Papers by Alla Rozovskaya
New Dataset and Strong Baselines for the Grammatical Error Correction of Russian (2021.findings-acl)
Copied to clipboard
| Challenge: | a new resource is created to evaluate grammatical error correction models in English . a subset of the dataset is annotated in Russian, which is hard to come by and expensive to annotate . |
| Approach: | They develop an annotated learner corpus of Russian extracted from the Lang-8 website. |
| Outcome: | The proposed dataset is compared against two state-of-the-art grammatical error correction models . the results show that the created corpus is more diverse than the existing one . |
Multi-Reference Benchmarks for Russian Grammatical Error Correction (2024.eacl-long)
Copied to clipboard
| Challenge: | Using the union of the references increases system scores by more than 10 points, but not across error types. |
| Approach: | They propose multi-reference benchmarks for the Grammatical Error Correction of Russian . they use two existing single-refer datasets for a total of 7,444 learner sentences . |
| Outcome: | The proposed benchmarks show that new raters tend to make more changes, especially at the lexical level, compared to the original rater. |
Using Neural Machine Translation for Generating Diverse Challenging Exercises for Language Learner (2023.acl-long)
Copied to clipboard
| Challenge: | a common challenge for language learners is understanding how to appropriately use words that may have similar meanings but are used in different contexts. |
| Approach: | They propose a method to automatically generate distractors for cloze exercises for English language learners using round-trip neural machine translation. |
| Outcome: | The proposed method generates distractors for cloze exercises for English learners . it shows that the generated distractors are of the same difficulty as human distractors . |
Automatic Classification of Russian Learner Errors (2022.lrec-1)
Copied to clipboard
| Challenge: | In the field of Grammatical Error Correction, evaluations of system output are performed overall without taking into account performance on individual error types. |
| Approach: | They propose a tool that automatically classifies errors in Russian learner texts . they compare the performance of two error correction systems with a grammatical error category tool . |
| Outcome: | The proposed tool shows that 93% of errors are judged correct or acceptable . the proposed edits are "reallistic", "realistic", and "a" ) and "are" - "were" |
Automatic Generation of Distractors for Fill-in-the-Blank Exercises with Round-Trip Neural Machine Translation (2022.acl-srw)
Copied to clipboard
| Challenge: | a fill-in-the-blank exercise involves removing one word from a sentence and generating distractors . a valid distractor is a word that does not fit the context, and distractors are invalid . |
| Approach: | They propose to automatically generate distractors using round-trip neural machine translation . they show that using hundreds of translations for a given sentence generates a rich set of distractors . |
| Outcome: | The proposed method outperforms two strong baselines against a real corpus of cloze exercises and manually checks for validity. |
Toward Robust Evaluation for Multilingual Grammatical Error Correction: Can Large Language Models Replace Human References? (2026.acl-long)
Copied to clipboard
| Challenge: | Prior work has shown that using aclosest-gold reference yields more accurate performance estimates, but producing such references for each system individually is costly. |
| Approach: | They propose a method for generating closest-gold references by prompting a large language model with system outputs and a standard reference-based evaluations show weak or no correlation. |
| Outcome: | The proposed method outperforms state-of-the-art models on 14 languages across 14 benchmarks. |
Low-Resource Grammatical Error Correction: Selective Data Augmentation with Round-Trip Machine Translation (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for grammatical error correction require large amounts of parallel training data. |
| Approach: | They propose to generate synthetic data through round-trip machine translation by generating a set of character-level errors using a technique known as SeLex-RT. |
| Outcome: | The proposed technique produces errors similar to those observed with language learners, but lacks gold-labeled training data. |
How Good (really) are Grammatical Error Correction Systems? (2021.eacl-main)
Copied to clipboard
| Challenge: | Standard evaluations of Grammatical Error Correction systems use a fixed reference text generated relative to the original text. |
| Approach: | They propose to use a gold reference text to evaluate Grammatical Error Correction systems that is generated relative to the original text and is independent of the system output. |
| Outcome: | The proposed evaluations show that the system performs 20-40 points better than standard evaluations. |
Universal Dependencies for Learner Russian (2024.lrec-main)
Copied to clipboard
| Challenge: | a pilot study of Russian learner data with syntactic dependency relations is presented . a focus of recent work in the NLP community has been on grammar errors . |
| Approach: | They propose to annotate Russian learner data with syntactic dependency relations using a subset of sentences from two error-corrected Russian learners. |
| Outcome: | The proposed annotations are performed on a subset of Russian learner datasets. |