Extracting information from English text

• Natural language parsing is really easy
• You just use a dozen regexps and fiddle with them until it appears to work
• Assuming you don't mind some mistakes, of course.

There are three manipulations we perform on programmes: splitting multipart programmes into constitutents with the 'clumpidx', filling in the programme category based on the description, and parsing the description to find names of actors.

(At this point the speaker shall perform a haphazard and probably broken demonstration by running tv_extractinfo_en on some input and diffing the results.)

• Hog-butchering algorithm
• Emphasis on reducing false positives in matching
• Still many improvements to be made, but it's a tarpit
• Fun though
• Only works well with UK listings, N. America too terse. Might add something for German listings if I get time.