The short answer is in this paragraph, with more detail in other paragraphs. The particle acts a bit like a blob of mist spread out over position and momentum space. The intensity of the mist in a given region tells your the probability of finding it there. If you squash the blob too much in position space then it will be more spread out in momentum space and vice versa. the velocity tells you something about how fast the blob will spread out in position space. This doesn't require that the particle has a single well defined position.
The particle has some state
$$|\psi(0)\rangle = \sum\sqrt{p_a}|R_a\rangle,$$
where the $|R_a\rangle$ means that the particle is in the region $R_a$. There is no state for a particle being at a single point since no measurement can tell whether a particle is at a single point. Rather, the relevant states are for a particle to be in a region and if you measure more precisely the regions are smaller. More precise measurements are said to be more fine grained.
If you measure the particle then you couple the particle to a measuring instrument, the resulting state is
$$|\psi(1)\rangle = \sum\sqrt{p_a}|R_a\rangle|M_a\rangle.$$
where the $|M_a\rangle$ means that the particle has been measured as being in the region $R_a$.
What this means is that there are multiple versions of the particle. Different versions of the particle can sometimes interact with one another in interference experiments, for a discussion see "The Fabric of Reality" by David Deutsch, Chapter 2. There are also multiple versions of the measuring instrument and each version of the instrument sees a particular version of the particle. The different versions of the measuring instrument can't interact with one another as a result of a phenomenon called decoherence:
http://arxiv.org/abs/quant-ph/0703160.
Decoherence also explains why you don't see other versions of yourself although their existence is a consequence of quantum theory.
The probability for finding the particle in $R_a$ is $p_a$ and this means that the proportion of versions of the particle with value $R_a$ is $p_a$. For an explanation of why this is the appropriate measure see
http://arxiv.org/abs/quant-ph/9906015.
Now, you could write down the state in a different way as
$$|\psi(0)\rangle = \sum\sqrt{q_b}|P_b\rangle,$$
using states $|P_b\rangle$ in which the particle has momentum in a particular range and the $q_b$ are the probabilities for each of those states.
The $q_b$ and $p_a$ are not independent and if you have a suitably fine grained set of position states, then if the $p_a$ are peaked near some particular value, the $q_b$ will have multiple non-zero values, and vice versa.