Abstract
We introduce a new kind of language model, which modelswhole sentences or utterances directly using the Maximum Entropyparadigm. The new model is conceptually simpler, and more naturallysuited to modeling whole-sentence phenomena, than the conditional MEmodels proposed to date. By avoiding the chain rule, the model treatseach sentence or utterance as a "bag of features", where features arearbitrary computable properties of the sentence. The model is unnormalizable,but this does not...

This publication has 7 references indexed in Scilit: