Conference heron::euro_swas_ai

308.0. "February 1991 MCC Technical Report Abstracts & Revi" by ULYSSE::ROACH (TANSTAAFL !) Tue Mar 26 1991 14:00

 

                  I N T E R O F F I C E   M E M O R A N D U M

                                        Date:     05-Mar-1991 00:17am CET
                                        From:     BEANE
                                                  BEANE@BIGRED@MRGATE@DPD03@DPD
                                        Dept:      
                                        Tel No:    

TO: See Below

Subject: February 1991 MCC Technical Report Abstracts & Revisions

	For Digital Internal Use Only

February, 1991 MCC Technical Report Abstracts & Revisions

Title:		DETENTE:  Practical Support for Practical Action
Author(s):	Dave Wroblewski, Tim McCandless & Will Hill
Number:		ACT-AI-014-91
Date:		January, 1991

Complex tasks consist of many threads of activity to remember and coordinate.  
Managing these threads is an important part of human problem solving.  Our 
research attempts to find ways to aid this process.  In this paper, we present 
DETENTE, an object-oriented system to embed personal agendas in complex 
application interfaces.  We will discuss the theoretical motivations for 
DETENTE and explain the three classes and two message protocols that comprise 
it.  This paper also appears in The Proceedings of the CHI'91 ACM Conference 
on Human Factors in Computing Systems.
================================================================================

Title:		Person-Computer Cooperation Through Collaborative Manipulation
Author:		Loren G. Terveen
Number:		ACT-AI-048-91
Date:		January, 1991

This report contains the text of a dissertation submitted to the University of 
Texas at Austin.  The work it describes was done in MCC's Human Interface Lab. 
The primary contribution of this research is to state, motivate, and evaluate 
an approach to the design of cooperative computer systems.  The approach has 
been embodied in a knowledge editing tool, the HITS Knowledge Editor (HKE).  
Empirical testing of HKE has shown the effectiveness of the approach.

The HKE approach grounds person-computer cooperation on collaborative 
manipulation of objects in a shared workplace.  It addresses two key issues in 
the design of cooperative systems -- how to distribute responsibility for 
doing the task between users and the system and how to manage the user-system 
interaction.  In HKE, users design knowledge structures in the shared 
workplace; by inspecting the partial designs, the system is able to offer 
several types of assistance, including noticing problems with the design and 
offering repairs, suggesting additional representational issues, and proposing 
completions of unfinished work.  Since significant amounts of interaction 
occur through manipulation objects in the shared workspace, many problems with 
computing and interpreting textual advice are avoided.

The collaborative manipulation approach has been empirically evaluated in a 
study that compared subjects' performance on HKE and another knowledge editing 
tool based on a different approach.  The studies showed that (1) the expertise 
about the knowledge editing task embodied in HKE facilitated the performance 
of novice users, and (2) providing a workspace and organizing the interaction 
around manipulation of objects in the workspace eased the interaction for all 
users.
================================================================================

Title:		The Construction of Human-Computer Interfaces Considered as a 
		Craft
Author:		David A. Wroblewski
Number:		ACT-AI-052-91
Date:		January, 1991

This paper advances the notion that building human-computer interfaces is a 
craft, and examines the implications for management, research, and teaching.  
Craft is defined as any process that attempts to create functional artifacts 
without separating design from manufacturing.  The state of the art in design 
theory, software design, and human interface design is reviewed and parallels 
are drawn between traditional craft work and the way most software and nearly 
all human interface development proceeds.  Nontraditional roles for 
human-interface researchers, such as articulater craftsman or 
craft-methodologist, are suggested as more appropriate approaches to doing 
meaningful and useful research.  Finally, apprenticeship learning is discussed 
as a viable method of teaching the skills involved in human interface 
construction.
================================================================================

Title:		Planning, Reacting, and Communicating
Author:		Elaine A. Rich
Number:		ACT-AI-066-91
Date:		January, 1991

This paper presents an outline of a model of language use, including both 
generation and understanding.  The model is a hybrid one that exploits both 
planning and reaction, and the issues of when to move from one to the other 
are discussed.  Other issues that are addressed include the role of literal 
meaning and the tradeoffs that occur between a uniform model of language 
processing that avoids arbitrary boundaries between stages and a modular 
system that is practical to build and maintain.
===============================================================================

Title:		Cyc KE Region User's Guide
Author:		Mary Shepherd
Number:		ACT-CYC-018-91-Q
Date:		January, 1991
================================================================================

Title:		FI Manual
Author:		Dexter Pratt
Number:		ACT-CYC-021-91-Q
Date:		January, 1991

This manual documents the Lisp functions which implement the Cyc Functional 
Interface (FI) in the Cyc-140 release.
================================================================================

Title:		The World According to Cyc, Part 4
Author(s):	Douglas B. Lenat, R.V. Guha, Dexter Pratt, Karen Pittman, 
		Wanda Pratt, Keith Goolsbey
Number:		ACT-CYC-022-91
Date:		January, 1991

This report, the fourth of the series describing the principles behind and 
contents of the Cyc KB, deals with an assortment of issues ranging from some 
of the foundational issues behind the design of the CycL language to more 
detailed issues such as the representation of money.  Also covered here are 
descriptions of the treatments of counterfactuals, information bearing 
objects, and resource based constraints on what people can simultaneously 
perform.
================================================================================

Title:		CYC 140/14 KB 67 (Tape 1 of 2 and Tape 2 of 2)
		AI Lab CYC Project
Number:		ACT-CYC-063-91-Q
Date:		February, 1991
===============================================================================

Title:		CYC 140 Documentation Tape
		AI Lab CYC Project
Number:		ACT-CYC-064-91-Q
Date:		February, 1991
===============================================================================

