From 2ecb4e0a5438a5cbc4293f607134671ce3d97fd8 Mon Sep 17 00:00:00 2001 From: egenthon-cmd Date: Mon, 9 Nov 2026 16:33:32 -0500 Subject: [PATCH 1/2] fix: drop dataset-specific French sentence replacements Remove bulletins, zaka nino, and c est cher mappings that collapse distinct ASR surface forms. Co-authored-by: Cursor --- normalization/languages/french/sentence_replacements.py | 3 --- 1 file changed, 3 deletions(-) diff --git a/normalization/languages/french/sentence_replacements.py b/normalization/languages/french/sentence_replacements.py index 5b60a49..585e074 100644 --- a/normalization/languages/french/sentence_replacements.py +++ b/normalization/languages/french/sentence_replacements.py @@ -12,8 +12,5 @@ "pour cent": "pourcent", # ASR / WER canonicalizations "my people doc": "mypeopledoc", - "bulletins de salaires": "bulletins de salaire", - "zaka nino": "zaccanino", - "c est cher": "c'est cher", "releve d heure": "releve d heures", } From a72495efb09cb0cae11b8fdbcc97ff67451c3551 Mon Sep 17 00:00:00 2001 From: egenthon-cmd Date: Mon, 9 Nov 2026 16:40:09 -0500 Subject: [PATCH 2/2] fix: map French square metres phrases to m2, not bare metres MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Word-level metres→m2 wrongly treated linear metres as area; use sentence rules for carré forms only. Co-authored-by: Cursor --- normalization/languages/french/replacements.py | 1 - normalization/languages/french/sentence_replacements.py | 4 ++++ 2 files changed, 4 insertions(+), 1 deletion(-) diff --git a/normalization/languages/french/replacements.py b/normalization/languages/french/replacements.py index dc2a6a6..4e166c9 100644 --- a/normalization/languages/french/replacements.py +++ b/normalization/languages/french/replacements.py @@ -32,7 +32,6 @@ "alpiq": "alpic", "traitee": "traite", "kilowattheures": "kilowattheure", - "metres": "m2", "monsieur": "m", "ouais": "oui", } diff --git a/normalization/languages/french/sentence_replacements.py b/normalization/languages/french/sentence_replacements.py index 585e074..36b68db 100644 --- a/normalization/languages/french/sentence_replacements.py +++ b/normalization/languages/french/sentence_replacements.py @@ -13,4 +13,8 @@ # ASR / WER canonicalizations "my people doc": "mypeopledoc", "releve d heure": "releve d heures", + # Surface area: converge spoken form and symbol (not bare ``metres``) + "metres carres": "m2", + "metre carre": "m2", + "metres carre": "m2", }