Upload 8 files
Browse files- .gitattributes +3 -55
- MSA_test.json +0 -0
- MSA_train.json +3 -0
- MSA_val.json +3 -0
- README.md +76 -106
- spoken_test.json +0 -0
- spoken_train.json +3 -0
- spoken_val.json +0 -0
.gitattributes
CHANGED
|
@@ -1,55 +1,3 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
| 5 |
-
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
| 6 |
-
*.ftz filter=lfs diff=lfs merge=lfs -text
|
| 7 |
-
*.gz filter=lfs diff=lfs merge=lfs -text
|
| 8 |
-
*.h5 filter=lfs diff=lfs merge=lfs -text
|
| 9 |
-
*.joblib filter=lfs diff=lfs merge=lfs -text
|
| 10 |
-
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
| 11 |
-
*.lz4 filter=lfs diff=lfs merge=lfs -text
|
| 12 |
-
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
| 13 |
-
*.model filter=lfs diff=lfs merge=lfs -text
|
| 14 |
-
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
| 15 |
-
*.npy filter=lfs diff=lfs merge=lfs -text
|
| 16 |
-
*.npz filter=lfs diff=lfs merge=lfs -text
|
| 17 |
-
*.onnx filter=lfs diff=lfs merge=lfs -text
|
| 18 |
-
*.ot filter=lfs diff=lfs merge=lfs -text
|
| 19 |
-
*.parquet filter=lfs diff=lfs merge=lfs -text
|
| 20 |
-
*.pb filter=lfs diff=lfs merge=lfs -text
|
| 21 |
-
*.pickle filter=lfs diff=lfs merge=lfs -text
|
| 22 |
-
*.pkl filter=lfs diff=lfs merge=lfs -text
|
| 23 |
-
*.pt filter=lfs diff=lfs merge=lfs -text
|
| 24 |
-
*.pth filter=lfs diff=lfs merge=lfs -text
|
| 25 |
-
*.rar filter=lfs diff=lfs merge=lfs -text
|
| 26 |
-
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
| 27 |
-
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
| 28 |
-
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
| 29 |
-
*.tar filter=lfs diff=lfs merge=lfs -text
|
| 30 |
-
*.tflite filter=lfs diff=lfs merge=lfs -text
|
| 31 |
-
*.tgz filter=lfs diff=lfs merge=lfs -text
|
| 32 |
-
*.wasm filter=lfs diff=lfs merge=lfs -text
|
| 33 |
-
*.xz filter=lfs diff=lfs merge=lfs -text
|
| 34 |
-
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 35 |
-
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 36 |
-
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 37 |
-
# Audio files - uncompressed
|
| 38 |
-
*.pcm filter=lfs diff=lfs merge=lfs -text
|
| 39 |
-
*.sam filter=lfs diff=lfs merge=lfs -text
|
| 40 |
-
*.raw filter=lfs diff=lfs merge=lfs -text
|
| 41 |
-
# Audio files - compressed
|
| 42 |
-
*.aac filter=lfs diff=lfs merge=lfs -text
|
| 43 |
-
*.flac filter=lfs diff=lfs merge=lfs -text
|
| 44 |
-
*.mp3 filter=lfs diff=lfs merge=lfs -text
|
| 45 |
-
*.ogg filter=lfs diff=lfs merge=lfs -text
|
| 46 |
-
*.wav filter=lfs diff=lfs merge=lfs -text
|
| 47 |
-
# Image files - uncompressed
|
| 48 |
-
*.bmp filter=lfs diff=lfs merge=lfs -text
|
| 49 |
-
*.gif filter=lfs diff=lfs merge=lfs -text
|
| 50 |
-
*.png filter=lfs diff=lfs merge=lfs -text
|
| 51 |
-
*.tiff filter=lfs diff=lfs merge=lfs -text
|
| 52 |
-
# Image files - compressed
|
| 53 |
-
*.jpg filter=lfs diff=lfs merge=lfs -text
|
| 54 |
-
*.jpeg filter=lfs diff=lfs merge=lfs -text
|
| 55 |
-
*.webp filter=lfs diff=lfs merge=lfs -text
|
|
|
|
| 1 |
+
MSA_train.json filter=lfs diff=lfs merge=lfs -text
|
| 2 |
+
spoken_train.json filter=lfs diff=lfs merge=lfs -text
|
| 3 |
+
MSA_val.json filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
MSA_test.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
MSA_train.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f3bd005333678f5b57ecdb29a24fe4f2391f4dfd380cb856667963a5080ef45f
|
| 3 |
+
size 134
|
MSA_val.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f8b50bc8af01148989819d153074647690e0b30fc0edb1121bbe816933647599
|
| 3 |
+
size 12654174
|
README.md
CHANGED
|
@@ -1,106 +1,76 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
דובר 2: כן, שרה באופן מקצועי)
|
| 78 |
-
|
| 79 |
-
Rejected:
|
| 80 |
-
שאלה: لمين صباح لخير؟ (למי בוקר טוב?)
