Support streaming UDHR dataset (#4487)

* Support streaming udhr dataset

* Add BCP 47 language ID

* Update homepage

* Update dataset card

* Update metadata JSON

* Update dummy data

* Remove unused import

Commit from https://github.com/huggingface/datasets/commit/dc41a8821afac7efa0db66be2e11e2e78ffc31d6

README.md

@@ -501,7 +501,7 @@ pretty_name: The Universal Declaration of Human Rights (UDHR)
 
 ## Dataset Description
 
-- **Homepage:** https://www.ohchr.org/EN/UDHR/Pages/UDHRIndex.aspx, https://unicode.org/udhr/index.html
+- **Homepage:** https://www.ohchr.org/en/universal-declaration-of-human-rights, https://unicode.org/udhr/index.html
 - **Repository:** https://github.com/unicode-org/udhr
 - **Paper:**
 - **Leaderboard:**
@@ -516,7 +516,7 @@ Assembly in Paris on 10 December 1948 during its 183rd plenary meeting.
 
 © 1996 – 2009 The Office of the High Commissioner for Human Rights
 
-This plain text version prepared by the “UDHR in Unicode” project, https://www.unicode.org/udhr.
+This plain text version prepared by the "UDHR in Unicode" project, https://www.unicode.org/udhr.
 
 ### Supported Tasks and Leaderboards
 
@@ -536,16 +536,21 @@ text.
 
 ### Data Fields
 
-- `text`: The full document text with each line of text delimitted by a newline (`\n`).
+- `text`: The full document text with each line of text delimited by a newline (`\n`).
 - `lang_key`: The unique identifier of a given translation.
 - `lang_name`: The textual description of language/dialect.
 - `iso639-3`: The [iso639-3](https://iso639-3.sil.org/) language identifier.
 - `iso15924`: The [iso15924](https://unicode.org/iso15924/iso15924-codes.html) language identifier.
+- `bcp47`: The [BCP 47](https://www.rfc-editor.org/info/bcp47) language identifier.
 
 ### Data Splits
 
 Only a `train` split included which includes the full document in all languages.
 
+|                    | train |
+|--------------------|------:|
+| Number of examples |   488 |
+
 ## Dataset Creation
 
 ### Curation Rationale
@@ -583,7 +588,7 @@ document in the world and as such can be useful for settings requiring paired te
 
 In addition to the social and political significance of the United Nations' Universal Declaration of Human Rights,
 the document set a world record in 1999 for being the most translated document in the world and as such can be useful
-for settings requiring paired text between many languagers including those that are low resource and significantly
+for settings requiring paired text between many languages including those that are low resource and significantly
 underrepresented in NLP research.
 
 ### Discussion of Biases

dataset_infos.json

@@ -1 +1 @@
-{"default": {"description": "The Universal Declaration of Human Rights (UDHR) is a milestone document in the history of human rights. Drafted by\nrepresentatives with different legal and cultural backgrounds from all regions of the world, it set out, for the\nfirst time, fundamental human rights to be universally protected. The Declaration was adopted by the UN General\nAssembly in Paris on 10 December 1948 during its 183rd plenary meeting. The dataset includes translations of the\ndocument in 464 languages and dialects.\n\n\u00a9 1996 \u2013 2009 The Office of the High Commissioner for Human Rights\n\nThis plain text version prepared by the \u201cUDHR in Unicode\u201d project, https://www.unicode.org/udhr.\n", "citation": "", "homepage": "https://www.ohchr.org/EN/UDHR/Pages/UDHRIndex.aspx", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}, "lang_key": {"dtype": "string", "id": null, "_type": "Value"}, "lang_name": {"dtype": "string", "id": null, "_type": "Value"}, "iso639-3": {"dtype": "string", "id": null, "_type": "Value"}, "iso15924": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "udhn", "config_name": "default", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 6749909, "num_examples": 488, "dataset_name": "udhn"}}, "download_checksums": {"https://unicode.org/udhr/assemblies/udhr_xml.zip": {"num_bytes": 2389690, "checksum": "a3350912790196c6e1b26bfd1c8a50e8575f5cf185922ecd9bd15713d7d21438"}, "https://unicode.org/udhr/assemblies/udhr_txt.zip": {"num_bytes": 2215441, "checksum": "cb87ecb25b56f34e4fd6f22b323000524fd9c06ae2a29f122b048789cf17e9fe"}}, "download_size": 4605131, "post_processing_size": null, "dataset_size": 6749909, "size_in_bytes": 11355040}}
+{"default": {"description": "The Universal Declaration of Human Rights (UDHR) is a milestone document in the history of human rights. Drafted by\nrepresentatives with different legal and cultural backgrounds from all regions of the world, it set out, for the\nfirst time, fundamental human rights to be universally protected. The Declaration was adopted by the UN General\nAssembly in Paris on 10 December 1948 during its 183rd plenary meeting. The dataset includes translations of the\ndocument in 464+ languages and dialects.\n\n\u00a9 1996 \u2013 2009 The Office of the High Commissioner for Human Rights\n\nThis plain text version prepared by the \u201cUDHR in Unicode\u201d project, https://www.unicode.org/udhr.\n", "citation": "", "homepage": "https://www.ohchr.org/en/universal-declaration-of-human-rights", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}, "lang_key": {"dtype": "string", "id": null, "_type": "Value"}, "lang_name": {"dtype": "string", "id": null, "_type": "Value"}, "iso639-3": {"dtype": "string", "id": null, "_type": "Value"}, "iso15924": {"dtype": "string", "id": null, "_type": "Value"}, "bcp47": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "udhr", "config_name": "default", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 6753383, "num_examples": 488, "dataset_name": "udhr"}}, "download_checksums": {"https://unicode.org/udhr/assemblies/udhr_xml.zip": {"num_bytes": 2389690, "checksum": "a3350912790196c6e1b26bfd1c8a50e8575f5cf185922ecd9bd15713d7d21438"}}, "download_size": 2389690, "post_processing_size": null, "dataset_size": 6753383, "size_in_bytes": 9143073}}

