Academic

This is what I've managed to get through today:

• Prepared an input data set consisting of "long" passages (speeches from Hamlet, Quarto 1 vs Folio 1, running from a few lines to a few dozen), and created a batch script that processes these, which are read from separate files as opposed to a single input file.
• Created a data set from lines in two versions of one speech (Quarto 2 vs Folio 1), and a script to process that data set.
• Created a third data set which consists of a variety of specific mutations and combinations of mutations of one input sentence, to demonstrate the situations in which the two similarity metrics throw up similar and different results.

MB has just sent me some more interesting sample data from Shakespeare, which I'll also prepare, and AM has sent the chapter of his thesis dealing with ShingleCloud, which I'll need to read carefully tomorrow. He may also be able to add the jaccard measure to ShingleCloud, which would probably give a better test of similarity than the crude averaging of needle and haystack containment that I'm currently using.

I'm also going to prepare some data sets using XML markup. I'm expecting to see that the USM performs a bit more effectively on this data than the pure textual data, but that remains to be seen.

Today I've written the bash script I'm going to be using to create arrays of data points to compare ShingleCloud with USM. This is what it does right now:

• Reads some template files from disk, to assist in building two kinds of output: an XHTML page with tabular data, and a spreadsheet in FODS (Flat Open Document Spreadsheet) format, for use in OOo Calc.
• Reads the test data in from a text file. Right now, the file contains all the test data, with one pair of text items on each line, separated by a slash. This may be quite sufficient, since I don't anticipate running tests with items larger than paragraphs, but it would also be feasible to read in a list of filenames from which to retrieve larger bits of data.
• Calls USM on each bit of the test data and retrieves the scores.
• Calls ShingleCloud twice, once with --containmentneedle and once with --containmenthaystack, and then averages their scores. With ShingleCloud, these two scores can be radically different, so averaging them is one crude approach to creating a single similarity measure. I've written to AM to ask if he uses anything more sophisticated in TEIComparator.
• Creates an XHTML document showing the input data and the results in a table.
• Creates a FODS document with the data in a table.
• Opens the XHTML file in Firefox, and the FODS file in OOo Calc.

Learned a few things I didn't know out of this process. For instance, bash can't do floating-point math (call awk instead); a FODS file is much easier to create than a full-scale Open Document file (it's just a single flat XML file); and this is how you call an external application from a bash script and return immediately, without getting your shell session hijacked by errors from the external application: `soffice -calc -nologo $FODSFILE 2>/dev/null` &

Easy to do with JOptSimple.

I'm back into working on the Universal Similarity Metric code and background reading in preparation for my DH 2010 presentation. I've done a fair bit of background reading today -- much more to do, I think -- and I've also gone back to work on my Java class, which I think will form the basis for the comparative work with ShingleCloud. I needed to add a lot of command-line switches and parameters, so I've incorporated the JOptSimple library to make it simpler to handle them. I've added these switches/parameters:

• -compare: indicates that a comparison operation should be done.
• -str1="some string": the first string for the comparison.
• -file1=somefile.txt: a path to a file, the contents of which will be used as the first string in the comparison. This is only used if str1 is not present.
• -str2="some string": the second string for the comparison.
• -file2=somefile.txt: a path to a file, the contents of which will be used as the second string in the comparison. Only used if str2 is not present.
• -reverse: indicates that the calculation should be done by performing the operation twice, the second time with the strings switched around, and averaging the results. This is available because slightly different results often appear when doing the comparison one way, as opposed to the other.

I need to add some more sophistication to this, including a -help switch to output explanations of all the switches, and possibly some other switches that allow component operations to be done without a full comparison (such as compress-and-measure).

AM has updated his Shingle Cloud code and very kindly added a command-line interface I can use to compare SC to the USM. I've downloaded and tried out the new SC version, and I'm now puzzling out the command-line switches, and trying to figure out what would actually be analogous to a straightforward USM score (between 0 and 1) from SC.

Up at the crack of dawn to present my little talk remotely over telephone conferencing, with someone at the other end advancing my slides. A very odd experience (no feedback, no real sense of what people can see on the screen, no idea if you're going too fast or too slow). Weird. People asked on-topic questions at the end, though, so I guess even if I failed to communicate properly, the slides did at least part of the job.

Made these changes, to make it easier to work remotely with someone clicking through the slides, and to clarify some things:

• Removed all the incremental bullet displays.
• Added explicit numbering to slide titles.
• Added highlighting to XML tags in the <biblStruct> example.
• Expanded the notes considerably.

Finished drafting my additional section on electronic documents, and then went back and finished the existing text. Sent it to LR and KH, with a summary of the main changes. It's a relief to get this off my plate for the moment; it was hard work. I've had some feedback from KH, which I've incorporated, and there will presumably be more before the meeting next week.

I've now written about half of my new section on electronic references for the revision to the TEI guidelines. It's slow work...

Chaired Micro-talks session at TRUTH (8 presenters in 1 hour). In advance, got key theme from each talk and prepared little bridging blurbs to provide continuity and cover while presenter got prepared. Informal feedback seemed to be that microtalks were successful in giving people a sense of the range of work being done and how various projects had connections with each other.
Also attended the keynote guy on empirical research into retention of language learning, and the afternoon talk-tables.