This paper uses the distribution and interpretation of antonymous adjective
s in comparative constructions as an empirical basis to argue that abstract
representations of measurement, or 'degrees,' must be modelled as interval
s on a scale, rather than as points, as commonly assumed. I begin by demons
trating that the facts in this domain must be accounted for in terms of the
interaction of the semantics of adjectival polarity and the semantics of t
he comparative, rather than principles governing the (overt) expression of
particular types of adjectives in comparatives. I then show that a principl
ed account of the full range of data under consideration can be constructed
within a model in which degrees are formalized as intervals on a scale and
adjectival polarity is characterized in terms of two structurally distinct
and complementary sorts of 'positive' and 'negative' degrees.