Datasets:

Languages:
German
Multilinguality:
monolingual
Size Categories:
10M<n<100M
ArXiv:
License:
PhilipMay commited on
Commit
b9a136e
1 Parent(s): 7a4bb33

add Anomalies

Browse files
Files changed (1) hide show
  1. README.md +21 -0
README.md CHANGED
@@ -54,6 +54,27 @@ For us it has proven useful to delete the following pairs of sentences:
54
  - **`en_de_token_count`**: number of tokens of the `de` text, tokenized with [deepset/gbert-large](https://huggingface.co/deepset/gbert-large)
55
  - **`cos_sim`**: the [cosine similarity](https://en.wikipedia.org/wiki/Cosine_similarity) of both sentences measured with [sentence-transformers/paraphrase-multilingual-mpnet-base-v2](https://huggingface.co/sentence-transformers/paraphrase-multilingual-mpnet-base-v2)
56
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
57
  ## Parallel text corpora used
58
  | Corpus name & link | Number of paraphrases |
59
  |-----------------------------------------------------------------------|----------------------:|
 
54
  - **`en_de_token_count`**: number of tokens of the `de` text, tokenized with [deepset/gbert-large](https://huggingface.co/deepset/gbert-large)
55
  - **`cos_sim`**: the [cosine similarity](https://en.wikipedia.org/wiki/Cosine_similarity) of both sentences measured with [sentence-transformers/paraphrase-multilingual-mpnet-base-v2](https://huggingface.co/sentence-transformers/paraphrase-multilingual-mpnet-base-v2)
56
 
57
+ ## Anomalies in the texts
58
+ It is noticeable that the [OpenSubtitles](https://opus.nlpl.eu/OpenSubtitles-v2018.php) texts have weird dash prefixes. This looks like this:
59
+
60
+ ```
61
+ - Hast du was draufgetan?
62
+ ```
63
+
64
+ To remove them you could apply this function:
65
+
66
+ ```python
67
+ import re
68
+
69
+ def clean_text(text):
70
+ text = re.sub("^[-\s]*", "", text)
71
+ text = re.sub("[-\s]*$", "", text)
72
+ return text
73
+
74
+ df["de"] = df["de"].apply(clean_text)
75
+ df["en_de"] = df["en_de"].apply(clean_text)
76
+ ```
77
+
78
  ## Parallel text corpora used
79
  | Corpus name & link | Number of paraphrases |
80
  |-----------------------------------------------------------------------|----------------------:|