^^h-'^^^Si-
I
Of
T
HD28
•
'
i
•M414
5
^AR
1
S 1988
Translating Type Hierarchies:
Framework Analysis and A Proposal.
Jintae Lee and
Thomas W. Malone
Massachusetts Institute of Technology
Abstract
Many
systems are based on various types of messages, forms, or other
users of such systems need to communicate with people who use
different document types, some kind of translation is necessary. In this paper, we
explore the space of general solutions to this translation problem and propose several
specific solutions to it. After first illustrating the problem in the Information Lens
electronic messaging system, we identify two partly conflicting objectives that any
translation scheme should satisfy: preservation of meaning and autonomous evolution
of group languages. Then we partition the space of possible solutions to this problem
in terms of the set theoretic relations between group languages and a common
language. This leads to four primary solution classes and we illustrate and evaluate
each one. Finally, we describe a composite scheme that combines many of the best
features of the other schemes.
Even though our examples deal primarily with
extensions to the Information Lens system, the analysis also suggests how other kinds
of office systems might exploit specialization hierarchies of document types to simplify
office
documents.
When
the translation problem.
Computer-based
office
systems often use various types of messages, forms, and other documents to
communicate information. When
the
same types
for
the people
who wish
to
communicate with each other use exactly
of documents, translation problems do not arise.
communicate with others who use
The needs
all
different kinds of documents,
such translation seem likely
to
However, when people want
some kind
of translation
is
to
necessary.
become increasingly common as more and more diverse
kinds of systems are linked into heterogeneous networks.
We
have been particularly concerned with one instance of
template-based communication systems
how users
of such systems can
(e.g.
Malone
et.al.,
this
problem that arises
1987b; Tsichritzis, 1982).
communicate with other users who have a
in the context of
The problem
different set of templates.
Other examples of the problem may arise when users of different word processing systems wish
exchange documents, or when difTerent companies wish
standards
to
to
to
use Electronic Data Interchange (EDI)
exchange business forms such as purchase orders and invoices.
WP //i974-88
is
In this paper,
we
several specific schemes for solving
Information Lens system (Malone
(2)
Our primary
it
(1)
et.al.,
may change over
we have been pleased
One
kinds of document interchange as well.
all,
in
same types
of messages,
analysis has implications for other
most important general lessons of our analysis
that several novel and attractive translation schemes are possible
arranged
of the
time. In addition to these extensions to
to find that the
of the
problem and propose
1987b) that allow different groups to
the groups use some, but not
the message types used by each group
the Information Lens system,
to this translation
goal has been to design extensions to the
1987a; Malone
et.al.,
communicate with each other when
and
general solutions
will explore the space of
when
the
is
document types are
an inheritance hierarchy with certain types of documents being regarded as specializations
of others.
In the first section of the paper,
Lens system.
we want
that
we
illustrate the
its
solution to satisfy.
suggesting a dimension along which
each
solutions
have
I.
for
we propose
In Section 4,
class.
we explored.
Finally,
state the problem
In Section 3,
to partition the
it
arises in the context of the Information
more precisely
we explore the space
terms of the objectives
of possible solutions by
space and examining a representative solution for
new scheme that combines most
a
in
we conclude by hinting
of the desirable features of the
at the implications that this research
might
more general contexts.
Illustration: Inter-group
1. 1.
we
In the second section,
problem as
communication
in the
Information Lens system
Description of the Information Lens system.
The Information Lens
is
an intelligent system
groups and organizations.
helps
them
supports
More
find useful
helps people
messages
common actions
specifically, the
It
people
supporting information sharing and coordination
filter, sort,
or other
may
for
and
prioritize electronic
messages they receive;
documents they would not otherwise have seen; and
it
Lens system enhances the usual capabilities of an electronic mail system with
may
or
may
not choose to use:
(11
Senders
can compose their messages using structured templates that suggest the kinds of information
included in the message and likely alternatives for each kind of information.
rules to automatically process their incoming messages on the basis of the
in
it
take on receiving certain kinds of messages.
four important optional capabilities, that individual users
senders
in
constructing messages.
folders; in other cases, the rules
In
may
some cases these
(2)
to
be
Receivers can specify
same dimensions used by
rules filter or classify incoming
automatically take actions such as replying
to or
messages
into
forwarding the
messages. In
still
other cases, the system takes no automatic action but simply suggests actions that a
person might take on receiving a message of a certain type
(3)
a message, in addition to specific individuals or distribution
"Anyone")
anyone
to indicate that the
who might
else
addressed
to
sender
Senders can include as an addressee of
a special mailbox (currently
lists,
willing to have this message automatically redistributed to
is
be interested, and
i4)
Receivers can specify rules that find and show messages
"Anyone" that the receiver would
not otherwise have seen.
In addition to electronic mail, bulletin boards,
and conferencing,
this basic
framework supports a
management.
surprising variety of other applications including task tracking and simple calendar
The system
is
described in
much more
detail elsewhere
(Malone
For our purposes here, a central feature of the system
is
that
message templates or "message types." For instance, Figure
announcement message. The template contains
fields
found
menus
field.
to
A
in all
fields for
similar editor
is
used
for difTerent fields,
to create rules for
given type based on the values of the same
fields
1
1
et.al.,
it
1987a, Malone et.al
1987b).
depends on a set of semi-structured
shows a sample template
The display-oriented
for a
meeting
editor uses pop-up
but users can put any value they desire in any
automatically processing incoming messages of a
used by senders
to
compose messages of that type.
Save
Deliver
,
Time. Place, and Date, as well as the usual
messages (such as To, From, and Subject)
suggest alternative values
named
Cancel
Message Editor for Meeting Announcement
Default
To:
E<plan.3t: ion
A 1 1ernat 1 /e t
From: Jin Lee: 3 loan
S c h CI 1
cc: Jin
III
:
M as
s I
nsrTech
LeeiSloan 3chool:MassInstTech
Subject:
In-Reply-to:
Topic:
Day:
Meeting Date: 4-Oct-S7
Time:
Place:
Text:
FIGURE
editor
l.l
A
Body
of
Message
<<
template for a meeting announcement message
is
displayed
in a
message
The
set of
message types currently
The subtypes
subtypes of others
processing incoming messages).
in a
Subject
Bug
Its child,
field
shown
As shown
in Fig. 1.2.
network with some templates designated as
properties (such as alternatives for field values, and rules for
Any subtype may
also, in turn,
add new
property values inherited by the parent (Fikes and Kehler, 1985).
message type has the
is
of a given template can automatically inherit from the parent
names and other
field
MIT
use by our research group at
message types are arranged
in the figure, the
template the
in
Action Deadline in addition
to the
fields or override
any of the
For example, the Action Request
usual fields such as To, From, and
Fix Request, inherits these fields and has additional fields such as
Bug Name
and Severity
Message Type Tree
Bug
Action Request
Fix
LENS Bug
Request
Meeting Proposal
..-
Meeting Announcement
"- Notice
.— — Software Release =^I
-
NYT
\\
'
\
FIGURE
We
1.2
hierarchy
that
parent.
