “Conclusions”
See all papers in Proc. ACL 2014 that mention cross validation.
See all papers in Proc. ACL that mention cross validation.
Back to top.
Maximum Entropy
- To predict whether two basic-edits address the same writing problem more discriminatively, we train a Maximum Entropy binary classifier based on features extracted from relevant contexts for the basic edits.Page 3, “A Classifier for Merging Basic-Edits”
- MaXEntMerger We use the Maximum Entropy classifier to predict whether we should merge the two edits, as described in Section 34.Page 4, “Experimental Setup”
- We use a Maximum Entropy classifier along with features suggested by Swanson and Yamangil for this task.Page 4, “Experimental Setup”
See all papers in Proc. ACL 2014 that mention Maximum Entropy.
See all papers in Proc. ACL that mention Maximum Entropy.
Back to top.
