Development of Emergent Leadership Measurement: Implications for Human-Machine Teams

Abstract

Emergent leadership refers to the dynamic by which, when there is no appointed leader in a group, one or more members assume leadership behaviors. Understanding emergent leadership in task-oriented human-machine teams is critical to optimize the role and input of machine agents. We find, however, a dearth of measures of emergent leadership to guide the development of machine agents. Here we describe the initial development of peer-report and natural language processing (NLP) -derived measurement techniques for indexing emergent leadership in a team context, rooted in the leaderplex model (Denison et al. 1995; Quinn 1984); we take a behavioral approach to indexing emergent leadership which emphasizes the diverse functions of leaders in the team context. We describe initial evidence of validity, areas of further exploration, and implications for human-machine teams. Overall, we find good concordance between peer-report measures of leadership behaviors and peer-report identification of emergent leaders, as well as with initial NLP behavioral marker extractions. Our mixed-method approach presents a first step in developing language-derived computational methods to enhance machine agent artificial social intelligence and theory of mind, ultimately improving their effectiveness in human-machine teams.

Appendix

Vars: "org/clulab/quickstart/grammars/ASIST/vars.yml"

rules:

  - name: "Entity"

    label: Entity

    priority: 1

    type: token

    pattern: |

      [entity="PERSON"]

      |

      [tag=/^N|PR*/ & !tag=/^V/]

  - name: "Action"

    label: Action

    priority: 2

    pattern: |

      trigger = [tag=/^V/]

      agent: Entity? = /${agents}/

      theme: Entity? = /${objects}/

  - name: "Medic"

    label: Roles

    priority: 1

    type: token

    pattern: |

      /(?i)medic/

      |

      [lemma=/medic|medical/]

  - name: "Hammer"

    label: Roles

    priority: 1

    type: token

    pattern: |

      /(!?)hammer/

  - name: "Specialist"

    label: Roles

    priority: 1

    type: token

    pattern: |

      /(!?)specialist/

  - name: "Search"

    label: Roles

    priority: 1

    type: token

    pattern: |

      /(!?)search/

  - name: "Obstacle"

    label: stuck

    priority: 1

    type: token

    pattern: |

      [lemma=/immobilize|frozen|stuck/]

      |

      [lemma=can] [lemma=not] [lemma=move]

      - name: "Approval"

    label: Approval

    priority: 2

    example: Sure, Bob!

    pattern: |

      trigger = [lemma=/yes|ok|okay|sure|alright|good|excellent|great|terrific|copy/]

      |

      /(?i)Sounds/ /good/

      |

      /(?i)I/ /agree/

      agent: Entity? = /${agents}/

      - name: "Responsibility"

    label: Responsibility

    priority: 2

    type: token

    example: I can do that.

    pattern: |

      [lemma=/(?i)I\’ll/]

      |

      /(?i)Can/ /do/

      |

      /(?i)Take/ /care/

      |

      /(?i)Let/ /me/

      |

      /(?i)I/ /will/

        # Planning Rules #

  #############################

  - name: "ConditionalIF"

    label: ContingentPlan

    priority: 3

    pattern: |

      trigger = [lemma=if]

      condition: Action = <mark

      solution: Action = <mark <advcl_if

  - name: "PlanTrig"

    label: Planning

    priority: 1

    type: token

    pattern: |

      /(?i)should|need/

      |

      [lemma=/plan|goal|bring|come|gather|get|be/]

  - name: "DeliberateFut"

    label: DeliberatePlan

    priority: 3

    pattern: |

      trigger = (?<= [tag=VBP & lemma=be]) [tag=VBG]

      # the first thing in the parantheses is a look behind so it’s going to look at everything that came before.

      # both inside the square brackets are constraints applied to a single token. A token comes before that is the tag restraints

      # and the verb "to be"

      # the second is capturing the gerund "am going"

      |

      (?<= [tag=MD & lemma=will]) [tag=VB]

      # to make this useful we need to know who is doing the planning.

      agent: Entity = >/${agents}/

      # agent is the name of the argument while Entity is the type.

      # the pattern is right hand side of the colon and is applied to the sentence.

      # from the trigger, follow an outgoing agent and it would need to be in previously found entity.

      theme: Param? = >/${objects}/

  - name: "HelpRequests"

    label: coordination

    example: Can someone help me?

    pattern: |

      trigger = /(?i)could|would|can/

      actor: Entity = <aux >/${agents}/

      # person doing the helping (someone)

      request: Action = <aux

      # not about linear order, we are traversing against an incoming aux to land on the end of the aux (from can to help)

      recipient: Entity? = <aux dobj

      # now, we’re traversing against an incoming aux to land on the dobj (from help to me).

  # - name: "Response"

  # label: coordination

  # example: I can help you.

  # type: token

  # pattern: |

  # /(?i)could|would|can/ [tag=/PRP|N*/]

  - name: "Unless"

    label: coordination

    type: token

    pattern: |

      /(?i)unless/ [tag=/PRP|N*/]

        # My Rule #

  # 5. Taking responsibility: Number of times participant takes responsibility #

  # (keywords: I can do that, I’ll take care of it, let me). #