thvitt
commented
6 years ago There are two general issues at hand here which we should discuss
-
Adjusting the graph to introduce relationships before we perform conflict resolution and sorting. This could, e.g., mean: a) introduce a directed, non-ignored edge from each inscription to its corresponding witness, indicating that all inscriptions happen before the witness is dated b) copying, or c) moving all outgoing edges from the inscription to the witness (if we don’t do a, then copy also all ingoing edges) Of course this might introduce inappropriate or conflicting edges depending on
-
Improving only the additional date span inference, which, of course, only works after all cycles have been removed. E.g., we could do the following:
- calculate the explicit and inferred absolute datings for each reference (already happens)
- for each witness that is linked to at least one inscription and copy the earliest start and the latest end date of the inscriptions to the witness, unless the witness already has an explicit absolute dating for this.
Of course, this step mostly only makes sense when we don’t implement 1b or 1c.
-
if we do so much inference on the inferred time span (i.e. 2.), we should think about how we could integrate the span as secondary or tertiary criterion into the ordering
gerritbruening
Example: http://dev.faustedition.net/macrogenesis/2_IV_H.22
It has turned out that mere links are not enough to avoid confusion. Datings of inscriptions are, as it were, sub nodes of witnesses. Is it possible to aggregate those inscriptions to somehow obtain directly visible absolute datings (it doesn't make sense to aggregate mere relative chronological order of inscriptions, of course). It is clear that we will end up with two or more intervals in many cases, for example
notBeforeandnotAfterones each of which refers to different inscriptions. But I am not sure whether this would make an aggregation impossible from the start.