tech,15-1-E06-1004,ak are of considerable importance to <term> Statistical Machine Translation ( SMT ) </term> but which have not been addressed
other,30-1-E06-1004,ak been addressed satisfactorily by the <term> SMT research community </term> . Over the last decade , a variety
tech,8-2-E06-1004,ak Over the last decade , a variety of <term> SMT algorithms </term> have been built and empirically tested
other,22-2-E06-1004,ak whereas little is known about the <term> computational complexity </term> of some of the fundamental problems
tech,31-2-E06-1004,ak some of the fundamental problems of <term> SMT </term> . Our work aims at providing useful
other,10-3-E06-1004,ak providing useful insights into the the <term> computational complexity </term> of those problems . We prove that
model,4-4-E06-1004,ak those problems . We prove that while <term> IBM Models 1-2 </term> are conceptually and computationally
model,22-4-E06-1004,ak involving the higher ( and more useful ) <term> models </term> are hard . Since it is unlikely that
other,8-5-E06-1004,ak it is unlikely that there exists a <term> polynomial time solution </term> for any of these <term> hard problems
other,15-5-E06-1004,ak time solution </term> for any of these <term> hard problems </term> ( unless P = NP and P #P = P ) ,
other,38-5-E06-1004,ak and justify the need for developing <term> polynomial time approximations </term> for these computations . We also
other,9-6-E06-1004,ak some practical ways of dealing with <term> complexity </term> . In this paper a novel solution
skrij podrobnosti