=^I._
'"
Bug
Fix
Commitment
Meeting Acceptance
network of message types as a
The multiple in/ien/ance
we use
Discussion Item
The message type hierarchy used
will refer to this
^- Bug Fix Announcement
Conference Opener
._----
Commitment
_— -New LENS Sysout
Article
Network
\
Meeting Cancellation
Announcement
Publication
\
Seminar Notice
LENS Meeting Announcement
•*?::
'
/
Message >-
Request
Request for information
""
/
Fix
at
MIT
type hierarchy
even though
capabilities of the underlying
(Stefik et.al., 1983) allow one
message type
it
need not be a
strict
knowledge representation system
to inherit properties
from more than one
Communicating with other groups.
1.2.
same network
In the current version of Lens, all users of Lens in a local group share the
Lens users are also connected
types
to
communicate with other people who do
not use Lens at
and they sometimes communicate with
all,
active test site for Lens, in addition to our research group at
is
how
Our current
local
to
handle messages
to
solution to this problem
group— it
is
and from other
is
(We currently have one other
When
the following:
a Lens
message type are sent as the
The body
message
electronic mail recipient can read a
first
of the
"Message Type" and another one called "Text" that contains
Thus any
The problem that
)
all
message from a Lens
in the
a Lens user receives a message, the message
called
Message Type and whether the name
is
electronic mail user.
in that field
If
the message
names followed by
is
into the
If not,
whether
to see
the message
Announcements
also receive
body of the messages
definitions for a given
it
the sender
message
message
one of the
contains a field
is
the
or
Bug
message type,
for
messages (such as To, From, Subject,
the sending site
this
up
Fix Requestsl. They can,
much
of the
Lens
users.
who simply send messages with
When
set
and process specialized message types from Lens
the appropriate field
and the receiving
site
have exactly
communication works as desired. However, there
and the receiver use the same name
types, then the
If so,
treated by Lens as the most
are two kinds of problems that can occur:
(1) If
field
colons) will
one recognized by the local group
fields present in all
users at other sites—or even from non-Lens users
same
is
checked
when communicating with non-Lens
With the current scheme. Lens users can
the
the remainder of the
the system degrades quite "gracefully," taking advantage of as
functionality as possible, even
names typed
header
message also includes a
These messages are not processed by any of the special purpose rules
however, be processed using rules that check
Thus
its
Thus Lens users can receive messages from any connected
particular kinds of messages (such as Meeting
Text).
is first
treated as being of that specialized type.
general message type of all--Message
and
sent--even within the
body of the message-
When
message
is
few lines of what the mail
user.
specialized types with extra fields, these fields (including the field
simply appear
arises in this
sites.
transport system regards as the body of the message.
called
MIT
sent as an ordinary message with only the standard mail header fields in
All the other fields specific to this
message
national electronic mail networks, however, so they often
users of Lens at other sites that use a difTerent set of message types.
situation
of
for
what are actually different
incoming message may have either extra
fields,
missing
In these cases, the extra fields are Inserted in the text of the
fields, or both.
and the missing
names are
fields are
simply treated as being empty. Even
may
present, the sender
if fields
message
with the same
be using these fields in ways that are different from
what the receiver expects.
(2)
and receiver use
the sender
If
message types
messages
Bug Report and Bug
(e.g.,
satisfactorily, but
have been applied
to
names
different
none of the automatic processing that could otherwise
problem we
will explore in the
much
remainder of
functionality as
possible
this
in
paper
is
fails to
how
to
be as helpful as
might be The
communication between groups without imposing
document type
definitions
The Problem
We
formulate our problem more precisely as follows:
Let there be
shared by
all
denoted by
group.
in
it
design systems like Lens that can
excessive burdens on the different groups to coordinate their changing
2.
same
the messages will be invoked.
occur in either of these cases, but the Lens system
as
in fact the
Fix Request), then they can read each other's
No disasters
retain
what are
for
We
some number
members
a/, 02, etc.
will
of the group
We
Each group has a
o( groups A, B, C, etc.
For instance, the types shared by
refer to the set of
set of
document
members
types that are
of group
A might
be
document types used by a group as the language of the
be particularly concerned with languages that are type hierarchies-- that
is,
languages
which some document types are specializations of others (their "parents") and automatically inherit
properties from their parents.
There are also some number o( translation schemes
for
translating types from one language to another
For instance, Txy(x) might be a scheme that translates the types used by group X into types used by
group Y. The problem
is
to
formulate translation schemes that
a given type of document as
their type definitions with as
detail.
much
as possible while
much autonomy
(2)
(1)
preserve the original "meaning" of
allowing different groups
as possible.
to create or
change
Let us explore these two objectives in more
2.
1.
Preservation of Meaning
The question
what a document,
of
or
any other expression, "means"
is,
philosophical and linguistic issue (eg see Putnam, 1975. JackendofT, 1983)
regard
not
will
"meaning"
the
correspondence with "reality"
Instead,
we
will focus only
take on receiving an "expression" (eg a document)
the
same actions
in the
same
means
expression, therefore,
situation,
to
Thus,
to
for
same range
an intuitive sense
To preserve meaning
We
what
it
way
to
of
an
that the receivers of the
would have wished
of actions they
of
falsifiablel
our purposes, two documents that evoke
have the same "meaning".
they received and "understood" the original expression
"understood" and appeal only
Perry, 1983).
on the actions that a receiver might wish
translate the expression in such a
translated expresion can perform the
&
an expression as having some (potentially
of
complex
For our purposes here, we
and ignore most of these complexities. Following (Barwise
will oversimplify greatly
we
of course, a
to
perform had
intentionally leave undefined the term
would mean
for receivers to
"understand"
the original expression.
In the context of Lens, for example,
Tab
unknown
that translates any type
Suppose someone
Message.
we can interpret
to
Take the simple scheme
a group into the most general type the group has, say
Group A sends
in
this criterion as follows.
to
someone
Group B
in
a
message of type Action
Request, and suppose that Group B does not have the message type Action Request. The operations
that the
members
Forward,
etc.
of the group
want
to
apply on objects of the generic type. Message, such as Reply To,
are preserved under this
scheme However, the operations
messages such as Making a Commitment would not be available because,
object received
it
is
now an
object of type
specific to Action
for all the
Request
system knows, the
Message no matter what the original group may have intended
to be.
