Webpage

What is the webpage for the dataset (if it exists)?

Website

Download

What is the link to where the original dataset is hosted?

Github

Paper

What is the link to the paper describing the dataset (open access preferred)?

ACL anthology

@inproceedings{kanerva2019newsgen,
Title = {Template-free Data-to-Text Generation of Finnish Sports News},
Author = {Jenna Kanerva and Samuel R{\"o}nnqvist and Riina Kekki and Tapio Salakoski and Filip Ginter},
booktitle = {Proceedings of the 22nd Nordic Conference on Computational Linguistics (NoDaLiDa’19)},
year={2019}
}

Has a Leaderboard?

Does the dataset have an active leaderboard?

no

Multilingual?

Is the dataset multilingual?

no

Covered Languages

What languages/dialects are covered in the dataset?

Finnish

License

What is the license of the dataset?

cc-by-nc-sa-4.0: Creative Commons Attribution Non Commercial Share Alike 4.0 International

Primary Task

What primary task does the dataset support?

Data-to-Text

Curation Organization Type(s)

In what kind of organization did the dataset curation happen?

academic

Data Fields

List and describe the fields present in the dataset.

The dataset is constructed of games, where each game is a list of events. If the event was annotated (corresponding sentence was found from the news article), it includes text field with value other than empty string ("").

For each game (dict), there are keys gem_id (string), id (string), news_article (string), and events (list).

For each event (dict), there are different, relevant keys available with non empty values depending on the event type (e.g. goal or penalty). The mandatory keys for each event are event_id (string), event_type (string), text (string, empty string if not annotated), and multi_reference (bool). The keys not relevant for the specific event type are left empty.

The relevant keys in the event dictionary are:

For each event type, the following keys are relevant: event_id: Identifier of the event, unique to the game but not globally, in chronological order (string) event_type: Type of the event, possible values are game result, goal, penalty, or saves (string) text: Natural language description of the event, or empty string if not available (string) multi_reference: Does this event refer to a text passage describing multiple events? (bool)

The rest of the fields are specific to the event type. The relevant fields for each event type are:

game result: event_id: Identifier of the event, unique to the game but not globally, in chronological order (string) event_type: Type of the event (string) home_team: Name of the home team (string) guest_team: Name of the guest team (string) score: Final score of the game, in the form of home–guest (string) periods: Scores for individual periods, each in the form of home–guest score in that period (list of strings) features: Additional features, such as overtime win or shoot out (list of strings) text: Natural language description of the event, or empty string if not available (string) multi_reference: Does this event refer to a text passage describing multiple events? (bool)

goal: event_id: Identifier of the event, unique to the game but not globally, in chronological order (string) event_type: Type of the event (string) player: Name of the player scoring (string) assist: Names of the players assisting, at most two players (list of strings) team: Team scoring with possible values of home or guest (string) team_name: Name of the team scoring (string) score: Score after the goal, in the form of home–guest (string) time: Time of the goal, minutes and seconds from the beginning (string) features: Additional features, such as power play or short-handed goal (list of strings) text: Natural language description of the event, or empty string if not available (string) multi_reference: Does this event refer to a text passage describing multiple events? (bool)

penalty: event_id: Identifier of the event, unique to the game but not globally, in chronological order (string) event_type: Type of the event (string) player: Name of the player getting the penalty (string) team: Team getting the penalty with possible values of home or guest (string) team_name: Name of the team getting the penalty (string) penalty_minutes: Penalty minutes (string) time: Time of the penalty, minutes and seconds from the beginning (string) text: Natural language description of the event, or empty string if not available (string) multi_reference: Does this event refer to a text passage describing multiple events? (bool)

saves: event_id: Identifier of the event, unique to the game but not globally, in chronological order (string) event_type: Type of the event (string) player: Name of the goalkeeper (string) team: Team of the goalkeeper with possible values of home or guest (string) team_name: Name of the team (string) saves: Number of saves in the game (string) text: Natural language description of the event, or empty string if not available (string) multi_reference: Does this event refer to a text passage describing multiple events? (bool)

Text passages describing multiple events (multi_reference):

Some text passages refer to multiple events in such way that separating them to individual statements is not adequate (e.g. "The home team received two penalties towards the end of the first period."). In these cases, multiple events are aligned to the same text passage so that the first event (in chronological order) include the annotated text passage, while the rest of the events referring to the same text passage include the identifier of the first event in the annotated text field (e.g. text: "E4").

