Skip to Main Content
Table 1: 
Dataset statistics. “Library Size” counts distinct function names (e.g., findEvent) plus keyword names (e.g., start=). “Length” and “Depth” columns show (.25, .50, .75) quantiles. For programs, “Length” is the number of function calls and “Depth” is determined from a tree-based program representation. “OOS” counts the out-of-scope utterances. MultiWOZ statistics were calculated after applying the data processing of Wu et al. (2019). Vocabulary size is less than reported by Goel et al. (2019) because of differences in tokenization (see code release).
# Dialogues# User TurnsUser Vocab. SizeLibrary SizeUtterance LengthProgram LengthProgram DepthOOS
SMCalFlow 41,517 155,923 17,397 338 (5, 8, 11) (11, 40, 57) (5, 9, 11) 10,466 
MultiWOZ 2.1 10,419 71,410 4,105 35 (9, 13, 17) (2, 29, 39) (2, 6, 6) 
# Dialogues# User TurnsUser Vocab. SizeLibrary SizeUtterance LengthProgram LengthProgram DepthOOS
SMCalFlow 41,517 155,923 17,397 338 (5, 8, 11) (11, 40, 57) (5, 9, 11) 10,466 
MultiWOZ 2.1 10,419 71,410 4,105 35 (9, 13, 17) (2, 29, 39) (2, 6, 6) 
Close Modal

or Create an Account

Close Modal
Close Modal