A Cooperative Lightweight Translation Algorithm Combined with Sparse-ReLU

Table 6

The comparison results between this model and others (German-English translation task using IWSLT14 dataset).

		BLEU	Size M	Model
Research [17]		23.1	—	Encoder model based on 6-layer CNN.
Research [28]		28.83	—	Tag-less backtranslation
Research [29]		29.9	—	Linear transformer
Research [16]		34.4	—	Random feature attention
Research [30]		34.8	285	Pay less attention with lightweight CNN
		35.20	296	Pay less attention with dynamic CNN
Traditional transformer model [4]		34.44	49.27	Traditional transformer model
This paper scheme	Small model + Sparse-ReLU	34.29	36.42	Small transformer
		35.16		Sparse-ReLU + Small transformer
	Small model + Sparse-ReLU + CNN	35.24	37.99	Sparse-ReLU + small transformer + CNN