Let $X$ be an homogeneous projective variety, written as the quotient $G/P$, where $G$ is a Lie group and $P$ is a parabolic subgroup of it. It seems it is wellknown that the monoid of effective curves on $X$, as a submonoid of $H_2(X, \mathbb{Z})$, is generated by finitely many curves $\beta_1, \ldots, \beta_p$. Moreover the $\beta_i$ are embedded $\mathbb{P}^1$'s in $X$. For instance in the expository notes http://arxiv.org/abs/alggeom/9608011, the authors use this fact (e.g. at the beginning of p. 37). The question is simply: how does one prove this fact? Is there a reference for a precise proof of it (perhaps in a slightly different generality, or proving the same statement for the effective monoid inside the Chow group, instead of homology)?

One place this is treated is in Brion's notes, http://arxiv.org/abs/math/0410240, particularly Section 1.4, and the references in the notes at the end of that section. A little more precisely, one shows that a divisor (line bundle) is nef iff it is globally generated, and this cone is generated by the Schubert divisors. Using the fact that the Schubert classes form a selfdual basis for $H^*(X,{\Bbb Z})$ (as well as $A^*X$) under the intersection pairing, it follows that the "dual classes" to the Schubert divisors generate the cone of curves. (As with many such things in $G/P$ world, the arguments for cohomology and Chow groups are exactly the same.) By the way, it's an easy consequence of the existence of a dense open $B$orbit that the pseudoeffective cone is generated by the complement of the open orbit  that is, again by the Schubert divisors. So for $G/P$, the nef and pseudoeffective cones are the same. 

