On the Use of Self-Organizing Map (SOM) in Linguistic Visualization

The availability of linguistic corpora in electronic form has made it possible to make various kinds of computer-based analyses on them. The use of Self-Organizing Map (SOM) in the analysis and visualization of various aspects of textual material is outlined. The basic approach for creating maps of lexical items, documents, or set of languages or dialects is described with reference to original research contributions. The Self-Organizing Map (SOM) was originally developed by professor Teuvo Kohonen for to explore and organize data in an unsupervised way (Kohonen, 1981; Kohonen, 1982a; Kohonen, 1982b). From the very beginning, one speciic application area has been speech analysis. The SOM algorithm has proven also to be a general-purpose method for analysis of complex phenemena and it has been used in a multitude of areas like process monitoring , image analysis, and categorization of economic data. A list of references of over 1500 articles on the SOM is available for anonymous ftp user at the Internet site cochlea.hut.., 130.233.168.48. SOM PAK, the software implementation of the algorithm, can also be obtained from the same location. The basic self-organizing map architecture is a combination of a map and a input vector (see Fig.1). The map consists of a set of units (cells) which are usually organized as a two-dimensional grid. All the elements of the input vector are connected to each unit in the map. The learning process is competitive and unsupervised, meaning that no prior classiications need to be done. During the adaptation the units become speciically tuned to various input patterns or classes of patterns. In the basic version, only one map unit (winner) at a time gives an active response to the current input. The locations of the responses in the map become ordered as if some meaningful nonlinear coordinate system for the diierent input features were being created over the network. The \image" of an input item on the map is deened to be in the location, where the map unit vector matches best with the input vector in some metric. Detailed description of SOM, its mathematical background, and many examples of its applications has been presented in (Kohonen, 1990) and (Kohonen, 1995).