2.1.0: run 2026-11-09-0626 (gold 182/104/6/0, Tatoeba 0.43%) + guard: a noun followed by its own verb is not a doubled word
Browse files- README.md +15 -11
- config.json +0 -0
- gector-guards.json +2 -1
- gector-meta.json +5 -5
- model.onnx +1 -1
README.md
CHANGED
|
@@ -22,7 +22,7 @@ datasets:
|
|
| 22 |
|
| 23 |
**Try it in your browser:** type a sentence and see the corrections. The model runs on your own computer; nothing to install, and your text never leaves the page.
|
| 24 |
|
| 25 |
-
**Version 2.
|
| 26 |
|
| 27 |
A GECToR-style sequence tagger for German grammatical error correction, exported to ONNX (fp32) to run on the CPU.
|
| 28 |
It is the on-device grammar model of [Proofbird](https://proofbird.app), a writing assistant for Windows, and it is
|
|
@@ -81,28 +81,32 @@ words, and capitalisation at the start of a sentence.
|
|
| 81 |
- **Data: synthetic only.** 3 000 000 sentences from German Wikipedia
|
| 82 |
([`wikimedia/wikipedia`](https://huggingface.co/datasets/wikimedia/wikipedia), `20231101.de`) and 250 000 sentences
|
| 83 |
from [Tatoeba](https://tatoeba.org) were corrupted by hand-written German error rules; about 10% stay clean, so the
|
| 84 |
-
model learns to leave correct text alone.
|
| 85 |
contains (`du` and `ihr` with their verb forms, `das` / `dass` with the same pronouns on both sides, tag questions,
|
| 86 |
-
errors in a later clause), with correct twins so the model also learns
|
|
|
|
| 87 |
- **No learner corpora** and no non-commercial data of any kind. No user text was used.
|
| 88 |
-
- One epoch (95
|
| 89 |
|
| 90 |
## How it was measured
|
| 91 |
|
| 92 |
| Check | Result |
|
| 93 |
|---|---|
|
| 94 |
-
| Gold set (
|
| 95 |
-
| False alarms on 4 000 everyday Tatoeba sentences (held out from training) | 17 changed (0.43%);
|
| 96 |
-
| Stability: the gold re-run with one extra token in front of each sentence |
|
| 97 |
|
| 98 |
The gold score is with the apply-side guards in `gector-guards.json` (below). The model was trained for precision: a
|
| 99 |
missed error costs less than a wrong correction. The Tatoeba changes include real fixes of Tatoeba's text (`muß` →
|
| 100 |
-
`muss`, `Du wird` → `wirst`, `Unsere Lehrer ist` → `sind`) and
|
| 101 |
-
and `
|
| 102 |
-
|
|
|
|
|
|
|
| 103 |
The guards keep `man` with the Konjunktiv I of instructions (`Man nehme zwei Eier`), `das das` after a form of `sein`
|
| 104 |
(`Ist das das Haus von Anna?`), a relative pronoun and an article after a comma (`diejenige, die die Entscheidungen
|
| 105 |
-
trifft`) and a plural after `die`, `meine` and similar
|
|
|
|
| 106 |
throughout with `ss` instead of `ß` (Swiss spelling) is left alone. It has not been measured on learner essays.
|
| 107 |
|
| 108 |
## Files
|
|
|
|
| 22 |
|
| 23 |
**Try it in your browser:** type a sentence and see the corrections. The model runs on your own computer; nothing to install, and your text never leaves the page.
|
| 24 |
|
| 25 |
+
**Version 2.1.0** · run `2026-11-09-0626` (2.0.0 = run `2026-11-08-1901`, 1.x = run `2026-11-08-0749`) · tag [`v2.1.0`](https://huggingface.co/proofbird/gector-bert-de/tree/v2.1.0), the files the Proofbird app downloads. Pin it with `revision="v2.1.0"`. Versions: a change to the model's design (encoder, inputs, outputs) raises the first number, a new training run or changed guards and rules the second, card or metadata fixes the third.
