Article Preview
TopIntroduction
Lately, one can see a tendency toward an increased use of computers on the so-called Social Web, for both leisure and work. There is an intense use of instant messaging (chats), blogs, wikis, forums, social networks, etc. for informal talks in our spare time, for collaborative knowledge building in different kinds of virtual communities. Moreover, lots of newspapers and libraries decided to provide their content as digital documents on the Internet; the number of websites sharing information about different topics is continuously growing and so on. Consequently, very large quantities of information are available in digital format to whole communities. These documents are the result of communities interactions mediated by language in a socio-cultural knowledge building perspective (Vygotsky, 1978). We may say that they incorporate the voices, the ideas of many persons in the community, in a dialogistic way (Bakhtin, 1981).
One of the most complex facets of text analysis is discourse and most of the applications for its analysis are biased towards one of the two types of texts: written by only one author (we will call them narrations) and those written collaboratively (we call them conversations, but we include here instant messenger chats, transcribed face-to-face conversations, forums, wikis, blogs, etc). We consider that this dichotomy is artificial. Texts, even narrations and essays are always written to be read by others; they are directed towards an implicit dialog with a community; they contain not only the voice of the author (Bakhtin, 1981). The latter should consider even implicitly the potential voices of the readers from the community.
Tools for text automatic analysis are very important for the development of the social web. Information retrieval and associated tools (for example, provided by Google) are now intensively used. Visualization means such as word clouds, tree or graph-based representations of key concepts are provided in many environments. However, they are based on a bag of words model, considering only the frequency of apparitions of these words, and not the discourse structures.
In this paper we present a unified method and a visualization tool for analyzing both kinds of texts (narrations and conversations) within the same framework. The method presented here combines the socio-cultural and cognitive paradigms using the concept of voices’ inter-animation from the polyphonic model (Trausan-Matu, 2010; Trausan-Matu & Rebedea, 2010), based on Bakhtin’s ideas (1981), and respectively the WordNet (http://wordnet.princeton.edu) linguistic database-based processing for analyzing texts. It uses Natural Language Processing (NLP) techniques (Jurafsky & Martin, 2009) (such as building lexical chains starting from the given text and a linguistic database) and the ideas related to identifying polyphonic threads (Trausan-Matu & Rebedea, 2010). Computer-Supported Collaborative Learning is one of the most important applications of the instruments of the social web, of the social construction of knowledge in small groups and in virtual communities (Stahl, 2006). They make use of instant messenger (chat), discussion forums, wikis, blogs or social networking and the participants now enter in massive open online courses. Automatic tools for analyzing discourse in the huge amount of interconnected documents and logs of interaction (in chats, wikis, blogs, forums and social networks) are obviously needed.
In the next section we will present the theoretical ideas that lead to the presented method and to the implementation of the system. The paper continues with the presentations of the system and of the possible uses of the provided visualizations. An evaluation of this system for the task of analyzing Computer-Supported Collaborative Learning (CSCL) chats (Stahl, 2006) is presented right before our conclusions regarding the proposed method and the implemented system.