Title:		CYC USER'S MANUAL FOR INTERFACE ON SYMBOLICS MACHINES (for CYC 
		version 140)
Author(s):	Mary Shepherd and The Cyclists
Number:		ACT-CYC-065-91-Q
Date:		January, 1991
================================================================================

Title:		Capturing Language-Specific Semantic Distinctions in 
		Interlingua-Based MT
Author(s):	J. Barnett, I. Mani, E. Rich, C. Aone, K. Knight, and J.C. 
		Martinez
Number:		ACT-NL-003-91
Date:		January, 1991

We describe an interlingua-based approach to machine translation motivated by 
two goals: maximizing the ability of the system to produce correct 
translations in cases of "non-information-preserving translation", and 
minimizing the cost of adding a new language to the system.  In this approach, 
each language in the system must have a reversible, independently-specified 
description which includes an indication of the distinctions that it forces 
speakers to make.
===============================================================================

Title:		A Functional Interface to a Knowledge Base for Use by a 
		Natural Language Processing System
Author(s):	Jim Barnett, Elaine Rich, Dave Wroblewski, Knowledge-Based 
		Natural Language Project 
Number:		ACT-NL-019-91
Date:		January, 1991

We are building the KBNL natural language system, which assumes that it has 
access to a knowledge base that describes both general concepts in the world 
as well as concepts in the domain(s) that are being considered.  KBNL is 
designed to be portable across knowledge bases and to the best it can with 
whatever knowledge it is able to retrieve from the knowledge base it is given. 
In this report we describe the functional interface that KBNL assumes exists 
between itself and its knowledge base.  This is a first draft of what will be 
an evolving document.
================================================================================

Title:		Context Maintenance 1
Author:		Charles J. Petrie, Jr.
Number:		ACT-RA-364-90
Date:		January 29, 1991

Traditional applications of "Truth Maintenance Systems" (TMSs) fail to 
adequately represent heuristic search in which some paths are initially 
preferences.  We show that it is useful, especially for plans with 
contingencies, to maintain the validity of the reason for context choices and 
rejections.  We demonstrate how to do so with a problem solver/TMS 
architecture called REDUX.
================================================================================


	For Digital Internal Use Only

February, 1991 MCC Technical Report Abstracts & Revisions

Title:		Integrated Segmentation and Recognition of Hand-Printed 
		Numerals
Author(s):	James D. Keeler, David E. Rumelhart, Wee-Kheng Leow
Number:		ACT-NN-010-91
Date:		January, 1991

Neural network algorithms have proven useful for recognition of individual, 
segmented characters.  However, their recognition accuracy has been limited by 
the accuracy of the underlying segmentation algorithm.  
Conventional,rule-based segmentation algorithms encounter difficulty if the 
characters are touching, broken, or noisy.  The problem in these situations is 
that often one cannot properly segment a character until it is recognized yet 
one cannot properly recognize a character until it is segmented.  We present 
here a neural network algorithm that does segmentation and recognition in an 
integrated system.  This algorithm has several novel features:  it uses a 
supervised learning algorithm (backpropagation), but is able to take 
position-independent information as targets and self-organize the activities 
of the units in a competitive fashion to infer the positional information.  We 
demonstrate this ability with overlapping hand-printed numerals.
================================================================================

Title:		Recognizing Handwritten Text
Author:		James A. Pittman
Number:		ACT-NN-013-91
Date:		January, 1991

Notebook computers, using stylus input, are currently a hot topic among PC 
manufacturers.  Handwriting recognition may be an important component of such 
systems, but only if everyday sloppy handwriting can be accommodated.  If 
recognizers require unnaturally neat or boxed character input, such systems 
may fail in the marketplace.  Neural nets have shown excellent performance at 
handwriting recognition.  I present three neural net approaches to recognizing 
lines of English text:  one using 2D image input, one using stroke sequence 
input, and one using context to combine the outputs of the other two networks. 
These networks can be combined to form a recognition engine that will handle 
natural lines of handwritten English text, including handprint, cursive 
script, and mixtures of both.
================================================================================

Title:		Semi-local Units for Prediction
Author(s):	Eric Hartman, James D. Keeler
Number:		ACT-NN-050-91
Date:		January, 1991

We consider a certain class of semi-local activation functions, which respond 
to more localized regions of input space than sigmoid functions but less 
localized regions than radial basis functions (RBFs).  In particular, we 
examine "Gaussian bar" functions, which sum the Gaussian responses from each 
input dimension.  We present evidence that Gaussian bar networks avoid the 
slow learning problems of sigmoid networks and deal more robustly with 
irrelevant inputs than RBF networks.  On the Mackey-Glass problem, the speedup 
over sigmoid networks is so dramatic that the difference in training time 
between RBF and Gaussian bar networks is minor.  Architectures that superpose 
composed Gaussians (Gaussians-of-Gaussians) to approximate the unknown 
function have the best performance.  An automatic connection pruning mechanism 
inherent in the Gaussian bar function is very likely a key factor in the 
success of this representation.  We find that when generalizing with noisy 
inputs, the more local the representation, the better the performance.
================================================================================

Distribution:

TO:  Pat Roach@VBE
TO:  Susan Sugar@MWO
TO:  Steve Becker@AQO
TO:  Ed Hurry@DVO
TO:  SHIRLEY CRIDER@DVO
TO:  STEVE DONOVAN@DLO
TO:  DENNIS DICKERSON@DLO
TO:  Gale Kleinberger@HSO
TO:  Mike Sievers@HSO
TO:  Mike Willis@HSO
TO:  Sherry Williams@HSO
TO:  Katherine Jones@HSO
TO:  Dale Stout@HSO
TO:  Tommy Gaut@HSO
TO:  Tom Wilson@HST
TO:  jim rather@HSO