"In the beginning, there was the word."

1) ears will be a program that
  a) combines listen and train_ears into one app
  b) add. allows starting of a subshell controlled by speech input with
     graphics
  c) with a cmdline option, at once opens the subshell without graphics
  d) has much fewer cmdline options, most config options will be set
     through the config file.
*** Should we use two programs for 1b and 1c? Discuss.
well we just decided to put training and listening both into the app,
so where's consistency?  otoh, until now the separation of t./l. didn't
yield significant space savings but it is expected that the mentioned
split will do that.  is it?  is a recognition-without-feedback-to-user
really what we want?  i think there is a need for it, later, so while
we're at it, we could introduce it now.

2) 1b and 1c together mean that ears won't have NCL train_ears functionality
   which simplifies UI handling.

3) ears will be a program that
  a) uses specific resources, like its configuration info and its screen,
     to start up.
  b) uses user input to navigate through its options
  c) with the training option, uses the recognizer, the current word list, 
     the pattern stream from the mic, its screen and user input to train 
     some words from the list.  
     It then writes the recognizer file, no, a better word might be 
     (recognizer) training data.
  d) with the speech shell option, uses a subshell, the utterance stream,
     the word-action list, user input (and with debug or later the
     alternative option, the screen) to link occurring utterances with
     a shell action.

4) 3c means that ears, even if it once has speaker-independent SR, will
   always have a means of further adapting to the speaker and its
   environment, respectively.
*** Should we rename 'train/ing' to 'adapt/ion' everywhere?

5) the notion of a word list in 3c isn't clear enough: in later versions,
   there might be a dictionary of possible words and a list for specifying
   actions when these words are spoken.
*** What name should the latter have?

6) the notion of words in 3 isn't broad enough: with continuos SR, we
   will have words, sentence parts, or sentences and more.  Let's name
   them utterances.

(to be cont'd...)