Skip to Main Content
Table 2: 
SMCalFlow results. Agent action accuracy is significantly higher than a baseline without metacomputation, especially on turns that involve a reference (Ref. Turns) or revision (Rev. Turns) to earlier turns in the dialogue (p < 10−6, McNemar’s test).
FullRef. TurnsRev. Turns
devtestdevtestdevtest
# of Turns 13,499 21,224 3,554 8,965 1,052 3,315 
 
Dataflow .729 .665 .642 .574 .697 .565 
inline .696 .606 .533 .465 .631 .474 
FullRef. TurnsRev. Turns
devtestdevtestdevtest
# of Turns 13,499 21,224 3,554 8,965 1,052 3,315 
 
Dataflow .729 .665 .642 .574 .697 .565 
inline .696 .606 .533 .465 .631 .474 
Close Modal

or Create an Account

Close Modal
Close Modal