Mova0.6 pre-release

the flag of Kazakhstan 3. Kazakh scripts and transliteration

Standard written Kazakh in Kazakhstan uses Cyrillic. Kazakh is also written in an Arabic-derived alphabet, particularly in China. Historical texts and modern Latin proposals add further conventions. Identify the convention before reading a spelling across scripts. A written letter, a letter’s pronunciation, and its romanization are three different things.

The Cyrillic alphabet and its Kazakh letters

The standard Cyrillic alphabet has 42 letters. It includes all 33 letters of the Russian Cyrillic alphabet and nine additional letters used in Kazakh. The table gives the full Kazakh alphabet in collation order. Each entry gives the uppercase letter followed by its lowercase form.

PositionsLetters in order
1–7А а, Ә ә, Б б, В в, Г г, Ғ ғ, Д д
8–14Е е, Ё ё, Ж ж, З з, И и, Й й, К к
15–21Қ қ, Л л, М м, Н н, Ң ң, О о, Ө ө
22–28П п, Р р, С с, Т т, У у, Ұ ұ, Ү ү
29–35Ф ф, Х х, Һ һ, Ц ц, Ч ч, Ш ш, Щ щ
36–42Ъ ъ, Ы ы, І і, Ь ь, Э э, Ю ю, Я я

The Kazakh KTB treebank and the Kazakh Cyrillic subset of FLORES-200 both provide examples of modern Cyrillic usage. Their scope is annotated text and parallel evaluation text respectively. The Cyrl in kaz_Cyrl identifies the script of that dataset subset. (Universal Dependencies contributors 2025, § UD Kazakh KTB. NLLB Team 2022, § kaz_Cyrl.)

The nine additional letters are ordinary letters with their own places in the alphabet. Their strokes and shapes are integral to the letters. In particular, Ғ, Қ, and Ң remain distinct from Г, К, and Н in lowercase, uppercase, and handwriting. The readings below give broad values. Phonetic detail can vary with the speaker and the surrounding sounds.

LetterUsual valueReading cue
Ә әA front open vowel, approximately [æ]Distinct from the back vowel а
Ғ ғA voiced back fricative, often uvular [ʁ]The back counterpart of г in the native harmony pattern
Қ қA voiceless uvular stop [q]The back counterpart of к in the native harmony pattern
Ң ңA velar nasal [ŋ]A single letter and sound, distinct from н and the written sequence нг
Ө өA front rounded vowel, approximately [ø] or [œ]Distinct from back о
Ұ ұA back rounded vowel, approximately [ʊ]Distinct from front ү and from у
Ү үA front rounded vowel, approximately [ʏ] or [y]Distinct from back ұ
Һ һA glottal fricative [h] where the distinction is maintainedA separate letter from х
І іA front unrounded vowel, approximately [ɪ]Distinct from ы, и, and й

In native stems, қ and ғ normally go with back vowels, and к and г with front vowels. This relationship helps identify the harmony of a stem. In loans, compounds, and other forms with mixed vowel sequences, read the spelling that belongs to the lexical form. In particular, the letter shapes give no license to substitute қ for к or ғ for г.

FormRomanizationGlossTranslationStatus
санsannumber‘number’(constructed)
сәнsänstyle‘style’(constructed)
торtornet‘net’(constructed)
төрtörplace.of.honor‘place of honor’(constructed)
ұнūnflour‘flour’(constructed)
үнünvoice‘voice’(constructed)
онonten‘ten’(constructed)
оңoñright.side‘right side’(constructed)
қырqyrridge‘ridge’(constructed)
кірkirenter.IMP.2SG‘Come in.’(constructed)

Several letters require context for pronunciation. Й represents a palatal glide. In inherited words и commonly covers a vowel plus that glide. The front or back value of the vowel comes from the word. In inherited words у can represent a labial glide, as after the vowel in тау, or cover a vowel plus the glide, as in су. Loans can have vowel readings of и and у. The spellings і, и, and й therefore remain distinct even when a phonetic transcription uses an i- or j-like symbol in more than one case. Likewise, ұ, ү, and у remain distinct in transliteration.

Е can have an initial glide, especially at the beginning of a word. Я and ю can write glide-plus-vowel sequences, including in familiar Kazakh words. In loans the readings of е, ё, ю, and я can also follow conventions associated with the source spelling and the preceding consonant. Each counts as one alphabet letter. Conversely, a pair of Cyrillic letters remains two written letters even where the sounds are pronounced together.

