jannikskytt commited on
Commit
1050884
1 Parent(s): 405e6ed

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +30 -1
README.md CHANGED
@@ -60,4 +60,33 @@ configs:
60
  - split: train
61
  path: Verb inflection analogy/train*
62
 
63
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60
  - split: train
61
  path: Verb inflection analogy/train*
62
 
63
+ ---
64
+
65
+ # Danish medical word embeddings
66
+
67
+ MeDa-We was trained on a Danish medical corpus of 123M tokens. The word embeddings are 300-dimensional and are trained using [FastText](https://fasttext.cc/).
68
+
69
+ The embeddings were trained for 10 epochs using a window size of 5 and 10 negative samples.
70
+
71
+ The development of the corpus and word embeddings is described further in our [paper](https://aclanthology.org/2023.nodalida-1.31/).
72
+
73
+ We also trained a transformer model on the developed corpus which can be found [here](https://huggingface.co/jannikskytt/MeDa-Bert).
74
+
75
+ ### Citing
76
+
77
+ ```
78
+ @inproceedings{pedersen-etal-2023-meda,
79
+ title = "{M}e{D}a-{BERT}: A medical {D}anish pretrained transformer model",
80
+ author = "Pedersen, Jannik and
81
+ Laursen, Martin and
82
+ Vinholt, Pernille and
83
+ Savarimuthu, Thiusius Rajeeth",
84
+ booktitle = "Proceedings of the 24th Nordic Conference on Computational Linguistics (NoDaLiDa)",
85
+ month = may,
86
+ year = "2023",
87
+ address = "T{\'o}rshavn, Faroe Islands",
88
+ publisher = "University of Tartu Library",
89
+ url = "https://aclanthology.org/2023.nodalida-1.31",
90
+ pages = "301--307",
91
+ }
92
+ ```