Anaphora resolution (Embodiment)

From OpenCog
Jump to: navigation, search


Anaphoric pronouns are pronouns that refer to antecedents in the text. For instance in the sentences "I ate an apple yesterday. It was very good.", the pronoun "it" in the second sentence refers to the "apple" said in the first sentence.

The Opencog Language Comprehension uses anaphora resolution to improve the language understanding. The Hobbs algorithm, which is alared implemented in Relex, was used to anaphora resolution.

The process of anaphora resolution is as following:

  • All sentences listened by a pet are stored in a historic set in the MV Proxy. This historic may use the number of sentences or the timestamp to select the sentences that must be considered in anaphora resolution;
  • When sending the sentences said by a avatar to be processed by Relex, the historic is also sent;
  • If a pronoun is identified in one sentence, the listener historic is used and all words that fit the pronoun (according to Hobbs algorithm) are selected as anaphora candidates;
  • These candidates are classified according to a score computed by the Hobbs algorithm;
  • For each candidate, it is created an EvaluationLink which contains a ConceptNode called "anaphoric reference" and a WordInstanceNode that represents the candidate;
  • When the message containing the candidates (the same message that contains the content of the parsed sentences) arrived in the OAC, the Reference Resolution module uses these atoms to match the pronoun with one or more already known entities in the world.