Map Of London

After discussions with KMF and JJ, I've added support for the @rendition attribute on <label> and <note> elements, referring to <rendition> elements in the header. This will eventually be extended to other elements such as <fw> which have consistent rendering features. This is only being done in the redesign, not in the old site code. There is a function which safely combines the contents of the <rendition> element with those of a local @style attribute which may also be present, supplying local modification.

We've also decided that all <note> elements that don't have @type, and label elements, must have @place. I'm putting that into the schema now. @place will be constrained to three values: leftmargin, rightmargin and foot.

I now have a working XSLT file which can generate a relatively sophisticated list of links in LINKS1.xml, based on parsing any TEI files it finds in the same folder as itself. That links file is now in the data on the server, and I've modified the XQuery, XSLT and caption boilerplate so that if there are any links, when processing a location file, the list of links is generated at the bottom of the page.

Marked up R2 and processed it through NER using the previous three plays as training set, and including an expanded gazetteer. Results slightly better again, but still not approaching really useful. Created a spreadsheet and chart of the results. Wrote some XSLT to process the tagged ISE plays into linkGrp and link elements in a new generated file called LINKS1.xml in the db/data collection. I can use this to add links to our location files. It still needs its prefixDef stuff defining properly.

Built a new classifier with a massively-expanded gazetteer covering Britain, and (as input texts) the concatenation of the R3_M and H8_F1 texts already tagged and analysed. Results were again slightly better than the last round, especially with regard to precision, but the still-poor recall gives cause for serious concern. I'm going to run one more round of NER and see if we can get any further improvement in recall.

Finished marking up 2H6 with placenames; some questions outstanding with JJ. Ready for NER part 3...

Tagged Famous Victories, but it lacks interesting place referencese. Tagged the first third of 2H6. This looks more promising. Found a couple of gazetteers; I think Britain is covered, but I'm lacking good sources for European country, city and town names.

I now have a training file for NER based on Richard III and a gazetteer, but we need much more training data to have a significant impact on success.

Project meeting notes to self:

• The personography and bibliography pages need a keyword search box which just acts as a filter for that specific page.
• We may need to assign responsibility for person and bibl entries -- add this to the @resp ticket on SourceForge.

I've been running the Stanford NER tool against the text of Richard III and analyzing the results. This is part of the preparation for our paper in Nebraska. Sent preliminary results to JJ for discussion. The next stage is to turn Richard III into training data, add a gazetteer, and then run it against Henry VIII.

Got a list of requirements from PS for the front page and the encyclopedia tile page. I've created the former, in such a way that we can also invoke identical processing for other pages if we want. The latter seems no different from a regular site page, so I've suggested to PS that we create it that way, with some special TEI markup that represents, and is converted to, the tiles. That would allow JJ and KMF to change the tile order or layout if they want to.

Spent most of the day cleaning up the database to remove duplicate copies of files, fix duplicate ids, and similar infelicities which might lead to the slowdown we're seeing in performance. Problems have gone away with regular files and searches, but the Stow 1598 is still proving to be a killer, and we're working with RE to figure out why it's taking so long.