sloie.py

""" SloIE is a manually labelled dataset of Slovene idiomatic expressions. """


import os

import datasets

_CITATION = """\
@article{skvorc2022mice,
title = {MICE: Mining Idioms with Contextual Embeddings},
journal = {Knowledge-Based Systems},
volume = {235},
pages = {107606},
year = {2022},
issn = {0950-7051},
doi = {https://doi.org/10.1016/j.knosys.2021.107606},
url = {https://www.sciencedirect.com/science/article/pii/S0950705121008686},
author = {{\v S}kvorc, Tadej and Gantar, Polona and Robnik-{\v S}ikonja, Marko},
}
"""

_DESCRIPTION = """\
SloIE is a manually labelled dataset of Slovene idiomatic expressions. 
It contains 29,400 sentences with 75 different expressions that can occur with either a literal or an idiomatic meaning, 
with appropriate manual annotations for each token. The idiomatic expressions were selected from the Slovene Lexical 
Database (http://hdl.handle.net/11356/1030). Only expressions that can occur with both a literal and an idiomatic 
meaning were selected. The sentences were extracted from the Gigafida corpus.
"""

_HOMEPAGE = "http://hdl.handle.net/11356/1030"

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

_URLS = {
    "sloie": "https://www.clarin.si/repository/xmlui/bitstream/handle/11356/1335/SloIE.zip"
}


class SloIE(datasets.GeneratorBasedBuilder):
    """ SloIE is a manually labelled dataset of Slovene idiomatic expressions. """

    VERSION = datasets.Version("1.0.0")

    def _info(self):
        features = datasets.Features(
            {
                "sentence": datasets.Value("string"),
                "expression": datasets.Value("string"),
                "word_order": datasets.Sequence(datasets.Value("int32")),
                "sentence_words": datasets.Sequence(datasets.Value("string")),
                "is_idiom": 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["sloie"]
        data_dir = dl_manager.download_and_extract(urls)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"file_path": os.path.join(data_dir, "SloIE.txt")}
            )
        ]

    def _generate_examples(self, file_path):
        idx_instance = 0
        with open(file_path, "r", encoding="utf-8") as f:
            line = f.readline().strip()
            while line:
                assert line.startswith("#")
                sent = line[1:]  # Remove initial "#"
                word_order = list(map(int, f.readline().strip().split(" ")))
                expression = ""
                sentence_words, idiomaticity = [], []

                line = f.readline().strip()
                while line:
                    token_info = line.split("\t")
                    word, is_idiomatic_str, expression = token_info
                    sentence_words.append(word)
                    idiomaticity.append(is_idiomatic_str)

                    line = f.readline().strip()
                    # Encountered start of the next sentence - Note that "#" may also be an annotated word, hence the second condition
                    if line.startswith("#") and len(line.split("\t")) == 1:
                        break

                yield idx_instance, {
                    "sentence": sent,
                    "expression": expression,
                    "word_order": word_order,
                    "sentence_words": sentence_words,
                    "is_idiom": idiomaticity
                }
                idx_instance += 1