|
--- |
|
language: |
|
- 'en' |
|
- 'me' |
|
license: afl-3.0 |
|
tags: |
|
- 'translation' |
|
datasets: |
|
- 'Qilex/EN-ME' |
|
metrics: |
|
- bleu |
|
|
|
model-index: |
|
- name: 'en-me' |
|
results: |
|
- task: |
|
type: 'translation' |
|
name: 'translation en-me' |
|
dataset: |
|
type: 'translation' |
|
name: 'Qilex/EN-ME' |
|
|
|
metrics: |
|
- type: 'bleu' |
|
value: 17.2 |
|
--- |
|
|
|
This is a BART-large model finetuned on roughly 58000 aligned sentence pairs in English and Middle English, collected from the works of Geoffrey Chaucer, John Wycliffe, and the Gawain Poet. |
|
<br> |
|
It includes special characters such as þ. |
|
<br> |
|
This model reflects the spelling inconsistencies characteristic of Middle English. |
|
<br> |
|
Because the model is trained largely on poetry and some prose, it is best at translating those sorts of tasks. |
|
<br> |
|
Performance can be improved by sentence tokenizing input data and translating sentence-by-sentence. |
|
<br> |
|
Removing contractions (hadn't -> had not) also boosts performance. |
|
|
|
|
