Appendix C — Turkic Language Reference
This appendix separates language identifiers, writing systems and computational evidence: identifiers and scope (Section C.1), the evidence needed before calling a resource available (Section C.2), a script reference (Section C.3) and the typological properties the chapters use (Section C.4).
The language table lists 23 contemporary languages. It differs from the 24 entries of TurkicNLP 0.5.0 (Chapter 1): the table adds Shor (cjs) and Chulym (clc), which the toolkit does not cover, and leaves out the toolkit’s historical Ottoman Turkish (ota) and Old Turkic (otk) and its separate South Azerbaijani entry (azb). For toolkit coverage, use its versioned registry; for corpus or model labels, use a resource-specific crosswalk.
C.1 Language Identifiers and Scope
| Language | ISO 639-3 | Descriptive grouping | Script examples |
|---|---|---|---|
| Turkish | tur | Oghuz | Latin |
| Azerbaijani | aze | Oghuz | Latin / Cyrillic / Arabic (variety-dependent) |
| Turkmen | tuk | Oghuz | Latin |
| Gagauz | gag | Oghuz | Latin |
| Kazakh | kaz | Kipchak | Cyrillic; versioned Latin schemes |
| Kyrgyz | kir | Kipchak | Cyrillic |
| Tatar | tat | Kipchak | Cyrillic |
| Bashkir | bak | Kipchak | Cyrillic |
| Kumyk | kum | Kipchak | Cyrillic |
| Nogai | nog | Kipchak | Cyrillic |
| Karachay-Balkar | krc | Kipchak | Cyrillic |
| Crimean Tatar | crh | Kipchak | Latin / Cyrillic |
| Karakalpak | kaa | Kipchak | Latin |
| Uzbek | uzb | Karluk | Latin / Cyrillic |
| Uyghur | uig | Karluk | Perso-Arabic |
| Sakha (Yakut) | sah | Siberian (Northern) | Cyrillic |
| Tuvan | tyv | Siberian (Southern) | Cyrillic |
| Khakas | kjh | Siberian (Southern) | Cyrillic |
| Southern Altai | alt | Kipchak (Eastern) | Cyrillic |
| Shor | cjs | Siberian (Southern) | Cyrillic |
| Chulym | clc | Siberian (Southern) | Cyrillic |
| Chuvash | chv | Oghur | Cyrillic |
| Khalaj | klj | Arghu | Arabic / Latin |
aze (Azerbaijani) and uzb (Uzbek) are macrolanguage identifiers; resources may instead use North/South Azerbaijani (azj/azb) or Northern/Southern Uzbek (uzn/uzs). alt denotes Southern Altai, not every variety called Altai. Do not join corpus labels, Wikipedia editions, Hub tags and toolkit codes by string substitution. Record a many-to-one or one-to-many mapping explicitly.
Speaker estimates need a source table or page, reference year, geography and a definition such as first language, self-reported ability or total speakers. Estimates with different definitions cannot support a single ranking, so this table omits them. Glottocodes also await individual verification. Consult the ISO 639-3 registry and Glottolog for identifiers. The codes here were checked against an installed ISO dataset, rather than a fresh registry snapshot.
C.2 NLP Resource Evidence
| Question | Versioned evidence | What it does not establish |
|---|---|---|
| Is a treebank officially released? | UD 2.17 inventory in Chapter 8 | A local silver file is not a released gold treebank |
| Is an analyser registered? | Toolkit/Apertium module and revision in Chapter 6 | Nonempty features are not accuracy |
| Are model weights released? | Pinned entries in Table B.6 | Availability is not target-domain quality |
| Is evaluation available? | Dataset release, labels and splits in Chapter 9 through Chapter 15 | Software support is not benchmark coverage |
MMS, for example, releases Chuvash and Crimean Tatar TTS checkpoints (Chapter 11); their suitability for production needs separate testing. The five-language audit in Section 17.6 records availability, unknowns and failed queries separately. Verify model architecture before assigning a label such as “BERT-class”, and retain each resource’s licence and task type.
