Amir Plivatsky

Sorry for not responding in a timely manner. This fix seems good. Meanwhile, by looking at the code around the fix I found ways to make it slightly faster (added...

It was also a problem with the initial size of `dx->category` (for the same reason). I tested it on **armv7l** virtual machine (**armhf** architecture).

> are you using QEMU to do that? Yes. For the **armhf** architecture I used a Ubuntu cloud image with a `virt-install` command line.

It even doesn't arrive the spell-guess step. By design (per the original one), the lookup order of the tokenizer is as follows, and it **stops** on the first successful one...

> If a word is capitalized and unknown, we should spell-guess the lower-case version. > If a word ends with an s and is unknown, we should spell-guess it. >...

> Post-processing needs to be avoided at all costs. Post-processing is a fail. But I think doing a kind of post-processing of sentences is a natural things people are doing...

> For alternatives to correctly-spelled words ... not sure. we need to add some sort alternatives file. > ... > > there#their: [FOO+]; > ... I don't know if you...

There is a slight problem that it makes "bad sentences" parsable without a way to switch this feature off (like it is possible with spell guessing). A way to load...

> Maybe in some ideal world, we should allow the dictionary to contain something like > `foo: BAr+ or ;` I would like to fix the "[idiom problem](https://groups.google.com/forum/?hl=en#!topic/link-grammar/JgFxbh079O4)", and since...

The Russian "idiom problem" is not so specific. A similar general problem exists for words which can only participate in idioms and have no current dictionary meaning otherwise (such as...