ABSTRACT

            Diagnosing Reading Strategies: Paraphrase Recognition
                              Chutima Boonthum
                        Old Dominion University, 2006
                     Director:   Dr. Irwin B. Levinstein

      Paraphrase recognition is a form of natural language  processing  used
in  various  types  of  applications  such  as  tutoring  systems,  question
answering (QA)  systems,  and  information  retrieval  (IR)  systems.   This
attempt to recognize the  most  common  paraphrase  patterns  amongst  these
applications led us to a definition of the  most  commonly  used  paraphrase
patterns.  The context of this paraphrase recognitionthe present work is  in
an automated reading strategy trainer called  iSTART  (Interactive  Strategy
Trainer for Active Reading and Thinking).  An The ability to  recognize  the
use of paraphrase - a complete, partial, or inaccurate paraphrase;  with  or
without extra information - in the  student's  input  is  highly  essential,
especially for if the trainer in orderis to give appropriate  feedback.   We
analysed the most common patterns of paraphrase and  developed  a  means  of
representing the semantic structure of sentences.  To accomplish this  goal,
internal representations of a given sentence and  the  student's  input  are
generated and used in the recognition process.  Paraphrases  are  recognized
by transforming sentences into this representation and comparing  them.   To
construct a precise semantic representation, it is  extremely  important  to
understand the meaning of a prepositions.   A  system  augmented  withAdding
preposition disambiguation to the original system improved its  accuracy  by
20%.shows significant improvement beyond one without by  over  20%  to  90%.
The preposition  sense  disambiguation  module  itself  achieves  about  80%
accuracy  on  average  for  the  top  10  most  frequency  frequently   used
prepositions.
      The main contributions of this work to the research community  is  are
on the preposition classification and generalized disambiguation  processes,
which are integrated into the paraphrase recognition system  and  are  shown
to be quite effective.  The recognition model also forms a significant  part
of this contribution.
      I  achieved  (1)The  present  effort  includes  the  modeling  of  the
paraphrase recognition process. featuring the Syntactic-Semantic Graph as  a
sentence representation, .the implementation of  a  significant  portion  of
this design demonstrating its effectiveness, the modeling  of  an  effective
preposition classification based  on  their  prepositional  usage,  and  the
designing  of  the  generalized  preposition  disambiguation   module,   (2)
modeling the paraphrase recognition process and implementing  a  significant
portion of this design demonstrating its effectiveness,  (3)  integratingand
the integration  of  the  preposition  disambiguationthat  module  into  the
paraphrase recognition system and so as  to  gain  significant  improvement,
and   (4)   featuring   the   Syntactic-Semantic   Graph   as   a   sentence
representation.