Difference between revisions of "NLP Status Report 2017-5-15"

From cslt Wiki
Jump to: navigation, search
 
Line 17: Line 17:
 
||
 
||
 
* test embedding untrained model
 
* test embedding untrained model
 +
* refine embedding untrained model
 
* fix reversed sentences problem and rerun on MOSES, NMT, M-NMT
 
* fix reversed sentences problem and rerun on MOSES, NMT, M-NMT
 
* implement UNK model  
 
* implement UNK model  

Latest revision as of 04:37, 15 May 2017

Date People Last Week This Week
2017/4/5 Jiyuan Zhang
Aodong LI
Shiyue Zhang
  • got result of M-NMT 28.92 ( +2.2, baseline=26.73)
  • trained word2vec on big zh,uy data
  • tested NMT baseline when there is UNK in ref, bleu=34.10 (better than MOSES=33.10), which means UNK is the biggest problem in NMT
  • found a problem in dataset, some sentences are reversed
  • test embedding untrained model
  • refine embedding untrained model
  • fix reversed sentences problem and rerun on MOSES, NMT, M-NMT
  • implement UNK model
Shipan Ren
  • configured environment and ran tf_translate code
  • read machine translation paper
  • learned lstm model and seq2seq model
  • learn the implement of seq2seq model
  • read tf_translate code
  • understand the meaning of main code