Making the CHAVE collection available

CHAVE em português

Since April 2007 a version syntactically anotated by PALAVRAS (Bick, 2000) has been made available.

Since January 2010 a named-entity annotated version of CHAVE, by REMBRANDT 0.7 (Cardoso, 2008) is also available.

In addition to a large set of documents, namely the 1994 and 1995 full editions of the Público and Folha de São Paulo newspapers, we make available the following resources, related to two diferent tracks:

• Information Retrieval (IR) Ad Hoc Track (2004, 2005, 2006, Robust: 2007)
  • a list of topics in Portuguese, compiled cooperatively with the other CLEF organizers
  • a pool of (binarily) judged documents for each topic
• Question Answering Evaluation QA@CLEF (2004, 2005, 2006, 2007, 2008)
  • a list of questions and answers in Portuguese, compiled cooperatively with the other QA@CLEF organizers (data for 2004 can be directly download from the organization (here)
  • a (non exhaustive) set of document ids that support the answer(s) for a subset of the above
• Geographic Information Retrieval GeoCLEF (2006, 2007, 2008)
  • a list of topics in Portuguese, compiled cooperatively with the other CLEF organizers
  • a pool of (binarily) judged documents for each topic

• Textos - Folder containing the complete texts of the newspapers PÚBLICO and Folha de São Paulo of 1994 and 1995. From April 2007, there is also available a version syntactically anotated by PALAVRAS (Bick, 2000). The files can be identified by the cg. prefix.
• 2004 - The Portuguese resources concerning CLEF2004
• 2005 - The Portuguese resources concerning CLEF2005
• 2006 - The Portuguese resources concerning CLEF2006
• 2007 - The Portuguese resources concerning CLEF2007
• 2008 - The Portuguese resources concerning CLEF2008
• 200x/Monte - Document pools for each topic relative to CLEF200x
• 200x/PerguntasRespostas - Questions and answers compiled by the organization of CLEF200x
• 200x/Topicos - Topics in Portuguese, compiled by the organization of CLEF200x

Information about the texts

For the CLEF 2004 edition, only PÚBLICO 1995 was used for adhoc IR, and PÚBLICO 1994 and 1995 for QA@CLEF. For CLEF 2005 and 2006 both newspapers and years were used for QA@CLEF and adhoc IR. In 2006 they were also used for GeoCLEF. In 2007 they continue to be used in QA@ and GeoCLEF, as well as for the robust track.

The following table presents a quantitative description of the collections:

Colections		Público	Folha de São Paulo
Years		1994-1995	1994-1995
Editions		726	730
Documents		106,821	103,913
Size		348,078 kB	226,690 kB
Units	Total	64,222,797	42,109,286
	Different	500,197	426,469
Words	Total	54,947,072	35,699,765
	Different	472,817	393,885

Notes:

• Público is not published on Christmas day or New Year's day, so there are 4 editions less.
• A word was defined as a letter followed by a (possibly void) sequence of letters and hyphens.
• Markup was not counted.

Here is a sample document from PÚBLICO (SGML 351KB, gz 135KB, with the corresponding DTD) and from Folha de São Paulo (SGML, gz and corresponding DTD).

Conditions for use

1. Register in order to get the collection
2. Reference the following facts: that the collection consists of the 1994 and 1995 complete editions of Público newspaper (www.publico.pt) and Folha de São Paulo (www.folha.com.br), that it was compiled by Linguateca (www.linguateca.pt), and that this compilation occurred in the framework of CLEF (www.clef-campaign.org)
3. Use it for research and development only; not for reselling or making profit from its direct distribution (on-line or off-line)
4. No results obtained outside the CLEF official campaigns can invoke CLEF's name in a way that implies that the system was assessed within CLEF, i.e., it is not acceptable to compare with results out of contest without clearly stating it. Ideally, one should simply refer to the CHAVE collection.
5. Mention that the syntactic annotation was made by PALAVRAS (Bick, 2000) and the NE-annotation by REMBRANDT (Cardoso, 2008).