A Comparative Study of Some Automatic Arabic Text Diacritization Systems

Algorithm 5

convert_A_Sequence_Of_Embeddings_to_A_Sequence_Of_Indexes

(i)	Input: 2D tensor representing the sequence of embeddings seq,
(ii)	//shape of seq is (time_steps, embedding_dim)
(iii)	2D tensor for the learned embedding matrix emb_Mat,
(iv)	//shape of Emb_Mat is (number of chars, embedding_dim)
(v)	Output: a tensor t//of shape (time_steps) where each row contains the index of the char
(vi)	seq_shape: = shape(seq)
(vii)	b_shape: = shape(emb_Mat)
(viii)	//tile seq along new dimension
(ix)	seq_tiled: = tile(seq, [1,b_shape[0]])
	//reshape
(x)	seq_tiled: = reshape(seq_tiled, [seq_shape[0], b_shape[0],seq_shape [1]])
(xi)	//Elementwise comparison
(xii)	eq: = equal(emb_Mat, seq_tiled)
(xiii)	//Reduce the last dimension
(xiv)	red: = reduce(eq, -1)
(xv)	//element where condition eq is True
(xvi)	z: = where(red)
(xvii)	t: = z[:,1]
(xviii)	RETURN t