Communicating meaning Sociocognitive semantic systems

1
Communicating meaningSocio-cognitive semantic
systems

• Robert McArthur
• CSIRO ICT CentreANU, Canberra

2
There is human communication
I Just met this guy, John Smith. He works for
Microsoft and is interested in Foo. Can we get a
contract for an evaluation licence to him
soon? Cheers, Peter Ill send him something
ASAP. Rupert Hi John, Peter said you are
interested in evaluating Foo. Can you please
sign and return the attached licence agreement
and we can send you the software. Rupert, Account
Manager Ive faxed the signed agreement to
you. John In the corridor, Peter to Rupert Hi
Rupert. Hows it going with John?
Conceptualising communication includes visualising
3
The problem
11Mbps?40bps
Must know
Become insularspecialised
Become insular
Too much information
!
Cant process it allepistemicallychallenged
Lack ofawareness

4
Solution
5
It all started one day a long time ago

• Monitor whats happening outside speciality
• Detect something interesting
• Flag potential importance
• One problem with many IT systems is that they
  abstract out the human. Dangerous here because
• How to know about whats happening? understand
  communication
• How to know whats outside vs inside? It
  changes humans know
• How to know what is already known? The person
  knows
• How to know what is interesting? The person
  (sometime) knows
• How to know when and what is the best way to flag?

6
Socio-cognitive theory

• Purposely aligned with the way people think cog
  and interact socio

To make sense of human communication using a
socio-cognitive framework, to reason about it,
and to make a person aware of something, the
computer system must know about what it is that
it is processing
It must compute with meanings not symbolic
surface structures
7
The whole picture
One person (or group)can only work on asmall
part of the bigpicture. This part is mine.
8
Goals and guiding principles

• Goal paradox of increasing informationwe are
  losing awareness
• for info processing technology to be more aligned
  with humans
• it needs to have some understanding of what it is
  that it is processing to compute with meanings
  not just symbolic surface structures
• these meanings need to be in accord with the
  meanings we harbor

Develop technology which enhances awareness, e.g.
pragmatic reasoning systems which draw
context-sensitive associations like those we
would make but dont, or cant, as we are
epistemically challenged
Utilise and adapt computational forms of
socio-cognitive semantic representation.
Semantics is a socio-cognitive issue - the
meanings, stored and manipulated by the system,
should accord with what we have in our heads
9
Cognition
Semantic space models
Philosophy
Geometry meaning
Experience management
Practical logics of cog systems
Sociology
Applied Logic
abduction
Social network analysis
Tacit/explicit K-
relevance
Perry semantics
Matching functions
POS
Term weighting
Language models
- Qualitative analysis - Experimentation -
Speculation
Information Science
10
Knowledge representation Gärdenfors cognitive
model
Cognitioncomputation symbol manipulationTuring
machine
symbolic
conceptual
Geometric representation
(Semantic space)
sub-conceptual
Connectionist representation
11
To do

• Take socio-cognitive theory from the bible
  Peter Gärdenfors book Conceptual Spaces which
  propounds a theory of geometric paradigm
• Take geometric-based implementations of cognitive
  practice on representing meaning HAL, LSA,
  COALS,
• Link the two together (not been done before)
• Test it all out on a variety of IT problems
  involving human communication (nada)

12
Practicalities

• create semantic space (creation)
• HAL, LSA, COALS (cog.sci)
• Augmented PoS, anaphora, index expressions
• muck-around with the space (normalisation)
• Lowe re-weighting, dimensional reduction
• get something out of it (associational inference)
• cosine, minkowski, info flow
• uncover tacit knowledge
• Inference within a conceptual space (semantic
  space) has a associational/analogical rather than
  deductive character

13
Anomaly detection projecting computational sense
of self

• Online community (email lists) with chronic
  disease
• Qualitative analysis when is an individual in
  transition?
• Extraordinariness turmoil axes are
• Fatigue and pain characterized via negative
  emotional words in utterances
• Decreased socialisation characterized (possibly)
  by kin words
• Ordinariness coping
• The experience of being in transition from
  Extraordinariness to Ordinariness (and sometimes
  back again) are non-linear processes that are
  sometimes cyclical, often convoluted and
  potentially recurring throughout peoples lives
  as changes both in life and illness create new
  challenges (Kralik 2002)
• Transition can be analyzed qualitatively in terms
  of self identity- computational sense of self

14
Computational sense of self
15
(No Transcript)
16
Example social network discovery

• Rob Rupert Peter
• found connections between people
• couldnt find them by link analysis
• also get connections people-information
• not possible by link analysis
• How does this work on social networks promote
  awareness?
• A means for carrying ad verecundiam (appeal to
  authority) Naomi found Rupert
• promoting introspection
• Warren Jones
• ask targeted questions, and internalise to be
  able to ask

17
Network of people guidebeam
Justin
Dave
Paul
Matt
David
David
Simon
Peter
Raymond
Ron
Robert
Yukio
Kim
guidebeam
Warwick
Nick
Rupert
Naomi
Kate
Nataliya
Wasim
Carol-Anne
Joe
John
Yvonne
Michael
18
Network of topics guidebeam
philosophy
tools
search
SIGIR
Government
ABC
CiTR
Peter
conference
PlanetMirror
Robert
capex
guidebeam
installation
management
Google
Rupert
Boeing
Yahoo
licence
sydney
documentation
CHIC
meeting
sqlator
19
Network of people guidebeam
Justin
Roberts
Justin
Peter
Roberts
Nick
all email
Robert
guidebeam
Nick
Nick
Peters
justin
Nick
Kate
Ruperts
Justin
Ruperts
Nick
Rupert
20
Human Communication

• Email, blogs, mailing lists etc. (ie. text) are
  the main artefacts of (current) online
  communication
• Strange language (not trad. letters, articles) ?
  dirty
• Not normal (computational) linguistics examples!
• But must be able to deal with all this
• Prevalence of English online (HCS implications
  cultural conditioning issues)
• Philosophical work on representing meaning
  (Gärdenfors, Grice, Perry, Woods, Freyd)
• Practical tests on representing meaning
  (cog.psych)

21
Human Communication

• Not so interested in the boundarieswhat is
  communication as much as what to do with/about
  it (assumed that it occurs!)
• (from Cliff) Mostly concentrating on messages
  passing forth (and text at that), but acknowledge
  the possible primacy of other forms. Its just
  that in computer settings were so far from
  getting the message that to work on the dangerous
  edge (people) is not terribly acceptable
• CogSci is (arguably) a mix between psych and
  ITWhere is the mix between cog.sci and
  sociology (communication in/among groups of
  people)?

22
Communicating meaningSocio-cognitive semantic
systems

• Robert McArthur
• CSIRO ICT CentreANU, Canberra