update

README.md

@@ -19,8 +19,8 @@ license: apache-2.0
 | scandi_langid |  | TRAIN: 239618, TEST: 59840 |  | [kardosdrur/scandi-langid](https://huggingface.co/datasets/kardosdrur/scandi-langid) |
 | nordic_langid | [Discriminating Between Similar Nordic Languages](https://aclanthology.org/2021.vardial-1.8/) | TRAIN: 226159, TEST: 10700 | 重点关注六种北欧语言之间的区别：丹麦语、瑞典语、挪威语（尼诺斯克语）、挪威语（博克马尔语）、法罗语和冰岛语。 | [strombergnlp/nordic_langid](https://huggingface.co/datasets/strombergnlp/nordic_langid) |
 | mike0307 | [Mike0307/language-detection](https://huggingface.co/datasets/Mike0307/language-detection) | TRAIN: 33095, VALID: 4040, TEST: 4048 |  |  |
-| tatoeba | [tatoeba](https://tatoeba.org/); [Tatoeba Paper](https://arxiv.org/abs/1812.10464v2) |  | Tatoeba 是句子和翻译的集合。 | [tatoeba](https://huggingface.co/datasets/tatoeba) |
-| bucc2018 | [bucc2018](https://comparable.limsi.fr/bucc2018/bucc2018-task.html) |  | 共享任务：识别可比语料库中的平行句子 |  |
+| tatoeba | [tatoeba](https://tatoeba.org/); [Tatoeba Paper](https://arxiv.org/abs/1812.10464v2) | TRAIN: 702895 | Tatoeba 是句子和翻译的集合。 | [tatoeba](https://huggingface.co/datasets/tatoeba) |
+| bucc2018 | [bucc2018](https://comparable.limsi.fr/bucc2018/bucc2018-task.html) | TRAIN: 2173318, TEST: 2125879 | 共享任务：识别可比语料库中的平行句子，语言：de, en, fr, ru, zh |  |
 
 
 

data/ducc2018.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8768a7e674e4aeb919fd745847a09feff90b641de8e0fcbb7dddc1a60cadbc6f
+size 877661718

data/tatoeba.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:027bab142da675b190f9bd09e72a20529232729fde8f532938057b32d690b52a
+size 82855499

examples/preprocess/preprocess_bucc2018.py

@@ -0,0 +1,133 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+from collections import defaultdict
+import json
+import os
+from pathlib import Path
+import shutil
+import sys
+import tarfile
+import tempfile
+
+pwd = os.path.abspath(os.path.dirname(__file__))
+sys.path.append(os.path.join(pwd, "../../"))
+
+from datasets import load_dataset, DownloadMode
+from tqdm import tqdm
+
+from language_identification import LANGUAGE_MAP
+from project_settings import project_path
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--dataset_dir",
+        default=r"E:\programmer\nlp_datasets\ducc2018",
+        type=str
+    )
+    parser.add_argument(
+        "--output_file",
+        default=(project_path / "data/ducc2018.jsonl"),
+        type=str
+    )
+
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+
+    dataset_dir = Path(args.dataset_dir)
+
+    # extract
+    out_root = Path(tempfile.gettempdir()) / "bucc2018"
+    if not out_root.exists():
+        out_root.mkdir(parents=True, exist_ok=True)
+        print(out_root.as_posix())
+
+        train_files = [
+            "bucc2018-de-en.training-gold.tar.bz2",
+            "bucc2018-fr-en.training-gold.tar.bz2",
+            "bucc2018-ru-en.training-gold.tar.bz2",
+            "bucc2018-zh-en.training-gold.tar.bz2",
+            "bucc2018-de-en.test.tar.bz2",
+            "bucc2018-fr-en.test.tar.bz2",
+            "bucc2018-ru-en.test.tar.bz2",
+            "bucc2018-zh-en.test.tar.bz2",
+        ]
+        for train_file in train_files:
+            file_path = dataset_dir / train_file
+
+            with tarfile.open(file_path, "r:bz2") as tar:
+                tar.extractall(path=out_root.as_posix())
+
+    # read
+    root_path = out_root / "bucc2018"
+    name_list = [
+        "de-en", "fr-en", "ru-en", "zh-en"
+    ]
+
+    split_map = {
+        "training": "train",
+    }
+
+    language_map = {
+        "zh": "zh-cn"
+    }
+
+    text_set = set()
+    counter = defaultdict(int)
+    with open(args.output_file, "w", encoding="utf-8") as fout:
+        for name in name_list:
+            name_path = root_path / name
+
+            for split_ in ["training", "test"]:
+                for language in name.split("-"):
+                    train_file = name_path / "{}.{}.{}".format(name, split_, language)
+
+                    with open(train_file, "r", encoding="utf-8") as fin:
+                        for row in fin:
+                            row = str(row).strip()
+                            splits = row.split("\t")
+
+                            if len(splits) != 2:
+                                print("skip row: {}".format(row))
+                                continue
+
+                            text = splits[1]
+                            text = text.strip()
+                            text = text.replace(" ", " ")
+                            text = text.replace("­", "-")
+
+                            if text in text_set:
+                                continue
+                            text_set.add(text)
+
+                            language = language_map.get(language, language)
+                            if language not in LANGUAGE_MAP.keys():
+                                raise AssertionError(language)
+
+                            if split_ in split_map.keys():
+                                split = split_map[split_]
+                            else:
+                                split = split_
+                            row = {
+                                "text": text,
+                                "language": language,
+                                "data_source": "bucc2018",
+                                "split": split
+                            }
+                            row = json.dumps(row, ensure_ascii=False)
+                            fout.write("{}\n".format(row))
+                            counter[split] += 1
+
+    print("counter: {}".format(counter))
+    shutil.rmtree(out_root.as_posix())
+    return
+
+
+if __name__ == '__main__':
+    main()

