Based on the Tutsu's answer and the ensuing comments, there is confusion about how to deal with such data. The phrase structure approach as presented in Haegeman (2006) and as advocated for by Tutsu and many others is a dense and complex approach to syntax. There is a simpler way to address sentences like the one from Greek in the question. The answer I now produce touches on some of the points from the comments, in particular concerning the inclusion of tree diagrams in answers and the status of so-called non-configurational languages and analyses.
Assuming a dependency grammar (DG) approach, the answer to the first question is that ο άντρας 'the man' is a postdependent of μιλάει 'speak', and the answer to the second question is "Yes, the subject NP is a sister V" insofar as if one converts the dependency analysis to a phrase structure analysis, the resulting tree would have ο άντρας 'the man' as a sister of the verb μιλάει 'speak'.
Part of the difficulty with the sentence is due to the presence of a discontinuity (i.e. a long distance dependency). The prepositional phrase Σε ποιόν φίλο 'To which friend' is a complement of the embedded verb μιλάει 'speak', which appears to right of the matrix verb νομίζεις 'think'. This situation results in crossing lines in the tree structure. Regardless of what type of tree analysis one produces, dependency or phrase structure, one has to have a means of addressing such discontinuities. Movement is of course one prominent means of doing this. Another, and one which I prefer, assumes feature passing through the tree structure from displaced constituent down to the governor of that constituent.
The next tree shows the presence of the discontinuity on a DG analysis:
The crossing lines identify the discontinuity. An approach to such discontinuities that assumes feature passing might look something like this:
The wh-focused constituent is attached higher up in the tree and identified with a dashed dependency edge. The words in red mark the path in the tree from that constituent down to its canonical position. The dangling dependency edge marks this canonical position. Red is hence used to mark the path in the tree along which information about the wh-fronted constituent is passed down to its governor and canonical position. This is akin to movement in a a transformational approach.
To provide a point of comparison, the English translation given for the Greek sentence is also analyzed here in terms of dependencies as follows:
The crossing lines again identify the discontinuity. And again assuming feature passing, the following analysis obtains:
The conventions are consistent. The dashed dependency edge marks the wh-focused constituent; the words in red identify the path through the tree from that constituent down to its governor; and the dangling dependency edge marks the canonical position of that constituent.
The DG analyses of the Greek sentence above are consistent with claims about non-configurational languages more generally. The key trait of such languages is that they reject the existence of a finite VP constituent and thus have flatter structures. DG by its vary nature has always denied the existence of such a constituent for all languages, including English!
Finally, I produce such trees using the drawing commands in Microsoft Word. Producing such DG trees is relatively easy once you get the hang of it. I can also use Microsoft Word to produce phrase structure trees, although such trees demand more effort because they are more complex by their very nature.