23. The written standard and common errors

🔴 Advanced · Lesson 23 of 23

You can now read and write Sorani grammatically. This lesson covers what separates a correct text from a merely understandable one: the spelling standard, and the mistakes that native writers and machines make most. Sections 1 to 6 are implemented in zimankit (skills_ckb/orthography.py); section 7 and the fuller written standard are only partly implemented, so this lesson is also the specification of a future grammar checker.

1. Use Kurdish letters, not Arabic or Persian ones

Keyboards often produce look-alike characters from other languages. They look the same but break searching, sorting and spell-checking:

wrong (Arabic/Persian code point)	right (Kurdish)
ك	ک
ي ى	ی
ة	ە
ه used as a vowel	ە
ھ	ه

2. Don’t drop the marks

ڕ ڵ ێ ۆ are distinct letters. Writing ر ل ی و instead changes the word: گول is not گوڵ flower, بەلام is not بەڵام but.

3. Initial vowels need ئ

A word never starts with a bare vowel: ئاو, ئێمە, ئەو, never او or ێمە.

4. Old encodings

Older websites and documents typed ە as ه + ZWNJ and used Arabic digits and punctuation. The normaliser converts all of this:

typed text	after normalising	problems the validator still finds	note
به‌ڕێوه‌به‌ر	بەڕێوەبەر	·	legacy ه + ZWNJ for ە
كتێبەكە	کتێبەکە	·	Arabic kāf
ئەو رۆژە	ئەو ڕۆژە	·	initial ر should be ڕ
ساڵی 2015	ساڵی ٢٠١٥	·	Latin digits
گول	گول	·	missing ڵ: a possible word shape, so not flagged here (see section 6)

5. Words the validator rejects

validate_word() reports spellings that cannot be standard Sorani:

typed text	after normalising	problems the validator still finds	note
ثەورە	ثەورە	ثەورە: arabic-only letter; non-Kurdish character(s): U+062B	Arabic-only letter ث
اسپ	ئاسپ	·	initial vowel without ئ
ذکر	ذکر	ذکر: arabic-only letter; non-Kurdish character(s): U+0630	Arabic-only letter ذ

6. Real words in the wrong spelling

Some errors produce a string that could be a word. گول is a possible letter sequence, so a validator cannot reject it; only a lexicon can tell you that گوڵ flower is the word meant. zimankit generates the confusable variants of a word (spelling_variants: ر/ڕ ل/ڵ ی/ێ و/ۆ و/وو ه/ە, ZWNJ), and the wordlist pipeline maps a rare spelling to a variant that is at least five times more frequent in the corpus. 24 924 such pairs were learned from real text. The most frequent:

non-standard	standard	seen	standard seen
وێستگه	وێستگە station	6 748	55 860
سێشەممه	سێشەممە Tuesday	1 409	7 681
ڕای‌گەیاند	ڕای گەیاند or ڕایگەیاند he announced	295	79 327 (solid)
دیناره	دینارە it is a dinar	234	13 721
نێردراوه	نێردراوە has been sent	209	3 322

Most are the legacy ه-for-ە problem of section 4 on words the normaliser could not safely fix on its own. ڕای‌گەیاند shows a ZWNJ inserted after the preverb clitic (lesson 20). No authority uses that ZWNJ. The spelling standard prefers a space (ڕای گەیاند) and allows the solid form, which is far more common in the corpus. zimankit’s lexicon stores the solid form, and its analyzer does not read the split one correctly yet.

7. The spelling standard

The rules below come from the written-standard sources in chapter 23, which also says where those sources disagree. The last column says what validate_word() checks, and most spacing needs grammar, so it is not checked:

write	not	why	checked
وشە, ورچ	ووشە, وورچ	وو is never initial	yes
چوو, مێژوو, وەکوو	چو, مێژو, وەکو	final û after a consonant	yes
کەرکووک	کەرکوک	medial û	listed words
تیر, کوردی	تییر	one ی for î after a consonant
کۆتایی, سنەیی		یی after a vowel
نییە, چییە, کوردییە	نیە, کوردی‌یە	ی + یە, solid	listed words, ZWNJ
بەرانبەر	بەرامبەر		yes
ڕۆژاوا, ماڵاوایی	ڕۆژئاوا, ماڵئاوایی	ئ drops inside a compound	yes
من و تۆ	منو تۆ, من‌و تۆ	and is a separate word	ZWNJ
ڕۆژی جیهانیی ژنان	ڕۆژی جیهانی ژنان	ezafe after ی
ئەو نووسیی, پرسییان	ئەو نووسی, پرسیان	keep the stem’s ی before a clitic
ئەمڕۆ, ئەمساڵ; ئەم جارە		a time word is solid until it takes an ending
نانکردن; نان دەکەم; نانم کرد		compound verbs: infinitive solid, conjugated apart
هەڵدەگرم; هەڵم گرت	ڕای‌گەیاند	preverbs split around a clitic; the solid هەڵمگرت is allowed
لە ماڵدا; من لە ماڵدام	لە ماڵ دا, من لە ماڵ دام	postpositions attach, and so does a copula after them
لەگەڵ, لەسەر	لە گەڵ, لە سەر	fixed prepositions
بەخێرایی; بە ئۆتۆمبیل	بە خێرایی	manner adverb against preposition
جوانتر; ڕۆژێکی تر	جوان تر	comparative against another
پێم وایە; دەستم لێ دا	پێموایە	پێ, لێ, تێ stand apart, but join a clitic
بیست و یەک, دوو سەد; دووشەممە		compound numbers apart; number + noun as a new word solid

8. Grammar errors to watch

These need context, not just spelling. zimankit’s sentence parser warns about the first two in its notes, and the rest are not checked:

error	example	correct	lesson
agent clitic missing in the past	* من نان خوارد	من نانم خوارد	14
agent clitic on an intransitive verb	* هاتمان	هاتین	13
wrong allomorph after a vowel	* دەرگاەکە	دەرگاکە	7
plural noun after a number (common usage; no explicit rule in the grammars checked)	* سێ پیاوان	سێ پیاو	7
ezafe missing between noun and adjective	* پیاو باش	پیاوی باش	9
definite + adjective order	* پیاوەکە باش (as a phrase)	پیاوە باشەکە	9

Summary

Use Kurdish code points; keep the marks on ڕ ڵ ێ ۆ; start vowels with ئ.
zimankit normalises legacy text and flags impossible spellings.
Lexicon-based variant mapping catches real-word misspellings.
The spelling standard includes spacing: compound verbs apart (نان دەکەم, هەڵم گرت), postpositions attached (لە ماڵدا).
Agreement errors need context: the sentence parser warns about some of them.

Congratulations, you have finished the Sorani course. The reference pages have every verb and noun of the course in full.