At some point in the past I broke the regular expression that handles input verification for verbs to parse. It was incorrectly refusing the $ used to indicate a shin, š. That has been fixed.
Finally got around to fixing the recent parsings link. It now queries off of the MySQL table of parsings. Not perfect, but at least it is there
There were problems with generation, both due to the mishandling of spaces in the weak stems and with missing quotes. Both have been fixed
Rewrote the entire CGI interface in perl, using YASWI as an interface between SWI-prolog and perl. I need to fix the recent query, but everything else seemed to work without much trouble.
Moving the codebase from RH 7.2, JPL 1.0.1 and SWIPL 5.2.10 to Fedora C3, and SWIPL 5.4.7 broke everything. The code was eventually rewritten, removing the cgi parsing stuff in java, and moving it to some AWK in the shell script. The goal was to use JPL3, but that broke oddly.
Upgraded the underlying SWI-Prolog to the most recent stable release, 5.2.10. Checked speeds and generation seems the same, but parsing a tiny bit slower, not enough to worry, change has been made permanent.
Status update on infinitive work: Prolog code is complete, pending some slight discrepancies for strong and first weak verbs. As the variations are far fewer, t, and tn derived stems will be included.
Finally got the new hard drive in the server and got everything back online.
Worked on the multi-threaded, socketed version of the Java code. Got it working, but noticed that the parser was no longer returning more than one value, so moved threaded changes to new code base. Tracked problem to double increment of a value, fixed. Threaded version still has problems, but looks promising
The java wrappers now generate an error if the root for verb generation length !=3. Also fixed the akk.pl prolog file so that there are two 'verb' bits, the old verb has become verbns(no suffix) and the part that used to do the -ma removal is now the only true 'verb' procedure, with a pass through to 'verbns'. The generator calls 'verbns' so that only the regular forms are generated, no -ma forms are ever generated, only parsed
