CREOLELINK® · FREE ENGLISH REFERENCE

Understanding Haitian Creole

Orthography, grammar conventions and language technology: a practical guide for translation agencies, language reviewers, educators and AI developers.

Use the contents below to jump to a topic or search the guide for words such as apostrophe, NG and tokenization.

Read the Haitian Creole companion guide

Source policy: This guide summarizes the Akademi Kreyòl Ayisyen’s 2017 and 2023 orthography resolutions. Where the documents disagree, the 2023 resolution has precedence. Professional implementation tips are CreoleLink guidance, not official AKA provisions.

Why are there “floating letters”?

M ap vini.

English: I am coming. The letter m here is a shortened form of mwen, not a typo. In this construction, AKA uses spaces rather than an apostrophe.

Agency takeaway: Do not autocorrect this to M’ap vini.

Rules and implementation guidance

SECTION 1

Floating letters and pronouns

Short forms such as m, w, l, n and y may appear as standalone words. They are not inherently typographical errors.

M ap vini. → I am coming.

N ap dòmi. → We are sleeping.

Y ap pran l. → They are taking it.

Professional takeaway: Do not merge or expand a single-letter token without grammatical analysis.

Reference: AKA 2017, Provisions 2–3, pp. 7–9

Back to contents ↑

SECTION 2

Full and shortened forms

The 2017 resolution describes mwen/m, ou/w, li/l, nou/n, yo/y and ki/k. These alternate according to context rather than automatic character substitution.

Mwen vini / M vini.

Yo ap travay / Y ap travay.

Se pa w / Se pa ou (different meanings).

Professional takeaway: Check semantics before changing a shortened form.

Reference: AKA 2017, Provision 2, pp. 7–8

Back to contents ↑

SECTION 3

Spaces and apostrophes

The official examples separate short forms from adjacent words with spaces. English- or French-style apostrophes should not automatically be inserted.

M ap vini.

Al pran liv pa w.

Anita frape pye l.

Professional takeaway: A spellchecker must not rewrite M ap as M’ap.

Reference: AKA 2017, Provision 3, p. 9

Back to contents ↑

SECTION 4

When W is appropriate

The shortened w has restrictions involving neighboring vowels. The 2017 resolution also distinguishes expressions where w and ou convey different meanings.

Ou manje manje a.

Se pa w.

Se pa ou.

Professional takeaway: Do not replace all instances of w with ou.

Reference: AKA 2017, Provision 2, p. 8

Back to contents ↑

SECTION 5

24 letters and 32 graphemes

The 2023 resolution distinguishes letters from graphemes. A grapheme may be written using one, two or three letters.

CH = one grapheme.

NG = one grapheme.

OUN = one trigraph.

Professional takeaway: Do not assume one character always equals one sound.

Reference: AKA 2023, Provision 1, pp. 6–9

Back to contents ↑

SECTION 6

Names versus pronunciations

The Akademi names graphemes and separately lists their phonological values. Letter names are not equivalent to the speech sounds they represent.

Y: ye, /j/.

W: we, /w/.

U: ui, /ɥ/.

Professional takeaway: For speech systems, distinguish spoken letter names from phoneme labels.

Reference: AKA 2023, Provision 1, p. 9

Back to contents ↑

SECTION 7

Nasal vowels and OUN

The AKA lists AN, EN, ON and OUN as nasal vowel graphemes. The phonological interpretation of letter sequences within whole words must be handled in context.

AN / EN / ON / OUN.

oungan.

moun (requires contextual segmentation).

Professional takeaway: Avoid assigning one pronunciation to every matching character sequence.

Reference: AKA 2023, Provision 1, p. 8; contextual caution by CreoleLink

Back to contents ↑

SECTION 8

Accents are meaningful

The grave accent distinguishes E/È, O/Ò, and oral A followed by N from nasal AN in relevant contexts.

pe / pè.

lo / lò.

pan / pàn.

Professional takeaway: Do not strip diacritics from user-visible output.

Reference: AKA 2017, Provision 7, p. 11

Back to contents ↑

SECTION 9

One word or two?

The 2023 resolution describes when expressions are combined or separated, including grammatical tests for word boundaries.

Li gentan ale.

Li gen tan pou l dòmi.

an bwa / anba.

Professional takeaway: Tokenization and spelling checks must consider meaning.

Reference: AKA 2023, Provisions 2–6, pp. 10–13

Back to contents ↑

SECTION 10

Numbers and ordinals

The AKA specifies conventions for numbers 0–19 and compound numbers, including special cases such as venteyen.

8 = uit.

21 = venteyen.

70 = swasann dis.

Professional takeaway: Test number spellings separately from English conventions.

Reference: AKA 2023, Provisions 9–14, pp. 15–16

Back to contents ↑

SECTION 11

Age, time and hyphenation

The 2023 resolution documents expressions for ages, hours, compound proper names, dates and selected uses of hyphens.

2 zan.

li senk è / li de zè.

Jan-Pyè.

Professional takeaway: Preserve legitimate local time and name conventions.

Reference: AKA 2023, Provisions 15–17, pp. 17–19

Back to contents ↑

SECTION 12

Translation agency QA checklist

Check whether a single letter is a grammatical short form, whether an accent changes meaning, and whether a sequence is a multi-letter grapheme. Compare applicable AKA provisions before submitting corrections.

M ap vini: retain spacing.

pan / pàn: preserve distinction.

CH / NG: check grapheme context.

Professional takeaway: Prioritize accuracy and contextual meaning, not unfamiliar-looking typography.

Reference: CreoleLink workflow recommendations based on AKA 2017/2023

Back to contents ↑

SECTION 13

AI, OCR and spellchecking

The resolutions are orthographic standards, not training datasets. The following recommendations translate their language distinctions into software testing requirements.

Keep legitimate short tokens.

Retain accents after normalization.

Distinguish grapheme names and phonemes.

Professional takeaway: Evaluate with consented, privacy-safe examples and qualified Haitian Creole reviewers.

Reference: CreoleLink engineering guidance; not an AKA rule

Back to contents ↑

SECTION 14

Evaluation and error analysis

Test tokenization, machine translation, OCR, ASR and TTS with a representative dataset. Include accepted variation where relevant and document both missed errors and false corrections.

False correction: M ap → M’ap.

Accent deletion: pàn → pan.

Word boundary: gentan / gen tan.

Professional takeaway: Track false positives and correct them with expert feedback.

Reference: CreoleLink engineering guidance; not an AKA benchmark

Back to contents ↑

Check your understanding

Open any question to reveal its answer. These controls also work with a keyboard, even without JavaScript.

QUESTION 1: Which is the AKA’s short-form spelling example?

Answer: M ap vini.

The short form m is separated from ap.

QUESTION 2: Can w represent a shortened pronoun?

Answer: Yes, in appropriate contexts.

The AKA also imposes restrictions on its use.

QUESTION 3: Which grapheme uses three letters?

Answer: OUN.

It is the listed trigraph.

QUESTION 4: Can accents be stripped without changing text?

Answer: No.

For example, pan and pàn are distinct.

QUESTION 5: Which AKA resolution has precedence if the documents differ?

Answer: The 2023 resolution.

The second resolution explicitly states this.

QUESTION 6: Should an AI system expand every m to mwen?

Answer: No.

It requires contextual grammatical evaluation.