BibTex

Provide the BibTex-formatted reference for the dataset. Please use the correct published version (ACL anthology, etc.) instead of google scholar created Bibtex.

@inproceedings{puduppully-etal-2019-data,
title = "Data-to-text Generation with Entity Modeling",
author = "Puduppully, Ratish  and
Dong, Li  and
Lapata, Mirella",
booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
month = jul,
year = "2019",
address = "Florence, Italy",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/P19-1195",
doi = "10.18653/v1/P19-1195",
pages = "2023--2035",
}

Intended Use

What is the intended use of the dataset?

The dataset can be used to study data-to-text generation. The dataset is in sports domain. It pairs statistics of Major League Baseball (MLB) game with its summary. The summary is in the form of a document containing an average of 540 tokens. Thus it is useful to study long document generation.

Dataset Creators

Who created the original dataset? List the people involved in collecting the dataset and their affiliation(s).

Ratish Puduppully, Li Dong, Mirella Lapata

Reason for Structure

How was the dataset structure determined?

The high level structure contains the following attributes: home_name, vis_name, home_city, vis_city, summary, summary_eval, day, gem_id, box_score, play_by_play, home_line, vis_line. The attributes home_name, vis_name, home_city, vis_city and day are string values. The attribute "summary" contains the summary in the form of a list of tokens. The attribute "summary_eval" contains the summary in the form of a string of tokens. The difference from "summary" field is that "summary_eval" doesn't contain "NEWPARAGRAPH" delimiters to separate the paragraphs. "summary_eval" field should be used to evaluate model outputs. "summary" field may be used during the training process. box_score contains the box score statistics of the players in the game. It is in the form of a list (of average size 90), with each element describing the statistics of a player. The box score statistics contain 53 attributes. The description of the attributes is given below. The descriptions of most of the attributes is obtained from mlb.com.

• r : Runs scored by a player in the game.
• rbi Runs Batted In (RBI): action of a batter results in a run scored by other players in the team.
• pos Position of the player.
• avg Batting Average. It is an indicator of the hits in the players' career.
• bb A walk occurs when a pitcher throws four pitches out of the strike zone, none of which are swung at by the hitter.
• hr Batter hits the ball in the air over the outfield fence.
• p_r Runs given by a pitcher in the game.
• p_bb Walks allowed by pitcher in a game.
• p_h Hits allowed by pitcher in a game.
• p_hr Home runs allowed by pitcher in a game.
• p_er Earned Run (ER): An earned run is any run that scores against a pitcher.
• p_era Earned Run Average (ERA): Earned run average represents the number of earned runs a pitcher allows per nine innings.
• p_np Number of Pitches: A pitcher's total number of pitches is determined by all the pitches he throws in game.
• p_ip1 Innings Pitched (IP1): Innings pitched measures the number of innings a pitcher remains in a game. Because there are three outs in an inning, each out recorded represents one-third of an inning pitched.
• p_ip2 Innings Pitched (IP2): Innings pitched measures the number of innings a pitcher remains in a game. Because there are three outs in an inning, each out recorded represents one-third of an inning pitched.
• p_w A pitcher receives a win when he is the pitcher of record when his team takes the lead for good.
• p_l A pitcher receives a loss when a run that is charged to him proves to be the go-ahead run in the game, giving the opposing team a lead it never gives up.
• p_so A strikeout occurs when a pitcher throws any combination of three swinging or looking strikes to a hitter.
• p_save Save: A save is awarded to the relief pitcher who finishes a game for the winning team. A pitcher cannot receive a save and a win in the same game.
• p_sv Saves: The count of saves recorded by a pitcher in his career.
• sac A sacrifice fly occurs when a batter hits a fly-ball out to the outfield or foul territory that allows a runner to score.
• p_bf Batters faced is simply a count of the number of total plate appearances against a certain pitcher or team. In a perfect game -- with 27 outs -- a pitcher will record 27 batters faced.
• cs A caught stealing occurs when a runner attempts to steal but is tagged out before reaching second base, third base or home plate.
• hbp A hit-by-pitch occurs when a batter is struck by a pitched ball without swinging at it. He is awarded first base as a result.
• ab An official at-bat comes when a batter reaches base via a fielder's choice, hit or an error (not including catcher's interference) or when a batter is put out on a non-sacrifice.
• p_bs A blown save occurs when a relief pitcher enters a game in a save situation, but allows the tying run to score.
• p_s The count of strikes thrown by a pitcher
• lob Left on base can be viewed as both an individual statistic or as a team statistic. In an individual batter's case, it refers to how many men remain on base after that batter makes an out at the plate, as the batter has failed to do his job to score those runners -- or at least put himself in a position to score. In a team's case or in an individual pitcher's case, it refers to the number of men who remain on base at the end of an inning.
• po A fielder is credited with a putout when he is the fielder who physically records the act of completing an out -- whether it be by stepping on the base for a forceout, tagging a runner, catching a batted ball, or catching a third strike
• ops OPS adds on-base percentage and slugging percentage to get one number that unites the two. It's meant to combine how well a hitter can reach base, with how well he can hit for average and for power.
• p_hld A hold occurs when a relief pitcher enters the game in a save situation and maintains his team's lead for the next relief pitcher, while recording at least one out.
• p_loss True/False- Indicates losing pitcher
• e A fielder is given an error if, in the judgment of the official scorer, he fails to convert an out on a play that an average fielder should have made.
• p_win True/False- Indicates winning pitcher
• a An assist is awarded to a fielder who touches the ball before a putout is recorded by another fielder.
• h A hit occurs when a batter strikes the baseball into fair territory and reaches base without doing so via an error or a fielder's choice.
• so A strikeout of a batter
• team Team of the player
• sb A stolen base occurs when a baserunner advances by taking a base to which he isn't entitled.
• slg Slugging percentage represents the total number of bases a player records per at-bat. Unlike on-base percentage, slugging percentage deals only with hits and does not include walks and hit-by-pitches in its equation.
• sf A sacrifice fly occurs when a batter hits a fly-ball out to the outfield or foul territory that allows a runner to score.
• obp OBP refers to how frequently a batter reaches base per plate appearance. Times on base include hits, walks and hit-by-pitches, but do not include errors, times reached on a fielder's choice or a dropped third strike.

