michsethowusu/akan-pos-tagger

Akan POS Tagger (SpaCy model)

This is a custom-trained Part-of-Speech (POS) tagging model for the Akan language using spaCy.

✨ Installation and Usage

Prerequisites

pip install spacy huggingface_hub

Loading and Using the Model

from huggingface_hub import snapshot_download
import spacy

# Download the model from Hugging Face Hub
model_path = snapshot_download(repo_id="michsethowusu/akan-pos-tagger")

# Load the model
nlp = spacy.load(model_path)

# Use the model for POS tagging
doc = nlp("bosom som nyɛ")
for token in doc:
    print(f"{token.text} -> {token.tag_}")

Expected Output

bosom -> N
som -> V
nyɛ -> ADV

🏷 Complete POS Tags Reference

This model uses a comprehensive custom POS tagset with over 100 tags for Akan linguistic structures:

Adjectives

• ADJ – adjective
• ADJatt – attributive adjective
• ADJC – comparative adjective
• ADJpret – predicative adjective
• ADJS – superlative adjective

Adverbs

• ADV – adverb
• ADVdir – directional adverb
• ADVm – manner adverb
• ADVneg – negative operator
• ADVplc – place adverb
• ADVtemp – temporal adverb

Articles and Determiners

• ART – article
• DET – determiner
• DEM – demonstrative

Auxiliary and Copula

• AUX – auxiliary
• COP – copula
• COPident – identity copula
• COPloc – locative copula
• COPneg – negative copula

Numerals

• CARD – cardinal numeral (e.g. 4, sixty-five)
• NUM – numeral
• NUMpart – partitive numeral
• ORD – ordinal

Conjunctions

• CONJ – conjunction
• CONJC – coordinating conjunction (e.g. and, or)
• CONJS – subordinating conjunction (e.g. although, when)
• CONJSinf – subordinating conjunction (introducing an infinitive clause)

Nouns

• CN – common noun
• N – common noun
• Nbare – bare noun
• Ncomm – noun with common gender
• NFEM – feminine noun
• NMASC – masculine noun
• NNEUT – neuter noun
• NNO – noun neutral for number (e.g. data, aircraft)
• Np – proper noun
• Npinst – name of an institution
• Nploc – name of a location
• Npname – personal name
• Nrel – relational noun
• Nspat – spatial noun

Pronouns

• PN – personal pronoun
• PNabs – absolute pronoun (Bantu)
• PNana – pronominal anaphor
• PNdem – demonstrative pronoun
• PNposs – possessive pronoun
• PNrefl – reflexive pronoun
• PNrel – relative pronoun
• PROint – interrogative pronoun
• PROPana – propositional anaphor

Prepositions and Postpositions

• PREP – preposition
• PREPdir – directional preposition
• PREPplc – locative preposition
• PREPsel – selected preposition
• PREPtemp – temporal preposition
• PPOST – postposition

Particles

• PRT – particle
• PRTexist – existential marker
• PRTinf – infinitive marker
• PRTint – interrogative particle
• PRTn – nominal particle
• PRTneg – negative particle
• PRTposs – possessive particle
• PRTpred – predicative particle
• PRTprst – presentational particle
• PRTresp – response words such as "thanks, please, no, yes..."
• PRTv – verbal particle

Verbs

• V – verb
• V1 – first verb in a verbal chain
• V2 – second verb in a verbal chain
• V3 – third verb in a verbal chain
• V4 – fourth verb in a chain
• Vbid – verb bid (Kwa)
• Vcon – converb
• Vdtr – ditransitive verb
• Vimprs – impersonal verb
• Vitr – intransitive verb
• VitrOBL – intransitive verb with a prepositional object
• Vlght – light verb
• Vmod – modal verb
• Vneg – negative verb
• Vpre – preverb
• Vrefl – reflexive verb
• Vtr – transitive verb
• VtrOBL – transitive verb with a prepositional object
• Vvector – vector verb

Other Categories

• CIRCP – circumposition
• CL – clitic
• CLF – classifier
• CLFnom – nominal classifier
• CLFnum – numeral classifier
• COMP – complementiser
• EXPL – expletive pronoun
• INTRJCT – interjection
• IPHON – ideophone, onomatopoeia
• MOD – modifier
• PTCP – participle
• QUANT – quantifier
• REL – relative clause marker
• TRUNC – truncation
• Wh – wh-word

Punctuation

• PUL – punctuation: left bracket (e.g. ( or [)
• PUN – punctuation: general separating mark (e.g. , ; . ! : ?)
• PUQ – punctuation: quotation marks (' or ")
• PUR – punctuation: right bracket (e.g. ) or ])

Special Categories

• NE – named entity
• XY – non-words such as "JU52"

🔧 Technical Details

• Framework: SpaCy
• Language: Akan (ak)
• Training Epoch: 1 (best-performing model)
• Tag Set: Custom 100+ tag dictionary based on Akan linguistic structures
• Model Type: Token classification for POS tagging

📊 Performance

The model achieves competitive accuracy on Akan POS tagging tasks, with specialized handling for the complex morphological and syntactic features of the Akan language.

🤝 Citation

If you use this model in your research, please cite:

@misc{{akan-pos-tagger,
  title={{Akan POS Tagger}},
  author={{{michsethowusu}}},
  year={{2025}},
  url={{https://huggingface.co/michsethowusu/akan-pos-tagger}}
}}

"""