Languages have a quite large inventory of expressions to refer to an imprecise
quantity such as n ≥ 4. In English, for instance, this quantity can be described
by the expressions 4 or more, at least 4, more than 3, minimally 4, over 3,
etc. Do these expressions actually convey the exact same meaning? No. Some
of them, i.e., at least 4, 4 or more, minimally 4 (vs. more than 3, over 3),
have been found to additionally convey speaker ignorance e↵ects and these effects
are pragmatic in nature. This dissertation probes experimentally speaker
ignorance e↵ects as well as two other types of meaning, i.e., variation e↵ects
and speaker indi↵erence e↵ects, by looking at the o↵-line and the real-time
comprehension of utterances with two distinct kinds of numeral modifiers, represented
by at least and more than. The aim is to find out where one should
draw the dividing line between the core meaning and the pragmatic meaning
of each kind of modifiers, and how di↵erent these two are, with a main focus on
their pragmatics. Experimental data reveal that all three types of inference are
available with both at least and more than, and are non-obligatory, contextdependent
pragmatic inferences. It is further shown that more than triggers
variation and speaker ignorance inferences to a smaller extent compared to at
least. These findings point to an account where the inferences of at least and
more than come about through di↵erent routes. Those associated with more
than are based on standard Quantity reasoning. The same holds partly for at
least, but in addition superlative modifiers trigger a Manner implicature based
on a conventional signal of anti-specificity. As to speaker indi↵erence inferences,
the opposite pattern seems to arise between the two numeral modifiers. This
could still be accommodated by the aforementioned account as long as we take
conventionalization processes into consideration.