examples/preprocess/preprocess_nbnn.py

@@ -41,6 +41,7 @@ def main():
         path=args.dataset_path,
         cache_dir=args.dataset_cache_dir,
         # download_mode=DownloadMode.FORCE_REDOWNLOAD
+        streaming=True
     )
     print(dataset_dict)
 

examples/preprocess/preprocess_tatoeba.py

@@ -0,0 +1,91 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+from collections import defaultdict
+import json
+import os
+import sys
+
+pwd = os.path.abspath(os.path.dirname(__file__))
+sys.path.append(os.path.join(pwd, "../../"))
+
+from datasets import load_dataset, DownloadMode
+from tqdm import tqdm
+
+from language_identification import LANGUAGE_MAP
+from project_settings import project_path
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--dataset_path", default="tatoeba", type=str)
+    parser.add_argument(
+        "--dataset_cache_dir",
+        default=(project_path / "hub_datasets").as_posix(),
+        type=str
+    )
+    parser.add_argument(
+        "--output_file",
+        default=(project_path / "data/tatoeba.jsonl"),
+        type=str
+    )
+
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+
+    name_list = ["en-mr", "eo-nl", "es-gl", "es-pt", "fr-ru"]
+
+    dataset_dict_list = list()
+    for name in name_list:
+        dataset_dict = load_dataset(
+            path=args.dataset_path,
+            name=name,
+            cache_dir=args.dataset_cache_dir,
+            # download_mode=DownloadMode.FORCE_REDOWNLOAD
+        )
+        print(dataset_dict)
+        dataset_dict_list.append((name, dataset_dict))
+
+    text_set = set()
+    counter = defaultdict(int)
+    with open(args.output_file, "w", encoding="utf-8") as f:
+        for _, dataset_dict in dataset_dict_list:
+            for k, v in dataset_dict.items():
+                split = k
+                if split not in ("train", "validation", "test"):
+                    print("skip split: {}".format(split))
+                    continue
+
+                for sample in tqdm(v):
+                    translation = sample["translation"]
+                    for language, text in translation.items():
+                        text = text.strip()
+
+                        if text in text_set:
+                            continue
+                        text_set.add(text)
+
+                        if language not in LANGUAGE_MAP.keys():
+                            raise AssertionError(language)
+
+                        row = {
+                            "text": text,
+                            "language": language,
+                            "data_source": "tatoeba",
+                            "split": split
+                        }
+                        row = json.dumps(row, ensure_ascii=False)
+                        f.write("{}\n".format(row))
+                        counter[split] += 1
+
+    print("counter: {}".format(counter))
+
+    return
+
+
+if __name__ == '__main__':
+    main()

language_identification.py

@@ -39,9 +39,11 @@ LANGUAGE_MAP = {
     "de": "german",
     "el": "modern greek",
     "en": "english",
+    "eo": "esperanto",
     "es": "spanish",
     "fo": "faroese",
     "fr": "french",
+    "gl": "galician",
     "hi": "hindi",
     "is": "icelandic",
     "it": "italian",
@@ -50,6 +52,7 @@ LANGUAGE_MAP = {
     "no": "norwegian",
     "no-b": "norwegian (bokmål)",
     "no-n": "norwegian (nynorsk)",
+    "mr": "marathi",
     "pl": "polish",
     "pt": "portuguese",
     "ru": "russian",

requirements.txt

@@ -1,5 +1,6 @@
-datasets==2.4.0
+datasets==2.10.1
 fsspec==2023.9.2
+pyarrow==9.0.0
 tqdm==4.66.1
 pandas==2.0.3
 xlrd==1.2.0