FormRomanizationGlossTranslationStatus
биbìdance‘dance’(constructed)
тауtaumountain‘mountain’(constructed)
суsuwater‘water’(constructed)
елelcountry‘country’(constructed)
аяқajaqfoot‘foot’(constructed)
аюajubear‘bear’(constructed)

Cyrillic letters in loans and names

Kazakh Cyrillic has letters that occur chiefly in loans and proper names. Their presence can reflect the spelling through which a word entered Kazakh, often Russian. An established borrowing may retain the source spelling closely, or it may have a conventional Kazakh spelling of its own. Proper names can likewise have established local spellings. Use the actual written form when identifying letters or transliterating a name. Predicting how a name might otherwise be adapted is a separate task.

LetterTypical role in loans and namesOther uses and qualifications
В вRepresents в in retained source spellingsIts spoken realization can vary
Ё ёRepresents ё, including in namesPreserve the dots when they are present. An attested е is a different written letter
Ф фRepresents фAlso occurs in established loans, so it gives little information about how recently a word entered Kazakh
Ц цRepresents ц, typically with a [ts]-like readingOne Cyrillic letter even when romanized with two Latin letters
Ч чRepresents ч, typically with an affricate readingOccurs in proper names as well as other loans
Щ щRetains щ in loan spellings and namesAlso occurs in Kazakh words such as ащы and тұщы, where the usual reading is a long or doubled sh-like sound
Ъ ъRetains the hard sign in borrowed spelling, typically as a separator before a vowel letterHas no consonant or vowel sound of its own
Ь ьRetains the soft sign in borrowed spelling and can mark palatalization or separation according to that spellingHas no consonant or vowel sound of its own
Э эRetains э, typically for an e-like vowelRemains distinct in writing from е
Х хOccurs in loans and in many long-established Kazakh formsIts realization varies, including velar and farther-back fricative readings
Һ һOccurs in a limited set of established spellingsSpeakers may differ in how clearly they distinguish it from х
И и, У уCan have vowel readings in borrowed formsAlso have the native uses described above
Ю ю, Я яCan preserve corresponding letters of the donor spellingAlso have native uses with a glide and vowel

Loan stems and names can take Kazakh endings. The source spelling of the stem and the selection of the ending are separate parts of the result. A silent sign remains part of a borrowed stem when the standard spelling retains it. Its presence does not supply a syllable or a vowel for the ending.

FormRomanizationGlossTranslationStatus
ВикторViktorViktor.PN‘Viktor’(constructed)
ФёдорFëdorFyodor.PN‘Fyodor’(constructed)
ЦойTsojTsoi.PN‘Tsoi’(constructed)
ЭммаÈmmaEmma.PN‘Emma’(constructed)
объектobʺektobject‘object’(constructed)
фильмдерfilʹmderfilm-PL‘films’(constructed)
ащыashchybitter‘bitter’(constructed)
тұщыtūshchyfresh‘fresh, unsalted’(constructed)
қаһарqaharwrath‘wrath’(constructed)

Arabic-script Kazakh and its reading conventions

Arabic-script Kazakh is read from right to left. The letters generally take different shapes depending on whether they join a letter on either side. A nonjoining edge makes a break in the stroke inside a word. Spaces, the sequence of letters, and the spelling convention determine word boundaries. Letter identity survives changes of joining shape.

There are two distinct historical contexts for reading this script. In the nineteenth-century material described by Melioransky, Kazakh appears in a literary tradition shared with other Central Asian Turkic writing. The spellings can reflect conventions of that tradition, and the language of a passage can include forms from literary registers alongside everyday Kazakh. Baitursynuly’s Til-qural belongs to the early twentieth-century use of a Kazakh-oriented Arabic orthography. Contemporary Arabic-script Kazakh, especially in China, belongs to the tradition of adapting the script to Kazakh. A reader should use the conventions of the particular text or edition. (Melioransky 1894, pp. 3–6, 11–14. Baitursynuly 1914, title page.)

Reading older writing

Initial alif can support the spelling of a word beginning with a vowel. Within or at the end of a word, it can itself help spell a vowel. Waw and ya have several possible readings, including vocalic ones, and a vowel may have no visible letter. The reader supplies such vowels from the possible word forms, vowel harmony, and the sentence. The same written base can therefore admit more than one reading in isolation. Traditional spellings of Arabic and Persian loans can preserve letters that say more about the donor spelling than about Kazakh pronunciation. Other spellings reflect local pronunciation more closely. Even within the material Melioransky describes, practice varies. (Melioransky 1894, pp. 11–14.)