|
| 81 |
-
תשובה: للجميع (לכולם)
|
| 82 |
-
קונטקסט: صباح الخير للجميع (בוקר טוב לכולם)
|
| 83 |
-
|
| 84 |
-
מדדי דיוק
|
| 85 |
-
בסוף שלב האיסוף נדגמו אקראית מהדאטה-סט כ-330 שאלות ותשובות. הסט הנדגם שימש להערכת איכות הדאטה-סט, אשר נמדדה באמצעות שני פרמטרים שנקבעו מראש:
|
| 86 |
-
א. אחוז שאלות ותשובות תקינות (האם התשובה X אכן עונה לשאלה Y בהינתן פסקת הקונטקסט Z): כ-92% מהשאלות שנדגמו לצורך הבדיקה נמצאו תקינות.
|
| 87 |
-
ב. אחוז תשובות בעלות טווח תשובה תקין מתוך השאלות שנמצאו תקינות (טווח תשובה תקין הוגדר בתור יחידת הטקסט המינימלית בפסקה שמספקת תשובה לשאלה): כ-97% מהשאלות התקינות היו בעלות טווח תשובה מינימלי.
|
| 88 |
-
מתוך השאלות והתשובות שנדגמו בבדיקה זו, כ-71% היו טובות (good), כ-17% עברו סף (verified), כ-4% היו מעולות (gold) וכ-8% נפסלו (rejected).
|
| 89 |
-
|
| 90 |
-
ניקוי הדאטה
|
| 91 |
-
בסיום שלב האיסוף נערכו מספר פעולות ניקוי על הדאטה: נפסלו שאלות זהות, נמחקו רווחים מיותרים בסוף או בתחילת שאלה, נוסף סימן שאלה בשאלות שבהן היה חסר, נוקו תווים מיותרים מהשאלות והתשובות, תוקנו שאלות ללא מילת שאלה, תוקנו תשובות שנחתכו באמצע, ועוד. חשוב לציין שהתיקונים שנעשו היו של דברים שזוהו כארקטיפקטים טכניים שעלולים להפריע למודל. תופעות טבעיות, כמו שגיאות כתיב בשאלות, לא תוקנו.
|
| 92 |
-
|
| 93 |
-
חלוקה לסטים
|
| 94 |
-
לאחר שלב האיסוף, חולק הדאטה-סט לשלושה סטים: סט אימון (90%), סט ולידציה (5%) וסט מבחן (5%). החלוקה של השאלות לסטים הייתה אקראית-נשלטת, כאשר מספר אילוצים על ההתפלגות נקבעו מראש:
|
| 95 |
-
א. כל השאלות שנשאלו על אותו ה"ערך" (כתבה חדשותית בספרותית או פסקה מתוך תמלול במדוברת) הוקצו לאותו סט, על מנת להימנע מאימון המודל על חומר שיהיה בסט המבחן ("זליגת מידע").
|
| 96 |
-
ב. רוב השאלות שקיבלו תווית איכות "gold" במהלך בקרת האיכות הוקצו לתתי-הסטים של הוולידציה והמבחן (ואיתן כמובן הערכים שהכילו אותן בשלמותם). פרמטר זה תורם לאיכותם של תתי-סטים אלה.
|
| 97 |
-
בין שלושת הסטים נשמר יחס קבוע ככל הניתן בין שאלות פתירות לבלתי פתירות, על מנת לא לתת משקל עודף לסוג מסוים של שאלה באחד הסטים.
