As mentioned earlier, the complexity of the alignment task is very dependent on the type of text. Using exclusively Hansard documents would probably have greatly simplified the task of assembling the BAF, not only because these texts are typically fairly easy to align, but also because they are widely available in plain-text format. However, we felt that concentrating one a single genre would make the corpus less useful as an evaluation resource. So, when selecting the pairs of documents that make up the BAF, we tried to include documents from various sources and of various genres.
Of course, the corpus is, to some extent, representative of the types of texts that are available in multilingual versions. For instance, it does not contain such things as newspaper articles or e-mail messages simply because such texts are usually not translated. On the other hand, documents produced by international organizations or governments of countries with two or more official languages are usually easy to find in multilingual versions, and so the BAF contains a lot of these. This is also true, although maybe to a lesser extent, of some technical documents, such as user's manuals, and of literary text. In general, scientific articles are not routinely translated, but as part of a Canadian government-owned research institution, we found that we had access to a number of bilingual technical reports. This explains their presence in the BAF.
The documents that currently make up the BAF corpus are presented in Table 1.
Genre
Reference
Source
Institutional
Hansard - Canadian Parliamentary Proceedings. March 14, 1994
House of Commons publication service.
Supreme court of Canada (1995). Terrence Wayne Burlingham v.
Her Majesty the Queen
Centre de recherche en droit
public of the Law Faculty of the University of Montreal.
UN International Labor Organization (1985). 241st and 242nd Reports
of the Committee on Freedom of Association
ECI Multilingual Corpus.
UN (1993). Report of the Secretary-General on the Work of the Organization
UN translation services.
Scientific
Geoffroy, Catherine (1994). Les technologies de communication de
l'information et les aîné(e)s. CITI technical report
All these documents were available as technical reports at the
CITI.
Lapointe, François (1995). Changement technologique et
organisation du travail. CITI technical report
Macklovitch, Elliott (1995), Peut-on vérifier automatiquement
la cohérence terminologique? in Actes des IVes journées
scientifiques, Lexicommatique et Dictionnairiques, Lyon, France
Simard, Michel (1995), Réaccentuation automatique de textes
français, CITI technical report
Technical
Xerox Corporation, ScanWorX User's Guide
ECI Multilingual Corpus.
Note : these documents contain in appendix a relatively large
glossary of terms, which could not be aligned, because the
order of the entries is completely different in French and English.
Literary
Verne, Jules. De la terre à la lune
The original French version was obtained from the WWW site of the
Association des Bibliophiles Universels. The English translation comes
from the Project Gutenberg.