Update README.md
Browse files
README.md
CHANGED
|
@@ -23,4 +23,17 @@ model-index:
|
|
| 23 |
metrics:
|
| 24 |
- type: 'bleu'
|
| 25 |
value: 17.2
|
| 26 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
metrics:
|
| 24 |
- type: 'bleu'
|
| 25 |
value: 17.2
|
| 26 |
+
---
|
| 27 |
+
|
| 28 |
+
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.
|
| 29 |
+
<br>
|
| 30 |
+
It includes special characters such as þ.
|
| 31 |
+
<br>
|
| 32 |
+
This model reflects the spelling inconsistencies characteristic of Middle English.
|
| 33 |
+
<br>
|
| 34 |
+
Because the model is trained largely on poetry and some prose, it is best at translating those sorts of tasks.
|
| 35 |
+
<br>
|
| 36 |
+
Performance can be improved by sentence tokenizing input data and translating sentence-by-sentence.
|
| 37 |
+
<br>
|
| 38 |
+
Removing contractions (hadn't -> had not) also boosts performance.
|
| 39 |
+
|