dummy/1.0.0/dummy_data.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:19063b07d325e85347e7f9ad82e55a8db619e739784ec6613049ce51d98ced17
-size 11931
+oid sha256:0ee4681ffc0c16a7f1ff227cc8e9c5eda924c9d7ce1b78184b01f60265baf542
+size 6205

udhr.py

@@ -17,7 +17,6 @@
 """Universal Declaration of Human Rights"""
 
 
-import glob
 import os
 import xml.etree.ElementTree as ET
 
@@ -36,11 +35,9 @@ document in 464+ languages and dialects.
 This plain text version prepared by the “UDHR in Unicode” project, https://www.unicode.org/udhr.
 """
 
-_WEBPAGE = "https://www.ohchr.org/EN/UDHR/Pages/UDHRIndex.aspx"
+_HOMEPAGE = "https://www.ohchr.org/en/universal-declaration-of-human-rights"
 
-# XML for meta-info about language and TXT for text
-_XML_DOWNLOAD_URL = "https://unicode.org/udhr/assemblies/udhr_xml.zip"
-_TXT_DOWNLOAD_URL = "https://unicode.org/udhr/assemblies/udhr_txt.zip"
+_URL = "https://unicode.org/udhr/assemblies/udhr_xml.zip"
 
 
 class UDHN(datasets.GeneratorBasedBuilder):
@@ -58,36 +55,31 @@ class UDHN(datasets.GeneratorBasedBuilder):
                     "lang_name": datasets.Value("string"),
                     "iso639-3": datasets.Value("string"),
                     "iso15924": datasets.Value("string"),
+                    "bcp47": datasets.Value("string"),
                 }
             ),
-            homepage=_WEBPAGE,
+            homepage=_HOMEPAGE,
         )
 
     def _split_generators(self, dl_manager):
-        xml_dir = dl_manager.download_and_extract(_XML_DOWNLOAD_URL)
-        txt_dir = dl_manager.download_and_extract(_TXT_DOWNLOAD_URL)
+        dl_dir = dl_manager.download_and_extract(_URL)
 
         return [
-            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"xml_dir": xml_dir, "txt_dir": txt_dir}),
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN, gen_kwargs={"file_paths": dl_manager.iter_files(dl_dir)}
+            ),
         ]
 
-    def _generate_examples(self, xml_dir, txt_dir):
-        xml_paths = glob.glob(os.path.join(xml_dir, "udhr_*.xml"))
-        for i, xml_path in enumerate(xml_paths):
-            lang_attr = ET.parse(xml_path).getroot().attrib
-            lang = lang_attr["key"]
-
-            txt_path = os.path.join(txt_dir, f"udhr_{lang}.txt")
-            assert os.path.exists(txt_path), f"xml file {xml_path} found with no corresponding txt file"
-
-            with open(txt_path, "r", encoding="utf-8") as f:
-                lines = f.readlines()[6:]  # ignore first 6 lines (metainfo)
-            text = "\n".join([line.strip() for line in lines if len(line.strip()) > 0])
-
-            yield i, {
-                "text": text,
-                "lang_key": lang,
-                "lang_name": lang_attr["n"],
-                "iso639-3": lang_attr["iso639-3"],
-                "iso15924": lang_attr["iso15924"],
-            }
+    def _generate_examples(self, file_paths):
+        for id_, path in enumerate(file_paths):
+            if os.path.basename(path).startswith("udhr_"):
+                root = ET.parse(path).getroot()
+                text = "\n".join([line.strip() for line in root.itertext() if line.strip()])
+                yield id_, {
+                    "text": text,
+                    "lang_key": root.get("key"),
+                    "lang_name": root.get("n"),
+                    "iso639-3": root.get("iso639-3"),
+                    "iso15924": root.get("iso15924"),
+                    "bcp47": root.get("{http://www.w3.org/XML/1998/namespace}lang"),
+                }