You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A third grammar-test probe exposed a different class of nahw failure: the model can often assign a familiar surface label, but it misses hidden grammatical structure or collapses محلّ الإعراب into a visible case ending.
This issue: structures where the visible word is not enough; the skill must detect hidden elements/operators or محل-based parsing.
The three relevant constructions are:
اسم التفضيل as predicate with a hidden grammatical element: زيدٌ أفضلُ.
الاختصاص with a deleted verb/operator: نحن العرب نكرم الضيف.
لا النافية للجنس where the noun is built in محلّ نصب, not normally منصوب: لا رجلَ في البيت.
Source / citation check
Evidence is from the uploaded GrammarProblems.pdf, titled:
اختبار دِقَّة الذكاء الاصطناعي في الإجابة عن مسائل علم النحو العربي - برنامج تشات جي بي تي (ChatGPT) المجاني الإصدار GPT-5 أنموذجًا
Verified PDF locations:
Question 57, printed p. 148, topic: إعمال أفعل التفضيل.
Prompt: في قولنا: زيدٌ أفضلُ، أي الأعاريب الآتية هو الإعراب الصحيح لهذه الجملة؟
Model answer: option (4):
زيدٌ مبتدأ، وأفضلُ خبر، وفيها فاعل مستتر يعود على زيد.
The tested model failed by choosing the flat answer:
زيدٌ مبتدأ، وأفضلُ خبر
and explicitly denied that أفضل needs a hidden فاعل.
Question 59, printed pp. 149–150, topic: الاختصاص.
Prompt: أعرب: نحن العرب نكرم الضيف.
Model answer:
نحن: ضمير منفصل مبني على الضم في محل رفع مبتدأ.
العرب: مفعول به منصوب بفعل محذوف تقديره أخصّ، وعلامة نصبه الفتحة الظاهرة، وهو المشهور بالمنصوب على الاختصاص.
نكرم الضيف: الجملة الفعلية في محل رفع خبر للمبتدأ.
The tested model failed by parsing العرب as بدل من نحن مرفوع مثله.
Question 60, printed p. 150, topic: لا النافية للجنس.
Prompt: أعرب: لا رجل في البيت.
Model answer:
لا: حرف نفي للجنس، ناسخ يعمل عمل إن، مبني على السكون، لا محل له من الإعراب.
رجل: اسم لا النافية للجنس، مبني على الفتح في محل نصب.
في البيت: جار ومجرور متعلقان بخبر محذوف.
The original tested model failed by saying merely:
رجل: اسم لا منصوب، وعلامة نصبه الفتحة.
In the current probe, the model answered this one correctly; keep it as a positive regression so future edits do not break it.
Regression probe that triggered this issue
The following questions were asked as a final set:
6) أعرب: زيدٌ أفضلُ.
7) أعرب: نحن العربَ نكرمُ الضيفَ. What is the iʿrāb of العربَ specifically?
8) أعرب: لا رجلَ في البيت.
The response was:
wrong on زيدٌ أفضلُ, because it gave only a flat مبتدأ + خبر analysis and missed the hidden فاعل/ضمير in أفضل.
wrong on نحن العرب نكرم الضيف, because it parsed العرب as بدل from نحن and even contradicted the supplied fatḥa by calling it مرفوع بالضمة.
correct on لا رجل في البيت, because it correctly said رجل is اسم لا النافية للجنس مبني على الفتح في محل نصب.
So this is a 2/3 failure, with the third item retained as a non-regression success.
Gap 1 — زيدٌ أفضلُ: اسم التفضيل is not only a flat خبر
Observed response
The response parsed:
زيدٌ: مبتدأ مرفوع.
أفضلُ: خبر مرفوع.
Then it described أفضل as simply:
اسم تفضيل على وزن أفعل، وهو هنا خبر.
This is not enough for the PDF’s expected answer.
Correct answer required by the regression
زيدٌ: مبتدأ مرفوع.
أفضلُ: خبر مرفوع، وفيه / فيها فاعل مستتر يعود على زيد.
Alternative acceptable wording:
أفضلُ خبر مرفوع، ويتحمّل ضميرًا مستترًا عائدًا على زيد.
or:
في أفضل ضمير مستتر / فاعل مستتر تقديره هو يعود على زيد.
Why this matters
This is a hidden-structure trap. The surface answer زيد مبتدأ وأفضل خبر is not false as far as it goes, but it is incomplete for advanced iʿrāb. The PDF’s test explicitly rejects the answer that stops there, because the intended analysis requires recognizing the hidden element inside/with أفضل.