The following signs are the ones described in Melioransky’s account. The illustrated code points identify ordinary modern Unicode characters for the signs where a combining character is given. The dot-circle is a display aid. The vowel readings are those reported for that older material and do not constitute a modern Kazakh vowel chart. (Melioransky 1894, pp. 11–13, § 3.)

SignPosition and nameReading in the older account
◌َ U+064EAbove, fathaA vowel with an a- or e-like reading
◌ِ U+0650Below, kasraUsually an i-like vowel
◌ُ U+064FAbove, dammaUsually a rounded u-like vowel
◌ْ U+0652Above, sukunAbsence of a following vowel on the marked consonant
◌ّ U+0651Above, shaddaDoubling of the marked consonant
◌ٓ U+0653Above, maddaFound especially on initial alif and helps distinguish its reading from an e-like value
HamzaWritten above a carrier or as required by the spelling traditionIn the account it also occurs in rare Persian-style expressions to write a linking vowel

The first five marks were generally sparse in the Kazakh texts described there, and the marks occur particularly in material of Arabic or Persian origin. The absence of a short-vowel mark leaves the reading open. The sukun is an explicit mark for absence of a vowel, so these two situations give different evidence. (Melioransky 1894, pp. 11–13, § 3.)

Melioransky prints the short phrase “бір бала” in his transcription of an Arabic-script tale. The quoted form is from his Cyrillic-based reading. It does not establish which vowels had visible letters in the Arabic-script original. (Melioransky 1894, p. 14.)

FormRomanizationGlossTranslationAttribution
бір балаbir balaone child‘one child’(Melioransky 1894, p. 14)

Reading Kazakh-oriented Arabic writing

Kazakh-oriented Arabic alphabets provide letters for Kazakh consonant distinctions, including q versus k and ñ versus n. They also make greater use of vowel spelling. In conventions that use a word-initial frontness sign, the sign affects the interpretation of the relevant vowels in the word. It is a harmony cue with scope over the word, and its use depends on what the other letters already tell the reader. For example, a convention can omit it when a front-only vowel or a consonant such as k or g already identifies the word as front. Words with disharmonic vowels, especially loans, require the spelling conventions of the edition. Older hamza uses, such as the linking-vowel use in the table, have a different function.

These sample letters illustrate the difference between consonant identity and the way vowel readings are recovered. They form part of the Kazakh-oriented inventory.

LetterReading or function in the examples and conventions discussed
قq, corresponding to Cyrillic қ
كk, corresponding to Cyrillic к
ڭñ, corresponding to Cyrillic ң
نn, corresponding to Cyrillic н
زz, corresponding to Cyrillic з
اAn a-series vowel letter. Its value depends on the orthography and the harmony of the word
ٴHigh hamza, used as a frontness cue in conventions that have that word-level sign
FormRomanized readingCyrillic readingGlossTranslationStatus
قازاقqazaqқазақKazakh‘Kazakh’(constructed)
قازاق بالاسىqazaq balasyқазақ баласыKazakh child-POSS.3‘a Kazakh child’(constructed)

A Roman form based on the Cyrillic reading can supply information inferred by the reader of the Arabic text. A diplomatic transliteration, which records the visible letters and signs of the Arabic text, serves a different purpose. For an older text, the literary form, the reader’s interpretation, and an equivalent expression in present-day standard Kazakh may all differ. (Melioransky 1894, pp. 4–6, 11–14.)

Historical Latin writing and modern Latin proposals

Kazakh used a Latin alphabet in the Soviet period from 1929 until the adoption of Cyrillic in 1940. Historical Latin print belongs to its own spelling standard. Its Roman letter shapes can be familiar while the letter values, alphabetic order, and treatment of loans differ from those of a modern proposal. A date and edition help determine which values to use. In republications of earlier works, the alphabet on the page may be the alphabet of the republication, so the original publication date and the script of the copy must be recorded separately.

More recent Latin proposals for Kazakhstan have used several ways to represent the Kazakh distinctions. The 2017 design used apostrophe spellings. The 2018 revision used acute accents and retained the digraphs sh and ch. A later draft used letters with signs such as the umlaut, breve, tilde, macron, and cedilla. These are distinct designs. A proposed letter value belongs to its dated proposal. Modern Cyrillic remains the reference spelling for the forms in this manual.

The following sample correspondences compare the designs for selected letters. The final column gives the manual’s romanization of the Cyrillic letter. The romanization is specified in full below.

