Hi, the goal of the language model is to guess/predict which word will follow any given group of words. Perplexity measures how well the LM is at guessing words following some given context. A perplexity of e.g. 100 means that the LM on average guesses the right word with chance 1/100. This number must be compared against the size of the vocabulary. PP=100 is a pretty good figure good for a very large vocabulary is very large, e.g. 1,500,000 word forms. It does not look so good for a small vocabulary, e.g. 15,000 words. The lower the perplexity the more useful the LM is. In MT a better language model often means better (more fluent) translations.
Marcello > On 07 Jan 2015, at 09:47, Leandra Vogel <[email protected]> wrote: > > hello mt folks, > > can someone please explain (if possible in layman's terms) what perplexity > means and why it is important? > > maybe also give a simple example? > > Best, > Lea > _______________________________________________ > Mt-list site list > [email protected] > http://lists.eamt.org/mailman/listinfo/mt-list _______________________________________________ Mt-list site list [email protected] http://lists.eamt.org/mailman/listinfo/mt-list
