NLLB-200-Distilled-600M-Middle-English-v1.5

All of this Readme is generated by Claude 4.6 Sonnet!

---

NLLB-200-distilled-600M: English ↔ Middle English (v1.5)

A LoRA adapter fine-tuned on top of facebook/nllb-200-distilled-600M for bidirectional translation between Modern English and Middle English (Wycliffe/Chaucer era, ~1380–1400).

This is v1.5 — an incremental improvement over v1.0 using the American Standard Version (ASV, 1901) as the Modern English source instead of KJV (1611), trained for 5 epochs instead of 3.

Version History

Model Description

This adapter teaches NLLB-200 to translate between Modern English and Middle English in the style of the Wycliffe Bible (~1382). It is the faithful back-translation component of a two-model pipeline — optimized for meaning preservation rather than stylistic fluency.

The companion fluency model (Qwen3-4B LoRA) is trained using synthetic data generated by this model.

Benchmark Results

Evaluated on 100 held-out verse pairs from the Wycliffe Bible vs. ASV.

Note: The gap between chrF++ and BLEU is expected — BLEU penalizes valid Middle English spelling variants (e.g. ech vs euery, figis vs figus). chrF++ is the more appropriate metric for this task. The theoretical ceiling is ~65-70 due to inconsistent spelling in Wycliffe manuscripts.

Training Details

Training Curve

Data

Training data sourced from two public domain texts, aligned by book/chapter/verse:

• Wycliffe Bible (~1382) — Middle English target
• American Standard Version / ASV (1901) — Modern English source

V1.5 improvement over v1.0: ASV (1901) is significantly more modern than KJV (1611) — no thou/thee/hath/spake constructions. This reduces register mismatch between training source and real-world modern English input.

Usage

from transformers import NllbTokenizer, AutoModelForSeq2SeqLM
from peft import PeftModel
import torch

MODEL_ID = 'facebook/nllb-200-distilled-600M'
ADAPTER  = 'MihaiPopa-1/NLLB-200-Distilled-600M-Middle-English-v1.5'  # update with your HF repo

# 1. Load tokenizer + add ME token
tokenizer = NllbTokenizer.from_pretrained(MODEL_ID)
ME_TOKEN = 'enm_Latn'
if ME_TOKEN not in tokenizer.additional_special_tokens:
    tokenizer.add_special_tokens({'additional_special_tokens': [ME_TOKEN]})

ME_TOKEN_ID = tokenizer.convert_tokens_to_ids(ME_TOKEN)
EN_TOKEN_ID = tokenizer.convert_tokens_to_ids('eng_Latn')

# 2. Load base model + resize + adapter (order matters!)
model = AutoModelForSeq2SeqLM.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.float16,
    device_map='auto'
)
model.resize_token_embeddings(len(tokenizer))
model = PeftModel.from_pretrained(model, ADAPTER)
model.eval()

def translate(text, src_lang='eng_Latn', tgt_token_id=ME_TOKEN_ID):
    tokenizer.src_lang = src_lang
    inputs = tokenizer(text, return_tensors='pt', truncation=True, max_length=64).to(model.device)
    with torch.no_grad():
        out = model.generate(
            **inputs,
            forced_bos_token_id=tgt_token_id,
            max_new_tokens=64,
            num_beams=4
        )
    return tokenizer.decode(out[0], skip_special_tokens=True)

# Modern English → Middle English
print(translate("I will call you later tonight."))
# → "Y schal clepe you later in this nyyt."

print(translate("Let's meet at the coffee shop at noon."))
# → "Meete we in the coffee shop at myddai."

# Middle English → Modern English
print(translate("Y haue not ete ony thing fro the morewtid.", 
                src_lang='enm_Latn', tgt_token_id=EN_TOKEN_ID))
# → "I have not eaten anything since breakfast."

Sample Outputs

Limitations

• Biblical register bias — training data is formal Biblical prose. Casual modern slang may produce stilted output.
• Post-1400 vocabulary — words like coffee, phone, traffic have no ME equivalent and are either left in modern form or phonologically naturalized.
• Sequence length — trained with MAX_LEN=64; very long sentences may be truncated.
• Spelling variance ceiling — Wycliffe spelling is inconsistent across manuscripts, capping chrF++ at ~65-70 regardless of model quality.

Pipeline

This model is part of a two-model English ↔ Middle English translation system:

NLLB-600M LoRA  →  faithful back-translation  →  synthetic training data
                                                          ↓
Qwen3-4B LoRA   →  fluent, natural ME output  →  daily use (texts, messages)

Citation

@misc{nllb-middle-english-lora,
  title   = {NLLB-200 LoRA: English ↔ Middle English Translation (v1.5)},
  year    = {2025},
  note    = {Fine-tuned on Wycliffe Bible / ASV parallel corpus}
}