Update README.md
Browse files
README.md
CHANGED
@@ -19,17 +19,19 @@ widget:
|
|
19 |
|
20 |
# ERWT-year
|
21 |
|
22 |
-
A
|
|
|
|
|
23 |
|
24 |
|
25 |
**Warning**: This model was trained for **experimental purposes**, please use it with care.
|
26 |
|
27 |
|
28 |
-
You find more detailed information below
|
29 |
|
30 |
## Background
|
31 |
|
32 |
-
ERWT was created using a
|
33 |
|
34 |
To create this ERWT model we fine-tuned [`distilbert-base-cased`](https://huggingface.co/distilbert-base-cased) on a random subsample of the Heritage Made Digital newspaper of about half a billion words. We slightly adapted to the training routine by adding the year of publication and a special token `[DATE]` in front of each text segment (i.e. a chunk of hundred tokens).
|
35 |
|
|
|
19 |
|
20 |
# ERWT-year
|
21 |
|
22 |
+
A Historical Language Model,
|
23 |
+
|
24 |
+
ERWT is a fine-tuned [`distilbert-base-cased`](https://huggingface.co/distilbert-base-cased) model trained on historical newspapers from the [Heritage Made Digital collection](https://huggingface.co/datasets/davanstrien/hmd-erwt-training) with temporal metadata.
|
25 |
|
26 |
|
27 |
**Warning**: This model was trained for **experimental purposes**, please use it with care.
|
28 |
|
29 |
|
30 |
+
You find more detailed information below, especially the "limitations" section (hey, seriously, read this ... very important, we don't write this just to look smart). You can also consult our working paper ["Metadata Might Make Language Models Better"](https://drive.google.com/file/d/1Xp21KENzIeEqFpKvO85FkHynC0PNwBn7/view?usp=sharing) for more background and nerdy evaluation stuff (still, work in progress, handle with care and kindness).
|
31 |
|
32 |
## Background
|
33 |
|
34 |
+
ERWT was created using a **M**eta**D**ata **M**asking **A**pproach (or **MDMA** 💊), in which we train a Masked Language Model simultaneously on text and metadata. Our intuition was that incorporating information that is not explicitly present in the text—such as the time of publication or the political leaning of the author—may make language models "better" in the sense of being more sensitive to historical and political aspects of language use.
|
35 |
|
36 |
To create this ERWT model we fine-tuned [`distilbert-base-cased`](https://huggingface.co/distilbert-base-cased) on a random subsample of the Heritage Made Digital newspaper of about half a billion words. We slightly adapted to the training routine by adding the year of publication and a special token `[DATE]` in front of each text segment (i.e. a chunk of hundred tokens).
|
37 |
|