AMI Meeting Multi-Tier Annotation

# AMI Meeting Multi-Tier Annotation Configuration
# Based on Carletta et al., MLMI 2005
# Paper: https://link.springer.com/chapter/10.1007/11677482_3
# Task: ELAN-style multi-tier annotation of multi-party meeting recordings

annotation_task_name: "AMI Meeting Multi-Tier Annotation"
task_dir: "."

# Data configuration
data_files:
  - sample-data.json
item_properties:
  id_key: "id"
  text_key: "video_url"

# Output
output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

# Annotation schemes - ELAN-style parallel tiers aligned to the video timeline
annotation_schemes:
  # Tier 1: Speaker turn segmentation
  - name: "speaker_turn_tier"
    description: |
      Segment the meeting timeline by speaker turns. Mark who is speaking at
      each point, including overlapping speech. Each segment should capture
      a continuous turn by one speaker or an overlap between multiple speakers.
    annotation_type: "video_annotation"
    mode: "segment"
    labels:
      - name: "speaker-A"
        color: "#3B82F6"
        tooltip: "Participant A is the primary speaker"
      - name: "speaker-B"
        color: "#EF4444"
        tooltip: "Participant B is the primary speaker"
      - name: "speaker-C"
        color: "#10B981"
        tooltip: "Participant C is the primary speaker"
      - name: "speaker-D"
        color: "#F59E0B"
        tooltip: "Participant D is the primary speaker"
      - name: "overlap"
        color: "#8B5CF6"
        tooltip: "Two or more participants speaking simultaneously"
    show_timecode: true
    video_fps: 25

  # Tier 2: Head gesture segmentation
  - name: "head_gesture_tier"
    description: |
      Annotate visible head gestures of the currently active speaker or the
      participant being focused on. Mark the type and duration of each
      distinct head movement.
    annotation_type: "video_annotation"
    mode: "segment"
    labels:
      - name: "nod"
        color: "#22C55E"
        tooltip: "Vertical head nod (agreement, acknowledgment, backchannel)"
      - name: "shake"
        color: "#EF4444"
        tooltip: "Horizontal head shake (disagreement, negation)"
      - name: "tilt"
        color: "#A855F7"
        tooltip: "Lateral head tilt (thought, consideration, uncertainty)"
      - name: "turn"
        color: "#F97316"
        tooltip: "Head turn toward a specific person or object"
      - name: "neutral"
        color: "#9CA3AF"
        tooltip: "No notable head movement; neutral or still position"
    show_timecode: true
    video_fps: 25

  # Tier 3: Focus of attention tracking
  - name: "focus_of_attention_tier"
    description: |
      Track where the active speaker or focal participant is directing their
      visual attention. Mark the target of gaze or visual focus at each
      point in time.
    annotation_type: "video_annotation"
    mode: "segment"
    labels:
      - name: "whiteboard"
        color: "#06B6D4"
        tooltip: "Looking at the whiteboard"
      - name: "slides"
        color: "#84CC16"
        tooltip: "Looking at the projected slides or screen"
      - name: "speaker-A"
        color: "#3B82F6"
        tooltip: "Looking at participant A"
      - name: "speaker-B"
        color: "#EF4444"
        tooltip: "Looking at participant B"
      - name: "speaker-C"
        color: "#10B981"
        tooltip: "Looking at participant C"
      - name: "speaker-D"
        color: "#F59E0B"
        tooltip: "Looking at participant D"
      - name: "notes"
        color: "#6366F1"
        tooltip: "Looking down at personal notes or documents"
      - name: "table"
        color: "#78716C"
        tooltip: "Looking at the table or objects on it"
    show_timecode: true
    video_fps: 25

  # Tier 4: Dialogue act classification
  - name: "dialogue_act"
    description: "Classify the dialogue act type of the current speaker turn or utterance."
    annotation_type: radio
    labels:
      - "inform"
      - "suggest"
      - "assess"
      - "comment"
      - "elicit-inform"
      - "elicit-suggest"
      - "elicit-assessment"
      - "backchannel"
      - "stall"
      - "fragment"
    keyboard_shortcuts:
      inform: "1"
      suggest: "2"
      assess: "3"
      comment: "4"
      backchannel: "5"