Cyrillic2017 design2018 revisionLater draftManual
әa'áää
ғg'ǵğğ
ңn'ńññ
өo'óöö
ұuuūū
үu'úüü
шs'shşsh

An apostrophe spelling uses a base letter and a punctuation-shaped character. An accented spelling attaches a diacritic to the letter and may have more than one Unicode encoding. A digraph consists of two letters of text that the spelling system can assign to a single sound or alphabetic unit. The significance of a digraph depends on the particular alphabet. At a stem or suffix boundary, a visually identical sequence can sometimes arise from two independent letters. This matters when converting text back to Cyrillic.

Cyrillic formLatin spelling in the indicated designGlossTranslationStatus
әнa'n, 2017 designsong‘song’(constructed)
әнán, 2018 revisionsong‘song’(constructed)
әнän, later draftsong‘song’(constructed)
шаңs'an', 2017 designdust‘dust’(constructed)
шаңshań, 2018 revisiondust‘dust’(constructed)
шаңşañ, later draftdust‘dust’(constructed)

Older Latin spelling, any particular modern proposal, and romanization for an English-language reference work should each be read with their own letter key. A form shared by several conventions, such as qazaq, can be read in context while leaving the exact convention undetermined.

Unicode characters and mixed-script text

Unicode assigns each of the nine additional Cyrillic letters its own uppercase and lowercase character. These are the ordinary plain-text characters for the Kazakh alphabet.

LetterUppercase code pointLowercase code point
Ә әU+04D8U+04D9
Ғ ғU+0492U+0493
Қ қU+049AU+049B
Ң ңU+04A2U+04A3
Ө өU+04E8U+04E9
Ұ ұU+04B0U+04B1
Ү үU+04AEU+04AF
Һ һU+04BAU+04BB
І іU+0406U+0456

A glyph can resemble one from another script. І and і in Cyrillic use U+0406 and U+0456. Latin I and i use U+0049 and U+0069. Cyrillic А, С, О, and Р likewise resemble Latin A, C, O, and P, respectively. A visual match provides no evidence that the underlying characters are identical. Lowercase һ can resemble Latin h, and ө can resemble a Latin or Greek glyph in some fonts. Script identity comes from the characters in the string. Language identity also requires consideration of the words and context, since the same script can serve several languages.

In Cyrillic ё and й can each be represented by one precomposed character or by a base letter and a combining mark. Latin ä and ñ have the same kind of encoding alternatives. For example, a single Latin ä U+00E4 and U+0061 U+0308 represent canonically equivalent text. Normalization to NFC gives these equivalent sequences a common representation for ordinary comparison. It preserves the distinctions between Latin and Cyrillic letters and between letters with different diacritics. Straight apostrophe ' U+0027, right single quotation mark ’ U+2019, and modifier letter apostrophe ʼ U+02BC are separate characters. Their treatment as equivalent in a Latin spelling requires a convention specific to that spelling.

Arabic text is stored in logical reading order. Fonts produce joined shapes, and bidirectional display places the runs of text on screen. For example, the logical order of قازاق is U+0642 U+0627 U+0632 U+0627 U+0642. The two instances of ق are the same character even when their displayed shapes differ. Spaces, digits, punctuation, and an adjacent Latin or Cyrillic run can affect how the line is laid out. For quotations and comparisons, take the letter sequence from the logical string. Arabic presentation-form characters can appear in legacy text and should be distinguished from ordinary encoded letters. Optional diacritics and a word-level frontness sign can carry reading information, so their presence or absence matters when preserving a spelling.

A line of Kazakh can contain a Latin brand name, a technical label, or an Arabic quotation. A Cyrillic word typed with accidental Latin lookalikes is also possible. These cases give different evidence about the text. A whole span in another script may be intentional, while an unexpected script change inside a familiar word can be a keyboard or transcription error. The following complete comparison shows how visually similar names can have different encoded sequences.

FormLetter sequenceReading or romanizationGlossTranslationStatus
АлматыU+0410 U+043B U+043C U+0430 U+0442 U+044BAlmatyAlmaty.PN‘Almaty’(constructed)
AлматыU+0041 U+043B U+043C U+0430 U+0442 U+044BAlmaty, if the intended name is readAlmaty.PN‘Almaty’(constructed)

Romanization and transcription in this manual

Cyrillic is the citation form for modern standard Kazakh. The italic Roman form beside it follows a letter-based system. It preserves such spelling distinctions as и versus і, е versus э, and х versus һ. It preserves the signs ъ and ь with distinct symbols. The italic form is useful for reading examples across scripts. Where a pronunciation point matters, a separate IPA form can be given in square brackets. In phonological discussion, slashes enclose phonemes.

