The dataset released with the submission of INLG 2016 paper "Crowd-sourcing NLG Data: Pictures Elicit Better Data" (https://aclweb.org/anthology/W/W16/W16-6644.pdf)
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
INLG2016_dataset.csv
README.md
images.zip

README.md

INLG_16_submission

This repository contains the dataset released with the submission of INLG 2016 paper "Crowd-sourcing NLG Data: Pictures Elicit Better Data" (https://aclweb.org/anthology/W/W16/W16-6644.pdf).

File descriptions:

  • INLG2016_dataset.csv - the dataset with textual meaning representations
  • images.zip - images used as pictorial meaning representations

Data fields:

mr - textual meaning representation

nl.utterance - natural language utterance

informativeness - human rating of informativeness, collected on 1-6 Likert scale (crowd evaluation)

naturalness - human rating of naturalness, collected on 1-6 Likert scale (crowd evaluation)

phrasing- human rating of phrasing, collected on 1-6 Likert scale (crowd evaluation)

attr.count - number of attributes in a corresponding MR

batch - pic: pictorial MR presented to crowd workers, txt: textual MR presented to crowd workers

mr.id - id number of MR (used as the name of image for pictorial MR)

min.length - minimal length metric, used for automatic pre-validation (see the paper)

utterance.length - length of the utterance, in characters

no.Of.Sentences - number of sentences in the utterance

len.Ratio - utterance.length/min.length

coll.channel - the work channel that the crowd worker accessed the job through

coll.country - the country the crowd worker is from

coll.region - a region code for the area the crowd worker is from

col.city - the city the crowd worker is from

col.inform - self-evaluation on informativeness

col.natur - self-evaluation on naturalness

col.phrasing - self-evaluation on phrasing

eval.inform - crowd-evaluation on informativeness

eval.natur - crowd-evaluation on naturalness

eval.phrasing - crowd-evaluation on phrasing

sem.coverage - semantic similarity score (see the paper)

norm.sem.coverage - normalised semantic similarity score

coll.task.duration - average duration of the task, sec

Citing the dataset:

If you use or refer to the dataset, please cite this paper:

Jekaterina Novikova, Oliver Lemon, Verena Rieser (2016). Crowd-Sourcing NLG Data: Pictures Elicit Better Data, in Proceedings of the 9th International Natural Language Generation Conference INLG'16. Edinburgh, UK