Now
suppose further that group B has a message type called Request and that another translation
scheme T'aB allows translation
translation
of all objects of type Action
Request
scheme would preserve the meaning of the original type
to this type.
to the
This
Request
extent that the operations
that the group has defined on Request objects overlap with the operations that the group would have
defined over the set of objects that the other group classify as Action Request
operations are the same, then the meaning of the type
the groups
type
is
name
it.
If
is
One can imagine
receivers cannot understand the
document as
its
the two sets of
fully preserved, despite the differences in
the two sets of operations do not have
not well-preserved.
If
much
in
common, then
situations where this definition
senders did--but
is
useful
enough
the
is
for
meaning
how
of the
problematic-e.g..
our purposes.
One way
to
ensure
semantic preservation
full
However, such a requirement has
2.2
is to
require
all
groups
share the same set of types.
to
The next objective captures the
its costs.
tradeoffs involved.
Autonomous evolution of group languages
Each group would,
their
in general, like to be able to create or
own needs with
change their
tj-pe definitions
as few constraints as possible from the other groups.
We
according to
separate three aspects
of this objective: maximizing expressive adequacy, minimizing the need for consensus, and minimizing
the need for propagation of changes.
22
Maximize expressive adequacy Each group would
1
are useful for
difficult to
its
meet
Requiring
purposes
different parts
them
to
company might want
make
to
company might want
to
group products
it
adequacy because the different goals and contexts of
conflicting distinctions
make many
and subassemblies of many
express the distinctions that
groups to sh£U'e the same set of message types makes
this objective of expressive
different groups often lead
division of a
all
like to be able to
For example, a manufacturing
detailed distinctions in their messages about
different products while the
marketing division
way and make
in a different
of the
same
detailed distinctions about
various market segments.
2.2
2 Minimize need for consensus
while requiring as
static,
little
Each group would
its t\-pe
definitions to
meet
its
2 2 3 Minimize need for propagation of changes
have
to
know about
example,
if
its
type structure
for
consensus from other groups, the more easily
current needs.
When
one group makes a change, other groups might
the change so that they can reflect this change in their translation schemes.
one group renames a type,
from that group would have
to
all the
know about
groups that maintain a dictionary
the change
translation scheme requires, the more desirable
In the next sections,
change
consensus as possible from other groups. In general, type structures are not
but evolve as needs change. The less the need
a group can change
also like to be able to
we
will discuss
how
it is
for translating types
In general, the fewer of these updates a
on this objective
these different objectives interact
translation schemes against these objectives.
For
when we evaluate different
3.
The space of possible translation schemes
There are a number of possible translation schemes that achieve,
we discussed
above.
In this section,
we
to dilTerent degrees, the objectives
will describe several general
problem, evaluate each solution with respect
to the objectives
examples of how each solution could be implemented
for the
solutions to the translation
discussed above, and provide specific
Lens system.
W'e partition the space of possible translation schemes by using a dimension that centers around the
notion of
common language and
categorizing possible solutions,
practically important. First
groups
its
relation to group language
we have found
we define
a
this
dimension useful
common language
in placing the
Given
this definition of
common language and
common
all
itself or
the group language
language, we characterize the space of possible general solutions to
the translation problem in terms of the set theoretic relationships between the group languages
common language
the
any).
(if
If
we consider only "pure"
languages have the same relationship
si.x
The
possibilities.
language, then
languages,
i4)
it
to
first possibility is
may
be
the
In order for all groups to be able to
communicate using the common language, each group must either use the common language
be able to translate between the
of
schemes that seem
groups as a language that
for a set of
communicate with one another.
in the set use to
Even though there are other ways
(2) identical
the
common
that
(
1)
cases, that
there
i5)
cases where
language, then, as shown in Figure
is
no
common
with the group languages,
a subset of the group languages.
is
language.
If
all
the group
there are
3.1,
there
is
and
a
common
non-overlapping with the group
i3)
a superset of the group languages, or
(6)
partially
overlapping with the group languages
The
first
and we
four of these possibilities represent interesting practical solutions to our general problem,
will descri'oe
(2) identical
fifth
the
and
them
group languages,
si.xth possibilities
common language and
first
in the
external
this section,
common
using the terms
language, and
(41
(1)
internal
no
common
common
language.
language. The
and "hybrid" cases where some group languages have one relationship
other groups have another relationship can
four "pure" possibilities.
several other possibilities.
(3)
remainder of
We discuss these special cases briefly at
all
to
be analyzed in terms of the
the end of the section along with
m
(a)
No Common
C =
Languag'^
Gi:
Gi/
..
C:
A Group Language
A Common Language
,y
(b) Identical
Group Languages
:x\
C
=
Gi for
all
r
i.
II ?
(c)
External
Common Language
Gi
cOgi
=56
^mmmy\
/
\<
I.
<,;
(d)
Internal
Common Language
y
Gi^
CcGi
(
(e)
Superset
A-y-AvH^S
IIBiii
:DGi
i.'O
Intersection
C
.<::>::•:>:>••:
C
C
C
C
n
^
<3;
p
FIGURE
Cn
;\
:Sf
^:/y:-
C^i
•:
Gi
Gi
3.1
Gi
y
Possible Relations between
Group Language,
Common Language
and
No Common Language
3. 1
In a
scheme
of this type, there
common language
no
is
communicate. Instead, each pair of groups must be able
out of each other's languages. This
to
that
is
shared by groups that wish
have pairwise translations made into and
in a sense, the situation that actually prevails in the real
is,
of natural languages such as English,
to
French, and Japanese
In the
documents, this scheme implies that there are translating programs
for
world
world of computer-based
each pair of groups. The most
general form of such programs would involve "dictionaries" that provide translation rules for each
type from a source language into some type in the target language.
