Rule POS¶
engine="rule" (the default, and the only engine wired into
BurmeseNLP.process()). Stages:
- Candidates — lexicon + closed-class heuristics (incl.
သည်asPOSTP/SFP/PART). - Context rules — filter per token (subject marker vs finite particle).
- Grammar / sequence rules — multi-token patterns.
- Preference —
TAG_PREFERENCEamong remaining tags.
BMWE indices override candidates to the span's resolved POS before
disambiguation. Confirmed gazetteer entity spans similarly override the
final tags to PROPN afterward — see Pipeline order.
engine="xlmr" is a second, unrelated engine (see
POS tagging): a
transformer model with its own myPOS-style tagset, standalone only.