For the skill, this means:
Do not flatten all predicates into خبر مرفوع only.
When the خبر is a مشتق-like predicate such as اسم فاعل / صفة مشبهة / اسم تفضيل, check whether it carries a hidden pronoun/فاعل returning to the مبتدأ.
If the prompt asks whether there is a hidden grammatical element, explicitly answer that part.
Negative regression
Reject or downgrade answers like:
أفضل خبر فقط، ولا يحتاج إلى فاعل مستتر لأنه اسم وليس فعلًا.
This is exactly the kind of error the PDF tested.
Gap 2 — نحن العرب نكرم الضيف: الاختصاص is not بدل
Observed response
The response parsed:
نحن: ضمير منفصل في محل رفع مبتدأ.
العربَ: بدل من الضمير نحن مرفوع وعلامة رفعه الضمة الظاهرة.
This has two problems:
It gives the wrong construction: بدل instead of اختصاص.
It internally contradicts the written form العربَ; it says it is مرفوع بالضمة even though the probe displayed fatḥa.
Correct answer required by the regression
نحن: ضمير منفصل مبني على الضم في محل رفع مبتدأ.
العربَ: مفعول به منصوب بفعل محذوف تقديره أخصّ / أعني، وعلامة نصبه الفتحة، وهو منصوب على الاختصاص.
نكرم: فعل مضارع مرفوع، والفاعل ضمير مستتر وجوبًا تقديره نحن.
الضيفَ: مفعول به منصوب.
وجملة نكرم الضيف في محل رفع خبر للمبتدأ نحن.
Why this matters
الاختصاص is a hidden-operator construction. The visible phrase can look like apposition, but the classical analysis supplies a deleted verb such as:
أخصُّ العربَ
أعني العربَ
So العرب is منصوب, not تابعًا مرفوعًا لـ نحن.
Skill rule to add
Add a detection step for الاختصاص:
If a first-person or second-person pronoun / speaker-group expression is followed by a definite noun that specifies the intended group, and that noun appears in accusative position, check for الاختصاص before treating it as بدل or عطف بيان.
Typical pattern:
نحن العربَ نكرم الضيف.
نحن المسلمينَ نعبد الله.
أنا الطالبَ أحتاج إلى المراجعة.
The second phrase is often:
منصوب على الاختصاص بفعل محذوف تقديره أخصّ أو أعني.
Negative regression
Reject or downgrade answers like:
العرب: بدل من نحن مرفوع مثله.
for this test context.
Also flag contradiction when a model says:
العربَ ... مرفوع بالضمة
The visible fatḥa and the claimed رفع cannot both be true.
Gap 3 — لا رجلَ في البيت: preserve the correct مبني في محل نصب distinction
Observed response
The current response answered this correctly:
رجلَ: اسم لا النافية للجنس مبني على الفتح في محل نصب.
The original PDF tested exactly this distinction and marked the flat answer wrong. The wrong answer was essentially:
رجل: اسم لا النافية للجنس منصوب، وعلامة نصبه الفتحة.
That is not precise enough for مفرد اسم لا النافية للجنس. The correct form is:
اسم لا النافية للجنس مبني على الفتح في محل نصب.
This is the same family of failure as the other two cases: a surface-looking fatḥa is not always an ordinary نصب sign.
Required rule
Add / preserve this rule:
If لا is لا النافية للجنس and its اسم is مفرد غير مضاف ولا شبيه بالمضاف, the اسم is مبني على ما ينصب به في محل نصب, not normally معرب منصوب.
Examples:
لا رجلَ في البيت.
رجلَ: اسم لا النافية للجنس مبني على الفتح في محل نصب.
لا طالبَيْنِ في الفصل.
طالبَيْنِ: اسم لا النافية للجنس مبني على الياء في محل نصب، لأنه مثنى.
لا مجتهدينَ خاسرون.
مجتهدينَ: اسم لا النافية للجنس مبني على الياء في محل نصب، لأنه جمع مذكر سالم.
For simple initial implementation, at minimum include the singular case:
لا رجلَ في البيت.
Proposed skill additions
1. Add a reference section: Hidden iʿrāb structures / البنية الإعرابية الخفية
Suggested subsections:
اسم التفضيل والخبر المشتق
الاختصاص
لا النافية للجنس: البناء في محل نصب
2. Add a procedure step before finalizing iʿrāb labels
Before returning a flat parse such as خبر, بدل, or منصوب, ask:
Is this a derived predicate that carries a hidden pronoun/فاعل?
Is this a known hidden-operator construction such as الاختصاص?
Is the word built in a syntactic slot, i.e. مبني في محل rather than visibly معرب?