Heimbigner and McLeod's work (1985) provides an e.xample
information sharing
it
implemented by
Whenever
communicates
would use the dictionary
group.
to translate the
This translation could involve
following rule might be contained
and
of such a dictionary in the context of
letting each group
a message
is
is
in a
have a dictionary
for
each other
received from one of these groups, Lens
incoming message into a type understood by the receiving
much more than
just changing the type
example, include moving and transforming values within the
IF the message type
in
among autonomous databases
In Lens, this solution could be
group with which
Type and map derivations
fields of the
name
It
might, for
message. For instance, the
dictionary for messages that group B receives from Group A;
Information Request
the topic field contains Lens,
THEN convert the
and map
message type
to
Lens-related Request,
the fields as follows:
ChangeDateFormat (Action Deadline) > By When,
-
Subject
The
of
rule says that
its
the
topic field
name
'Topic'
is
if
-
>
Topic.
the received
message
is
of type Information Request from
Lens, then group B will treat
of the field 'Action Deadline' to 'By
Evaluation.
message
This scheme
fine-grained details.
condition
is
so that
is
it
will
appear
very Hexible
In this way. the
in
When' and change
in the
the
name
will be applied to the
it
value of Action Deadline
11
YY/MM/DD,
can provide case-specific translations
fully preserve
met:
the field 'Subject' to
of
format used by Group B, say
the sense that
scheme can
the value
as a message of type Lens-related Request; change
it
Furthermore, the function ChangeDateFormat
in the original
Group A and
in
meaning provided that the following
If there
is
a one-to-many mapping from a Group
X
other characteristics of documents of type
A
X.
type.
's
to
Group B
's
unambiguous
that allow
types, Yi, then there
should be
selection of the Yi to
which
X
should be mapped.
Group
In the above example, the rule that translates
Lens-related Request
is
map
only a partial
type are mapped to the type in B
properly, there has to be
A's Information
Request type
to
Group
B's
because only some of the messages of Information Request
To be able
more than one type
in
to translate all the
B
to
Information Request messages
which Information Request
is
mapped
Hence,
there has to be some characteristics in such messages that allow the translation scheme to determine
as which types in
B
the messages should be translated
In the example, the value
Lens
in the
Topic
served as such a characteristic.
field
This scheme also satisfies some of the objectives of autonomous evolution fairly well
is
not constrained to have any part of
is
not limited.
consensus
have
to
is
its
hierarchy shared with other groups, the expressive adequacy
Since setting up translation rules within a group does not affect any other group, no
required. However,
be propogated to
all
when
a group changes
could be quite costly to set up and maintain.
it
communicates.
If
to set
up a dictionary
type structure, the change will, in general,
A problem with
for
this
scheme, therefore,
Each group needs a dictionary
that
is
it
groups
for all the other
there are n groups communicating with one another, there would, in
general, need to be nln-l) dictionaries.
have
its
other groups with which this group communicates so that the other groups
can change their translation rules or dictionaries.
with which
Since the group
Also,
whenever
that group, and the
a
new group joins,
new group would have
all
the other groups would
to set
up dictionaries
for all
the existing groups.
Applicability.
and when the
efforts
or
if
may
This scheme would be appropriate when the groups already use their
efforts to build translation rules
be justified, for example,
if
the
and dictionaries
number
for the
own langauges
other groups are justified. Such
of groups that need to
communicate may be small
the groups themselves are stable and their type hierarchies do not change often.
3.2 Identical
Group Languages
In this scheme, the
required to use the
common language
is
the
same
same language and communicate through
include the adoption of global standards such as
libraries or
as any group language
Computing Reviews
Classification
Dewey Decimal
Scheme used
12
that language.
for
That
all
is,
Examples
Classification
groups are
of this
scheme
System among
categorizing literature
in
computer
science.
In the context of Lens, adopting this
scheme means that
the
all
^oups share
the
same
message type hierarchy.
In this scheme,
Eualuation.
on them are the same
for
consensus
for a
group
groups.
for all the
make
all
preserved because the types and the operations defined
fully
is
However, expressive adequacy
groups
quite large. Since
is
to
meaning
groups have
quite limited and the need
agree on changes, the only changes that are easy
to
will be those that do not affect
Even these changes may require
is
other groups or those that are valuable to
a significant overhead to
come
to
all
agreement. This scheme
also requires that all changes be propagated to all groups.
This scheme would be useful when there are not already well-established groups and
Applicability
when
the groups do not differ greatly in their needs. This
very restricted or
artificial, or
because there
is
may
may
be a small set of generic message types such as Request and
useful in almost all office environments
In this scheme, each group uses
is
a single
another by
Winograd
first
its
know
its
&
common language
translating
its
for
communication between groups
language into the
own language and how
to
its
common
own group language
Commitment
that are
translate
it
into
Lens were
to
use this scheme, there would be a
and out
how
to
translate
common
vice versa.
For example, suppose the
specialized
message types Action Request,
common
its
common
of the
to
group only needs
Another example can be found
implemented
in
any
locally
the outsiders iGuttag, 1977).
type hierarchy shared by
own
An example
language.
all
groups
in
the
types into the types in this hierarchy and
hierarchy consists of the root node. Message, and three
.Votice,
and Commitment, as
sends a message of type Information Request with "Lens"
translated into the type .\ction Request of the
A group communicates with
In this way. each
the idea of abstract data type that allows data structures to be
sense that each group would know
A
communicate. For
language, and then having the receiving
convenient ways as long as they maintain consistent interface
If
to
communication within the group, but
for
would be the adoption of an international language such as Esperanto
in
is
Flores, 1986).
own separate language
group translate the common language into
to
domain
Common Language
3.3 External
there
(
the
a 'vell-accepted theory about the domain, or simply
because the groups share the same goals and environments insofar as they need
example, there
when
occur, for instance,
common
13
in
hierarchy.
in
Fig 3
1
the topic field,
Then when group
it
might
Then group B might
first
be
interpret the
-
-.=-~fT'
zzz.
joc.
re rooe
=rc Si ^r^rr—rr
i_
tc_-r
^ s~-<T~i? ill ire ir>_js need lo srC'W r-nlj !jqw
Tj.-
it
.isiesi ^- i-ecii.on
—g-r-r
'.1
ze tC
-St
is~"-^g^
giTritKs'
ICC <^-g":^ge
:ij'c= fe-c *11
A-pp^^-zhujx-}
^'=:n:.
---
hizx F:
.
a$ icg
Ais«.
.
.
TrJ^
r-i.
scrj^rrza -»'':^c. zft
^rzjtr:- rr:^
:^_-..T it
dxs
J-
tea.-
ixrx^zjt -:
r-:te-srairjK >r
sro*.:? -ar^z'^a^e
are
>laK
am
rj;-TB^ :*
=cc<iild ise
~
:Ee gri'^j^s ha-re x- trarig*. so sorir
sr: _3s
:,-a
'
''
"^t
:
; :
ize srouo
:
i^i^
i^
.. : r.
already u=e ci5er*r.; -ar^_agei want VJ
s Ssr 'UKse rs^^rs -.r^t 3<kcj'1* propose a-
'j3- -T&tC
^"^ -^gs^
^
i'sr :::« t2S»s
cls^jT-i^c*
».-j=r-
be
L^rs.
I'jcsiri.
