SUBJECT: [ &NAME ] Re : read a text &NAME &NAME , I 'm working on this topic as part of my Ph. &NAME thesis on &NAME &NAME Signals from &NAME ' . In this work I have shown that the position of words in the text is more successful at recognizing topic words than non-positional word statistics usually used in &NAME . By ' topic words ' I mean words that are generally used in discussin &CHAR the topic . For example : ' Belief Networks ' and ' &NAME &NAME ' are topic phrases for Artificial Intelligence . Note , however , that this work focuse &CHAR on words only at this stage . Given a training set for a topic &CHAR , our approach is used to extract ' topic words ' for &NAME Thus , &CHAR is represented by this set of topic words . When the system is introduced with some new document &CHAR , it uses &CHAR 's topic words to evaluate the intensity of discussion on topic &CHAR in the new document . The more &CHAR topic words there are , and the more condensed they are , the more intense the discussion at that point in &NAME I can send you more details as soon as the writeup is done . Please let me know if you are interested . &NAME &NAME . &NAME &NAME &NAME &NAME | ' Do n't go where the path may lead . Department of Computer Science | Instead , go where there is no path an &CHAR University of &NAME | leave a trail ' On Mon , &NUM &NAME &NUM , &NAME &NAME wrote : <QUOTE> Hi , I am want to explore some other way to classify and / or understand texts . I have some experience in words-occurrence statistic techniques , but I ju st want more :- ) . In fact I am looking for works about algorithms able to learn grammar basics , phrase structure or just words relationships ( location in text , common topic , .. ) . The idea is to get an algorithm able to ' read ' a &NAME &CHAR , extract some ' topics ' just in time- and even ' predict ' what will be read in order to get an opinion about the text . I would prefer some non-natural-language-processing as long as animals can learn a language without grammar knowledge ( .. It seems .. ) . .. I know , that goes away from classic text classification. . Does it already exist something going in this direction ? Every weird inputs are welcome &SMILEY &NAME <END_QUOTE>