v3: real code-switching vocabulary (intern, sprint, stakeholder...), 500K examples

README.md

@@ -46,9 +46,9 @@ This corpus trains a model to repair that, **without touching text that is genui
 | **Pure Arabic in** | `أنا رايح البيت دلوقتي` |
 | **Pure Arabic out** | `أنا رايح البيت دلوقتي` — unchanged |
 
-- **475,945 examples** across **9 dialect labels**
-- **30% pass-through rows** where the correct answer is *change nothing*
-- **1,380,376 labelled spans**, tagged by category
+- **499,998 examples** across **9 dialect labels**
+- **29% pass-through rows** where the correct answer is *change nothing*
+- **1,416,686 labelled spans**, tagged by category
 - Zero train/test leakage, zero token/tag misalignment (both verified — see *Quality guarantees*)
 
 ---
@@ -121,10 +121,10 @@ The category is part of the label because the same characters convert differentl
 
 | Category | Count | Input | Output |
 |---|---|---|---|
-| `CS` | 1,362,606 | `انترن` | `intern` |
-| `ACRONYM` | 9,001 | `ايه اي` | `AI` |
+| `CS` | 1,394,636 | `انترن` | `intern` |
+| `ACRONYM` | 12,392 | `ايه اي` | `AI` |
 | `EMAIL` | 4,370 | `احمد ات جيميل دوت كوم` | `ahmed@gmail.com` |
-| `ENTITY` | 4,399 | `اورانج انوفيشن ايجيبت` | `Orange Innovation Egypt` |
+| `ENTITY` | 5,288 | `اورانج انوفيشن ايجيبت` | `Orange Innovation Egypt` |
 | `NUMBER` | 0 | `تو زيرو تو فور` | `2024` |
 
 ---
@@ -133,9 +133,9 @@ The category is part of the label because the same characters convert differentl
 
 | Split | Rows | Pass-through | Spans |
 |---|---|---|---|
-| `train` | 428,383 | 128,581 (30.0%) | 1,242,621 |
-| `validation` | 23,595 | 7,061 (29.9%) | 68,259 |
-| `test` | 23,967 | 7,141 (29.8%) | 69,496 |
+| `train` | 450,022 | 130,321 (29.0%) | 1,275,285 |
+| `validation` | 24,842 | 7,159 (28.8%) | 70,162 |
+| `test` | 25,134 | 7,230 (28.8%) | 71,239 |
 
 Splits are assigned by **hashing a normalised sentence key** rather than shuffling, so a rebuild
 reproduces exactly the same test set — important if this corpus is to serve as a shared benchmark.
@@ -145,14 +145,14 @@ reproduces exactly the same test set — important if this corpus is to serve as
 | Code | Dialect | Rows |
 |---|---|---|
 | `unk` | Mixed / unlabelled (monolingual pass-through) | 142,599 |
-| `msa` | Modern Standard Arabic | 109,929 |
-| `egy` | Egyptian | 104,818 |
-| `glf` | Gulf (Saudi/Emirati/Kuwaiti) | 104,004 |
+| `msa` | Modern Standard Arabic | 114,741 |
+| `egy` | Egyptian | 111,590 |
+| `glf` | Gulf (Saudi/Emirati/Kuwaiti) | 109,522 |
+| `lev` | Levantine (Syrian/Lebanese/Jordanian/Palestinian) | 7,823 |
+| `mgr` | Maghrebi (Moroccan/Algerian/Tunisian) | 3,935 |
 | `yem` | Yemeni | 3,566 |
-| `lev` | Levantine (Syrian/Lebanese/Jordanian/Palestinian) | 2,939 |
-| `mgr` | Maghrebi (Moroccan/Algerian/Tunisian) | 2,912 |
+| `irq` | Iraqi | 3,521 |
 | `sdn` | Sudanese | 2,701 |
-| `irq` | Iraqi | 2,477 |
 
 `unk` is monolingual Arabic contributing pass-through examples; it is dialectally mixed and is
 labelled honestly rather than guessed at.
@@ -203,7 +203,7 @@ emails symbol by symbol (`@` → `ات`, `.` → `دوت`), and digits as words
 |---|---|---|
 | [SDAIA ArE-CSTD](https://huggingface.co/datasets/SDAIANCAI/Ar-En-Code-Switching-Textual-Dataset) | 310,578 rows — MSA, Saudi, Egyptian | CC-BY-SA-4.0 |
 | Gemini synthesis | 22,584 rows — the missing dialects and rare categories | generated for this corpus |
-| Arabic Wikipedia + dialectal ASR transcripts | 142,783 rows — pass-through examples | CC-BY-SA |
+| Arabic Wikipedia + dialectal ASR transcripts | 144,710 rows — pass-through examples | CC-BY-SA |
 
 **Why synthesis was necessary.** SDAIA covers three dialects and is almost entirely plain
 word-level code-switching. Measured on 20K of its sentences: ~60,000 ordinary code-switch spans but

dataset_stats.json

@@ -1,38 +1,39 @@
 {
-  "total_seen": 352582,
-  "kept": 475945,
+  "total_seen": 432582,
+  "kept": 499998,
   "skipped_no_latin": 3246,
-  "skipped_duplicate": 16170,
-  "passthrough_examples": 142783,
-  "passthrough_ratio": 0.3,
-  "total_tokens": 5602916,
-  "total_spans": 1380376,
+  "skipped_duplicate": 74044,
+  "passthrough_examples": 144710,
+  "passthrough_ratio": 0.2894,
+  "total_tokens": 5745052,
+  "total_spans": 1416686,
   "by_split": {
-    "train": 428383,
-    "validation": 23595,
-    "test": 23967
+    "train": 450022,
+    "validation": 24842,
+    "test": 25134
   },
   "by_dialect": {
-    "msa": 109929,
-    "egy": 104818,
-    "glf": 104004,
+    "msa": 114741,
+    "egy": 111590,
+    "glf": 109522,
     "sdn": 2701,
     "yem": 3566,
-    "irq": 2477,
-    "mgr": 2912,
-    "lev": 2939,
+    "irq": 3521,
+    "mgr": 3935,
+    "lev": 7823,
     "unk": 142599
   },
   "by_source": {
     "sdaia": 310578,
     "gemini": 22584,
+    "vocab": 22126,
     "monolingual": 142599,
-    "sdaia-mono": 184
+    "sdaia-mono": 2111
   },
   "by_category": {
-    "CS": 1362606,
-    "ACRONYM": 9001,
-    "ENTITY": 4399,
+    "CS": 1394636,
+    "ACRONYM": 12392,
+    "ENTITY": 5288,
     "EMAIL": 4370
   }
 }

test.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa830dacd06165c4bfceccedf2d7fa37f27ddc7ed4647a9774fff0a07d50f106
-size 3410183
+oid sha256:d74446addf1897646dd218e0383b7479bc6bfa3f9cc888e9243904bd479f35d5
+size 3491165

train.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1da670d5b822c33438b95cfc4f4e16df2ddfbcb8b86c3aaf6273cf3b53ab7245
-size 63554055
+oid sha256:10dd8994e73b96b471f45bcf3c35d296cd89bb37c28be48b535d259acad383b6
+size 65349126

validation.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fbd32d077933b0039a669be0977edeb2ae7ed035e50aa607dd5f2140da6085f4
-size 3341783
+oid sha256:af0579427b1151726bbcee38cd630ea06c500195db000ae8d4b424a255c97c1a
+size 3437466