The description of attributes in play-by-play is below:

• batter Batter in the play.
• pitcher Pitcher in play.
• b1 Player/s at first base position.
• b2 Player/s at second base position.
• b3 Player/s at third base position.
• scorers Player/s scored in the play.
• fielder_error Player committed field error.
• event Event of the play such as single, double, home run etc.
• event2 Second event of the play such as wild pitch, error etc.
• inning Inning of the play.
• top/ bottom If home team is batting it is bottom and if away team is batting it is top.
• o Count of outs
• b Count of balls
• s Count of strikes
• r Count of runs
• rbi Count of runs batted in (rbi)
• error_runs Runs due to error
• home_team_runs Score of home team
• vis_team_runs Score of visiting team

home_line and vis_line contain string value pairs for team_name, team_city, team_runs, team_hits, team_error, result, and a list of runs scored in each inning.

Splitting Criteria

Describe any criteria for splitting the data, if used. If there are differences between the splits (e.g., if the training annotations are machine-generated and the dev and test ones are created by humans, or if different numbers of annotators contributed to each example), describe them here.

The splits are random.

Why is the Dataset in GEM?

What does this dataset contribute toward better generation evaluation and why is it part of GEM?

This dataset can verify if models are capable of long document generation. The challenges in long document generation conditioned on input tables include ensuring coherent output, staying faithful to the input, ensuring fluent output and avoiding repetition of text. Such aspects can be verified on models trained on this dataset

Difference from other GEM datasets

What else sets this dataset apart from other similar datasets in GEM?

Compared to the existing RotoWire (Wiseman et al. 2017) dataset, MLB summaries are longer (approximately by 50%) and the input records are richer and more structured (with the addition of play-by-play). Moreover, the MLB dataset is five times larger in terms of data size (i.e., pairs of tables and game summaries).

Modification Details

For each of these changes, described them in more details and provided the intended purpose of the modification

Some examples have been removed from training dataset which satisfied the below criteria:

1. The examples in training dataset which overlapped with validation/test.
2. Some examples which described washed out games.

Pointers to Resources

Getting started with in-depth research on the task. Add relevant pointers to resources that researchers can consult when they want to get started digging deeper into the task.

The research paper is a good resource

Proposed Evaluation

List and describe the purpose of the metrics and evaluation methodology (including human evaluation) that the dataset creators used when introducing this task.

We have reused the automatic metrics based on Information Extraction evaluation introduced by Wiseman et al (2017). For human evaluation, we conducted studies to evaluate the factuality, coherence, grammaticality and conciseness.

Relevant Previous Results

What are the most relevant previous results for this task/dataset?

The most relevant previous results for dataset are in the TACL 2021 paper on Data-to-text Generation with Macro Planning

Language Producers

What further information do we have on the language producers?

The game summaries are produced by professional writers.

Data Preprocessing

How was the text data pre-processed? (Enter N/A if the text was not pre-processed)

Game summaries were tokenized using NLTK (Bird et al., 2009) and hyphenated words were separated. Sentences containing quotes were removed as they included opinions and non-factual statements unrelated to the input tables. Sometimes MLB summaries contain a "Game notes" section with incidental information which was also removed.

Justification for Using the Data

If not, what is the justification for reusing the data?

The copyright remains with the original data creators and the usage permission is restricted to non-commercial uses.