Method and apparatus for evaluating natural language input to identify actions and landmarks

An apparatus configured to evaluate natural language input to identify actions and landmarks, the apparatus comprising processing circuitry and at least one memory including computer program code instructions, the computer program code instructions configured to, when executed by the processing circuitry, cause the apparatus to:

• encode one or more word representations in context based upon one or more other word representations in at least one of a forward direction or a backward direction within a sentence of the natural language input;
• generate an action context vector defining actions that occur during each of a plurality of time steps;
• generate a state context vector defining one or more landmarks associated with each of the plurality of time steps;
• predict the state for a respective time step by defining the one or more landmarks associated with the respective time step, wherein the computer program code instructions configured to generate the state context vector and predict the state are based upon a combination of a result of encoding the one or more word representations in at least one of the forward direction or the backward direction within the sentence, the one or more word representations and the state predicted during a prior time step;
• and
• predict an action for the respective time step based on the action context vector and the state context vector.

An apparatus configured to evaluate natural language input to generate a landmark graph, the apparatus comprising processing circuitry and at least one memory including computer program code instructions, the computer program code instructions configured to, when executed by the processing circuitry, cause the apparatus to:

• receive the natural language input comprising one or more sentences;
• encode one or more word representations in context based upon one or more other word representations in at least one of a forward direction or a backward direction within a sentence of the one or more sentences of the natural language input;
• generate an action context vector and a state context vector defining actions that occur during each of a plurality of time steps and one or more landmarks associated with each of the plurality of time steps, respectively;
• predict an action and a state for a respective time step, wherein the action is predicted based upon the action context vector and the state context vector and the state is predicted based upon a combination of a result of encoding the one or more word representations in at least one of the forward direction or the backward direction within the sentence, the one or more word representations and the state predicted during a prior time step;
• and
• generate the landmark graph from the natural language input based upon the actions and the states that are predicted for respective time steps of the sentence.