TRIAL CORPORA FOR DIACHRONIC EVALUATION TASK SEMEVAL 2015
=========================================================
(Octavian Popescu May 2014)


The technical details:
~~~~~~~~~~~~~~~~~~~~~~
The resource was obtained by inspecting large archives of newspapers.

Requirements/prerequisites:
~~~~~~~~~~~~~~~~~~~~~~~~~~~
There are no specific requirements. The packages contains text files with no special encoding.

Install:
~~~~~~~~
unzip trial.zip

The directory trial contains three files, trialT1 , trialT2 and trialT3, each one containing the examples for task 1, 2 and 3 respectively (see webpage http://alt.qcri.org/semeval2015/task7/)

The following train and test corpora will contain exactly the same type of files.

Specifications:

The files trialT1 and trialT2 are organised in simple units, delimited by the "<text ID="...">" and "</text>" tags. Each unit contains three types of time interval and a piece of news.
 
Each piece of news is introduced by a line containing the ID (unique for each piece of text) followed by three lines as exemplified below:

<text id="607bc11532116102">
<textF 
<textM
<textC

Each of the last three lines continues with a series of time interval like:
. . . no="1734-1736" no="1737-1739" yes="1740-1742" no="1743-1745" no="1746-1748" . . .
covering the period from 1700 to present divided into intervals with a certain time span:

<textF  contains fine graded intervals: for task1 the time span is 2 years, and for task2 the time span is 6 years

<textM contains medium graded interval: for task1 the time span is 6 years, and for task2 the time span is 12 years

<textC contains coarse graded interval: for task1 the time span is 12 years, and for task2 the time span is 20 years

On each of the lines above, all the time intervals are marked as "no", except for ONLY one time interval which is marked as "yes". The interval of time marked "yes" is the time interval within which the piece was news was created.

The above lines are followed on the next line by the piece of news, which is followed on the line by "</text>" tag which signals the end of one text unit. 

In the test corpora the "<textF", "<testM", "<textC" lines will have all time intervals marked with "no". The systems have to choose ONE and ONLY ONE interval that will be annotated with "yes", overwriting the initial "no" mark. Example:

system input (test corpora task 1 or 2):
. . . no="1734-1736" no="1737-1739" no="1740-1742" no="1743-1745" no="1746-1748". . .
system output: 
. . . no="1734-1736" no="1737-1739" yes="1740-1742" no="1743-1745" no="1746-1748" . . .

There is no fine, medium or coarse annotation for task3. There is just one time period specified, and certain phrases from the piece of news are selected and marked by "yes" or "no" according to their time relevance for the specified time period . There may be more than one "yes" marked phrase. The marks "yes", "no" are indicated in the trial and train corpora. For test corpora, all phrases will be mark "no" and the system has to change the marks to "yes" for those considered related to the specified period:

system input (test corpus task 3):
<text no="French Interest in Hague" no="Republick" no="which were accused" period="1700-1800">
system output:
<text yes="French Interest in Hague" yes="Republick" no="which were accused" period="1700-1800">

Please address your questions to the task organisers, see webpage http://alt.qcri.org/semeval2015/task7/index.php?id=contact or subscribe to our discussion group (details on the same web page).