22s: oc sectlr^ j3 trani^:lvn r-il*s 3r clcticriaries for all the gnrups
gii
5ir "rra* r%a5cr'
^^ul
T-r«3'.ijd
oeinz
wr*re
r:e
-s=f-l «rjir: irjere
*e
r :r ^^tanple.
;:«--
-z lr.*Vi:res:
j-i
a -r:>>c
ar.T
sr's^
tcrr.n-jtr.
a .ar-auage
car. 'ikV-Jt a
Is r::ear.:r.gf^[
t,» -raeh cj>sr„t: -e x^sri or.
muzitt serve as the
1=
-y--.
trjs
-.
Tiigr*
user
.arz_a«e
-^J^rrjzr^^-.rjiL).
lar^-aae
5*jer.
rje
5'jcr.
:r-at 1=
a
a
Vj
is
too
Lar^.iafe or jse English
ejoress.ve eoo^igr., but
be
-a-':g-.;age
•.rj£
^li/ir.
of all group
ould expres
cLnur-lt Vj use
:f
al^
the
the •>'pes are too
Lar.g-uage a.tr»'igh rjot suitaoie as a
c
»•
-jr
=r^^—
tf
IT
5£T.
~>-
.-
^Ti— r^
?-n.T;
s-
see:
••i—^—rrrtT. .rtf^=~=.
"Tm
IfiS.
FIGlTx-
3
::
A r^~:2r2
hierarrhj i::-
its loci! j.;§:::e^ti
JI
,
We
all
have explored a scheme of this type
for
Lens called the Local Augmentation Scheme
groups might share the message types that are useful almost everywhere, such as Message, Action
Commitment, and
Request,
Notice.
Then each group
creating subtypes of these shared types (See Fig 3
Meeting Announcement, the type
type, say,
the
For instance,
common
not in the ancestor (such as
and put
in the
already in the
One way
of
common
this
A
augment the hierarchy by
sends
to
the message should also contain:
Of course,
common
(a)
is to
If this
the
is
if
group
A
sends a message of a type that
type
name
is
not in the
common
its
type in the message type
hierarchy shared by
sending group and
of the
Evaluation.
(b)
is
meaning
will be abstracted
properties.
many
of a type in the
common
away
to the
the nearest "ancestor" of
to the
degree that
hierarchy
its
its
meaning
is
ancestor in the
in expressive
common
such as Action Request or
group does not want
and use them
to
Commitment
common
shared.
the
If
its
hierarchy preserves
its
If not,
adequacy since each group can add as
The only constraint on
made
in
the
common
hierarchy defines speech act categories
as the immediate children of the root type. Message.
use this speech-act based categorization,
for internal
is
then
fully preserved
that groups cannot completely ignore distinctions
For example, suppose, as before, that the
other groups can
common language
types as desired to express distinctions that are important to them.
is
in all
extent that the type structure
hierarchy, then
This scheme allows a great latitude
expressive adequacy
groups,
all
hierarchy. In this way, receivers in the sending group can process the
This scheme preserves meaning
message
is
necessary.
have each message include
as being of the nearest "ancestor" type in the
it
in
Meeting Announcement (such as
message as being of exactly the type intended by the sender, and receivers
process
group B a message of
ancestor, N'otice, are preserved as they are: those that are
field
scheme
hierarchy used by the sending group
the message tj-pe in the
group
fields of the type
hierarchy, then no translation
implementing
When
2).
to locally
Meeting Time and Meeting Place) are treated as a part of the main text
beginning of the Text
common
allowed
automatically translated into the nearest ancestor
Those
hierarchy, in this case N'otice
Topic) that are also in the nearest
is
is
it
If
a
can add other children of Message
communication, but the group members may
still
receive messages from
other groups based on the speech act categorization
This scheme only requires consensus
among groups on what
the
common
only requires the propagation of changes in this
common
group need not be propagated
However, a change
to
other groups
hierarchy should be, and
it
hierarchy The changes made locally within a
in the
common
hierarchy can have
far-reaching implications. All the groups that have rules or subtypes of the modified type would have
to
make necessary adjustments
16
This scheme would be useful
Applicability
when
well-established and incompatible language groups,
groups, and
(3)
useful in the Lens
environment
to
have
a set of
there are important commonalities across
(2)
there are also significant local variations
already exist numerous
there do not
(I)
among
For instance,
groups.
common message
it
seems quite
types used by all groups (such as
Action Request, Commitment, and Meeting Announcement), and
to also let
own
The X.400 standards
for electronic
field for specifying a
message type,
message types
specialized
messaging are compatible with
and leave open the
for their
this
own unique
situations.
approach since they include a
possibility of different groups developing their
each group develop their
own conventions about how
this
field is used.
This approach also seems quite applicable
definition of
in
many
unchanging universal standards
is
other document interchange situations where the
For example,
difficult
been focused on developing Electronic Data Interchange standards
(such as invoices, purchase orders, etc
)
much
recent attention has
for a variety of
The general approach suggested here would be
generic versions of these forms that contained the fields and values used in almost
well as one or
more additional "miscellaneous"
companies are ensured a certain
version.
fields (like the
level of interoperability,
i
At the same time, some of the companies can
e
to
Text
included "hooks"
A
it
would be possible
for industry-specii^ic
to
field in
all
Lens).
to
develop
industries as
Then
all
the
the extent that they share the generic
locally
augment
accomodate their special needs by establishing conventions about how
Also, in this scheme,
business forms
to
this generic version to
use the miscellaneous
develop packaged software
for
EDI
fields.
applications that
and firm-specific routines.
similar approach might be used for interchanging formatted documents between difl"erent word
processing systems.
Current approaches
to
this
problem usually involve writing many pairwise
translation programs to convert between different formats, or trying to agree on a single standard that
is
expressive enough
to
represent
agreeing on a generic standard
all
for the
other formats
The approach described here would suggest
most basic kinds of documents and formatting, and then having
additional miscellaneous "fields" whose interpretation could be standardized by different subgroups.
Then, when group B's system reads a document from group
in its
usual
when
it
way and merely
.\. it
can display the standard information
use indications such as "The original document contained a figure here"
encounters non-standard formats
17
3.5 Other translation schemes
Other
set theoretic possibilities
There are two other
set theoretic possibilities listed
language and the group languages:
and
(2)
first
case
the
common language
amounts
to
(
1)
the
for the
common language
assuming that each group
is
groups translate
The difference
which
it
is
it
is
Thus, this case
all
all
probably no reason
for
of the group languages,
part of every group language as well as
similar to the
know how
to
The second
amounts
e
&
f).
