Amir Plivatsky

It turns out elimination of the empty word in favor of "optional words" was not a good idea after all, since tit works fine only with null_count==0. After much thinking...

After thinking on this problem some time, I was able to fix it without introducing again the empty word device. However, now the fix is not as simple as before....

I the post above I said: > I tested the new code by producing detailed linkages of the batch files and comparing them to those of version 5.3.13 (the last...

> To replace the linear array, you would have to replace constructs like sent->word[i] by some kind of pointer into the word-graph, and replace i+1 with some kind of "get...

Since by elimination the empty-word device I created a problem of inaccurate null count in some cases of `null_count!=0` (no such examples in the corpus files, since the number of...

I will send it for review after the current PRs are applied (so I can sync my code for now conflicts).

I have sent it (PR #588) for your review.

I will try now to eliminate the empty-word as far as possible. In order to eliminate it altogether, I will need to find a way to allow those quotation marks...

Status summary of this issue: - The bug is fixed by PR #588. - The empty-word device is still used for quotation marks (and BTW, due to ZZZ distance=1, this...

Status update: - The empty-word device is still used for quotation marks (and BTW, due to ZZZ distance=1, this doesn't work when a word before the quotation mark is a...