Evita 1.2 — Release Notes

Roser Saurí and Marc Verhagen.
Brandeis University, July 2006.

Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 2.5 License.

Installation

Evita 1.2 requires at least Python version 2.2.3 or higher, version 2.4 is recommended. Python is available at http://python.org/. Evita 1.2 has been succesfully tested on Linux with Python 2.2.3, Linux with Python 2.4.2, Windows 2000 with Python 2.4.1, Windows XP with Python 2.4.2, and Mac OS 10.4.2 with Python 2.3.5.

To install, first download Evita 1.2 from timeml.org. Then use gunzip and tar to unpack the archive:

% gunzip -c evita-1.2.tar.gz | tar xvp

This creates a directory named evita-1.2 in the current directory. On Windows, use WinZip or something similar to unpack the archive.

Running Evita

Input is a TimeML file with chunk information. Evita assumes that noun chunks are tagged with <NG> and verb chunks with <VG> or any tag containing VG, such as <VG_INF>. In addition, Evita expects the following tags and attributes:

  1. Sentences are marked with <s> tags.
  2. Possessive phrases are marked with a <POS> tag.
  3. Tokens are marked with <lex> tags, with "pos" as the attribute that contains a part-of-speech conform to the Penn Treebank tagset.
  4. Chunk heads are marked with <HEAD> tags.

Evita will not produce any sensible output without the <s>, <NG>, <VG> and <lex> tags. Absence of <HEAD> and <POS> tags will affect performance. Example input is given in the ExampleInput directory of the distribution.

To run Evita, you need to change the current directory to evita-1.2/Code.

To run Evita on Windows, first open a command prompt, change the current directory to the evita-1.2/Code directory and type the commands as printed above.

Please email us at evita@timeml.org if you have any problems installing or running Evita. Bug reports and other comments can also be sent to that address.

Optimizing Evita

In the nominal disambiguation phase, Evita uses binary search on text files. Equivalent dbm dictionaries can be created by setting the current directory to evita-1.2/Code and running the buildEventNominals2.py script:

% cd evita-1.2/Code
% python buildEventNominals2.py

This step is not necessary to run Evita 1.2 but it may increase performance.

Additional Documentation

Roser Saurí, Robert Knippen, Marc Verhagen and James Pustejovsky (2005) Evita: A Robust Event Recognizer for QA Systems. In Proceedings of the HLT05. Vancouver, Canada, October 6-8, 2005.

Release Schedule

Bugs fixes will be released reqularly as minor versions with version number 1.x. Version 2.0 of Evita is scheduled to arrive in December 2006 as part of the TARSQI Toolkit.

Revision History

Evita 1.2, July 2006

Evita 1.1, June 2006

Evita 1.0, October 2005