Gazetteers (NER)

Rule-based entity lists under corpus/gazetteers/. In V1 they run inside process() after POS, on post-BMWE words, and fill doc.entities.

from burmesenlp import BurmeseNLP, EntityType

nlp = BurmeseNLP()  # gazetteer=True by default (lazy-loaded)
doc = nlp.process("ဒေါ်အောင်ဆန်းစုကြည် ရန်ကုန်ကို သွားသည်။")

for e in doc.entities:
    print(e.text, e.entity_type, e.attributes)

# Skip NER when you only need syntax
nlp = BurmeseNLP(gazetteer=False)

Pipeline placement

… → BMWE → POS → Gazetteer → Phrase chunking (entity spans → NP) → …
                      │                    │
                      ▼                    ▼
                doc.entities         doc.chunks (NP with features.entity)

Entities remain a separate semantic list. Matching spans are also locked as one NP in chunks so names are not split by POS patterns (e.g. မင်း PRON + လှိုင်အစိုးရရဲ့ PP).

Attributes

Name lists store gender metadata:

{"gender": "male", "source": "male_names"}
{"gender": "female", "source": "female_names"}

Standalone use

from burmesenlp import GazetteerManager

gaz = GazetteerManager()
gaz.contains("ရန်ကုန်")
gaz.lookup("ရန်ကုန်")

Most files are JSON string arrays; type comes from the filename (towns.json → TOWN, male_names.json → PERSON, …).