  # Tier 5: Topic segment classification
  - name: "topic_segment"
    description: "Classify the meeting topic or phase that this segment belongs to."
    annotation_type: radio
    labels:
      - "opening"
      - "agenda"
      - "design-discussion"
      - "budget"
      - "action-items"
      - "closing"
      - "off-topic"
    keyboard_shortcuts:
      opening: "q"
      agenda: "w"
      design-discussion: "e"
      budget: "r"
      action-items: "t"
      closing: "y"
      off-topic: "u"

# HTML layout
html_layout: |
  <div style="max-width: 900px; margin: 0 auto;">
    <h3 style="margin-bottom: 8px;">AMI Meeting: Multi-Tier Meeting Annotation</h3>
    <p style="color: #666; font-size: 14px; margin-bottom: 16px;">
      Annotate multi-party meeting recordings across parallel tiers for speaker turns,
      head gestures, visual attention, dialogue acts, and topic segments.
    </p>
    <div style="text-align: center; margin-bottom: 20px;">
      <video controls width="720" style="max-width: 100%; border-radius: 8px; border: 1px solid #ddd;">
        <source src="{{video_url}}" type="video/mp4">
        Your browser does not support video playback.
      </video>
    </div>
    <div style="background: #f8f9fa; padding: 12px; border-radius: 6px; margin-bottom: 16px; font-size: 13px;">
      <strong>Multi-Tier Instructions:</strong> Annotate the meeting across five parallel tiers:
      speaker turns, head gestures, focus of attention, dialogue acts, and topic segments.
      Use the video controls to navigate frame by frame for precise boundary placement.
    </div>
  </div>

# User configuration
allow_all_users: true

# Task assignment
instances_per_annotator: 30
annotation_per_instance: 2

# Instructions
annotation_instructions: |
  ## AMI Meeting Multi-Tier Annotation

  This task uses ELAN-style multi-tier annotation for multi-party meeting
  recordings from the AMI Meeting Corpus.

  ### Tier 1: Speaker Turn Segmentation
  - Segment the meeting timeline by who is speaking:
    - **Speaker A/B/C/D**: The identified participant holds the floor
    - **Overlap**: Two or more participants speaking simultaneously
  - Mark clean turn boundaries at the start and end of each contribution
  - Short backchannels (e.g., "mm-hmm") during another speaker's turn
    should be marked as overlap if they are audible

  ### Tier 2: Head Gesture Annotation
  - Mark visible head gestures of the focal participant:
    - **Nod**: Vertical movement (agreement, acknowledgment)
    - **Shake**: Horizontal movement (disagreement, negation)
    - **Tilt**: Lateral tilt (consideration, uncertainty)
    - **Turn**: Deliberate head turn toward a person or object
    - **Neutral**: No notable head movement

  ### Tier 3: Focus of Attention
  - Track the visual attention target of the focal participant:
    - **Whiteboard/Slides**: Looking at shared visual resources
    - **Speaker A-D**: Looking at a specific participant
    - **Notes**: Looking at personal notes or documents
    - **Table**: Looking at the table or objects on it
  - Gaze shifts should be marked at the moment of transition

  ### Tier 4: Dialogue Act Classification
  - Classify each speaker turn by its communicative function:
    - **Inform**: Providing information or facts
    - **Suggest**: Making a suggestion or proposal
    - **Assess**: Evaluating or judging something
    - **Comment**: Personal reaction or remark
    - **Elicit-inform/suggest/assessment**: Requesting information, suggestions, or opinions
    - **Backchannel**: Minimal response showing attention (mm-hmm, yeah, ok)
    - **Stall**: Hesitation or time-buying (well, so, let me think)
    - **Fragment**: Incomplete or abandoned utterance

  ### Tier 5: Topic Segment
  - Classify the meeting phase or topic:
    - **Opening**: Greetings and meeting start
    - **Agenda**: Setting or reviewing the agenda
    - **Design discussion**: Core design-related conversation
    - **Budget**: Budget or resource-related discussion
    - **Action items**: Assigning tasks and next steps
    - **Closing**: Wrap-up and meeting end
    - **Off-topic**: Social chat or tangential discussion

  ### Quality Notes
  - Focus on one participant at a time for head gesture and attention tiers
  - Speaker turns should have no gaps (silence = previous speaker's turn end)
  - Dialogue acts apply per utterance within a turn, not per entire turn