The
to
all the
its
other
No Common Language scheme.
translate not only the languages with
common language
case,
where the common language partly
assuming that the common language consists of a
some language external
to all
the group languages. Thus,
Common Language and
the External
it
Common
cases.
have hybrid cases, where some groups have one relationship with the
also possible to
language and other groups have another relationship with the
groups might use only the
Languages
Internal
1
So we dismiss this case as a theoretical
such extra efforts
can be analyzed as a combination of the Internal
Language
is
own language and have
its
the group languages because they are all part of the
possiblity but a practically implausible one
overlaps with
a superset of all the group languages,
is
able to translate any other group's language into
that in this case, a group needs to
communicates but
In practice, there
own
into their
common
relationship between the
partly overlaps with all the group languages. (See Fig 3
For example, a group can always send a message in
own.
It is
above
common language
case), while other
Common Language
within their
case).
own groups
groups augment this language with their
case),
and
still
We
For example, some
(as in the Identical
own
other groups might use their
the groups but have translation tools into and out of the
Common Language
common language
common
local
own
language (as
local
common language
Group
in the
languages within
(as in the External
believe that all these hybrid cases can be analyzed as combinations of
the "pure" cases discussed above.
Sequential composition of translation schemes
The schemes discussed above can be used sequentially
and powerful schemes
Common Language
finally into
G4
of course, "lose
using
various ways to create even more complex
For example, Fig 3.3 shows an example of a group that uses language
communicating with another group that uses G4
Internal
in
message
is
first
scheme, then into G3 using an External
No Common Language scheme
something
Its
in the translation."
18
translated to
Common
G2
Gl
using an
Language, and then
Such complicated sequential translations may,
L.
U--'
—
C)
C)
explcitly agree on the types of messages used, because the types can be automatically determined by
the fields the messages have.
new message
individual, can create a
combination of
what
Although
Consensus
fields.
how
use and
fields to
this
Expressive adequacy
scheme
is
to
is
is
uncompromised because a group,
or Ln fact
new
type whenever needed by composing a message with a
not required on
what message types
to use,
but
is
it
any
required on
change them when needed.
attractive in
many
respects,
satisfied in addition to the implicit type hierarchy
it
would work only
if
assumption we made.
certain assumptions are
For example, the type
hierarchy should allow multiple parents, because the classification algorithm based on fields would
not determine a unique parent. Also, this scheme does not allow two different types to have the
Yet there are cases where this assumption
set of fields.
questions any further here
The point
of
mentioning
this
is
scheme
is
framework we provided above
on whether the set of fields shared are
scheme with the difTerent trandeofTs
3.6
Summary of translation schemes
3
1
summarizes the
discussions above are
objective.
in
the objectives as
difTerent classes of translation
summarized
In each column, the
in
For example, depending
we can have
different versions of
we have discussed above
schemes we have considered. The detailed
the table by rough qualitative rankings of the schemes on each
schemes that
general satisfy the objective better receive more stars
in
For example, the schemes that receive more stars in the column. Need
for
Nevertheless, the general
holds for these cases as well
identical, internal, or external,
this
Table
still
pursue these
will not
that often specialized schemes can
be designed that exploit particular features of the language used.
set-theoretic
We
too restrictive
same
Consensus, have
for
less
need
achieve
For
consensus.
As we have seen above, there are complex
tradeoffs
example, as highlighted
pairwise translation scheme used
language allows
more
full
in the table, the
expressive adequacy within groups and
diiTiculty in preserving
opposite virtue:
It
meaning across groups.
Internal
objectives
little
need
Common Language scheme
is
own
we are trying
for
L'sing Identical
preserves meaning quite well across groups, but
individual groups to adequately express their
The
among the
it
when
to
there
is
no
consensus, but
it
common
leads to
Group Languages has the
severely restricts the ability of
distinctions without consensus from other groups.
intermediate on
all
the dimensions, suggesting that
be a reasonable compromise in situations where no single objective
20
is
most important
it
may
TABLE 3.1
Evaluation of the different classes of translation schemes.
Meaning
Expressive
Preservation
Adequacy
Need
Need for
Consensus
for
Propagation
of Change
No Common
**
Language
Group
Languages
Identical
External
Common
Language
Internal
Common
Language
We
have shown how the classes of translation schemes we have discussed exhaustively partition the
space of possible solutions.
means
However, the examples of these schemes that we considered are by no
the only possible ones. For instance, there
of particular languages just as the Local
may
be other schemes that exploit special properties
Augmentation Scheme
exploits the Inheritance property of
type hierarchies.
4.
A composite scheme:
In this section,
we present
schemes discussed above.
defintions
to
is
types in
a composite
scheme that combines some
OMEGA or layers
in
V2
explicit.
The notion
PIE (Barber. 1982, Goldstein
are specializations of ("children"
Lens and then present
its
of)
A
view V2
is
message types
in
algorithm
listed above.
21
We
of the best features of the basic
uiews to modularize related type
This scheme exploits the notion of
a set of message types and their relations
in the context of
we
Shared Views
and make relations among such modules
viewpoints in
a uiew
Partially
&
of views used here
[
similar
Bobrow, 1981). For our purpose,
a child of \'i
V
is
We
if
some
of the
message
first illustrate this
then discuss how well
it
scheme
meets the objectives
4. 1
Illustration
Consider the following scenario. Suppose the
Message
All the groups can be
assumed
instance of the generic type Message
initial
view
V'q consists of
share this view because any message can be treated as an
to
Suppose Group A creates a new view,
specialized types of Message: Action Request, Notice, and
new view
declares a
V'l,
makes Vq
subclasses of the Message type.
Since
Vq
is
the only view that
its
Commitment
containing three
\'i,
In other words,
group
A
parent view, and within the new view creates the three
Now Group A
B has adopted
sends a message of type Action Request
so far, the type of the
message
is
unknown
to
group
B.
to B.
^T7-Hf
LElts-relatrd
Information
only the single message type,
FLequ-fit
FIGURE
When
4.1
Example
a message of
of
unknown
Views
type
is
received. Lens treats
type that both the sender and the receiver
message
itself as
and the most
apply
view
to the
it
we
will see
specific ancestor type
does not share, the group
in the
Suppose that B adopts
to
Identical
is still
may wish
unshared view
know about
In this case,
message are those that apply
from the type
known
below
to a
\'i in this case.
is
Message.
to the type
to (a)
This information can be obtained from the
the most specific view that both groups share,
Hence, the only operations that Lens
Message.
If
its
own
far as
necessary
the
that has
\'\
as
its
(b)
22
V'l
So the type T
The scheme becomes an instance
groups A and B are concerned
parent.
create translation rules
same view,
create a type called Information Request as a child of Action Request. So
\'2,
able to
views.