|
| 98 |
-
לאחר החלוקה, סט המבחן וסט הוולידציה עברו עיבוד נוסף:
|
| 99 |
-
א. בקרה הרמטית של איכות השאלות וטיוב שאלות באופן ידני במידת הצורך (כך שאין בסטים אלו שאלות לא תקינות).
|
| 100 |
-
ב. סימון תשובות נוספות: במקרים שבהם את התשובה לשאלה ניתן לסמן במגוון דרכים נכונות, נוספו אפשרויות סימון נוספות אלו כדי להפוך את הסטים לרובסטיים יותר.
|
| 101 |
-
לדוגמה: אם התשובה נתונה בתוך מרכאות, היא עשויה לכלול או שלא לכלול את המרכאות (ועדיין תהיה נכונה); תשובות עשויות לכלול או שלא לכלול מילות יחס המקדימות את תוכן התשובה (ב-, על-) או רכיבי תמורה ("הד"ר אחמד סעיד" או "אחמד סעיד") וכו'. לכל שאלה פתירה נוספו בין 0 ל-3 אפשרויות נוספות לסימון תשובה.
|
| 102 |
-
|
| 103 |
-
סיכום
|
| 104 |
-
ArQ הוא דאטה-סט גדול ואיכותי מסוג Question Answering הכולל 32,625 שאלות ותשובות בערבית מדוברת של אזור הלבנט ובערבית ספרותית מודרנית סטנדרטית (MSA). אנו תקווה ש-ArQ יתרום לקהילת ה-NLP, ולצד מאמצים נוספים שנעשים במסגרת תוכנית ה-NLP הלאומית, יאפשר לתרום לייצוג של ערבית מדוברת של אזור הלבנט בתחום ה-NLP. אנו מקווים שהדאטה-סט ישמש לקידום תשתיות וטכנולוגיות לעיבוד והבנת השפה הערבית, עליהן ניתן יהיה לבסס יישומים מתקדמים של עיבוד שפה טבעית (מנועי חיפוש, צ'טבוטים, וכיו"ב). בנוסף, אנחנו מקווים שהדאטה-סט שיצרנו, יחד עם הידע והכלים ששיתפנו, יתרמו למחקר בתחום ה-NLP, יהוו השראה ויסייעו ביצירת דאטה-סטים חדשים.
|
| 105 |
-
|
| 106 |
-
תודות לכל השותפים שקצרה היריעה מלהכיל את כולם לפרוייקט חשוב זה
|
|
|
|
| 1 |
+
# ArQ: Arabic Question Answering Dataset
|
| 2 |
+
## Summary
|
| 3 |
+
ArQ is a question answering dataset in Levantine Spoken Arabic and Modern Standard Arabic (MSA), consisting of 32,625 triplets (context-question-answer).
|
| 4 |
+
|
| 5 |
+
## Introduction
|
| 6 |
+
The dataset follows the format and methodology of HeQ (<a href="https://github.com/NNLP-IL/Hebrew-Question-Answering-Dataset">Hebrew Questions and Answers Dataset</a>). A team of annotators were given random context paragraphs in either spoken Arabic or MSA, and were asked to write relevant questions and mark the correct answers. The answer to each question was segment of text (span) included in the relevant paragraph.
|
| 7 |
+
|
| 8 |
+
Paragraphs were sourced using two types of sources: (1) for MSA we used short news articles from an online Israeli-Arabic weekly newspaper, and (2) for spoken Arabic we used transcriptions of short videos and recorded interviews in Levantine Arabic.
|
| 9 |
+
|
| 10 |
+
**Questions on both sources were written in Levantine Spoken Arabic (no MSA questions were written)**.
|
| 11 |
+
|
| 12 |
+
## Question Features
|
| 13 |
+
__Two types of questions were collected:__
|
| 14 |
+
|
| 15 |
+
Answerable questions (24,124; 74%): Questions for which a single correct answer is present in the paragraph.
|
| 16 |
+
|
| 17 |
+
Unanswerable questions (8,501; 26%): Questions related to the paragraph's content, where a correct answer is not present in the paragraph, but the paragraph provides a "plausible" incorrect answer in terms of logic.
