latin-language-toolkit/llt-segmenter

Segments text into sentences.

llt-segmenter's Issues

Fix Regexp in jruby

jruby breaks with different lookbehind handling.

\n handling with optionality

Let the user decide how many \n might constitute a sentence boundary.

Switch approach of handling abbreviations?

Right now we split after every . and check if it probably is an abbreviation.

Would it be safer to change this approach? I.e. regarding everything that is one or two characters long and with a trailing . as abbreviation and add execptions for that rule?

Fix the dreaded little bugs once and for all

No more ad hoc fixes, especially complete TEI files seem to cause trouble.

• Show real commitement to the test suite
  • Add real files as fixtures, don't rely on contrived examples
• Test the regexps independently
• Try to work around the while loop and use something more robust or at least something that doesn't send the service into the land of no return
• Test output on a higher level as well, see latin-language-toolkit/llt#11

segmenter doesn't like empty tag at the end.

The segmenter hangs (or crashes not sure which) with XML that ends with an empty tag.

E.g.

Numen inest vati, vatum mens consona caelo est, Nec certus scit fallere Apollo.

Take the milestone tag at the end away and it returns fine.

ignore punctuation in element names and attribute values

the segmenter should ignore punctuation in element names and attribute values.

ability to override definition of SENTENCE_CLOSER ?

I wonder if newline_boundary could be combined with an option to ignore all except new lines as the segmentation?

I'm experimenting with a simple UI for creating sentence alignments which allows the user to break up the text into the the chunks they want to align by new lines, in which case I think we might want to just ignore punctuation for the case of segmentation.

I'm not sure of the utility of this though, so for discussion before proceeding.

Investigate build failure of API Branche under jruby

flaky behavior with xml=true and newline_boundary

The segmenter just hangs (or maybe crashes not sure which) with xml content which includes new lines but the newline_boundary isn't set (or is set to something other than 1). It works okay if the xml in the input doesn't have any new lines. I didn't take the time to look at the code for this, but I think maybe we should just not try to calculate segments from new lines at all when the input type is xml. What do you think?