I started implementing this now. The idea is to have a flexible and powerful method of
ingesting tabular stand-off annotation data in STAM and, if needed,
either automatically align this with a text file (i.e. compute offsets if not
explicitly provided), or even reconstruct the text file from zero.

Users should be able to provide simple TSV data like:

Text	pos
Hello	interjection
world	noun

Here Text is a recognized column and pos is not so it translates to an
AnnotationSet (undefined here) and DataKey (pos). When loaded against an existing
resource file (like below), the offsets are computed automatically

Hello world!

Alternatively, this text (without the exclamation mark) can be reconstructed on
the basis of the input data (with space as an output delimiter). Note that text
input doesn't need to be constrained to words/tokens. Reconstruction and
alignment both assumes the input rows are sequential. If rows are explicitly
marked as not sequential (via some parameter), we can fall back on a tagging
mechanism to simple tag all found matches (e.g. with natural word boundaries).

The above illustrates the more complex case I want to support where input data
is incomplete, when more predefined columns are used parsing can be much
simpler and no alignment or reconstruction is needed in the first place:

Text	BeginOffset	EndOffset	pos
Hello	0	5	interjection
world	6	10	noun

TSV to STAM conversion #1

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions