Abstract
We have various kinds of data of time series such as a temperature record. We understand them via their linguistic expressions in a natural language rather than conventional stochastic models. We have proposed a method to extract their linguistic expressions of time series with a global trend and some local features. It goes without saying that a global trend is most important for summarization of a time series. However, other features such as some local features, oscillation and positions in a plotting area are important for us to understand details of time series. In this paper, therefore, we add new features made from a time series of changes of values. The feature represent a difference from a trend around the point. It is not a local feature in previous research such as the position of locally large deffirence between the original time seires and the time series representing the global trend. And we use the time series of changes of values to tune the fuzzy intervals in the temporal axis.