Then Group A and B share
is
is
a group receives messages from a
adopt the whole view, or
type within
both groups now and no translation
Group Languages scheme as
Vq
as an instance of the most specific ancestor
it
Now
of the
suppose B wants
B creates
a view of
is
its
to
own,
Suppose B sends a message of this newly created type, Information Request,
and adopts
V'l
But C does not
it.
object that the request
is
like
V9 because
it
about (eg Information Request)
messages by the subject domain of the requests
which there
classifies
Group
C.
Group C
likes
Action Request messages by the kind of
Instead,
C wants
to classify
So C creates a view V3, with
C
a subclass of Action Request, Lens-related Request.
is
to
\'l
also sets
Action Request
as
its
parent, in
up the following
translation rule:
If the
and
message
its
of type Information Request in view V'2,
topic field has as its value Lens,
then change
If
is
its
type to Lens-related Request in view V3.
the appropriate value
incoming message
is
in the topic field of the
incoming message,
this rule will fire
will be translated into a Lens-related Request.
Algorithm
4.2
First,
we make
a few assumptions:
which consists of the most general
1.
All groups share the default view, \'q.
2.
When
3.
Each message includes the following information:
a group adopts a view,
also adopts all the views
it
a.
the type of the message,
b.
a globally unique
C-
{(ti, V2),
specific to
.,
name
and one
simplicity,
to
its
make
ancestors.
v, is
t, is
the
name
which
of
this type
is
defined,
an ancestor view of
In these cases,
in
Fig 4
With these assumptions, the algorithm
1,
is
is
V'l),
only a single parent.
(Message, Vq)}.
the following.
M from Group G,
t
v,.
But
for
For example, in the
the message type information for type Information
be the type of M;
be the view to which
from most
the information has to be presented as a nested
we assume below that there
Request, V2), (Action Request,
let V
v ordered
the most specific ancestor type oft within
Request would consist of the following sequence of (type, view)
t
v.
sure that we deal with multiple parents appropriately.
view structure shown
Given a message
Message.
actually a bit more complex than presented here because multiple
parents are possible
list
of the view within
more general, and
is
which are
t>'pe,
t.
where
(tn, Vn)},
This information
let
and the
belongs;
23
pairs:
{(Information
if
Adopted'' v
.Adopte<i?(vj
;If the
.
then
proicess
retams TRUE
M as
if v is
ie as of t>-pe
is,
t
in
Wew
v
one of the views that has been adopted,
view v has been adopted, then there
FALSE
know what
no need to translate because we
is
if not.
it is.
else
if TranslatiorJlule?(ft,v)),
the t}-pe to which
i:
then apply the translation rules and process the message as of
has been translated
.Tran5!a:eRuIe?((t,v)) returns
.FALSE
;If
TRUE
there
if
is
a translati:-
-
"-
"or
the tv-pe
t
in
view
v,
if not.
the view v has not been adopted, then first check
if
a transla::.'.
.--.e
has been set up
for
;this tj-pe.
:If
make
there are more th^n one translation rules, then
multiple copies of the message, M,
:and apply each translation rule to a copy.
else
From
-ne most specific
searching and process
;So the
vj)
'ti,
M as of t^.-pe
is
else check the next general
message as
its
is
then stop
'ti,vi).
no translation rule
for the type
t.
nearest ancestor in the most specific view
an ancestor of v as well as which has been adopted.
;This algorithm
:Afler the
ti
view has not been adopted and there
;So in this case, process the
which
in the inheritance path, if Adopted?(vi),
is
guaranteed
to
terminate because of the assumption
message has been processed and presented
to the user,
an option
is
1
availabe for the
user to examine the view of the rr.essage. adopt the view, or set up translation rules for the
t.vpe of
the message
4.3 Evaluation
This view-based scheme
is
a composite of several of the basic schemes discussed in the previous
section.
As noted above, between groups
Identical
Group Language scheme
views,
it
serves as an instance of
translation methods
-
an instance of
When
rules
between
the groups share only some
up between the views not shared, the scheme acts as an
et
among different groups
a composite one,
its
while
still
providing a unifying framework.
evaluation with respect to the objectives listed in Section 2
is
will be best if the type belongs to the shared
if
is
To the extent that the groups have translation
depends much on which of the basic methods
preserved
view, the scheme
Common Language scheme The scheme thereby allows finer-grained adoption of
instance of Internal
is
same
No Common Language scheme
views and do not yet have translatic
As the scheme
that share the
adopted between two groups. Semantic preservation
view
its
semantics will be fully
the translations can be set up oetween the types in the two views
Otherwise, some of the
semantics will be
lost
between two
If it
does not, then
view.s without translation rules
24
But the semantics
will be
preser^ea ro zne exten: tna:
have
to
scheme
and
be compromised
is
that
it
Lf
:wo viewa have
select a translation
same
important enough
method
apprc'prlate for eacn case
enough
ixtth impc^rtan;
tite
two views.
If.
No
view
group's \iew;
If a
is
needed
group cho-oses net
is
t.~.e
its
can create
own view
its
useful because
acout t^pes at anot.ter
not
is
jut m.ay not set _p
m
is
required
to
tne
adopt any other
adopt an e.xistir^ \-iew V. the group can simply create a
The
irJteritance structure of the v:ewr.ase
i
e
.
new
the collectic^n of
allows groups to share agreements about t>"pes at one level and disagree
it
.evei.
.Another useful propert>- of this approacn
is tiiat
the viewbase
is
monotonic. that
is,
existing views are
never modified but only new -.iews are added. This means tnat ctner groups can co.ntinue
views even
own view
type into the nearest ancestor
scheme because no group
in this
:c
V as anotner child of Vs parent.
view?^
adequacy and
wil! be used.
cor^sensus betw-een groups
^new
It
direr=-t cases
on tne other hand. 5em.antic preser-.^ation
gnoup can create
to Justify the eiTorts, tne
for
if express:i.-=
then
niig:::
3u: a benec: of this
:-c;ecti-.-es
For example,
to justify tne efforts,
translation rules, in which case the default translation of
sr.ared
CT^^^55:^? aaecjacy
ir.cesior .-.ew
a group wanii :o share ihe 5a— e v-.ew with anotrier
up translation rules between
set
tr.e
allows a group to weigh the importance of the diftere^t
semantic preservation are
and
:r.e
when a group adopts
a
new
to
use old
view, and therefore com.munication can continue without
propagating any updates. However,
m order :o achieve greater preservation of ~ieaning. other groups
may wish
so that
to
know about new \iews
new views could be
In practice, the i.nform.ation abcut
posted in a global database that
'c^
they can adopt them or create translation rules for them
distributed in a variety of wa>-s such as
'a'
accessible to all the groups, b' broadcast to all the other grvurs. or
is
distributed in ph.vsical docum.entation such as newsletters
4.4 Applicability
Since the Partially Shared Views scneme combines m;any of the best features of the other schemes
potentially relevant in ver>this
one
many
situations
This scheme, therefore,
may
In a sense, each of the other schem.es
require
somewhat greater im^plementation
is
it is
a special case of
com.ple.xity.
