POS tagging

Rule-based tagger (engine="rule"): lexicon candidates → context filters → grammar filters → preference fallback.

from burmesenlp import pos_tag, POS_TAGS, BurmeseNLP

pos_tag(["ကျွန်တော်", "ကျောင်း", "သို့", "သွား", "သည်"], engine="rule")
list(POS_TAGS)  # NOUN, VERB, POSTP, SFP, IDIOM, ...

Ambiguous particles

သည် / တယ် can be:

Context	Tag
Subject / topic marker (သူသည် ကျောင်း…)	POSTP
Verb-phrase final (…သွားသည်။)	SFP

BMWE spans force their resolved POS (e.g. IDIOM) before disambiguation.

Pipeline

nlp = BurmeseNLP()
doc = nlp.process(text)
doc.pos_tags

Alternate engine: engine="xlmr"

An XLM-RoBERTa model fine-tuned on the myPOS tagset is available as a second engine, for standalone tagging only:

from burmesenlp import pos_tag

pos_tag(words, engine="xlmr")  # myPOS labels: n, v, ppm, part, conj, ...

Requires the xlmr extra (pip install burmesenlp[xlmr]; pulls in transformers/torch, imported lazily). Its label set does not match the rule engine's tagset above (no PROPN, no SFP/PART/AUX split), so it is not wired into BurmeseNLP.process() — chunking, sentence segmentation, and clause parsing all assume engine="rule" tags.