ht_core_news_sm

Language: Haitian Creole (ht)
Type: spaCy pipeline (tokenizer, POS tagger, dependency parser)
Size: Small (optimized for efficiency)

Training

• Trained on ~3,300 manually corrected CoNLL-U sentences following Universal Dependencies guidelines.
• Data includes formal Haitian Creole texts (e.g., articles, religious texts, educational material).
• No pretrained word vectors were used (pure end-to-end pipeline).
• CoNLL-U Data: https://github.com/JephteyAdolphe/UD_Haitian_Creole-Adolphe

Capabilities

• Tokenization (including contractions and informal forms common in Haitian Creole)
• Part-of-Speech (POS) tagging based on Universal POS tags
• Dependency parsing (basic syntactic parsing)

Intended Use

• NLP research on low-resource languages
• Language technology development for Haitian Creole
• Educational or linguistic applications

Limitations

• No named entity recognition (NER) currently included.
• Trained primarily on formal Haitian Creole; performance may vary on very informal or highly dialectal texts.
• Small dataset: best for prototyping, research, and early-stage projects.

Example Usage

import spacy

texts = [
        "Si'm ka vini, m'ap pale ak li.", "M ap teste model lan (pou kounye a).",
         "Map manje gato a pandan map gade televizyon lem lakay mwen.",
         "M ap pale ak ou le w vini demen.", "M'ap vini, eske wap la avek lajan'm? Si ou, di'l non pou fre'w."
         ]

nlp = spacy.load("ht_core_news_sm")

for text in texts:
    doc = nlp(text)

    # Tokenization, POS tagging, Lemmatization, Dependency parsing
    print("Tokens, NORM, POS (tag), Dependency:")
    print(len(doc))
    for token in doc:
        print(f"{token.text} | {token.norm_} | {token.tag_} | {token.dep_}")
    print("\n")