|
| 18 |
+
|
| 19 |
+
## Quality Labels
|
| 20 |
+
As part of ongoing quality control during the collection process, and additional checks on the test and validation sets, approximately 12% of the final data was manually chekced for quality.
|
| 21 |
+
|
| 22 |
+
Triplets received one of the following quality labels:
|
| 23 |
+
|
| 24 |
+
*Verified*: Questions that passed the threshold and were relatively easy, with wording exactly or similar to the relevant sentence in the paragraph, or very common questions.
|
| 25 |
+
|
| 26 |
+
*Good*: Questions with wording that was significantly different (lexically or syntactically) from the wording of the relevant sentence in the paragraph.
|
| 27 |
+
|
| 28 |
+
*Gold*: Questions that involve more complex inference-making.
|
| 29 |
+
|
| 30 |
+
*Rejected*: Questions that did not pass the threshold and therefore not indcluded in the published data.
|
| 31 |
+
|
| 32 |
+
## Additional Answers
|
| 33 |
+
After splitting the data, the test and validation subsets underwent additional processing. In cases where there were multiple correct answer spans for an answerable question, the additional possible answer spans were added by annotators to both subsets to enhance robustness.
|
| 34 |
+
|
| 35 |
+
For example, if the answer appears in quotation marks, another possible answer could be the same answer without the quotation marks. Another example involves answers that may or may not include prepositions preceding the content or appositions. Each answerable question in the test and validation sets received 0 to 3 additional possible answers.
|
| 36 |
+
|
| 37 |
+
## Dataset Statistics
|
| 38 |
+
The table below shows the number of answerable and unanswerable questions, by source:
|
| 39 |
+
|
| 40 |
+
| | MSA | Spoken | Total |
|
| 41 |
+
|-----------|-------|--------|--------|
|
| 42 |
+
| Answerable| 12421 | 11703 | 24124 (74%) |
|
| 43 |
+
| Unanswerable| 4425 | 4076 | 8501 (26%) |
|
| 44 |
+
|
| 45 |
+
The table below shows the number of triplets, by sub-set:
|
| 46 |
+
|
| 47 |
+
| | MSA | Spoken | Total |
|
| 48 |
+
|-----------|-------|--------|--------|
|
| 49 |
+
| Train | 15080 | 14197 | 29277 (90%) |
|
| 50 |
+
| Val | 928 | 745 | 1673 (5%) |
|
| 51 |
+
| Test | 838 | 837 | 1675 (5%) |
|
| 52 |
+
|
| 53 |
+
The table below shows the number of unique questions and paragraphs, by source:
|
| 54 |
+
|
| 55 |
+
| | MSA | Spoken |
|
| 56 |
+
|-----------|-------|--------|
|
| 57 |
+
| Questions | 16846 (52%) | 15779 (48%) |
|
| 58 |
+
| Unique Paragraphs | 1016 | 1024 |
|
| 59 |
+
|
| 60 |
+
The table below shows the question word distribution in the dataset:
|
| 61 |
+
|
| 62 |
+
| What | Who | How Much/Many | Which | Where | When | How | Why |
|
| 63 |
+
|-------|-----|---------------|-------|-------|------|-----|-----|
|
| 64 |
+
| 11517 (34%) | 7323 (22%) | 4451 (13%) | 4103 (12%) | 3293 (10%) | 1351 (4%) | 880 (3%) | 700 (2%) |
|
| 65 |
+
|
| 66 |
+
## Code
|
| 67 |
+
upcoming.
|
| 68 |
+
|
| 69 |
+
## Model
|
| 70 |
+
upcoming.
|
| 71 |
+
|
| 72 |
+
## Contributors
|
| 73 |
+
ArQ was annotated by Webiks for MAFAT, as part of <a href="https://github.com/NNLP-IL">NNLP-IL</a>, the Israeli national initiative in the field of NLP in Hebrew and Arabic.
|
| 74 |
+
|
| 75 |
+
Contributors: Amir Shufaniya (Webiks), Carinne Cherf (Webiks) and Yossy Eizenrouah (MAFAT).
|
| 76 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
spoken_test.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
spoken_train.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bb20a36986c27d87485248935b0d46d02917b7999f98465669b9b75b5f3f922d
|
| 3 |
+
size 133
|
spoken_val.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|