Good question. Below isn't the best answer you could get, but it was too long for a comment.

Perhaps it is better for the understanding of these theories to have the point of view of the Galois representation (as in the survey you put in your question). First, a $p$-adic Galois representation $\rho$ is said to be modular if it is equivalent to a $p$-adic Galois representation $\rho_{f}$ attached to a cuspidal newform $f$.

Then the modularity of elliptic curves can be stated as "for some $p$, the Galois representation $\rho_{E,p}$ attached to the elliptic curve $E$ is modular". Note that this is equivalent to your statement "$a_{l}(E)=a_{l}(f)$" according to Carayol's theorem (and the fact that the trace of $\rho_{f,p}(\mathbf{Frob }_{l})$ is $a_{l}(f)$). That's what people try to generalize, and that's also the Langlands setting.
Finally, in a perspective of generalization, it is good to keep in mind that $\rho_{E}$ is in fact $H^{1}_{et}(E,\mathbb{Q}_{ p})$ . In fact, for any projected smooth curve $X$, the space $H^{1}_{et}(X,\mathbb{Q}_{p})$ will always be a (nice i.e. de Rham) Galois representation.

Therefore, you can first ask if (a subrepresentation of) the $H^{1}_{et}(X,\mathbb{Q}_{p})$ is isomorphic to a representation from of the *automorphic world*.

But for a genus $g$ curve the space $H^{1}_{et}(X,\mathbb{Q}_{p})$ has dimension $2g$, so the Langlands philosophy tells you that 'it should be associated with a cuspidal representation of $\mathbf{Gl}_{2g}$ (or maybe if your representation have nice proprieties something a bit smaller like $\mathbf{GSpin}_{2g+1}$. **Edit** actualy this is always happening, see the comment of David Loeffler bellow). I believe that we cannot hope for a more precise conjecture, it is too general. But my opinion must be considered with reserve as this is really deep and complicated.