diff --git a/README.md b/README.md
index 1dd5071a22dced29bc104941f2895068e1666a5b..27b2789ec172fb995e3bb09c0b248060341124c4 100755
--- a/README.md
+++ b/README.md
@@ -3,15 +3,29 @@
 **GenTUS** is a user simulator for task-oriented dialogues, which consists of an encoder-decoder structure and is able to optimise both the user policy and natural language generation jointly. GenTUS generates both semantic actions and natural language utterances, preserving interpretability and enhancing language variation. In addition, by representing the inputs and outputs as word sequences and by using a large pre-trained language model we can achieve generalisability in feature representation.
 
 ## Training
-Building dataset
+* Building dataset
 ```
 python3 convlab2/policy/genTUS/build_data.py --add-history
 ```
-Train the model
+The data will be under `convlab2/policy/genTUS/data` as default.
+
+* Train the model
 ```
 python3 convlab2/policy/genTUS/train_model.py --batch-size 8
 ```
+The result `YOUR_MODEL_DIR` may be named as `genTUS-YY-MM-DD-hh-mm` as default.
+
+* Evaluate the semantic level performance
+The test data `TEST_DATA` will be `convlab2/policy/genTUS/data/data_test_v1.json` as default.
 
+```
+python3 convlab2/policy/genTUS/evaluate.py --do-semantic --dataset multiwoz21 --model-checkpoint YOUR_MODEL_DIR --evaluate-file TEST_DATA --only-action
+```
+
+* Evaluate the natural language performance
+```
+python3 convlab2/nlg/corpus_evaluation.py --model-checkpoint YOUR_MODEL_DIR --input-fname TEST_DATA --do-generation --model-name gentus
+```
 ## Citing
 
 If you use GenTUS in your research, please cite: