eXtended WordNet  
The University of Texas at Dallas
home  |  about xwn  |  downloads  |  papers  |  people  |  news  |  related links  |  contacts  

Logic Form Transformation

A logic form is an intermediary step between the syntactic parse and the deep semantic form.

The logic form transformation takes as input the parse-tree of a sentence, and assigns predicates and arguments to the words.

The LFT codifies the syntactic subjects, syntactic objects, prepositional attachments, complex nominals, and modifiers.

The Logic Form of a gloss is represented in the following manner for each part-of-speech:

  • 'noun': For each synset, the first word representing the synset is taken and assigned the argument 'x1'. In the gloss of this synset, the argument 'x1' refers to the same entity as the first word in the synset. For eg,

    plant:NN(x1) -> living:JJ(x1) organism:NN(x1) lack:VB(e1, x1, x2) power:NN(x2) of:IN(x2, x3) locomotion:NN(x3)

    Here, plant(x1) on the LHS and living(x1) organism(x1) on the RHS refer to the same entity.

  • 'verb': For each synset, the first word representing the synset is taken and assigned the argument 'e1', its subject is assigned the arguement 'x1' and the object 'x2'. Please note that the object argument can be a dummy since the verb could be an intransitive one. For eg,

    recognize:VB(e1, x1, x2) -> show:VB(e1, x1, x5) approval:NN(x3) or:CC(x5, x3, x4) appreciation:NN(x4) of:IN(x5, x2)

    Here, if 'x1' (subject) recognizes ('e1' action) 'x2' (object) on the LHS, then 'x1' shows ('e1' action) approval or appreciation ('x5') of 'x2' on the RHS. There could be intransitive verbs, for instance,

    tremble:VB(e1, x1, x2) -> move:VB(e1, x1, x4) with:IN(e1, x3) tremor:NN(x3)

    Here, if 'x1' trembles on the LHS, then 'x1' moves with tremor 'x3' on the RHS. Note that 'x2' (object of tremble) and 'x4' (object of move) are dummies.

  • 'adjective': For each synset, the first word representing the synset is taken and assigned the argument 'x1'. In the gloss of the synset on the RHS, the argument 'x1' refers to the same entity as the one described by the first word in the synset. For eg,

    ascetic:JJ(x1) -> practice:VB(e1, x1, x2) great:JJ(x2) self-denial:NN(x2)

    Here, if 'x1' is ascetic on the LHS, then 'x1' (subject) practices ('e1' action) great self-denial ('x2').

  • 'adverb': For each synset, the first word representing the synset is taken and assigned the argument 'e1'. In the gloss of the synset on the RHS, the argument 'e1' refers to the same action (or a modification of the same action). For eg,

    grossly:RB(e1) -> in:IN(e1, x1) gross:JJ(x1) manner:NN(x1)
    largely:RB(e1) -> mainly:RB(e1) chiefly:RB(e1)

Currently, in the first release, the Logic Forms have been generated using a simple bottom-up processing of the parse-tree. In this technique, arguments are first assigned to the leaves of the tree, and then propogated up the tree.

Work is currently in progess to implement a much more powerful recursive top-down processing of the parse-tree. This technique recursively applies grammar rules (for instance 'S -> NP VP PP') starting from the root of the tree, and is expected to yield much better results.

An overview of the quality of the Logic Forms:

POS TOTAL GOLD SILVER NORMAL
Noun87810140002411449696
Verb133731337300
Adj203371116009177
Adv3922392200