Error-annotated AN Combinations (COLING 2014)

We have released a dataset of adjective–noun (AN) combinations which, on the one hand, exemplify the typical errors committed by language learners in the choice of content words within such combinations, and, on the other hand, are challenging for an error detection and correction (EDC) system.

Starting from an analysis of the typical errors in AN combinations committed by language learners, we compiled a list of 61 adjectives that often present problems for learners. AN combinations involving those adjectives were then extracted from the unannotated part of the Cambridge Learner Corpus. We have focused on AN combinations previously unseen in the British National Corpus.

This set contains 798 AN combinations. Each combination is marked up as correct or incorrect, and, for the incorrect combinations, the locus of the error is identified (adjective, noun or both), as is the type of confusion involved (incorrect synonym, form-related word, or non-related word). The most appropriate corrections are included in the dataset.

Further details can be found in the document ‘Error annotation in adjective–noun (AN) combinations’, which is included with the dataset.

Download

You may download the dataset if you agree to the licence. Publications using it should include a reference to the following publication:

• Kochmar, Ekaterina and Ted Briscoe: ‘Detecting learner errors in the choice of content words using compositional distributional semantics’, Proceedings of the 25th International Conference on Computational Linguistics (COLING 2014).

CLC-FCE AN Combinations (RANLP 2013)

This dataset of adjective–noun (AN) combinations is extracted from the parsed version of the publicly-available CLC FCE Dataset (see first tab). The error coding is used to divide the set into two subsets — correctly used ANs and those that are annotated as errors due to inappropriate choice of an adjective or/and noun. For the ANs that are used correctly in some contexts and incorrectly in others, the most frequent annotation from the CLC is used. The dataset contains 4681 correct and 530 incorrect combinations.

The set of ANs is further divided into corpus-attested and corpus-unattested examples, where parsed version of the British National Corpus (BNC) is used for reference with the frequency threshold set to 3 occurrences in the corpus.

Both the CLC FCE Dataset and the BNC corpus are lemmatised, tagged and parsed using the RASP system (Briscoe et al., 2006; Andersen et al., 2008).

Download

You may download the adjective–noun Dataset if you agree to the licence. Publications using it should include a reference to the following publication:

• Yannakoudakis, Helen, Ted Briscoe and Ben Medlock: ‘A New Dataset and Method for Automatically Grading ESOL Texts’, Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies.

Cambridge English Readability Dataset

The Cambridge English Readability Dataset is composed of reading passages from the five main suite Cambridge English Exams (KET, PET, FCE, CAE, CPE). These five exams are targeted at learners at A2–C2 levels of the Common European Framework of Reference (CEFR). The documents are harvested from all the tasks in the past reading papers for each of the exams. The Cambridge English Exams are designed for L2 learners specifically and the A2–C2 levels assigned to each reading paper can be treated as the level of reading difficulty of the documents for the L2 learners.

Download

You may download the Cambridge English Readability Dataset if you agree to the licence.

Publications using it must include a reference to the following publication:

• Menglin Xia, Ekaterina Kochmar and Ted Briscoe (2016). Text Readability Assessment for Second Language Learners. Proceedings of the 11th Workshop on Innovative Use of NLP for Building Educational Applications.