slownet.py

"""SloWNet is a semantic lexicon of Slovene."""


import os
import xml.etree.ElementTree as ET
import datasets


_CITATION = """\
@inproceedings{fiser2012slownet,
  title={sloWNet 3.0: development, extension and cleaning},
  author={Fi{\v{s}}er, Darja and Novak, Jernej and Erjavec, Toma{\v{z}}},
  booktitle={Proceedings of 6th International Global Wordnet Conference (GWC 2012)},
  pages={113--117},
  year={2012}
}
"""

_DESCRIPTION = """\
sloWNet is the Slovene WordNet developed in the expand approach: it contains the complete Princeton WordNet 3.0 and 
over 70 000 Slovene literals. These literals have been added automatically using different types of existing resources, 
such as bilingual dictionaries, parallel corpora and Wikipedia. 33 000 literals have been subsequently hand-validated.
"""

_HOMEPAGE = "http://hdl.handle.net/11356/1026"

_LICENSE = "Creative Commons - Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)"

_URLS = {
    "slownet": "https://www.clarin.si/repository/xmlui/bitstream/handle/11356/1026/slownet.zip"
}

XML_NAMESPACE = "{http://www.w3.org/XML/1998/namespace}"
UNK_INT = -1
UNK_STR = "N/A"


class SloWNet(datasets.GeneratorBasedBuilder):
    """SloWNet is a semantic lexicon of Slovene."""

    VERSION = datasets.Version("3.1.0")

    def _info(self):
        features = datasets.Features(
            {
                "id": datasets.Value("string"),
                "pos": datasets.Value("string"),
                "bcs": datasets.Value("int32"),
                "en_synonyms": {
                    "words": datasets.Sequence(datasets.Value("string")),
                    "senses": datasets.Sequence(datasets.Value("int32")),
                    "pwnids": datasets.Sequence(datasets.Value("string"))
                },
                "sl_synonyms": {
                    "words": datasets.Sequence(datasets.Value("string")),
                    "is_validated": datasets.Sequence(datasets.Value("bool"))
                },
                "en_def": datasets.Value("string"),
                "sl_def": datasets.Value("string"),
                "en_usages": datasets.Sequence(datasets.Value("string")),
                "sl_usages": datasets.Sequence(datasets.Value("string")),
                "ilrs": {
                    "types": datasets.Sequence(datasets.Value("string")),
                    "id_synsets": datasets.Sequence(datasets.Value("string"))
                },
                "semeval07_cluster": datasets.Value("string"),
                "domains": datasets.Sequence(datasets.Value("string"))
            }
        )

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION
        )

    def _split_generators(self, dl_manager):
        urls = _URLS["slownet"]
        data_dir = dl_manager.download_and_extract(urls)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"file_path": os.path.join(data_dir, "slownet-2015-05-07.xml")}
            )
        ]

    def _generate_examples(self, file_path):
        curr_doc = ET.parse(file_path)
        root = curr_doc.getroot()

        for idx_synset, synset in enumerate(root.iterfind("SYNSET")):
            id_synset = synset.find("ID").text.strip()
            pos_tag = synset.find("POS").text.strip()

            # <BCS>
            base_concept_set = synset.find("BCS")
            if base_concept_set is not None:
                base_concept_set = int(base_concept_set.text.strip())
            else:
                base_concept_set = UNK_INT

            en_synonyms = {"words": [], "senses": [], "pwnids": []}
            sl_synonyms = {"words": [], "is_validated": []}
            # <SYNONYM>
            for curr_synonyms in synset.findall("SYNONYM"):
                lang = curr_synonyms.attrib[f"{XML_NAMESPACE}lang"]

                if lang == "en":
                    for syn in curr_synonyms.iterfind("LITERAL"):
                        en_synonyms["words"].append(syn.text.strip())
                        en_synonyms["senses"].append(int(syn.attrib["sense"]))
                        en_synonyms["pwnids"].append(syn.attrib["pwnid"])
                else:
                    for syn in curr_synonyms.iterfind("LITERAL"):
                        sl_synonyms["words"].append(syn.text.strip())
                        sl_synonyms["is_validated"].append(syn.attrib.get("lnote", "auto") == "manual")

            # <DEF>: 1 or 2
            sl_def, en_def = UNK_STR, UNK_STR
            for curr_def in synset.findall("DEF"):
                lang = curr_def.attrib[f"{XML_NAMESPACE}lang"]
                if lang == "en":
                    en_def = curr_def.text.strip()
                else:
                    sl_def = curr_def.text.strip()

            en_usages, sl_usages = [], []
            # <USAGE> - TODO: what does "lnote" indicate in Slovene usages? (some kind of ID?)
            for curr_usage in synset.findall("USAGE"):
                lang = curr_usage.attrib[f"{XML_NAMESPACE}lang"]
                TARGET_LIST = en_usages if lang == "en" else sl_usages
                TARGET_LIST.append(curr_usage.text.strip())

            # <ILR>
            ilrs = {"types": [], "id_synsets": []}
            for curr_ilr in synset.findall("ILR"):
                ilrs["types"].append(curr_ilr.attrib["type"])
                ilrs["id_synsets"].append(curr_ilr.text.strip())

            # <CLUSTER>
            semeval07_cluster = synset.find("CLUSTER")
            if semeval07_cluster is not None:
                semeval07_cluster = semeval07_cluster.text.strip()
            else:
                semeval07_cluster = UNK_STR

            # <DOMAIN>
            domains = list(map(lambda el: el.text.strip(), synset.findall("DOMAIN")))

            yield idx_synset, {
                "id": id_synset,
                "pos": pos_tag,
                "bcs": base_concept_set,
                "en_synonyms": en_synonyms,
                "sl_synonyms": sl_synonyms,
                "en_def": en_def,
                "sl_def": sl_def,
                "en_usages": en_usages,
                "sl_usages": sl_usages,
                "ilrs": ilrs,
                "semeval07_cluster": semeval07_cluster,
                "domains": domains
            }