metadata

license: apache-2.0
task_categories:
  - text-classification
  - text2text-generation
language:
  - el
tags:
  - text
  - keypoint-analysis
  - keypoint-matching
  - keypoint-generation
size_categories:
  - 10K<n<100K

Dataset Card for ArgKP_2021_GR

Dataset Summary

The ArgKP-2021-GR dataset is a Greek-language benchmark of 27.519 <argument, key point> pairs labeled as matching/non-matching, for 31 controversial topics for evaluating Key Point Analysis systems in Greek. Key Point Analysis is a task introduced by Bar-Haim et al. (2020) that enables summarization of multiple arguments for or against a debatable topic into a set of concise key points, ranked based on their prevalence in the original arguments. The task obtained more attention during the KPA-2021-Shared task introduced by IBM the following year. This dataset introduces the Greek version of the official Shared task's dataset ArgKP-2021 generated through machine and human translation.

Supported Tasks and Leaderboards

Fully automatic Key Point Analysis systems consist of a Key Point Generation and a Key Point Matching model.

- Key Point Generation (KPG): Given a debatable topic and a set of crowd arguments supporting or contesting the topic, generate a set of key points for each stance of the topic.

Evaluation: ROUGE, BERTScore--> for details refer to my MSc Thesis Project

- Key Point Matching (KPM): Given a debatable topic, a set of key points per stance, and a set of crowd arguments supporting or contesting the topic, report for each argument its match score for each of the key points under the same stance towards the topic.

Official Evaluation of KPM systems: mean Average Precision (mAP) --> for details refer to KPA-2021 Shared Task description

For an overview of the official Shared tasks's subsystem results refer to : Overview of the 2021 Key Point Analysis Shared Task

Languages

The BCP-47 code for Modern Greek is el-GR.

Dataset Structure

Data Instances

For each instance, there is a string for the argument_id, a string for the key point_id, a string for the argument, a string for the key point,an integer for the label, an integer for the stance. In the following two example instances, we see for the debatable topic 'We should subsidize vocational training' and the arguments supporting this topic (stance:1), one argument (arg_27_128) that matches the key point (kp_27_4) and one argument (arg_27_129) that does not.

Example_1 (matching argument-kp pair)
arg_id: arg_27_128
kp_id: kp_27_4,
label: 1
argument: "οι φτωχοί έχουν επίσης δικαίωμα στην εκπαίδευση"
key_point: "η επιδότηση της επαγγελματικής εκπαίδευσης προωθεί όσους έχουν λιγότερους πόρους"
topic: "Πρέπει να επιδοτήσουμε την επαγγελματική εκπαίδευση"
stance: 1

Example_2 (non-matching argument-kp pair)
arg_id:arg_27_129,
kp_id: kp_27_4,
label: 0,
argument: "Υπάρχει ένα τρέχον στίγμα στην κοινωνία που λέει ότι όλοι οι απόφοιτοι του γυμνασίου θα πρέπει να πάνε στο κολέγιο, και αυτή η επιδότηση θα βοηθήσει στην καταπολέμηση αυτού του στίγματος και θα οδηγήσει τους ανθρώπους προς πολύτιμους τομείς σταδιοδρομίας."
key_point: "η επιδότηση της επαγγελματικής εκπαίδευσης προωθεί όσους έχουν λιγότερους πόρους"
topic: "Πρέπει να επιδοτήσουμε την επαγγελματική εκπαίδευση"
stance: 1

The average token count for the arguments and the key points are provided below:

Feature	Mean Token Count
argument	20
key point	9

Data Fields

arg_id: the unique identification number of each argument
key_point_id: the unique identification number of each key point
label: binary (matching(1)/non-matching(0)) annotation of <argument, key point> pairs
arg: the argument supporting/contesting a debatabe topic
key_point: the key point reflecting (or not-depending on the 'label') the content of the argument as writen originally by an expert debater
topic: the debatable topic
stance: binary label (suporting(1)/contesting(-1)) the debatable topic

Data Splits

Split	Number of Examples
Train	20.635 pairs
Validation	3.458 pairs
Test	3.426 pairs

Use

from datasets import load_dataset
dataset = load_dataset("Kleo/ArgKP_2021_GR")
dataset

Dataset Creation

Curation Rationale

KPA systems are valuable tools for informed decision making, that should not be limited to English-speaking populations. We attempt to transfer the task of KPA in a low-resource language by providing a relevant dataset created through machine and human translation. Due to its relatively larger size, the train set has been zero-shot-translated with madlad400_3b_mt, a multilingual machine translation model, while the validation and tests set have been manually translated from the official KPA benchmark ArgKP-2021.