but
it
allows a great flexibility in using a com.bination of different translation schemes for different
situations
For instance, this approach seems clearly desirarle
implementing a version of
other otilce svstems
it
there
for
the Lens system,
and we are currently
For sim.ilar reasons, tne approach also seems desirable
We described above m
section 3 4
how
for
m.any
the Local .Aucm-entation schem.e could
let
'
difTerent
EDI
-
agree:.- r
Partia.
or
:
-
.
;
ration systems achieve a certain degree of interoperability by
-:"
generic giobai staraards that
---ed Views
schenie seenis eve.- ~
many overlapping subgroups to
use their
subgroups, and also to sat up translati:
—
-
.
:^ppropriate in these situations because
'
own standards
' -
'
for
-
augmentation by different subgroups.
for
.ation with
_ -
:
members
it
The
allows
of the
same
commurucation with other subgroups, even when
there are no significant global standards.
5.
Conclusion
We began this paper with a description of the problem of translating between different document types
(such as messages and forms) in various kinds of office systems. At that level, our analysis suggested
several immediate
.
t',".
extensions to the !-.;;-—.ation Lens system and several intriguing
systems such as forms processing sv'Stems, document preparation systems,
possibilities for other office
and Electronic Data Interchange systems.
Most of our analysis of
this prcr
^^
-
.
A-ever.
was conducted
general than just documents and ofnce systems.
'-'
-:
have
for
an abstract
some of the issues that
suggest
new
seems
of agents, including both
arise in
in
more general contexts. For example, schemes
an
object oriented database
e g
be seen as translation schemes of the sort we have analyzed here
may
is
much more
to
apply
to
human and
intriguing to speculate about the implications this line of research might
It is
between different versions of objects
outlined
level that
In most cases, our analysis
among any kind
--ges in general and to communication
com.putational agents.
at
t>-pe
management schemes
corr.
—
,-
.^.
-
Skarra
The space
&
Zdonik, 1986) can
of solutions
we have
for object-oriented databases, or existing type
management schemes may suggest new translation schemes
general translation problem occur in
,
for translating
-:•
for office
systems. Other instances of this
A-een computational agents in a "blackboard
architecture" 'Nii, 1986) and in communication between heterogeneous databases in a networked
environment 'Heimbigner
&
McLeod,
1985).
We suspect that the approach we have outlined
eventually help illuminate and show connections
office syste
itions
among
here
may
these other applications, in addition to the
we have explored here
Acknowledgements
This research wais supported in part by
and Bankers Trust Company
Rosenblitt for their
We
Wang
Laboratories, Inc
thank Kevin Crowston,
comments that helped
identify
not have seen
26
.
Xerox Corporation, General Motors,
Kum-Yew
Lai,
Wendy MacKay, and David
and elucidate problems that we otherwise would
References
Barber, G. 1982. Office Semantics.. Ph.D. thesis. Massachusetts Institute of Technolog>-, 1982.
Barwise, J
&
Perr>-.
J 1983
Situations
and Attitudes MIT
R.J. & Schmolze, J.G. 1985.
system. Cognitive Science, 9(2). April-June
Brachman,
Fikes, R.
&
Kehler. T. 1985.
The
role of
An
Press.
over%-iew of the
Cambridge, MA.
KL-OXE
knowledge representation
171-216
frame-based representation in reasoning. In
ACM
Comm.
v.28(9) September. pp.90-i-920
Goldstein, IP. & Bobrow, D G. 1981. Layered networks as a tool for sofbvgu^ development. Proc. 7th
IntlConf. on Artificial Intelligence, 1981.
Guttag, J.V. 1977. Abstract data tN-pes and the development of data structures.
Comm. AC3f, v.2(X6)-
June 1977.pp.396-404
Heimbigner. D. & McLeod, D. 1985 A federated architecture
Trans. Office Information Systems 3(3) July, 1985.
JackendofT, R. 1983. Semantics
and Cognition MIT
Press.
for
information management. In
ACM
Cambridge,MA.
Lee, J. & Malone, T.W. 1988. How can groups communicate when they use different languages?
Translating between partially shared t>-pe hierarchies. Proc. 5th Conf. on Office Information System.
Mar. 1988. Palo Alto. CA.
Malone, T.W., Grant, K.R.. Turbak. F.A., Brobst. S.A., Cohen.
information-sharing systems InComm. .\CM v30(5>May pp 390-402
M.
Intelligent
19S7a.
Malone, T.W., Grant, K.R.. Lai, K.Y.. Rao. R.. & Rosenblitt. D. 1987b Semi-structured messages are
surprisingly useful for computer supported coordination. In Trans, on Office Information System
v.5(2) pp.li5-131
Nii, P.
1986 The blackboard model of problem solving The AI Magazine. Spring, 1986 pp. 38-53
Putnam. H. 1975 The meaning of 'meaning'. In Language. Mind, and Knowledge,
Gunderson. Minneapolis: Univ. of Minnesota Press.
&
Schmolze, J G.
Lipkis. T. 1983.
Proc. 6th IntlJoint
Skarra,
AH
&
Conf on
Classification in the
by K.
representation system.
Artificial Intelligence. 1983.
Zdonick, S.B. 1986.
database. In Proc.
KL-ONE knowledge
ed.
OOPSLA, September
The management
of changing t>-pes in
an object-oriented
1986. pp 483-495.
Stefik, M Bobrow, D G., Mittal. S. & Conway, L. 1983. Knowledge programming in LOOPS; Report
on an experimental course. The AI Magazine, pp. 3-13, Fall 1983
,
Tsichritzis. D.
Winograd, T
1982.
&
Form management.
Flores.
F
1986.
Design. Ablex. Norwood, N.J.
In
Comm. ACAf v. 25i71 July
pp.453-478
Understanding Computers and Cognition:
A New
Foundation for
fi5^f^
Q3h
Date Due
Lib-26-67
III
llh I,
^'''fS
„..„|
3
.aWoT'-i
'S 53^"
li
SSij
liiifiiiiiiwiBWi