The study of architectures for Natural Language and Speech applications has been necessitated by the large scale proliferation of such applications into desktop and embedded computers. This paper proposes NLPBench a benchmark for analyzing statistical Natural Language applications. This study validates the need for such a benchmark by demonstrating the lack of such applications in the currently existing benchmarks. A comparison between architectural characteristics of the applications in NLPBench is done with currently existing benchmarks to further prove the need for such a benchmark.