|
| 26 |
|
| 27 |
A GECToR-style sequence tagger for German grammatical error correction, exported to ONNX (fp32) to run on the CPU.
|
| 28 |
It is the on-device grammar model of [Proofbird](https://proofbird.app), a writing assistant for Windows, and it is
|
|
|
|
| 81 |
- **Data: synthetic only.** 3 000 000 sentences from German Wikipedia
|
| 82 |
([`wikimedia/wikipedia`](https://huggingface.co/datasets/wikimedia/wikipedia), `20231101.de`) and 250 000 sentences
|
| 83 |
from [Tatoeba](https://tatoeba.org) were corrupted by hand-written German error rules; about 10% stay clean, so the
|
| 84 |
+
model learns to leave correct text alone. 50 000 additional template sentences teach patterns Wikipedia rarely
|
| 85 |
contains (`du` and `ihr` with their verb forms, `das` / `dass` with the same pronouns on both sides, tag questions,
|
| 86 |
+
errors in a later clause, and the same errors inside long sentences), with correct twins so the model also learns
|
| 87 |
+
what to leave alone.
|
| 88 |
- **No learner corpora** and no non-commercial data of any kind. No user text was used.
|
| 89 |
+
- One epoch (95 794 steps, batch 64, learning rate 3e-5) on one NVIDIA L4. Run id `2026-11-09-0626`.
|
| 90 |
|
| 91 |
## How it was measured
|
| 92 |
|
| 93 |
| Check | Result |
|
| 94 |
|---|---|
|
| 95 |
+
| Gold set (292 hand-written sentences: 188 with one error each, 104 correct) | **182 fixed, 104 left alone, 6 missed, 0 made worse** |
|
| 96 |
+
| False alarms on 4 000 everyday Tatoeba sentences (held out from training) | 17 changed (0.43%); 14 of them correct an error or an old spelling in Tatoeba's own text |
|
| 97 |
+
| Stability: the gold re-run with one extra token in front of each sentence | 1 sentence changes result |
|
| 98 |
|
| 99 |
The gold score is with the apply-side guards in `gector-guards.json` (below). The model was trained for precision: a
|
| 100 |
missed error costs less than a wrong correction. The Tatoeba changes include real fixes of Tatoeba's text (`muß` →
|
| 101 |
+
`muss`, `Du wird` → `wirst`, `Unsere Lehrer ist` → `sind`) and three false alarms, for example `den Talern` → `Tälern`
|
| 102 |
+
and `der jüngere der Brüder` → `Jüngere`. Since 2.0.0 it also fixes `nicht war` and `ß` inside long sentences
|
| 103 |
+
(`Er ist schon seit gestern krank, nicht war?`). Known misses: `schon` → `schön` without a word like `sehr` or `wirklich`
|
| 104 |
+
before it (`Die Blumen sind trotzdem schon.`; `Die Blumen sind wirklich schon.` is fixed), and `hätte` → `hatte`, which
|
| 105 |
+
the model leaves alone because both are often right.
|
| 106 |
The guards keep `man` with the Konjunktiv I of instructions (`Man nehme zwei Eier`), `das das` after a form of `sein`
|
| 107 |
(`Ist das das Haus von Anna?`), a relative pronoun and an article after a comma (`diejenige, die die Entscheidungen
|
| 108 |
+
trifft`), a noun followed by its own verb (`Wir müssen das Lernen lernen`) and a plural after `die`, `meine` and similar
|
| 109 |
+
words (`Ich stelle hier die Fragen`). A sentence written
|
| 110 |
throughout with `ss` instead of `ß` (Swiss spelling) is left alone. It has not been measured on learner essays.
