Perplexity Of A Language Model