Dataset: MSCOCO 
Learning Rate: 5e-5

Semantic Similarity: SBERT
Syntactic Diversity: Dependency Parse Tree edit distance
Lexical Diversity: Character-level edit distance 
Phonological Diversity: Rhythmic Diversity
Morphological Diversity: POS edit distance.

Train Loss (MSE): 0.0127
Dev Loss (MSE): 0.0136