Depictional Metonymies

Josef Meyer-Fujara (University of Applied Sciences, Stralsund) and
Hannes Rieser (Bielefeld University and SFB 360)
Abstract
Our research on metonymy is based on a corpus of construction dialogues. The dialogues were elicited in experimental settings, where two agents, an Instructor and a Constructor, had to see to it that the Constructor built a toy airplane according to the directives of the Instructor (cf. fig. 1). Despite using only general-purpose construction elements, the agents used airplane terminology in referring and predicating expressions. We call these occurrences of airplane terminology "depictional metonymies". From the data it is clear that one can’t develop theories of reference, predication, etc., without some understanding of how these metonymies work with respect to semantics and pragmatics. The constructed examples (1) and (2) should make the basic issues clear:

(1)
I.: We are going to build an airplane.
C.: O.K.

(2)
I.: The wing goes onto the five-holes bar.
C.: The seven-holes bar we just talked about?

Many of the corpora we know contain metonymies of a similar sort, for example, the Edinburgh Map Task Corpus. Also, psycholinguistic research settings involving pictures show this kind of metonymies. We expect that the investigation of our domain will yield findings relevant for other kinds of depicting objects like maps, pictures, icons, models or software-robots.


Fig. 1: Toy airplanes of the experimental setting, seen from different perspectives

In order to discuss metonymy we use a simple logical language (a PL1 version) whose predicates can get a non-literal (i.e. a metonymical) interpretation. Expressions of this language are evaluated with respect to a model, the domain of which subsumes depicting as well as non-depicting objects. Depicting is thus considered a property of things. Concerning depicting objects we have developed a separate theory, which we do not want to treat in greater detail here. Roughly, the idea is that depicting objects behave very much like natural language expressions in the sense that they can be equipped with structure and meaning. Their meanings are the objects depicted.

The main difference between a normal PL1-model and the model we use is that instead of one interpretation function we have a set of interpretation functions for constants. The different interpretation functions capture the literal and various possible metonymical interpretations. This is because an expression, say police, in the utterance

(3)
Walk up to the police!

can be interpreted as building or institution or representative of the institution or a place contiguous to the police building or whatever.

The problem which interpretation function to choose is handled as follows: We assume that inter alia the context determines which interpretation is salient. Intuitively, this captures a speaker’s knowledge to decide whether, say, a toy airplane or a real airplane is at stake. In order to model this role of context, we use a special selection function over the set of interpretation functions.

With this tool we investigate the semantics of simple utterances like This is an airplane. said in a context where toy airplanes as well as Airbusses are around. Our approach enables a discussion about possible or necessary relations between natural and metonymical extensions of predicates. We suppose that our approach can be generalized to capture non-depictional metonymies or even metaphors. Another interest we developed of late is to discuss problems of compositionality with respect to metaphors like in (4),

(4)
die Katze im Sack kaufen (literally: buy the cat in the bag; buy a pig in the poke)

where constituents of different complexity can be affected by non-literal meaning. Compositionality considerations are also relevant for metonymical contexts, as extensions of (2) such as

(2a)
I.: The wing of the airplane goes on the five-holes bar.
(2b)
I.: The wing you have goes on the five-holes bar.

clearly show.

Bibliography

sfb-logo to SFB-Homepage created by: Anke Weinberger (2000-09-06).
maintained by: Anke Weinberger (2000-09-06).