Fostering Summarisation Skills via the Use of Automated Feedback
The main source of knowledge for students are texts, i.e. students mainly learn by reading. It has long been noticed that summarizing improves text comprehension processes and by this the efficiency of learning from texts. To this end the intelligent tutoring system conText is being developed, that assists by writing of summaries in German language.
The system uses Latent Semantic Analysis (LSA), a computational model of word and documents similarities. LSA analyzes the co-occurrence of words in huge corpora and generates a high-dimensional vector space in which single words, sentences and even paragraphs are commonly represented by corresponding vectors. Thus, the vector of a word (or a document) represents its co-occurrence with all other words (documents). Accordingly, the vectors can be regarded as representing the meaning of words and paragraphs at least to the extent to which the meaning of a text is determined by its use.
The project aims at creating “semantic” vector spaces for German corpora in different subjects in order to use them for a tutorial program for writing summaries. Our work strongly profits from close cooperation with the LSA research group and the “Summary Street” project at the University of Colorado, Boulder (Profs. Eileen & Walter Kintsch) as well as from cooperation with Profs. Sandra Jean La-Rose and Guy Denhière (Paris).
The project is currently under development. A prototype of the system is running under http://www.summa.psychologie.uni-wuerzburg.de/summa/ (-> LSA). Please keep in mind that due to maintenance and software development, you may encounter malfunctions. Moreover, the system is only in German language so far. Please contact us, if you would like to get access to the system. We will send you the login details.