P1247176256CuYjP

1
Agents Communication and Interaction
Students Simon Coffey, Hywel Dunn-Davies,
Jaspreet Shaheed Supervisors Keith Clark, Jim
Cunningham
Lexical Descriptions and Agent Models
A Diagrammatic Formalism for the Representation
of Agent Interaction Protocols
Jaspreet Shaheed and Jim Cunningham
Understanding natural language (the language with
we speak to each other) is regarded by many as an
AIhard problem and core to the question of
intelligence itself. Our work focuses on
understanding verbs, which are often the locus of
meaning in sentences- in John gave Mary the
vase", John', Mary' and the vase' could be
considered as participants in the giving
event'. Our work looks at two different
contemporary perspectives to understanding verbs.
The first is linguistically motivated, semantic
roles, as employed in electronic lexical
resources such as FrameNet and VerbNet. The
second is an ontological perspective, in which
the type' of the event' denoted by the verb
(Physical', Social, etc.) is considered. We
are interested in whether these perspectives can
aid us in an agent-oriented' understanding of
language, in which the focus is on the mental
activities of participants described.
Hywel Dunn-Davies and Jim Cunningham

• Motivation Common Interaction Protocols in Open
  Agent Environments
• Interaction Protocols provide a context for agent
  communications, ensuring that they are
  interpreted in a consistent manner
• Diagrammatic formalisms for representing
  Interaction Protocols have proven to be most
  popular among protocol designers. However,
  current diagrammatic methodologies, such as Agent
  UML, are semi-formal in nature, leading to
  protocol definitions that can be ambiguous,
  resulting in misunderstandings between agents,
  which compromises the usefulness of the protocol
• We need a formal diagrammatic language for
  describing Interaction Protocols, so that
  executable protocol threads can be implemented,
  and properties of protocols can be verified,
  automatically

• Electronic Lexical Resources
• There are a number of electronic lexical
  resources, FrameNet, VerbNet, The Praguian
  Treebank, PropBank which are concerned with how
  verbs may be represented
• FrameNet is based upon Fillmore's theory of Frame
  semantics in which understanding a word or a
  concept such as sell' means understanding
  related concepts, such as commercial transfer',
  seller', buyer', etc. In FrameNet, Frames
  correspond to events or concepts such as
  revenge'. The description of a frame provides a
  definition of the concept, the roles of
  participants in it, the lexical units that
  pertain to the concept and finally relations to
  other frames-- such as inheritance, or being
  temporally preceded by', etc
• All of the electronic lexical resources mentioned
  above employ semantic roles which to greater or
  lesser extents are inspired by thematic roles

• Propositional Statecharts
• Propositional Statecharts are a variant of UML
  statecharts intended to provide formal
  descriptions of Interaction Protocols

• Thematic Roles
• Thematic roles exist at the syntax/semantics
  interface. They provide labels to the noun
  phrases (NPs) that accompany the main' verb of a
  sentence, often analogous to participants in an
  occurrence. Whilst many different taxonomies
  exist, some commonly employed thematic roles
  include
• Agent Deliberate undertaker of the action, John
  cut the cake with the knife
• Theme/Patient Affected by the action, John cut
  the cake with the knife".
• Instrument Used in performing the action, John
  cut the cake with the knife"
• Criticism. David Dowty role types are simply
  not discrete categories at all, but rather
  cluster concepts'. Dowty suggested instead that
  participants be described by their semantic
  entailments directly. In Dowtys theory of
  thematic proto-roles, participants are described
  by whether they volitionally engage in an event,
  are aware of it, affect or are affected, move,
  etc.
• Dowtys theory provides a much better starting
  point for an agent-oriented understanding of
  language, but an investigation into it identified
  that it left too much unspecified (what does
  affecting mean?)

• They are so called because the states are taken
  to denote logical propositions
• Propositional statecharts can provide joint
  protocol representations, treating the interation
  as a single sequential process (as above), or
  agent centred representations, which give an
  explicit representation of the messages sent and
  received by the agents, and include an explicit
  representation of the communication medium (shown
  below)

• A mechanism for modular decomposition of protocol
  representations has also been added, allowing the
  formalism to describe recursively defined
  protocols
• We intend to provide a formal semantics for the
  formalism allowing us to
• Verify properties of protocols subject to
  different communication media
• Implement protocols expressed using the formalism
  automatically as stand alone protocol threads

• Conclusion and Future Work
• Existing lexical descriptions do not describe
  well the mental activities of their participants
  (what is the 'Recipient', Mary, thinking in the
  sentences Mary accepted a vase from John and
  John left Mary a vase?)
• Can we use an agent model as a starting point in
  helping us to design a lexical description?

Metro - A Multi-Threaded Teleo-Reactive Robot
Architecture
Simon Coffey and Keith Clark
To date, behavioural approaches have dominated
research in single-robot systems, while
cognitive architectures have largely failed to
make a significant impact. However, behavioural
attempts at robot collaboration are presently
somewhat limited in scope (although interesting
projects certainly exist). By contrast,
Multi-Agent research has demonstrated the
benefits of a cognitive architecture for
cooperative software, offering a wide variety of
interaction schemes and techniques. We propose a
hybrid architecture that seeks to enable the
application of these agent techniques without
compromising the behavioural techniques that have
met with so much success.

• Project Goals
• To establish that there is some level of
  behaviour at which symbolic reasoning is
  beneficial
• To test the applicability of symbolic reasoning
  to the field of multi-robot coordination
• To examine the challenges of integrating such
  reasoning successfully with behavioural
  techniques
• To demonstrate further that this integration does
  not restrict the application of behavioural
  techniques where appropriate.