synterr: rule-grounded synthetic error generation for Russian GEC

Fuente: Zenodo
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Smirnova, Anna, Kopan, Artyom, Makeev, Vladislav, Chernishev, George
Format: Recurso digital
Langue:anglais
Publié: Zenodo 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866901729570717696
author Smirnova, Anna
Kopan, Artyom
Makeev, Vladislav
Chernishev, George
author_facet Smirnova, Anna
Kopan, Artyom
Makeev, Vladislav
Chernishev, George
contents <p>synterr is an open-source synthetic-error generator for Russian grammatical error correction. It corrupts clean text into realistic learner-like errors and labels each error with the prescriptive rule it violates, anchored in a 98-category taxonomy from Rozental's reference grammar.</p><p>Distinguishing properties: every error carries a defensible taxonomic label (Rozental §, RLC tag, or ERRANT tag) and a syntactic justification (dependency-tree-driven handlers for agreement, government, and punctuation). Output formats include GECToR token tags, parallel TSV, rich JSONL with rule labels, instruction-tuning chat format, and rule-targeted SFT.</p><p>This release accompanies the BEA 2026 paper <em>What Aggregate Scores Hide: Per-Rule Evaluation of Russian Grammatical Error Correction</em>.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_20182862
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle synterr: rule-grounded synthetic error generation for Russian GEC
Smirnova, Anna
Kopan, Artyom
Makeev, Vladislav
Chernishev, George
grammatical error correction
GEC
Russian
synthetic data
data augmentation
natural language processing
NLP
Rozental
rule-based generation
dependency parsing
<p>synterr is an open-source synthetic-error generator for Russian grammatical error correction. It corrupts clean text into realistic learner-like errors and labels each error with the prescriptive rule it violates, anchored in a 98-category taxonomy from Rozental's reference grammar.</p><p>Distinguishing properties: every error carries a defensible taxonomic label (Rozental §, RLC tag, or ERRANT tag) and a syntactic justification (dependency-tree-driven handlers for agreement, government, and punctuation). Output formats include GECToR token tags, parallel TSV, rich JSONL with rule labels, instruction-tuning chat format, and rule-targeted SFT.</p><p>This release accompanies the BEA 2026 paper <em>What Aggregate Scores Hide: Per-Rule Evaluation of Russian Grammatical Error Correction</em>.</p>
title synterr: rule-grounded synthetic error generation for Russian GEC
topic grammatical error correction
GEC
Russian
synthetic data
data augmentation
natural language processing
NLP
Rozental
rule-based generation
dependency parsing
url https://doi.org/10.5281/zenodo.20182862