%* Last edited: Jul 21 13:54 1997 (viola)
%* Last edited: Jul  9 10:20 1997 (viola)
%\documentstyle[doublespace,authdate]{article}

\documentstyle[12pt,fancyheadings]{article}


\setlength\topmargin{-0.75in} 
\setlength\oddsidemargin{0.4in}
\setlength\evensidemargin{0.4in}
\setlength\textheight{9.25in} 
\setlength\textwidth{5.6in}
\setlength{\parskip}{\medskipamount}

\lhead{}
\chead{{\bf \sf PROJECT SUMMARY}}
\rhead{}
\lfoot{NSF Form 1358}
% \cfoot{NSF Career Development Grant}
\rfoot{\bf Paul Viola (MIT)}
\setlength{\headsep}{0.3in}
\setlength{\footskip}{0.7in}
\setlength{\headrulewidth}{1pt}
\setlength{\footrulewidth}{1pt}

\begin{document}
\pagestyle{fancy}

\bibliographystyle{authdate}

\begin{center}
  {\bf \large Summary}
\end{center}

We propose research in three areas: i) handwritten text and diagram
recognition; ii) information retrieval by semantic category from very
large databases; iii) human computer interfaces.  Much general purpose
work has been done in these areas, but we intend to focus our efforts and
evaluate our results in one particular area: a flexible pen-based computer
interface called the {\em Natural Log}.  
The Natural Log is a notebook-sized flat display that combines the
merits of a technical notebook with the those of a consulting expert.  It
uses handwriting input to accept text, diagrams, and formulae and combines
this with a powerful information retrieval capability to allow access to
textbooks, previous work, and tools such as circuit simulators and
numerical analysis packages.  

Handwritten diagram and equation recognition is a difficult
and unsolved problem that lies somewhere between classical handwriting
recognition and computer vision.  It requires both low level reasoning
to analyze ambiguous stroke information, and high-level geometric
reasoning to drive the interpretation of a configuration of symbols.

Information retrieval by semantic category will require capabilities
that go far beyond those available from keyword retrieval systems like
Alta-Vista.  The retrieval of natural language documents requires the
construction of intermediate representations which are more closely
related to document content than a simple list of keywords.  We intend
to learn these representations directly from large corpora using
statistical techniques.

We believe that by focusing on a single encompassing application such as the
Natural Log several benefits arise.  
Firstly, we are not trying to solve a
dauntingly large ``generic AI'' problem.  Construction of the Natural
Log will provide meaningful evaluation criteria and stopping points.
Secondly, the breadth of the project allows graduate students with
disparate interests to explore, interact and collaborate.  For these
students the Natural Log will act to reunify the now disparate
disciplines of artificial intelligence such as vision, reasoning, and
natural language.  Thirdly, the notion of the Natural Log has been
very motivating for undergraduate students.  We have recruited several
undergraduate thesis students who are quite excited about the project
and each has learned a great deal.  Finally, we believe that the
resulting system will be useful in university level education.

\end{document}


%%% Local Variables: 
%%% mode: latex
%%% TeX-master: t
%%% End: