Under scale/multidimensional scaling/Model in SPSS you are asked for level of measurement. These are usually either ordinal or interval. The default is ordinal. This makes the basic assumption that the measurements, of how close stimuli are, may be ranked. It does not take into account the relative levels of closeness. If one can also make the assumption that the measures are linear then we have interval scaled measures. In the Reaction Time example the reaction time for responding to an animal word (goat) immediately after seeing a person word (eskimo) will not equal the reaction time to a person word immediately after an animal word. So the difference in reaction times will depend upon the relative ordering of the words. This is not interval scoring which assumes the differences between any pair of stimuli is the same (ignoring sign), irrespective of ordering.
The relationship could still be ordinal if the RTs between pairs of living things are faster than the RTs involving living things and household objects. Ie there is a ranking in stimuli RTs.
In summary the ordinal level of measurement is the safer bet as it makes fewer assumptions.