In the paper Phrase-Based & Neural Unsupervised Machine Translation on page 4 the authors give a probability equation:
$$\Large{ p(t_j|s_i) = \frac{e^{\frac{1}{T}\cos(e(t_j), W e(s_i))}}{\sum_ke^{\frac{1}{T}\cos(e(t_k),W e(s_i))}} }$$
Here, twice the authors use $cos$ with two arguments, both at the top and the bottom of the fraction. There is no reference to $\cos$ indicating any other function in the paper. The paper does mention a rotation matrix, which leads me to believe that $\cos$ is used in its classical trigonometric form. I have looked online for instances where this notation is used, but could not find any.
Can someone explain what the double arguments mean in this instance?
This answer was provided by @zwim in the comments. When we refer to the Conneau article cited in the paper, we find that $\cos$ refers to the cosine similarity function, which can take two vectors as an input.