Only released in EOL distros:   fuerte

What are Semantic Frames?

In RoboFrameNet, a semantic frame describes a scene being played out. It consists of an overall concept, the frame name, as well as related actors and objects, its frame elements.

Creating, Adding, and Registering to Semantic Frames

The natural language processing capabilities of RoboFrameNet can be extended by creating new semantic frames, adding them to the frame_registrar, and registering your own code to semantic frames.

• To create new semantic frames for RoboFrameNet, follow the guide below.

• To add the created semantic frames, see rfnserver.

• To register to a frame, see rfnserver.

Creating Semantic Frames

Overview and Examples

Files are in standard yaml notation. The layout is most explained by example; individual elements will be described below.

This is a minimal example, which describes a scene where the robot emits a mooing sound:

name: mooing
description: The robot moos.

This is the first "maximal" example, which describes a robot turning. Additionally, it describes its relation to two semantic frames, which specifically describe the robot moving forwards or backwards:

name: moving
description: Robot moves in a direction (forwards or backwards) for distance distance_unit(s).  (eg, 3 feet)
frame_elements:
  - name: direction
    is_core: True
  - name: distance
    is_core: True
  - name: distance_unit
    is_core: True
children:
  - name: moving_forward
    frame_element_relations:
      - parent_name: distance
        child_name: distance
      - parent_name: distance_unit
        child_name: distance_unit
      - parent_name: direction
        child_value: [forward, forwards]
  - name: moving_backward
    frame_element_relations:
      - parent_name: distance
        child_name: distance
      - parent_name: distance_unit
        child_name: distance_unit
      - parent_name: direction
        child_value: [backward, backwards]

This is the second "maximal" example, which describes the robot retrieving an object. Additionally, it describes its relation to a series of other semantic frames, whose sequence performs the same task as the master:

name: retrieving
description: Agent goes to object, picks up object, and gives it to the operator.
frame_elements:
  - name: object
    is_core: True
children:
  # This is a sequence of frames
  - - name: navigating_to_object
      frame_element_relations:
        - parent_name: object
          child_name: object
    - name: picking_up
      frame_element_relations:
        - parent_name: object
          child_name: object
    - name: navigating_to_master

See also some corresponding lexical units in the semantic_framer, which describe mappings for the first two examples.

Notation

Below, the phrase "one or more" for an entry signifies that either a single instance can be entered, or a list of multiple instances can be entered. For instance, with verb, "one or more verbs" means all the following entries are legal:

verb: moo
verb: [moo]
verb: [moo, speak]

Required Values

Each semantic frame must specify the following values:

• name: The name of the semantic frame.

• description: A description of the semantic frame's scene.

Optional Values

Each semantic frame may specify the following values:

frame_elements

frame_elements is a list of frame elements. Each frame element describes an object or actor in the frame's scene and must have two values:

• name: The name of the frame element.

• is_core: A boolean describing whether or not the frame element is core -- that is, whether or not a value needs to be given to the frame element for a mapping to the semantic frame to be valid. (If no value is specified, the argument of the frame element will be "" (the blank string).)

children

children is a list of relation sequences from the parent frame to child frames. A relation sequence describes a set of frames which -- when executed serially -- complete an action equivalent to the parent frame. Each relation sequence contains one or more relations.

For each entry in a relation sequence, one value is required:

• name: The name of the child frame.

Additionally, one value is optional:

• frame_elements_relations: A list of mappings between the current frame's frame elements and the child frame's frame elements.

Each relation in frame_element_relations must have exactly one of the following values:

• parent_name: The name of the frame element in the current (parent) frame.

• parent_value: An argument to fill into the child's frame element. (See below.)

Additionally, each frame element relation must have exactly one of the following values:

• child_name: The name of the frame element in the child frame.

• child_value: One or more values which the parent's frame element's argument must have to allow the child's semantic frame to be filled.