Inconsistent tagging for single word sentences

[desilinguist]

For single word sentences, ZPar tagging is not always consistent. For example, given the following input file:

REBELLION

I am going away .
The rebellion is just another word for change and change is necessary to live .
REBELLION
REBELLION
The rebellion is just another word for change and change is necessary to live .
REBELLION
This is just another sentence .
REBELLION

I get the following tagger output:

REBELLION/NN

I/PRP am/VBP going/VBG away/RB ./.
The/DT rebellion/NN is/VBZ just/RB another/DT word/NN for/IN change/NN and/CC change/NN is/VBZ necessary/JJ to/TO live/VB ./.
REBELLION/NNP
REBELLION/NNP
The/DT rebellion/NN is/VBZ just/RB another/DT word/NN for/IN change/NN and/CC change/NN is/VBZ necessary/JJ to/TO live/VB ./.
REBELLION/NNP
This/DT is/VBZ just/RB another/DT sentence/NN ./.
REBELLION/IN

As you can see, the word REBELLION is tagged as NN, NNP and IN all in the same text. This is obviously inconsistent.

[desilinguist]

Looks like the issue is underflow. In tagger.cpp, we check whether there are 2 words between the current word and the end of the sentence but we never check whether the sentence actually has more than 2 words. For a single word sentence, i.e., where m_CacheSize is 1, m_CacheSize-2 will underflow and become a large positive number. And, therefore, the condition will be satisfied when it shouldn't be.