|
| 111 |
|
| 112 |
## Files
|
config.json
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
gector-guards.json
CHANGED
|
@@ -45,7 +45,8 @@
|
|
| 45 |
"type": "delete_only_duplicates",
|
| 46 |
"words": "das|die|der|den|dem",
|
| 47 |
"whenPrevIs": "ist|war|wäre|sei|wird|bleibt|,",
|
| 48 |
-
"
|
|
|
|
| 49 |
},
|
| 50 |
{
|
| 51 |
"type": "redirect_replace",
|
|
|
|
| 45 |
"type": "delete_only_duplicates",
|
| 46 |
"words": "das|die|der|den|dem",
|
| 47 |
"whenPrevIs": "ist|war|wäre|sei|wird|bleibt|,",
|
| 48 |
+
"keepCaseDiffering": true,
|
| 49 |
+
"note": "a $DELETE is honoured only on a repeated word ('ist ist'); never on a real word ('wirklich echt?'); 'das das' after a form of sein is a demonstrative + an article (Ist das das Haus?) and stays (DE 1.1.0); so does a relative pronoun + article after a comma (diejenige, die die Entscheidungen trifft, DE 2.0.0) - Mach bitte die die Tür zu still loses one; a pair differing only in case mid-sentence is a noun + its verb (Wir müssen das Lernen lernen, DE 3.0.0)"
|
| 50 |
},
|
| 51 |
{
|
| 52 |
"type": "redirect_replace",
|
gector-meta.json
CHANGED
|
@@ -18,20 +18,20 @@
|
|
| 18 |
"min_error_prob": 0.0,
|
| 19 |
"n_iteration": 5
|
| 20 |
},
|
| 21 |
-
"run_id": "2026-11-
|
| 22 |
-
"exported_at": "2026-11-
|
| 23 |
"checkpoint": "ckpt_best.pt",
|
| 24 |
"git_commit": null,
|
| 25 |
"recipe_sha256": {
|
| 26 |
"build_dataset.py": "311ab76557eb",
|
| 27 |
"preprocess.py": "384e6c617e5e",
|
| 28 |
"train.py": "489238d21b2c",
|
| 29 |
-
"export_onnx.py": "
|
| 30 |
"runrecord.py": "718118cd4cb0",
|
| 31 |
"tatoeba_split.py": "4183ce16b93d",
|
| 32 |
-
"corrupt_de.py": "
|
| 33 |
"morph_de.py": "a9361cce0af0",
|
| 34 |
"language-rules.de.json": "901b350a9c75"
|
| 35 |
},
|
| 36 |
-
"version": "2.
|
| 37 |
}
|
|
|
|
| 18 |
"min_error_prob": 0.0,
|
| 19 |
"n_iteration": 5
|
| 20 |
},
|
| 21 |
+
"run_id": "2026-11-09-0626",
|
| 22 |
+
"exported_at": "2026-11-09T10:33:32+00:00",
|
| 23 |
"checkpoint": "ckpt_best.pt",
|
| 24 |
"git_commit": null,
|
| 25 |
"recipe_sha256": {
|
| 26 |
"build_dataset.py": "311ab76557eb",
|
| 27 |
"preprocess.py": "384e6c617e5e",
|
| 28 |
"train.py": "489238d21b2c",
|
| 29 |
+
"export_onnx.py": "51b98fd3d7a6",
|
| 30 |
"runrecord.py": "718118cd4cb0",
|
| 31 |
"tatoeba_split.py": "4183ce16b93d",
|
| 32 |
+
"corrupt_de.py": "9ab57aed5ce8",
|
| 33 |
"morph_de.py": "a9361cce0af0",
|
| 34 |
"language-rules.de.json": "901b350a9c75"
|
| 35 |
},
|
| 36 |
+
"version": "2.1.0"
|
| 37 |
}
|
model.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 452972247
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6d7b43b4bd95c7a00c02c8babc7569d28558df894afe4874d96c202e6ab9a464
|
| 3 |
size 452972247
|