The table gives the complete mapping. The entries are lowercase. For the uppercase letter, uppercase the Latin letter or all letters of its sequence. The two sign symbols have no case. Thus Ш gives SH under this convention, and ш gives sh. Preserve visible word boundaries and punctuation. A hyphen in the gloss may indicate a morpheme boundary that has no hyphen in the form or its romanization.

CyrillicRomanizationCyrillicRomanizationCyrillicRomanization
аaқqүü
әäлlфf
бbмmхkh
вvнnһh
гgңñцts
ғğоoчch
дdөöшsh
еeпpщshch
ёëрrъʺ
жzhсsыy
зzтtіi
иìуuьʹ
йjұūэè
кkюju
яja

The compact Roman form can hide Cyrillic letter boundaries. For example, ц and the sequence тс can each give ts. Likewise, я and йа each give ja. If the boundaries themselves are under discussion, the manual separates the Roman units with a middle dot for that word. A single Cyrillic letter then has no interior dot, so ц is ts, while тс is t·s. For case-sensitive quotation of a hard or soft sign, the Cyrillic form retains the exact case.

The sequences zh, kh, ts, ch, sh, shch, ju, and ja each appear as one entry in the mapping table. They are romanization conventions, so a two-letter Roman unit can correspond to a single Cyrillic letter. The Latin y in this system gives ы, and j gives й when it occurs as its own unit. The acute symbols of the 2018 proposal and the apostrophe spellings of the 2017 design carry their proposal-specific values. The grave accents in the manual’s ì and è distinguish letters and give no stress information.

FormRomanizationPronunciation cueGlossTranslationStatus
елelInitial glide in a reading such as [jel]country‘country’(constructed)
тауtauA final labial glide in a reading such as [tɑw]mountain‘mountain’(constructed)
қиqìA back vowel and glide in a reading such as [qɯj]dung‘dung’(constructed)
шаңshañш and ң each represent a single consonant, [ʃ] and [ŋ]dust‘dust’(constructed)
атсызatsyz, a·t·s·y·zт and с belong to different morphemes herehorse-PRIV‘horseless’(constructed)

For an Arabic-script example, the form column shows the Arabic letters. The romanized reading gives the reading in Kazakh, using the manual’s Roman values for the corresponding Cyrillic form when a Cyrillic reading is supplied. Where an older expression has more than one defensible reading, identify the chosen reading or retain the uncertainty. A historical Latin form is already Roman script and can stand as written. If its reading requires clarification, the example can give the corresponding modern Cyrillic form or a phonetic transcription in a separate column. Glosses identify words or morphemes, and translations give the meaning of the full example. In this chapter, IMP.2SG means second-person singular imperative, PL plural, PN proper name, POSS.3 third-person possessive, and PRIV privative, expressing absence.

How Mova handles script and language identification

The multilingual parser tools/mova-parse has two levels. Level 1 uses tools/langseg to assign a language label to each sentence and span. Its Latin-script identification distinguishes English and French using a Latin table, a lexicon for each language, and a character four-gram model. It also distinguishes Hebrew-script varieties and recognizes Hebrew or Aramaic in scholarly transliteration. Other scripts fall under other. Cyrillic and Arabic Kazakh therefore have no Kazakh-specific identification through this script coverage, and the Latin-script lexicons cover English and French. Adding support for another Latin-script language requires a lexicon and a Lang code. The langseg latin-lexicon command builds the lexicon from an open Universal Dependencies training split.

Level 2 parses a span with a model for the language assigned to it. Kazakh (kk) has no level-2 model. tools/mova-parse reports the missing model in the # parse comment and leaves LEMMA, UPOS, XPOS, FEATS, HEAD, and DEPREL as _, with Parse=None in MISC. Mova provides no Kazakh tokenization, tagging, lemmatization, or parsing at this stage. It holds no annotated Kazakh corpus and has measured no Kazakh treebank baseline.

Squid records each grammatical point in the manual as a claim and queues formalizable claims for possible implementation. The script and transliteration rules in this chapter are unimplemented. The first candidates are the distinctions needed for tokenization, tagging, and a Universal Dependencies treebank baseline.

Sources for this chapter

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/kk/03-scripts/