This paper presents multi-reference benchmarks for the Grammatical Error Correction (GEC) of Russian, based on two existing single-reference datasets, for a total of 7,444 learner sentences from a variety of first language backgrounds. Each sentence is corrected independently by two new raters, and their corrections are reviewed by a senior annotator, resulting in a total of three references per sentence. Analysis of the annotations reveals that the new raters tend to make more changes, compared to the original raters, especially at the lexical level. We conduct experiments with two popular GEC approaches and show competitive performance on the original datasets and the new benchmarks. We also compare system scores as evaluated against individual annotators and discuss the effect of using multiple references overall and on specific error types. We find that using the union of the references increases system scores by more than 10 points and decreases the gap between system and human performance, thereby providing a more realistic evaluation of GEC system performance, although the effect is not the same across the error types. The annotations are available for research.

We present our system that participated in the shared task on the grammatical error correction of Ukrainian. We have implemented two approaches that make use of large pre-trained language models and synthetic data, that have been used for error correction of English as well as low-resource languages. The first approach is based on fine-tuning a large multilingual language model (mT5) in two stages: first, on synthetic data, and then on gold data. The second approach trains a (smaller) seq2seq Transformer model pre-trained on synthetic data and fine-tuned on gold data. Our mT5-based model scored first in “GEC only” track, and a very close second in the “GEC+Fluency” track. Our two key innovations are (1) finetuning in stages, first on synthetic, and then on gold data; and (2) a high-quality corruption method based on roundtrip machine translation to complement existing noisification approaches.

We propose a novel approach to automatically generate distractors for cloze exercises for English language learners, using round-trip neural machine translation. A carrier sentence is translated from English into another (pivot) language and back, and distractors are produced by aligning the original sentence with its round-trip translation. We make use of 16 linguistically-diverse pivots and generate hundreds of translation hypotheses in each direction. We show that using hundreds of translations allows us to generate a rich set of challenging distractors. Moreover, we find that typologically unrelated language pivots contribute more diverse candidate distractors, compared to language pivots that are closely related. We further evaluate the use of machine translation systems of varying quality and find that better quality MT systems produce more challenging distractors. Finally, we conduct a study with language learners, demonstrating that the automatically generated distractors are of the same difficulty as the gold distractors produced by human experts.

In a fill-in-the-blank exercise, a student is presented with a carrier sentence with one word hidden, and a multiple-choice list that includes the correct answer and several inappropriate options, called distractors. We propose to automatically generate distractors using round-trip neural machine translation: the carrier sentence is translated from English into another (pivot) language and back, and distractors are produced by aligning the original sentence and its round-trip translation. We show that using hundreds of translations for a given sentence allows us to generate a rich set of challenging distractors. Further, using multiple pivot languages produces a diverse set of candidates. The distractors are evaluated against a real corpus of cloze exercises and checked manually for validity. We demonstrate that the proposed method significantly outperforms two strong baselines.