Updating global variables without locking

[Google Code Exporter]

Some of the global variables are updated in the function TrainModelThread 
without the use of locking. Here is one of them:

word_count_actual += word_count - last_word_count;

Can someone please explain how this works. 

Also, can someone please help me with the following question:

For negative sampling, why do we need a unigram table to choose the negative 
samples from? Why can't we just choose random words from the vocabulary?

Sincerely,

Vishal

Original issue reported on code.google.com by [e-mail] on 21 Jul 2015 at 6:26