C.3 Script Reference
C.3.1 Latin Characters and Letter Marks
Turkish dotted İ/i and dotless I/ı need locale-aware casing. Azerbaijani Latin ə/Ə (U+0259/U+018F) must not be confused with Cyrillic ә/Ә (U+04D9/U+04D8). Similar appearance does not imply identical code points or script.
Uzbek oʻ/gʻ and the tutuq sign appear with several digital apostrophe-like characters. Choose a declared encoding convention for a derived copy, retaining the original; using U+0027 does not by itself prove that every tokeniser will segment incorrectly. The context-sensitive search-key helper in Section 4.3 only normalises o/g letter marks and leaves other punctuation alone.
Kazakh Latin schemes are versioned proposals and orthographic conventions, not interchangeable transliterations. Use the dated sources in Chapter 3 and record the chosen converter instead of combining apostrophe, acute and later diacritic inventories into one “current alphabet”.
C.3.2 Cyrillic Characters
| Lowercase | Uppercase | Code points (lower / upper) | Unicode name |
|---|---|---|---|
| ң | Ң | U+04A3 / U+04A2 | CYRILLIC SMALL LETTER EN WITH DESCENDER |
| ү | Ү | U+04AF / U+04AE | CYRILLIC SMALL LETTER STRAIGHT U |
| ұ | Ұ | U+04B1 / U+04B0 | CYRILLIC SMALL LETTER STRAIGHT U WITH STROKE |
| ә | Ә | U+04D9 / U+04D8 | CYRILLIC SMALL LETTER SCHWA |
| ғ | Ғ | U+0493 / U+0492 | CYRILLIC SMALL LETTER GHE WITH STROKE |
| қ | Қ | U+049B / U+049A | CYRILLIC SMALL LETTER KA WITH DESCENDER |
| ө | Ө | U+04E9 / U+04E8 | CYRILLIC SMALL LETTER BARRED O |
| һ | Һ | U+04BB / U+04BA | CYRILLIC SMALL LETTER SHHA |
| ҙ | Ҙ | U+0499 / U+0498 | CYRILLIC SMALL LETTER ZE WITH DESCENDER |
| ҫ | Ҫ | U+04AB / U+04AA | CYRILLIC SMALL LETTER ES WITH DESCENDER |
| ҥ | Ҥ | U+04A5 / U+04A4 | CYRILLIC SMALL LIGATURE EN GHE |
The table gives no single IPA value per character: pronunciation and phonemic analysis depend on the language. Kazakh қ/ғ are not letters of the Kyrgyz alphabet. Chapter 2’s source-specific vowel table and Chapter 5’s harmony descriptions provide the linguistic context.
C.3.3 Perso-Arabic Uyghur
Distinguish vowel letters from sequences that include an initial hamza carrier; strings such as ئو contain more than one Unicode code point. Preserve right-to-left text logically, including joining behaviour and meaningful controls. Inspect NFC, code points and rendered output separately. Removing every invisible control as “noise” can damage text. Conversion to ULY or a Cyrillic convention needs a declared scheme and loss report (Chapter 3). The recurring Uyghur sentence (Table 1.4) shows one text in three views: the original, the treebank’s ULY transliteration and the toolkit’s CTS rendering (Section 3.6).
C.4 Typology and Computational Interpretation
| Property | Interpretation for this book |
|---|---|
| Vowel inventories and harmony | Use source-specific analyses in Chapter 2 and Chapter 5; avoid a binary strict/partial scale |
| Constituent order | SOV is the usual base order; contact varieties and discourse patterns require separate treatment (Section 8.1) |
| Null subjects | Grammar and context may license an absent subject; agreement is not gender marking |
| Case and morphology | Specify the grammatical/annotation scheme instead of ranking languages by raw case counts |
| Morphological density | Measure a defined corpus statistic; “high/moderate” is not an observed morpheme ratio |
These properties suggest what to test; they guarantee no particular NLP outcome. SOV order does not by itself imply longer dependencies, and morphological richness alone does not explain every model error. See Chapter 5 and Chapter 8 for constructions, annotation choices and controlled comparisons (Johanson 2021).