Page source:

<webl>     WubCall("sataire.incHeader", []);      </webl>
<webl>     WubCall("sataire.incTop", []);         </webl>



<!-- content here 
============================================================ -->
				<div id="header">
					<!-- img src="/images/sataire.jpg" width="332" height="453" alt="Sataire" id="person" / -->
					<h1><img src="/images/logo.gif" width="259" height="107" alt="SATAIRE" /></h1>
					<div id="wide">

						
<p id="heading">Thesis Proposal</p>				
						
					
IBM has successfully defined and achieved two milestone ?challenge problems? in the field of AI that captured the imagination of the general public by having a computer perform a task that seems innately human in its intelligence: playing chess (Deep Blue) and playing jeopardy (Watson).  The objective of this thesis project is to create a system that can take and ?pass? the online SAT (Scholastic Aptitude Test) that american students use to get into universities.  
<p>
The SAT is made up of the following components: Reading, Writing, and Mathematics.  The reading section includes Passage Based Reading questions and Sentence Completion questions.  The Math section includes both multiple choice question as well as student produced responses.  The system will have to understand word problems, graphs and charts, and so forth.  Writing has sections on Improving Sentences, Identifying Sentence Errors, Improving Paragraphs to apply conventions of standard written English, and then includes a 25 minute original essay where the student will be asked to present a point of view on a particular issue.
<p>
Various artificial intelligence techniques will be developed and applied to produce a system capable of responding at human-level performance in these areas, but the underlying common theory that will tie together the work throughout each area is the notion of Analogy, or transferring information from one subject to another.  Early work on analogical reasoning was done in the 1960?s and a fair amount of work occurred in the late 1990?s and early 2000?s, but little has been done since the ready availability of text-based machine learning algorithms across big-data infrastructures.  In the SAT case, analogy can be useful in all the test areas, from reading comprehension (e.g. ?bird::nest AS dog::doghouse?), mathematics (detecting what class of math problem is being presented and then transferring and applying previous solutions to the specific example presented), and even writing (many essays fall into various patterns, and transferring and adapting previously-seen relevant essays to the topic presented could be an effective strategy for composing a response to an essay question).
<p>
Evaluation: comparisons will be drawn to other test-taking systems (e.g. HALO), but the end evaluation will be how well the system performs on an actual SAT test, and whether this score would be a credible entrance score at a four-year US university.	
						
					</div>
					
<webl>     WubCall("sataire.incNav", []);         </webl>

				</div>

<!-- END CONTENT
============================================================ -->

<webl>     WubCall("sataire.incFooter", []);      </webl>