Using word base n-gram language models in continuous speech recognition systems is so prevalent. For using this type of language models, we should extract them from large corpora. Since some corpora are not rich, the extracted language models are not credible. For this reason, most researchers extract class n-grams instead of finding word n-grams. For building class n-grams, all of the word needs to be clustered. In this research, two algorithms based on a new idea for soft word clustering are represented that each word can be assigned to more that one class. These algorithms are Fuzzy C-Means and Possibilistic C-Means. The various parameters of it have been deliberated and finally, this algorithm was applied on 20000 most frequent Persian words extracted from "Persian Text Corpus". The extracted language models are evaluated by perplexity criterion and the results show that a considerable reduction in perplexity has been achieved in comparison to other fuzzy methods. Also, the results of this language model were evaluated on Persian speaker independent continuous speech recognition system and improved the system accuracy.