diff --git a/README.md b/README.md
index b50c65f92479889cd4b394378409d7217364f845..5a6883a3ef87c0b862b743b25dea64e86d02d981 100644
--- a/README.md
+++ b/README.md
@@ -8,6 +8,7 @@ Codebase for [LAVA: Latent Action Spaces via Variational Auto-encoding for Dialo
             
 ### Data
 The pre-processed MultiWoz 2.0 data is included in data.zip
+
 Unzip the compressed file and access the data under **data/norm-multi-woz**.
             
 ### Over structure:
@@ -31,23 +32,27 @@ The supervised training step of the variational encoder-decoder model could be d
     - sl_cat: train a latent action model with categorical latent variables using SL,
     - sl_gauss: train a latent action model with continuous latent varaibles using SL,
 
+
 2. using the VAE models as pre-trained model (equivalent to LAVA_pt):
 
     - finetune_cat_ae: use the VAE with categorical latent variables as weight initialization, and then fine-tune the model on response generation task
     - finetune_gauss_ae: as above but with continuous latent variables 
     - Note: Fine-tuning can be set to be selective (only fine-tune encoder) or not (fine-tune the entire network) using the "selective_finetune" argument in config
 
+
 3. using the distribution of the VAE models to obtain informed prior (equivalent to LAVA_kl):
 
     - actz_cat: initialized new encoder is combined with pre-trained VAE decoder and fine-tuned on response generation task. VAE encoder is used to obtain an informed prior of the target response and not trained further.
     - actz_gauss: as above but with continuous latent variables
     - actz_e2e_cat: same model as actz_cat, but without using database pointers and belief state from corpus
 
+
 4. or simultaneously from scrath with VAE task in a multi-task fashion (equivalent to LAVA_mt):
 
     - mt_cat
     - mt_gauss
 
+
 No.1 and 4 can be directly trained without Step 1. No. 2 and 3 requires a pre-trained VAE model, given via a dictionary 
     pretrained = {"2020-02-26-18-11-37-sl_cat_ae":100}
 
@@ -66,6 +71,9 @@ The script takes a file containing list of test results from the SL step.
 
 
 ### Checking the result
-The evaluation result can be found at the bottom of the test_file.txt 
-We provide the best model in this repo
+The evaluation result can be found at the bottom of the test_file.txt. We provide the best model in this repo
+
 NOTE: when re-running the experiment some variance is to be expected in the numbers due to factors such as random seed and hardware specificiations. Some methods are more sensitive to this than others.
+
+### Contact
+Any questions or bug reports can be sent to lubis@hhu.de
\ No newline at end of file