diff --git a/convlab2/dst/sumbt/README.md b/convlab2/dst/sumbt/README.md
index 68f4f5968eff65a400ef9ea26645cd9f7b176534..9d8f6b250afa0fae6d3bb131ba0b473a7cbaec2d 100755
--- a/convlab2/dst/sumbt/README.md
+++ b/convlab2/dst/sumbt/README.md
@@ -36,8 +36,6 @@ We use the multiwoz data.
 
 ## Performance on Multiwoz
 
-`mode` determines the data we use: if mode=`usr`, use user utterances to train; if mode=`sys`, use system utterances to train.
-
 We evaluate the Joint accuracy and Slot accuracy on Multiwoz 2.0 validation and test set. 
 The accuracy on validation set are slightly higher than the results reported in the paper,
 because in the evaluation code all undefined values in ontology are set `none` but predictions