The next four(ish) lectures will be concerned with two
``mundane'' tasks: natural language and vision. Both tasks
are concerend primarily with understanding: in natural
language we are primarily concerned understanding spoken or
typed language
,
while in vision we are concerned with understanding images.
In both we are trying to get from some input sense data to
some representation of what that data really means.
Of course ``what it really means'' is rather vague - we aren't generally concerned with the deep philosophical implications of the data (e.g., beautiful sunrise means there is sense to the universe ..). We're just concerned with getting a sufficient interpretation for our purposes. In vision we might just want to recognise wonky widgets so our robot can remove them from the conveyor belt - the ``meaning'' of the image would just be a classification of images into wonky and non-wonky widgets. In natural language we might want to be able to answer the user's natural language questions given some database of information - the meaning of the sentence might then look something like a formal database query. For other tasks the meaning representation might be quite different, and the understanding task more or less easy.
Rich &Knight, ch 14 discusses the problem of understanding in general. The rest of this section though will be concerned solely with natural language understanding.