Texts are language tools to transfer and store information in written form. In this thesis, two analysis methods: function word detection and collocation extraction which support the studies in extraction of meaning and measuring the amount of information are presented.Function words, playing grammatical roles, appear in high frequencies and rarely contribute to the meaning of the text. In this thesis, a statistical method, discriminant analysis, which merges several linguistic features, is proposed to discriminate function words. The proposed method is utilized on Turkish texts. The results show that the method brings a significant improvement in distinction compared to well known frequency of occurrence and tf-idf methods.Collocations are the groups of words which collocate to complete or enforce their meaning integrity. In this thesis, collocative tendency method which states that any word in a collocation must suggest or at least imply the following words composing the collocation is presented. Firstly, current collocation extraction methods are utilized on a Turkish corpora and the effect of stemming is investigated. Following, the proposed method is tested on a base data set extracted by some statistical techniques and it is evaluated by precision and recall measures. It is found that collocation tendency method gives a remarkable improvement in current techniques of collocation extraction. |