danielschnell commited on
Commit
6d050fe
1 Parent(s): b06d53c

Updated dataset with combined full documents and improved secondary stress labels

Browse files
IGC-Wiki-News1-22.10.TEI-plbert.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c965c5ba3cf1f40805f69bb114d6ef903243f2396457e6f878803c9f9f558a3a
3
- size 1328985159
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ceb0d29d489310b17a54fc536eb978895f4e9cb108ee839baed2333c67fc4de4
3
+ size 1083454239
README.md CHANGED
@@ -5,30 +5,55 @@ language:
5
  ---
6
  ## Introduction
7
 
8
- This dataset, derived from the Icelandic Gigaword Corpus, is designed as a more comprehensive alternative to the existing dataset found at https://huggingface.co/datasets/styletts2-community/multilingual-pl-bert/tree/main/is. The original dataset, derived from just 52MB of raw text from the Icelandic Wikipedia, was processed using the espeak-ng backend for normalization and phonemization. However, the Icelandic module of espeak-ng, which has not been updated for over a decade, employs an outdated IPA dialect and a simplistic approach to stress marking. Additionally, the limited phonemization capabilities of the module independently contribute to inaccuracies in the phonetic transcriptions.
 
 
 
 
 
9
 
10
- Significant advancements in the normalization and G2P (Grapheme-to-Phoneme) conversion of Icelandic have been made through the Icelandic Language Technology program. More information about this program can be found [here](https://clarin.is/en/links/LTProjectPlan/). The tools developed in this program have been extensively used to enhance the quality of this dataset.
 
 
11
 
12
  ## Dataset
13
 
14
- This dataset surpasses its predecessor in size, incorporating not only text from the relatively small Icelandic Wikipedia but also from the extensive Icelandic Gigaword corpus. Specifically, we have enriched the [Wikipedia text](https://repository.clarin.is/repository/xmlui/handle/20.500.12537/252) with material from the [News1 corpus](https://repository.clarin.is/repository/xmlui/handle/20.500.12537/237). To adhere to the maximum size limit of 512 MB for the raw text, we combined the complete Wikipedia text with randomly shuffled paragraphs from the News1 corpus until reaching the size cap.
 
 
 
 
15
 
16
- In total, the dataset contains `2,212,618` rows, each corresponding to a paragraph in the IGC corpus' XML format. This structure differs from the original dataset, where each row represented an entire Wikipedia article. This change accounts for the significantly increased row count. The dataset allows for merging of paragraphs belonging to the same original document, as the URL and title rows accurately identify their source and order.
17
 
18
  ### Cleaning
19
 
20
- Prior to processing with the [Bert](https://huggingface.co/bert-base-multilingual-cased) tokenizer, the dataset underwent cleaning, deduplication, and language detection to filter out most non-Icelandic text. Paragraphs containing fewer than five words were also removed. These steps eliminated approximately 15% of the original dataset.
 
 
 
 
 
21
 
22
  ### Normalization
23
 
24
- For normalization, we adapted the [Regina Normalizer](https://github.com/grammatek/regina_normalizer), which employs a BI-LSTM Part-of-Speech (PoS) tagger. Although this makes the process somewhat time-consuming, the adaptions were necessary to handle a variety of edge cases in the diverse and sometimes unclean text within the IGC. The processing of approximately 2.5 GB of raw text took about one day, utilizing 50 CPU cores.
 
 
25
 
26
  ### Phonemization
27
 
28
- Phonemization was conducted using [IceG2P](https://github.com/grammatek/ice-g2p), which is also based on a BI-LSTM model. We made adaptations to ensure the IPA phoneset output aligns with the overall phoneset used in other PL-Bert datasets. Initially, we created and refined a new vocabulary from both the Wikipedia and News1 corpora. Following this, the BI-LSTM model was employed to generate phonetic transcriptions for the dictionary. We also enhanced stress labeling and incorporated secondary stresses after conducting compound analysis. A significant byproduct of this effort is a considerably improved G2P dictionary with more than 2.1 million transcriptions, which we plan to integrate into the G2P module and various other open-source projects involving Icelandic G2P.
 
 
 
 
 
 
29
 
30
  Ultimately, to ensure textual coherence, all paragraphs with incorrect Grapheme-to-Phoneme (G2P) transcriptions were excluded from the dataset.
31
 
32
  ## License
33
 
34
- The dataset is distributed under the same CC-by-4.0 license as the original source material from which the data was derived.
 
