Skip to Main Content
Table 2: 
BLEU scores on NIST dev set MT06 from rerankers which are incorporated with various language models. In the language model column X: Y means the language model X is trained on dataset Y. A bigger language model improves the doc-reranker but does not help the sent-reranker.
Proposal modelLanguage modelSent-rerankerDoc-reranker
Sent-transformer LSTM: NIST 49.92 50.24 
transformer-XL: NIST 50.29 50.56 
transformer-XL: NIST + Gigaword 50.19 50.93 
 
Doc-transformer LSTM: NIST 50.75 51.20 
transformer-XL: NIST 51.27 51.68 
transformer-XL: NIST + Gigaword 51.33 51.99 
Proposal modelLanguage modelSent-rerankerDoc-reranker
Sent-transformer LSTM: NIST 49.92 50.24 
transformer-XL: NIST 50.29 50.56 
transformer-XL: NIST + Gigaword 50.19 50.93 
 
Doc-transformer LSTM: NIST 50.75 51.20 
transformer-XL: NIST 51.27 51.68 
transformer-XL: NIST + Gigaword 51.33 51.99 
Close Modal

or Create an Account

Close Modal
Close Modal