NERGAL

Named Entity Recognition with Grounded Additive Labels

SlayerLab hybrid PII cleaner for Polish. Not a chat model. Not a drop-in pipeline("token-classification").

TL;DR

Python rules do the identifiers they can prove. A transformer NER head adds phone and other PII spans the regex misses. The cleaner unions the two on the original text, then replaces hits with [Telefon] or [PII].

• Ground: frozen scrub_pii regex
• Additive labels: XLM-RoBERTa-large token classifier, BIO tags phone / pii, threshold 0.95
• This snapshot: seed 202609160, epoch 5 of a seven-epoch schedule

841-dev

Tables use one development split: 841 passages, 215 with gold PII, 354 spans (169 phone, 185 other PII). It is the dev side of a 4,500-passage labelled tranche (3,655 train / 841 dev). Sources match Dynaword (EUR-Lex, HPLT, Wikipedia, parliamentary and government text, plus smaller news/literary slices). Labels mix unchanged silver with human review.

The files contain real identifiers, so they are not released with the weights.

Why XLM-R

GLiNER, HerBERT-large, and XLM-R-large were trained on the same split and unioned with the same regex. Plot: diagnostic threshold 0.50; selection used the full threshold grid. GLiNER covers more at 0.50 and then dumps precision. XLM-R is the architecture we kept.

Why epoch 5

Fresh XLM-R, seven epochs. 133 epoch/threshold combinations. Epoch 5 at 0.95 was the only point that both beat the historical GLiNER∪regex incumbent on coverage and introduced zero new false-mask characters. Epoch 7 covers more PII (334/354) but adds 15 new false characters.

Epoch	Covered @ 0.95 /354	Residual passages	False characters	New false vs incumbent
1	272	59	175	42
2	291	49	141	8
3	317	30	140	7
4	320	28	143	10
5	323	25	133	0
6	328	20	147	14
7	334	16	148	15

Compared with other systems

Same 841-dev split, threshold 0.95. Naked is the transformer alone. ∪ regex is that model unioned with the frozen rules (the NERGAL recipe). Character scores are gold vs masked characters.

System	Mode	Whole /354	Residual	False chars	Char P	Char R
Regex (scrub_pii)	rules	245	73	133	97.36%	81.01%
GLiNER 2.5-multi zero-shot	naked	81	188	970	56.98%	21.22%
GLiNER 2.5-multi zero-shot	∪ regex	262	63	1,103	82.36%	85.02%
Historical GLiNER email12	naked	249	70	10	99.81%	85.39%
Historical GLiNER email12	∪ regex	289	51	143	97.54%	93.48%
XLM-R epoch 5	naked	298	42	57	98.96%	89.38%
NERGAL (this snapshot)	∪ regex	323	25	133	97.76%	95.95%

Zero-shot GLiNER 2.5 is not competitive here, especially on non-phone PII (7/185 whole vs 160 naked / 179 union for this snapshot). Fine-tuned historical GLiNER is the precise naked baseline (10 false characters) and still trails XLM-R on coverage. NERGAL is XLM-R epoch 5 plus the regex: 144/169 phone, 179/185 other PII. Exact-span precision 87.50%, recall 88.98%, F1 88.24%. The 133 false characters are the glued-email regex error; this seed does not add to that floor.

Trained GLiNER, HerBERT-large, and XLM-R-large were also compared on this split (plot above). GLiNER’s 0.50 coverage lead is the precision collapse in that figure; no GLiNER or HerBERT operating point passed the content-preservation gate.

Extra seeds

Seed	Whole /354	False chars	New false vs historical union
202609160 (this repo)	323	133	0
202609161	322	134	1
202609162	316	151	18

Load

This repo is the PII island: frozen scrub_pii.py plus nergal.py. pipeline("token-classification") will not match. Regex runs on the original text, the model adds spans at 0.95, then the two are unioned and replaced with [Telefon] / [PII].

from pathlib import Path
from huggingface_hub import snapshot_download

root = Path(snapshot_download("SlayerLab/NERGAL"))
import sys
sys.path.insert(0, str(root))
from nergal import Nergal

nergal = Nergal.from_pretrained(root, local_files_only=True)
masked, counts = nergal.scrub(text)

hybrid.json records threshold 0.95 and gap ids 250002 / 250003. test_nergal.py is synthetic (no corpus text). From this snapshot: python -m unittest test_nergal.

Base weights: FacebookAI/xlm-roberta-large revision c23d21b0620b635a76227c604d44e43a9f0ee389 (MIT).