If I understand your question correctly, it's not too difficult
 (if I have misunderstood, please let me know)
Firstly we're dealing with two axes, let's call them x (along the horizontal) and y (vertical)
as usual.
With a "normal" graph, each axis is marked off in linear increments, that is to say moving the
same distance (say 1 cm) along each axis away from the origin (the meeting point of the two
 axes) corresponds to an increase in the relevant (x or y) coordinate of the same amount (say 1 unit)
With a "semi-log" graph, the x axis (for example) is the same as above, in the "normal" graph, BUT
the y axis is marked off differently. Now, moving the same distance (say 1 cm) along the y axis
away from the origin, from ANY point, corresponds to a scaling (i.e. multiplication) of y 
coordinate by a fixed amount (say 10 times), rather than an addition. Thus, travelling outwards
 from the origin, equidistant points would be marked 1, 10, 100, 1000, 10000, etc. rather than
1, 2, 3, 4, etc. You will note that this corresponds, mathematically to a logarithmic mapping - 
hence the name. This compression of the y coordinate in a logarithmic way can be useful when dealing
with plotting points with widely varying values. Bear in mind, though that when correlating the 
plotted data, that a straight line on the graph paper no longer refers to a linear relationship
 between x and y, but rather a logarithmic one.
With a "log-log" graph, as you might expect, BOTH the axes, x and y, are delineated in this non-linear
fashion described immediately above.