{
'gem_id': 'gem-turku_hockey_data2text-train-0',
'id': '20061031-TPS-HPK',
'news_article': 'HPK:n hyvä syysvire jatkuu jääkiekon SM-liigassa. Tiistaina HPK kukisti mainiolla liikkeellä ja tehokkaalla ylivoimapelillä TPS:n vieraissa 1–0 (1–0, 0–0, 0–0).\nHPK hyödynsi ylivoimaa mennen jo ensimmäisessä erässä Mikko Mäenpään maalilla 1–0 -johtoon.\nToisessa ja kolmannessa erässä HPK tarjosi edelleen TPS:lle runsaasti tilanteita, mutta maalia eivät turkulaiset millään ilveellä saaneet. Pahin este oli loistavan pelin Hämeenlinnan maalilla pelannut Mika Oksa.\nTPS:n maalissa Jani Hurme ei osumille mitään mahtanut. Joukkueen suuri yksinäinen kenttäpelaaja oli Kai Nurminen, mutta hänelläkään ei ollut onnea maalitilanteissa.',
'events':
{
'event_id': ['E1', 'E2', 'E3'],
'event_type': ['game result', 'penalty', 'goal'],
'text': ['HPK kukisti TPS:n vieraissa 1–0 (1–0, 0–0, 0–0).', '', 'HPK hyödynsi ylivoimaa mennen jo ensimmäisessä erässä Mikko Mäenpään maalilla 1–0 -johtoon.'],
'home_team': ['TPS', '', ''],
'guest_team': ['HPK', '', ''],
'score': ['0–1', '', '0–1'],
'periods': [['0–1', '0–0', '0–0'], [], []],
'features': [[], [], ['power play']],
'player': ['', 'Fredrik Svensson', 'Mikko Mäenpää'],
'assist': [[], [], ['Jani Keinänen', 'Toni Mäkiaho']],
'team': ['', 'guest', 'guest'],
'team_name': ['', 'HPK', 'HPK'],
'time': ['', '9.28', '14.57'],
'penalty_minutes': ['', '2', ''],
'saves': ['', '', ''],
'multi_reference': [false, false, false]
}
}

Similar Datasets

Do other datasets for the high level task exist?

yes

Modificatied for GEM?

Has the GEM version of the dataset been modified in any way (data, processing, splits) from the original curated data?

yes

Additional Splits?

Does GEM provide additional splits to the dataset?

no

Previous results available?

Are previous results available?

yes

Original Curation Rationale

Original curation rationale

The dataset is designed for text generation (data2text), where the original source of natural language descriptions is news articles written by journalists. While the link between structural data (ice hockey game statistics) and the news articles describing the game was quite weak (news articles including a lot of information not derivable from the statistics, while leaving many events unmentioned), the corpus includes full manual annotation aligning the events extracted from game statistics and the corresponding natural language passages extracted from the news articles.

Each event is manually aligned into a sentence-like passage, and in case a suitable passage was not found, the annotation is left empty (with value None). The extracted passages were manually modified not to include additional information not derivable from the game statistics, or not considered as world knowledge. The manual curation of passages is designed to prevent model hallucination, i.e. model learning to generate facts not derivable from the input data.

Sourced from Different Sources

Is the dataset aggregated from different data sources?

no

How was Language Data Obtained?

How was the language data obtained?

Other

Data Validation

Was the text validated by a different worker or a data curator?

not validated

Was Data Filtered?

Were text instances selected or filtered?

algorithmically

Additional Annotations?

Does the dataset have additional annotations for each instance?

expert created

Number of Raters

What is the number of raters

1

Annotation Service?

Was an annotation service used?

no

Any Quality Control?

Quality control measures?

validated by data curators

Any Consent Policy?

Was there a consent policy involved when gathering the data?

yes

Contains PII?

Does the source language data likely contain Personal Identifying Information about the data creators or subjects?

yes/very likely

Any Maintenance Plan?

Does the original dataset have a maintenance plan?

no

Usage of Models based on the Data

Are you aware of cases where models trained on the task featured in this dataset ore related tasks have been used in automated systems?

no

Addresses needs of underserved Communities?

Does this dataset address the needs of communities that are traditionally underserved in language technology, and particularly language generation technology? Communities may be underserved for exemple because their language, language variety, or social or geographical context is underepresented in NLP and NLG resources (datasets and models).

no

Any Documented Social Biases?

Are there documented social biases in the dataset? Biases in this context are variations in the ways members of different social categories are represented that can have harmful downstream consequences for members of the more disadvantaged group.

no