The notion of symmetric operator on Hilbert space can be generalized to locally convex spaces as follows: A map $T \colon D \to X^a$ between a subspace $D$ of a locally convex topological vector space $X$ and its algebraic dual $X^a$ is called symmetric, if
$$(Tx,y)_{X^a,X} = (x,Ty)_{X,X^a}$$,
where the brackets denote duality pairings. This condition automatically ensures that $T$ is linear.
For the special case where $X$ is a Banach space $B$ and $T$ is defined a) on a dense linear subspace or b) on all of $B$, does this symmetry condition ensure that $T$ must map into the continuous dual $B^{\ast}$?
As the trivial example of an operator defined on $\{0\}$ shows, this is not true in general.
Here are some related results:
If $T \colon B \to B^{\ast}$ is symmetric, then it is continuous (this is a generalization of the Helling-Toeplitz theorem on Hilbert spaces and can be found as Proposition 1.1 on p. 145 in the book "Probability on Banach spaces" by Vakhania, Tarieladze and Chobanyan). From this
If $T \colon B^{\ast} \to B^{\ast a}$ is the covariance operator of some Borel measure on $B$ (so in particular positive symmetric), then $T$ maps into $B^{\ast\ast}$. This is claimed (referring to an article in Russian and not easily available) on page 12, before stating Theorem 6 of Covariance operators of probability measures in locally convex spaces by Vakhania and Tarieladze.