Source Data

Original Dataset : IBM/KPA_2021_shared_task

Data Collection and Processing

Arguments: As noted in the original dataset (Bar-Haim et al., ACL 2020), the arguments were collected from the publicly available IBM-Rank-30k dataset, with strict length limitations, accompanied by extensive quality control measures. Out of the 71 controversial topics in this dataset, a subset of 28 topics were selected, for which a corresponding motion exists in the Debatabase repository of the iDebate website. This requirement guaranteed that the selected topics were of high general interest. Fo mor details refer to the original ArgKP Dataset paper.
Key points: Were generated by an expert debater following specific guidelines (see ArgKP dataset paper for details)

Annotations

Annotation process

Label:The labelling of argument-key point pairs as matching/non-matching was performed through crowdsourcing, using the Figure Eight crowd labeling platform. For annotation details refer to the original paper
Stance:Refer to ibm/argument_quality_ranking_30k

Personal and Sensitive Information

The dataset contains arguments supporting or contesting debatable topics, that might be considered sensitive, such as Assisted suicide, Capital punishment, Atheism, cannabis legalization

Considerations for Using the Data

Social Impact of Dataset

The purpose of this dataset is to help develop models that can summarize a collection of opinionated texts (arguments, comments, reviews, survey responses) for or against a topic of interest into a set of representative key points, which are ranked based on their prevalence in the original argument data. Key Point Analysis introduces a quantitative aspect in the field of Text summarization and can be applied to various fields apart from debates, such as customer reviews, survey responses, twitter comments and many more.

Bias, Risks, and Limitations

The train set of this dataset is the result of machine translation.

Additional Information

Curated by: https://huggingface.co/Kleo
Language(s) (NLP): el
License: Apache license 2.0

Dataset Sources

Repository: https://github.com/Kleo-Karap/KPA_thesis/tree/main
Paper [optional]: [More Information Needed]

Citation Information

Bar-Haim, R., Lilach, H., Friedman, E., Kantor, Y., Lahav, D., & Slonim, N. (2020). From Arguments to Key Points: Towards Automatic Argument Summarization (pp. 4029–4039). Association for Computational Linguistics. https://aclanthology.org/2020.acl-main.371.pdf
Karapanagiotou, K. (2025). Key Point Analysis in Greek: A New Dataset and Baselines [MSc Thesis, National and Kapodistrian University of Athens]. Pergamos.https://pergamos.lib.uoa.gr/uoa/dl/frontend/el/browse/3456844

@inproceedings{bar-haim-etal-2020-arguments,
    title = "From Arguments to Key Points: {T}owards Automatic Argument Summarization",
    author = "Bar-Haim, Roy  and
      Eden, Lilach  and
      Friedman, Roni  and
      Kantor, Yoav  and
      Lahav, Dan  and
      Slonim, Noam",
    editor = "Jurafsky, Dan  and
      Chai, Joyce  and
      Schluter, Natalie  and
      Tetreault, Joel",
    booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2020.acl-main.371/",
    doi = "10.18653/v1/2020.acl-main.371",
    pages = "4029--4039",
    abstract = "Generating a concise summary from a large collection of arguments on a given topic is an intriguing yet understudied problem. We propose to represent such summaries as a small set of talking points, termed \textit{key points}, each scored according to its salience. We show, by analyzing a large dataset of crowd-contributed arguments, that a small number of key points per topic is typically sufficient for covering the vast majority of the arguments. Furthermore, we found that a domain expert can often predict these key points in advance. We study the task of argument-to-key point mapping, and introduce a novel large-scale dataset for this task. We report empirical results for an extensive set of experiments with this dataset, showing promising performance."
}

@masterthesis{3456844,
    title = "Key Point Analysis in Greek: A New Dataset and Baselines",
    authorField = "Καραπαναγιώτου, Κλεοπάτρα",
    year = "2025",
    school = "ΠΜΣ Γλωσσική Τεχνολογία, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών"
}

Dataset Card Contact

https://huggingface.co/Kleo