Unification

An example lexical entry

AVOCADO = {'meaning': {'shape': 'pear',
                       'length': 15,
                       'width': 8,
                       'surface': {'reflectivity': 'high',
                                   'thickness': 'thin',
                                   'texture': 'hard',
                                   'color': 'dark-green',
                                   'edibility': 'no'},
                       'interior': {'color': 'green',
                                    'texture': 'mushy',
                                    'edibility': 'yes'}}
           }

An example utterance instance to match against lexical entries

UTT1 = {'meaning': {'shape': 'pear',
                    'surface': {'reflectivity': 'high',
                                'texture': 'hard',
                                'color': 'dark-green'},
                    'interior': {'color': 'green',
                                 'texture': 'mushy'}},
        'speaker': 'ego',
        'hearer': 'mabel'}

Unification: a procedure that takes two feature structures (Python dicts for us) and either a merger of the two if they match or False if they don't
A first cut at a unification function: unify1.py

Variables

The need to represent sameness within an utterance: variables

Two deictic lexical entries

ME = {'form': 'me',
      'speaker': '?S',
      'meaning': '?S'}

YOU_SING = {'form': 'you',
            'hearer': '?H',
            'meaning': '?H'}

An extended unification function that handles variables: unify2.py

A more complex example: representing the meaning of an an event type

GIVE = {'form': 'give',
        'meaning':
        # Agent gives Patient to Recipient
        {'agent': '?A',
         'patient': '?P',
         'recipient': '?R',
        # Precondition: Agent owns Patient
         'prec': {'owner': '?A',
                  'object': '?P'},
        # Effect: Recipient owns Patient
         'effect': {'owner': '?R',
                    'object': '?P'}}}

To include all variable bindings in feature structure that's returned, we need to replace all variables by their bindings
An extended unification function that performs this substitution step: unify3.py