This is our abstractive and extractive summary corpus for personal narrative.
The directory includes narratives, abstractive summaries, extractive summaries, and rewrite-labeled alignments.

The data is structured as follows:

-- ‘narrative’ -- 

The original text of the narratives being summarized.
These are taken from Ouyang and McKeown's Reddit personal narrative
data set (www.cs.columbia.edu/~ouyangj/reddit-data).  Each file is
named using Ouyang and McKeown's narrative IDs. There are 476 narratives 
in total.

Within each annotator’s directory:

-- ‘abstractive’ -- 

The abstractive summaries written by that annotator. The
files are named in the format '<narrative id>.abstractive’.

-- ‘extractive’ --

The extractive summaries corresponding to the abstractive summaries written 
by the annotator.  The files are named in the format '<narrative id>.<type>', 
where type is one of the following:

'1', '2', or '3' -- The extractive summary constructed by one of the
three Turkers.  Note that the three Turkers are not the same from
summary to summary, so it is NOT the case that all extractive
summaries with the same suffix were constructed by the same person.

'union' -- The extractive summary produced by union aggregation of the
three Turker summaries.  This aggregation scheme selects all sentences
chosen by at least one Turker.

'majority' -- The extractive summary produced by majority aggregation of the
three Turker summaries.  This aggregation scheme selects all sentences
chosen by at least two Turkers.

'intersect' -- The extractive summary produced by intersect aggregation of the
three Turker summaries.  This aggregation scheme selects all sentences
chosen by all three Turkers.

-- ‘alignments’ --

The phrase alignments between each abstractive summary written by the annotator
and its corresponding extractive summary (the union aggregation). The files are
named in the format ‘<narrative id>.alignments’.

Each file contains all of the alignments made for that abstractive/extractive pair.

For each alignment, the whole abstractive and extractive summaries are displayed,
with alignments indicated within double brackets: [[ . . . ]]. 

Below that, we include our evaluation of the alignment (precision, confidence), 
and then we indicate whether the rewrite technique of Fusion was used to transform 
the extractive phrase into the abstractive phrase. Under Fusion, we list the 
results from our Phrase Transformation HIT. For each rewrite technique, we show 
the number of Turkers (out of three) that believed the technique in question 
was employed for this alignment, and the number that did not.