This project considers the process of conventionalization of metaphors, and
the way in which conventionalization can be incorporated into the Rational
Speech Act (RSA) based modeling approach described in Kao et al. (2014).
Conventionalization of metaphors is discussed in Bowdle and Gentner (1999),
where it is analyzed as arising due to relationships between subjects,
predicates, and abstract metaphoric categories. This project takes a
different approach to analyzing conventionalization, following the
pragmatic account in Kao et al., and proposes that it arises due to a
restriction on the space of features which are a part of the interpretation
of a predicate term.
As in the basic RSA model, the model in Kao et al. posits that listener and
speaker layers reason recursively about one another, in order to arrive at
a pragmatically enriched meaning of an utterance. The primary addition to
the basic RSA model in Kao et al. is the introduction of speaker goals and
sets of features. The listener must consider what the speaker’s goal
potentially is, and use the speaker’s goals to determine the
characteristics of the real-world object described by the metaphor.
While the model in Kao et al. (2014) gives the correct predictions for
novel metaphors, it does not capture the distinction between conventional
and novel metaphors. An extension to the model is therefore proposed, which
will allow for this distinction to arise, while still maintaining a feature
and goal-based analysis of metaphors.