9 Sentence alignment

An alignment between two parallel corpora (e.g. a collection of source texts and their translations into some other language) can be encoded as a corpus attribute within CWB.

• Alignment attributes (a-attributes) are unlike other types of attribute because alignment presupposes the existence of the source and target corpora. That is, first we need to encode the two corpora independently; then we can add the alignment attribute that links them.
• Alignment attributes are usually employed for sentence alignment, and we will assume throughout that it is sentences that we are aligning.
• However, you can also align at some other level (e.g. clauses or paragraphs or chapters). Aligning regions that are much smaller than a sentence will not be very useful because of the limitations of how CQP deals with a-attributes.
• Only one a-attribute linking any particular pair of corpora can exist.
• There are two ways that a pair of corpora can be aligned.
  • First, the cwb-align tool can be used to automatically align the sentences of the two corpora, with its output subsequently encoded as an a-attribute using cwb-align-encode.
  • Second, an existing alignment scheme encoded in the corpus markup can be imported as an a-attribute using cwb-align-import.

CWB supports many types of alignment link: one to one, many to many, and crossing. However, the regions in the corpora that are the units to be aligned with one another cannot be discontinuous.