3. Add a contradiction check
If the answer says a word is written with fatḥa but parsed as مرفوع بالضمة, flag it.
Example contradiction from the failed probe:
العربَ: بدل من نحن مرفوع وعلامة رفعه الضمة الظاهرة.
A robust tutor should catch this immediately.
Files likely needing updates
Depending on current repo organization, likely candidates are:
references/irab-case-mood.md
procedures/irab-case-mood.md
fusha/nahw/SKILL.md if present or planned
If the repo has examples/tests, add regressions in the appropriate location, e.g.:
examples/nahw-regressions.md
tests/nahw/hidden-irab-structures.*
tests/nahw/irab-case-mood.*
Minimal regression examples
### زيدٌ أفضلُ
Expected:
زيدٌ: مبتدأ مرفوع.
أفضلُ: خبر مرفوع، وفيه / فيها فاعل مستتر يعود على زيد.
Reject:
أفضلُ خبر فقط ولا يحتاج إلى فاعل مستتر لأنه اسم.
### نحن العرب نكرم الضيف
Expected:
نحن: ضمير منفصل في محل رفع مبتدأ.
العربَ: مفعول به منصوب بفعل محذوف تقديره أخصّ / أعني، وهو منصوب على الاختصاص.
نكرم الضيف: جملة فعلية في محل رفع خبر للمبتدأ.
Reject:
العرب بدل من نحن مرفوع مثله.
### لا رجلَ في البيت
Expected:
لا: حرف نفي للجنس يعمل عمل إن.
رجلَ: اسم لا النافية للجنس مبني على الفتح في محل نصب.
في البيت: جار ومجرور متعلقان بخبر محذوف.
Reject:
رجل منصوب بالفتحة فقط.
Acceptance criteria
The skill parses زيدٌ أفضلُ with the hidden فاعل/ضمير in أفضل returning to زيد.
The skill does not answer زيد مبتدأ وأفضل خبر as a complete final parse when the prompt asks for full iʿrāb or hidden elements.
The skill rejects the claim that أفضل cannot carry a hidden فاعل/ضمير because it is an اسم.
The skill parses نحن العرب نكرم الضيف as اختصاص.
The skill gives العربَ as منصوب على الاختصاص بفعل محذوف تقديره أخصّ or أعني.
The skill rejects parsing العرب as بدل من نحن in this regression context.
The skill flags internal contradictions such as writing العربَ but saying مرفوع بالضمة.
The skill preserves the correct لا النافية للجنس analysis: رجلَ is مبني على الفتح في محل نصب.
The skill does not reduce رجلَ to ordinary منصوب بالفتحة.
Regression examples for all three cases are added.
Together, these prevent the skill from producing plausible but shallow parses: the common failure is not ignorance of the topic name, but flattening the construction before checking the deeper nahw mechanism.
Problem
A third grammar-test probe exposed a different class of nahw failure: the model can often assign a familiar surface label, but it misses hidden grammatical structure or collapses محلّ الإعراب into a visible case ending.
This is distinct from the previous issues:
كتبْتُ.The three relevant constructions are:
زيدٌ أفضلُ.نحن العرب نكرم الضيف.لا رجلَ في البيت.Source / citation check
Evidence is from the uploaded
GrammarProblems.pdf, titled:Verified PDF locations:
Question 57, printed p. 148, topic: إعمال أفعل التفضيل.
في قولنا: زيدٌ أفضلُ، أي الأعاريب الآتية هو الإعراب الصحيح لهذه الجملة؟زيدٌ مبتدأ، وأفضلُ خبر، وفيها فاعل مستتر يعود على زيد.زيدٌ مبتدأ، وأفضلُ خبرأفضلneeds a hidden فاعل.Question 59, printed pp. 149–150, topic: الاختصاص.
أعرب: نحن العرب نكرم الضيف.نحن: ضمير منفصل مبني على الضم في محل رفع مبتدأ.العرب: مفعول به منصوب بفعل محذوف تقديره أخصّ، وعلامة نصبه الفتحة الظاهرة، وهو المشهور بالمنصوب على الاختصاص.نكرم الضيف: الجملة الفعلية في محل رفع خبر للمبتدأ.العربas بدل مننحنمرفوع مثله.Question 60, printed p. 150, topic: لا النافية للجنس.
