| { | |
| "version": 1, | |
| "language": "en", | |
| "_comment": "Per-bundle language facts (see C# LanguageRules / ADR-004). stripTrailingTerminator true: the shippable EN model is GECToR BERT/WordPiece (same recipe as FIGS+PT, which is period-brittle) — validate at eval. NOTE: the non-shippable dev gector-roberta-base-en bundle is RoBERTa and NOT period-brittle; its locally-installed copy of this file keeps stripTrailingTerminator false. clauseConjunctions: connectors that, straight after a comma, mark an independent-clause boundary — GectorAnalyzer.Correct splits there and corrects each clause alone, recovering errors the model under-detects right before a comma (training gap; measured on Italian comma_context 14/6 -> 20/0). 'for' is deliberately absent (', for example' would false-split). Validate against lang/en/gold comma_context at eval.", | |
| "stripTrailingTerminator": true, | |
| "clauseConjunctions": ["but", "and", "or", "nor", "so", "yet", "then", "because", "while", "whereas", "although", "though", "however", "therefore", "meanwhile", "instead", "otherwise"] | |
| } | |