 
5
  ---
6
  ## Introduction
7
 
8
+ This dataset, derived from the Icelandic Gigaword Corpus, is designed as a more comprehensive alternative to the existing dataset found at
9
+ https://huggingface.co/datasets/styletts2-community/multilingual-pl-bert/tree/main/is.
10
+ The original dataset, derived from just 52MB of raw text from the Icelandic Wikipedia, was processed using the espeak-ng backend for
11
+ normalization and phonemization. However, the Icelandic module of espeak-ng, which has not been updated for over a decade, employs an outdated
12
+ IPA dialect and a simplistic approach to stress marking. Additionally, the limited phonemization capabilities of the module independently
13
+ contribute to inaccuracies in the phonetic transcriptions.
14
 
15
+ Significant advancements in the normalization and G2P (Grapheme-to-Phoneme) conversion of Icelandic have been made through the Icelandic
16
+ Language Technology program. More information about this program can be found [here](https://clarin.is/en/links/LTProjectPlan/).
17
+ The tools developed in this program have been extensively used to enhance the quality of this dataset.
18
 
19
  ## Dataset
20
 
21
+ This dataset surpasses its predecessor considerably in size, incorporating not only text from the relatively small Icelandic Wikipedia but also
22
+ from the extensive Icelandic Gigaword corpus. Specifically, we have enriched the
23
+ [Wikipedia text](https://repository.clarin.is/repository/xmlui/handle/20.500.12537/252) with material from the
24
+ [News1 corpus](https://repository.clarin.is/repository/xmlui/handle/20.500.12537/237). To adhere to the maximum size limit of 512 MB for the
25
+ raw text, we combined the complete Wikipedia text with randomly shuffled paragraphs from the News1 corpus until reaching the size cap.
26
 
27
+ In total, the dataset contains `400.676` rows, each corresponding to corresponding document in the IGC corpus' XML format.
28
 
29
  ### Cleaning
30
 
31
+ Prior to processing with the [Bert](https://huggingface.co/bert-base-multilingual-cased) tokenizer, the dataset underwent cleaning, deduplication,
32
+ and language detection to filter out most non-Icelandic text. Documents containing fewer than 10 words were also removed.
33
+ This preprocessing resulted in the elimination of 8,146 documents from the initial 55,475 in the Wikipedia corpus (approximately 14.7%)
34
+ and 28,869 from 1,545,671 in the News1 corpus (about 1.9%). The notably higher reduction in the Wikipedia corpus primarily arose from the
35
+ minimum word count criterion. However, this did not significantly diminish the total volume of text, which only saw a modest decrease from
36
+ 52.3MB to 49.68MB, a reduction of around 5%.
37
 
38
  ### Normalization
39
 
40
+ For normalization, we adapted the [Regina Normalizer](https://github.com/grammatek/regina_normalizer), which employs a BI-LSTM Part-of-Speech
41
+ (PoS) tagger. Although this makes the process somewhat time-consuming, the adaptions were necessary to handle a variety of edge cases in the diverse
42
+ and sometimes unclean text within the IGC. The processing of approximately 2.5 GB of raw text took about one day, utilizing 50 CPU cores.
43
 
44
  ### Phonemization
45
 
46
+ Phonemization was conducted using [IceG2P](https://github.com/grammatek/ice-g2p), which is also based on a BI-LSTM model. We made adaptations
47
+ to ensure the IPA phoneset output aligns with the overall phoneset used in other PL-Bert datasets. Initially, we created and refined a new vocabulary
48
+ from both the normalized Wikipedia and News1 corpora. Following this, the BI-LSTM model was employed to generate phonetic transcriptions for the dictionary.
49
+ We also enhanced stress labeling and incorporated secondary stresses after conducting compound analysis.
50
+
51
+ A significant byproduct of this effort is a considerably improved G2P dictionary with more than 2.1 million transcriptions, which we plan to
52
+ integrate into the G2P module and various other open-source projects involving Icelandic G2P.
53
 
54
  Ultimately, to ensure textual coherence, all paragraphs with incorrect Grapheme-to-Phoneme (G2P) transcriptions were excluded from the dataset.
55
 
56
  ## License
57
 
58
+ The dataset is distributed under the same CC-by-4.0 license as the original source material from which the data was derived.
59
+