أعرب: لا رجل في البيت.لا: حرف نفي للجنس، ناسخ يعمل عمل إن، مبني على السكون، لا محل له من الإعراب.رجل: اسم لا النافية للجنس، مبني على الفتح في محل نصب.في البيت: جار ومجرور متعلقان بخبر محذوف.رجل: اسم لا منصوب، وعلامة نصبه الفتحة.Regression probe that triggered this issue
The following questions were asked as a final set:
The response was:
زيدٌ أفضلُ, because it gave only a flat مبتدأ + خبر analysis and missed the hidden فاعل/ضمير inأفضل.نحن العرب نكرم الضيف, because it parsedالعربas بدل fromنحنand even contradicted the supplied fatḥa by calling it مرفوع بالضمة.لا رجل في البيت, because it correctly saidرجلis اسم لا النافية للجنس مبني على الفتح في محل نصب.So this is a 2/3 failure, with the third item retained as a non-regression success.
Gap 1 —
زيدٌ أفضلُ: اسم التفضيل is not only a flat خبرObserved response
The response parsed:
Then it described
أفضلas simply:This is not enough for the PDF’s expected answer.
Correct answer required by the regression
Alternative acceptable wording:
or:
Why this matters
This is a hidden-structure trap. The surface answer
زيد مبتدأ وأفضل خبرis not false as far as it goes, but it is incomplete for advanced iʿrāb. The PDF’s test explicitly rejects the answer that stops there, because the intended analysis requires recognizing the hidden element inside/withأفضل.For the skill, this means:
خبر مرفوعonly.Negative regression
Reject or downgrade answers like:
This is exactly the kind of error the PDF tested.
Gap 2 —
نحن العرب نكرم الضيف: الاختصاص is not بدلObserved response
The response parsed:
This has two problems:
العربَ; it says it is مرفوع بالضمة even though the probe displayed fatḥa.Correct answer required by the regression
Why this matters
الاختصاص is a hidden-operator construction. The visible phrase can look like apposition, but the classical analysis supplies a deleted verb such as:
So
العربis منصوب, not تابعًا مرفوعًا لـنحن.Skill rule to add
Add a detection step for الاختصاص:
Typical pattern:
The second phrase is often:
Negative regression
Reject or downgrade answers like:
for this test context.
Also flag contradiction when a model says:
The visible fatḥa and the claimed رفع cannot both be true.
Gap 3 —
لا رجلَ في البيت: preserve the correct مبني في محل نصب distinctionObserved response
The current response answered this correctly:
This should be kept as a positive regression.
Why this still belongs in Issue #3
The original PDF tested exactly this distinction and marked the flat answer wrong. The wrong answer was essentially:
That is not precise enough for مفرد اسم لا النافية للجنس. The correct form is:
This is the same family of failure as the other two cases: a surface-looking fatḥa is not always an ordinary نصب sign.
Required rule
Add / preserve this rule:
Examples:
For simple initial implementation, at minimum include the singular case:
Proposed skill additions
1. Add a reference section:
Hidden iʿrāb structures / البنية الإعرابية الخفيةSuggested subsections:
اسم التفضيل والخبر المشتقالاختصاصلا النافية للجنس: البناء في محل نصب2. Add a procedure step before finalizing iʿrāb labels
Before returning a flat parse such as
خبر,بدل, orمنصوب, ask:3. Add a contradiction check
If the answer says a word is written with fatḥa but parsed as مرفوع بالضمة, flag it.
Example contradiction from the failed probe:
A robust tutor should catch this immediately.
Files likely needing updates
Depending on current repo organization, likely candidates are:
references/irab-case-mood.mdprocedures/irab-case-mood.mdfusha/nahw/SKILL.mdif present or plannedIf the repo has examples/tests, add regressions in the appropriate location, e.g.:
examples/nahw-regressions.mdtests/nahw/hidden-irab-structures.*tests/nahw/irab-case-mood.*Minimal regression examples
Acceptance criteria
زيدٌ أفضلُwith the hidden فاعل/ضمير inأفضلreturning toزيد.زيد مبتدأ وأفضل خبرas a complete final parse when the prompt asks for full iʿrāb or hidden elements.أفضلcannot carry a hidden فاعل/ضمير because it is an اسم.نحن العرب نكرم الضيفas اختصاص.العربَas منصوب على الاختصاص بفعل محذوف تقديرهأخصّorأعني.العربas بدل مننحنin this regression context.العربَbut sayingمرفوع بالضمة.لا النافية للجنسanalysis:رجلَis مبني على الفتح في محل نصب.رجلَto ordinaryمنصوب بالفتحة.Relationship to Issues #1 and #2
This issue should be treated as a third layer of nahw robustness:
Together, these prevent the skill from producing plausible but shallow parses: the common failure is not ignorance of the topic name, but flattening the construction before checking the deeper nahw mechanism.